Type 2 diabetes mellitus (T2DM) in children and adolescents (hereinafter, youth) is becoming increasingly common worldwide.1 2 A recent meta-analysis estimated that approximately 41 600 new cases of T2DM were identified among youth in 2021.3 Type 2 DM in youth exhibits relatively rapid clinical progression with a sharp decline in beta-cell function and high risk of complications.4 In a study recently published by the TODAY (Type 2 Diabetes in Adolescents and Youth) Study Group, which analysed 500 young adults with youth-onset T2DM, 60.1% of patients developed at least one microvascular complication (diabetic kidney, nerve, or retinal disease) and 28.4% of patients developed at least two complications.5 In addition to hyperglycaemia, the presence of co-morbidities (eg, hypertension and dyslipidaemia) was associated with an increased risk of complications.5

A similar increase in the incidence of T2DM has been observed in Hong Kong. We previously reported that the crude incidence rate increased from 1.27 per 100 000 person-years in 1997-2007 to 3.42 per 100 000 person-years in 2008-2017.6 However, there have been limited data regarding the outcomes of paediatric T2DM patients in Hong Kong. In this study, we reviewed the glycaemic control findings and microvascular complication rates among recently diagnosed paediatric T2DM patients in Hong Kong, with a focus on outcomes at 2 years after diagnosis; we sought to identify factors associated with poor glycaemic control and the development of microalbuminuria (MA).

Data analysed in this study were retrieved from the Hong Kong Childhood Diabetes Registry, a database established in 2016. The Registry was approved by the Research Ethics Committee of the Hospital Authority of Hong Kong, which includes 11 public hospitals. Investigators retrieved information from medical records and entered relevant data into the Registry. Standardised data entry forms for recording baseline clinical characteristics and annual entry forms were provided for investigators to enter data into the Registry at diagnosis and annually thereafter. Data were cross-checked by at least two investigators.

All patients aged <18 years who had been diagnosed with DM at public hospitals in Hong Kong were recruited. All recruited patients met the diagnostic criteria for DM according to the International Society for Paediatric and Adolescent Diabetes Clinical Practice Consensus Guidelines: patients were symptomatic and had either fasting blood glucose level ≥7 mmol/L, 2-hour blood glucose level ≥11.1 mmol/L during an oral glucose tolerance test, random blood glucose level ≥11.1 mmol/L, or haemoglobin A1c (HbA1c) level ≥6.5%.4 Asymptomatic patients underwent repeat testing with a different test, as suggested in the Guidelines.4 The classification of DM was based on the attending clinician’s assessment of clinical symptoms and laboratory findings, including obesity status, family history, autoimmunity, and clinical course. Patients diagnosed with T2DM from 2014 to 2018 were included in the analysis, including those who had an initial diagnosis of type 1 DM that was subsequently revised to T2DM. Patients who refused Registry recruitment and patients whose diagnosis was revised to type 1 DM or maturity-onset DM of the young were not included in the analysis.

The following data were retrieved from the Registry: patient age, sex, family history of T2DM (in first- or second-degree relatives), symptoms at presentation, anti-islet cell antibody test results, body mass index (BMI), HbA1c level, presence of co-morbidities (non-alcoholic fatty liver disease, dyslipidaemia, hypertension, and obstructive sleep apnoea), presence of complications (MA, retinopathy, and neuropathy), treatments received, and frequency of blood glucose self-monitoring. Overweight and obesity were defined using age- and sex-specific cut-offs established by the International Obesity Task Force, which predicted BMI values at 18 years (25, 30, and 35 kg/m²) by the respective standard deviations to define overweight, obesity and morbid obesity, respectively; standard deviations of BMI were calculated according to age- and sex-specific reference data provided by the International Obesity Task Force.7 In this study, weight loss was defined as any decrease in BMI z-score, and improvement in HbA1c level was defined as any decrease in HbA1c level. Non-alcoholic fatty liver disease was defined as an elevated alanine transferase level (based on age- and sex-specific reference data) or compatible ultrasound findings. Dyslipidaemia was defined as an elevated low-density lipoprotein level of ≥2.6 mmol/L, a triglyceride level ≥1.7 mmol/L, or the receipt of lipid-lowering agents. Hypertension was defined as an elevated systolic blood pressure ≥95th percentile for age, height, and sex—on at least two occasions—or the receipt of anti-hypertensive medication. Obstructive sleep apnoea was defined as the presence of clinical symptoms indicating sleep-disordered breathing, along with polysomnography findings of obstructive apnoeas/hypopneas. Microalbuminuria was defined as an elevated spot urine albumin-creatinine ratio >2.5 mg/mmol for boys and >3.5 mg/mmol for girls in at least two of three samples within a 6-month period, or as the receipt of any treatment for MA. Retinopathy (eg, non-proliferative and proliferative diabetic retinopathy, as well as macula oedema) was identified by digital fundus photography and confirmed via referral to an ophthalmologist. Neuropathy was clinically identified by the presence of symptoms (numbness and paraesthesia) and through clinical examinations including the 10-g monofilament test, vibration sense assessment, and ankle reflex evaluation. Suboptimal glycaemic control was defined as HbA1c level ≥7%, as suggested by the International Society for Paediatric and Adolescent Diabetes Clinical Practice Consensus Guidelines.4

Statistical analyses were performed using SPSS software (Windows version 23; IBM Corp, Armonk [NY], United States). All available data were included in the statistical analysis, and the numbers of available values are listed in the tables. Continuous variables, including age, HbA1c level, and BMI z-score, were tested for normality using the Shapiro–Wilk test. Data with skewed distributions were expressed as medians and interquartile ranges (IQRs), and comparisons were made using the Mann-Whitney U test. Analyses of relationships between two continuous variables were assessed by Spearman rank correlation and expressed using the Spearman correlation coefficient (ρ). Categorical variables were expressed as exact numbers of patients with percentages. Binary logistic regression analysis was used to assess risk factors for suboptimal glycaemic control at 2 years and MA at 2 years. Univariate analyses were performed to determine unadjusted odds ratios and 95% confidence intervals. Multivariate analyses of factors associated with suboptimal glycaemic control at 2 years were performed while including HbA1c level at diagnosis to adjust for initial disease severity. Multivariate analyses of factors associated with MA at 2 years were performed while including HbA1c level at 2 years to eliminate the effect of glycaemic control at 2 years; this approach was intended to independently assess the effects of co-morbidities. Missing data were not included in regression analyses. Statistical tests were two-sided and were performed with a 5% significance threshold (ie, alpha=0.05). The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist for cohort studies was used when reporting the study findings.

In total, 212 patients diagnosed with T2DM between 2014 and 2018 were recruited into the Registry. Their baseline demographics are summarised in Tables 1 and 2. Of these patients, 71.3% had a family history of T2DM, and 21.7% were symptomatic at diagnosis (Table 1). At 2 years after the diagnosis of DM, 143 patients (67.5%) continued attending follow-up visits. There were no significant differences in baseline characteristics between the groups with and without follow-up at 2 years, except for a higher age at diagnosis in the loss to follow-up group (online supplementary Table).

Haemoglobin A1c levels at diagnosis, 1 year after diagnosis, and 2 years after diagnosis were available for 203, 176, and 135 patients, respectively. The median HbA1c levels at diagnosis, 1 year after diagnosis, and 2 years after diagnosis were 7.5% (IQR=6.5%-10.6%), 6.3% (IQR=5.8%-7.4%), and 6.5% (IQR=5.8%-8.0%), respectively; at these times, 65.5%, 29.0%, and 40.7% of patients had suboptimal glycaemic control (ie, HbA1c level ≥7%), respectively [Table 2].

There was an overall improvement in BMI z-score at 2 years after diagnosis (median BMI z-score decreased from 2.5 at diagnosis to 2.3 at 2 years). Overall, 146 of 191 patients (76.4%) had dyslipidaemia at diagnosis, whereas 62 of 95 patients (65.3%) had dyslipidaemia at 2 years. However, more patients had hypertension at 2 years—the number increased from 45 of 212 patients (21.2%) at diagnosis to 55 of 143 patients (38.5%) at 2 years (Table 2).

Overall, 21 of 113 (18.6%) patients screened at 2 years after diagnosis developed MA, compared with 9.0% at diagnosis. Two patients (1.8%) developed retinopathy, whereas one patient (0.9%) developed neuropathy, at 2 years after diagnosis (Table 2).

At diagnosis, 24.1% of patients were not receiving any pharmacological treatment, 58.0% were receiving anti-diabetic drugs, and 18% required insulin. At 2 years, only 12.6% of patients were not receiving any medications. The proportions of patients requiring insulin were similar at diagnosis and 2 years (2.1%). In total, 64.9% of patients did not perform daily blood glucose self-monitoring (Table 2).

A higher initial HbA1c level was associated with suboptimal glycaemic level at 2 years (correlation coefficient=0.39, P<0.001; n=130). There were no significant correlations of HbA1c level at 2 years with age at diagnosis (correlation coefficient=0.02, P=0.852; n=135), BMI z-score at diagnosis (correlation coefficient=-0.10, P=0.277; n=133), or BMI z-score at 2 years (correlation coefficient=0.04, P=0.638; n=131). Greater decline in BMI z-score was associated with a lower HbA1c level at 2 years (correlation coefficient=-0.22, P=0.011; n=129). However, there was no correlation between the change in BMI z-score and the change in HbA1c level (correlation coefficient <0.01, P=0.973; n=126).

Table 3 shows factors associated with suboptimal glycaemic control at 2 years. The effect of a family history of T2DM was not statistically significant after adjustment for initial HbA1c level (adjusted odds ratio [aOR]=2.29; P=0.075). A similar result was observed regarding the effect of symptomatic disease at diagnosis (aOR=2.01; P=0.174) and weight loss (aOR=0.53; P=0.128). The presence of fatty liver (aOR=2.50; P=0.021) and dyslipidaemia (aOR=3.19; P=0.033) at 2 years were associated with suboptimal glycaemic control at 2 years, even after adjustment for initial HbA1c level.

Patients with MA had higher HbA1c levels at 2 years compared with patients who did not exhibit MA (median HbA1c level: 7.2% vs 6.4%; P=0.037) [Table 4]. Dyslipidaemia at 2 years was associated with MA at 2 years in the univariate analysis (unadjusted odds ratio=5.51; P=0.030), but the effect did not remain statistically significant after adjusting for glycaemic control at 2 years (Table 5). The presence of hypertension at 2 years was a risk factor for MA at 2 years, independent of glycaemic control at 2 years (aOR=4.61; P=0.008) [Table 5].

The results of this study provide insights into the early post-diagnosis clinical course of T2DM among youth in Hong Kong. Nearly 60% of patients (59.3%) in our cohort had optimal glycaemic control with HbA1c level <7% at 2 years after diagnosis. Previous studies regarding glycaemic control among youth with T2DM showed variable results, presumably due to heterogeneity in the study populations, follow-up periods, and glycaemic targets.8 9 10 11 A clinical trial by the TODAY Study Group8 followed up 234 youth with DM, who were put on metformin and lifestyle modifications and with initial HbA1c level <8%, for 3.86 years on average. It showed that 46.6% of youth with DM exhibited loss of glycaemic control, defined by HbA1c level >8%.8 In a study of 301 paediatric T2DM patients with initial HbA1c level ≥7% in the United States, Barr et al9 found that after 1 year, 37% of patients achieved optimal control (HbA1c level ≤6.5%) and 58% achieved durable glycaemic control (HbA1c level ≤8%). However, at 3 years, only 26% of patients achieved HbA1c level ≤6.5%, whereas 59% exhibited HbA1c level ≤8%.9 Candler et al10 followed 100 paediatric T2DM patients in the United Kingdom; the median HbA1c level was 7% after 1 year, and 38.8% of patients exhibited HbA1c level <6.5%. Notably, no data regarding HbA1c levels at diagnosis were provided in the study.10 In a study of 96 patients in Israel, Meyerovitch et al11 found that the median HbA1c level was 7.97% after an average follow-up period of 3.11 years, compared with 7.8% at diagnosis. Additionally, >50% of patients required insulin at the end of the follow-up period.11 Although our cohort appeared to have better glycaemic control compared with the previous studies, our patients might have had lower initial HbA1c levels at diagnosis, considering that most of them were asymptomatic (78.3%). Our study also showed that patients with a higher initial HbA1c level tended to have a persistently high HbA1c level at 2 years. These findings emphasise the importance of early diagnosis and treatment before patients develop clinically significant hyperglycaemia, which makes DM more difficult to control. Most of our patients were overweight or obese (89.8%); many of them also had co-morbidities including hypertension, dyslipidaemia, and fatty liver (Table 2). Previous studies in Hong Kong showed a high risk of metabolic syndrome (OR up to 32.2) in overweight and obese children.12 13 Active screening for metabolic syndrome would enable early diagnosis and treatment of DM and its co-morbidities.

Factors associated with suboptimal glycaemic control were dyslipidaemia and fatty liver at 2 years after diagnosis. A recent study showed that each 1% increase in HbA1c level was associated with 13% and 20% increases in the risks of abnormal triglyceride and low-density lipoprotein levels, respectively.14 The importance of weight loss has been emphasised in various guidelines, for example, The American Diabetes Association recommends weight loss of at least 5% in adult overweight or obese DM patients.15 However, a specific weight loss target cannot be established in growing children. The study by Candler et al10 regarding youth with T2DM showed that each one-unit increase in BMI z-score was associated with a 34.9% increase in HbA1c level. Although the present study indicated that a greater drop in BMI z-score was associated with lower HbA1c level at 2 years, the association between weight loss and prevention of suboptimal glycaemic control at 2 years was not significant after adjustment for initial HbA1c level.

Diabetic retinopathy and neuropathy were rare among youth with T2DM. However, the proportion of patients with MA increased from 9.0% at diagnosis to 18.6% at 2 years (Table 2). Previous studies regarding the prevalence of diabetic complications in youth have shown mixed results, probably due to genetic variation and differences in DM duration. High prevalences have been observed in cohorts with long DM durations.16 The MA prevalence has been approximately 20% to 30% in most studies of youth with a short duration of T2DM. In the SEARCH for Diabetes in Youth study, the MA prevalence in youth with T2DM was 22.2%, and the average duration of disease was 1.9 years.17 In an Australian population, Eppens et al18 found that 28% of patients had MA, with a median disease duration of 1.3 years. Candler et al10 showed that the MA prevalence increased from 4.2% to 16.4% within 1 year after diagnosis. Our cohort showed a similar prevalence compared with other cohorts. Nevertheless, the increasing trend is concerning, particularly because MA has been identified as an independent predictor of mortality risk in adults.19 Thus, we conducted further analysis of risk factors for MA, revealing the associations of higher HbA1c level and hypertension at 2 years, consistent with the SEARCH for Diabetes in Youth study.17 The deleterious effects of hypertension on the kidneys explain the additional increase in MA risk, independent of glycaemic control.

Strengths of this study included its provision of insights regarding the early outcomes of youth with T2DM in Hong Kong, particularly concerning glycaemic control and associated factors. First, our findings supported the implementation of active screening in overweight and obese individuals to allow early diagnosis of DM, considering the high prevalence of overweight or obesity and the relationship of lower initial HbA1c level with better glycaemic control at 2 years. Second, our findings indicated that the presence of co-morbidities at 2 years, rather than baseline, was associated with suboptimal glycaemic control and MA, demonstrating the reversibility of the risk factors and highlighting the importance of co-morbidity management. Third, our study identified challenges in managing youth with T2DM, including a high loss to follow-up rate (n=69, 23.5%) [online supplementary Table], suboptimal glycaemic control in >40% of patients at 2 years, infrequent blood glucose self-monitoring by the patients, and increasing trends in MA and hypertension.

Indeed, the high loss to follow-up rate was a major limitation of our study. Many patients did not return for clinical assessment or were transferred to an adult endocrinology clinic. Although the loss to follow-up group had a higher age at diagnosis (online supplementary Table), this presumably did not have a substantial impact on the results because age was not a significant risk factor for poor glycaemic control or the likelihood of MA onset. Although a high loss to follow-up rate is a common phenomenon in studies of children with T2DM,20 this obstacle hindered the achievement of good glycaemic control and prevention of complications. It also created difficulty in acquiring long-term follow-up data. Another limitation was that our patients were managed by different doctors in different hospitals; there remains no standardised protocol for the management of paediatric T2DM patients in Hong Kong, which may be a confounding factor for multicentre studies such as ours.

Approximately 60% of youth with T2DM in Hong Kong achieved HbA1c level <7% at 2 years after diagnosis. A higher HbA1c level at diagnosis was associated with worse glycaemic control at 2 years. The presence of dyslipidaemia and fatty liver at 2 years were factors associated with suboptimal glycaemic control. Overall, 18.6% of patients developed MA at 2 years; other microvascular complications were rare. These results highlight the importance of early diagnosis and holistic management, including co-morbidity management. The high loss to follow-up rate, high proportion of patients with suboptimal glycaemic control, and increasing number of patients with MA and hypertension are ongoing challenges in the management of youth with DM. The establishment of a standardised protocol may improve outcomes in our patient population. Future research could include studies regarding the effects of insulin resistance and beta-cell function on metabolic outcomes in youth with DM.

Concept or design: All authors.
Acquisition of data: All authors.
Analysis or interpretation of data: WI Yam, SMY Wong.
Drafting of the manuscript: WI Yam.
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 received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

The research was conducted as a part of the Hong Kong Childhood Diabetes Registry, which was approved by the Kowloon Central / Kowloon East Cluster Research Ethics Committee of Hospital Authority, Hong Kong (Ref No.: KC/KE-16-0087/ER-3). Written informed consent was obtained from parents for patients aged <16 years and from both parents and patients for patients aged between ≥16 and <18 years.

The supplementary material was provided by the authors and some information may not have been peer reviewed. Accepted supplementary material will be published as submitted by the authors, without any editing or formatting. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by the Hong Kong Academy of Medicine and the Hong Kong Medical Association. The Hong Kong Academy of Medicine and the Hong Kong Medical Association disclaim all liability and responsibility arising from any reliance placed on the content.

1. Lawrence JM, Divers J, Isom S, et al. Trends in prevalence of type 1 and type 2 diabetes in children and adolescents in the US, 2001-2017. JAMA 2021;326:717-27. Crossref

2. Haynes A, Kalic R, Cooper M, Hewitt JK, Davis EA. Increasing incidence of type 2 diabetes in indigenous and non-indigenous children in Western Australia, 1990-2012. Med J Aust 2016;204:303. Crossref

3. Wu H, Patterson CC, Zhang X, et al. Worldwide estimates of incidence of type 2 diabetes in children and adolescents in 2021. Diabetes Res Clin Pract 2022;185:109785. Crossref

4. Shah AS, Zeitler PS, Wong J, et al. ISPAD Clinical Practice Consensus Guidelines 2022: type 2 diabetes in children and adolescents. Pediatr Diabetes 2022;23:872-902. Crossref

5. TODAY Study Group; Bjornstad P, Drews KL, et al. Long-term complications in youth-onset type 2 diabetes. N Engl J Med 2021;385:416-26. Crossref

6. Tung JY, Kwan EY, But BW, et al. Incidence and clinical characteristics of pediatric-onset type 2 diabetes in Hong Kong: the Hong Kong Childhood Diabetes Registry 2008 to 2017. Pediatr Diabetes 2022;23:556-61. Crossref

7. Cole TJ, Lobstein T. Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity. Pediatr Obes 2012;7:284-94. Crossref

8. TODAY Study Group; Zeitler P, Hirst K, et al. A clinical trial to maintain glycemic control in youth with type 2 diabetes. N Engl J Med 2012;366:2247-56. Crossref

9. Barr MM, Aslibekyan S, Ashraf AP. Glycemic control and lipid outcomes in children and adolescents with type 2 diabetes. PLoS One 2019;14:e0219144. Crossref

10. Candler TP, Mahmoud O, Lynn RM, Majbar AA, Barrett TG, Shield JP. Treatment adherence and BMI reduction are key predictors of HbA1c 1 year after diagnosis of childhood type 2 diabetes in the United Kingdom. Paediatr Diabetes 2018;19:1393-9. Crossref

11. Meyerovitch J, Zlotnik M, Yackobovitch-Gavan M, Phillip M, Shalitin S. Real-life glycemic control in children with type 2 diabetes: a population-based study. J Paediatr 2017;188:173-80.e1. Crossref

12. Ozaki R, Qiao Q, Wong GW, et al. Overweight, family history of diabetes and attending schools of lower academic grading are independent predictors for metabolic syndrome in Hong Kong Chinese adolescents. Arch Dis Child 2007;92:224-8. Crossref

13. Kong AP, Ko GT, Ozaki R, Wong GW, Tong PC, Chan JC. Metabolic syndrome by the new IDF criteria in Hong Kong Chinese adolescents and its prediction by using body mass index. Acta Paediatr 2008;97:1738-42. Crossref

14. Brady RP, Shah AS, Jensen ET, et al. Glycemic control is associated with dyslipidemia over time in youth with type 2 diabetes: the SEARCH for Diabetes in Youth study. Pediatr Diabetes 2021;22:951-9. Crossref

15. American Diabetes Association Professional Practice Committee. 8. Obesity and weight management for the prevention and treatment of type 2 diabetes: standards of medical care in diabetes—2022. Diabetes Care 2022;45(Supp 1):S113-24. Crossref

16. Amutha A, Mohan V. Diabetes complications in childhood and adolescent onset type 2 diabetes—a review. J Diabetes Complications 2016;30:951-7. Crossref

17. Maahs DM, Snively BM, Bell RA, et al. Higher prevalence of elevated albumin excretion in youth with type 2 than type 1 diabetes: the SEARCH for Diabetes in Youth study. Diabetes Care 2007;30:2593-8. Crossref

18. Eppens MC, Craig ME, Cusumano J, et al. Prevalence of diabetes complications in adolescents with type 2 compared with type 1 diabetes. Diabetes Care 2006;29:1300-6. Crossref

19. Chronic Kidney Disease Prognosis Consortium; Matsushita K, van der Velde M, et al. Association of estimated glomerular filtration rate and albuminuria with all-cause and cardiovascular mortality in general population cohorts: a collaborative meta-analysis. The Lancet 2010;375:2073-81. Crossref

20. Shoemaker A, Cheng P, Gal RL, et al. Predictors of loss to follow-up among children with type 2 diabetes. Horm Res Paediatr 2017;87:377-84. Crossref

Heart disease is the third leading cause of death in Hong Kong. In 2019, an average of approximately 10.2 people died from coronary heart disease each day.1 International guidelines recommend prehospital 12-lead electrocardiogram (ECG) for the assessment of patients with suspected acute coronary syndrome who present to emergency medical services (EMS).2 3 Prehospital triage with direct transfer to the cardiac catheterisation laboratory for primary percutaneous coronary intervention is a strategy adopted by various healthcare systems to reduce reperfusion time in patients with ST-elevation myocardial infarction (STEMI).4 5 Previous studies have investigated the diagnostic performances of prehospital electrocardiograms (PHECGs) for STEMI by various automated algorithms,6 7 8 9 10 trained onsite EMS personnel,11 12 and emergency department (ED) physicians remotely interpreting the tele-transmitted ECGs13; the findings have implications for policymakers involved in planning systems of care to minimise inappropriate resource mobilisation.

In Hong Kong, the Hospital Authority, the local public healthcare service, and Hong Kong Fire Services Department, the primary EMS provider, jointly launched the Prehospital 12-Lead Electrocardiogram for Chest Pain Protocol on 1 February 2021. The Protocol covers the catchment areas of all EDs in Hong Kong and serves a population of 7.41 million. This study utilised data from a territory-wide audit of the Protocol to determine the diagnostic performance of the PHECG algorithm for STEMI.

This prospective observational study analysed data from the territory-wide audit project regarding the Prehospital 12-Lead Electrocardiogram for Chest Pain Protocol, led by the Hong Kong Hospital Authority Coordinating Committee in Accident and Emergency. The Protocol was designed to include all patients with complaints of chest pain, excluding those <12 years of age; in cardiac arrest; with unmanageable airway or breathing; a Glasgow Coma Scale score of ≤13; a first systolic blood pressure of <90 mm Hg; a respiratory rate of <10 or >29 breaths per minute; or refusal or inability to give consent.

Ambulances were equipped with 12-lead ECG machines capable of automatic algorithm-based diagnosis. The selected machine model was a corpuls3 Monitor and Defibrillator (GS Elektromedizinische Geräte G Stemple GmbH, Kaufering, Germany), with the telemedicine application corpuls.mission (GS Elektromedizinische Geräte G Stemple GmbH, Kaufering, Germany). The selected ECG algorithm was the ECG diagnostic algorithm of the Hannover ECG System (Corscience GmbH & Co KG, Erlangen, Germany).

Upon encountering a patient who met the Protocol’s criteria, the ambulance personnel performed a 12-lead ECG on scene or in the stationary ambulance compartment. The ECG was immediately analysed by the computer algorithm and classified as ‘STEMI’, ‘Not STEMI’, or ‘N/A’ (not interpretable due to suboptimal ECG quality). Additional ECGs were performed as necessary to improve quality. The ECG(s) were tele-transmitted to the ED serving the particular catchment area for reading and interpretation by the ED attending physician. If the ECG was classified as ‘STEMI’ by the computer algorithm, the EMS personnel also directly called to alert the ED. The ED prepared the resuscitation room for patient arrival if the ECG was classified as STEMI by the ED physician.

This study adhered to the STARD (Standards for Reporting of Diagnostic Accuracy Studies) 2015 reporting guideline. Patients with PHECGs performed in accordance with the Protocol throughout Hong Kong were prospectively recruited from 1 October to 31 December 2021.

Prehospital ECGs performed and tele-transmitted during the study period were obtained from corpuls.mission’s online database and matched to clinical data from the Clinical Data Analysis and Reporting System and Accident and Emergency Information System (Information Technology and Health Informatics Division, Hospital Authority, Hong Kong). Electrocardiograms without matching patient data and those classified as ‘N/A’ by the algorithm were excluded from the analysis.

Three reference standards were used to investigate the diagnostic accuracy of the computer algorithm. The first reference standard, the primary outcome, was adjudicated blinded rating of the ECG. Each ECG was de-identified and independently interpreted as ‘STEMI’, ‘Not STEMI’ or ‘Not interpretable’ by two investigators: an emergency physician with ≥5 years of experience in emergency medicine practice and a specialist in Emergency Medicine. Electrocardiograms for which there was disagreement between the interpretations of the two blinded raters were classified according to the blinded interpretation of an adjudicator (a second Emergency Medicine specialist). The diagnosis of STEMI was based on the Fourth Universal Definition of Myocardial Infarction14 and the modified Sgarbossa criteria for left bundle branch block or ventricular paced rhythm.15 16 ST-elevation myocardial infarction mimics17 and STEMI equivalents, according to the 2022 ACC Expert Consensus Decision Pathway on the Evaluation and Disposition of Acute Chest Pain in the Emergency Department,18 were regarded as ‘Not STEMI’. ‘Not interpretable’ ECGs were those with substantial motion artefacts, wavering baseline, or disconnected lead(s); these ECGs were excluded from the analysis.

The second reference standard was the ED attending physician’s diagnosis, which considered the patient’s clinical condition, along with additional ECGs and other investigations performed upon arrival in the ED. Patients without ECGs performed in the ED were excluded from the analysis.

The third reference standard was the diagnosis on hospital discharge from the index admission. We excluded patients who died in the ED without an established diagnosis, who developed STEMI after admission, or were discharged with acknowledgement of medical advice and no definitive diagnosis.

Interrater agreement analysis was performed in three dimensions, namely, between the two blinded raters, between the adjudicated blinded rating and the ED diagnosis, and between the adjudicated blinded rating and the diagnosis on hospital discharge. If there was disagreement between the adjudicated blinded rating and the ED diagnosis, the prehospital and ED ECGs were reviewed by the principal investigator to differentiate between dynamic change or true disagreement. Dynamic change was defined as the lack of ST-segment elevation and ECG criteria fulfilment on the initial PHECG, with subsequent evidence on serial ECG performed in the ED.

False-positive and false-negative ECGs were reviewed and classified by the principal investigator. The following categories of ECG morphology were determined based on criteria described in existing literature: Brugada pattern,19 early repolarisation,20 left bundle branch block or paced rhythm not matching STEMI criteria,15 16 left ventricular hypertrophy,21 pericarditis,22 and ventricular ectopics.23

Continuous variables were presented as mean ± standard deviation and were analysed with the independent t test. Categorical variables were reported as absolute frequencies and percentages and were analysed with the Chi squared test or Fisher’s exact test. Interrater agreement regarding ECG diagnosis was analysed using Cohen’s kappa. Sensitivity, specificity, and predictive values were derived from 2 × 2 contingency tables and analysed with the Chi squared test.

The threshold for statistical significance was regarded as P<0.05. All statistical analyses were performed using SPSS software (Windows version 26.0; IBM Corp, Armonk [NY], US).

During the study period, 2801 PHECGs were performed, one for each patient who presented with chest pain. Of these ECGs, 2437 were matched to electronic patient records. After the exclusion of 103 ECGs classified as ‘N/A’ by the computer algorithm, 2334 ECGs were included in the analysis (Fig 1).

The characteristics of the study population are presented in Table 1. Overall, 62.9% of the patients were men. The mean age of male patients, female patients, and both sexes were 63.9 years, 74.1 years, and 67.7 years, respectively. In total, 83.6% of patients were placed on stretchers upon arrival at the ED. Furthermore, 8.2% of patients were institutionalised in residential homes. Of the ECGs, 42.4% were performed during 0800 to 1559 hours, 35.4% were performed during 1600 to 2359 hours, and 22.3% were performed during 0000 to 0759 hours. A total of 405 (17.4%) PHECGs were classified as STEMI by the algorithm.

The primary outcome was diagnostic accuracy based on the adjudicated blinded rating. The prevalence of STEMI was 5.1% (Table 2). There was good interrater observed agreement (96.9%) between the two blinded ECG assessors. Cohen’s kappa was 0.84 (95% confidence interval [CI]=0.81-0.88; P<0.05) [Table 3]. The algorithm had a sensitivity of 94.6% (95% CI=88.2%-97.8%), specificity of 87.9% (95% CI=86.4%-89.2%), positive predictive value of 29.4% (95% CI=24.8%-34.4%), negative predictive value of 99.7% (95% CI=99.3%-99.9%), positive likelihood ratio of 7.8 (95% CI=6.9-8.8), and negative likelihood ratio of 0.06 (95% CI=0.03-0.13) [all P<0.05] (Table 2).

Secondary outcomes were the algorithm’s diagnostic accuracy with reference to the ED attending physician’s diagnosis and to the diagnosis on hospital discharge.

Substantial agreement was observed between the diagnosis based on the adjudicated blinded rating and these two reference standards. Discrepancies in agreement between the adjudicated blinded rating of ECGs and these two reference standards reflected the presence of dynamic ECG changes. Observed agreement between the adjudicated blinded rating and ED physician’s diagnosis was 97.1%, with Cohen’s kappa of 0.69. Excluding patients with dynamic ECG changes in the ED, the observed agreement was 98.2% and Cohen’s kappa was 0.78. Observed agreement between the adjudicated blinded rating and final discharge diagnosis was 97.5%, with Cohen’s kappa of 0.74. Excluding patients with dynamic ECG changes in the ED, the observed agreement was 98.4% and Cohen’s kappa was 0.80 (Table 3). The diagnostic performance based on the three reference standards and the analysis of interrater agreement are summarised in Tables 2 and 3, respectively.

The 255 false-positive ECGs with the adjudicated blinded rating as the reference standard were reviewed and characterised as shown in Figure 2. The leading causes were early repolarisation (n=97; 38.0%), left bundle branch block (n=40; 15.7%), and tachycardia of >140 beats per minute (n=34; 13.3%). Excluding ECGs with suboptimal quality (classified as ‘N/A’ by the algorithm and ‘Not interpretable’ according to adjudicated blinded rating), false-positive ECGs due to artefacts constituted 8.6% (n=22).

Using the diagnosis on hospital discharge as the reference standard, 22 STEMI cases were missed by the algorithm (Fig 3). Thirteen (59.1%) of the false-negative ECGs were due to the development of dynamic ECD changes in the ED; four (18.2%) of these had subtle ST-segment elevation. ST-segment elevation in lead augmented vector right and the STEMI equivalent morphology of de Winter’s T wave were noted in two (9.1%) ECGs each. One ECG was classified as ‘Not interpretable’ according to the adjudicated blinded rating because of substantial artefacts.

This prospective observational study examined the diagnostic performance of a rule-based PHECG algorithm universally utilised in Hong Kong, based on three levels of reference standards. The primary outcome, adjudicated blinded rating, closely reflects diagnostic performance without the addition of patient clinical history and presentation or any other diagnostic aids. The American Heart Association recommends three levels of PHECG diagnosis, namely, EMS interpretation, computerised algorithm diagnosis, and ECG transmission for remote interpretation.24 However, in healthcare systems such as the Hospital Authority in Hong Kong, EMS are trained to perform but not interpret PHECGs. Thus, it is important to understand reliance on the computerised algorithm using the benchmark of physician-based remote interpretation; these data can guide the establishment and improvement of care systems.

One in eight of the PHECGs in this study showed false-positive results. Considering the fair specificity and positive predictive value of only 29.4% for the automated ECG diagnostic programme, this high false-positive rate reflected limitations in guiding prehospital treatment and streamlining care systems (eg, prehospital diversion to percutaneous coronary intervention–capable centres or prehospital triage for direct transfer to a cardiac catheterisation laboratory). A hybrid two-step ECG interpretation model, involving a physician’s remote (ie, telemedicine-based) interpretation of ECGs that are classified as STEMI by the computerised algorithm, could be adopted to minimise overactivation and ensure prudent use of healthcare resources. Nonetheless, the algorithm exhibited good sensitivity in terms of identifying STEMI patients. Its high negative predictive value allowed STEMI to be reasonably excluded based on ECG results. Although remote PHECG interpretation is considered relatively accurate, it generally results in STEMI misdiagnosis rates of 6% to 8%.13 Therefore, we included secondary outcomes, namely, the algorithm’s diagnostic performance based on the ED attending physician’s diagnosis and the final discharge diagnosis; we assessed interrater agreement between these reference standards. We adopted an operational approach focused on the ‘appropriateness of cardiac catheterisation laboratory activation’, rather than a strictly patient-centred approach based on primary percutaneous coronary intervention findings or cardiac biomarkers.

The inclusion of three reference standards was intended to address the heterogeneous estimates of PHECG diagnostic performance for STEMI in existing literature. Prior studies have been based on various reference standards, including blinded physician rating,25 ED attending physician’s diagnosis,6 26 hospital discharge diagnosis,7 and the appropriateness of coronary angiography activation.8 The results have varied according to STEMI prevalence in the study population, as well as the reference standard, ECG machine, and computerised algorithm. Using ED clinical diagnosis as the reference standard, a single-centre pilot study in Hong Kong by Cheung et al6 utilising the X Series Monitor/Defibrillator and Inovise 12L Interpretive Algorithm (Zoll Medical Corporation, Chelmsford [MA], US) demonstrated a low sensitivity (53.8%) and high specificity (99.6%). Bhalla et al26 utilised LIFEPAK 12 monitors (Physio-Control, Redmond [WA], US) equipped with a Marquette 12SL ECG analysis programme (General Electric Company, Fairfield [CT], US) to evaluate PHECGs from 100 STEMI patients and 100 control participants; they found a similarly low sensitivity (58%) and very high specificity (100%). Bosson et al7 examined ECGs obtained with the LIFEPAK 15 monitor (Physio-Control, Inc, Minneapolis [MN], US) and analysed using the University of Glasgow 12-Lead ECG Analysis Programme (version 27); their results showed 92.8% sensitivity and 98.7% specificity, based on the reference standard of appropriateness for emergency coronary angiography.7 The prevalence of STEMI was much lower in their study than in our study (1.4%7 vs 5.1% [Table 2]) because their dataset also included PHECGs performed for symptoms other than chest pain. Using the same ECG machine model as the aforementioned study,7 Fakhri et al8 tested an automated analysis method with a high-specificity STEMI configuration. In a carefully selected STEMI population, the sensitivity and specificity were 69.8% and 51.5%, respectively, based on discharge diagnosis.8 A meta-analysis conducted by Tanaka et al27 suggested that computer-assisted ECG interpretation had a high pooled specificity (95.4%; 95% CI=87.3%-98.4%) with an acceptable estimated number of false-positive results, whereas the pooled sensitivity was relatively low (85.4%; 95% CI=74.1%-92.3%), for identifying STEMI on PHECG. All of these studies utilised ECG machines and diagnostic algorithms that differed from our method, emphasising that diagnostic performance varies across models; evaluations of specific ECG machines and algorithms should be conducted by individual healthcare systems to suit their operational needs.

The Hannover ECG System algorithm utilised in our study was one of nine computer programmes investigated in the international Common Standards for Quantitative Electrocardiography Diagnostic Study,28 using clinical diagnosis as the reference standard. This statistics-based algorithm exhibited one of the highest sensitivities (79.0%) for detecting myocardial infarction compared with all algorithms combined (72.2%); its sensitivity also was similar to that of the combined independent ratings of eight cardiologists (80.3%). However, its ability to correctly classify normal ECGs (86.6%) was lower than that of the combined ratings of cardiologists (97.1%) and the combined algorithms (96.7%). Our findings are consistent with the results of the Common Standards for Quantitative Electrocardiography Diagnostic Study. The presence of artefacts contributed to 8.6% of false-positive ECGs; this rate could be improved by enhancing ECG technique. The major patterns of misdiagnosis were early repolarisation (38.0%), left bundle branch block (15.7%), and tachycardia of >140 beats per minute (13.3%) [Fig 2]. Artefacts on ECG were responsible for the largest proportion of false-positive ECGs8; they contributed a smaller proportion in our dataset because we excluded ECGs considered ‘Not interpretable’ by the algorithm or blinded raters. Early repolarisation remained a leading cause of false-positive ECGs, and existing consensus papers on early repolarisation may help guide future algorithm development.20 29 Further collaboration with the software provider to optimise the algorithm may enhance its accuracy.

Among cases of STEMI missed by the algorithm using diagnosis on hospital discharge as the reference standard, more than half were caused by ECG changes after patient arrival in the ED. False-negative ECGs due to subtle ST-segment elevation represented only 3.45% of all STEMI patients. Remote physician interpretation of these PHECGs would likely be equivocal and uncertain. It presumably would not be beneficial to adjust the algorithm to correct this margin of error, considering the potential for additional false-positives. However, it might be useful to refine the algorithm for enhanced detection of STEMI equivalents, which were missed in the current cohort.

Although the diagnostic limitations of rule-based algorithms are recognised, Zhao et al9 described an artificial intelligence diagnostic algorithm that showed promising results (96.8% sensitivity and 99% specificity) using coronary angiography findings as the reference standard. The potential role of artificial intelligence in PHECG diagnosis merits further exploration to increase accuracy.

Meyers et al30 proposed a new paradigm of occlusion myocardial infarction (OMI) vs non-OMI, which they compared with the conventional STEMI vs non-STEMI paradigm. Occlusion myocardial infarction refers to type 1 myocardial infarction that involves acute total or near-total occlusion of a major epicardial coronary vessel with insufficient collateral circulation, causing acute infarction. Meyers et al30 showed that 38% of OMI patients did not meet ECG-based STEMI criteria, as stated in the 4th Universal Definition of Myocardial Infarction.14 Compared with OMI patients who met STEMI criteria, patients not meeting the criteria experienced significant delays in cardiac catheterisation but exhibited similar adverse outcome profiles. These findings highlight the need to re-evaluate classification strategies for acute coronary syndrome, with a focus on rapidly recognising this underserved and poorly understood subgroup of patients who would benefit from emergent reperfusion therapy. Future research should emphasise identifying ECG features of OMI beyond the STEMI criteria.

First, 13% of PHECGs were not matched to electronic patient records, resulting in the loss of data for interpretation. Second, during adjudicated blinded rating of the ECGs, STEMI equivalents were not included in the definition of STEMI because the algorithm was not designed to include these characteristics. This exclusion differs from real-world scenarios in which the recognition of STEMI equivalents would prompt ED physicians to implement STEMI management. Third, this study evaluated a single rule-based algorithm combined with a single ECG machine model utilised by a single urban EMS service provider serving a predominantly ethnic Chinese population. Fourth, intraobserver variability was not assessed for each ECG reviewer. Finally, ECGs considered ‘Not interpretable’ by ECG reviewers due to substantial artefacts were excluded from data analysis, which might lead to underestimation regarding the contributions of artefacts to false positivity.

In this territory-wide study, a rule-based PHECG algorithm demonstrated good sensitivity and fair specificity for the diagnosis of STEMI.

Concept or design: JHY Lai, CT Lui.
Acquisition of data: JHY Lai, TWT Chan, BCP Wong.
Analysis or interpretation of data: JHY Lai, CT Lui.
Drafting of the manuscript: JHY Lai.
Critical revision of the manuscript for important intellectual content: CT Lui, TWT Chan, BCP Wong, MSH Tsui, BKA Wan, KL Mok.

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

The research was approved by the New Territories West Cluster Research Ethics Committee of Hospital Authority, Hong Kong (Ref No.: NTWC/REC/21097). The requirement for patient consent was waived by the Committee because the study was conducted within a preexisting prehospital clinical service.

1. HealthyHK, Hong Kong SAR Government. Coronary heart diseases. 2021. Available from: https://www.healthyhk.gov.hk/phisweb/en/chart_detail/24/. Accessed 3 Dec 2022.

2. O’Gara PT, Kushner FG, Ascheim DD, et al. 2013 ACCF/AHA guideline for the management of ST-elevation myocardial infarction: executive summary: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2013;61:485-510. Crossref

3. Ibánez B, James S, Agewall S, et al. 2017 ESC Guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation [in English, Spanish]. Rev Esp Cardiol (Engl Ed) 2017;70:1082. Crossref

4. Kontos MC, Gunderson MR, Zegre-Hemsey JK, et al. Prehospital activation of hospital resources (PreAct) ST-segment-elevation myocardial infarction (STEMI): a standardized approach to prehospital activation and direct to the catheterization laboratory for STEMI recommendations from the American Heart Association’s mission: lifeline program. J Am Heart Assoc 2020;9:e011963. Crossref

5. Brunetti ND, De Gennaro L, Correale M, et al. Prehospital electrocardiogram triage with telemedicine near halves time to treatment in STEMI: a meta-analysis and meta-regression analysis of non-randomized studies. Int J Cardiol 2017;232:5-11. Crossref

6. Cheung KS, Leung LP, Siu YC, et al. Prehospital 12-lead electrocardiogram for patients with chest pain: a pilot study. Hong Kong Med J 2018;24:484-91. Crossref

7. Bosson N, Sanko S, Stickney RE, et al. Causes of prehospital misinterpretations of ST elevation myocardial infarction. Prehosp Emerg Care 2017;21:283-90. Crossref

8. Fakhri Y, Andersson H, Gregg RE, et al. Diagnostic performance of a new ECG algorithm for reducing false positive cases in patients suspected acute coronary syndrome. J Electrocardiol 2021;69:60-4. Crossref

9. Zhao Y, Xiong J, Hou Y, et al. Early detection of ST-segment elevated myocardial infarction by artificial intelligence with 12-lead electrocardiogram. Int J Cardiol 2020;317:223-30. Crossref

10. Goebel M, Vaida F, Kahn C, Donofrio JJ. A novel algorithm for improving the diagnostic accuracy of prehospital STelevation myocardial infarction. Prehosp Disaster Med 2019;34:489-96. Crossref

11. Le May MR, Dionne R, Maloney J, et al. Diagnostic performance and potential clinical impact of advanced care paramedic interpretation of ST-segment elevation myocardial infarction in the field. CJEM 2006;8:401-7. Crossref

12. Ducas RA, Wassef AW, Jassal DS, et al. To transmit or not to transmit: how good are emergency medical personnel in detecting STEMI in patients with chest pain? Can J Cardiol 2012;28:432-7. Crossref

13. Tanguay A, Lebon J, Brassard E, Hébert D, Bégin F. Diagnostic accuracy of prehospital electrocardiograms interpreted remotely by emergency physicians in myocardial infarction patients. Am J Emerg Med 2019;37:1242-7. Crossref

14. Thygesen K, Alpert JS, Jaffe AS, et al. Fourth universal definition of myocardial infarction (2018). J Am Coll Cardiol 2018;72:2231-64. Crossref

15. Meyers HP, Limkakeng AT Jr, Jaffa EJ, et al. Validation of the modified Sgarbossa criteria for acute coronary occlusion in the setting of left bundle branch block: a retrospective case-control study. Am Heart J 2015;170:1255-64. Crossref

16. Smith SW, Dodd KW, Henry TD, Dvorak DM, Pearce LA. Diagnosis of ST-elevation myocardial infarction in the presence of left bundle branch block with the ST-elevation to S-wave ratio in a modified Sgarbossa rule. Ann Emerg Med 2012;60:766-76. Crossref

17. Wang K, Asinger RW, Marriott HJ. ST-segment elevation in conditions other than acute myocardial infarction. N Engl J Med 2003;349:2128-35. Crossref

18. Writing Committee; Kontos MC, de Lemos JA, et al. 2022 ACC Expert Consensus Decision Pathway on the evaluation and disposition of acute chest pain in the emergency department: a report of the American College of Cardiology Solution Set Oversight Committee. J Am Coll Cardiol 2022;80:1925-60. Crossref

19. Wilde AA, Antzelevitch C, Borggrefe M, et al. Proposed diagnostic criteria for the Brugada syndrome: consensus report. Circulation 2002;106:2514-9. Crossref

20. Patton KK, Ellinor PT, Ezekowitz M, et al. Electrocardiographic early repolarization: a scientific statement from the American Heart Association. Circulation 2016;133:1520-9. Crossref

21. Armstrong EJ, Kulkarni AR, Bhave PD, et al. Electrocardiographic criteria for ST-elevation myocardial infarction in patients with left ventricular hypertrophy. Am J Cardiol 2012;110:977-83. Crossref

22. Bischof JE, Worrall C, Thompson P, Marti D, Smith SW. ST depression in lead aVL differentiates inferior ST-elevation myocardial infarction from pericarditis. Am J Emerg Med 2016;34:149-54. Crossref

23. Mond HG, Haqqani HM. The electrocardiographic footprints of ventricular ectopy. Heart Lung Circ 2020;29:988-99. Crossref

24. Ting HH, Krumholz HM, Bradley EH, et al. Implementation and integration of prehospital ECGs into systems of care for acute coronary syndrome: a scientific statement from the American Heart Association Interdisciplinary Council on Quality of Care and Outcomes Research, Emergency Cardiovascular Care Committee, Council on Cardiovascular Nursing, and Council on Clinical Cardiology. Circulation 2008;118:1066-79. Crossref

25. Wilson RE, Kado HS, Percy RF, et al. An algorithm for identification of ST-elevation myocardial infarction patients by emergency medicine services. Am J Emerg Med 2013;31:1098-102. Crossref

26. Bhalla MC, Mencl F, Gist MA, Wilber S, Zalewski J. Prehospital electrocardiographic computer identification of ST-segment elevation myocardial infarction. Prehosp Emerg Care 2013;17:211-6. Crossref

27. Tanaka A, Matsuo K, Kikuchi M, et al. Systematic review and meta-analysis of diagnostic accuracy to identify ST-segment elevation myocardial infarction on interpretations of prehospital electrocardiograms. Circ Rep 2022;4:289-97. Crossref

28. Willems JL, Abreu-Lima C, Arnaud P, et al. The diagnostic performance of computer programs for the interpretation of electrocardiograms. N Engl J Med 1991;325:1767-73. Crossref

29. Macfarlane PW, Antzelevitch C, Haissaguerre M, et al. The early repolarization pattern: a consensus paper. J Am Coll Cardiol 2015;66:470-7. Crossref

30. Meyers HP, Bracey A, Lee D, et al. Comparison of the ST-elevation myocardial infarction (STEMI) vs. NSTEMI and occlusion MI (OMI) vs. NOMI paradigms of acute MI. J Emerg Med 2021;60:273-84. Crossref

Non-visualisation of the fetal gallbladder (NVFGB) in the second trimester is a rare condition affecting 0.1% of pregnancies.1 It might be a transient finding—the gallbladder is visible later in pregnancy or after birth in 70% of cases.2 Persistent NVFGB may be associated with benign conditions (eg, gallbladder agenesis); it may also be a manifestation of serious disorders, such as biliary atresia (BA), cystic fibrosis, or chromosomal abnormalities.3 Although chromosomal abnormalities and cystic fibrosis can be identified prenatally via chromosomal microarray (CMA) and sequencing of the cystic fibrosis transmembrane conductance regulator gene, respectively, it remains challenging to diagnose BA before birth because no diagnostic prenatal test is currently available. Biliary atresia is a devastating condition and is the leading indication for liver transplantation in childhood.4 Prenatal suspicion of BA allows prompt postnatal assessment and early diagnosis, permitting timely intervention via Kasai hepatoportoenterostomy and resulting in improved outcomes.4

The measurement of gamma-glutamyl transferase (GGT) in amniotic fluid (AF) has been suggested as a method for prenatal detection of BA. This transferase is secreted by the fetal biliary tract, passes into the intestines, and is ultimately excreted into the amniotic cavity. It becomes detectable in AF at around 14 weeks of gestation upon maturation of the intestinal villi and opening of the cloacal membrane.5 6 7 Biliary tract obstructions, such as BA, hinder the passage of GGT into the intestines and AF, leading to reduced levels of amniotic fluid gamma-glutamyl transferase (AFGGT). Muller et al5 first reported extremely low AFGGT levels in three fetuses with extrahepatic bile duct obstruction at 18 to 19 weeks of gestation; they concluded that a low AFGGT level could be a useful indicator of BA. Subsequently, Burc et al,6 Dreux et al,8 and Bardin et al9 reported similar findings. Nevertheless, existing platforms for analysis of GGT in serum or plasma samples have not been thoroughly validated for use with AF samples. Furthermore, an appropriate reference range, essential for the interpretation of AFGGT results, is difficult to establish due to the limited availability of AF samples from normal pregnancies. Thus, publications regarding gestational age–specific reference ranges for AFGGT have been scarce.

Burc et al6 established reference values for five AF enzymes (AFEs), including GGT, using the Hitachi 911 analyser (Roche Diagnostics). Despite a large sample size of 508, no separate reference range was constructed for each gestational age from 20 to 24 weeks, limiting clinical use of the findings.6 Bardin et al7 established another reference range for AFGGT, from 16 to 22 weeks, using the Integra 800 analyser (Roche). However, the numbers of samples at weeks 16, 21, and 22 were small and had relatively large standard deviations, precluding clinical application.7 Both reference ranges were derived from Caucasian populations. Because the GGT level in adult blood varies according to ethnicity,10 it is likely that AFGGT levels also vary according to ethnicity; thus, there is a need to establish a local AFGGT reference range. Furthermore, both previous reference ranges covered the 5th to 95th percentiles. A wider reference range, from the 2.5th to 97.5th percentiles, would allow greater flexibility in selecting cut-off values for clinical application.

In this study, we aimed to validate a serum/plasma matrix–based GGT assay for AF samples, establish a local gestational age–specific reference range for AFGGT (from the 2.5th to 97.5th percentiles), and evaluate the efficacy of AFGGT for predicting fetal BA in pregnancies with NVFGB using the constructed reference range.

This retrospective study used archived AF supernatant obtained from amniocentesis procedures that had been conducted to exclude fetal chromosomal abnormalities between January 2012 and December 2021. All amniocenteses were performed under ultrasound guidance with aseptic technique by fetal-maternal medicine specialists who also performed detailed ultrasound examinations to document the presence or absence of fetal structural abnormalities. All AF samples were centrifuged at 100 g for 10 minutes; 1 mL of the resulting supernatant was stored at -80℃ in plain aliquots without additives (Axygen; BioGene, Union City [CA], United States), whereas the pellet was used for either CMA and/or karyotyping by G-banding analysis. The cytogenetic laboratory information system recorded the ultrasound findings, indication for amniocentesis, gestational age at amniocentesis, and CMA and/or karyotype results. Pregnancy and fetal outcomes, including the presence of any abnormalities at birth or autopsy findings, were recorded in each patient’s electronic records.

Archived AF samples with known fetal and pregnancy outcomes, taken at 16⁺⁰ to 22⁺⁶ weeks of gestation from singleton pregnancies with a euploid fetus (confirmed by CMA or karyotype) during the period from January 2018 to December 2020, were retrieved to establish a gestational age–specific reference range for GGT. Pregnancies with fetal chromosomal, gastrointestinal, or hepatobiliary anomalies (particularly BA) polyhydramnios, or oligohydramnios were excluded to eliminate potential confounding effects of these conditions.

Non-visualisation of the fetal gallbladder was incidentally detected during fetal morphology scans in pregnancies with risk factors for fetal abnormalities because the fetal gallbladder was not assessed in low-risk routine anomaly scans, in accordance with the International Society of Ultrasound in Obstetrics and Gynecology guideline.11 It was defined as failure to visualise the fetal gallbladder on two targeted ultrasound examinations performed 1 week apart. Isolated NVFGB was defined as NVFGB in the absence of other abnormal ultrasound findings. Pregnant women were counselled regarding possible differential diagnoses and offered amniocentesis for chromosomal analysis, as well as repeated ultrasound scans until the fetal gallbladder was visible. After birth, babies with persistent NVFGB were referred to paediatricians for hepatobiliary ultrasound and liver function tests. If the gallbladder was visible during prenatal scans, paediatricians did not order further tests in the absence of clinical suspicion. We followed the progress of all babies until the time of writing, using electronic hospital records for those delivered in public hospitals and phone calls for those delivered in private hospitals. With parental consent, post-mortem examinations were arranged in pregnancies terminated for serious associated fetal abnormalities.

Amniotic fluid samples were removed from -80℃ storage in batches, thawed, and equilibrated to room temperature immediately prior to analysis. Gamma-glutamyl transferase levels were determined using an International Federation of Clinical Chemistry and Laboratory Medicine–standardised L-gamma-glutamyl-3-carboxy-4-nitroanilide (GGCNA) enzymatic colorimetric assay on a Cobas c502 analyser (Roche, Basel, Switzerland). Internal quality controls were performed before and after each batch.

Details of the AFGGT assay validation and analytical performance, including matrix effects; linearity; intra- and inter-run precision at various GGT levels; interference due to haemolysis, icterus, and lipaemia; and sample stability, are summarised in online supplementary Appendix 1.

An AFGGT reference range was developed using the Generalised Additive Models for Location (μ), Scale (υ) and Shape (σ) [GAMLSS] package in R statistical software (version 3.3.2; R Foundation for Statistical Computing, Vienna, Austria). All GGT values were transformed to their natural log equivalent before model construction; the final model balanced between percentile smoothness, goodness-of-fit, and simplicity. Model fit was assessed using the generalised Akaike information criterion and by inspection of residuals with quantile-quantile plots for all measurements, detrending of quantile-quantile plots, and comparison of empirical percentiles to fitted percentiles. Empirical percentiles were determined for comparative purposes by grouping GGT levels according to gestational age (in weeks).

The final model was used to determine reference values for the 2.5th, 5th, 50th, 95th (z = ± 1.645), and 97.5th percentiles (z = ± 1.964). Percentiles were determined using the expression μ × (1 + z_pυσ)1/υ, where z_p is the percentile of interest and μ, υ, and σ are dependent on the time covariate, gestational age. The GGT reference range was constructed using R statistical software and Microsoft Excel (Microsoft Corporation, Redmond [WA], United States).

We estimated a priori that 293 AFGGT measurements were needed to achieve a standard error of 10% of the gestational age–specific standard deviation for the 2.5th and 97.5th (z=1.96) reference percentiles, assuming that the standard error of the percentile of interest is expressed as a multiple of standard deviation using the formula , where each gestational age between 16⁺⁰ and 22⁺⁶ weeks has a minimum of 42 measurements.

Measured levels of AFGGT in pregnancies with NVFGB were transformed to their gestational age–specific percentile values using the final model. We then determined the positive predictive value (PPV) and negative predictive value (NPV) of AFGGT level <2.5th percentile for identifying BA in euploid pregnancies with NVFGB.

The performance of the GGT assay using AF is summarised in online supplementary Appendix 2. Validation studies indicated that the GGCNA enzymatic colorimetric assay for determination of GGT activity in AF had linearity, precision, recovery, interference profiles, and stability comparable to the values reported for measurement of GGT in plasma and serum samples. The verified analytical measurement range was 10 to 1200 U/L. No significant interference was observed in the presence of 0.25 g/dL haemoglobin, 103 μmol/L bilirubin, or 16.8 mmol/L triglyceride. The limits of haemolysis/icterus/lipaemia indices above which interference occurred were significantly higher than the degrees of those indices in all analysed samples.

A database search identified 518 stored amniotic samples (502 [97%] Chinese and 16 [3%] other Asian ethnicities) suitable for use in establishing a local gestational age–specific AFGGT reference range. The median number of samples per week was 65; there were 2 weeks with <42 samples (27 and 28 samples in the 16th and 19th week of gestation, respectively). The online supplementary Table lists the regressed values of AFGGT levels according to percentile for each week of gestation.

The simplest best-fit model indicated a linear relationship between natural log-transformed GGT and gestational age. Table 1 and the online supplementary Figure show the final smoothing equations for median, coefficient of variation, and skewness, along with gestational age–specific smoothed percentile curves and values of AFGGT for our local population. Residuals of the final model had a mean skewness of 0 and a variance of 1; they were almost mesokurtic, with kurtosis values close to 3.

The database search identified 32 pregnancies with NVFGB from 2012 to 2021, of which 18 had an available AF sample (four isolated NVFGB and 14 non-isolated NVFGB). There were no cases of cystic fibrosis. Nine cases were excluded from analysis, including five with chromosomal abnormalities, one with renal hypoplasia and oligohydramnios, one with hydrops and polyhydramnios, and two in which the biliary tract anatomy could not be identified. The nine remaining cases for analysis included four with isolated NVFGB and five with non-isolated NVFGB (Table 2). The median gestational age at amniocentesis was 21.0 weeks (interquartile range=20.7-22.1).

The Figure depicts the AFGGT levels in the nine pregnancies for analysis compared with our local gestational age–specific AFGGT reference range. Four cases had AFGGT level <2.5th percentile, including one with BA (AFGGT level of 27 U/L at 20⁺³ weeks) and three with transient non-visualisation (AFGGT levels of 32, 68, and 37 U/L at 20⁺⁰, 20⁺⁶, and 22⁺² weeks, respectively). Five had AFGGT level ≥2.5th percentile, including one with gallbladder agenesis and four with transient non-visualisation (Table 2). Using AFGGT level <2.5th percentile as the cut-off, the NPV and PPV for identifying fetal BA in pregnancies with NVFGB were 100% and 25% (95% confidence interval=0, 53), respectively (Table 3). Repeated analysis using AFGGT level <5th percentile or the reference ranges from Burc et al6 or Bardin et al7 yielded identical results.

Biliary atresia is a rare congenital anomaly, with a prevalence of 1 in 15 000 to 20 000 live births among Caucasian populations.12 However, BA is more common in East Asians; the prevalence is 1 in 5000 to 7000 among Chinese populations.13 14 Untreated BA is a progressive and devastating disease that can cause cirrhosis and death by 2 years of age.14 This outcome can be prevented by early intervention via palliative Kasai hepatoportoenterostomy, which is essential for re-establishing biliary drainage. If biliary drainage cannot be re-established, liver transplantation is necessary. Indeed, BA is the most common indication for liver transplantation in children, contributing to 75% of transplantations in children before 2 years of age.12 It is therefore imperative to diagnose BA early, preferably during the prenatal period. Nevertheless, prenatal diagnosis of BA is challenging because ultrasound cannot directly examine the patency of fetal bile ducts. When NVFGB is associated with a hepatic hilar cyst or heterotaxy, it is highly suggestive of BA.15 Non-visualisation of the fetal gallbladder with a hepatic hilar cyst is an indicator of cystic BA, a rare subtype representing 5% to 10% of BA cases16; therefore, this prenatal combination is uncommon. Heterotaxy is another rare condition with a prevalence of 1 in 10 000 live births,17 and concurrent BA is present in only 10.4% of left atrial isomerism cases18; therefore, this prenatal combination is even less common. Consequently, NVFGB may be the only prenatal sign indicating the possibility of BA. However, in cases of isolated NVFGB, it is difficult to differentiate between BA and gallbladder agenesis. The discovery of the association between low AFGGT levels and fetal BA has led to interest regarding the role of AFGGT in the management of NVFGB.

To our knowledge, this study is the first to validate the International Federation of Clinical Chemistry and Laboratory Medicine–standardised GGCNA enzymatic colorimetric assay on the Cobas c502 analyser, a common locally available plasma and serum analyser, for use with AF. We have demonstrated that accurate and precise measurements of GGT can be achieved with AF samples, enabling adoption in clinical settings. We have also established that the analytical measurement range is 10 to 1200 U/L. This is particularly important for an AFGGT assay because AFGGT levels in euploid pregnancies can vary across multiple orders of magnitude, whereas plasma GGT levels in healthy individuals usually remain below 100 U/L. This verification of the lower limit of quantification and linearity range improves confidence in our measurements. Furthermore, we have excluded potential interference, particularly from haemoglobin, because AF samples may sometimes contain maternal blood. We were initially concerned about GGT stability because some samples had been stored for several years; however, consistent with the World Health Organization report that GGT is stable for years in frozen serum and plasma samples,19 we found that AF stored frozen in plain bottles without additives at -20℃ or -80℃ remained stable for at least 6 months (online supplementary Appendix 1). This finding indicates that supernatants can be stored and subsequently retrieved for GGT assays, an important consideration if amniocentesis is performed in the early second trimester but an indication for AFGGT testing is identified during a mid-trimester morphology scan.

We have established a reference range for AFGGT levels at 16⁺⁰ to 22⁺⁶ weeks of gestation using a large local reference population of 518 samples (all Asian, 97% Chinese); this reference population is the largest compared with similar previous publications. The presence of an adequate sample size for each week of gestation allowed us to establish a reference range for each gestational age, thus overcoming the aforementioned limitations regarding clinical use of the two previous reference ranges.6 7 With respect to the two previous reference ranges, the 5th, 50th, and 95th percentiles in our study are similar to those reported by Bardin et al7 but higher at most gestational ages than those reported by Burc et al.6 The larger sample size in our study permitted the calculation of the 2.5th and 97.5th percentiles, such that a 95% central reference range could be established; this allows greater flexibility in selecting cut-off values for clinical application.

It has been reported that AFGGT levels are increased in oesophageal atresia and duodenal atresia, whereas they are decreased in anal atresia without fistula.20 21 Although duodenal atresia can easily be diagnosed prenatally by detection of the double bubble sign, this sign usually appears after 24 weeks of gestation.22 Prenatal diagnosis of oesophageal atresia relies on the indirect sign of non-visualisation of the stomach bubble and polyhydramnios; the sensitivities of these ultrasound findings range from 8.9% to 42%.20 Additionally, prenatal detection of anal atresia without fistula depends on the absence of the perianal muscular complex, but the anal sphincter does not fully mature until after 28 weeks of gestation.23 Further research regarding the use of AFGGT for early detection of these congenital gastrointestinal tract obstructions is valuable, and the availability of a local gestational age–specific reference range is crucial for supporting such research.

In the present study, the NPV and PPV of AFGGT level <2.5th percentile for identifying fetal BA in NVFGB were 100% and 25%, respectively (Table 3). Bardin et al9 assessed the efficacy of low AFGGT levels in predicting BA among cases of NVFGB between 17 and 22 weeks of gestation. Of the 26 cases with AFGGT level ≥5th percentile, none had BA; of the four cases with AFGGT level <5th percentile, three had BA. The corresponding NPV and PPV were 100% and 75%, respectively. The PPV in their study may have been higher because amniocentesis was performed before 22 weeks of gestation in all cases.9 In our cohort, after exclusion of the two cases in which amniocentesis was performed after 22 weeks of gestation, the PPV only marginally improved to 33.3%. Our figures are more consistent with those of Dreux et al,8 who analysed the efficacy of AFEs for predicting BA in NVFGB before and after 22 weeks of gestation. In that study, abnormal AFE was defined as GGT and/or intestinal alkaline phosphatase level <0.5 multiples of the median.8 Before 22 weeks of gestation, there were three cases of BA among seven cases with abnormal AFE and no cases of BA among 16 cases with normal AFE. The corresponding NPV and PPV were 100% and 43%, respectively. However, after 22 weeks of gestation, there was only one case of BA among six cases with abnormal AFE and four cases of BA among 56 cases with normal AFE. The corresponding NPV and PPV were 93% and 17%, respectively (Table 3). These findings confirmed the expected decrease in AFE efficacy for predicting BA after 22 weeks of gestation. By that gestational age, the passage of GGT from the intestine into the AF is impeded by mature anal sphincter muscles in normal fetuses; thus, the AFGGT level is very low after 22 weeks of gestation, and it is difficult to distinguish between a low level due to BA and a low level related to normal development.5 6 7

In our cohort, there were three fetuses without BA who had AFGGT level ≤2.5th percentile; one of these fetuses had an extremely low AFGGT level (Case 4) [Table 2]. In addition to its association with fetal BA, low AFGGT is linked to chromosomal abnormalities, cystic fibrosis, anal atresia without fistula, and polyhydramnios. However, none of the three fetuses had any of these conditions. Although one fetus (Case 2) [Table 2] had a small choledochal cyst which might have impeded biliary drainage and caused a mild decrease in AFGGT, we could not find a potential explanation for the low AFGGT levels in the other two fetuses.

Limitations of our study include its retrospective nature and small cohort size. The incidence of NVFGB is low (0.1%).1 In the largest systematic review concerning the outcomes of NVFGB, encompassing seven studies, the total number of cases was 280; among 170 cases of isolated NVFGB, only six (3.5%) had BA.2 In Hong Kong, the fetal gallbladder is not routinely assessed in low-risk routine anomaly scans, in accordance with the International Society of Ultrasound in Obstetrics and Gynecology guideline.11 24 Among the 32 cases of NVFGB in our cohort, one (3.1%) had BA; this incidence is comparable to the findings in the aforementioned review. However, only 18 cases had an AF sample available for AFGGT testing; after the exclusion of cases with abnormalities that might affect the AFGGT level, we included nine cases in the analysis. Because there was only one case of BA in our cohort and 11 more were described in the literature (Table 3), clinical application of these research findings requires caution, as well as careful pre- and post-test counselling.

In addition to the one case of BA in this cohort, two additional cases of BA were not included in this study because prenatal ultrasound scans did not indicate NVFGB. The AFGGT levels in all three cases were <30 U/L and <2.5th percentile (27, 28, and 29 U/L at 20⁺³, 21⁺⁵, and 21⁺⁶ weeks, respectively), whereas the AFGGT levels in all samples used to establish our reference range were >30 U/L. Thus, an AFGGT level <30 U/L may be a useful absolute cut-off for the prediction of BA. Notably, all three cases of extrahepatic bile duct obstruction reported by Muller et al5 also had an AFGGT level <30 U/L (20 U/L for all three cases [<1st percentile]). Further research with a larger cohort is required to confirm the efficacy of using an AFGGT level <30 U/L as the absolute cut-off for predicting BA in NVFGB.

Based on the present findings and published literature, we conclude that AFGGT testing is useful for the exclusion of fetal BA in pregnancies with NVFGB. With a consistent NPV of 100% across all published series, AFGGT level ≥2.5th percentile can provide reassurance for parents that the fetus is unlikely to have BA. However, considering its PPV of 33.3% to 75% before 22 weeks of gestation and 17% to 43% after 22 weeks of gestation, AFGGT level <2.5th percentile cannot be considered diagnostic for BA. Instead, it serves as a warning sign, indicating the need for prompt postnatal investigation of possible BA. Because NVFGB is also associated with chromosomal abnormalities, amniocentesis is recommended; the advantages of detecting underlying chromosomal abnormalities by CMA and excluding BA through AFGGT testing likely outweigh the 0.3% risk of procedure-related miscarriage.25 Follow-up prenatal ultrasound scans to visualise the fetal gallbladder should be arranged. Paediatricians should also be alerted for prompt postnatal assessment to facilitate early detection of BA. Timely performance of the Kasai operation can reduce the need for liver transplantation in childhood and improve the rate of overall survival into adulthood to 90%.26 27

Concept or design: YH Ting, DS Sahota, FCK Wong, TYT Cheung.
Acquisition of data: TYT Cheung, SR Subramaniam, FCK Wong, MHM Chan, YH Ting, NKL Wong, SL Lau, X Zhu, WT Lui, EKW Chan, YKY Kwok, KW Choy, TY Leung.
Analysis or interpretation of data: TYT Cheung, DS Sahota, FCK Wong, YH Ting, NKL Wong.
Drafting of the manuscript: YH Ting, TYT Cheung, DS Sahota, NKL Wong, FCK Wong, SR Subramaniam.
Critical revision of the manuscript for important intellectual content: DS Sahota, YH Ting, TYT Cheung, FCK Wong, NKL Wong.

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

This research was approved by the Joint Chinese University of Hong Kong—New Territories East Cluster Clinical Research Ethics Committee, Hong Kong (Ref No.: CRE 2020.060). Informed consent for amniocentesis in the current study and storage and use of excess amniotic fluid in future research was obtained from the patients at the time of amniocentesis.

The supplementary material was provided by the authors and some information may not have been peer reviewed. Accepted supplementary material will be published as submitted by the authors, without any editing or formatting. Any opinions or recommendations discussed are solely those of the authors and are not endorsed by the Hong Kong Academy of Medicine and the Hong Kong Medical Association. The Hong Kong Academy of Medicine and the Hong Kong Medical Association disclaim all liability and responsibility arising from any reliance placed on the content.

1. Blazer S, Zimmer EZ, Bronshtein M. Nonvisualization of the fetal gallbladder in early pregnancy: comparison with clinical outcome. Radiology 2002;224:379-82. Crossref

2. Di Pasquo E, Kuleva M, Rousseau A, et al. Outcome of non-visualization of fetal gallbladder on second-trimester ultrasound: cohort study and systematic review of literature. Ultrasound Obstet Gynecol 2019;54:582-8. Crossref

3. Ting YH, So PL, Cheung KW, Lo TK, Ma TW, Leung TY. Non-visualisation of fetal gallbladder in a Chinese cohort. Hong Kong Med J 2022;28:116-23. Crossref

4. Parolini F, Boroni G, Milianti S, et al. Biliary atresia: 20-40-year follow-up with native liver in an Italian centre. J Pediatr Surg 2019;54:1440-4. Crossref

5. Muller F, Gauthier F, Laurent J, Schmitt M, Boué J. Amniotic fluid GGT and congenital extrahepatic biliary damage. Lancet 1991;337:232-3. Crossref

6. Burc L, Guibourdenche J, Luton D, et al. Establishment of reference values of five amniotic fluid enzymes. Analytical performances of the Hitachi 911. Application to complicated pregnancies. Clin Biochem 2001;34:317-22. Crossref

7. Bardin R, Danon D, Tor R, Mashiach R, Vardimon D, Meizner I. Reference values for gamma-glutamyl-transferase in amniotic fluid in normal pregnancies. Prenat Diagn 2009;29:703-6. Crossref

8. Dreux S, Boughanim M, Lepinard C, et al. Relationship of non-visualization of the fetal gallbladder and amniotic fluid digestive enzymes analysis to outcome. Prenat Diagn 2012;32:423-6. Crossref

9. Bardin R, Ashwal E, Davidov B, Danon D, Shohat M, Meizner I. Nonvisualization of the fetal gallbladder: can levels of gamma-glutamyl transpeptidase in amniotic fluid predict fetal prognosis? Fetal Diagn Ther 2016;39:50-5. Crossref

10. Stewart SH, Connors GJ, Hutson A. Ethnicity and gamma-glutamyltransferase in men and women with alcohol use disorders. Alcohol Alcohol 2007;42:24-7. Crossref

11. Salomon LJ, Alfirevic Z, Berghella V, et al. Practice guidelines for performance of the routine mid-trimester fetal ultrasound scan. Ultrasound Obstet Gynecol 2011;37:116-26. Crossref

12. Hartley JL, Davenport M, Kelly DA. Biliary atresia. Lancet 2009;374:1704-13. Crossref

13. Hsiao CH, Chang MH, Chen HL, et al. Universal screening for biliary atresia using an infant stool color card in Taiwan. Hepatology 2008;47:1233-40. Crossref

14. Zhan J, Chen Y, Wong KK. How to evaluate diagnosis and management of biliary atresia in the era of liver transplantation in China. World J Paediatr Surg 2018;1:e000002. Crossref

15. Chalouhi GE, Muller F, Dreux S, Ville Y, Chardot C. Prenatal non-visualization of fetal gallbladder: beware of biliary atresia! Ultrasound Obstet Gynecol 2011;38:237-8. Crossref

16. Rahamtalla D, Al Rawahi Y, Jawa ZM, Wali Y. Cystic biliary atresia in a neonate with antenatally detected abdominal cyst. BMJ Case Rep 2022;15:e246081. Crossref

17. Lin AE, Ticho BS, Houde K, Westgate MN, Holmes LB. Heterotaxy: associated conditions and hospital-based prevalence in newborns. Genet Med 2000;2:157-72. Crossref

18. Gottschalk I, Stressig R, Ritgen J, et al. Extracardiac anomalies in prenatally diagnosed heterotaxy syndrome. Ultrasound Obstet Gynecol 2016;47:443-9. Crossref

19. Ehret W. Use of Anticoagulants in Diagnostic Laboratory Investigations. Geneva: World Health Organization; 1999.

20. Czerkiewicz I, Dreux S, Beckmezian A, et al. Biochemical amniotic fluid pattern for prenatal diagnosis of esophageal atresia. Pediatr Res 2011;70:199-202. Crossref

21. Muller C, Czerkiewicz I, Guimiot F, et al. Specific biochemical amniotic fluid pattern of fetal isolated esophageal atresia. Pediatr Res 2013;74:601-5. Crossref

22. The Fetal Medicine Foundation. Duodenal atresia. Available from: https://fetalmedicine.org/education/fetal-abnormalities/gastrointestinal-tract/duodenal-atresia. Accessed 19 Jan 2023.

23. Ochoa JH, Chiesa M, Vildoza RP, Wong AE, Sepulveda W. Evaluation of the perianal muscular complex in the prenatal diagnosis of anorectal atresia in a high-risk population. Ultrasound Obstet Gynecol 2012;39:521-7. Crossref

24. Chen M, Leung TY, Sahota DS, et al. Ultrasound screening for fetal structural abnormalities performed by trained midwives in the second trimester in a low-risk population—an appraisal. Acta Obstet Gynecol Scand 2009;88:713-9. Crossref

25. Salomon LJ, Sotiriadis A, Wulff CB, Odibo A, Akolekar R. Risk of miscarriage following amniocentesis or chorionic villus sampling: systematic review of literature and updated meta-analysis. Ultrasound Obstet Gynecol 2019;54:442-51. Crossref

26. Serinet MO, Wildhaber BE, Broué P, et al. Impact of age at Kasai operation on its results in late childhood and adolescence: a rational basis for biliary atresia screening. Pediatrics 2009;123:1280-6. Crossref

27. Davenport M. Biliary atresia: clinical aspects. Semin Pediatr Surg 2012;21:175-84. Crossref

Cardiovascular disease (CVD) constitutes a spectrum of diseases that affect the heart and blood vessels. Worldwide, CVD is the leading cause of death as well as a major cause of premature death and chronic disability in numerous regions.1 In 2017, CVD was responsible for an estimated 17.8 million deaths, representing 31% of all global deaths.2 Hypertension, smoking status, hyperlipidaemia, and diabetes mellitus are prominent risk factors for CVD.3 The global prevalence of CVD risk factors is increasing. Currently, an estimated 15% of the world’s population (1.13 billion people) has hypertension, and the prevalence is expected to increase to 29% by 2025.4 5 Additionally, the global prevalence of diabetes increased from 211 million in 1990 to 476 million in 2017, representing a 129.7% increase in nearly three decades.6 The prevalences of CVD and its risk factors are expected to continue to increase in the near future due to industrialisation and population ageing. Fortunately, the World Health Organization (WHO) has estimated that premature CVD is preventable in >75% of cases, and risk factor amelioration can help reduce the burden caused by CVD.7

The risk of CVD over the next 10 years for an individual can be estimated using prediction models. Cardiovascular disease risk prediction is important at the individual and population levels. At the individual level, CVD risk prediction allows primary care medical professionals to identify high-risk patients. Risk factors for CVD, such as hypertension, can be treated accordingly to reduce the patient’s future risk of CVD. At the population level, CVD risk trends allow health policy planners to make evidence-based decisions and review current public health strategies used in CVD prevention.8

Changes in CVD risk can serve as a reference to indicate changes in public health status within a population. Two studies have evaluated changes in the 10-year CVD risk in the United States (US) population. In a study by Ajani and Ford,9 risk models adopted by the National Cholesterol Education Program Adult Treatment Panel III were utilised to estimate coronary heart disease risk, rather than the more precise Framingham formulae for estimation of overall CVD risk. Notably, their analysis lacked age stratification. On the other hand, Lopez-Jimenez et al10 evaluated the changes in CVD risk between 1976 and 2004. To our knowledge, no previous study has evaluated the change in 10-year CVD risk in an Asian population, and insights are needed regarding the cardiovascular health status of the population in a more recent time period.

In this study, we aimed to calculate and compare the changes in sex- and age-specific CVD risks predicted by the Framingham model in a Hong Kong general population by using the Hong Kong Population Health Survey (PHS) 2014/15 in combination with two previous surveys conducted in 2003 to 2005, namely, PHS 2003/2004 and Heart Health Survey (HHS) 2004/2005.

The data in this study were sourced from PHS 2003/200411 and PHS 2014/15.12 Population Health Surveys are territory-wide cross-sectional surveys conducted by the Department of Health of the Hong Kong SAR Government. These surveys target the land-based non-institutional population of individuals aged ≥15 years in Hong Kong, excluding foreign domestic helpers and visitors. Systematic replicate sampling was adopted to select living quarters that were representative of the Hong Kong general population. All domestic households in the selected living quarters and household members in the target population were individually surveyed. Written consent was obtained from individuals who agreed to participate in a PHS. Participants were invited to complete a face-to-face interview, where they provided information about their socio-demographic characteristics, disease status, and daily lifestyle habits. After the interview, consenting participants aged 15 to 84 years were randomly selected to undergo a health examination that included physical measurements and biochemical testing.12 Health examinations for participants of PHS 2003/2004 were conducted as part of the HHS 2004/2005.

The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist was followed for preparing this manuscript.

The main outcome of this study was the predicted risk of CVD over the next 10 years among people aged 30 to 74 years. Thus far, no specific prediction models have been developed for the Hong Kong population. In this study, we adopted the prediction model for primary care developed by the Framingham Heart Study Cohort in the general adult population aged 30 to 74 years.13 The Framingham CVD risk model was validated and can be applied to the Chinese population, but requires recalibration in men.14 A CVD event was defined as a composite of coronary heart disease, cerebrovascular events, peripheral artery disease, and heart failure in the Framingham model. The predicted CVD risk over the next 10 years was calculated for each participant using the following information: age, total cholesterol level, high-density lipoprotein cholesterol level, systolic blood pressure, use of antihypertensive medication, current smoking status, and diabetes status.

A complete-case analysis approach was utilised in this study. Descriptive statistics were used to present the characteristics of included individuals. Sex- and age-specific predicted CVD risks in 2003-2005 and 2014-2015 were calculated to summarise the change in risk according to sex and age-group. Significant differences in the above factors between the two PHSs were tested by independent t test or Chi squared test, as appropriate.

To summarise results at the population level, population weighting established by the Department of Health was applied according to age-group and sex for participants in each PHS. To compare results at the population level, age-standardised predicted CVD risk was calculated using WHO 2000-2025 world standard population data15 and the US population census data in 2000.16

All statistical analyses were performed with SPSS software (Windows version 26.0; IBM Corp, Armonk [NY], US). All significance tests were two-tailed, and P values <0.05 were considered statistically significant.

In total, 1662 and 818 participants were included in the PHS 2014/15 cohort and the PHS 2003/2004 and HHS 2004/2005 cohort, respectively. These individuals represented populations of 4 445 868 and 3 495 074, respectively. Table 1 shows the baseline characteristics of the PHS and HHS cohorts according to Framingham predictors. The mean age was similar in both cohorts (2003-2005: 48.6 years vs 2014-2015: 50.8 years) and the proportions of male and female participants were similar (female participants in 2003-2005: 54.1% vs 2014-2015: 53.0%).

Predicted CVD risk increases with age and is higher in men (Table 2). In PHS 2003/2004 and HHS 2004/2005, the mean CVD risk increased with age in both sexes, from 1.6% among women aged 30-44 years to 17.8% among women aged 65-74 years, and from 5.1% among men aged 30-44 years to 33.4% among men aged 65-74 years. Between the two surveys, there was no significant difference in overall 10-year CVD risk (2003-2005: 10.2% vs 2014-2015: 10.6%; P=0.29). After age standardisation according to WHO world standard population data, the age-standardised predicted CVD risks in the 2003-2005 cohort and the 2014-2015 cohort were 9.4% and 8.8%, respectively (Table 3). The small decrease in predicted CVD risk may be due to the decrease in smokers among men (30.5% vs 24.0%; P<0.001) [Table 1].

The risk of cardiovascular events over the next 10 years was classified as low (CVD risk <10%), medium (CVD risk ≥10% and <20%), and high (CVD risk ≥20%); the distributions of risk groups are presented in Table 4. Among participants aged 30 to 74 years, the risk group distributions were similar in both cohorts. When analysed according to sex, 29.1% of men and 5.1% of women were classified as high-risk in PHS 2014/15, whereas 28.2% of men and 6.4% of women were classified as high-risk in PHS 2003/2004 and HHS 2004/2005. When analysed according to age-group, more participants aged 65 to 74 years were classified as high-risk in PHS 2003/2004 and HHS 2004/2005 (66.8% vs 2014-2015: 53.1%; P=0.028).

Using representative samples from the Hong Kong PHS and HHS, we found that the 10-year CVD risk increases with age and is consistently higher in men (men: 15.5% vs women: 6.2%; P<0.001, in PHS 2014/15) [Table 2]. This trend is consistent with previous reports; for example, an analysis of NHANES (National Health and Nutrition Examination Survey) III data showed that men have a significantly higher CVD risk (11.8% in men vs 5.1% in women) and that CVD risk is higher in older age-groups (16.2% for participants aged 60-74 years vs 11.2% for participants aged 50-59 years).10 Thus, risk management for older adults may represent a challenge to the current health infrastructure. The burden of CVD is directly associated with increased morbidity and mortality in patients, and it translates to substantial healthcare costs. In Hong Kong, the number of people aged ≥65 years is predicted to reach 2.58 million (35.9% of the population) by 2064.17 Thus, there is a critical need to achieve a more comprehensive understanding of the aetiologies associated with CVD in older adults.

The results have extensive public health implications. Although age-standardised rates of death from CVD in Hong Kong greatly decreased from 93.4 per 100 000 standard population in 2001 to 56.0 per 100 000 standard population in 2017,18 there was no significant difference in overall 10-year CVD risk between 2003-2005 and 2014-2015 (10.2% vs 10.6%; P=0.29) [Table 2]. After age standardisation according to WHO world standard population data, a small decrease in CVD risk was observed, from 9.4% in 2003-2005 to 8.8% in 2014-2015 [Table 3]. A study of NHANES data revealed a similar trend, consisting of a small decrease in CVD risk from 1988 to 2004 (7.9% to 7.4%; P<0.001).10 The stagnation in 10-year CVD risk between 2003-2005 and 2014-2015 suggests that despite improvements in treatment, more effective prevention strategies (eg, improvements in diet and physical activity levels) are needed.19 Primary care intervention to manage modifiable risk factors, such as hypertension, dyslipidaemia, and diabetes, can complement population-based policies. Efforts to ensure access to appropriate healthcare and affordable medications will help control abnormal risk factor levels.20 Furthermore, despite the importance of targeting individual risk factors to reduce prevalence, the long-term objective should be reduced overall risk of CVD. This objective requires a holistic approach simultaneously targeting multiple risk factors.

Our results highlight the need for overall risk assessment, in addition to targeted efforts focused on specific CVD risk factors. During the past decade, numerous public health initiatives (eg, smoking cessation campaigns) have been implemented to reduce the prevalence of established risk factors for CVD. It is reasonable to expect that CVD risk would decrease over time in conjunction with the decreased prevalences of various risk factors, such as the declining number of current smokers. Surprisingly, these changes were not strongly reflected in the change in 10-year CVD risk. One explanation is that decreases in the prevalence of some risk factors are offset by increases in others. For example, population ageing in Hong Kong may shift more adults into the high-risk group. However, because the Framingham model was limited to individuals aged 30 to 74 years, the change in mean age between the two surveys is inconsequential. Other possible changes in risk factors include increases in the prevalence of diabetes and hypertension, as reflected by the increased proportion of participants receiving antihypertensive medications (Table 1). Although CVD mortality has considerably decreased, it is concerning that CVD risk has not substantially diminished over the past decade. This lack of improvement in CVD risk may lead to an increasing community burden of CVD in the near future, especially in the context of population ageing.

The overall 10-year CVD risk for participants aged 30 to 74 years in 2003-2005 after age adjustment to US census population data in 2000 was 10.7%. During a similar time period (1999-2004), the US population had a 10-year CVD risk of 7.4%.10 When Hong Kong men were stratified according to risk group in 2014-2015, 48.8% were classified as low-risk, 22.1% were classified as medium-risk, and 29.1% were classified as high-risk (Table 4). For comparison, among men in the United Kingdom population during 2012, 46.5% were classified as low-risk, 39.9% were classified as medium-risk, and 13.6% were classified as high-risk using the National Institute for Health and Care Excellence Framingham risk model.21 In terms of risk distribution, a greater proportion of Hong Kong men were classified as high-risk compared with men in the United Kingdom in the early 2010s. In terms of age-standardised risk, the Hong Kong population had a higher 10-year CVD risk than the US population in 2003.

Analysis according to age showed that the proportion of low-risk participants increased from 2003-2005 to 2014-2015, particularly in the age-groups of 55-64 and 65-74 years (2003-2005: 28.5% vs 2014-2015: 44.7% for participants aged 55-64 years, and 6.1% vs 15.4% for participants aged 65-74 years) [Table 4]. This increased proportion may be the result of primary care clinicians recognising age as a prominent risk factor for CVD, leading to more aggressive treatment of modifiable risk factors. Furthermore, there were fewer male smokers in 2014-2015 than in 2003-2005 (24.0% vs 30.5%; P<0.001) [Table 1], suggesting that the success of recent anti-smoking campaigns also contributed to this paradigm shift.

Contrary to a previous report addressing changes in several CVD risk factors,22 we used the widely validated Framingham risk model to predict the risk of CVD. By measuring the net change in cardiovascular risk, we more comprehensively estimated the impacts of prevention strategies; this is particularly important because some risk factors worsened, some improved, and some remained stable over time. To our knowledge, this is the first study to evaluate the change in 10-year CVD risk in an Asian population. The results reinforce the need for more aggressive community-wide and clinic-based preventions, with an emphasis on increased exercise hours during leisure time, dietary changes that promote higher intake of vegetables and fruits, as well as lower intake of salt and saturated fats, weight maintenance, and smoking cessation. The findings also suggest that the discovery and use of lipid-lowering drugs and antihypertensive medications does not effectively translate into overall risk reduction in an Asian population, unless these treatments are simultaneously paired with prevention strategies targeting CVD risk factors.

This study was based on two Hong Kong population health surveys, and thus the sample is highly representative of the general population. Baseline data were collected through laboratory tests and face-to-face interviews, suggesting that these data are highly reliable. The long interval between the two health surveys (2003-2005 to 2014-2015) also provides insights regarding the effectiveness of current CVD prevention strategies. Limitations of the current study involve its use of the Framingham risk prediction model and the PHS and HHS. Similar to other surveys, the PHS and HHS are susceptible to participation bias because the sample data might not provide an accurate representation of the overall population. Many PHS variables are self-reported and may lead to reporting bias. Considering the effort required to complete the long questionnaires, the collected data may be susceptible to non-response bias and recall bias. Furthermore, because the Framingham cohort primarily consisted of Caucasians, the predicted 10-year CVD risk in the Hong Kong population calculated using the Framingham model should be interpreted cautiously. The Framingham risk model was developed for people without a history of CVD; thus, a small proportion of the overall sample lacking a history of CVD might have been included in the study, causing inaccuracies in the results. Additionally, the model may overestimate the 10-year CVD risk for men in Hong Kong,14 and a recalibrated model with greater predictive power may be required. Although several limitations and assumptions may hinder the prediction of CVD risk, these potential sources of bias were present in both surveys. Therefore, the effects of such biases may be less important when comparing the change in predicted 10-year CVD risk between the two time points.

During the period from 2003-2005 to 2014-2015, the change in predicted 10-year CVD risk was not statistically significant. However, the proportion of low-risk participants within older age-groups was higher in PHS 2014/15 than in PHS 2003/2004 and HHS 2004/2005. More aggressive CVD prevention strategies and primary care interventions are needed to address CVD risk factors.

Concept or design: BMY Cheung, CLK Lam.
Acquisition of data: BYC Sung, EHM Tang, BMY Cheung, CLK Lam.
Analysis or interpretation of data: BYC Sung, EHM Tang.
Drafting of the manuscript: BYC Sung, EHM Tang.
Critical revision of the manuscript for important intellectual content: L Bedford, CKH Wong, ETY Tse, EYT Yu, BMY Cheung, CLK Lam.

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 advisers of the journal, CKH Wong and EYT Yu were not involved in the peer review process. Other authors have disclosed no conflicts of interest.

The authors thank the Department of Health of the Hong Kong SAR Government for granting approval to use the data from Population Health Survey 2003/2004, Heart Health Survey 2004/2005, and Population Health Survey 2014/15 for this research. The authors also thank Dr Weinan Dong and Ms Tingting Wu from the Department of Family Medicine and Primary Care of The University of Hong Kong for their assistance in data interpretation and revising the final draft of the manuscript.

Part of this study was presented as a poster at the ESC Congress 2021 – The Digital Experience (virtual), 27-30 August 2021, and was published as an abstract (Sung BY, Tang EH, Bedford L, et al. Change in Framingham cardiovascular disease risk between 2003 and 2014 in the Hong Kong Population Health Survey [abstract]. Eur Heart J 2021;42:1).

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

The requirement for ethics approval for this research was waived by the Institutional Review Board of The University of Hong Kong/Hospital Authority Hong Kong West Cluster, Hong Kong as this study involved secondary analysis of de-identified governmental data.

1. Roth GA, Johnson C, Abajobir A, et al. Global, regional, and national burden of cardiovascular diseases for 10 causes, 1990 to 2015. J Am Coll Cardiol 2017;70:1-25. Crossref

2. GBD 2017 Causes of Death Collaborators. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2018;392:1736-88. Crossref

3. Romero CX, Romero TE, Shlay JC, Ogden LG, Dabelea D. Changing trends in the prevalence and disparities of obesity and other cardiovascular disease risk factors in three racial/ethnic groups of USA adults. Adv Prev Med 2012;2012:172423. Crossref

4. World Health Organization. Hypertension. 2019. Available from: https://www.who.int/news-room/fact-sheets/detail/hypertension. Accessed 8 Jan 2021.

5. Kearney PM, Whelton M, Reynolds K, Muntner P, Whelton PK, He J. Global burden of hypertension: analysis of worldwide data. Lancet 2005;365:217-23. Crossref

6. Lin X, Xu Y, Pan X, et al. Global, regional, and national burden and trend of diabetes in 195 countries and territories: an analysis from 1990 to 2025. Sci Rep 2020;10:14790. Crossref

7. World Health Organization. Cardiovascular diseases. 2016. Available from: https://www.who.int/health-topics/cardiovascular-diseases#tab=tab_1. Accessed 8 Jan 2021.

8. World Health Organization. Prevention of cardiovascular disease. Guidelines for assessment and management of cardiovascular risk. 2007. Available from: https://apps.who.int/iris/bitstream/handle/10665/43685/9789241547178_eng.pdf?sequence=1&isAllowed=y. Accessed 8 Jan 2021.

9. Ajani UA, Ford ES. Has the risk for coronary heart disease changed among U.S. adults? J Am Coll Cardiol 2006;48:1177-82. Crossref

10. Lopez-Jimenez F, Batsis JA, Roger VL, Brekke L, Ting HH, Somers VK. Trends in 10-year predicted risk of cardiovascular disease in the United States, 1976 to 2004. Circ Cardiovasc Qual Outcomes 2009;2:443-50. Crossref

11. Department of Health, Hong Kong SAR Government; Department of Community Medicine, The University of Hong Kong. Population Health Survey 2003/2004 2015. Available from: https://www.chp.gov.hk/files/pdf/report_on_population_health_survey_2003_2004_en.pdf. Accessed 8 Jan 2021.

12. Surveillance and Epidemiology Branch, Centre for Health Protection, Department of Health, Hong Kong SAR Government. Report of Population Health Survey 2014/2015. 2017. Available from: https://www.chp.gov.hk/files/pdf/dh_phs_2014_15_full_report_eng.pdf. Accessed 8 Jan 2021.

13. D’Agostino RB Sr, Vasan RS, Pencina MJ, et al. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation 2008;117:743-53. Crossref

14. Lee CH, Woo YC, Lam JK, et al. Validation of the Pooled Cohort equations in a long-term cohort study of Hong Kong Chinese. J Clin Lipidol 2015;9:640-6.e2. Crossref

15. World Health Organization. Age standardization of rates: a new WHO standard. 2001. Available from: https://cdn.who.int/media/docs/default-source/gho-documents/global-health-estimates/gpe_discussion_paper_series_paper31_2001_age_standardization_rates.pdf. Accessed 6 May 2024.

16. United States Census Bureau. Introduction to Census 2000 Data Products. Available from: https://usa.ipums.org/usa/resources/voliii/pubdocs/2000/mso-01icdp.pdf. Accessed 1 May 2024.

17. Census and Statistics Department, Hong Kong SAR Government. Hong Kong population projections 2017-2066. 2017. Available from: https://www.censtatd.gov.hk/media_workers_corner/pc_rm/hkpp2017_2066/index.jsp. Accessed 8 Jan 2021.

18. Department of Health, Hong Kong SAR Government. Non-Communicable Diseases Watch. Overview of cardiovascular diseases. September 2018. Available from: https://www.chp.gov.hk/files/pdf/ncd_watch_sep_2018.pdf. Accessed 8 Jan 2021.

19. National Prevention Council, United States Government. National Prevention Strategy. America’s plan for better health and wellness. June 2011. Available from: https://www.hhs.gov/sites/default/files/disease-prevention-wellness-report.pdf. Accessed 8 Jan 2021.

20. Chow CK, Nguyen TN, Marschner S, et al. Availability and affordability of medicines and cardiovascular outcomes in 21 high-income, middle-income and low-income countries. BMJ Glob Health 2020;5:e002640. Crossref

21. Collins GS, Altman DG. Predicting the 10-year risk of cardiovascular disease in the United Kingdom: independent and external validation of an updated version of QRISK2. BMJ 2012;344:e4181. Crossref

22. Carroll MD, Lacher DA, Sorlie PD, et al. Trends in serum lipids and lipoproteins of adults, 1960-2002. JAMA 2005;294:1773-81. Crossref

The evolution of optimal axillary management for breast cancer patients has led to emphasis on the de-escalation of axillary surgery and minimisation of surgical morbidity. Favourable results from the American College of Surgeons Oncology Group (ACOSOG) Z0011 phase 3 randomised clinical trials have redefined the indications for completion axillary lymph node dissection (ALND) in patients with positive sentinel lymph nodes (SLNs). Early-stage breast cancer patients who undergo upfront breast-conserving surgery and have one or two positive SLNs can safely forgo ALND while maintaining good overall survival and disease-free survival.1 2 Consequently, the ASCO (American Society of Clinical Oncology)3 and the NCCN (National Comprehensive Cancer Network)4 have revised their clinical practice guidelines to recommend against completion ALND in this subset of patients. Although this guidance has led to a significant decline in the rate of completion ALND among ACOSOG Z0011–eligible patients,5 6 7 a similar reduction was observed among patients undergoing mastectomy.8 9 This reduction was particularly pronounced among patients with SLN micrometastases.8 Further evidence was obtained in the phase 3 IBCSG (International Breast Cancer Study Group) 23-01 randomised controlled trials, where approximately 10% of patients with SLN micrometastases underwent mastectomy; subgroup analysis demonstrated that disease-free survival among patients without axillary dissection was non-inferior to those with axillary dissection after 10 years of follow-up.10 11 Similarly, in the AMAROS trial (After Mapping of the Axilla: Radiotherapy Or Surgery?), 17% of patients with tumour (T) staging T1 to T2 primary breast cancer underwent mastectomy.12 Axillary radiotherapy led to an oncological outcome comparable to completion ALND but was associated with a lower rate of lymphoedema.

In Hong Kong, factors such as the relatively small breast sizes among Chinese women13 and more conservative cultural attitudes13 14 have contributed to a higher rate of mastectomy. The decision to perform mastectomy has prevented a substantial number of breast cancer patients from meeting the ACOSOG Z0011 criteria. Our previous study evaluated the applicability of ACOSOG Z0011 criteria in Hong Kong.15 Patients with clinical nodal (N) staging N0 breast cancer and one or more positive SLNs were stratified into eligible and ineligible groups according to the ACOSOG Z0011 criteria, with 93% of patients in the ineligible group underwent mastectomy.15 Importantly, only 24% of patients in that study met the ACOSOG Z0011 criteria and could potentially avoid ALND.15 Therefore, it is important to identify a low-risk subset of SLN-positive mastectomy patients who could benefit from this non-ALND approach. This retrospective study was conducted to compare the clinical characteristics of SLN-positive breast cancer patients in Hong Kong with the ACOSOG Z0011–eligible cohort.

This retrospective analysis of a prospectively maintained database was conducted at Queen Mary Hospital, a university-affiliated tertiary breast cancer centre in Hong Kong, from June 2014 to May 2019. Potentially eligible patients in the database were identified by an independent research assistant according to whether they met the ACOSOG Z0011 criteria, irrespective of breast surgery type. Patients were excluded if they had positive non-SLNs or positive SLNs only detected by immunohistochemical staining. Relevant data were extracted in July 2020 and missing information was verified using the Clinical Management System, a central computer system for medical records across public hospitals in Hong Kong. Recruited patients were divided into two groups, namely, the mastectomy group and the breast-conserving treatment (ie, ACOSOG Z0011–eligible) group.

All breast cancer patients underwent mammography and ultrasound of the breasts and axillae for clinical tumour and nodal staging. Sentinel lymph node biopsy (SLNB), offered to patients with clinically node-negative disease, was performed with a dual tracer of radioisotope and patent blue dye. Sentinel lymph nodes were defined as lymph nodes with ex vivo gamma probe counts exceeding 10% of the highest ex vivo reading or lymph nodes that displayed blue staining. Non-SLNs were defined as suspicious nodes that were neither hot (high gamma probe counts) nor blue-stained during SLNB, or nodes that were removed during completion ALND. During the study period, intraoperative frozen sections of SLNs or suspicious non-SLNs were routinely collected; these were analysed by standard haematoxylin and eosin staining. Immunohistochemistry was performed in cases of suspected nodal metastasis. Completion ALND was conducted if frozen or paraffin sections showed evidence of nodal metastasis. All final pathological results were reviewed in multidisciplinary meetings. The pathologies of SLNs were considered normal or containing one of the following: macrometastases (>2 mm), micrometastases (>0.2 to ≤2 mm), or isolated tumour cells (≤0.2 mm). For patients undergoing breast-conserving surgery, ‘no ink on tumour’ was regarded as an adequate resection margin16; alternatively, a second operation was performed to ensure a clear resection margin. Adjuvant treatment was administered by breast oncology specialists according to decisions made in multidisciplinary meetings.

Patient demographic characteristics and tumour characteristics were retrieved from database records; percentages were calculated. Missing information was evaluated and managed by pairwise deletion. Comparisons were made between the mastectomy and breast-conserving treatment groups in our cohort, along with the sentinel-alone arm in the ACOSOG Z0011 trial (n=436, in intention to treat).1 2 Analyses followed the per-protocol approach and calculations were performed with SPSS software (Windows version 24.0; IBM Corp, Armonk [NY], United States). Comparisons between cohorts were conducted with Student’s t test or the Chi squared test, as appropriate. Human epidermal growth factor receptor 2 (HER2) status was not assessed in the ACOSOG Z0011 study; therefore, HER2 statuses were only compared within our cohort. The Memorial Sloan Kettering Cancer Center (MSKCC) breast cancer nomogram,17 a well-validated prediction tool to assess the likelihood of non–sentinel node metastases18 19 20 (including external validation in the Chinese population19 20), was used to calculate probability through an online calculator that considered nine variables; comparisons were made between the breast-conserving treatment and mastectomy groups. P values <0.05 were considered statistically significant.

In our centre, the ACOSOG Z0011 criteria have been used to manage patients undergoing breast-conserving surgery since June 2019. From June 2014 to May 2019, 1249 breast cancer patients underwent SLNB in our institution; 171 patients (13.7%) met the study inclusion criteria of clinical T1 or T2 invasive breast cancer and one or two positive SLNs. One hundred and twelve patients (65.5%) underwent mastectomy and 59 patients (34.5%) underwent breast-conserving treatment. The median follow-up period was 58 months (range, 25-84).

Patient demographic characteristics and tumour characteristics of our mastectomy group and the sentinel-alone arm in the ACOSOG Z0011 trial are presented in Table 1. Invasive ductal carcinoma was more common in our patient population than in the ACOSOG Z0011 group. A higher prevalence of clinical T2 breast cancers (~50%) was observed in our mastectomy group (P<0.001). There were also significantly more patients with lymphovascular invasion in our cohort than in the sentinel-alone arm in the ACOSOG Z0011 trial (P<0.001). Although nearly half of the original ACOSOG Z0011 cohort had micrometastatic SLNs, approximately 70% of mastectomy patients had macrometastatic SLNs (P=0.004). These findings suggested that the clinicopathological profile was more aggressive in patients requiring mastectomy.

In our patient cohort, the mastectomy group exhibited many clinicopathological characteristics similar to the breast-conserving treatment group (Table 2). There were no statistically significant differences in terms of age, tumour grade, lymphovascular invasion status, oestrogen receptor/progesterone receptor status, or HER2 status. The mastectomy group had relatively larger tumours than the breast-conserving treatment group (mean: 2.2 cm vs 1.8 cm; P=0.005). Although the difference was not statistically significant, the mastectomy group tended to have larger proportions of patients with two metastatic SLNs (24.1% vs 13.6%; P=0.1) and SLN macrometastases (70.5% vs 57.6%; P=0.11) than the breast-conserving treatment group. Furthermore, the MSKCC probability for additional metastatic non-SLNs was slightly higher in the mastectomy group than in the breast-conserving treatment group (37.1% vs 31.4%; P=0.03) [Table 2].

Ninety-seven patients (86.6%) in the mastectomy group and 45 patients (76.3%) in the breast-conserving treatment group underwent completion ALND. Among patients who underwent mastectomy and completion ALND, 26 patients (26.8%) had additional non-SLN metastases (range, 1-18). In contrast, eight patients (17.8%) in the breast-conserving treatment group had additional non-SLN metastases (range, 1-8). There was no statistically significant difference in the rate of non-SLN metastases between the two treatment arms (P=0.24). Twenty-nine patients (17.9%) underwent SLNB alone; 15 of these patients were in the mastectomy group. Most patients with SLNB alone had micrometastatic SLNs (89.7%) and one patient had isolated tumour cells. None of the patients with SLNB alone experienced recurrence.

In the mastectomy group, 97 patients (86.6%) underwent post-mastectomy irradiation targeting the chest wall and third field regional nodes. Third field regional nodes refer to level III axillary and supraclavicular lymph node regions. None of these patients developed chest wall or axillary recurrence during the follow-up period. Among the 15 patients who did not undergo post-mastectomy irradiation, eight (53.3%) had micrometastatic SLNs and six (40.0%) had macrometastatic SLNs. There were two recurrences (13.3%). First, a 38-year-old patient with one macrometastatic SLN developed ipsilateral chest wall recurrence 4 years after the index operation; this recurrence was managed by a second operation. Second, a patient with two macrometastatic SLNs refused adjuvant systemic treatment and died of breast cancer&dash;related distant metastases. One hundred and ten patients in the mastectomy group (98.2%) received adjuvant systemic treatment: 10 patients (8.9%) required chemotherapy only, 22 patients (19.6%) required hormonal treatment only, and 78 patients (69.6%) required both of these treatments. Seven patients (6.3%) in the mastectomy group developed distant recurrence, and there were three (2.7%) breast cancer–related deaths.

In the breast-conserving treatment group, 58 of the 59 patients underwent adjuvant whole-breast irradiation; 61.0% of these patients underwent additional third field nodal irradiation. Fifty-eight patients (98.3%) in the breast-conserving treatment group received adjuvant systemic treatment involving hormonal therapy and/or chemotherapy. Three patients (5.1%) had distant recurrence; among them, one (1.7%) died at 39 months after the initial diagnosis. One patient experienced ipsilateral breast recurrence at 30 months and underwent completion mastectomy.

The favourable oncological results of the ACOSOG Z0011 trial1 2 have challenged the conventional approach of performing completion ALND in patients with SLN metastases. Patients with one or two SLN metastases who underwent breast-conserving surgery, whole-breast irradiation, and adjuvant systemic treatment could safely forgo completion ALND. This paradigm shift has led to substantial de-escalation of axillary surgery worldwide.5 A meta-analysis by Schmidt-Hansen et al,21 which involved 2020 patients and findings from the IBCSG 23-0110 11 and the AATRM (Agència d’Avaluació de Tecnologia i Recerca Mèdiques) 048/13/200022 trials, concluded that SLNB alone was sufficient for locoregional control in early breast cancer, without adverse effects on survival.

Despite widespread adoption of the ACOSOG Z0011 criteria, the study has been criticised in several ways. The low locoregional relapse rate of 1.5% indicates that the study was underpowered.23 Furthermore, significant deviation in the radiotherapy protocol, such that 18.9% of patients received ‘high tangents’ radiotherapy, has raised questions concerning the oncological safety of SLNB alone in patients without third field nodal irradiation.24 Combined with the insufficient numbers of mastectomy patients in the IBCSG 23-01,10 11 AMAROS,12 and AATRM 048/13/200022 trials, it has been unclear whether this non-ALND approach can be extrapolated to SLN-positive breast cancer patients who undergo mastectomy with or without radiotherapy.

In this study, we compared the clinicopathological characteristics among our mastectomy group, our breast-conserving treatment group, and the sentinel-alone arm in the original ACOSOG Z0011 study. Unsurprisingly, our mastectomy group exhibited more aggressive tumour characteristics than the sentinel-alone arm in the Western population; specifically, it had a larger tumour size, more frequent lymphovascular invasion, and a greater proportion of patients with SLN macrometastases. These differences in clinicopathological features have also been reported in Western populations. For example, Hennigs et al8 analysed a large German cohort that included 4093 SLN-positive mastectomy patients. Compared with the entire study cohort of 166 074 patients, T2 tumour and lymphovascular invasion were more commonly found in patients requiring mastectomy. Additionally, the study by Milgrom et al25 included 535 early-stage breast cancer patients with a positive SLNB and no ALND. In their mastectomy group, patients had significantly larger tumours and more frequently displayed multifocal/multicentric disease. However, these adverse pathological features among mastectomy patients did not justify a more aggressive axillary approach. Similarly, the low rates of local and regional failure observed in our cohort were consistent with previous reports, suggesting that axillary-specific treatment can be considered in this group of patients with low-volume SLN disease.25 26 27 28 Debate persists regarding the comparatively large proportions of patients with micrometastatic disease in the original ACOSOG Z0011 trial1 2 and other studies.25 26 Cowher et al29 published a retrospective analysis of patients who underwent mastectomy and conservative axillary regional excision (ie, removal of SLNs and other palpable nodes). Among 144 patients with pathological N1 disease, a small proportion (24%) had micrometastatic disease; only three axillary recurrences (2.1%) were reported.29 Notably, the low locoregional failure rate was not attributed to post-mastectomy irradiation25 26 27 28 29 or increased use of chemotherapy.26 27 28

In our previous study, we demonstrated differences in clinical characteristics between Asian and Western populations.15 In the present study, our breast-conserving treatment group had a higher rate of clinical T2 tumours and more frequent lymphovascular invasion compared with the Western population. Similar findings were observed in Korean30 and Japanese31 studies, which revealed larger and higher-grade tumours, increased lymphovascular permeation, and more frequent SLN macrometastases. Despite these disparities, the Korean30 and Japanese31 studies both demonstrated safe application of ACOSOG Z0011 criteria in Asia, with low incidences of disease recurrence. These intrinsic differences in tumour characteristics between Eastern and Western populations have presumably reduced the gap in clinicopathological features between patients undergoing mastectomy and those undergoing breast-conserving surgery. In the head-to-head comparison between our mastectomy cohort and our breast-conserving treatment group, the only notable difference involved the mean pathological size of the invasive focus (2.2 cm vs 1.8 cm; P=0.005); the clinical tumour stage distribution did not differ (P=0.69) [Table 2]. The small difference in mean MSKCC breast cancer nomogram probability (37.1% vs 31.4%; P=0.03) could also be related to the difference in pathological size, which is one of the nine variables considered in the nomogram. Therefore, we believe that a non-ALND approach in this low-risk subset of SLN-positive mastectomy patients is acceptable.

The primary concern regarding extrapolation of this non-ALND approach is the risk of undertreatment for patients with an extensive nodal burden. The original ACOSOG Z0011 trial revealed a non-SLN macrometastasis rate of 27.3% in the ALND group.1 2 The AMAROS trial also showed that 33% of patients in the ALND group had additional positive lymph nodes.12 Importantly, the axillary recurrence rate remained low in both of these studies. In our SLN-positive mastectomy and breast-conserving treatment groups, the proportions of patients with additional non-SLN metastases were 26.8% and 17.8%, respectively. Among patients undergoing adjuvant irradiation and adjuvant systemic treatment, it is likely that some non-SLN metastases do not progress to clinically detectable disease.

This study had several limitations. First, its retrospective design could result in recall bias and the potential for missing clinical information. Although data from the ACOSOG Z0011 trial were limited with respect to HER2 status, extracapsular extension, and multifocality, we attempted to mitigate this issue by including some of the affected variables in the comparison of our mastectomy and breast-conserving treatment groups. Second, we could not address the need for post-mastectomy irradiation among patients in this study. The value of such irradiation for breast cancer patients with <4 positive lymph nodes remains controversial. The meta-analysis by the Early Breast Cancer Trialists’ Collaborative Group,32 which included 1314 breast cancer patients with one to three positive nodes after mastectomy and ALND, suggested that radiotherapy provided oncological benefit in terms of locoregional recurrence, overall recurrence, and breast cancer mortality. However, this meta-analysis has been criticised for including some very early studies from the 1970s, in which the reported recurrence rates were much higher than rates in later studies. In 2016, a focused update by the American Society of Clinical Oncology, American Society for Radiation Oncology, and Society of Surgical Oncology acknowledged the use of post-mastectomy radiotherapy for this group of patients but recommended clinical judgement for patients with a low risk of locoregional recurrence.33 In our centre, post-mastectomy irradiation was generally administered to patients with pathological N1 disease during the study period; 86.6% of patients in the present study underwent adjuvant radiotherapy. Considering the similarities in clinicopathological features and adjuvant systemic treatment use between our SLN-positive mastectomy and breast-conserving treatment groups, we suspect that it is safe for selected low-risk SLN-positive mastectomy patients to forgo ALND through the expansion of AMAROS eligibility12 to ACOSOG Z0011–ineligible patients. Several ongoing randomised studies, such as the English POSNOC (POsitive Sentinel NOde: adjuvant therapy alone versus adjuvant therapy plus Clearance or axillary radiotherapy)34 and the Dutch BOOG 2013-07,35 are recruiting breast cancer patients who undergo mastectomy and have a maximum of two to three positive SLNs; these studies aim to compare completion axillary treatment (ALND or axillary radiotherapy) and the lack of completion axillary treatment. Additionally, the SINODAR-ONE trial36 recently published their subgroup analysis and found non-inferior overall survival and recurrence-free survival among mastectomy patients receiving SLNB and ALND. The ongoing studies are expected to provide more robust evidence concerning the optimal treatment for SLN-positive mastectomy patients.

This study demonstrated the clinicopathological similarities between SLN-positive mastectomy and breast-conserving treatment groups among breast cancer patients in Hong Kong. Cautious application of the non-ALND approach in mastectomy patients with low-volume SLN disease is reasonable, considering the low locoregional recurrence rate. However, additional research is needed to standardise the adjuvant post-mastectomy radiotherapy protocol, especially among patients who forego ALND.

Concept or design: V Man.
Acquisition of data: V Man.
Analysis or interpretation of data: V Man.
Drafting of the manuscript: Both authors.
Critical revision of the manuscript for important intellectual content: A Kwong.

Both 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.

Both authors have disclosed no conflicts of interest.

This study has been presented and awarded the Young Investigator Award Best Scientific Paper in the Hong Kong Society of Breast Surgeons 5th Annual Scientific Meeting (19 September 2021, Hong Kong).

This study 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, Hong Kong (Ref No.: HKU/HA HKW UW 09-045). Written informed consent was obtained from patients for all treatments, procedures, and publication.

1. Giuliano AE, Hunt KK, Ballman KV, et al. Axillary dissection vs no axillary dissection in women with invasive breast cancer and sentinel node metastasis: a randomized clinical trial. JAMA 2011;305:569-75. Crossref

2. Guiliano AE, Ballman KV, McCall L, et al. Effect of axillary dissection vs no axillary dissection on 10-year overall survival among women with invasive breast cancer and sentinel node metastasis: the ACOSOG Z0011 (Alliance) randomized clinical trial. JAMA 2017;318:918-26. Crossref

3. Lyman GH, Termin S, Edge SB, et al. Sentinel lymph node biopsy for patients with early-stage breast cancer: American Society of Clinical Oncology clinical practice guideline update. J Clin Oncol 2014;32:1365-83. Crossref

4. Gradishar WJ, Anderson BO, Balassanian R, et al. NCCN guidelines insights: breast cancer, version 1.2017. J Natl Compr Canc Netw 2017;15:433-51. Crossref

5. Camp MS, Greenup RA, Taghian A, et al. Application of ACOSOG Z0011 criteria reduces perioperative costs. Ann Surg Oncol 2013;20:836-41. Crossref

6. Delpech Y, Bricou A, Lousquy R, et al. The exportability of the ACOSOG Z0011 criteria for omitting axillary lymph node dissection after positive sentinel lymph node biopsy findings: a multicenter study. Ann Surg Oncol 2013;20:2556-61. Crossref

7. Verheuvel NC, Voogd AC, Tjan-Heijnen VC, Roumen RM. Potential impact of application of Z0011 derived criteria to omit axillary lymph node dissection in node positive breast cancer patients. Eur J Surg Oncol 2016;42:1162-8. Crossref

8. Hennigs A, Riedel F, Feißt M, et al. Evolution of the use of completion axillary lymph node dissection in patients with T1/2N0M0 breast cancer and tumour-involved sentinel lymph nodes undergoing mastectomy: a cohort study. Ann Surg Oncol 2019;26:2435-43. Crossref

9. Poodt IG, Spronk PE, Vugts G, et al. Trends on axillary surgery in nondistant metastatic breast cancer patients treated between 2011 and 2015: a Dutch population-based study in the ACOSOG-Z0011 and AMAROS era. Ann Surg 2018;268:1084-90. Crossref

10. Galimberti V, Cole BF, Zurrida S, et al. Axillary dissection versus no axillary dissection in patients with sentinel-node micrometastases (IBCSG 23-01): a phase 3 randomised controlled trial. Lancet Oncol 2013;14:297-305. Crossref

11. Galimberti V, Cole BF, Viale G, et al. Axillary dissection versus no axillary dissection in patients with breast cancer and sentinel-node micrometastases (IBCSG 23-01): 10-year follow-up of a randomised, controlled phase 3 trial. Lancet Oncol 2018;19:1385-93. Crossref

12. Donker M, van Tienhoven G, Straver ME, et al. Radiotherapy or surgery of the axilla after a positive sentinel node in breast cancer (EORTC 10981-22023 AMAROS): a randomised, multicentre, open-label, phase 3 non-inferiority trial. Lancet Oncol 2014;15:1303-10. Crossref

13. Sinnadurai S, Kwong A, Hartman M, et al. Breast-conserving surgery versus mastectomy in young women with breast cancer in Asian settings. BJS Open 2018;3:48-55. Crossref

14. Chan SW, Cheung C, Chan A, Cheung PS. Surgical options for Chinese patients with early invasive breast cancer: data from the Hong Kong Breast Cancer Registry. Asian J Surg 2017;40:444-52.Crossref

15. Man V, Lo MS, Kwong A. The applicability of the ACOSOG Z0011 criteria to breast cancer patients in Hong Kong. Chin Clin Oncol 2021;10:27. Crossref

16. Moran MS, Schnitt SJ, Giuliano AE, et al. Society of Surgical Oncology–American Society for Radiation Oncology consensus guideline on margins for breast-conserving surgery with whole-breast irradiation in stages I and II invasive breast cancer. Ann Surg Oncol 2014;21:704-16. Crossref

17. Memorial Sloan Kettering Cancer Center. Breast cancer nomogram: breast additional non-SLN metastases. Available from: https://nomograms.mskcc.org/breast/BreastAdditionalNonSLNMetastasesPage.aspx. Accessed 19 Mar 2024.

18. Gur AS, Unal B, Ozbek U, et al. Validation of breast cancer nomograms for predicting the non–sentinel lymph node metastases after a positive sentinel lymph node biopsy in a multi-center study. Eur J Surg Oncol 2010;36:30-5. Crossref

19. Kuo YL, Chen WC, Yao WJ, et al. Validation of Memorial Sloan-Kettering Cancer Center nomogram for prediction of non–sentinel lymph node metastasis in sentinel lymph node positive breast cancer patients: an international comparison. Int J Surg 2013;11:538-43. Crossref

20. Bi X, Wang Y, Li M, et al. Validation of the Memorial Sloan Kettering Cancer Center nomogram for predicting non–sentinel lymph node metastasis in sentinel lymph node–positive breast-cancer patients. Onco Targets Ther 2015;8:487-93. Crossref

21. Schmidt-Hansen M, Bromham N, Hasler E, Reed MW. Axillary surgery in women with sentinel node–positive operable breast cancer: a systematic review with meta-analyses. Springerplus 2016;5:85. Crossref

22. Solá M, Alberro JA, Fraile M, et al. Complete axillary lymph node dissection versus clinical follow-up in breast cancer patients with sentinel node micrometastasis: final results from the multicenter clinical trial AATRM 048/13/2000. Ann Surg Oncol 2013;20:120-7. Crossref

23. Huang TW, Kuo KN, Chen KH, et al. Recommendation for axillary lymph node dissection in women with early breast cancer and sentinel node metastasis: a systematic review and meta-analysis of randomized controlled trials using the GRADE system. Int J Surg 2016;34:73-80. Crossref

24. Jagsi R, Chadha M, Moni J, et al. Radiation field design in the ACOSOG Z0011 (Alliance) trial. J Clin Oncol 2014;32:3600-6. Crossref

25. Milgrom S, Cody H, Tan L, et al. Characteristics and outcomes of sentinel node–positive breast cancer patients after total mastectomy without axillary-specific treatment. Ann Surg Oncol 2012;19:3762-70. Crossref

26. FitzSullivan E, Bassett RL, Kuerer HM, et al. Outcomes of sentinel lymph node–positive breast cancer patients treated with mastectomy without axillary therapy. Ann Surg Oncol 2017;24:652-9. Crossref

27. Snow R, Reyna C, Johns C, et al. Outcomes with and without axillary node dissection for node-positive lumpectomy and mastectomy patients. Am J Surg 2015;210:685-93. Crossref

28. Joo JH, Kim SS, Son BH, et al. Axillary lymph node dissection does not improve post-mastectomy overall or disease-free survival among breast cancer patients with 1-3 positive nodes. Cancer Res Treat 2019;51:1011-21. Crossref

29. Cowher MS, Grobmyer SR, Lyons J, O’Rourke C, Baynes D, Crowe JP. Conservative axillary surgery in breast cancer patients undergoing mastectomy: long-term results. J Am Coll Surg 2014;218:819-24. Crossref

30. Jung J, Han W, Lee ES, et al. Retrospectively validating the results of the ACOSOG Z0011 trial in a large Asian Z0011-eligible cohort. Breast Cancer Res Treat 2019;175:203-15. Crossref

31. Kittaka N, Tokui R, Ota C, et al. A prospective feasibility study applying the ACOSOG Z0011 criteria to Japanese patients with early breast cancer undergoing breast-conserving surgery. Int J Clin Oncol 2018;23:860-6. Crossref

32. EBCTCG (Early Breast Cancer Trialists’ Collaborative Group); McGale P, Taylor C, et al. Effect of radiotherapy after mastectomy and axillary surgery on 10-year recurrence and 20-year breast cancer mortality: meta-analysis of individual patient data for 8135 women in 22 randomised trials. Lancet 2014;383:2127-35. Crossref

33. Recht A, Comen EA, Fine RE, et al. Postmastectomy radiotherapy: an American Society of Clinical Oncology, American Society for Radiation Oncology, and Society of Surgical Oncology focused guideline update. Pract Radiat Oncol 2016;6:e219-34. Crossref

34. Goyal A, Dodwell D. POSNOC: a randomised trial looking at axillary treatment in women with one or two sentinel nodes with macrometastases. Clin Oncol (R Coll Radiol) 2015;27:692-5. Crossref

35. van Roozendaal LM, de Wilt JH, van Dalen T, et al. The value of completion axillary treatment in sentinel node positive breast cancer patients undergoing a mastectomy: a Dutch randomized controlled multicentre trial (BOOG 2013-07). BMC Cancer 2015;15:610. Crossref

36. Tinterri C, Canavese G, Gatzemeier W, et al. Sentinel lymph node biopsy versus axillary lymph node dissection in breast cancer patients undergoing mastectomy with one to two metastatic sentinel lymph nodes: sub-analysis of the SINODAR-ONE multicentre randomized clinical trial and reopening of enrolment. Br J Surg 2023;110:1143-52. Crossref