The clinical decision regarding a need for treatment of menstrual bleeding relies on the patient’s perception of flow amount and its effects on her physical, emotional, and social well-being.1 However, retrospective recall regarding the amount of menstrual flow in previous cycles is heavily influenced by a woman’s subjective perception and is not always associated with the measured blood loss.2 The ‘gold standard’ approach for assessment of menstrual blood loss is the alkaline haematin method, which requires a woman to collect all soiled sanitary products for laboratory assessment2; however, this is a cumbersome non-hygienic impractical method outside the research setting.

The Pictorial Blood Loss Assessment Chart (PBAC) is a scoring system developed as a semi-quantitative evaluation of menstrual blood loss, which considers the number of sanitary products used, the degree to which these products are soiled with blood, the number and size of blood clots passed, and the number of flooding episodes.3 The PBAC has been validated with the alkaline haematin method to diagnose heavy menstrual bleeding in several studies in other populations.3 4 5 Furthermore, the PBAC has been used as a measurement tool to evaluate menstrual blood loss in systematic reviews and randomised controlled clinical trials.6

In the clinical setting, it can be difficult for a physician to determine the amount and implication of menstrual flow in a patient reporting heavy menstrual bleeding. Menstruation is a taboo topic in many communities, including among Asian women in Hong Kong.7 8 9 10 Some women may prefer not to, or find it difficult or embarrassing to, articulate details regarding their menstrual problems.7 8 9 Furthermore, some women may be unaware of heavy menstrual bleeding.

The objective of this study was to evaluate the usefulness of the PBAC as a self-evaluation tool for heavy menstrual bleeding. Additionally, we sought to determine the acceptability of the PBAC and whether PBAC scores were associated with menstrual blood loss severity among Asian women in Hong Kong.

This prospective cohort study compared PBAC scores between women who presented with and without heavy menstrual bleeding. Women were recruited between November 2014 and January 2016 through the gynaecology ward or the general gynaecology out-patient clinic of a university-affiliated hospital. They attended the out-patient clinic for routine follow-up or were admitted to the ward for elective or emergent treatment. Inclusion criteria included good general health, absence of other medical conditions which might lead to anaemia, no prior PBAC use, and age ≥18 years. Women were excluded if they were pregnant, in menopause, receiving hormonal treatment, mentally incompetent, and/or undergoing treatment/monitoring of a gynaecological malignancy. Ethics approval was obtained from the Institutional Review Board of The University of Hong Kong/Hospital Authority Hong Kong West Cluster. Written informed consent was obtained from all study participants.

Women were approached by the research nurse and were placed into heavy menstrual bleeding and normal menstrual bleeding groups based on their self-reported menstrual cycle symptoms over the preceding 6 months. All group allocations were noted by the nurse. All participants, regardless of perceived menstrual flow, were instructed by the research nurse to fill in a PBAC for one cycle in the next cycle. They were also instructed to answer a question regarding whether they found the PBAC acceptable (yes/no) and a question regarding the ease of use of the PBAC (scale of 1-5; 1=easiest and 5=hardest). The PBAC originally described by Higham et al3 was used, but diagrams of clot sizes were modified to the sizes of local coins. The PBAC consisted of a series of diagrams representing lightly, moderately, and heavily soaked towels and tampons (depending on the degree of staining) to evaluate menstrual blood loss.3 The numbers of pads or tampons used each day were recorded. In the event of clot passage, the number and size were recorded; flooding episodes were also recorded. A total score was calculated by multiplying by a factor of 1 for each lightly soiled item, 5 for each medium soiled item, 10 for a fully soaked tampon, and 20 for a fully soaked pad.3 Small and large clots were given a score of 1 and 5, respectively.3 Women continued to use their own sanitary products (ie, products used prior to the study) and were asked to document the types and sizes of sanitary products used. Each woman was asked to return the completed PBAC to the research nurse by mail in a stamped envelope. The following clinical data were retrieved from the women’s electronic medical records and used in the analysis: age, haemoglobin level within 3 months before the consultation or on the day of consultation (if available), and the iron supplement status (using/not using).

The sample size was determined based on an anticipated 20% difference in accuracy endpoints between study groups and a standard deviation of 40%. Allowing for 10% non-responders, the calculated sample size per group was 70 women. Statistical tests were performed using SPSS Statistics (Windows version 24; IBM Corp, Armonk [NY], United States). Comparisons between groups were made using the Chi squared test for categorical variables and the non-parametric Mann–Whitney U test for continuous variables. Continuous variables were expressed as median and range. A P value of <0.05 was considered statistically significant. The kappa statistic was used to test agreement between subjective evaluation of heavy menstrual bleeding and the PBAC score at various cut-off scores. Predictions of heavy menstrual bleeding according to the PBAC score and haemoglobin level were determined using area under the receiver-operating characteristic curve analysis.

The response rate was better than expected and more women than expected were recruited in each clinic session; this yielded a final sample size larger than originally planned. However, among 292 women who were asked to complete the PBAC, the return rate was only 206/292 (70.5%). In all, 118 women had self-perceived heavy menstrual bleeding and 88 women had self-perceived normal menstrual flow. Haemoglobin level data were available in 179/292 (61.3%) women (116 in the heavy menstrual bleeding group and 63 in the normal menstrual bleeding group). Table 1 summarises the reasons for presentation in both groups of women. The PBAC scores based on different diagnoses are shown in Table 2. Women with heavy menstrual bleeding were older than women with normal menstrual bleeding (median age 44 years, [interquartile range=40-48] vs 38 years [interquartile range=31-43], respectively, P<0.001). There was no significant difference in education level between groups (Table 3).

Nearly all women in the study used pads; one woman used both pads and tampons. In total, 147/206 (71.4%) women used various brands and sizes of pads with distinct absorbency characteristics during the menstrual cycle; the remaining 59/206 (28.6%) women used only one type of pad. Seven women used diapers and three women used postpartum pads. The median PBAC scores of women who reported heavy and normal menstrual bleeding were 497 (interquartile range=152-1112) and 54 (interquartile range=41-65), respectively (Table 3). Compared with women who had normal menstrual flow, women with heavy menstrual bleeding had significantly higher total PBAC scores and numbers of flooding episodes, larger clot sizes and numbers, more days of bleeding, and lower haemoglobin levels (Table 3). Using cut-off scores of 76, 80, 100, 130, 150, and 185, levels of agreement between PBAC score and self-reported symptoms in the diagnosis of heavy menstrual bleeding are shown in Table 4. Women with anaemia, defined as haemoglobin level <11.0 g/dL, had significantly higher median PBAC scores than did women without anaemia (508 [interquartile range=168-1087] vs 58 [interquartile range=46-84], P<0.01). Receiver-operating characteristic curves demonstrating the predictive abilities of the PBAC and haemoglobin level for heavy menstrual bleeding are shown in the Figure. The area under the receiver-operating characteristic curves of the PBAC and haemoglobin level for prediction of heavy menstrual bleeding were 0.961 (95% confidence=0.940-09.982) and 0.876 (95% confidence=0.821-0.931), respectively. The PBAC cut-off score with the highest Youden index was 76, which yielded a sensitivity of 93.2% and a specificity of 83.0% for predicting self-perceived heavy menstrual bleeding.

All women in our study were able to complete the PBAC. Missing information was filled in with the help of the research nurse via phone contact after return of the PBAC. Twenty-eight women (13.6%) who began the PBAC on the day of consultation were contacted by phone to urge them to return the PBAC using the stamped envelopes. Another 11 women (5.4%) with prolonged menstrual bleeding did not provide full details regarding their menstrual bleeding; they were contacted by phone for confirmation. In all, 200/206 women (97.1%) found the PBAC acceptable: 113/118 (95.8%) in the heavy menstrual bleeding group and 87/88 (98.9%) in normal menstrual bleeding group. Assuming that the reason for non-response was that those women found the PBAC to be unacceptable, the acceptability rate was 200/292 (68.5%). There was no significant difference in the perceived ease of use of the PBAC; the median rating was 2 in both groups (P=0.618; Table 3). Notable written comments from the women concerning the PBAC were that it could not accurately describe their menstrual blood loss (n=19), it required explanation (n=11), it was inconvenient or involved recall problems (n=3), and it did not record other symptoms which were more distressing (n=1).

Our results suggested that the reported PBAC scores in this group of Asian women comprised a useful tool for differentiating self-perceived heavy and normal menstrual bleeding. Heavy menstrual bleeding considerably impacts a woman’s quality of life; interventions should be designed to improve the quality of life, rather than focusing on the exact amount of menstrual blood loss.1 Nevertheless, some women may be unaware of heavy bleeding or find it difficult to describe the amount of menstrual flow. The PBAC offers a semi-objective method for initial self-evaluation of the amount of menstrual bleeding in women whose cultural backgrounds may cause reluctance in discussing their gynaecological or menstrual problems. This self-evaluation can alert women to seek medical attention, thus facilitating clinical evaluation and treatment. The PBAC cut-off scores included in Table 4 have been used in previous studies to imply heavy menstrual bleeding.3 4 5 11 The recommendation of a particular cut-off score depends on the clinical context (ie, whether a higher sensitivity or specificity is required). For example, if the PBAC is used as a screening tool, a lower cut-off score may be appropriate to alert women to seek medical attention. In contrast, if the PBAC is used to evaluate women with heavy menstrual bleeding for potential participation in a research study, a higher cut-off score may be used to recruit women with more severe symptoms to evaluate their response to treatment.

In our study, 10 women (11.4%) in the self-perceived normal menstrual bleeding group had PBAC scores of >100, although they reported normal menstrual bleeding. In the self-perceived normal menstrual bleeding group, 14 women had anaemia (haemoglobin level <11.0 g/dL), among which five women had a haemoglobin level of <10.0 g/dL. Twelve women who reported normal menstrual bleeding were using iron supplements. Although most women accurately recognised heavy menstrual bleeding, use of the PBAC identified an additional 10% of women who might have unperceived abnormal bleeding. Of the 10 women with self-perceived normal menstrual bleeding (PBAC scores of 101-180), seven (70%) had anaemia. Thus, use of the PBAC might enable identification of a small group of apparently asymptomatic women who had unrecognised anaemia, thereby facilitating earlier medical attention.

In our study, women were asked to use their own sanitary products, rather than using specific brands and sizes of pads; thus, our findings are more representative of realistic PBAC use, compared with results acquired in a research setting. Most women used different brands and sizes of pads with different absorbency characteristics, even within a single cycle. In addition, several women used adult diapers or postpartum pads, which implied substantial difference in blood loss compared with the usual sanitary pads. The range of PBAC scores was much larger in our study than in previous studies.3 4 5 11 12 One woman in our study had a PBAC score of 32 301; she had prolonged vaginal bleeding for 56 days and had a haemoglobin level of 4.5 g/dL. Women with adenomyosis and uterine fibroids had significantly higher PBAC scores than did women with other diagnoses. Therefore, the PBAC may be useful for evaluation of responses to interventions during randomised controlled trials involving these groups.

Although women in our study who returned the PBAC found it acceptable and generally easy to use, the return rate should be considered. Notably, 19/206 (9.2%) women commented that the range of icons in the PBAC did not accurately reflect their blood loss on pads or clots because they experienced difficulty in evaluating the amount of blood loss (based on a particular stain) when comparing among pads with different absorbency characteristics. The clots were of irregular size and women felt that a scale or use of items encountered daily (such as ‘tofu’ or ‘palm’, rather than coins) could more accurately describe these clots. Women (particularly in the heavy menstrual bleeding group) who had to sit on the toilet during flooding episodes could not quantify their bleeding; several women with prolonged bleeding did not continue the PBAC evaluation because they felt that continuing the documentation was time-consuming and annoying. In total, 5.3% of the women commented that clearer instructions could be provided. This is consistent with the findings by Zakherah et al,5 who reported that improved instructions led to greater accuracy when a physician or nurse reviewed the documentation with the patient. The role of the nurse in our study was crucial. Our research nurse found it helpful to demonstrate to the women how to fill in the PBAC using their current or previous cycle; the nurse also helped the women to complete the PBAC in the event of substantial missing information, especially among women with prolonged menstrual bleeding. Some women probably completed the PBAC by recall, rather than in a day-by-day manner. This aspect should be considered when the PBAC is applied as a self-screening tool. The development of PBAC-containing mobile apps or websites accessible by the public may improve the usability of the PBAC as a self-screening tool in terms of better convenience and less recall bias, especially among younger women.

Our study had some limitations. First, we only evaluated use of the PBAC in a small group of patients who presented for clinical treatment, rather than the general population; this may limit the generalisability of the results. Second, we did not study the inter-cycle variability in PBAC score or the effects of other demographic factors (eg, household income) which may affect the use of the PBAC. Although only one cycle of menstrual bleeding was charted in our study and women may have unusual menstrual flow in subsequent cycles, previous studies have demonstrated high consistency with low inter-cycle variation in women who completed a second PBAC evaluation without treatment.11 Third, patients may have been offered treatment during the consultation; because the PBAC was completed in the cycle after consultation, the PBAC score may not fully reflect the pre-consultation reported symptoms, especially among women with self-perceived heavy menstrual bleeding. Fourth, compliance with iron therapy was not checked; this could have affected the haemoglobin results. However, the aim of our study was to evaluate the relationship between the PBAC score and self-perceived menstrual flow. Overall, the results of this population-specific study might support the use of the PBAC as a potential self-screening tool for heavy menstrual bleeding among Asian women in Hong Kong.

There is considerable endpoint heterogeneity in the current literature with respect to the outcomes of various treatment options for heavy menstrual bleeding. Furthermore, there is currently no core outcome set for valid comparison and interpretation of data from research studies and assessments regarding abnormal uterine bleeding.6 Although PBAC scores have shown high inter-individual variation, they had low intra-individual variation;11 thus, the PBAC may be useful in future studies of treatment responses in individual women. Despite the large variety of commercially available sanitary products, the PBAC remains a reliable screening tool for semi-quantitative evaluation of menstrual blood loss, which can alert women to seek medical attention for heavy menstrual bleeding. Additional studies are needed to confirm the clinical usefulness of the PBAC, especially in the context of the evolution and advancement of superabsorbent sanitary products currently available. Overall, the advantages of the PBAC are its relative objectivity and flexibility as a tool for screening, diagnosis, and evaluation of treatment effect.

Concept or design: JKY Ko, VYT Cheung.
Acquisition of data: JKY Ko, VYT Cheung.
Analysis or interpretation of data: All authors.
Drafting of the manuscript: JKY Ko.
Critical revision of the manuscript for important intellectual content: All authors.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

The authors thank Ms Wai-ki Choi for patient recruitment, teaching women about the Pictorial Blood Loss Assessment Chart, and managing the database.

The study was presented in an oral presentation at the FOCUS in O&G 2018 Congress in Hong Kong (17-18 November 2018).

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Ethics approval was obtained from the Institutional Review Board of The University of Hong Kong/Hospital Authority Hong Kong West Cluster (Ref: UW 14-299). Written informed consent was obtained from all study participants.

1. National Institute for Health and Care Excellence guideline. Heavy menstrual bleeding: assessment and management. 14 Mar 2018 (last updated 24 May 2021). Available from: http://www.nice.org.uk/guidance/ng88. Accessed 25 Nov 2021.

2. Magnay JL, O’Brien S, Gerlinger C, Seitz C. A systematic review of methods to measure menstrual blood loss. BMC Womens Health 2018;18:142. Crossref

3. Higham JM, O’Brien PM, Shaw RW. Assessment of menstrual blood loss using a pictorial chart. Br J Obstet Gynaecol 1990;97:734-9. Crossref

4. Janssen CA, Scholten PC, Heintz AP. A simple visual assessment technique to discriminate between menorrhagia and normal menstrual blood loss. Obstet Gynecol 1995;85:977-82. Crossref

5. Zakherah MS, Sayed GH, El-Nashar SA, Shaaban MM. Pictorial blood loss assessment chart in the evaluation of heavy menstrual bleeding: diagnostic accuracy compared to alkaline hematin. Gynecol Obstet Invest 2011;71:281-4. Crossref

6. Herman MC, Penninx J, Geomini PM, Mol BW, Bongers MY. Choice of primary outcomes evaluating treatment for heavy menstrual bleeding. BJOG 2016;123:1593-8. Crossref

7. Garg S, Anand T. Menstruation related myths in India: strategies for combating it. J Family Med Prim Care 2015;4:184-6. Crossref

8. The Lancet Child Adolescent Health. Normalising menstruation, empowering girls. Lancet Child Adolesc Health 2018;2:379. Crossref

9. Agampodi TC, Agampodi SB. Normalising menstruation, empowering girls: the situation in Sri Lanka. Lancet Child Adolesc Health. 2018;2:e16. Crossref

10. Wong WC, Li MK, Chan WY, et al. A cross-sectional study of the beliefs and attitudes towards menstruation of Chinese undergraduate males and females in Hong Kong. J Clin Nurs 2013;22:3320-7. Crossref

11. Hald K, Lieng M. Assessment of periodic blood loss: interindividual and intraindividual variations of pictorial blood loss assessment chart registrations. J Minim Invasive Gynecol 2014;21:662-8. Crossref

12. Reid PC, Coker A, Coltart R. Assessment of menstrual blood loss using a pictorial chart: a validation study. BJOG 2000;107:320-2. Crossref

Burnout among doctors is increasingly recognised as a serious threat to medical and dental practice across all specialties; its prevalence is increasing worldwide.1 Burnout is a spectrum of clinical syndromes that were first categorised into three dimensions by Maslach as emotional exhaustion, depersonalisation, and a low sense of personal accomplishment.2 Subsequently, it was added to the International Classification of Diseases as a syndrome that results from poorly managed chronic workplace stress.3

Burnout among doctors can lead to decreased effectiveness and shortening of professional lifespan.4 Burnout exacerbates negative emotions, thereby impeding cognitive performance; it may result in biased decision making. Hence, the well-being of doctors is important for maintaining manpower, quality of care, and equity of care delivery. Multiple studies in different countries have shown that the incidence of burnout among doctors is rising. In the US, a Medscape nationwide survey showed that 59% of emergency medicine doctors experienced burnout symptoms, and the incidence had increased steadily over time.1 In Australia, the National Mental Health Survey found that the level of very high psychological distress was significantly greater in doctors (3.4%) than in the general population (2.6%) or other professionals (0.7%).5 A cross-sectional online survey in the United Kingdom also revealed a high rate of mental health disorders among junior doctors and medical students.6

To our knowledge, studies regarding well-being and burnout among doctors in Asia are limited. Gan et al7 performed a cross-sectional study of general practitioners in Hubei, China; they found a combined prevalence of 2.46% across all three dimensions of emotional exhaustion, depersonalisation, and personal accomplishment. However, that study only included doctors within a single specialty in one province. Huang et al8 found the prevalences of personal burnout and client-related burnout were 44.0% and 14.8%, respectively, among residents in Taiwan; however, they used a non-standardised questionnaire. In Hong Kong, a previous cross-sectional survey showed that 31.4% of respondents among doctors in the public sector had a high rate of burnout, but the sampling criteria were random and non-specific; moreover, that study did not include a substantial proportion of doctors who worked in the private sector.9 Another survey also suggested that burnout was prevalent among doctors in Hong Kong, but it only included graduates from one medical school in Hong Kong.10

Hence, this study aimed to evaluate the prevalence of burnout in the Hong Kong medical and dental workforce by administering standardised questionnaires to a broad population of residents-in-training and doctors within 10 years of their specialist registration. The study also explored well-being among doctors in terms of job satisfaction, depression, lifestyle behaviours, and factors associated with these states.

In this study, all doctors within 10 years of their specialist registration registered with the Hong Kong Academy of Medicine, as well as residents-in-training registered with one of the Academy’s 15 constituent Colleges, were invited to complete a voluntary cross-sectional survey between February 2019 and June 2019. The cut-off of 10 years was selected because the Hong Kong Academy of Medicine considers doctors within 10 years of their specialist registration to be “young Fellows”. The survey consisted of self-reported demographic data, year of entry into medical school, and current professional details. Burnout was assessed using the validated Copenhagen Burnout Inventory (CBI).11 Depression was assessed using the Patient Health Questionnaire-9 (PHQ-9).12 Lifestyle factors were assessed with reference to the respondents’ drinking habits, sleep patterns, and levels of both exercise and activities. Items concerning job satisfaction and lifestyle behaviours were adapted from existing doctor questionnaires and health surveys.13 14

An online survey was developed in-house by the Hong Kong Jockey Club Innovative Learning Centre for Medicine of the Hong Kong Academy of Medicine, then administered electronically. The invitations to participate were sent via e-mail; two separate reminder emails were sent after the initial invitation. As an incentive, respondents were offered coffee or food coupons after completion of the survey. The study protocol was approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Ref No: UW 19-062).

PASS 2000 (NCSS, LLC., Kaysville [UT], US; www.ncss.com) power analysis software was used for sample size calculation. The prevalence of personal burnout among young doctors in Hong Kong was assumed to be similar to the prevalence of personal burnout among residents in Taiwan (44.0%)8; thus, to achieve a 95% confidence interval (CI) with a precision of 4.5%, 458 participants were required. Our final sample of 514 doctors was sufficient to achieve the desired statistical power.

This instrument consists of three scales that measure personal burnout, work-related burnout, and client-related burnout; the scales can be applied to workers in all industries and cultures. Personal burnout measures the degree of fatigue experienced by the respondent, irrespective of work experience or occupational status. Work-related burnout measures the degree of fatigue related to work; it explores how the respondent’s perception of work contributes to fatigue. Client-related burnout is the perceived degree of fatigue related to work with clients. The burnout level is calculated as a mean score; therefore, each scale has a value between 0 and 100. A score of ≥50 indicates a high degree of burnout.15 16 17 18

The PHQ-9 is a depression assessment tool, which scores each of the nine Diagnostic and Statistical Manual of Mental Disorders IV criteria for depression on a scale ranging from “0” (not at all) to “3” (nearly every day). A PHQ-9 score >9 has a reported sensitivity of 88% and specificity of 88% for major depression.19

The prevalence of burnout is shown using point estimates and 95% CIs. Descriptive statistics were presented concerning demographic characteristics and lifestyle behaviours. Bivariate logistic models were used to describe the distinct relationships of suicide, depression, and burnout with demographic, educational, and professional characteristics. The data were analysed using SPSS software (Windows version 26.0; IBM Corp, Armonk [NY], US). Statistical significance was set at P<0.05.

There were 746 total respondents; of these, 232 did not complete the entire survey and were excluded from the analysis. Of the included 514 respondents, 284 were doctors within 10 years of their specialist registration, while 230 were residents-in-training. The total number of doctors within 10 years of their specialist registration invited to participate in the survey was 2879; thus, the response rate was estimated as 9.9%. However, it was not possible to calculate the response rate for residents-in-training. The respondents included 277 women (54%); the mean age among all respondents was 33.7 ± 6.1 years. The respondents’ demographic data are summarised in Table 1; their current professional statuses are summarised in Table 2.

Overall, 24% (n=125) of respondents were “somewhat dissatisfied” with their present job position, while 4% (n=19) of respondents were “very dissatisfied” with their present job position. Furthermore, 15% (n=76) of respondents were “somewhat dissatisfied” with being a medical doctor, whereas 2% (n=10) of respondents were “very dissatisfied” with being a medical doctor. Finally, 3% (n=14) of respondents indicated they planned to stop practising medicine in the next 12 months, with stress or burnout (86%) cited as the most common reason for such plans.

As measured by the CBI, the mean personal burnout score was 59.6 ± 20.5, work-related burnout score was 57.3 ± 20.1, and client-related burnout score was 49.0 ± 22.3. Using a CBI subscale cut-off score of ≥50 (moderate and higher), 72.6% (n=373, 95% CI=68.5%-76.4%) of respondents reported personal burnout; 70.6% (n=363, 95% CI=66.4%-74.5%) of respondents reported work-related burnout; and 55.4% (n=285, 95% CI=51.0%-59.7%) of respondents reported client-related burnout (Table 3).

The mean physical component summary score of the 12-Item Short Form Survey was 49.6 ± 7.8; the mean mental component summary score of the 12-Item Short Form Survey was 42.3 ± 10.6 (Table 3).

As measured by the PHQ-9, the mean depression score was 6.2 ± 5.1. However, the prevalence of depression among respondents, defined as a score of ≥10, was 21% (n=110) [Table 3].

In total, 79% (n=404) of respondents did not report any suicidal ideation or attempt. The remaining respondents stated that life was “not worth living” or “wished he or she was dead”; some also reported a history of suicidal ideation or attempts. The most commonly cited source of stress in the past year was clinical responsibilities/job demands.

In terms of health conditions, there was a perception among respondents that their health status was “worse” (29%; n=148) or “much worse” (3%; n=17) than among other individuals of the same age. The mean duration of sleep each night was 6.2 ± 1.5 hours. However, most respondents frequently experienced inadequate sleep when at work; 70% (n=361) of respondents indicated that this occurred weekly or more often. In terms of personal habits, the prevalences of alcohol drinking, drug addiction, and smoking were low. However, the prevalences of regular physical activity and personal physical assessments were not high (Table 4).

Logistic regression modelling was performed to investigate bivariate associations of demographic and professional factors with burnout, the presence of depression, or suicide ideation and/or attempts.

The number of working hour(s) per week (odds ratio [OR]=1.02; 95% CI=1.01-1.04; P=0.001) was positively associated with depression (online supplementary Table 1); having children (OR=0.58; 95% CI=0.36-0.93; P=0.024) was negatively associated with suicidal ideation/attempts. Doctors who completed a project-based learning curriculum during undergraduate study were less likely to be depressed or report suicidal ideation/attempts (depression: OR=0.60; 95% CI=0.39-0.91; P=0.017; suicidal ideation/attempts: OR=0.65; 95% CI=0.43-1.00; P=0.049) [online supplementary Table 2].

Older age (OR=0.97; 95% CI=0.94-0.99; P=0.026), possession of a first university degree in medicine or dental surgery (OR=0.37; 95% CI=0.15-0.89; P=0.027), and possession of Academy fellowship status (OR=0.61; 95% CI=0.41-0.92; P=0.017) were associated with lower likelihood of personal burnout. Engagement in longer working hour(s) per week (OR=1.04; 95% CI=1.02-1.05; P<0.001) and working in Hospital Authority clinics (OR=1.95; 95% CI=1.05-3.62; P=0.034; compared with working in government clinics) were positively associated with personal burnout (online supplementary Table 3). Marital statuses of single, separated, or divorced (OR=1.71; 95% CI=1.16-2.53; P=0.007) and engagement in longer working hour(s) per week (OR=1.03; 95% CI=1.02-1.05; P<0.001) were positively associated with work-related burnout (online supplementary Table 4). Conversely, having children (OR=0.66; 95% CI=0.44-0.98; P=0.038), consultant seniority level (OR=0.27; 95% CI=0.09-0.88; P=0.029; compared with associate consultant seniority level), and working in the private sector (OR=0.40; 95% CI=0.17-0.94; P=0.035; compared with working in government) were negatively associated with work-related burnout. Provision of primary care (OR=1.5; 95% CI=1.04-2.16; P=0.031) was associated with client-related burnout (online supplementary Table 5).

This study attempted to quantify well-being and burnout in young doctors (both resident-in-training, and doctors within 10 years of their specialist registration) throughout Hong Kong; there were three main findings. First, the mean burnout score was high in this group of doctors; mean personal and work-related scores of ≥50 were observed on the CBI. Second, there was a high prevalence of job dissatisfaction (28%) in this group of doctors. Third, the self-perceived personal well-being and mental health were worse in this group of doctors than in members of the general population with similar ages.

Burnout is a well-known occupational hazard in people-oriented professions; doctors are at particular risk of burnout because of their frequent engagement in intense personal and emotional contact with patients. Although these therapeutic and service relationships are highly rewarding and engaging, they can also be a source of stress. Burnout among doctors has been recognised as a global crisis20; its effects on personal, patient, and institutional levels can be substantial. The expectation to meet job demands can lead to maladaptive practices which will ultimately compromise relationships with patients and colleagues, with long-term consequences on patient care.21 Hence, efforts to acknowledge that such a problem exists represents the first step in establishing a systematic strategy to address this crisis.

Although there have been multiple published reports regarding burnout among doctors, territory-wide data focusing on junior doctors in Hong Kong are lacking. Siu et al9 conducted a random sample survey of 226 public doctors in 2012; they found that 31.4% of respondents satisfied the criteria for high burnout. Moreover, young but moderately experienced doctors needing to work shifts were most vulnerable to high burnout. However, the questionnaire used in that study was not comprehensive, the random sampling method did not produce a representative cohort, and only public doctors were invited to the survey. More recently, a more comprehensive survey involving medical graduates of one university in Hong Kong found high prevalences of personal (63.1%) and 55.9% (work-related) burnout using the standardised CBI.10 The more comprehensive survey represents the most comprehensive and robust study in Hong Kong thus far, but it only included graduates from one university in Hong Kong; it did not include any doctors trained elsewhere.

The present study of young doctors throughout Hong Kong found high mean personal (59.6 ± 20.5) and work-related (57.3 ± 20.1) scores on the CBI. The mean client-related score was 49.0 ± 22.3, slightly below the score of 50 that constituted the threshold for burnout. These scores were higher than in the previous study performed in Hong Kong by Ng et al,10 which showed mean CBI scores of 57.4 ± 21.4 (personal), 48.9 ± 7.4 (work-related), and 41.5 ± 21.8 (client-related). Moreover, when compared with studies worldwide that used the CBI to measure burnout in doctors,15 16 18 22 the levels of burnout in the present study were among the highest. Contributing factors may differ among regional healthcare systems; causes of burnout and well-being in junior doctors may not be consistent worldwide. Our study attempted to identify sources of stress among junior doctors in Hong Kong; the most commonly cited sources were clinical responsibilities/job demands and professional examinations. Additional in-depth studies are necessary to determine how these factors can be modified to alleviate stress in junior doctors.

The respondents’ general health statuses (in terms of medical conditions) were not substantially worse than the general population, although 32% of the respondents indicated self-perceived health worse than their peers. The present study also showed that the prevalence of depression was 21%, according to the PHQ-9. This is more than double the prevalence previously reported in Hong Kong (8.4%).23 Despite the high prevalence of depression in the present study, respondents indicated low rates of suicidal ideation/attempts. Although a causal relationship could not be established because of the observational nature of the study, the number of working hours per week and having children were factors that affected risk of depression and suicidal ideation/attempts, respectively. The mean number of hours worked per week was 53.5 ± 14.8 hours. Junior doctors who work >55 hours per week are reportedly twofold more likely to have frequent health problems (OR=2.05, 95% CI=1.62-2.59; P<0.001) and suicidal ideation (OR=2.0, 95% CI=1.42-2.82; P<0.001).24 A previous systemic review showed an association between long working hours and a depressive state in other professions in general.25 Positive effects of reduced working hours among junior doctors have been found in some studies,26 27 but this relationship is not consistently observed. For example, in the United Kingdom, the Working Time Regulations were fully applied to junior doctors beginning in 2009; these comprised a limit of 48 hours per week, averaged across a reference period of 26 weeks, with additional minimum rest periods. However, implementation of the Working Time Regulations has not fully resolved the effects of long hours and fatigue.28 Furthermore, there are implications for professional training and manpower planning if rigid enforcement of such working hours is performed.

Our study did not find any substantial evidence that young doctors were reliant on alcohol, smoking, or drugs as coping mechanisms. This contrasts with findings from the US, which indicated that high levels of alcohol and substance abuse were associated with burnout among doctors.29 It was beyond the scope of the present study to explore other avenues that junior doctors in Hong Kong might use to alleviate their stress levels and burnout. Other health and lifestyle behaviours (eg, exercise levels and personal physical assessments) may be indicative of time constraints related to work or personal obligations; they may also be indicative of self-neglect caused by such constraints and work-related burnout.

This study had several limitations. First, it was a cross-sectional study with voluntary participation, and the results might not be representative of all doctors throughout public and private sectors in Hong Kong. However, to our knowledge, this study performed the most comprehensive survey regarding burnout among doctors in Hong Kong thus far. Second, the study was not designed to avoid selection bias concerning doctors who were more prone to burnout and therefore more interested to participate in such surveys. Third, because the survey did not allow free text entry in the questionnaire responses, more in-depth analysis was not possible in some instances. Fourth, because this was a cross-sectional survey, no causal relationships or risk factors could be established regarding the development of burnout or depression. Fifth, our definition of “young” was based on the 10 years of specialist registration, which included doctors with various levels of experience and responsibilities; thus, the results might not be representative of a specific subset of doctors.

The present study showed that junior doctors in Hong Kong had a high level of burnout, and there was a high prevalence of depression among the respondents. A substantial proportion of the respondents were dissatisfied with their present job position. Future studies to determine causal factors will allow the development and implementation of specific strategies to address these problems within Hong Kong. The maintenance of well-being in junior doctors is vital for sustaining a healthy medical workforce and long-term patient care.

Concept or design: All authors.
Acquisition of data: KYH Kwan, LWY Chan, PW Cheng.
Analysis or interpretation of data: KYH Kwan, LWY Chan, PW Cheng.
Drafting of the manuscript: KYH Kwan.
Critical revision of the manuscript for important intellectual content: All authors.

The authors have no conflicts of interest to disclose.

The authors thank all members of the Young Fellows Chapter of the Hong Kong Academy of Medicine for their active participation in this study; the secretariat and staff of the Hong Kong Academy of Medicine and its Hong Kong Jockey Club Innovative Learning Centre for Medicine for their administrative and information technology support; and the Council of the Hong Kong Academy of Medicine for their active support, encouragement, and funding of the coupons. The authors especially thank Dicken CC Chan for statistical assistance.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This study was approved by the Institutional Review Board of The University of Hong Kong/Hospital Authority Hong Kong West Cluster (Ref No: UW 19-062).

1. Kane L. Medscape national physician burnout and depression report 2019. Available from: https://www.medscape.com/slideshow/2019-lifestyle-burnout-depression-6011056. Accessed 2 Dec 2020.

2. Maslach C. What have we learned about burnout and health? Psychol Health 2001;16:607-11.Crossref

3. World Health Organization. International Classification of Diseases 11th Revision. QD85 Burnout. Available from: https://icd.who.int/browse11/l-m/en#/http://id.who.int/icd/entity/129180281. Accessed 3 Dec 2020.

4. Lin M, Battaglioli N, Melamed M, Mott SE, Chung AS, Robinson DW. High prevalence of burnout among US emergency medicine residents: results from the 2017 National Emergency Medicine Wellness Survey. Ann Emerg Med 2019;74:682-90. Crossref

5. BeyondBlue. National Mental Health Survey of Doctors and Medical Students 2019. Available from: https://www.beyondblue.org.au/docs/default-source/beyond-blue-s-strategic-plan-2020-2023/research-project-files/bl1132-report---nmhdmss-full-report_web.pdf?sfvrsn=845cb8e9_14. Accessed 3 Dec 2020.

6. Bhugra D, Sauerteig SO, Bland D, et al. A descriptive study of mental health and wellbeing of doctors and medical students in the UK. Int Rev Psychiatry 2019;31:563-8. Crossref

7. Gan Y, Jiang H, Li L, et al. Prevalence of burnout and associated factors among general practitioners in Hubei, China: a cross-sectional study. BMC Public Health 2019;19:1607. Crossref

8. Huang EC, Pu C, Huang N, Chou YJ. Resident burnout in Taiwan Hospitals—and its relation to physician felt trust from patients. J Formos Med Assoc 2019;118:1438-49. Crossref

9. Siu CF, Yuen SK, Cheung A. Burnout among public doctors in Hong Kong: cross-sectional survey. Hong Kong Med J 2012;18:186-92.

10. Ng AP, Chin WY, Wan EY, Chen J, Lau CS. Prevalence and severity of burnout in Hong Kong doctors up to 20 years post-graduation: a cross-sectional study. BMJ Open 2020;10:e040178. Crossref

11. Kristensen TS, Borritz M, Villadsen E, Christensen KB. The Copenhagen Burnout Inventory: a new tool for the assessment of burnout. Work Stress 2005;19:192-207. Crossref

12. Rizzo R, Piccinelli M, Mazzi MA, Bellantuono C, Tansella M. The Personal Health Questionnaire: a new screening instrument for detection of ICD-10 depressive disorders in primary care. Psychol Med 2000;30:831-40. Crossref

13. Heponiemi T, Kouvonen A, Vänskä J, et al. Health, psychosocial factors and retirement intentions among Finnish physicians. Occup Med (Lond) 2008;58:406-12. Crossref

14. Taylor K, Lambert T, Goldacre M. Future career plans of a cohort of senior doctors working in the National Health Service. J R Soc Med 2008;101:182-90. Crossref

15. Benson S, Sammour T, Neuhaus SJ, Findlay B, Hill AG. Burnout in Australasian younger fellows. ANZ J Surg 2009;79:590-7. Crossref

16. Chou LP, Li CY, Hu SC. Job stress and burnout in hospital employees: comparisons of different medical professions in a regional hospital in Taiwan. BMJ Open 2014;4:e004185. Crossref

17. Creedy DK, Sidebotham M, Gamble J, Pallant J, Fenwick J. Prevalence of burnout, depression, anxiety and stress in Australian midwives: a cross-sectional survey. BMC Pregnancy Childbirth 2017;17:13. Crossref

18. Wright JG, Khetani N, Stephens D. Burnout among faculty physicians in an academic health science centre. Paediatr Child Health 2011;16:409-13. Crossref

19. Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med 2001;16:606-13. Crossref

20. Physician burnout: a global crisis [editorial]. Lancet 2019;394:93. Crossref

21. Graham J, Potts HW, Ramirez AJ. Stress and burnout in doctors. Lancet 2002;360:1975-6. Crossref

22. Thomas LR, Ripp JA, West CP. Charter on physician well-being. JAMA 2018;319:1541-2. Crossref

23. Cheng CM, Cheng M. To validate the Chinese version of the 2Q and PHQ-9 questionnaires in Hong Kong Chinese patients. HK Pract 2007;29:381-90.

24. Petrie K, Crawford J, LaMontagne AD, et al. Working hours, common mental disorder and suicidal ideation among junior doctors in Australia: a cross-sectional survey. BMJ Open 2020;10:e033525. Crossref

25. Bannai A, Tamakoshi A. The association between long working hours and health: a systematic review of epidemiological evidence. Scand J Work Environ Health 2014;40:5-18. Crossref

26. Kiernan M, Civetta J, Bartus C, Walsh S. 24 Hours on-call and acute fatigue no longer worsen resident mood under the 80-hour work week regulations. Curr Surg 2006;63:237-41. Crossref

27. Kort KC, Pavone LA, Jensen E, Haque E, Newman N, Kittur D. Resident perceptions of the impact of work-hour restrictions on health care delivery and surgical education: time for transformational change. Surgery 2004;136:861-71. Crossref

28. Morrow G, Burford B, Carter M, Illing J. Have restricted working hours reduced junior doctors' experience of fatigue? A focus group and telephone interview study. BMJ Open 2014;4:e004222. Crossref

29. Oreskovich MR, Shanafelt T, Dyrbye LN, et al. The prevalence of substance use disorders in American physicians. Am J Addict 2015;24:30-8. Crossref

Considerable physical and psychosocial benefits are associated with participation in sporting activities.1 2 Physical activity has been demonstrated to reduce the risks of coronary heart disease, some cancers, obesity, hypertension, and type 2 diabetes mellitus.2 3 4 5 6 Its obvious merits have prompted several national health programmes, including the health programme in Hong Kong, to promote sports to the general public.7 8 9 However, sports participation carries a risk of injury, especially traumatic brain injury (TBI). It has been estimated that 20% of all TBIs are sports-related.10 In addition, up to 20% of sports-related TBI survivors (usually adolescents or young adults) experience chronic symptoms including headache, fatigue, and cognitive and balance difficulties.11 There is a global trend of increasing sports-related TBI incidence: from 3.5 to 31.5 per 100 000 population in the past decade.12 13 Because many patients with mild TBI do not seek medical attention, these figures likely underestimate the total burden of this condition.13 Population-based studies reviewing sports-related TBI are sparse; most target specific groups of individuals (eg, professional athletes) or rely on self-reporting surveys that lack a uniform definition and comprehensive assessment of brain injury.14 For similar reasons, studies reviewing outcome predictors have also been inadequate, thus hindering the generation of meaningful conclusions to guide governmental policy initiatives.14

Hong Kong is highly urbanised with an established public transport system, such that cycling is mainly regarded as a recreational activity.15 16 In terms of fatality rate, Hong Kong is among the most dangerous areas for cycling, compared with other cities such as New York, the US, or countries such as France.17

This study was performed to document the epidemiology of sports-related TBI among patients who required inpatient care by adopting territory-wide uniform diagnostic coding criteria, clear data definitions, and systematic assessments of radiologic findings. Because cycling is a popular sport in Hong Kong, factors predictive of intracranial haemorrhage (eg, the effect of helmet use) and poor functional outcomes among cyclists hospitalised with TBI were determined.

Patients who required inpatient care at any Hospital Authority institution for sports-related TBI from 1 January 2015 to 31 December 2019 were reviewed. The Hospital Authority is a public health service highly subsidised by the Hong Kong SAR Government; it is responsible for 90% of inpatient bed days in the city. Patients were identified by the International Classification of Diseases, Tenth Revision, Clinical Modification code (ICD-10-CM) designation for TBI 854.0 secondary to a sports-related external cause (E codes: E006-10). Data from clinical records, operation notes, medication prescriptions, and computed tomography (CT) brain scans from a central digital imaging repository were reviewed. In particular, the type of sport played, the clinical presentation of symptoms, the Injury Severity Score (ISS), the need for neurosurgery, length of hospitalisation, and diagnosis of post-concussion syndrome were recorded. Head injury was classified into mild (presenting Glasgow Coma Scale [GCS] score, 14-15); moderate (presenting GCS score, 9-13), and severe (presenting GCS score, 3-8), in accordance with criteria established by the Neurotraumatology Committee of the World Federation of Neurosurgical Societies.18 Post-concussion syndrome was defined in accordance with ICD-10 criteria. This required a 4-week duration of symptoms from at least three categories following a traumatic loss of consciousness. The symptom categories were headache, irritability, concentration impairment, insomnia, and a preoccupation with the aforementioned symptoms. For neurosurgical patients with cycling related–TBI, the mechanism of injury and their experience level (ie, professional athlete or amateur rider) were documented. All CT scans were reviewed by an independent assessor with 6 months of neurosurgical training experience who was blinded to the patients’ clinical characteristics and outcomes. The scans were first evaluated using the Rotterdam CT score, a commonly utilised validated radiological assessment system for the prognosis of patients with TBI. The classification has four distinct elements that require the appraisal of the degree of basal cistern obliteration, degree of midline shift, the presence (or absence) of an epidural mass lesion, and the presence (or absence) of intraventricular or traumatic subarachnoid haemorrhage (Table 1). In addition, the scans were assessed for skull fractures, cerebral contusions, and acute subdural haematomas (ASDHs). The primary outcome of the study was the presence of intracranial haemorrhage on the admitting CT scan. All potential predictors were categorised into patient-related, trauma-related, and radiological factors. The secondary outcome was unfavourable functional performance, defined as a Glasgow Outcome Scale (GOS) score of 3 to 5 on discharge from the hospital (3, severe disability; 4, persistent vegetative state; and 5, death).

Statistical analyses utilised the Chi squared test and Fisher’s exact test were used for categorical data such as patient gender or the use of antiplatelet medication. Independent-samples t test was used for continuous data such as patient age or duration of hospitalisation. Multivariate binary logistic regression was used to identify independent factors for the presence (or absence) of traumatic intracranial haemorrhage. A P value of <0.05 was considered statistically significant. Statistical analysis was conducted using SPSS (Windows version 20.0; IBM Corp, Armonk [NY], US).

In total, 720 consecutive patients were hospitalised with sports-related TBI during the 5-year study period, and 705 (97.9%) of them were admitted under neurosurgical care. This was equivalent to a crude incidence of 1.9 per 100 000 general population. The mean (± standard deviation [SD]) age was 32 ± 19 years; 521 (72.4%) patients were adults (≥18 years) and 568 (78.9%) were male. The most common sport was cycling (59.2%), followed by football (21.3%) and basketball (7.5%) [Fig a]. On admission, most (86.1%) patients were fully conscious. Overall, 658 (91.4%) patients had mild TBI, 41 (5.7%) patients had moderate TBI, and 21 (2.9%) patients had severe TBI. Post-traumatic seizures occurred in 36 (5.0%) patients. Furthermore, 324 (45.0%) patients had a loss of consciousness and 269 (37.4%) patients experienced post-traumatic retrograde amnesia. Only 19 (2.6%) patients were taking either antiplatelet or anticoagulant medication. Extracranial injuries were sustained by 208 (28.9%) patients; among them, injuries were mainly either limb abrasions or contusions (62.1%). The median ISS was 2 (interquartile range=2-8).

Intracranial haemorrhage was noted in 283 (39.3%) patients with TBI; 166 (23.1%) patients exhibited traumatic subarachnoid haemorrhage and 157 (21.8%) exhibited ASDH. Skull fractures were detected in 179 (24.9%) patients, with a median Rotterdam CT score of 2 (interquartile range=2-2). In total, 59 (8.2%) patients required neurosurgical intervention; 32 (54.2%) of them had good recovery with a median GOS score of 5 on discharge from the hospital and at 6 months. The mean (± SD) duration of hospitalisation was 5 ± 28 days. Among 307 (42.6%) patients in whom 6-month GOS scores could be assessed, good recovery was observed in 260 (84.7%). Post-concussion syndrome was diagnosed in 30 (6.2%) of 482 patients who attended scheduled follow-up outpatient consultations.

The crude incidence of recreational cycling-related TBI requiring hospitalisation was 1.1 per 100 000 population. A comparison was performed between cyclists with sports-related TBI and patients who had TBI because of other sports (Table 2). Cyclists were significantly older (P<0.001) [Table 2; Fig b]. Among 426 cyclists, 306 (71.8%) were male and 120 (28.2%) were female. However, the proportion of patients who were female was significantly higher among those who had TBI because of cycling (28.2%) than among those who had TBI because of other sports (10.9%; P<0.001). Cyclists were likely to exhibit more severe TBI with an almost three-fold greater risk of sustaining extracranial injury (odds ratio [OR] 2.8; 95% CI: 1.9-4.0), resulting in a higher ISS (P<0.001). Of cyclists admitted for head injury, 201 (47.2%) had intracranial haemorrhage, which was radiologically more extensive in terms of the Rotterdam CT score, compared with haemorrhage in non-cyclists (P<0.01). As a consequence, a greater proportion of cyclists had a worse GOS score on discharge from hospital (OR 2.8; 95% CI: 1.3-6.2) and at 6 months (OR 4.7; 95% CI: 2.1-10.5). The cause of death for all cyclists with 30-day mortality was severe TBI with medically refractory intracranial hypertension. Although the overall incidence was low, cyclists also had a greater risk of post-concussion syndrome (OR 2.5; 95% CI: 1.2-5.4).

Among the 426 cyclists in this study, 128 (30.0%) experienced bicycle accidents during the weekend; 10 (2.3%) of the cyclists were professional athletes. Most cyclists (273; 64.1%) accidently fell off their bicycle on their own (ie, without colliding into another object) on level ground. Of the injuries, 103 (24.2%) were sustained when the cyclist was traveling downhill; for the 28 patients with records of self-reported velocities, the estimated mean (± SD) velocity at the moment before injury was 40 ± 15 km/h. At the time of injury, 361 (84.7%) cyclists had not been wearing a helmet. Among eight (1.9%) patients who subsequently died, none had been wearing protective head gear.

Risk factors for traumatic intracranial haemorrhage among cyclists are shown in Table 3. Univariate analysis identified the following risk factors: age ≥60 years, use of antiplatelet medication, involvement in a motor vehicle collision, presence of moderate to severe TBI, and skull fracture. In univariate analysis, helmet wearing was protective against intracranial haemorrhage. Multivariate logistic regression identified the following independent risk factors: age ≥60 years, antiplatelet medication intake, moderate or severe TBI, and the presence of a skull fracture. Table 4 shows independent significant predictors for unfavourable GOS score on discharge from hospital: age ≥60 years, antiplatelet intake, severe TBI, intracranial haemorrhage, and the need for neurosurgical operative intervention.

As shown in Table 2, 375 (88.0%) hospitalised cyclists had mild TBI, whereas only 36 (8.5%) cyclists had moderate TBI and 15 (3.5%) had severe TBI. No protective effect of helmet use was noted in terms of reducing TBI severity across these GCS-defined categories (Table 3). However, among cyclists with mild TBI, helmets were significantly protective against intracranial haemorrhage and skull fracture, regardless of age, antiplatelet medication intake, or mechanism of injury (Table 5). Although the median Rotterdam CT score was comparable between cyclists with mild TBI who did or did not wear helmets (P=0.68), significantly fewer patients with head protection had epidural haematoma or ASDH. For patients with mild TBI who had intracranial haemorrhage, this difference in radiological factors led to a significantly shorter mean (± SD) duration of hospitalisation for patients who wore helmets (2.6 ± 2.9 days), compared with patients who did not (7.1 ± 11.6 days, P<0.001). However, there was no difference in the need for neurosurgical intervention among patients with mild TBI who had intracranial haemorrhage according to head protection status (P=0.17). Similarly, unfavourable GOS scores on discharge from hospital (P=0.43) and at 6 months (P=0.71) were comparable among patients with mild TBI who had intracranial haemorrhage, regardless of head protection status (Table 5).

The incidence of sports-related TBI in Hong Kong is 2 per 100 000 general population; this is lower than in other countries (eg, the US, Australia, or Italy), where the incidences range from 4 to 32 per 100 000 population.12 The lower incidence in Hong Kong is consistent with a previous finding that Hong Kong residents (especially children and adolescents) have lower physical activity and fitness levels than in other regions, according to a global evidence-based evaluation of such indicators from 49 countries.19 Considering the health benefits of an active lifestyle, there is a clear need to promote sports engagement in Hong Kong. A survey of 5701 residents performed by the Transport Department of the Hong Kong SAR Government estimated that 10% of households had bicycles available for use; moreover, 69% (4 million) of residents aged 15 years or older knew how to ride one.16 In addition, most survey respondents (73%) cycled for recreational or fitness purposes.15

Previous epidemiological studies of sports-related brain injuries revealed that cycling was one of the most frequent activities involved.10 12 20 21 In 2019, 1738 road traffic accidents involving cyclists were reported to the Hong Kong Transport Department.22 Half of these accidents (50.6%, 879/1738) occurred in recreational areas such as cycling tracks, parks, or playgrounds; eight (0.5%) patients experienced fatal injuries.22 In the past 10 years, the number of cyclist injuries in Hong Kong has increased by 5.2% per year.17 Compared with other regions worldwide, Hong Kong is one of the most dangerous areas for cycling.17 The fatality rate (per billion minutes cycled) in the city was 34, substantially higher than the rates in Stockholm, Sweden (3), France (4), and other metropolitan areas (eg, New York City [18] and Los Angeles [8]).17 These studies included riders primarily involved in commuting and the causes of death were not elucidated, but they indicate a growing need to enhance the safety of vulnerable road users.

To our knowledge, this is the first multicentre study to comprehensively document the outcomes of inpatients with recreational cycling-related TBI using standard assessment criteria. By comparison with patients who had TBI because of other sports, we found that cyclists in Hong Kong exhibited greater risks of more severe injury, intracranial haemorrhage, unfavourable GOS score at discharge from hospital, and post-concussion syndrome. Despite these findings, our results suggest that cycling is generally safe and hospitalised patients had a high (92.7%) likelihood of favourable functional outcomes on discharge from hospital.

Single-centre reviews of cycling-related injuries among various suburban districts in Hong Kong found that limb injuries were the most common form of trauma followed by head injury (10%-39% of patients).23 24 25 26 Among patients with TBI, 16% to 53% exhibited “severe” injury; however, the studies did not provide explicit definitions to qualify this categorisation, and did not describe radiological data regarding the extent of injury or the need for neurosurgical intervention.23 26 In the present study, 12% of hospitalised cyclists with head injury had moderate to severe TBI. There was also a high incidence of intracranial haemorrhage involving almost half of the patients. Both these factors were independent predictors of poor GOS score on discharge from hospital. Our results are consistent with the findings in a previous study where 75% of all cycling-related deaths were caused by severe TBI.27 The lack of an independent association with motor vehicle collisions, which constituted only a minority of injuries in this cohort, suggests that recreational cycling at comparatively low speeds can be fatal. Notably, the mechanism of injury for four (50%) of the eight recreational cyclists who died in this study was a loss of balance, followed by a fall on level ground without colliding into another object, person, or motor vehicle.

Previous studies in Hong Kong, the most recent of which was performed >10 years ago, revealed that recreational cyclists rarely wore protective headgear (eg, frequencies of 0.2% to 2.2% among emergency department attendees).24 25 26 Our findings revealed that significantly more patients (15%) wore helmets at the time of injury. Governmental advocacy initiatives for promoting helmet wearing in recent years may have resulted in heightened public awareness regarding the risks of head trauma.28 There is little doubt that helmets are protective. In the past 30 years, several case-control and epidemiological studies have delivered compelling evidence to support the efficacy of bicycle helmet wearing in reducing the risk of life-threatening TBI.29 30 31 32 33 34 35 36 37 In a case-control prospective multicentre study of over 3000 patients, Thompson et al33 noted that helmets (irrespective of design) conferred up to a 74% reduction in TBI during accidents. A subsequent meta-analysis of five studies observed that helmets provided a 63% to 88% reduction in the risk of head, brain, and severe TBI for all ages of cyclists; this included equal levels of protection for collisions involving motor vehicles and collisions due to other causes.38 In the present study, helmet wearing did not reduce TBI severity according to our broadly predefined categories. However, among hospitalised recreational cyclists with mild head injury, helmets did provide significant protection against intracranial haemorrhage, including potentially life-threatening epidural haematomas and ASDHs, as well as skull fractures. Thus, our findings may have important public health implications with regard to introducing mandatory bicycle helmet wearing legislation in the city.

Whether such laws should exist is a particularly divisive issue among public health experts and interest groups.39 40 41 42 43 In Australia, a nation with all-age helmet wearing safety laws, an overall 46% decline in cyclist fatalities per 1 000 000 population has been reported, compared with the pre-legislation period.44 Similar findings were noted in New Zealand: a 67% decline in severe TBI was recorded after the introduction of helmet laws.45 In the US, a significant reduction in paediatric cyclist fatalities involving motor vehicles was observed in states with such laws.46 A meta-analysis of the effectiveness of bicycle helmet legislation revealed that it increased helmet usage, while significantly reducing head injuries and mortality.47 Several medical associations have expressed support for introducing such legislation; these include the World Health Organization, the British Medical Association, the American Medical Association, and the Royal Australasian College of Surgeons.48 49 50 51 However, critics of such compulsory policies have hypothesised that helmets could encourage risk-compensation behaviour, whereby cyclists may be more willing to engage in potentially injurious risks or for motorists to exercise less caution when encountering them.39 52 53 Other reasons for opposition include infringement on individual liberties; some public health scholars have theorised that such laws could discourage cyclists from participating in gainful physical activity.39 40 41 42 54 From a Hong Kong Transport Department survey (5701 respondents) regarding attitudes towards possible helmet law and enforcement measures, the majority of respondents (78%-90%) were in favour of introducing such legislation, especially when riding on carriageways.16 However, among respondents who knew how to ride a bicycle (3933 respondents), 23% declared they would ride less frequently if mandatory helmet wearing was required.16

An inherent limitation of a retrospective study of this nature was the likely under-reporting of the number of patients with sports-related head injuries. In the only existing population-based study of TBI epidemiology that included community-based injuries, 95% were considered mild and 28% of respondents did not seek medical attention.55 Among professional or university-level athletes, under-reporting is more apparent: questionnaire surveys reveal that 31% to 78% of respondents neglected to pursue medical care despite experiencing a concussion during the preceding 12 months.56 57 At the emergency department level, no territory-wide TBI registry exists in Hong Kong; moreover, diagnostic coding to facilitate data retrieval is typically not performed after consultations. Therefore, we could only identify hospitalised patients with sports-related TBI by means of an administrative database that utilised the ICD-10 coding system. However, the validity of such administrative data for research has been questioned.58 Studies have shown significantly lower TBI rates among young adults, men, and patients with less severe injuries when the ICD system was utilised, compared with thorough medical record review.59 Analysis of a population-based TBI sample showed that only 19% of individuals were assigned a TBI-related diagnostic ICD code.60 In addition, a degree of selection bias may have existed because some non-hospitalised helmet-wearing cyclists with mild head injury may have been discharged from the emergency department, mitigating the protective effects of helmet use. This may explain why no considerable differences in outcomes were detected for patients with moderate or severely injured patients. Despite the low rate of helmet use among recreational cyclists (15%), significant protective effects were detected among mildly injured patients with regard to intracranial haemorrhage and skull fracture. This limited participant identification approach also allowed for a pragmatic review of patients with clinically significant TBI who were hospitalised following evaluation by an emergency care physician. Computed tomography scans are generally performed only for hospitalised patients with head injury in our public healthcare system; this approach offered an opportunity to evaluate imaging data for intracranial haemorrhage. Because the ICD coding system for traumatic intracranial haemorrhage reportedly has high sensitivity and specificity (both >80%),59 we adopted this coding outcome as the study’s primary endpoint. Another important limitation was the definition of mild TBI, which affected most patients in this study. The definitions offered by several authorities range from conventional GCS-based criteria such as the US Centers for Disease Control and Prevention,61 and the American College of Surgeons62 to additional symptoms of confusion, memory impairment, transient loss of consciousness, and irritability proposed by the World Health Organisation and the American Congress of Rehabilitation Medicine.18 63 64 A better delineation of these symptoms would have enhanced the identification of patients with “high-risk” mild TBI; however, because these relevant symptoms were often not systematically documented in most medical records retrieved in our study, we used GCS-based criteria to reduce the overall rate of underdiagnosis. Using GOS score on discharge from hospital as a secondary study endpoint, we found that only 31 (7%) patients had unfavourable outcomes. Although statistically significant predictors for TBI were identified, the wide confidence intervals for these predictors suggest that the sample size was insufficient to draw robust conclusions. Finally, we could only retrospectively assess GOS score as a fundamental measure of functional outcome. More sensitive instruments (eg, the extended GOS or the Sport Concussion Assessment Tool65 66) might have been better for assessing the psychosocial and cognitive aspects of TBI, considering that a large proportion of mildly injured cyclists had intracranial haemorrhage.

The incidence of sports-related TBI in Hong Kong is low and cycling is the most frequently associated activity. Almost half of hospitalised recreational cyclists sustained intracranial haemorrhage. Compared with patients who had head injury because of other sports, cyclists are more likely to experience severe consequences. There is evidence that helmet use offers protection against intracranial haemorrhage and skull fracture among cyclists with mild head injury. Cycling is a safe physical activity, but further legislative measures should be introduced to promote and protect the welfare of individuals enjoying this sport.

Concept or design: PYM Woo, E Cheung.
Acquisition of data: PYM Woo, E Cheung, FWY Lau, NWS Law, CKY Mak, P Tan, B Siu, A Wong.
Analysis or interpretation of data: PYM Woo, E Cheung, CKY Mak.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: All authors.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

This research has not been presented or published in any form prior to submission.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This study was approved by the Kowloon Central Cluster/Kowloon East Cluster research ethics committee (Ref KCC/KEC-2020-0331). All patients were treated in accordance with the Declaration of Helsinki. Informed consent was obtained from either the patient, next-of-kin, or their legal guardian.

1. Allender S, Cowburn G, Foster C. Understanding participation in sport and physical activity among children and adults: a review of qualitative studies. Health Educ Res 2006;21:826-35. Crossref

2. Warburton DE, Nicol CW, Bredin SS. Health benefits of physical activity: the evidence. CMAJ 2006;174:801-9. Crossref

3. Batty D, Thune I. Does physical activity prevent cancer? Evidence suggests protection against colon cancer and probably breast cancer. BMJ 2000;321:1424-5. Crossref

4. Batty GD, Lee IM. Physical activity and coronary heart disease. BMJ 2004;328:1089-90. Crossref

5. Aune D, Norat T, Leitzmann M, Tonstad S, Vatten LJ. Physical activity and the risk of type 2 diabetes: a systematic review and dose–response meta-analysis. Eur J Epidemiol 2015;30:529-42. Crossref

6. Liu X, Zhang D, Liu Y, et al. Dose-response association between physical activity and incident hypertension: a systematic review and meta-analysis of cohort studies. Hypertension 2017;69:813-20. Crossref

7. Malm C, Jakobsson J, Isaksson A. Physical activity and sports-real health benefits: a review with insight into the Public Health of Sweden. Sports (Basel) 2019;7:127. Crossref

8. Puckett M, Neri A, Underwood JM, Stewart SL. Nutrition and physical activity strategies for cancer prevention in current national comprehensive cancer control program plans. J Community Health 2016;41:1013-20. Crossref

9. Department of Health, Hong Kong SAR Government. Towards 2025: strategy and action plan to prevent and control non-communicable diseases in Hong Kong. 2018 Available from: https://www.chp.gov.hk/files/pdf/saptowards2025_fullreport_en.pdf. Accessed 2 Aug 2020.

10. Theadom A, Starkey NJ, Dowell T, et al. Sports-related brain injury in the general population: an epidemiological study. J Sci Med Sport 2014;17:591-6. Crossref

11. Manley G, Gardner AJ, Schneider KJ, et al. A systematic review of potential long-term effects of sport-related concussion. Br J Sports Med 2017;51:969-77. Crossref

12. Theadom A, Mahon S, Hume P, et al. Incidence of sports-related traumatic brain injury of all severities: a systematic review. Neuroepidemiology 2020;54:192-9. Crossref

13. Daneshvar DH, Nowinski CJ, McKee AC, Cantu RC. The epidemiology of sport-related concussion. Clin Sports Med 2011;30:1-17. Crossref

14. Selassie AW, Wilson DA, Pickelsimer EE, Voronca DC, Williams NR, Edwards JC. Incidence of sport-related traumatic brain injury and risk factors of severity: a population-based epidemiologic study. Ann Epidemiol 2013;23:750-6. Crossref

15. Transport Department, Hong Kong SAR Government. Cycling study. 2004. Available from: https://www.td.gov.hk/filemanager/en/publication/cyclingstudy.pdf. Accessed 22 Aug 2020.

16. Transport Department. Hong Kong SAR Government. Travel Characteristics Survey 2011 Final Report. Available from: https://www.td.gov.hk/filemanager/en/content_4652/tcs2011_eng.pdf. Accessed 22 Aug 2020.

17. Xu P, Dong N, Wong SC, Huang H. Cyclists injured in traffic crashes in Hong Kong: a call for action. PLoS One 2019;14:e0220785. Crossref

18. Servadei F, Teasdale G, Merry G, Neurotraumatology Committee of the World Federation of Neurosurgical Societies. Defining acute mild head injury in adults: a proposal based on prognostic factors, diagnosis, and management. J Neurotrauma 2001;18:657-64. Crossref

19. Huang WY, Wong SH, Sit CH, et al. Results from the Hong Kong’s 2018 report card on physical activity for children and youth. J Exerc Sci Fit 2019;17:14-9. Crossref

20. Harris AW, Jones CA, Rowe BH, Voaklander DC. A population-based study of sport and recreation-related head injuries treated in a Canadian health region. J Sci Med Sport 2012;15:298-304. Crossref

21. Beck AJ, Kee J. The epidemiology of sports head injuries in a rural population. Br J Sports Med 2011;45:A20. Crossref

22. Transport Department, Hong Kong SAR Government. Road Traffic Accident Statistics. Year 2019. Road traffic accidents involving bicycle by location and severity 2019. Available from: https://www.td.gov.hk/en/road_safety/road_traffic_accident_statistics/2019/index.html. Accessed 9 Aug 2020.

23. Lee LL, Yeung KL, Chan JT, Chen RC. A profile of bicycle-related injuries in Tai Po. Hong Kong J Emerg Med 2003;10:81-7.Crossref

24. Yeung JH, Leung CS, Poon WS, Cheung NK, Graham CA, Rainer TH. Bicycle related injuries presenting to a trauma centre in Hong Kong. Injury 2009;40:555-9. Crossref

25. Ng CP, Siu AY, Chung CH. Bicycle-related injuries: a local scene. Hong Kong J Emerg Med 2001;8:78-83. Crossref

26. Sze NN, Tsui KL, Wong SC, So FL. Bicycle-related crashes in Hong Kong: is it possible to reduce mortality and severe injury in the metropolitan area? Hong Kong J Emerg Med 2011;18:136-43. Crossref

27. Rivara FP, Thompson DC, Thompson RS. Epidemiology of bicycle injuries and risk factors for serious injury. Inj Prev 1997;3:110-4. Crossref

28. Transport Department, Hong Kong SAR Government. Be a smart cyclist. Wear bicycle helmet always. 2020. Available from: https://www.td.gov.hk/filemanager/en/content_4551/221201047_leaflet_a.pdf. Accessed 21 Aug 2020.

29. Attewell RG, Glase K, McFadden M. Bicycle helmet efficacy: a meta-analysis. Accid Anal Prev 2001;33:345-52. Crossref

30. Maimaris C, Summers CL, Browning C, Palmer CR. Injury patterns in cyclists attending an accident and emergency department: a comparison of helmet wearers and non-wearers. BMJ 1994;308:1537-40. Crossref

31. McDermott FT, Lane JC, Brazenor GA, Debney EA. The effectiveness of bicyclist helmets: a study of 1710 casualties. J Trauma 1993;34:834-44. Crossref

32. Thomas S, Acton C, Nixon J, Battistutta D, Pitt WR, Clark R. Effectiveness of bicycle helmets in preventing head injury in children: case-control study. BMJ 1994;308:173-6. Crossref

33. Thompson DC, Rivara FP, Thompson RS. Effectiveness of bicycle safety helmets in preventing head injuries. A case-control study. JAMA 1996;276:1968-73. Crossref

34. Thompson RS, Rivara FP, Thompson DC. A case-control study of the effectiveness of bicycle safety helmets. N Engl J Med 1989;320:1361-7. Crossref

35. Strotmeyer SJ, Behr C, Fabio A, Gaines BA. Bike helmets prevent pediatric head injury in serious bicycle crashes with motor vehicles. Inj Epidemiol 2020;7(Suppl 1):24. Crossref

36. Sethi M, Heidenberg J, Wall SP, et al. Bicycle helmets are highly protective against traumatic brain injury within a dense urban setting. Injury 2015;46:2483-90. Crossref

37. Persaud N, Coleman E, Zwolakowski D, Lauwers B, Cass D. Nonuse of bicycle helmets and risk of fatal head injury: a proportional mortality, case-control study. CMAJ 2012;184:E921-3. Crossref

38. Thompson DC, Rivara FP, Thompson R. Helmets for preventing head and facial injuries in bicyclists. Cochrane Database Syst Rev 2000;1999(2):CD001855. Crossref

39. Robinson DL. No clear evidence from countries that have enforced the wearing of helmets. BMJ 2006;332:722-5. Crossref

40. Hong Kong Cycling Alliance. Helmets. Available from: http://hkcyclingalliance.org/on-the-road/helmets. Accessed 21 Aug 2020.

41. Bateman-House A. Bikes, helmets, and public health: decision-making when goods collide. Am J Public Health 2014;104:986-92. Crossref

42. Goldacre B, Spiegelhalter D. Bicycle helmets and the law. BMJ 2013;346:f3817. Crossref

43. Rivara FP, Thompson DC, Patterson MQ, Thompson RS. Prevention of bicycle-related injuries: helmets, education, and legislation. Annu Rev Public Health 1998;19:293-318. Crossref

44. Olivier J, Boufous S, Grzebieta R. The impact of bicycle helmet legislation on cycling fatalities in Australia. Int J Epidemiol 2019;48:1197-203. Crossref

45. Tin Tin S, Woodward A, Ameratunga S. Injuries to pedal cyclists on New Zealand roads, 1988-2007. BMC Public Health 2010;10:655. Crossref

46. Meehan WP 3rd, Lee LK, Fischer CM, Mannix RC. Bicycle helmet laws are associated with a lower fatality rate from bicycle-motor vehicle collisions. J Pediatr 2013;163:726-9. Crossref

47. Macpherson A, Spinks A. Bicycle helmet legislation for the uptake of helmet use and prevention of head injuries. Cochrane Database Syst Rev 2008;2008(3):CD005401. Crossref

48. World Health Organisation. Why are helmets needed? 2006. Available from: https://www.who.int/roadsafety/projects/manuals/helmet_manual/1-Why.pdf. Accessed 13 Oct 2021.

49. British Medical Association. Healthy transport=healthy lives. 2012. Available from: https://thepep.unece.org/sites/default/files/2017-06/Healthy%20transport%20healthy%20lives%20Britsh%20Medical%20Association.pdf. Accessed 13 Oct 2021.

50. American Medical Association. Helmets for riders of motorized and non-motorized cycles H-10.964. Available from: https://policysearch.ama-assn.org/policyfinder/detail/bicycle helmets?uri=%2FAMADoc%2FHOD.xml-0-3.xml. Accessed 20 Sep 2021.

51. McDermott FT. Helmet efficacy in the prevention of bicyclist head injuries: Royal Australasian College of Surgeons initiatives in the introduction of compulsory safety helmet wearing in Victoria, Australia. World J Surg 1992;16:379-83. Crossref

52. Gamble T, Walker I. Wearing a bicycle helmet can increase risk taking and sensation seeking in adults. Psychol Sci 2016;27:289-94. Crossref

53. Walker I. Drivers overtaking bicyclists: objective data on the effects of riding position, helmet use, vehicle type and apparent gender. Accid Anal Prev 2007;39:417-25. Crossref

54. Hooper C, Spicer J. Liberty or death; don’t tread on me. J Med Ethics 2012;38:338-41. Crossref

55. Theadom A, Barker-Collo S, Feigin VL, et al. The spectrum captured: a methodological approach to studying incidence and outcomes of traumatic brain injury on a population level. Neuroepidemiology 2012;38:18-29. Crossref

56. Delaney JS, Lamfookon C, Bloom GA, Al-Kashmiri A, Correa JA. Why university athletes choose not to reveal their concussion symptoms during a practice or game. Clin J Sport Med 2015;25:113-25. Crossref

57. Meehan WP 3rd, Mannix RC, O’Brien MJ, Collins MW. The prevalence of undiagnosed concussions in athletes. Clin J Sport Med 2013;23:339-42. Crossref

58. van Walraven C, Bennett C, Forster AJ. Administrative database research infrequently used validated diagnostic or procedural codes. J Clin Epidemiol 2011;64:1054-9. Crossref

59. Carroll CP, Cochran JA, Guse CE, Wang MC. Are we underestimating the burden of traumatic brain injury? Surveillance of severe traumatic brain injury using Centers for Disease Control International Classification of Disease, Ninth Revision, Clinical Modification, traumatic brain injury codes. Neurosurgery 2012;71:1064-70. Crossref

60. Barker-Collo S, Theadom A, Jones K, Feigin VL, Kahan M. Accuracy of an International Classification of Diseases Code Surveillance System in the identification of traumatic brain injury. Neuroepidemiology 2016;47:46-52. Crossref

61. Centers for Disease Control and Prevention. Glasgow Coma Scale. Available from: https://www.cdc.gov/masstrauma/resources/gcs.pdf. Accessed 20 Sep 2021.

62. American College of Surgeons. Best practices in the management of traumatic brain injury. Available from: https://www.facs.org/-/media/files/quality-programs/trauma/tqip/tbi_guidelines.ashx. Accessed 20 Sep 2021.

63. Carroll LJ, Cassidy JD, Holm L, Kraus J, Coronado VG, WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury. Methodological issues and research recommendations for mild traumatic brain injury: the WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury. J Rehabil Med 2004;(43 Suppl):113-25. Crossref

64. Kay T, Harrington DE, Adams R, et al. Definition of mild traumatic brain injury. J Head Trauma Rehabil 1993;8:86-7. Crossref

65. Kean J, Malec JF. Towards a better measure of brain injury outcome: new measures or a new metric? Arch Phys Med Rehabil 2014;95:1225-8. Crossref

66. McCrory P, Meeuwisse WH, Aubry M, et al. Consensus statement on concussion in sport: the 4th International Conference on Concussion in Sport, Zurich, November 2012. J Athl Train 2013;48:554-75. Crossref

Klebsiella pneumonia causes various clinically important infections. In 2017, K pneumoniae was the third most common isolate in intensive care units (ICUs) and second most common isolate in all patients in the Hong Kong East Cluster.1

The emergence of multidrug-resistant K pneumoniae infections is an increasing concern.2 There have been outbreaks of extended-spectrum beta-lactamase (ESBL)–producing strains, carbapenem-resistant (CR) and carbapenemase-producing (CP) strains, and hypervirulent K pneumoniae infections both in Hong Kong and worldwide.2 3 4 In Greece and Italy, CP K pneumoniae comprises 68.3% of all K pneumoniae strains.5

The importance of appropriate early empirical antibiotics has been repeatedly emphasised in the management of septic shock by the Surviving Sepsis Campaign 2016.6 Previously, we highlighted the importance of appropriate early antibiotics for successful patient outcomes.4 To the best of our knowledge, risk factors for receiving inappropriate empirical antibiotics have not yet been explored. Therefore, in the present study, we aimed to evaluate the impact of appropriate empirical antibiotics on outcomes in patients with K pneumoniae bacteraemia; we also performed subgroup analysis on ICU patients with K pneumoniae bacteraemia.

We conducted a retrospective analysis of adult patients with K pneumoniae bacteraemia who were admitted to Pamela Youde Nethersole Eastern Hospital within the period from 1 January 2009 to 30 June 2017. Pamela Youde Nethersole Eastern Hospital is a 1700-bed hospital in Hong Kong which provides extensive services except cardiothoracic surgery, transplant surgery, and burns. Patients were excluded if they were aged <18 years or had incomplete information. Patient medical records were reviewed, as were data in clinical management and clinical information systems (IntelliVue Clinical Information Portfolio; Philips Medical, Amsterdam, The Netherlands). The clinical management system is a database that stores patients’ demographics, laboratory results, and drug administration records for all public hospitals in Hong Kong. In the event of missing data, medical records were reviewed manually. Baseline demographics, clinical characteristics, and microbiological data for all included patients were identified from the above databases and medical records.

Disease severity was quantified using the maximum Sequential Organ Failure Assessment (SOFA) score.7 The following clinical outcome data were investigated: use of invasive organ supports (eg, inotropic use, mechanical ventilation, and renal replacement therapies), ICU and hospital lengths of stay (LOSs), ICU ventilator duration, and mortality. The primary outcome was 90-day all-cause mortality; secondary outcomes were ICU and hospital mortalities, ICU and hospital LOSs, and ICU ventilator duration.

Klebsiella pneumoniae bacteraemia was defined as the growth of K pneumoniae in one or more blood cultures. If more than one positive blood culture result was recorded, only the first sample was included. Empirical antibiotic treatment was defined as the antibiotic used within 24 hours after a culture sample was collected. The empirical antibiotic treatment was considered appropriate if at least one of the antibiotic agents was consistent with the in vitro susceptibility results.8 9 10 Community-acquired infection was defined as K pneumoniae identified in patients upon admission or within 48 hours after admission; hospital-acquired infection was defined as K pneumoniae identified in patients at >48 hours after admission.11 Hepatobiliary sepsis comprised liver abscess, cholangitis, and cholecystitis; gastrointestinal sepsis comprised spontaneous bacterial peritonitis, peritonitis caused by bowel perforation, and intra-abdominal abscesses (excluding liver abscess). Medical co-morbidities (eg, diabetes mellitus, cirrhosis, congestive heart failure, chronic renal impairment, haematological malignancy and solid tumour) were defined in accordance with the International Classification of Disease coding. Prior steroid use was defined as oral or intravenous steroid consumption within 30 days before the index positive blood culture result. Chemotherapy use was defined as oral or intravenous infusion of biological agents or chemotherapy administered within 30 days before the index positive blood culture result. Any antibiotics usage within 30 days before the index hospital admission was regarded as prior antibiotics usage.

Blood cultures were incubated and processed. Blood culture results were considered negative if no positive growth occurred after 5 to 7 days. Susceptibility interpretation was based on Clinical and Laboratory Standards Institute interpretive criteria. The ESBL testing was based on Clinical and Laboratory Standards Institute testing criteria12 13 14 15 16 17 18 or the double-disk synergy test described by Jarlier et al.19 Regarding CR K pneumoniae, the isolates were sent to the Public Health Laboratory Centre of Hong Kong if further genetic testing was required to confirm carbapenemase production. Multiplex real-time polymerase chain reaction assays were performed to detect Classes A, B, and D carbapenemase gene targets.

We compared the characteristics and clinical parameters between patients treated with appropriate and inappropriate empirical antibiotics, as well as between 90-day survivors and non-survivors. Results are expressed as the median ± interquartile range (IQR) or as the number (percentage) of patients, as appropriate.

Categorical variables were compared by the Pearson Chi squared test or Fisher’s exact test, as appropriate for univariate analysis. Continuous variables were compared by using the Mann–Whitney U test. Variables with P<0.2 in univariate analysis or with clinical significance from previous studies were included in the multivariate analysis. Independent predictors for 90-day mortality were assessed by Cox regression analysis. Logistic regression analysis was used to assess independent predictors for receiving appropriate and inappropriate empirical antibiotics. Post hoc analysis was performed for patients with chronic renal failure and resistant organisms.

The Statistical Package for Social Sciences (Windows version 24.0; IBM Corp, Armonk [NY], US) was used to perform statistical analyses.

During the 8.5-year study period, we identified 984 patients with K pneumoniae bacteraemia; of these, 686 (69.7%) and 298 (30.3%) received appropriate and inappropriate empirical antibiotics, respectively. Table 1 shows the baseline demographics of patients who received appropriate and inappropriate empirical antibiotics. The median patient age was 75 years (IQR=63-83 years). The most common types of infection were hepatobiliary tract infection (33.1%), urosepsis (24.4%), and respiratory tract infection (18.4%). The overall 90-day mortality was 32.7%, hospital mortality was 22.5%, and median hospital LOS was 10.68 days (5.38-22.81 days, P<0.001).

Univariate analysis (Table 1) revealed that the risk factors for receiving inappropriate empirical antibiotics were age >65 years (P=0.044), chronic renal impairment (P<0.001), respiratory tract infection (P=0.002), mechanical ventilation (P=0.001), CR or ESBL-producing isolates (both P<0.001), and higher total SOFA score (P=0.048). Hepatobiliary sepsis was associated with a higher rate of appropriate empirical antibiotic treatment (P=0.009).

Table 2 demonstrates the logistic regression analysis of the predictors for the appropriateness of empirical antibiotics for all patients and ICU subgroup. These include older patients (P=0.010), chronic kidney disease (P=0.007), mechanical ventilation (P=0.005), respiratory tract infection (P=0.034), and either carbapenem resistance or ESBL production (P<0.001). Table 1 shows that the 90-day and hospital mortalities were significantly higher in patients with inappropriate empirical treatment (both P<0.001). Moreover, the hospital LOS was shorter in patients who received inappropriate empirical antibiotics (P<0.001).

Receipt of inappropriate empirical antibiotics was associated with higher hospital mortality; this finding was consistent in ICU subgroup (56% vs 23%). The absolute risk increases in mortality associated with the receipt of inappropriate empirical antibiotics were 18.3% and 33% in ICU subgroup. The number of inappropriate empirical antibiotics associated with each mortality was five in the ICU subgroup and three in all patients.

Antibiograms showing the proportions of non-susceptible K pneumoniae isolates are described in the Online Supplementary Table 1. Twenty (2.0%) patients had CR K pneumoniae bacteraemia, but molecular tests in the Public Health Laboratory Centre of Hong Kong revealed that none of them had CP strains. Overall, 113 (11.5%) patients had ESBL-producing infections.

Post hoc analysis revealed that patients with chronic renal failure were more likely to have ESBL infections (18.3% vs 10.5%; P=0.011) and CR infections (0.056% vs 0.015%; P=0.003).

The 90-day all-cause mortalities were 32.7% in all patients with K pneumoniae bacteraemia and 34.6% in the ICU subgroup. Univariate analysis (Table 3) showed that 90-day non-survivors were more likely to be aged >65 years (P<0.001), admitted through the Department of Medicine (P<0.001), have septic shock (P=0.005), have a higher total SOFA score (P<0.001), receive inappropriate or no empirical antibiotics (P<0.001 and P<0.001, respectively), have solid tumour (P<0.001), have respiratory tract infection (P<0.001), be mechanically ventilated (P<0.001), have gastrointestinal infections (P<0.001), and require renal replacement therapy (P=0.044). Patients with diabetes (P=0.001), hepatobiliary sepsis (P<0.001), and urosepsis (P<0.001) had lower 90-day mortalities.

Table 4 shows the Cox regression analysis of predictors for 90-day mortality. Independent predictors for increased 90-day mortality were respiratory tract infection (P<0.001), gastrointestinal infection (P<0.001), inappropriate empirical antibiotics (P<0.001), older age (P<0.001), solid tumour (P<0.001), patients admitted through the Department of Medicine (P=0.012), and higher total SOFA score (P<0.001). Patients with diabetes had lower 90-day mortality (P=0.001). The Figure shows the Kaplan–Meier survival plot and according to log rank analysis, the results demonstrated a statistically significant improvement in survival among patients who received appropriate empirical antibiotics (P<0.001).

Online Supplementary Table 2 shows the demographics for inappropriate empirical antibiotics by logistic regression analysis in ICU subgroup. Overall, 205 (20.8%) patients required intensive care; among them, 148 (72.2%) received appropriate empirical antibiotics, while 57 (27.8%) received inappropriate or no empirical antibiotics. The median patient age was 68 years (IQR=58-78 years). The commonest infection was hepatobiliary infections (42.9%), followed by respiratory tract (23.4%) and urosepsis (14.6%). Furthermore, 82.4% of the patients had septic shock, 33.2% received renal replacement therapy, 60% received mechanical ventilation, and 78.5% had vasopressor use. The ICU and overall 90-day mortalities were 18.5% and 34.6%, respectively. The receipt of inappropriate empirical antibiotics was significantly associated with higher 90-day mortality (59.6% vs 25.0%; P<0.001), higher ICU mortality (35.1% vs 12.2%; P<0.001), higher hospital mortality (56.1% vs 23.0%; P<0.001), and longer ventilator duration (2 d vs 1 d, P=0.026).

Cox regression analysis showed that the receipt of inappropriate or no empirical antibiotics (P<0.001; Table 4) was the strongest independent predictor of 90-day mortality in critically ill patients with K pneumoniae bacteraemia. Other independent predictors were congestive heart failure (P=0.02), admitted through the Department of Medicine (P=0.016), and a higher total SOFA score (P<0.001). Ninety-day non-survivors had longer hospital LOS (P<0.001).

Among all patients in this study, 686 (69.7%) received appropriate empirical antibiotics. Furthermore, 148 (72.2%) critically ill patients received appropriate empirical antibiotics. Importantly, 93 (9.5%) patients did not receive any empirical antibiotics. The median hospital LOS for these 93 patients was significantly shorter than the LOS for all patients in the study. We performed event-free survival analysis and found that the LOSs were similar in both groups; these findings suggested that patients who received inappropriate antibiotics had more severe disease and earlier death, leading to a shorter hospital LOS.

Micek et al20 found that prior antibiotic exposure was a risk factor for inappropriate empirical therapy. Lautenbach et al21 described a positive correlation between the total cumulative dose of antibiotics and ESBL K pneumoniae infection. In our cohort, prior antibiotics exposure was not significantly correlated with inappropriate empirical antibiotic treatment; furthermore, prior antibiotics exposure was not associated with ESBL infections. We examined the presence of prior antibiotics use 30 days prior to positive blood culture results, but information regarding the total cumulative antibiotics exposure in terms of dosing and duration were not available. Other information was unavailable regarding antibiotics prescribed outside hospital settings.

Patients with chronic renal failure are reportedly more prone to developing resistant infections.22 They were at greater risk of receiving inappropriate antibiotics. Additionally, hospital-acquired infection has been associated with a higher rate of inappropriate empirical antibiotic treatment and higher 90-day mortality.11

The rates of CR and CP K pneumoniae bacteraemia were much lower than the rates reported in other endemic countries.5 A study from Shanghai reported that approximately 22% of all patients with K pneumoniae bacteraemia had CR infections; moreover, approximately 59% of the isolates were CR infections in patients who required intensive care.23

The prevalence of ESBL K pneumoniae bacteraemia in Hong Kong is low.4 The ESBL infection rate in our cohort was 11.5%, similar to previous local studies.1 24 Another study in China demonstrated a much higher rate of ESBL infections (approximately 39%).25 Both ESBL and CR infections were not associated with increased mortality in our study in either the overall patient population or the ICU subgroup. However, ESBL and CR infections were significantly associated with longer hospital LOS.

The antibiotics sensitivity in our cohorts was comparable with the antibiogram data in the IMPACT guideline.1 Ampicillin-sulbactam or amoxicillin-clavulanate is recommended as the drug of choice for treatment of K pneumomiae infections in local guidelines.1 In our cohort, more than 80% of K pneumomiae isolates were susceptible to amoxicillin-clavulanate, indicating that it is a reasonable option for broad empirical coverage. Add-on therapy with aminoglycoside improves the coverage of this regimen, because more than 90% of the isolates in our cohort were sensitive to either gentamicin or amikacin. While combination therapy improves the chance of successful empirical therapy, routine use of combination therapy remains controversial.6

The 90-day all-cause mortalities in our study were comparable with the findings in previous studies.24 26 Respiratory tract infections and gastrointestinal infections have consistently been associated with a worse outcome and greater mortality.27 28 In contrast, urosepsis and hepatobiliary sepsis have repeatedly associated with better survival outcomes.29 These sources of infections may be amendable to percutaneous, endoscopic, or surgical drainage, allowing more rapid and definitive control of sepsis, which leads to better survival.28

A greater proportion of patients with respiratory tract infection did not receive any empirical antibiotics, which might explain the worse outcomes in these patients. The symptoms and signs of respiratory tract infection may overlap with other diseases (eg, heart failure) and treatment may be delayed while waiting for laboratory results and imaging. Given the greater proportion of patients with respiratory tract infections who did not receive any empirical antibiotics, there is a need for early consideration of empirical antibiotics in patients with signs and symptoms of respiratory tract infections.

Consistent with the findings of previous studies, we found that older age, solid tumour, and admission through the Department of Medicine were factors associated with higher 90-day mortality.30 31 Patients with these factors were more likely to have other pre-existing co-morbidities, worse premorbid functional status, and be institutionalised. They may also have received a more conservative approach to treatment overall.

Diabetes mellitus is well-known to predispose patients to infections, such that affected patients are reportedly 4.4-fold more likely to develop bloodstream infection.32 Similar to the findings by Peralta et al,32 we did not find increased mortality among patients with diabetes. In our cohort, diabetic patients were more likely to had urosepsis than respiratory tract infections. Greater frequency of urosepsis and smaller frequency of respiratory tract infection may have an overall positive effect on survival. Moreover, advances in diabetes care in recent decades (eg, newer generations of medication and integrated multidisciplinary care) have led to neutral effects of diabetes on short-term mortality in patients undergoing major operations and patients with sepsis.34 34 35 Glycaemic status, haemoglobin A1c levels, and diabetes severity were not available in the present study; thus, we could not delineate how diabetes control affected bacteraemia outcomes. Our results are limited to demonstrating that 90-day mortality and diabetes have a greater impact on long-term survival, although this conclusion may not be apparent in the current study.33

In addition to mortality, we demonstrated that the inappropriate use of empirical antibiotics was associated with longer ICU ventilator duration, which leads to greater costs and more extensive use of ICU resources. A large retrospective cohort from the US regarding Enterobacteriaceae infections showed that each additional day without appropriate antibiotics was associated with an increased hospital expenditure of US$750 and an increased risk of 30-day readmission.36

To the best of our knowledge, the appropriateness of empirical antibiotics has consistently been identified as one of the strongest independent predictors of 90-day mortality in all affected patients and in critically ill patients.37 38 39 Kumar et al8 demonstrated a fivefold increase in mortality among patients with sepsis who received inappropriate initial antibiotics. In this study, we demonstrated a twofold increase in mortality in all patients with K pneumoniae bacteraemia, as well as a threefold increase in mortality in critically ill patients with K pneumoniae bacteraemia. Furthermore, receipt of inappropriate initial antibiotics was the strongest independent predictor of 90-day mortality in the ICU subgroup. Zilberberg et al40 found that the detrimental effect of inappropriate empirical antibiotic treatment could not be corrected despite subsequent targeted antibiotic treatments. A meta-analysis revealed a slower rate of bacterial clearance and increased treatment failure rate when patients were administered inappropriate empirical antibiotic therapy.41 Appropriate early antibiotics allow rapid reduction of bacterial load and modulate host defences, thus alleviating some organ dysfunction.42 A more pragmatic approach would be the early administration of broader-spectrum empirical antibiotics and timely de-escalation, according to bacterial sensitivity and the patient’s clinical progression.

The chain of sepsis management begins during the first encounter in the Emergency Department and in general hospital wards, rather than in the ICU. Early administration of antibiotics within 3 hours of hospital admission and aggressive sepsis care (beginning in the Emergency Department) can improve survival.42 Another prospective observational study demonstrated that the adequate use of empirical antibiotics prior to ICU admission was the strongest independent factor associated with survival.29 Extensive efforts are needed to facilitate early, appropriate use of empirical antibiotics, including the use of a current antibiogram, implementation of multidisciplinary sepsis management guidelines, and establishment of protocols among pharmacists, microbiologists, clinicians, and nurses.

To the best of our knowledge, this is the largest study of K pneumoniae bacteraemia in Hong Kong and in the Asia-Pacific region. By including both ICU and general ward patients, we achieved clarity regarding the diverse characteristics of K pneumoniae bacteraemia. We also identified many potential predictors of K pneumoniae bacteraemia–related mortality, based on our extensive literature review and previous publications, then tested these predictors using real-world patient data. By evaluating 90-day mortality, hospital LOS, and ventilator duration, our study more comprehensively evaluated immediate and longer-term complications of bacteraemia; it also provided information for future studies of cost-effectiveness in terms of empirical antibiotics and resource utilisation. Finally, we used the maximum total SOFA score for severity assessment. This score has been repeatedly validated in determining disease severity and predicting mortality in critically ill patients.43 44 45

This retrospective study was subject to potential confounding factors, including selection bias that could not be completely eliminated from the analysis. Notably, the results of this single-centre study may not be generalisable to other countries with higher CR or CP K pneumoniae infections. Furthermore, this study encompassed a long duration, in which the definitions of sepsis or septic shock might have changed.45 The care of patients with sepsis evolved over time, including advances in source control by percutaneous and endoscopic means that potentially improved patient survival. Nonetheless, the role of empirical antibiotics in patients with sepsis remains an essential sepsis consideration.

Antibiotic pharmacodynamics also has a fundamental role in bacteraemia treatment. In this study, we could not collect information regarding the timing of first-dose antibiotics, time to appropriate antibiotics, duration of antibiotics, or time to surgical treatments. Moreover, antibiotic stewardship and therapeutic de-escalation efforts, as well as their impacts on patient outcomes, were not assessed in the present study. Future studies may be needed concerning the prolonged infusion of beta-lactam antibiotics, use of combination therapies, duration of antibiotics, and serum monitoring of antibiotics.

The receipt of inappropriate empirical antibiotics led to twofold greater 90-day mortality in patients with K pneumoniae bacteraemia. In critically ill patients, inappropriate use of empirical antibiotics was the strongest independent predictor of mortality. Early identification of high-risk patients and administration of appropriate empirical antibiotics can improve patient outcomes.

Concept or design: MY Man, HP Shum.
Acquisition of data: MY Man, HP Shum.
Analysis or interpretation of data: MY Man, HP Shum.
Drafting of the manuscript: MY Man.
Critical revision of the manuscript for important intellectual content: WW Yan.

All authors read and approved the final manuscript. All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

This research was presented by KC Li as an abstract at the 31st Annual Congress of the European Society of Intensive Care Medicine (ESICM), 21-24 October 2018, Paris, France.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This study was approved by the Hong Kong Easter Cluster Ethics Committee of the Hospital Authority (HKECREC-2018-018). The requirement for written informed consent was waived because of the retrospective nature of the study.

1. Ho PL, Wu TC, Chao DV, et al, editors. Reducing bacterial resistance with IMPACT. Interhospital multi-disciplinary programme on antimicrobial chemotherapy. 5th edition. 2017. Available from: https://www.chp.gov.hk/files/pdf/reducing_bacterial_resistance_with_impact.pdf. Accessed 7 Jun 2020.

2. Ben-David D, Kordevani R, Keller N, et al. Outcome of carbapenem resistant Klebsiella pneumoniae bloodstream infections. Clin Microbiol Infect 2012;18:54-60. Crossref

3. Wong MH, Shum HP, Chen JH, et al. Emergence of carbapenem-resistant hypervirulent Klebsiella pneumoniae. Lancet Infect Dis 2018;18:24. Crossref

4. Man MY, Shum HP, Chan YH, et al. Clinical predictors and outcomes of Klebsiella pneumoniae bacteraemia in a regional hospital in Hong Kong. J Hosp Infect 2017;97:35-41.Crossref

5. Alicino C, Giacobbe DR, Orsi A, et al. Trends in the annual incidence of carbapenem-resistant Klebsiella pneumoniae bloodstream infections: a 8-year retrospective study in a large teaching hospital in northern Italy. BMC Infect Dis 2015;15:415. Crossref

6. Rhodes A, Evans LE, Alhazzani W, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock: 2016. Intensive Care Med 2017;43:304-77.Crossref

7. Ferreira FL, Bota DP, Bross A, Mélot C, Vincent JL. Serial evaluation of the SOFA score to predict outcome in critically ill patients. JAMA 2001;286:1754-8. Crossref

8. Kumar A, Ellis P, Arabi Y, et al. Initiation of inappropriate antimicrobial therapy results in a fivefold reduction of survival in human septic shock. Chest 2009;136:1237-48. Crossref

9. Vazquez-Guillamet C, Scolari M, Zilberberg MD, Shorr AF, Micek ST, Kollef M. Using the number needed to treat to assess appropriate antimicrobial therapy as a determinant of outcome in severe sepsis and septic shock. Crit Care Med 2014;42:2342-9. Crossref

10. Micek ST, Welch EC, Khan J, et al. Empiric combination antibiotic therapy is associated with improved outcome against sepsis due to Gram-negative bacteria: a retrospective analysis. Antimicrob Agents Chemother 2010;54:1742-8. Crossref

11. Juan CH, Chuang C, Chen CH, Li L, Lin YT. Clinical characteristics, antimicrobial resistance and capsular types of community-acquired, healthcare-associated, and nosocomial Klebsiella pneumoniae bacteremia. Antimicrob Resist Infect Control 2019;8:1. Crossref

12. Clinical and Laboratory Standards Institute. M100-S16 Performance Standards for Antimicrobial Susceptibility Testing; Sixteenth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2006.

13. Clinical and Laboratory Standards Institute. M100-S17 Performance Standards for antimicrobial susceptibility testing; Seventeenth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2007.

14. Clinical and Laboratory Standards Institute. M100-S19 Performance Standards for antimicrobial susceptibility testing; Nineteenth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2009.

15. Clinical and Laboratory Standards Institute. M100-S20 Performance Standards for Antimicrobial Susceptibility Testing; Twentieth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2010.

16. Clinical and Laboratory Standards Institute. M100-S23 Performance Standards for Antimicrobial Susceptibility Testing; Twenty-Third Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2013.

17. Clinical and Laboratory Standards Institute. M100-S25 Performance Standards for Antimicrobial Susceptibility Testing; Twenty-Fifth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2015.

18. Clinical and Laboratory Standards Institute. M100-S26 Performance Standards for Antimicrobial Susceptibility Testing; Twenty-Sixth Informational Supplement. Clinical and Laboratory Standards Institute, Wayne, PA; January 2016.

19. Jarlier V, Nicolas MH, Fournier G, Philippon A. Extended broad-spectrum beta-lactamases conferring transferable resistance to newer beta-lactam agents in Enterobacteriaceae: hospital prevalence and susceptibility patterns. Rev Infect Dis 1988;10:867-78. Crossref

20. Micek ST, Hampton N, Kollef M. Risk factors and outcomes for ineffective empiric treatment of sepsis caused by Gram-negative pathogens: stratification by onset of infection. Antimicrob Agents Chemother 2017;62:e01577-17.

21. Lautenbach E, Patel JB, Bilker WB, Edelstein PH, Fishman NO. Extended-spectrum beta-lactamase-producing Escherichia coli and Klebsiella pneumoniae: risk factors for infection and impact of resistance on outcomes. Clin Infect Dis 2001;32:1162-71. Crossref

22. Eilertson B, Cober E, Richter SS, et al. Carbapenem-resistant Enterobacteriaceae infections in patients on renal replacement therapy. Open Forum Infect Dis 2017;4:ofx216.

23. Tian L, Tan R, Chen Y, et al. Epidemiology of Klebsiella pneumoniae bloodstream infections in a teaching hospital: factors related to the carbapenem resistance and patient mortality. Antimicrob Resist Infect Control 2016;5:48. Crossref

24. Pau CK, Ma FF, Ip M, You JH. Characteristics and outcomes of Klebsiella pneumoniae bacteraemia in Hong Kong. Infect Dis (Lond) 2015;47:283-8. Crossref

25. Li L, Huang H. Risk factors of mortality in bloodstream infections caused by Klebsiella pneumonia: a singlecenter retrospective study in China. Medicine (Baltimore) 2017;96:e7924.

26. Melot B, Colot J, Guerrier G. Bacteremic community-acquired infections due to Klebsiella pneumoniae: clinical and microbiological presentation in New Caledonia, 2008-2013. Int J Infect Dis 2015;41:29-31. Crossref

27. Russo A, Falcone M, Gutiérrez-Gutiérrez B, et al. Predictors of outcome in patients with severe sepsis or septic shock due to extended-spectrum beta-lactamase-producing Enterobacteriaceae. Int J Antimicrob Agents 2018;52:577-85.Crossref

28. Zilberberg MD, Shorr AF, Micek ST, Vazquez-Guillamet C, Kollef MH. Multi-drug resistance, inappropriate initial antibiotic therapy and mortality in Gram-negative severe sepsis and septic shock: a retrospective cohort study. Crit Care 2014;18:596. Crossref

29. Garnacho-Montero J, Gutiérrez-Pizarraya A, Escoresca-Ortega A, Fernández-Delgado E, López-Sánchez JM. Adequate antibiotic therapy prior to ICU admission in patients with severe sepsis and septic shock reduces hospital mortality. Crit Care 2015;19:302. Crossref

30. Antonio M, Gudiol C, Royo-Cebrecos C, Grillo S, Ardanuy C, Carratalà J. Current etiology, clinical features and outcomes of bacteremia in older patients with solid tumors. J Geriatr Oncol 2019;10:246-51. Crossref

31. Tseng CP, Wu HS, Wu TH, Lin YT, Fung CP. Clinical characteristics and outcome of patients with community-onset Klebsiella pneumoniae bacteremia requiring intensive care. J Microbiol Immunol Infect 2013;46:217-23. Crossref

32. Peralta G, Sánchez MB, Roiz MP, Garrido JC, Teira R, Mateos F. Diabetes does not affect outcome in patients with Enterobacteriaceae bacteremia. BMC Infect Dis 2009;9:94. Crossref

33. Filsoufi F, Rahmanian PB, Castillo JG, Mechanick JI, Sharma SK, Adams DH. Diabetes is not a risk factor for hospital mortality following contemporary coronary artery bypass grafting. Interact Cardiovasc Thorac Surg 2007;6:753-8. Crossref

34. Teo BJ, Chong HC, Yeo W, Tan AH. The impact of diabetes on patient outcomes after total knee arthroplasty in an Asian population. J Arthroplasty 2018;33:3186-9. Crossref

35. Stegenga ME, Vincent JL, Vail GM, et al. Diabetes does not alter mortality or hemostatic and inflammatory responses in patients with severe sepsis. Crit Care Med 2010;38:539-45. Crossref

36. Zilberberg MD, Nathanson BH, Sulham K, Fan W, Shorr AF. 30-day readmission, antibiotics costs and costs of delay to adequate treatment of Enterobacteriaceae UTI, pneumonia, and sepsis: a retrospective cohort study. Antimicrob Resist Infect Control 2017;6:124. Crossref

37. Gutiérrez-Gutiérrez B, Salamanca E, de Cueto M, et al. Effect of appropriate combination therapy on mortality of patients with bloodstream infections due to carbapenemase-producing Enterobacteriaceae (INCREMENT): a retrospective cohort study. Lancet Infect Dis 2017;17:726-34. Crossref

38. Cheng WL, Hsueh PR, Lee CC, et al. Bacteremic pneumonia caused by extended-spectrum beta-lactamase-producing Escherichia coli and Klebsiella pneumoniae: appropriateness of empirical treatment matters. J Microbiol Immunol Infect 2016;49:208-15. Crossref

39. Kohler PP, Volling C, Green K, Uleryk EM, Shah PS, McGeer A. Carbapenem resistance, initial antibiotic therapy, and mortality in Klebsiella pneumoniae bacteremia: a systematic review and meta-analysis. Infect Control Hosp Epidemiol 2017;38:1319-28. Crossref

40. Zilberberg MD, Shorr AF, Micek ST, Mody SH, Kollef MH. Antimicrobial therapy escalation and hospital mortality among patients with health-care-associated pneumonia: a single-center experience. Chest 2008;134:963-8. Crossref

41. Raman G, Avendano E, Berger S, Menon V. Appropriate initial antibiotic therapy in hospitalized patients with gram-negative infections: systematic review and meta-analysis. BMC Infect Dis 2015;15:395. Crossref

42. Seymour CW, Gesten F, Prescott HC, et al. Time to treatment and mortality during mandated emergency care for sepsis. N Engl J Med 2017;376:2235-44. Crossref

43. Moreno R, Vincent JL, Matos R, et al. The use of maximum SOFA score to quantify organ dysfunction/failure in intensive care. Results of a prospective, multicentre study. Working Group on Sepsis related Problems of the ESICM. Intensive Care Med 1999;25:686-96. Crossref

44. Minne L, Abu-Hanna A, de Jonge E. Evaluation of SOFA-based models for predicting mortality in the ICU: a systematic review. Crit Care 2008;12:R161.

45. Singer M, Deutschman CS, Seymour CW, et al. The third international consensus definitions for sepsis and septic shock (sepsis-3). JAMA 2016;315:801-10. Crossref

Pill splitting by patients is common globally. A German study observed that 24.1% of all drugs required splitting,1 and a Swiss study found that 10% of all discharged prescriptions contained pill splitting.2 In Sweden, 10% of 600 000 investigated prescriptions required splitting, and over 30% of the Swedish patients stated that they had problems dividing the tablets.3 The observed prevalence of pill splitting has been observed to range from 10% to >35% worldwide1 4 5 6 and was even higher in elderly patients (35-67%).6 7 One of the reasons for pill splitting is cost saving because it may result in institutions not needing to stock too many drug items in their formularies.1 In addition, drug splitting may achieve dose flexibility, particularly for patients requiring frequent dosing adjustment.8 Furthermore, some dosages may not be commercially available, especially those for off-label drug use. In these cases, splitting drugs may be essential.9 10 11 Nevertheless, it can also create other clinical issues including medication non-compliance, difficulties experienced by patients in handling unscored pills, drugs that crumble after splitting, and inappropriate drug splitting of extended release formulations, which may lead to treatment failure or toxicity.12 A published study reported that most hypertensive patients preferred not to split pills and that over 70% of patients were willing to pay more for medications with dosages that they did not need to split.13 Limited published studies have addressed this drug-related problem. In this study, we aimed to identify the effects of pill splitting on drug physiochemical properties and clinical outcomes among elderly cardiac patients.

A parallel design was adopted in this study. Patients were recruited from the Cardiac or Hypertension clinics of the Prince of Wales Hospital, Hong Kong. After medical records review, it was found that lisinopril, amlodipine, simvastatin, metformin, and perindopril were among the most commonly prescribed medications that required splitting in the two clinics. Therefore, patients who needed to split their pills when taking lisinopril, amlodipine, simvastatin, metformin, or perindopril were recruited. Patients were randomised into either group A (pill splitting with self-technique) or group B (pill splitting with instructions and training) after their first clinic visit. All patients were asked to sign an informed consent form before enrolment. Group A patients were asked to split drugs using their own technique and continue until their next clinic visit. Group B patients were given proper instruction by pharmacists or pharmacy students on using pill cutters at their first visit and were asked to cut their pills accordingly until their next clinic visit. Patients watched a 2-minute video that explained the reasons for using pill cutters and described the proper way to open the pill cutter, position the drug in the pill cutter, clean the pill cutter, and store the split pills. Subsequently, the pharmacist answered patients’ enquiries regarding the video or other questions related to pill splitting. The current study did not change any drug or dosage of the patient’s existing treatment regimens. Follow-up clinic visits were scheduled with mean duration between first and follow-up clinic visits of 23.1±7.3 weeks.

Chinese patients aged ≥65 years, both male and female, and currently prescribed one or more of metformin, lisinopril, perindopril, amlodipine, or simvastatin (which require splitting) were included in the current study. Patients with dementia or severe physical limitations such as hemiplegia, blindness, or upper limb contractures were excluded from the study.

In our pilot study, we found that the change in drug assays of metformin, atenolol, and amlodipine varied from 52.7% to 147.2% after splitting.14 In our previously published study, systolic blood pressure decreased significantly (from 152.38±18.80 mm Hg to 147.04±20.72 mm Hg, P=0.021) after pharmacist intervention.15 The sample size for the primary outcome was calculated based on a population standard deviation of 0.75, and that for the secondary outcomes was based on a population standard deviation of 0.85. To achieve a 5% significance level and 80% power, 30 and 45 patients in each group would be needed for the primary and secondary outcomes, respectively. We expected a 10% dropout rate, and therefore, at least 80 patients were recruited for each group. There were 193 total participants in this study.

The participants were randomised into group A or B using a computerised dynamic allocation programme and stratification according to types of medications taken, sex, age, and visit dates to ensure balanced patient allocation. All operations were performed by the same pharmacist who conducted the survey.

The primary outcome was the change of drug content before and after the pill splitting training. At baseline, patients in groups A and B were asked to split three tablets of the drugs that they were currently taking using their usual technique. At follow-up visit, group A patients were asked to split three tablets using their own technique, and group B patients were asked to split three tablets using a pill cutter. Two halved tablets were randomly selected for analysis each time. The halved tablets were weighed, and standard drug assays were performed. The drug content of metformin was assessed by ultraviolet-visible spectrophotometry; that of amlodipine, simvastatin, and perindopril was assessed by ultra-performance liquid chromatography; and that of lisinopril was assessed by high-performance liquid chromatography.

The secondary outcomes were the changes in clinical outcomes between before and after the pill splitting training, including the change in blood pressure measurements, haemoglobin A1c, and cholesterol levels, the changes in patients’ attitudes towards and acceptance of pill splitting, and the changes in patients’ knowledge about pill splitting. Haemoglobin A1c and lipid levels were usually collected in hospital 1 week before patients’ clinic visit. Upon initiating their clinic visits, patients had their blood pressures measured in the hospital’s nurse station before they met their physicians. Blood pressure, haemoglobin A1c, and cholesterol levels were collected at baseline and at follow-up visit, and the patients’ knowledge about and attitudes towards pill splitting were assessed by questionnaires.

Paired-samples t tests were used for intra-group comparisons, and independent-samples t tests were used for inter-group comparisons of mean tablet weight between before and after pill splitting training. Paired-samples t tests were also used to assess the changes in clinical outcomes. McNemar’s test was used for intra-group comparisons, and Fisher’s exact test (two-sided) was used for inter-group comparisons of content uniformity, change in patients’ acceptance towards pill splitting, and change in patients’ knowledge about pill splitting. A P value of <0.05 was considered statistically significant. All analyses were performed using the SPSS statistical programme (Windows version 25.0; IBM Corp, Armonk [NY], United States).

A total of 193 eligible patients were enrolled on or before 17 January 2019, and they had follow-up visits on or before 30 April 2019. The patients were randomised into group A (n=106) and group B (n=87). A total of 101 patients participated, of whom 47 from group A and 54 from group B returned for follow-up visit. Among them, 46 patients from group A and 47 patients from group B provided samples for assay. The primary outcome analysis was conducted on those patients, and the secondary outcomes analysis was conducted on the 101 patients who returned for follow-up visit. The patients’ demographic data are shown in Table 1.

The primary outcome was the change of drug content between before and after the pill splitting training. Patients were asked to split three tablets during each visit. Two halved tablets were randomly selected as samples. The samples were weighed, and assays were performed. The mean weight of the halved tablets at baseline and at follow-up visit is documented in Table 2. Table 3 shows the percentage of halved tablets that were within the assay specifications at baseline and at follow-up visit. The percentage of samples with both halved tablets within range was compared between groups A and B. In group A, the percentage in range increased from 39.13% to 47.82% (P=0.523), and the corresponding increase for group B was from 48.94% to 51.06% (P=1.000). The difference in drug assay results between groups A and B at baseline (P=0.406) and at follow-up visit (P=0.837) also did not reach statistical significance.

The correlation between pill cutting training and clinical outcomes is summarised in Table 4. The mean triglyceride in group B decreased significantly from 1.62±1.05 to 1.36±0.80 mmol/L (P=0.049), whereas the mean heart rate increased significantly from 73.97±11.01 to 77.92±12.72 bpm (P=0.026). In group B, there was also improvement in the mean diastolic blood pressure (from 73.40±14.39 to 73.05±9.33 mm Hg), high-density lipoprotein (from 1.40±0.39 to 1.46±0.44 mmol/L), low-density lipoprotein (from 1.87±0.88 to 1.85±0.73 mmol/L), and total cholesterol (from 3.98±0.93 to 3.91±0.85 mmol/L), but those differences did not reach statistical significance. In the overall cohort, improvements were seen in diastolic blood pressure (from 74.21±12.47 to 74.01±9.98 mm Hg), high-density lipoprotein (from 1.42±0.37 to 1.43±0.40 mmol/L), total cholesterol (from 3.90±0.97 to 3.88±0.96 mmol/L), and triglyceride (from 1.46±0.91 to 1.31±0.72 mmol/L), but the changes did not reach statistical significance.

In total, 57.43% of patients split their pills with their bare hands, followed by pill cutters (24.75%), knives (13.86%), and scissors (10.89%). The major reasons for not using pill cutters included: (1) the current method could split pills evenly (68.18%), (2) using pill cutters was time consuming (34.09%), and (3) the pills could not be split evenly by pill cutters (15.91%). The major reasons for using pill cutters included: (1) pills could be cut evenly (80.95%) and (2) the patient was able to exert force more easily (33.33%). In total, 29.70% and 24.75% of patients found pill splitting troublesome at baseline and at follow-up visit, respectively (the difference was not significant, P=0.063). The three major problems encountered by patients while splitting pills were (1) difficulty splitting the pills evenly (17.00%), (2) the pills easily fragmented (10.00%), and (3) difficulty seeing the pills clearly, as they were too small (9.00%). Overall, 61.00% of the patients claimed that they had no difficulties. Nevertheless, 98.21% preferred to take tablets with exact dosages so that no splitting would be required. Patients’ responses to other questions are listed in Table 5.

Table 6 shows that a significantly higher portion of patients in group B had a correct understanding of the following three questions after training: ‘Using pill cutters allows pills to be divided into more accurate doses’ (from 7.41% to 31.48%; P=0.002); ‘The pills should be put into the triangular tip of the pill cutter’ (from 9.26% to 31.48%; P=0.008); and ‘Pill cutters should be stored in a cool and dry place, away from sun or moisture’ (from 9.26% to 35.19%; P=0.003). In contrast, patients in group A did not show a statistically significant improvement in their understanding of any question. During the interview and evaluation of patients’ knowledge about pill splitting at baseline and at follow-up visit, we did not detect any patients with major physical or cognitive abnormalities.

Tablet splitting is a common practice in in-patient and out-patient settings,16 and it may be desirable in terms of dose adjustment, cost saving, and ease of swallowing.3 11 12 17 18 19 20 Nevertheless, it has been reported that splitting pills may cause drug instability, loss of drug due to powdering, uneven dosage, and reduced drug strength.21 22 It is generally understood that using tablet splitting devices can provide a more consistent dose.10 21 Previous studies have identified some characteristics that might affect the quality of halved tablets. Coated, unscored, and small tablets were found to be more difficult to cut.23 Individual pill cutting skill was another crucial factor that determined tablets’ uniformity.23 In the current study, only 24.75% of patients split pills using pill cutters, and only 14.43% of patients had received pill splitting training. Therefore, it is likely that the drug content in the halved tablets did not reach assay standards.

Previous studies mainly focused on the weight deviations among halved tablets, not on drug content.21 22 One study showed that more than one-third of sampled half-tablets did not meet the United States Pharmacopeia specifications.24 The measured drug content variations among half-tablets were: warfarin sodium (90.01%-109.40%), simvastatin (95.21%-111.35%), metoprolol succinate (82.77%-115.92%), metoprolol tartrate (94.83%-112.37%), citalopram (96.50-111.93%), and lisinopril (81.15%-125.72%). In another study, five of eight drugs failed to meet European Pharmacopoeia recommendations for tablet weight deviation after splitting, with 25% of samples deviating by >15% and 10% of samples deviating by >25%.23 The study drugs used were phenobarbitone (maximum deviation: 80.45%), digoxin (maximum deviation: 56.69%), chloroquine (maximum deviation: 48.97%), atenolol (maximum deviation: 45.37%), and doxycycline (maximum deviation: 43.97%). In the present study, both halves of the tablet were within the assay standard at baseline for 39.13% and 48.94% of the patients in groups A and B, respectively. After training, this percentage increased to 47.82% and 51.06%, respectively, but the improvement was not significant, and the percentage of tablets in range was still relatively low. The results corroborated those of previous studies.

Few studies have examined the effect of patient education on the drug content of split pills.25 26 In the current study, we found no significant improvement in content uniformity after pill splitting training. This may be because our patients were elderly patients who may not have been able to perform the task well after a single training session. Content uniformity after pill splitting may be improved if pills are split by pharmacists or qualified staff. A study of paediatric pharmacists suggested that tablets >8 mm could be split once to achieve an approximate half dose for paediatric use.27 Another study found a significant difference in splitting accuracy between nurses and laypersons.23 Nevertheless, only 39.29% of that study’s patients were eager to partake of pill splitting service, and only 18.18% were willing to pay extra money for it. Therefore, pill cutting service may not be practical without a financial incentive.

Triglyceride levels decreased significantly and heart rate increased significantly in group B patients after the intervention. Nevertheless, we did not evaluate the patients’ diet consumption or exercise levels, which may impact their triglyceride levels. No significant change in clinical outcomes was observed in other groups or other parameters. Because the studied drugs were lisinopril, perindopril, simvastatin, and amlodipine, which are not narrow-therapeutic-index drugs, these results were predictable and coincide with other studies that concluded that drugs with long half-life and wide therapeutic index are less likely to be affected.20 28 In view of the high variability of blood pressure measurements in the clinic, all patients were originally instructed to conduct daily blood pressure measurements at home using a portable blood pressure monitor. However, many patients did not measure their blood pressure daily or did not keep a proper self-record, so the clinical outcomes relied on the readings at clinic visits, which may not be consistent with their usual readings. In addition, management of chronic diseases like hypertension, diabetes mellitus, and dyslipidaemia could be influenced by multiple factors, and 3 months was a relatively short period for observation. The effect of drug content deviation after splitting on clinical outcomes may be more obvious in antibiotics or drugs with narrow therapeutic index (eg, digoxin).23 29 30

In the current study, we focused on the effect of pill splitting on drug content. Nevertheless, pill splitting may have other effects on drugs. The pill may carry a bitter taste, as the coating is broken, and the active ingredients may be more susceptible to moisture after exposure.31 Over 70% of patients prepared a sufficient quantity of pills for more than 1 day each time. In total, 36.63% of patients cut for 2 days to 1 week, 26.73% cut for 1 week to 1 month, and 6.93% cut for more than 1 month each time. Exposing the cut pills for too long may increase the risk of crushing or cracking.19 In total, 83.93% of patients found pill splitting training helpful, and the intervention produced significant improvement in patients’ knowledge about pill splitting. This study has identified the major difficulties encountered by patients and the reasons behind their choices. Those problems should be addressed in future patient education. More than half of patients split pills with their bare hands, and the majority of patients who did not use pill cutters thought their own methods could divide pills evenly and that the use of pill cutters was time consuming. The major obstacles patients faced were the difficulties in splitting pills evenly and that the pills fragmented easily. Overall, 98.21% of patients preferred to take tablets with the exact dosage instead of splitting pills. Previous studies also found that dispensing the exact dosage would be more favourable.23 32 Nevertheless, if pill splitting is unavoidable, pharmacists should encourage patients to split coated, unscored, or irregularly shaped tablets with pill cutters to reduce crushing or fragmenting. Pharmacists should also educate patients about the appropriate way to use and clean pill cutters and remind patients to seek doctors’ or pharmacists’ advice before cutting any pills.21 23 33

This project has several limitations. First, the participants’ dropout rate was high, which might result in attrition bias. Compared with group A, a higher proportion of group B patients returned for follow-up visits. The statistically significant improvement in clinical outcomes among group B patients might be caused by their higher awareness about their own health instead of the effectiveness of the pill splitting method. There were limited human resources to make phone calls to patients between the baseline and face-to-face follow-up visits, which could have served as a reminder for patients to attend follow-up visits and perform home monitoring of their blood pressure and their pill splitting methods.

Second, dietary consumption and exercise levels were not evaluated, even though they may affect the clinical outcomes. Third, participants’ education level, household income, and major caregivers were not collected at baseline. Only approximately 65% of participants who attended follow-up visits provided such information. These confounding factors might affect patients’ ability to understand and memorise the steps of using pill cutters, thus affecting the content uniformity of their split pills. The effects of patients’ characteristics on their knowledge and pill splitting skills were not assessed in the current study.

This study revealed that content uniformity can hardly be achieved after pill splitting by patients. No significant difference in clinical outcomes was observed after pill splitting training. It is preferable for pills with doses that do not require splitting to be provided, considering the assay results and patients’ preference. Currently, there is inadequate patient education about pill splitting. Pharmacists should educate patients to use pill cutters properly if splitting is inevitable.

Concept or design: VWY Lee, FYH Fong, BPY Yan.
Acquisition of data: VWY Lee, FYH Fong, BPY Yan.
Analysis or interpretation of data: JTS Li.
Drafting of the manuscript: JTS Li.
Critical revision of the manuscript for important intellectual content: VWY Lee.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

As an editor of the journal, BPY Yan was not involved in the peer review process. Other authors have disclosed no conflicts of interest.

We thank the technicians at the School of Pharmacy, The Chinese University of Hong Kong for carrying out the drug physiochemical tests and the staff at Prince of Wales Hospital for arranging the logistics for patient counselling.

This study was supported by the Health and Medical Research Fund, Food and Health Bureau, Hong Kong SAR Government (#14152111). The funder had no role in study design, data collection/analysis/interpretation, or manuscript preparation.

Ethical approval was obtained from The Joint Chinese University of Hong Kong–New Territories East Cluster Clinical Research Ethics Committee (Clinical trial registration no.: CREC Ref No. 2017.014). Patient consent was obtained upon enrolment. The trial protocol can be obtained as requested.

1. Quinzler R, Gasse C, Schneider A, Kaufmann-Kolle P, Szecsenyi J, Haefeli WE. The frequency of inappropriate tablet splitting in primary care. Eur J Clin Pharmacol 2006;62:1065-73. Crossref

2. Berg C, Ekedahl A. Dosages involving splitting tablets: common but unnecessary? J Pharm Health Serv Res 2010;1:137-41. Crossref

3. Rodenhuis N, De Smet PA, Barends DM. The rationale of scored tablets as dosage form. Eur J Pharm Sci 2004;21:305- 8. Crossref

4. Rodenhuis N, de Smet PA, Barends DM. Patient experiences with the performance of tablet score lines needed for dosing. Pharm World Sci 2003;25:173-6. Crossref

5. Stanković I, Pešić M, Lalić M. Breaking of tablets as a cost-saving strategy. Timočki Medicinski Glasnik 2007;32:5-10.

6. Fischbach MS, Gold JL, Lee M, Dergal JM, Litner GM, Rochon PA. Pill splitting in a long-term care facility. CMAJ 2001;164:785-6.

7. Denneboom W, Dautzenberg MG, Grol R, De Smet PA. User-related pharmaceutical care problems and factors affecting them: the importance of clinical relevance. J Clin Pharm Ther 2005;30:215-23. Crossref

8. Peek BT, Al-Achi A, Coombs SJ. Accuracy of tablet splitting by elderly patients. JAMA 2002;288:451-2. Crossref

9. Duncan MC, Castle SS, Streetman DS. Effect of tablet splitting on serum cholesterol concentrations. Ann Pharmacother 2002;36:205-9. Crossref

10. Rindone JP. Evaluation of tablet-splitting in patients taking lisinopril for hypertension. JCOM Wayne PA 2000;7:22-30.

11. Gee M, Hasson NK, Hahn T, Ryono R. Effects of a tablet-splitting program in patients taking HMG-CoA reductase inhibitors: analysis of clinical effects, patient satisfaction, compliance and cost avoidance. J Manag Care Pharm 2002;8:453-8. Crossref

12. Fawell NG, Cookson TL, Scranton SS. Relationship between tablet splitting and compliance, drug acquisition costs, and patient acceptance. Am J Health Syst Pharm 1999;56:2542-5. Crossref

13. McDevitt JT, Gurst AH, Chen Y. Accuracy of tablet splitting. Pharmacotherapy 1998;18:193-7.

14. Lee VW, Yu Y, Tang L, Yan B. Impact of pill-splitting training on drug physiochemical properties, compliance and clinical outcomes in elderly population–a cross-over cohort study. Value Health 2013;16:A514. Crossref

15. Lee VW, Pang PT, Kong KW, Chan PK, Kwok FL. Impact of pharmacy outreach services on blood pressure management in the elderly community in Hong Kong. Geriatr Gerontol Int 2013;13:175-81. Crossref

16. Crawford DJ. Is tablet splitting safe? Nursing made incredibly easy 2011;9:18-20. Crossref

17. Cohen CI, Cohen SI. Potential savings from splitting newer antidepressant medications. CNS Drugs 2002;16:353-8. Crossref

18. Stafford RS, Radley DC. The potential of pill splitting to achieve cost savings. Am J Manag Care 2002;8:706-12.

19. Bachynsky J, Wiens C, Melnychuk K. The practice of splitting tablets: cost and therapeutic aspects. Pharmacoeconomics 2002;20:339-46. Crossref

20. Helmy SA. Tablet splitting: is it worthwhile? Analysis of drug content and weight uniformity for half tablets of 16 commonly used medications in the outpatient setting. J Manag Care Spec Pharm 2015;21:76-86. Crossref

21. Verrue C, Mehuys E, Boussery K, Remon JP, Petrovic M. Tablet-splitting: a common yet not so innocent practice. J Adv Nurs 2010;67:26-32. Crossref

22. Rosenberg JM, Nathan JP, Plakogiannis F. Weight variability of pharmacist-dispensed split tablets. J Am Pharm Assoc (Wash) 2002;42:200-5. Crossref

23. Elliott I, Mayxay M, Yeuichaixong S, Lee SJ, Newton PN. The practice and clinical implications of tablet splitting in international health. Trop Med Int Health 2014;19:754-60. Crossref

24. Hill SW, Varker AS, Karlage K, Myrdal PB. Analysis of drug content and weight uniformity for half-tablets of 6 commonly split medications. J Manag Care Pharm 2009;15:253-61. Crossref

25. Gharaibeh SF, Tahaineh L. Effect of different splitting techniques on the characteristics of divided tablets of five commonly split drug products in Jordan. Pharm Pract (Granada) 2020;18:1776. Crossref

26. Ashrafpour R, Ayati N, Sadeghi R, et al. Comparison of treatment response achieved by tablet splitting versus whole tablet administration of levothyroxine in patients with thyroid cancer. Asia Ocean J Nucl Med Biol 2018;6:108-12. Crossref

27. Andersson ÅC, Lindemalm S, Eksborg S. Dividing the tablets for children—good or bad? Pharm Methods 2016;7:23-7. Crossref

28. Chou CL, Hsu CC, Chou CY, Chen TJ, Chou LF, Chou YC. Tablet splitting of narrow therapeutic index drugs: a nationwide survey in Taiwan. Int J Clin Pharm 2015;37:1235-41. Crossref

29. Simpson JA, Watkins ER, Price RN, Aarons L, Kyle DE, White NJ. Mefloquine pharmacokinetic-pharmacodynamic models: implications for dosing and resistance. Antimicrob Agents Chemother 2000;44:3414-24. Crossref

30. White NJ, Pongtavornpinyo W, Maude RJ, et al. Hyperparasitaemia and low dosing are an important source of anti-malarial drug resistance. Malar J 2009;8:253. Crossref

31. Margiocco ML, Warren J, Borgarelli M, Kukanich B. Analysis of weight uniformity, content uniformity and 30-day stability in halves and quarters of routinely prescribed cardiovascular medications. J Vet Cardiol 2009;11:31-9. Crossref

32. Madathilethu J, Roberts M, Peak M, Blair J, Prescott R, Ford JL. Content uniformity of quartered hydrocortisone tablets in comparison with mini-tablets for paediatric dosing. BMJ Paediatr Open 2018;2:e000198. Crossref

33. Allemann SS, Bornand D, Hug B, Hersberger KE, Arnet I. Issues around the prescription of half tablets in Northern Switzerland: the irrational case of quetiapine. Biomed Res Int 2015;2015:602021. Crossref