In 2017, breast cancer was the most common cancer and third leading cause of cancer death among women in Hong Kong.1 Additionally, an estimated 20% of breast cancers in Hong Kong were human epidermal growth factor receptor-2 (HER2)-positive.2 3

Intravenous (IV) trastuzumab, in combination with chemotherapy, is licensed for the treatment of HER2-positive early-stage breast cancer and metastatic breast cancer. It must be reconstituted into solution for loading dose infusion over a duration of 90 minutes, followed by maintenance dose infusion over a duration of 30 minutes.4 Additionally, IV trastuzumab is dosed according to each patient’s body weight, with a loading dose of 8 mg/kg followed by a maintenance dose of 6 mg/kg every 3 weeks.4 This regimen consumes considerable healthcare resources, including drug preparation and administration time, clinic and chair time, and physician time dedicated to patient interaction.5

A fixed-dose subcutaneous (SC) formulation of trastuzumab was developed to allow drug administration over approximately 5 minutes, which is much shorter than the duration of IV infusion. The 600-mg dose of SC trastuzumab every 3 weeks is non-dinferior to the IV formulation with respect to efficacy and tolerability.6 7 Furthermore, approximately 90% of patients preferred SC over IV administration of trastuzumab in the PrefHer (Preference for subcutaneous or intravenous administration of trastuzumab in patients with HER2-positive early breast cancer) randomised crossover trials,8 9 which were designed to assess patient preference and healthcare professional satisfaction with both treatment options.

Data from other countries have demonstrated that for SC formulation of trastuzumab, less time is required for drug preparation and administration; moreover, fewer consumables are used.10 11 12 13 A cost-minimisation analysis (CMA) study in Greece demonstrated that the total cost of therapy per patient was 21 870 euros (€) when using the SC formulation of trastuzumab, whereas it was €23 118 when using the IV formation of trastuzumab. The investigators concluded that use of the SC formulation of trastuzumab would provide cost savings for the Greek healthcare system.10 A study in Spain revealed similar findings: the use of the SC formulation of trastuzumab led to a 19.4 to 28.8% cost savings in the hospital.11 Additionally, a time-and-motion study in New Zealand compared medical resource utilisation between the IV and SC formulations of trastuzumab in patients with HER2-positive breast cancer. The potential cost saving was NZ$96.94 per patient per cycle.12 Furthermore, a time-and-motion sub-study13 from the PrefHer trials involving eight countries (Canada, France, Switzerland, Denmark, Italy, Russia, Spain, and Turkey) demonstrated time savings for patient chair, administration by healthcare professionals, and drug preparation.

The SC formulation of trastuzumab is expected to provide cost savings in other countries. However, healthcare systems and modes of clinical services differ between Hong Kong and other countries. Therefore, this study was conducted to investigate cost differences between IV and SC trastuzumab regimens in Hong Kong medical settings, using medical resources utilisation data from other countries.

A CMA model was developed to compare the cost of total care. The CMA approach was used because the clinical efficacy and safety profiles of IV and SC trastuzumab regimens are similar, as demonstrated in the previous studies7 14 15; this fulfils the CMA requirement for two treatments to demonstrate similar efficacy. The following steps were followed in the CMA. We compared direct medical costs related to the IV and SC trastuzumab regimens that produced equivalent health outcomes. The CMA solely focuses on selection of the least costly option. In this study, the CMA was conducted from a hospital perspective. All direct medical costs and full-time equivalent (FTE) hours were included in this study. Drugs, clinic visits for drug administration, specialist out-patient clinic visits, and consumables were regarded as direct medical costs. The time horizon was 18 cycles of treatment, which mimics the duration of treatment for early-stage HER2-positive breast cancer. Drug acquisition cost data were obtained from the manufacturer, whereas costs for hospitalisation and clinic visits were acquired from the 2017 Hong Kong Gazette.16 The drug acquisition cost was based on the dose used in previous clinical trials: IV loading dose of 8 mg/kg and maintenance dose of 6 mg/kg every 3 weeks versus SC fixed dose of 600 mg every 3 weeks. A mean body weight of 57.3 kg was used, based on data from the 2016 Hong Kong Cancer Registry.3

Estimated FTE hour values were obtained from previous literature. These values were regarded as the time (in hours) required for drug preparation and administration, divided by 44 hours, the weekly average working hours for such tasks. The FTE hour values were then converted to monetary values, calculated as the median hourly rate received by individuals in each position. In Hong Kong, nurses and pharmacists are mainly involved in drug preparation and administration; thus, the salaries of these positions were used for estimation of FTE hour values.

All costs were expressed in US dollars (US$1 = HK$7.8), using 2016 as the fiscal year. Because of the short time horizon in the study, no costs were discounted.

Medical resources and FTE hour values were determined by literature review in Embase and MEDLINE, using the key words ‘subcutaneous’, ‘trastuzumab’, ‘time’, ‘cost’, and ‘medical resources’.

The CMA was conducted from the healthcare payer perspective. All continuous variables were described as means ± standard deviations and medians with ranges.

A drug budget impact forecast analysis was performed to determine how changes in the total cost of treatment regimens, including direct medical costs and FTE hours, would impact healthcare expenditures in Hong Kong. Each individual parameter, namely drug acquisition cost for each formulation (±20%), patient body weight (±20%), and time and consumables reported in the literature (based on confidence intervals reported) were analysed independently within specified ranges, whereas other factors were fixed at base-case values. The analysis parameters were chosen based on the findings in previous cost-effectiveness studies.17 A simulation model was used to run 10 000 iterations of the forecast model; for each iteration, model parameters were input as shown in Table 1. We assumed that cost changes were consistent with the beta distribution around the mean. One-way deterministic sensitivity analysis was also performed to evaluate the extent to which the total cost would be affected by changes in the drug acquisition cost for each formulation (±20%), changes in times and consumables obtained from literature (based on confidence intervals reported), and changes in body weight (±20%); this approach is consistent with the methodology used in another cost-effectiveness analysis focused on trastuzumab.17 Figure 1 summarises the analysis process of this study.

In total, 11 studies were identified, eight of which were eligible for analysis.12 18 19 20 21 22 23 24 Three studies were excluded because they did not report the time required for administration or preparation. There are a total of six studies with information on pharmacist time on preparation and nursing time on administration for IV and SC trastuzumab; the remaining two only reported time differences between the two formulations. Among the six studies that reported the time for preparation, four reported the total drug preparation time required for IV and SC trastuzumab, whereas the remaining two only reported time differences. If the SC formulation was used, 0.18 FTE hour of nursing time (7.9 hours) and 0.14 FTE hour of pharmacist time (6.2 hours) could be saved each week. Table 2 summarises the findings from these studies.

After 18 cycles of treatment with SC trastuzumab, the drug acquisition and healthcare professional time costs were reduced by US$9451.28 and US$566.16, respectively, compared with IV trastuzumab. Therefore, US$10 017.44 could be saved for each patient who completed 18 cycles of treatment. The cost of consumables was excluded because only two studies reported this information, and the contributions to overall costs were minimal (NZ$15.2712 and GBP0.6421, respectively). Table 3 summarises the direct medical costs of IV and SC formulations.

The drug budget impact forecast model was most affected by body weight and drug acquisition cost. Cost differences between the IV and SC formulations were reduced by decreases in body weight and IV trastuzumab cost, as well as an increase in SC trastuzumab cost. The effects of changes in nursing time and pharmacist time were smaller. Table 1 summarises the model parameters, and Figure 2 illustrates the effects of each variable on cost differences.

In 2017, 4373 women were diagnosed with invasive breast cancer,1 and approximately 20% of them were HER2-positive.2 3 Furthermore, trastuzumab was the most commonly used targeted therapy (95.3%).3 Assuming that the SC formulation was used (instead of the IV formulation) for all HER2-positive patients receiving trastuzumab and using the 2017 data stated here, an annual saving of over US$8.3 million could be achieved in Hong Kong.

The results of this study suggest that SC trastuzumab would be more cost-effective than its IV counterpart in Hong Kong. Even if lower-cost biosimilar trastuzumab becomes available, the SC formulation will remain less expensive unless there is a substantial reduction in the acquisition cost of IV trastuzumab.

As body weight decreases, the necessary dosage and corresponding expenditures are expected to decrease. Paradoxically, ≤20% increases in body weight had a neutral effect in the analysis. This result could be related to a substantial amount of drug wastage when using weight-based IV trastuzumab, which is consistent with previous findings.19 25 Therefore, further studies are needed to determine the optimal route of administration for patients who are underweight or do not require full doses of trastuzumab because of their clinical conditions.

Although the SC formulation is expected to save time for healthcare professionals,26 27 28 the present analysis suggests that its contribution to the total cost of care is minimal. The cost of drug acquisition has the greatest effect on financial burden.

The use of data from previous time-and-motion studies in other countries may not be appropriate for medical settings in Hong Kong. Further studies should be conducted in Hong Kong to estimate the actual cost savings with respect to healthcare professional time, although theoretical time savings may not accurately represent actual time savings because of clinical activities conducted during administration of trastuzumab.29 Furthermore, data from other countries exhibited wide distributions in terms of standard deviation and range. Nevertheless, the influence of the SC formulation on the total cost-saving effect may be limited, as demonstrated in the sensitivity analysis.

Although the costs of clinic visits and chemotherapy were assumed to be identical throughout 18 cycles of treatment between the two formulations, some patients can receive SC trastuzumab in ambulatory care settings. Thus, the mean savings may have been underestimated in our model.

There were several limitations in this study. First, because of the small number of studies identified in the literature review, consumables could not be included in the CMA. Second, societal cost and patient preferences were not considered because such information is unavailable in Hong Kong. A more patient-centred approach would provide greater insights. Third, time-and-motion analysis and waste handling in Hong Kong were not considered; these factors may have specific impact on drug preparation time and administration time and costs. Fourth, costs for adverse drug reactions were not included because these costs were assumed to be equal for IV and SC trastuzumab regimens. However, this assumption may be incorrect, particularly with regard to infusion-related reactions.

The results of this study suggest that the SC formulation of trastuzumab would be a cost-saving therapy for HER2-positive breast cancer patients in Hong Kong. The drug acquisition cost was the parameter with the greatest effect on the total cost of treatment.

Concept or design: VWY Lee.
Acquisition of data: FWT Cheng.
Analysis or interpretation of data: Both authors.
Drafting of the manuscript: FWT Cheng.
Critical revision of the manuscript for important intellectual content: VWY Lee.

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 disclosed no conflicts of interest.

The datasets generated and/or analysed in this study are available from the corresponding author on reasonable request.

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

Not applicable because this study did not involve human participants.

1. Hong Kong Cancer Registry, Hospital Authority, Hong Kong SAR Government. Female breast cancer in 2017. Available from: https://www3.ha.org.hk/cancereg/pdf/factsheet/2017/breast_2017.pdf. Accessed 11 May 2020.

2. Yau TK, Sze H, Soong IS, Hioe F, Khoo US, Lee AW. HER2 overexpression of breast cancers in Hong Kong: prevalence and concordance between immunohistochemistry and in-situ hybridisation assays. Hong Kong Med J 2008;14:130-5.

3. Hong Kong Breast Cancer Foundation. Hong Kong Breast Cancer Registry Report No. 8. 2016. Available from: http://www.hkbcf.org/download/bcr_report8/hkbcf_report_2016_full_report.pdf. Accessed 21 May 2017.

4. Genentech. Highlights of prescribing information. 2016. Available from: http://www.gene.com/download/pdf/herceptin_prescribing.pdf. Accessed 8 Aug 2016.

5. Kruse GB, Amonkar MM, Smith G, Skonieczny DC, Stavrakas S. Analysis of costs associated with administration of intravenous single-drug therapies in metastatic breast cancer in a U.S. population. J Manag Care Pharm 2008;14:844-57. Crossref

6. Ismael G, Hegg R, Muehlbauer S, et al. Subcutaneous versus intravenous administration of (neo)adjuvant trastuzumab in patients with HER2-positive, clinical stage I-III breast cancer (Hannah study): a phase 3, open-label, multicentre, randomised trial. Lancet Oncol 2012;13:869-78. Crossref

7. Jackisch C, Kim SB, Semiglazov V, et al. Subcutaneous versus intravenous formulation of trastuzumab for HER2-positive early breast cancer: updated results from the phase III HannaH study. Ann Oncol 2015;26:320-5. Crossref

8. Pivot X, Gligorov J, Müller V, et al. Preference for subcutaneous or intravenous administration of trastuzumab in patients with HER2-positive early breast cancer (PrefHer): an open-label randomised study. Lancet Oncol 2013;14:962-70. Crossref

9. Pivot X, Gligorov J, Müller V, et al. Patients’ preferences for subcutaneous trastuzumab versus conventional intravenous infusion for the adjuvant treatment of HER2- positive early breast cancer: final analysis of 488 patients in the international, randomized, two-cohort PrefHer study. Ann Oncol 2014;25:1979-87. Crossref

10. Mylonas C, Kourlaba G, Fountzilas G, Skroumpelos A, Maniadakis N. Cost-minimization analysis of trastuzumab intravenous versus trastuzumab subcutaneous for the treatment of patients with HER2+ early breast cancer and metastatic breast cancer in Greece. Value Health 2014;17:A640-1. Crossref

11. Gutierrez F, Nazco G, Viña M, Bullejos M, Gonzalez I, Valcarcel C. Economic impact of using subcutaneous trastuzumab. Value Health 2014;17:A641. Crossref

12. North RT, Harvey VJ, Cox LC, Ryan SN. Medical resource utilization for administration of trastuzumab in a New Zealand oncology outpatient setting: a time and motion study. Clinicoecon Outcomes Res 2015;7:423-30. Crossref

13. De Cock E, Pivot X, Hauser N, et al. A time and motion study of subcutaneous versus intravenous trastuzumab in patients with HER2-positive early breast cancer. Cancer Med 2016;5:389-97. Crossref

14. Van den Nest M, Glechner A, Gold M, Gartlehner G. The comparative efficacy and risk of harms of the intravenous and subcutaneous formulations of trastuzumab in patients with HER2-positive breast cancer: a rapid review. Syst Rev 2019;8:321. Crossref

15. Jackisch C, Stroyakovskiy D, Pivot X, et al. Subcutaneous vs intravenous trastuzumab for patients with ERBB2-positive early breast cancer: final analysis of the HannaH phase 3 randomized clinical trial. JAMA Oncol 2019;5:e190339. Crossref

16. Government Logistics Department, Hong Kong SAR Government. Hospital Authority Ordinance (Chapter 113). Revisions to list of charges. Available from: https://www.gld.gov.hk/egazette/pdf/20172124/egn201721243884.pdf. Accessed 30 May 2017.

17. Kurian AW, Thompson RN, Gaw AF, Arai S, Ortiz R, Garber AM. A cost-effectiveness analysis of adjuvant trastuzumab regimens in early HER2/neu-positive breast cancer. J Clin Oncol 2007;25:634-41. Crossref

18. Olofsson S, Norrlid H, Karlsson E, Wilking U, Ragnarson Tennvall G. Societal cost of subcutaneous and intravenous trastuzumab for HER2-positive breast cancer—an observational study prospectively recording resource utilization in a Swedish healthcare setting. Breast 2016;29:140-6. Crossref

19. Ponzetti C, Canciani M, Farina M, Era S, Walzer S. Potential resource and cost saving analysis of subcutaneous versus intravenous administration for rituximab in non-Hodgkin’s lymphoma and for trastuzumab in breast cancer in 17 Italian hospitals based on a systematic survey. Clinicoecon Outcomes Res 2016;8:227-33. Crossref

20. De Cock E, Pan YI, Tao S, Baidin P. Time savings with trastuzumab subcutaneous (SC) injection verse trastuzumab intravenous (IV) infusion: a time and motion study in 3 Russian centers. Value Health 2014;17:A653. Crossref

21. Nawaz S, Samanta K, Lord S, Diment V, Mcnamara S. Cost savings with Herceptin® (trastuzumab) SC vs IV administration: a time & motion study. Breast 2013;22(S1):S112.

22. De Cock E, Tao S, Alexa U, Pivot X, Knoop A. Abstract P5- 15-07: Time savings with trastuzumab subcutaneous (SC) injection vs. trastuzumab intravenous (IV) infusion: first results from a Time-and-Motion study (T&M). Cancer Res 2014;72(24 Suppl):P5-15-07. Crossref

23. De Cock E, Tao S DM-P, Millar D CN. Time savings with trastuzumab subcutaneous (SC) injection vs. trastuzumab intravenous (IV) infusion: a time and motion study in 5 Canadian centres. Proceedings of the Canadian Association for Population Therapeutics (CAPT) Annual Conference; 2013 Nov 17-19; Toronto, Canada.

24. Samanta K, Moore L, Jones G, Evason J, Owen G. PCN39 potential time and cost savings with herceptin (trastuzumab) subcutaneous (SC) injection versus herceptin intravenous (IV) infusion: results from three different English patient settings. Value Health 2012;15:A415. Crossref

25. Nestorovska A, Naumoska Z, Grozdanova A, et al. Subcutaneous vs intravenous administration of trastuzumab in HER2+ breast cancer patients: a Macedonian cost-minimization analysis. Value Health 2015;18:A463. Crossref

26. Rojas L, Muñiz S, Medina L, et al. Cost-minimization analysis of subcutaneous versus intravenous trastuzumab administration in Chilean patients with HER2-positive early breast cancer. PLoS One 2020;15:e0227961. Crossref

27. O’Brien GL, O’Mahony C, Cooke K, et al. Cost minimization analysis of intravenous or subcutaneous trastuzumab treatment in patients with HER2-positive breast cancer in Ireland. Clin Breast Cancer 2019;19:e440-51. Crossref

28. Lopez-Vivanco G, Salvador J, Diez R, et al. Cost minimization analysis of treatment with intravenous or subcutaneous trastuzumab in patients with HER2-positive breast cancer in Spain. Clin Transl Oncol 2017;19:1454-61. Crossref

29. Papadmitriou K, Trinh XB, Altintas S, Van Dam PA, Huizing MT, Tjalma WA. The socio-economical impact of intravenous (IV) versus subcutaneous (SC) administration of trastuzumab: future prospectives. Facts Views Vis Obgyn 2015;7:176-80.

Osteosarcoma arises from primitive bone-forming mesenchymal cells.1 It is the most common primary malignant bone tumour worldwide,1 2 with an annual incidence of 4.8 per million population in the US.2 Moreover, it was one of the most common types of childhood malignancy in Hong Kong in 2017 to 2019.3

In Hong Kong, paediatric patients with high-grade osteosarcoma are treated in accordance with the Hong Kong Paediatric Haematology and Oncology Study Group Treatment Protocol for High-Grade Osteosarcoma (Fig 1), which consists of neoadjuvant chemotherapy, followed by tumour resection and subsequent adjuvant chemotherapy. Histological response to neoadjuvant chemotherapy, defined as tumour chemonecrosis according to the Huvos system,4 is used to stratify patients into good responders (patients with ≥90% tumour chemonecrosis) and poor responders (patients with <90% tumour chemonecrosis). These two groups receive different chemotherapy regimens.

A few prognostic factors for survival have been recognised; these factors include the presence of metastasis at diagnosis5 6 7 and poor tumour chemonecrosis.5 6 7 8 9 However, no specific studies have examined the validity of these factors for patients with osteosarcoma in Hong Kong. Moreover, the presence of lung nodules in computed tomography (CT) scans is a common finding at diagnosis and reassessment. In this study, we reviewed paediatric patients with osteosarcoma at two of the largest paediatric oncology centres in Hong Kong. We sought prognostic factors for event-free survival (EFS) and overall survival (OS), and we assessed lung nodules and metastasis in these patients.

This study included paediatric patients (aged <19 years at diagnosis) with biopsy-proven high-grade osteosarcoma, who were diagnosed from 1 January 2009 to 31 December 2018 at Queen Elizabeth Hospital and Prince of Wales Hospital. The patients’ medical records were reviewed and the following information was collected: demographic data (sex, age at diagnosis, ethnicity, and time from symptom onset to presentation), clinical characteristics (location of tumour, largest dimension of tumour, staging, maximal alkaline phosphatase [ALP] level, calcium and phosphate levels before the start of treatment, and presence of pathological fracture), treatment-related characteristics (time from diagnosis to start of chemotherapy, surgical treatment approach, and histological results of the resected tumour [including tumour chemonecrosis and surgical margin]), and details of metastasis (timing, location, size, and treatment). Radiological reports for the primary tumour were reviewed to determine the presence of local anatomical aggressiveness, which was defined as intra-articular tumour involvement or neurovascular bundle involvement. These risk factors made surgical resection with a negative tumour margin difficult or impossible. Thus, amputation was expected to provide the best local control. There were three indications for limb-salvage surgery (ie, tumour resection and reconstruction). First, clinical and radiological responses were observed during neoadjuvant chemotherapy. Clinical responses were reduction or stabilisation of tumour size and a significant decrease in pain. Radiological responses were an increase in consolidated calcification of the tumour on plain radiographs, reduction or stabilisation of tumour size, and decreased peritumoral oedema on magnetic resonance images. Second, magnetic resonance images showed no neurovascular involvement. Third, there was no need for extensive muscle resection that would render the limb non-functional. Amputation was considered when patients did not meet the above criteria for limb-salvage surgery. It was also considered as a palliative treatment for patients who had large painful tumours with metastatic disease.

The characteristics of lung nodules identified in chest CT scans were recorded from CT reports. The initial and final sizes, laterality, timing of appearance, and mediastinal lymph node involvement were analysed to determine whether their characteristics were sufficiently different for clear distinction. Non-specific lung nodules were either small or remained stable in subsequent scans; they were not biopsied or surgically excised for histological diagnosis. Lung metastases were either large when first observed, had radiological features of metastasis, or demonstrated enlargement in subsequent follow-up scans.

Descriptive data were expressed as median (interquartile range) or frequency (percentage). The Pearson Chi squared test or Fisher’s exact test was used for comparisons of categorical variables. The Mann-Whitney U test was used for comparisons of continuous variables.

The primary outcomes were OS and EFS. Overall survival was defined as the time from diagnosis to death. Event-free survival was defined as the time from diagnosis to the appearance of a new metastasis, progression of an existing metastasis, or death (whichever occurred first). The study end date was 31 December 2020. Patients who had no events were censored at the time of the last follow-up (if they had been lost to follow-up) or at the study end date. Kaplan-Meier curves and log-rank tests were used for survival analysis. To identify prognostic factors for OS and EFS, unadjusted hazard ratios (with 95% confidence intervals) were determined for each potential factor by using Cox proportional hazards models. Significant factors in univariate analyses were included in subsequent multivariate analyses; adjusted hazard ratios (with 95% confidence intervals) were generated in the multivariate analyses.

The secondary outcome was the differentiation of lung nodules. Their initial and final sizes, initial and final lateralities, timing of appearance, and mediastinal lymph node involvement were compared to identify statistical differences.

The SPSS software (Windows version 23.0; IBM Corp, Armonk [NY], US) was used for statistical analysis. P values <0.05 were considered statistically significant.

In total, 52 paediatric patients (22 girls and 30 boys; age 5-18 years) with high-grade osteosarcoma were included in this study. Their baseline demographics are shown in Table 1. Two patients (3.8%) had a predisposing condition: one had Rothmund–Thomson syndrome and the other had osteofibrous dysplasia, from which the high-grade osteosarcoma developed.

For the histological diagnosis, 45 patients (86.5%) had conventional high-grade osteosarcomas. The other subtypes of osteosarcoma were: two giant cell-rich osteosarcomas, one telangiectatic osteosarcoma, one chondrosarcomatous-predominant osteosarcoma, one osteoblastoma-like osteosarcoma, and one chondroblastic osteosarcoma. One patient had features suggestive of small round-cell sarcoma; the tumour was subsequently treated as a conventional high-grade osteosarcoma.

All patients received neoadjuvant chemotherapy with doxorubicin, cisplatin, and high-dose methotrexate (Fig 1). After two courses of chemotherapy, the patients underwent surgical resection of tumours. Almost all patients (n=50, 96.2%) underwent resections of primary tumours. Most patients (n=44, 84.6%) completed the whole course of treatment with adjuvant chemotherapy. Five patients (9.6%) experienced disease progression during treatment. They had terminated chemotherapy early, received palliative care, and eventually died. One patient (1.9%) had disease progression and left Hong Kong to seek a second opinion. In one patient (1.9%), the last course of chemotherapy was omitted because previous chemotherapy had induced clinically significant renal impairment. In one patient (1.9%), the last course of chemotherapy was omitted because of disease progression while receiving treatment.

The lung was the most common site of distant metastasis, both synchronous (n=10, 90.9%) and metachronous (n=12, 80.0%) [Table 2]. Lung nodules in chest CT scans were observed in 39 patients (75.0%). Seventeen nodules (43.6%) were non-specific, while 22 nodules (56.4%) were lung metastases. Factors that could potentially be used to differentiate lung metastases from non-specific lung nodules included laterality of nodules in serial follow-up scans (P=0.004), number of initial nodules (P=0.006), maximal number of nodules (P<0.001), size of initial nodule (P<0.001), and size of the largest nodule (P<0.001) [Table 3].

For the 15 patients who had no lung nodules throughout the course of treatment, OS was very good (93%) [Fig 2]. Only one patient died of local recurrence. For the 17 patients with non-specific lung nodules (Figs 2 and 3), OS was excellent (100%), regardless of the timing of nodule appearance (ie, at diagnosis or later). Among these 17 patients, survival ranged from 2 years 4 months to 11 years 9 months from diagnosis. For the 10 patients with synchronous lung metastasis (Fig 3), OS was 60%, whereas for the 12 patients with metachronous lung metastasis (Fig 2), OS was 33.3%. This difference was not statistically significant (P=0.666).

All patients were followed up until 31 December 2020; thus, the follow-up interval for the last recruited patient was 2 years from diagnosis. The median follow-up interval for all patients was 53 months from diagnosis (range, 4 months to 11 years 11 months). The EFS and OS were 55.8% and 71.2%, respectively. The EFS for patients with localised disease at diagnosis was significantly better than the EFS for patients with metastatic disease at diagnosis (65.9% vs 18.2%; P=0.007). The OS for patients with localised disease was better than the OS for patients with metastatic disease (75.6% vs 54.5%), but the difference was not statistically significant (P=0.26). Similarly, patients with good tumour chemonecrosis had better EFS and OS, compared with poor responders. These values were 78.3% versus 42.3% (P=0.019) and 82.6% versus 65.4% (P=0.209), respectively. Notably, patients with localised disease and good tumour chemonecrosis had very good outcomes, with EFS of 80% and OS of 86.7%. All deaths in this study were caused by disease progression. No patients died of treatment complications or other causes.

The prognostic factors identified for both EFS and OS included the presence of metastasis at diagnosis and throughout, as well as the need for amputation to manage the primary tumour. Local aggressiveness and a larger tumour dimension contributed to OS, while poor tumour chemonecrosis contributed to EFS (Table 4). The following factors did not have a statistically significant impact on EFS or OS: sex, age at diagnosis, primary tumour location in the femur, the presence of a pathological fracture, time from symptom onset to presentation, time from diagnosis to start of chemotherapy, and the histological subtype of the primary tumour.

Multivariate analysis (Table 5) revealed that the presence of metastasis at diagnosis, the need for amputation, and poor tumour chemonecrosis were prognostic factors independently associated with poor EFS, while the presence of metastasis at diagnosis and the need for amputation were prognostic factors independently associated with poor OS. Figure 4 shows the Kaplan-Meier analysis of the effects of the above three independent prognostic factors on EFS.

To our knowledge, this is the first multicentre review of the demographic characteristics and prognostic factors of paediatric high-grade osteosarcoma in Hong Kong. The number of patients included in the study constituted 76.5% of children with osteosarcoma diagnosed in Hong Kong during the study period (unpublished data). Thus, our findings are likely to be representative of the courses of disease and treatment for children with osteosarcoma in Hong Kong. In this study, slightly more patients with high-grade osteosarcoma were boys, the median age at diagnosis was 13.5 years, and the femur was the most common site of involvement. All of these results are comparable to findings in western countries.1 10 Only one patient had a high-grade osteosarcoma in the humerus; no patients had a primary nodule in the axial skeleton, demonstrating the rarity of such nodules. Although our study was limited by a short follow-up interval in some patients, the EFS and OS were comparable to the findings of large studies conducted in other countries,6 8 10 11 as well as the results of a study performed in Hong Kong in 2009.12 The shortest follow-up interval was 2 years from diagnosis. Most events occurred in the first 2 to 3 years, but some poor responders experienced late relapse at 4 to 5 years after initial diagnosis. Thus, a longer follow-up interval is necessary to better determine patient outcomes and more comprehensively assess the incidence of late toxicity.

In terms of prognostic factors, our findings in a cohort of Hong Kong patients confirm the validity of some important prognostic factors recognised in studies performed elsewhere, including the largest dimension of the primary tumour,8 13 14 presence of metastasis at diagnosis,6 8 9 11 and poor tumour chemonecrosis.6 8 9 13 14 15 Additionally, our study identified the need for amputation as an independent prognostic factor for EFS and OS. With respect to mortality, the prognostic effect of the need for amputation has varied among studies.6 13 14 15 16 This variation is presumably because the decision to amputate depends on many factors, including the opinions of orthopaedic surgeons and parental acceptance. In our study, 10 patients underwent amputation; in two of these patients, the procedure was performed with palliative intent to achieve symptomatic control. Both of those patients had metastatic disease at diagnosis, which involved large and painful primary tumours. Thus, risk stratification of patients according to the need for amputation may enable the selection of patients with more advanced disease. Nonetheless, in our cohort, all surgical margins were negative in both limb-salvage surgery and amputation groups. Moreover, local recurrence was uncommon. Thus, our findings may provide insights that can be used to update risk stratification protocols for paediatric patients with osteosarcoma. In the current treatment protocol, there is a considerable delay between initial diagnosis and risk stratification (at week 16), which is performed after tumour resection and when tumour chemonecrosis data are available. However, the surgical approach is usually determined after approximately 4 to 5 weeks of treatment, when reassessment imaging is conducted. If aggressive features are observed at diagnosis (eg, a massive tumour, early intra-articular tumour involvement, or early neurovascular bundle involvement), the discussion of possible amputation may have already begun. Thus, the presence of such features, or the early recognition of the need for amputation, may be suitable for risk stratification after validation in larger-scale studies.

In recent decades, there has been extensive research into surrogate markers for aggressiveness in osteosarcoma. In some studies, the levels of ALP8 9 14 15 17 and lactate dehydrogenase18 19 20 were identified as significant prognostic factors. Although the maximal ALP level was not associated with EFS or OS in our study, some studies have demonstrated a positive association between the serum ALP level and tumour volume.21 This association is presumably related to the increased rate of bone remodelling in the tumour. However, the serum ALP level is also elevated in teenagers because of increased bone remodelling during periods of rapid growth. Thus, a universal cut-off for all paediatric patients may not provide the greatest prognostic accuracy.22 Some studies23 have explored methods to increase the age specificity of serum ALP levels. However, more validation and larger-scale studies are required before these methods can be widely adopted. Whereas the serum lactate dehydrogenase level showed a weaker association with tumour volume,21 a change in this level is more likely to be associated with a non-specific increase in tumour metabolism. Currently, the serum lactate dehydrogenase level is not included in pretreatment staging and investigation protocols; thus, it is not routinely checked. For investigation purposes, it should also be included in baseline investigations.

Because most biochemical parameters are not accurate surrogate markers for the aggressiveness of osteosarcoma, technological advancements have enabled molecular and genetic profiling of osteosarcomas to become the focus of research in the past 10 to 15 years.22 24 25 These studies may provide insights concerning the ‘non-conforming’ behaviour of certain tumours in our patients; examples include locally aggressive primary tumours that warrant amputation but demonstrate good tumour chemonecrosis, or tumours that show disease progression despite good tumour chemonecrosis. The current literature suggests that paediatric osteosarcoma is a heterogeneous disease,21 26 although general knowledge of the disease remains incomplete. Despite advancements in surgical techniques for primary tumour resection,27 improvements in the accuracy of staging imaging, and enhancements of supportive care, further revisions are needed concerning the medical treatment of paediatric osteosarcoma. Thus, molecular and genetic studies are essential and may facilitate further stratification of patients with osteosarcoma into different risk groups, which may require tailored treatment regimens for better outcomes. Future studies may also enable the identification of molecular nodules for targeted therapy or immunotherapy, which lead to considerable advances in the treatment of osteosarcoma.

Lung nodules were common in the initial staging and serial follow-up CT scans of our patients. The small size of some nodules hindered characterisation. They might represent benign lung pathologies. However, they may also represent micro-metastases which were responsive to chemotherapy. Thus, they remained stable in size and number on subsequent scans, suggesting that they persisted as scars. Patients with non-specific lung nodules had an excellent prognosis, with an OS rate of 100% in our study. However, repeated scans are needed for the follow-up and characterisation of such lung nodules. Additionally, the appearance of any new lung nodules on CT scans creates a considerable psychological burden for patients and their families. Our study identified parameters that can help to differentiate true lung metastases from non-specific nodules, including the number of initial nodules, maximal number of nodules, initial nodule size, and largest nodule size. A study in Hong Kong in 201128 also identified number (≤5 vs >5), size, and laterality of lung nodules as important prognostic factors for survival, whereas a study in the US in 201129 found that survival was worse for central lung metastases than for peripherally located lung metastasis. Additional larger-scale risk stratification studies are needed to clearly delineate the cut-offs for size, number, laterality, and location of initial lung nodules. The establishment of a scoring system that considers parameters of lung nodules observed in chest CT scans may enable prediction of the risk of malignancy, thus improving early detection of lung metastasis and reducing unnecessary anxiety for patients and their families. Furthermore, standardised reviews of CT scans will help to ensure more uniform classification of lung nodules, particularly when the nodules are small.

With respect to confirmed lung metastasis, the situation becomes increasingly complicated. For a single synchronous or metachronous lung metastasis, metastasectomy was conducted whenever surgically feasible because it has been regarded as the primary treatment approach in multiple studies.30 31 32 However, given the diversity of treatments, there was no consensus regarding treatment strategies for multiple metastases or local recurrence. This is presumably because the effects of different chemotherapy regimens and targeted therapies remain under investigation.11 33 34 35 36 In our study, surgical resection of lung metastasis whenever possible, together with zoledronic acid, generally achieved durable clinical remission for >5 years. Long-term randomised controlled trials of different chemotherapy or targeted therapy regimens should be conducted to determine the most cost-effective regimens that improve survival for relapsed paediatric patients with high-grade osteosarcoma. Many centres in other nations are performing karyotyping and genetic analysis of osteosarcoma tumour cells to identify targetable mutations.11 24 37 However, these tests are not routinely conducted for standard care in Hong Kong. Collaborations with centres in other nations should be pursued to facilitate the implementation of international standards in Hong Kong.

To our knowledge, this represents the first multicentre review of paediatric high-grade osteosarcoma in Hong Kong. Important prognostic factors, including metastatic disease at diagnosis and poor tumour chemonecrosis, were validated. The need for amputation may reflect local aggressiveness, which influences OS. Larger-scale studies of high-grade osteosarcoma in paediatric patients should be conducted over a longer period of time to better understand the characteristics, patterns, and prognostic factors.

Concept or design: GPY Tong, CK Li.
Acquisition of data: GPY Tong.
Analysis or interpretation of data: GPY Tong, WF Hui, CK Li.
Drafting of the manuscript: GPY Tong, WF Hui, KC Wong, CK Li.
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.

The authors have no conflicts of interest to disclose.

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

This study was approved by the Research Ethics Committee (Kowloon Central/Kowloon East Cluster and New Territories East Cluster), Hospital Authority Hong Kong (Ref: KC/KE-19-0301/ER-1, 2019.712). The requirement for patient informed consent was waived.

1. Ottaviani G, Norman J. The epidemiology of osteosarcoma. In: Jaffe N, Bruland OS, Bielack S, editors. Pediatric and Adolescent Osteosarcoma. Boston, MA: Springer; 2009: 3-13. Crossref

2. North American Association of Central Cancer Registries (NAACCR), 2021.

3. Hong Kong Cancer Registry. Hong Kong Cancer Statistics 2015-2019.

4. Huvos AG, Rosen G, Marcove RC. Primary osteogenic sarcoma: pathologic aspects in 20 patients after treatment with chemotherapy en bloc resection, and prosthetic bone replacement. Arch Pathol Lab Med 1977;101:14-8.

5. Hung GY, Yen HJ, Yen CC, et al. Experience of pediatric osteosarcoma of the extremity at a single institution in Taiwan: prognostic factors and impact on survival. Ann Surg Oncol 2015;22:1080-7. Crossref

6. Pakos EE, Nearchou AD, Grimer RJ, et al. Prognostic factors and outcomes for osteosarcoma: an international collaboration. Eur J Cancer 2009;45:2367-75. Crossref

7. Abou Ali B, Salman M, Ghanem KM, et al. Clinical prognostic factors and outcome in pediatric osteosarcoma: effect of delay in local control and degree of necrosis in a multidisciplinary setting in Lebanon. J Glob Oncol 2019;5:1-8. Crossref

8. Vasquez L, Tarrillo F, Oscanoa M, et al. Analysis of prognostic factors in high-grade osteosarcoma of the extremities in children: a 15-year single-institution experience. Front Oncol 2016;6:22. Crossref

9. Mialou V, Philip T, Kalifa C, et al. Metastatic osteosarcoma at diagnosis: prognostic factors and long-term outcome—the French pediatric experience. Cancer 2005;104:1100-9. Crossref

10. Marina NM, Smeland S, Bielack SS, et al. Comparison of MAPIE versus MAP in patients with a poor response to preoperative chemotherapy for newly diagnosed highgrade osteosarcoma (EURAMOS-1): an open-label, international, randomised controlled trial. Lancet Oncol 2016;17:1396-408. Crossref

11. Gorlick R, Janeway K, Lessnick S, Randall RL, Marina N. Children’s Oncology Group’s 2013 blueprint for research: bone tumors. Pediatr Blood Cancer 2013;60:1009-15. Crossref

12. Yang JY, Cheng FW, Wong KC, et al. Initial presentation and management of osteosarcoma, and its impact on disease outcome. Hong Kong Med J 2009;15:434-9.

13. Xin S, Wei G. Prognostic factors in osteosarcoma: a study level meta-analysis and systematic review of current practice. J Bone Oncol 2020;21:100281. Crossref

14. Bramer JA, van Linge JH, Grimer RJ, Scholten RJ. Prognostic factors in localized extremity osteosarcoma: a systematic review. Eur J Surg Oncol 2009;35:1030-6. Crossref

15. Hu J, Zhang C, Zhu K, et al. Treatment-related prognostic factors in managing osteosarcoma around the knee with limb salvage surgery: a lesson from a long-term follow-up study. Biomed Res Int 2019;2019:3215824. Crossref

16. Jauregui JJ, Nadarajah V, Munn J, et al. Limb salvage versus amputation in conventional appendicular osteosarcoma: a systematic review. Indian J Surg Oncol 2018;9:232-40. Crossref

17. Zamzam MA, Moussa EA, Ghoneimy AE, et al. Outcomes and prognostic factors for non-metastatic osteosarcoma of the extremity. SM J Pediatr 2017;2:1013.

18. González-Billalabeitia E, Hitt R, Fernández J, et al. Pre-treatment serum lactate dehydrogenase level is an important prognostic factor in high-grade extremity osteosarcoma. Clin Transl Oncol 2009;11:479-83. Crossref

19. Fu Y, Lan T, Cai H, Lu A, Yu W. Meta-analysis of serum lactate dehydrogenase and prognosis for osteosarcoma. Medicine (Baltimore) 2018;97:e0741. Crossref

20. Chen J, Sun M, Hua Y, Cai Z. Prognostic significance of serum lactate dehydrogenase level in osteosarcoma: a meta-analysis. J Cancer Res Clin Oncol 2014;140:1205-10. Crossref

21. Lindsey BA, Markel JE, Kleinerman ES. Osteosarcoma overview. Rheumatol Ther 2017;4:25-43. Crossref

22. Savitskaya YA, Rico-Martínez G, Linares-González LM, et al. Serum tumor markers in pediatric osteosarcoma: a summary review. Clin Sarcoma Res 2012;2:9. Crossref

23. Shimose S, Kubo T, Fujimori J, Furuta T, Ochi M. A novel assessment method of serum alkaline phosphatase for the diagnosis of osteosarcoma in children and adolescents. J Orthop Sci 2014;19:997-1003.Crossref

24. de Nigris F, Zanella L, Cacciatore F, et al. YY1 overexpression is associated with poor prognosis and metastasis-free survival in patients suffering osteosarcoma. BMC Cancer 2011;11:472. Crossref

25. Morrow JJ, Khanna C. Osteosarcoma genetics and epigenetics: emerging biology and candidate therapies. Crit Rev Oncog 2015;20:173-97. Crossref

26. Poos K, Smida J, Maugg D, et al. Genomic heterogeneity of osteosarcoma–shift from single candidates to functional modules. PLoS One 2015;10:e0123082. Crossref

27. Wong KC, Kumta SM. Use of computer navigation in orthopedic oncology. Curr Surg Rep 2014;2:47. Crossref

28. Rasalkar DD, Chu WC, Lee V, Paunipagar BK, Cheng FW, Li CK. Pulmonary metastases in children with osteosarcoma: characteristics and impact on patient survival. Pediatr Radiol 2011;41:227-36. Crossref

29. Letourneau PA, Xiao L, Harting MT, et al. Location of pulmonary metastasis in pediatric osteosarcoma is predictive of outcome. J Pediatr Surg 2011;46:1333-7. Crossref

30. Daw NC, Chou AJ, Jaffe N, et al. Recurrent osteosarcoma with a single pulmonary metastasis: a multi-institutional review. Br J Cancer 2015;112:278-82. Crossref

31. Matsubara E, Mori T, Koga T, et al. Metastasectomy of pulmonary metastases from osteosarcoma: prognostic factors and indication for repeat metastasectomy. J Respir Med 2015;2015:1-5. Crossref

32. de Bree E, Drositis I, Michelakis D, Mavroudis D. Resection of pulmonary metastases in osteosarcoma. Is it justified? Hellenic J Surg 2018;90:293-8. Crossref

33. Geller DS, Gorlick R. Osteosarcoma: a review of diagnosis, management, and treatment strategies. Clin Adv Hematol Oncol 2010;8:705-18.

34. Warwick AB, Malempati S, Krailo M, et al. Phase 2 trial of pemetrexed in children and adolescents with refractory solid tumors: a Children’s Oncology Group study. Pediatr Blood Cancer 2013;60:237-41. Crossref

35. Jacobs S, Fox E, Krailo M, et al. Phase II trial of ixabepilone administered daily for five days in children and young adults with refractory solid tumors: a report from the children’s oncology group. Clin Cancer Res 2010;16:750-4. Crossref

36. Geoerger B, Chisholm J, Le Deley MC, et al. Phase II study of gemcitabine combined with oxaliplatin in relapsed or refractory paediatric solid malignancies: an innovative therapy for children with Cancer European Consortium Study. Eur J Cancer 2011;47:230-8. Crossref

37. Saraf AJ, Fenger JM, Roberts RD. Osteosarcoma: accelerating progress makes for a hopeful future. Front Oncol 2018;8:4. Crossref

Perinatal and neonatal mortality rates are important measures of the quality of medical care during pregnancy, childbirth, and the neonatal period. Although the global neonatal death (NND) rate has demonstrated a decreasing trend over the past 30 years, from 37/1000 livebirths to 17/1000 livebirths between 1990 and 2020,1 NND rates considerably vary among regions. According to the 2020 report by the World Health Organization, the NND rates were the highest in Africa (27/1000 livebirths), Eastern Mediterranean (25/1000 livebirths) and South-East Asia (18/1000 livebirths); the NND rates were the lowest in Americas (7/1000 livebirths), Western Pacific (5/1000 livebirths) and Europe (4/1000 livebirths).1 In Hong Kong, territory-wide statistics indicated NND rates of 1.2/1000 and 1.0/1000 livebirths in 2004 and 2014, respectively2; these rates are lower than the rates in most regions, according to the above report by the World Health Organization. However, there have been few in-depth studies concerning the trends and underlying causes of NND in Hong Kong. Our group recently published two epidemiological studies regarding singleton pregnancies in Hong Kong, which revealed a decreasing trend in the rate of stillbirths among singleton pregnancies from 3.61/1000 in 2000-2009 to 3.09/1000 in 2010-2019.3 The rate of perinatal mortality in multiple pregnancies also decreased from 5.52/1000 to 4.59/1000 during the same period.4 These improvements in mortality rates have mainly occurred because of advances in the prenatal diagnosis and management of fetal malformations and genetic diseases, as well as improvements in the antenatal management of multiple pregnancies. The present study investigated the trends of NNDs among singleton pregnancies in the largest tertiary perinatal centre in Hong Kong, as well as changes in the characteristics and aetiologies of NND over the past two decades, with the goal of improving perinatal care in Hong Kong.

This retrospective study included all singleton pregnancies that delivered at the Prince of Wales Hospital from 1 January 2000 to 31 December 2019. The STROBE reporting guideline was followed when writing this manuscript. The Prince of Wales Hospital is affiliated with The Chinese University of Hong Kong and serves a large population of 1.7 million in the New Territories East region of Hong Kong; the hospital’s annual delivery rate is 6000 to 7000 (approximately one-sixth of the total births in all public hospitals in Hong Kong, and one-ninth of the total births in Hong Kong). Both the obstetric unit and the neonatal unit are the largest in Hong Kong. The neonatal unit is a Level III centre that consists of a 22-bed neonatal intensive care unit (NICU). The staff of the neonatal unit worked closely with the staff of the obstetric unit to manage high-risk deliveries from complicated pregnancies, as well as pregnancies that required fetal intervention after referral from other hospitals.

Complicated pregnancies were discussed in weekly perinatal meetings attended by the staff of both the obstetric unit and the neonatal unit; discussions of these pregnancies focused on management plans and the optimal timing of delivery. Relevant disciplines (eg, paediatric surgery, cardiology, neurosurgery, radiology, or otolaryngology) were included as appropriate. In cases where a specialist service outside of Prince of Wales Hospital (eg, cardiothoracic surgery) was anticipated after delivery, specialists from other centres were invited to participate in management planning. Active resuscitation was provided for all viable neonates delivered at ≥24 weeks. For extremely premature neonates with borderline viability (ie, delivered at 22-23 weeks of gestation), considering the high risks of mortality and long-term morbidity, comprehensive counselling was provided to affected families, which allowed them to select active resuscitation or no resuscitation at birth. In accordance with the departmental protocol, the paediatric unit was requested to prepare for rapid management of deliveries that involved specific maternal or fetal conditions (eg, prematurity, fetal distress, instrumental deliveries, and antenatally diagnosed severe congenital abnormalities). Neonates were managed in the NICU in accordance with the standard unit protocols and guidelines. These protocols were updated regularly according to the latest evidence-based consensus guidelines and recommendations established by clinicians in Hong Kong and other nations. In accordance with departmental guidelines, comprehensive investigations were performed to determine the cause of death in all cases of NND. All NNDs were referred for autopsy unless the cause of death was clearly identified (eg, trisomy 13 or 18 confirmed by genetic tests). If the cause of NND could not be identified, the case was reported to the coroner. If NND was caused by multiple pathologies, the most clinically significant pathology that contributed to death was selected for analysis.

All cases of NND in livebirths of singleton pregnancies during the study period were retrieved using the Hospital Authority’s Clinical Data Analysis and Reporting System. Cases of NND in livebirths of multiple pregnancies were excluded. The included cases of NND were divided into two groups according to the decade of birth. The first group (ie, first decade) included cases of NND among singleton pregnancies that delivered between 1 January 2000 and 31 December 2009. The second group (ie, second decade) included cases of NND among singleton pregnancies that delivered between 1 January 2010 and 31 December 2019. Obstetric data including maternal demographics (maternal age, maternal illnesses, antenatal complications, and treatment) and birth history (gestation, mode of delivery, sex, birth weight, Apgar scores, and neonatal resuscitation) were collected from the Obstetric Specialty Clinical Information System. Neonatal data comprising neonatal diagnoses, interventions, and length of survival were retrieved from the Hospital Authority’s Clinical Management System. When further details were needed, individual case records were retrieved for analysis.

All NNDs were categorised as early NND (within 7 days after birth) or late NND (within 8-28 days after birth). Early, late, and total rates of NND, as well as baseline demographics, were compared between the two groups. Causes of death were divided into four main categories: prematurity, hypoxic-ischaemic encephalopathy (HIE), congenital abnormalities, and sepsis. Congenital abnormalities were defined by characteristic features on physical examination, confirmed by either genetic tests, diagnostic investigations, or autopsy. Sepsis was defined on the basis of positive cultures established using samples of blood, urine, cerebrospinal fluid, or tissue from the affected neonate. Hypoxic-ischaemic encephalopathy was diagnosed in accordance with criteria derived from international guidelines.5

The analysis was performed using data that overlapped with a previous study.3 Categorical variables were compared by the Chi squared test or Fisher’s exact test. The threshold of statistical significance was defined as a two-sided P value of <0.05. Data analysis was performed with the SPSS software (Windows version 22.0; IBM Corp, Armonk [NY], United States).

There were 124 281 livebirths from singleton pregnancies between 2000 and 2019 (Table 1). The number of livebirths increased by 12.7% from 58 442 in the first decade (2000-2009) to 65 839 in the second decade (2010-2019). There were 184 NNDs (1.48/1000 livebirths) between 2000 and 2019, including 97 in the first decade (1.66/1000 livebirths) and 87 in the second decade (1.32/1000 livebirths). Overall, there were 136 cases (73.9%) of early NND and 48 cases (26.1%) of late NND.

The maternal demographic characteristics of all singleton pregnancies during the study period were reported in our previous paper.3 The distribution of gestational age among all livebirths did not differ between the two decades (P=0.237) [Table 2]. The highest rate of NND (195.12/1000 livebirths) was observed in extremely preterm neonates (≤27 weeks of gestation) [Table 3]. The rate of NND decreased with increasing gestational age, such that NND rates were 48.42/1000, 20.13/1000, 4.98/1000, and 0.37/1000 for neonates delivered at gestational ages of 28-30 weeks, 31-33 weeks, 34-36 weeks, and ≥37 weeks, respectively. Compared with the first decade, there was a significant reduction (75.2%) in the rate of NND among neonates delivered at 31-33 weeks of gestation during the second decade (34.73/1000 vs 8.63/1000; P=0.001); however, there were no significant differences in the rates of NND among neonates in other gestational groups.

The primary causes of NND in the two decades are shown in Table 4. Congenital or genetic abnormalities was the most common cause of NND (82 of 184; 44.6%) during the 20-year study period. Other common causes of NND were prematurity (69 cases; 37.5%), sepsis (16 cases; 8.7%), and HIE (15 cases; 8.2%) [Supplementary Fig].

Chromosomal abnormalities caused 18.3% (15 of 82) of NNDs related to congenital or genetic abnormalities; all of these abnormalities were trisomy 13 or 18. Structural abnormalities caused 63.4% (52 of 82) of NNDs related to congenital or genetic abnormalities, and respiratory system abnormalities were the most common causes in both decades (22 cases). These respiratory system abnormalities included congenital diaphragmatic hernia (13 cases), pulmonary hypoplasia (5 cases), alveolar capillary dysplasia (2 cases), and tracheal stenosis or atresia (2 cases). The next most common causes were congenital cardiac abnormalities (8 cases), including transposition of the great arteries (2 cases), total anomalous pulmonary venous drainage (2 cases), endocardial cushion defect (2 cases), hypoplastic left heart syndrome (1 case), and congenital heart block (1 case); central nervous system abnormalities (8 cases), including anencephaly (3 cases), central nervous system malformation (4 cases), and brain tumour (1 case); and musculoskeletal abnormalities (7 cases), including fetal akinesia syndrome with arthrogryposis (3 cases), spinal muscular atrophy (2 cases), and skeletal dysplasia (2 cases). There were two cases of gastrointestinal abnormalities (volvulus and bowel atresia with meconium peritonitis), two cases of sacrococcygeal teratoma, two cases of multiple abnormalities, and one case of bilateral renal agenesis (a urogenital abnormality). There were also nine cases of haemoglobin Barts disease and six cases of idiopathic hydrops. There was a statistically significant decline in the rate of NND caused by congenital or genetic abnormalities, from 0.82/1000 livebirths in the first decade to 0.52/1000 livebirths in the second decade (P=0.037).

There were no significant differences in the rates of NND caused by prematurity, sepsis, or HIE between the two decades. Cases of NND due to sepsis were mainly caused by Group B Streptococcus in the first decade and Escherichia coli in the second decade. The majority of HIE cases (67.7%) were related to acute intrapartum events, including placenta abruption (5 cases), uterine rupture (2 cases), vasa praevia (1 case), cord accident (1 case), and chorioamnionitis (1 case).

The NND rate in our tertiary centre is consistent with the rate of 1.2/1000 livebirths in the territory-wide report2 and lower than the rates in many developed countries (eg, the United States, Australia, and nations located in Europe; neonatal mortality rates of 2-3/1000 livebirths).1 2 The global NND rate has been decreasing over the past two decades because of advances in perinatal care.2 Our overall NND rate decreased by 20%, from 1.66/1000 in the first decade to 1.32/1000 in the second decade. This decrease is mainly the result of a decrease in NNDs related to congenital or genetic disorders, as well as a decrease in NNDs among neonates delivered at 31-33 weeks of gestation.

Similar to our previous report, which showed a reduction in the rate of congenital or genetic abnormality-related stillbirths,3 the present study showed that the rate of congenital or genetic abnormality-related NNDs decreased from 0.82/1000 livebirths in the first decade to 0.52/1000 livebirths in the second decade. This decline was presumably because of improvements in antenatal screening and the early detection of lethal congenital abnormalities, which resulted in termination of pregnancy before 24 weeks of gestation. Universal first trimester combined screening for Down syndrome was implemented by the Hospital Authority in 2010.6 In 2011, non-invasive cell-free fetal DNA tests for common trisomies, as well as chromosomal microarrays for the diagnosis of chromosomal microdeletion syndromes, became available in the private sector.7 8 Expanded antenatal screening of inborn errors of metabolism was launched in the private sector in 2013; this expanded screening has gradually become available in the public sector since 2018.9 Although we expected a decline in the rate of trisomy-related NNDs after universal aneuploidy screening became available in 2011, there was an increase in the rate of trisomy 18–related NNDs (from 2 cases to 7 cases). A review of the individual cases revealed that the rate of trisomy 13–related NNDs decreased from six cases in the first decade to none in the second decade. Conversely, five of the seven cases of trisomy 18–related NND in the second decade were in pregnancies that had not received any screening; all of these five cases occurred during the period from 2010 to 2013. The other two cases of trisomy 18–related NND were diagnosed during prenatal screening, but the parents chose conservative management rather than termination of pregnancy. To further reduce mortality associated with hereditary genetic disorders such as spinal muscular atrophy and fetal akinesia syndrome (which caused NND in 5 cases), there is a need for expanded carrier screening of parents, particularly in families with a history of consanguineous marriage.10 11

The other main congenital abnormalities that caused NND in our cohort were cardiorespiratory and neuromusculoskeletal disorders, among which congenital diaphragmatic hernia was the most common. Although survival was common among neonates with mild to moderate congenital diaphragmatic hernia, neonates with severe congenital diaphragmatic hernia had a survival rate of 10% to 20% because of pulmonary hypoplasia. A recent large randomised controlled trial showed that fetoscopic endoluminal tracheal occlusion can improve the survival rate to 40% to 50%.12 In our unit, a baby survived after treatment with fetoscopic endoluminal tracheal occlusion in 2020.13 Pulmonary hypoplasia caused by hydrothorax or lung tumours can also be effectively and safely treated before birth with newly designed instruments such as the Somatex^® shunt for pleuro-amniotic shunting, and radiofrequency ablation of the tumour feeding artery, respectively.14 15 Fetal tumours such as sacrococcygeal teratoma, placental chorioangioma, and lung tumours remain challenging to manage because the rapid growth of tumours in utero increases the risk of preterm birth and leads to impaired neonatal cardiac function. We have demonstrated improvements in survival after in utero embolisation of chorioangioma using cyanoacrylate, and after in utero radiofrequency ablation of lung sequestration.15 16 Although spinal muscular atrophy has no cure, it can be prevented by accurate parental carrier screening using genomic technology and prenatal diagnosis.10

The rate of idiopathic hydrops fetalis–related NND decreased from 5.2% in the first decade to 1.1% in the second decade. Advances in antenatal diagnostic techniques in recent years have identified the underlying causes of many conditions which may have previously been regarded as ‘idiopathic hydrops fetalis’.17 The early diagnosis of treatable conditions in the antenatal period can prevent the development of severe hydrops fetalis and subsequent NND.17 18 Intrauterine blood transfusion for fetal anaemia and anti-arrhythmic treatment19 has significantly reduced the rate of hydrops fetalis, resulting in improved survival and long-term outcomes.

Our study showed a significant (75.2%) decrease in the rate of NND among moderately preterm neonates (31-33 weeks of gestation) from 34.73/1000 in 2000-2009 to 8.63/1000 in 2010-2019; however, the rate of NND did not change in other gestational groups (Table 3). The decrease in mortality among moderately preterm neonates could be attributed to the implementation of multiple approaches for the management of such neonates since 2010, including improved ventilation strategies with early extubation to non-invasive ventilation, new methods for surfactant administration (eg, the ‘less invasive surfactant administration’ method), and improvements in NICU care through continuous quality improvement programmes. The rate of NND among extremely preterm neonates (24-27 weeks of gestation) was 175-211/1000, which is comparable with the rates in other developed countries (139-326/1000).20 It is difficult to reduce the rate of NND among extremely preterm neonates. Research is ongoing regarding artificial placenta and womb technology, and the results may improve the survival of extremely preterm neonates in the future.21

The rate of prematurity-related NND can be reduced by preventing preterm delivery; however, this prevention remains a challenging goal. Although our overall preterm delivery rate of 7% is lower than the rates in other developed countries,1 22 it has remained at this level for the past two decades, and there has been no variations in gestation age-specific neonatal mortality among preterm categories. In a previous study, we demonstrated that measurements of cervical length can help to identify pregnant women who are at higher risk of preterm delivery, although the risk prediction values for Chinese women in Hong Kong are lower than the corresponding values for women in non-Asian countries.23 Additional methods to predict the onset of labour (eg, cervical elastography, immune markers, and genetic markers) should be explored to improve accuracy.24 25 Prophylactic progesterone is effective in reducing the risk of preterm delivery among women who have a short cervix.26 Although the use of a cervical ring pessary reportedly had a similar effect in a Spanish study,27 this result was not confirmed by a randomised controlled trial in Hong Kong28 or by subsequent meta-analysis.29 Pre-eclampsia is a common complication that requires medically induced preterm delivery. First trimester screening of pre-eclampsia, followed by prophylactic aspirin treatment in high-risk cases, is a proven strategy to effectively delay the onset of pre-eclampsia and the associated preterm births.29 Our recent study confirmed the accuracy of a screening programme for pre-eclampsia.30 Reductions in pre-eclampsia–related preterm births and mortality may be achieved by the implementation of a universal screening programme in the future.

Despite advances in NICU management of HIE and the use of therapeutic hypothermia since 2011, the rate of HIE-related NND did not improve during the study period. Approximately 67% of HIE-related NNDs were caused by acute and unpredictable perinatal events such as cord prolapse, uterine rupture, vasa praevia, or placental abruption. We previously reported an infant death secondary to severe cerebral palsy as a result of prolonged shoulder dystocia, which occurred during the first study decade.31 Therefore, team-based training for the above perinatal events is needed to ensure that the obstetric team can respond appropriately and efficiently so that the risk of HIE and associated perinatal mortality can be reduced. During these situations that involved irreversible peripartum hypoxia, we showed that umbilical cord arterial pH decreased as the length of the bradycardia-to-delivery interval increased.31 32 33 With appropriate training, we were able to achieve a median bradycardia-to-delivery interval of 10 minutes and a median decision-to-delivery interval of 11 minutes,32 which was effective in preventing peripartum mortality. Furthermore, we showed that during umbilical cord prolapse, the knee-chest position is the most effective approach for relieving fetal compression of the prolapsed cord34; we also formulated an algorithm for acute resolution of cord prolapse.35 Shoulder dystocia is associated with macrosomia, but the optimal fetal weight cut-off for prophylactic elective caesarean delivery has not been established. Our previous study suggested a cut-off of 4.2 kg may help to prevent shoulder dystocia.36 With effective training and correct use of manoeuvres such as posterior arm delivery, we recently showed that the head-to-delivery interval can be shortened and the Apgar scores can be improved.37 38 We also proposed a modified posterior axillary sling technique to relieve severe shoulder dystocia.39

The rate of severe sepsis-related NND is low and has been decreasing over the past two decades. Since the implementation of universal Group B Streptococcus screening and peripartum antibiotic prophylaxis in 2012, the rate of early onset Group B Streptococcus infection has significantly decreased from 1/1000 to 0.24/1000 births.40 Despite the reduced risk of neonatal Group B Streptococcus infection, recent reports have shown an increase in Escherichia coli–related early-onset neonatal sepsis.41 Clinicians should remain vigilant concerning the presence of chorioamnionitis and risk factors for sepsis.

To our knowledge, this is the largest and most comprehensive analysis of neonatal mortality during a 20-year period in Hong Kong. Nevertheless, there were a few limitations in this study. First, it was performed in a single large centre, rather than in a large segment of the population. Because the Prince of Wales Hospital is the main centre for fetal intervention in Hong Kong, many high-risk pregnancies are referred from adjacent hospitals, which may have led to an over-representation of complex cases and a bias towards worse outcomes. Second, some case details were not available for analysis because of the retrospective nature of the study. Third, our study excluded cases of NND among neonates with borderline viability (gestational age: 22-23 weeks and 6 days) because such NNDs are regarded as miscarriages based on the legal definition in Hong Kong. Although some parents of neonates with borderline viability requested resuscitation, the survival rate in this small group was zero according to a recent study in our centre.42 Finally, because the rate of NND is very low in Hong Kong, this study could have been strengthened by including data regarding the rates of major morbidities (eg, cerebral palsy). Nonetheless, our findings provide a basis for future territory-wide reviews of perinatal outcomes.

Hong Kong has one of the lowest rates of NND worldwide. The neonatal mortality in our centre has decreased from 1.66/1000 livebirths to 1.32/1000 livebirths over the past two decades, mainly because of improvements in the prenatal diagnosis and treatment of congenital or genetic abnormalities, as well as an improved survival rate among moderately preterm neonates. Future improvements should focus on in utero treatment, expanded carrier screening for genetic abnormalities, and the prevention of preterm birth and pre-eclampsia.

Concept or design: GPG Fung, TY Leung.
Acquisition of data: All authors.
Analysis or interpretation of data: GPG Fung, TY Leung.
Drafting of the manuscript: GPG Fung, TY Leung.
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.

Ethical approval was obtained from the Joint Chinese University of Hong Kong–New Territories East Cluster Clinical Research Ethics Committee (Ref No.: CRE 2017.442).

1. World Health Organization. Neonatal mortality rate (0 to 27 days per 1000 live births) (SDG 3.2.2). Available from: https://www.who.int/data/gho/indicator-metadata-registry/imr-details/67. Accessed 5 Oct 2022.

2. Hong Kong College of Obstetricians & Gynaecologists. Territory-wide audit in obstetrics & gynaecology 2014. Available from: https://www.hkcog.org.hk/hkcog/Download/Territory-wide_Audit_in_Obstetrics_Gynaecology_2014.pdf. Accessed 15 Jun 2022.

3. Wong ST, Tse WT, Lau SL, Sahota DS, Leung TY. Stillbirth rate in singleton pregnancies: a 20-year retrospective study from a public obstetric unit in Hong Kong. Hong Kong Med J 2022;28:285-3. Crossref

4. Lau SL, Wong ST, Tse WT, et al. Perinatal mortality rate in multiple pregnancies: a 20-year retrospective study from a tertiary obstetric unit in Hong Kong. Hong Kong Med J 2022;28:347-56. Crossref

5. Executive summary: Neonatal encephalopathy and neurologic outcome, second edition. Report of the American College of Obstetricians and Gynecologists’ Task Force on Neonatal Encephalopathy [editorial]. Obstet Gynecol 2014;123:896-901. Crossref

6. Sahota DS, Leung WC, Chan WP, To WW, Lau ET, Leung TY. Prospective assessment of the Hong Kong Hospital Authority universal Down syndrome screening programme. Hong Kong Med J 2013;19:101-8.

7. Chan YK, Leung WC, Leung TY, et al. Women’s preference for non-invasive prenatal DNA testing versus chromosomal microarray after screening for Down syndrome: a prospective study. BJOG 2018;125:451-9. Crossref

8. Hui AS, Chau MH, Chan YM, et al. The role of chromosomal microarray analysis among fetuses with normal karyotype and single system anomaly or nonspecific sonographic findings. Acta Obstet Gynecol Scand 2021;100:235-43. Crossref

9. Chong SC, Law LK, Hui J, Lai CY, Leung TY, Yuen YP. Expanded newborn metabolic screening programme in Hong Kong: a three-year journey. Hong Kong Med J 2017;23:489-96. Crossref

10. Chan OY, Leung TY, Cao Y, et al. Expanded carrier screening using next-generation sequencing of 123 Hong Kong Chinese families: a pilot study. Hong Kong Med J 2021;27:177-83. Crossref

11. Siong KH, Au Yeung SK, Leung TY. Parental consanguinity in Hong Kong. Hong Kong Med J 2019;25:192-200. Crossref

12. Deprest JA, Nicolaides KH, Benachi A, et al. Randomized trial of fetal surgery for severe left diaphragmatic hernia. N Engl J Med 2021;385:107-18. Crossref

13. TOPick. 3次流產40歲婦終懷孕胎兒27周時橫膈膜穿洞威爾斯婦產科團隊施宮內手術力保胎兒順利出世. Available from: https://topick.hket.com/article/2757416. Accessed 15 Jun 2022.

14. Chung MY, Leung WC, Tse WT, et al. The use of Somatex Shunt for fetal pleural effusion: a cohort of 8 procedures. Fetal Diagn Ther 2021;48:440-7. Crossref

15. Tse WT, Poon LC, Wah YM, Hui AS, Ting YH, Leung TY. Bronchopulmonary sequestration successfully treated with prenatal radiofrequency ablation of the feeding artery. Ultrasound Obstet Gynecol 2021;58:325-7. Crossref

16. Cheng YK, Yu SC, So PL, Leung TY. Ultrasound-guided percutaneous embolisation of placental chorioangioma using cyanoacrylate. Fetal Diagn Ther 2017;41:76-9. Crossref

17. Swearingen C, Colvin ZA, Leuthner SR. Nonimmune hydrops fetalis. Clin Perinatol 2020;47:105-21. Crossref

18. Songdej D, Babbs C, Higgs DR; BHFS International Consortium. An international registry of survivors with Hb Bart’s hydrops fetalis syndrome. Blood 2017;129:1251-9. Crossref

19. Donofrio MT, Moon-Grady AJ, Hornberger LK, et al. Diagnosis and treatment of fetal cardiac disease: a scientific statement from the American Heart Association. Circulation 2014;129:2183-242. Crossref

20. Ancel PY, Goffinet F; EPIPAGE-2 Writing Group, et al. Survival and morbidity of preterm children born at 22 through 34 weeks’ gestation in France in 2011: results of the EPIPAGE-2 cohort study. JAMA Pediatr 2015;169:230-8. Crossref

21. De Bie FR, Davey MG, Larson AC, Deprest J, Flake AW. Artificial placenta and womb technology: past, current, and future challenges towards clinical translation. Prenat Diagn 2021;41:145-58. Crossref

22. Hui AS, Lao TT, Leung TY, Schaaf JM, Sahota DS. Trends in preterm birth in singleton deliveries in a Hong Kong population. Int J Gynaecol Obstet 2014;127:248-53. Crossref

23. Leung TN, Pang MW, Leung TY, Poon CF, Wong SM, Lau TK. Cervical length at 18-22 weeks of gestation for the prediction of spontaneous preterm delivery in Hong Kong Chinese women. Ultrasound Obstet Gynecol 2005;25:713-7. Crossref

24. Feng Q, Chaemsaithong P, Duan H, et al. Screening for spontaneous preterm birth by cervical length and shear-wave elastography in the first trimester of pregnancy. Am J Obstet Gynecol 2022;227:500.e1-14. Crossref

25. Chim SS, Lee WS, Ting YH, Chan OK, Lee SW, Leung TY. Systematic identification of spontaneous preterm birth-associated RNA transcripts in maternal plasma. PLoS One 2012;7:e34328. Crossref

26. Romero R, Nicolaides KH, Conde-Agudelo A, et al. Vaginal progesterone decreases preterm birth ≤34 weeks of gestation in women with a singleton pregnancy and a short cervix: an updated meta-analysis including data from the OPPTIMUM study. Ultrasound Obstet Gynecol 2016;48:308-17. Crossref

27. Goya M, Pratcorona L, Merced C, et al. Cervical pessary in pregnant women with a short cervix (PECEP): an open-label randomised controlled trial. Lancet 2012;379:1800-6. Crossref

28. Hui SY, Chor CM, Lau TK, Lao TT, Leung TY. Cerclage pessary for preventing preterm birth in women with a singleton pregnancy and a short cervix at 20 to 24 weeks: a randomized controlled trial. Am J Perinatol 2013;30:283-8. Crossref

29. Conde-Agudelo A, Romero R, Nicolaides KH. Cervical pessary to prevent preterm birth in asymptomatic high-risk women: a systematic review and meta-analysis. Am J Obstet Gynecol 2020;223:42-65.e2. Crossref

30. Chaemsaithong P, Pooh RK, Zheng M, et al. Prospective evaluation of screening performance of first trimester prediction models for preterm preeclampsia in Asian population. Am J Obstet Gynecol 2019;221:650.e1-16. Crossref

31. Leung TY, Stuart O, Sahota DS, Suen SS, Lau TK, Lao TT. Head-to-body delivery interval and risk of fetal acidosis and hypoxic ischaemic encephalopathy in shoulder dystocia: a retrospective review. BJOG 2011;118:474-9. Crossref

32. Leung TY, Chung PW, Rogers MS, Sahota DS, Lao TT, Chung TK. Urgent cesarean delivery for fetal bradycardia. Obstet Gynecol 2009;114:1023-8. Crossref

33. Wong L, Tse WT, Lai CY et al. Bradycardia-to-delivery interval and fetal outcomes in umbilical cord prolapse. Acta Obstet Gynecol Scand 2021;100:170-7. Crossref

34. Kwan AH, Chaemsaithong P, Wong L, et al. Transperineal ultrasound assessment of fetal head elevation by maneuvers used for managing umbilical cord prolapse. Ultrasound Obstet Gynecol 2021;58:603-8. Crossref

35. Wong L, Kwan AH, Lau SL, Sin WT, Leung TY. Umbilical cord prolapse: revisiting its definition and management. Am J Obstet Gynecol 2021;225:357-66. Crossref

36. Cheng YK, Lao TT, Sahota DS, Leung VK, Leung TY. Use of birth weight threshold for macrosomia to identify fetuses at risk of shoulder dystocia among Chinese populations. Int J Gynaecol Obstet 2013;120:249-53. Crossref

37. Chan EH, Lau SL, Leung TY. Changes in the incidence and management of shoulder dystocia over 20 years from a tertiary obstetric unit in Hong Kong. Hong Kong Med J. In press.

38. Leung TY, Stuart O, Suen SS, Sahota DS, Lau TK, Lao TT. Comparison of perinatal outcomes of shoulder dystocia alleviated by different type and sequence of manoeuvres: a retrospective review. BJOG 2011;118:985-90. Crossref

39. Kwan AH, Hui AS, Lee JH, Leung TY. Intrauterine fetal death followed by shoulder dystocia and birth by modified posterior axillary sling method: a case report. BMC Pregnancy Childbirth 2021;21:672. Crossref

40. Ma TW, Chan V, So CH, et al. Prevention of early onset group B streptococcal disease by universal antenatal culture-based screening in all public hospitals in Hong Kong. J Matern Fetal Neonatal Med 2018;31:881-7. Crossref

41. Stoll BJ, Puopolo KM, Hansen NI, et al. Early-onset neonatal sepsis 2015 to 2017, the rise of Escherichia coli, and the need for novel prevention strategies. JAMA Pediatr 2020;174:e200593. Crossref

42. Hon KL, Liu S, Chow JC, et al. Mortality and morbidity of extremely low birth weight infants in Hong Kong, 2010-2017: a single-centre review. Hong Kong Med J 2018;24:460-5. Crossref

Selection of the mode of delivery in a twin pregnancy is always challenging for obstetricians, although vaginal delivery is theoretically feasible for diamniotic twins if the first twin is in cephalic presentation.1 In the past 15 years, two cohort studies2 3 and a multicentre randomised trial4 concluded that when the first twin was in cephalic presentation, planned caesarean delivery did not significantly decrease or increase the risk of fetal/neonatal death or serious neonatal morbidity, compared with planned vaginal delivery. These findings suggest that vaginal delivery of twins is a safe and reasonable mode of delivery. However, attempts to deliver vaginally are not always successful, and the intrapartum risks of adverse outcomes for second twins should be carefully considered.

In a study of factors that were predictive of successful vaginal delivery, Easter et al5 found that the vaginal delivery rates of second twins in non-vertex presentation were comparable with the vaginal delivery rates of second twins in vertex presentation. Successful vaginal delivery was associated with higher parity. In the subgroup of second twins in non-vertex presentation, successful vaginal delivery was associated with the presence of more experienced birth attendants. The rates of neonatal morbidity and mortality were low in both groups, and they did not differ between groups. However, that study only included twins with gestational ages of ≥32 weeks.

In a study that examined caesarean delivery of the second twin after successful vaginal delivery of the first twin, Breathnach et al6 found that the most common indication for caesarean delivery of the second twin was malpresentation (transverse/shoulder/brow) or compound presentation. Second twins who were delivered by emergency caesarean section after vaginal delivery of the first twin had a perinatal morbidity rate of 29%, but there were only 14 such twins; thus, the sample size was insufficient for robust statistical analysis.

There is a need for additional information concerning factors predictive of successful vaginal delivery of the second twin, which will allow better case selection and avoid combined vaginal-caesarean delivery (ie, failed vaginal delivery of the second twin). To our knowledge, there have been few studies of these factors in Asian populations. Here, we examined the medical records of second twins born to mothers who had attempted vaginal delivery of twins in Hong Kong; we sought to identify factors that could affect the perinatal outcomes and predict failure of vaginal delivery in a predominantly Asian population. We also included deliveries of preterm gestations (23-32 weeks), which were not extensively investigated in previous studies.

This retrospective study focused on twin pregnancies that were delivered between 1 January 2006 and 31 December 2015 in Princess Margaret Hospital, a regional public hospital in Kowloon, Hong Kong. Inclusion criteria were vaginal delivery of the first twin at gestational viability or beyond. Exclusion criteria were miscarriage (delivery before gestational viability) or delivery of the first twin by elective or emergency caesarean section. Under Hong Kong law, 24 full weeks of gestation is generally regarded as the threshold of gestational viability. In exceptional cases, the threshold may be reduced to 23 weeks if, after full discussion with the obstetric and neonatal care teams, the parents demonstrate a strong preference for earlier initiation of active neonatal management.

Eligible cases were identified from the Obstetric Clinical Information System (OBSCIS); for each case, the mother’s demographic and clinical data were retrieved. The OBSCIS is a territory-wide electronic database that contains the prenatal, intrapartum, and postpartum information of all mothers who receive care in public hospitals in Hong Kong. Clinical information in the system is updated in a timely manner by each patient’s midwives and physicians before the patient is discharged from the hospital. Data entry integrity is continuously monitored by a dedicated information technology team within the Hospital Authority, and each obstetrics unit is asked to provide missing data promptly. Each infant’s clinical information was retrieved from the Electronic Patient Record, a comprehensive system that contains all health information (except obstetric records) of patients from birth to death and is shared by all public hospitals and out-patient clinics under the Hong Kong Hospital Authority.

The following maternal data were retrieved: age, parity, gestational age at delivery, chorionicity, and mode of delivery of the second twin. The following infant data were retrieved: birth weight, Apgar score, cord blood pH, delivery time, inter-twin delivery interval, presentation at delivery, and neonatal intensive care unit (NICU) admission status.

The primary outcome of the study was a composite adverse perinatal outcome that included any of the following: Apgar score <6 at 5 minutes after birth, cord blood pH <7, NICU admission, birth trauma, and presence of neonatal complications. For infants with a hospital stay of >28 days, complications until hospital discharge were included. The following complications were considered: respiratory morbidity, intracranial haemorrhage, hypoxic ischaemic encephalopathy, sepsis, metabolic disturbance, birth defects, and neonatal death. The secondary outcome was mode of delivery. Gestational age was established by the patient’s last menstrual period and verified by ultrasound in the first or early second trimester. Chorionicity was established by prenatal ultrasound and confirmed by placental histology after delivery. The likelihood of vaginal delivery may be adversely impacted by considerably larger second twin size, compared with the first twin. Breathnach et al6 found that the rate of caesarean section was higher if the first twin had ≥20% lower weight than the second twin. Therefore, clinically significant weight discordance was regarded as ≥20% in the present study, where weight discordance was defined as the weight difference between the second and first twin divided by the weight of first twin.

Vaginal deliveries of twins were managed in accordance with our labour ward protocol, which does not regard estimated fetal weight discordance as a contra-indication to vaginal delivery. All deliveries were attended by two physicians (as described below) and assisted by ≥2 midwives. Specialist supervision was recommended. In this context, a specialist is an obstetrician who has completed ≥6 years of postgraduate residency training and received accreditation as a Fellow of the Hong Kong College of Obstetricians and Gynaecologists (FHKCOG). Membership in the Royal College of Obstetricians and Gynaecologists (MRCOG) is a prerequisite for FHKCOG accreditation. When a specialist was unavailable (particularly at night), deliveries were conducted or supervised by an MRCOG-qualified physician. Paediatricians were present for all deliveries of second twins. Prenatal steroids (either betamethasone or dexamethasone depending on pharmacy availability and initial treatment at the referral unit) were administered in cases of delivery before 34 weeks of gestation. If necessary, oral nifedipine was used as a first-line tocolytic. Intravenous salbutamol was used as a second-line tocolytic until 2012; since 2013, atosiban has been used as a second-line tocolytic.

Statistical analysis was carried out using SPSS software (Windows version 17.0; SPSS Inc., Chicago [IL], United States). Categorical data were analysed by the Chi squared test or Fisher’s exact test, as appropriate. Among the factors that showed statistical significance in univariate analysis, binary logistic regression was used to identify factors that were independently predictive of vaginal delivery and adverse perinatal outcomes. P values <0.05 were considered statistically significant.

During the 10-year study period, 47 595 deliveries were performed in Princess Margaret Hospital; 718 twin pairs were delivered. Among these twin pairs, 182 and 305 were delivered by elective and emergency caesarean section, respectively; they were excluded from the study. In the remaining 231 cases, the mothers delivered the first twin vaginally and intended to deliver the second twin vaginally. The second twins in this group of patients were included for analysis.

Table 1 shows the demographic and obstetric characteristics of the 231 cases, stratified according to the mode of delivery of the second twin. Emergency caesarean delivery was required in 10 cases (4.3%). Among the three second twins in vertex presentation, two were delivered by caesarean section because of second twin retention; the remaining twin was delivered by caesarean section because of fetal distress. Among the seven second twins in non-vertex presentation, the indications for caesarean delivery were second twin retention (two cases), fetal distress (four cases), and transverse lie (one case). Of the factors shown in Table 1, only an inter-twin delivery interval of >30 minutes and non-vertex presentation of the second twin at delivery were associated with the mode of delivery of the second twin. Logistic regression analysis showed that an inter-twin delivery interval of >30 minutes (odds ratio [OR]=26.952, 95% confidence interval [CI]=5.924-122.619) and non-vertex presentation of the second twin at delivery (OR=5.003, 95% CI=1.101-22.743) were independently associated with caesarean delivery of the second twin.

In subgroup analyses, we examined the relationships of the demographic and obstetric factors in Table 1 to determine their relationships with the mode of delivery for second twins in breech presentation. Univariate analysis revealed that only an inter-twin delivery interval >30 minutes and the presence of less experienced birth attendants were significantly associated with the mode of delivery. Logistic regression showed that an inter-twin delivery interval >30 minutes (OR=36.492, 95% CI=3.035-438.712) and the presence of less experienced birth attendants (OR=10.252, 95% CI=1.001-104.956) were independently associated with caesarean delivery of second twins in breech presentation.

Perinatal outcomes of second twins are shown in Table 2. Univariate analysis revealed that the composite adverse perinatal outcome was only associated with gestational age <32 weeks (P<0.001; OR=12.1, 95% CI=2.738-53.481) [Table 3]. Similarly, gestational age <32 weeks was the only factor significantly associated with NICU admission (P<0.001; OR=6.420, 95% CI=2.073-19.878) [Table 4].

Among the 47 cases with delivery before 34 weeks of gestation, 27 completed steroid treatment before delivery. In 17 cases, delivery occurred before the completion of steroid treatment because of rapid labour that did not respond to tocolytics. Steroid treatment was not administered in three cases; two of these cases involved delivery before 24 weeks of gestation, which is the threshold for beginning steroid treatment in our hospital. In the third case, the mother was admitted in advanced labour. Completion or non-completion of steroid treatment was not associated with the composite adverse perinatal outcome (25/27 vs 18/20, P=0.753).

To our knowledge, this is the first study in Hong Kong concerning the short-term composite adverse perinatal outcomes of second twins in cases where vaginal delivery was attempted. Our approach enabled simultaneous consideration of multiple outcome parameters. The inclusion of additional clinical information until hospital discharge for infants with prolonged hospital stay (>28 days) allowed a more comprehensive assessment of outcomes. Notably, cases of gestation <32 weeks were included; there are minimal published data for this group of infants because they have been excluded from many large trials. Additionally, we examined the effects of birth attendant experience and birth timing.

Prior to this study, there were two analyses of twin deliveries in a predominantly Asia population, both from Hong Kong. The first analysis mainly focused on patient preference regarding the mode of delivery; it also included few vaginal deliveries (35 cases).7 The second analysis, reported by Tang et al,8 was performed in the same obstetric unit as the first analysis; it reviewed neonatal and maternal outcomes after an increase in the rate of vaginal delivery of twins. The authors did not find any significant differences in neonatal morbidities between the vaginal delivery group and the elective caesarean delivery group. However, there were fewer successful vaginal deliveries of ≥1 twin (72 cases) and the effect of inter-twin delivery interval was not evaluated.

In this study, the main factor that affected the composite adverse perinatal outcome was gestational age; complications were mainly related to prematurity. Similarly, NICU admission was mainly related to complications of prematurity, rather than complications of vaginal delivery. There were no statistically significant differences in adverse perinatal outcomes, even for twins who were not delivered in cephalic presentation. Thus, non-cephalic presentation alone should not be considered sufficient to recommend caesarean delivery for twin pregnancies.

A study in Hong Kong by Leung et al,9 published in 2002, showed that all umbilical cord blood gas parameters in the second twin were significantly associated with the inter-twin delivery interval. The risk of severe fetal acidosis was 27% if the second twin was not delivered ≤30 minutes after delivery of the first twin, but the outcomes of second twins were not analysed. In our study, an inter-twin delivery interval of >30 minutes alone did not increase the risk of short-term adverse perinatal outcomes, although it increased the risk of caesarean delivery of the second twin. Schneuber et al10 also reported similar findings in their series, which suggested that an increased inter-twin delivery interval was not associated with adverse fetal outcomes. If monitoring of the second twin is possible and the findings are reassuring, obstetricians may thus consider a conservative approach, even 30 minutes after delivery of the first twin; however, emergency caesarean delivery should be readily available if necessary.

Our study also showed no increase in adverse perinatal outcomes for infants who were delivered after midnight. In general, delivery of twins in daytime or early evening is preferable because additional staff are present, and those staff are often more experienced. Therefore, when there are no indications for urgent delivery, the usual practice in our unit is to begin labour induction for twin pregnancies in the early morning. Deliveries after midnight usually follow spontaneous labour and are thus unplanned. However, such deliveries are supervised by the most senior on-call obstetrician (MRCOG-qualified or FHKCOG-accredited) during the intrapartum period.

The vaginal delivery of second twins in non-cephalic presentation is challenging. Our findings showed a higher rate of caesarean delivery for second twins in non-cephalic presentation (9.0% vs 2.0%, P=0.03). In a large cohort study using the World Health Organization Global Survey dataset, Vogel et al11 showed that caesarean rates were 6.2% and 0.9% for second twins in non-cephalic and cephalic presentation, respectively. Another study by Kong et al12 revealed the caesarean delivery rates of second twins were 4.7% in cephalic presentation, 11.1% in breech presentation, and up to 90% in transverse lie. In both of these studies, analyses were conducted based on the presentation of the second twin at the onset of labour; their findings were consistent with our results. The presence of more experienced obstetricians who are able to perform artful manoeuvres (ie, internal podalic version and external cephalic version) can increase the likelihood of successful vaginal delivery of the second twin. Regular training and rehearsal of the vaginal delivery of twins is important for obstetricians to maintain their skills.

In our study, caesarean delivery of the second twin was necessary in 4.3% of cases, which is similar to or lower than the proportions in other series.6 8 13 14 Regardless of whether the second twin was delivered by caesarean section, there were no significant increases in short-term adverse perinatal outcomes; however, this mode of delivery is less favourable for mothers. These results are contrary to the findings by Breathnach et al,6 in which the perinatal morbidity rate was 29% among second twins delivered by emergency caesarean section after vaginal delivery of the first twin. A systematic review by Rossi et al15 also showed a higher rate of morbidity in second twins after caesarean delivery (19.8% vs 9.5% after vaginal delivery). Thus, combined vaginal-caesarean delivery of twins should be avoided whenever possible.

In the present study, the presence of a larger second twin (≥20% weight discordance) did not significantly increase the risk of caesarean delivery. The second twin was larger in only 12 cases (5.2%). We suspect that many other cases with a larger second twin were scheduled for caesarean delivery without a trial of vaginal delivery. Decisions concerning the mode of delivery are affected by the estimated fetal weight, fetal presentation, and whether the mother has a history of successful vaginal delivery. Various factors must be carefully considered in each case.

There were some limitations in this study. First, the retrospective design may have resulted in missing data or incomplete data collection. This is not a large problem because clinical information in the OBSCIS and the Electronic Patient Record is required to be updated when each patient is discharged from the hospital; therefore, these systems are reliable sources of patient data. Nevertheless, some information was not retrievable, such as the presentation of the second twin at the time of first twin delivery and whether birth attendant manoeuvres were necessary to deliver the second twin. Second, the non-randomised analysis might have led to selection bias concerning the mode of delivery, such that low-risk cases were over-represented in the study. The number of second twins delivered by caesarean section was small; a larger trial is needed to more comprehensively evaluate such cases.

Among second twins born to mothers who had attempted vaginal delivery, we found that adverse perinatal outcomes were mainly related to prematurity, rather than actual mode of delivery. For all second twins, an inter-twin delivery interval <30 minutes was associated with a higher rate of vaginal delivery; for second twins in breech presentation, the presence of more experienced birth attendants was also associated with a higher rate of vaginal delivery. Overall, the risk of caesarean delivery of the second twin was low. Our findings in a predominantly Asian population in Hong Kong support vaginal delivery of the second twin when the first twin is delivered in cephalic presentation.

This study was planned and designed by both authors. Both authors also jointly performed the data analysis. TK Lo provided leadership and supervision, while SL Mok wrote and managed the manuscript. 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 research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This study was approved by the Kowloon West Cluster Research Ethics Committee (Ref KW/EX-17-154 (118-02)). The requirement for patient informed consent was waived because this was a retrospective review of medical records that did not involve patient participation.

1. Monson M, Silver RM. Multifetal gestation: mode of delivery. Clin Obstet Gynecol 2015;58:690-702. Crossref

2. Peaceman AM, Kuo L, Feinglass J. Infant morbidity and mortality associated with vaginal delivery in twin gestations. Am J Obstet Gynecol 2009;200:462.e1-6. Crossref

3. Fox NS, Silverstein M, Bender S, Klauser CK, Saltzman DH, Rebarber A. Active second-stage management in twin pregnancies undergoing planned vaginal delivery in a U.S. population. Obstet Gynecol 2010;115:229-33. Crossref

4. Barrett JF, Hannah ME, Hutton EK, et al. A randomized trial of planned cesarean or vaginal delivery for twin pregnancy. N Engl J Med 2013;369:1295-305.Crossref

5. Easter SR, Lieberman E, Carusi D. Fetal presentation and successful twin vaginal delivery. Am J Obstet Gynecol 2016;214:116.e1-10.Crossref

6. Breathnach FM, McAuliffe FM, Geary M, et al. Prediction of safe and successful vaginal twin birth. Am J Obsetet Gynecol 2011;205:237.e1-7. Crossref

7. Liu AL, Yung WK, Yeung HN, et al. Factors influencing the mode of delivery and associated pregnancy outcomes for twins: a retrospective cohort study in a public hospital. Hong Kong Med J 2012:18:99-107.

8. Tang HT, Liu AL, Chan SY, et al. Twin pregnancy outcomes after increasing the rate of vagina twin delivery: retrospective cohort study in a Hong Kong regional obstetrics unit. J Maternal Fetal Neonatal Med 2016;29:1094-100. Crossref

9. Leung TY, Tam WH, Leung TN, Lok IH, Lau TK. Effect of twin-to-twin delivery interval on umbilical cord blood gas in the second twins. BJOG 2002;109:63-7. Crossref

10. Schneuber S, Magnet E, Haas J, et al. Twin-to-twin delivery time: neonatal outcome of second twin. Twin Res Hum Genet 2011;14:573-9. Crossref

11. Vogel JP, Holloway E, Cuesta C, Carroli G, Souza JP, Barrett J. Outcomes of non-vertex second twins, following vertex vaginal delivery of first twin: a secondary analysis of the WHO Global Survey on maternal and perinatal health. BMC Pregnancy Childbirth 2014;14:55. Crossref

12. Kong CW, To WW. The predicting factors and outcomes of caesarean section of the second twin. J Obstet Gynaecol 2017;37:709-13. Crossref

13. Yang Q, Wen SW, Chen Y, Krewski D, Fung KF, Walker M. Neonatal death and morbidity in vertex-nonvertex second twins according to mode of deliverya and birth weight. Am J Obstet Gynecol 2005;192:840-7. Crossref

14. Persad VL, Baskett TF, O’Connell CM, Scott HM. Combined vaginal-cesarean delivery of twin pregnancies. Obstet Gynecol 2001;98:1032-7. Crossref

15. Rossi AC, Mullin PM, Chmait RH. Neonatal outcomes of twins according to birth order, presentation and mode of delivery: a systematic review and meta-analysis. BJOG 2011;118:523-32. Crossref

Heart failure is a major public health problem that has been shown to increase with age, such that incidence rates rapidly increase after 80 years of age.1 Among older patients with heart failure, 18% to 54% show signs of frailty, a state of reduced physical and cognitive function that results in weakness.2 There are some overlapping symptoms between heart failure and frailty; they interact to accelerate the vicious cycle of frailty.3 One study showed that frail patients with cardiovascular disease had a 7-year survival rate of 12%, which was much lower than the survival rate in non-frail patients with cardiovascular disease (43%).4 Moreover, frailty among patients with heart failure has been associated with poor prognosis5 and reduced cardiac output capacity.6 Pre-frailty is the first step towards frailty; approximately 34.6% to 46.1% of individuals with pre-frailty progress to frailty in Japan.7 Therefore, improvements in frailty and pre-frailty are important considerations in the care of older patients with heart failure.

Age,8 nutrition,9 walking speed,9 heart function,8 and grip strength10 have been shown to influence frailty improvement among older patients with heart failure. However, the predictive accuracies of such factors remain unknown. Moreover, a combination of the predictors has been suggested to increase their predictive accuracy,11 although this hypothesis has not yet been tested. Additionally, nutrition,12 physical function,12 and quality of life12 have been reported to influence pre-frailty improvement among community-dwelling older individuals. To our knowledge, there are no reports of factors that influence pre-frailty improvement among patients with heart failure.

In this study, we used the classification and regression tree (CART) method,13 which facilitates the establishment of clinical prediction models that can identify the best combinations of medical signs, symptoms, and other findings to predict prognoses or treatment outcomes.13 Several clinical prediction models have been developed using the CART method to predict mortality in patients with heart failure.14 15 However, no clinical prediction models have been established to predict the prognosis of patients with heart failure complicated by pre-frailty and frailty.

Here, we aimed to use the CART method to develop models that could predict the prognosis of older patients with heart failure complicated by pre-frailty and frailty, then confirm the predictive accuracies of those models.

This pilot prospective cohort study followed the STROBE reporting guidelines. All included patients provided written informed consent to participate in the study. Identifying information was not collected to protect each patient’s privacy. This study was performed in accordance with the Declaration of Helsinki.

Recruitment, follow-up, and data collection were performed at two acute hospitals (Saiseikai Kure Hospital, Kure, Japan; Kure Kyosai Hospital, Kure, Japan) between July 2018 and December 2019. Potential participants were recruited by therapists at the rehabilitation department.

This study included patients who met the following inclusion criteria: age ≥65 years; hospitalisation for the treatment of heart failure; and presence of pre-frailty or frailty. The exclusion criteria were complications during hospitalisation and/or severe dementia (defined as a revised Hasegawa’s Dementia Scale score ≤9).

Cardiac rehabilitation was performed by physical or occupational therapists to improve physical condition, restore walking ability during hospitalisation, and expand the activities of daily living. Rehabilitation programmes were established by physical or occupational therapists in accordance with physicians’ orders. Initially, aerobic exercises and resistance training programmes were provided according to each patient’s physical condition. Exercise intensity was determined using multiple indices, including target heart rate (convenient method: resting heart rate + 20 bpm), talk test, and Borg scale (11-13) for the chest and lower limbs. The type of exercise was modified (ie, duration extended and load increased) with consideration of each patient’s symptoms and haemodynamics. If necessary, functional exercises (eg, neuromuscular facilitation, joint range of motion, and muscle strengthening exercises), exercises for activities of daily living, and psychological support were implemented for patients and their families. Exercises for activities of daily living were customised according to the functions that each patient needed for discharge. Overall, the duration and frequency of intervention were 30 minutes to 1 hour per day and 5 days per week, respectively.

Variables included demographic and clinical characteristics, frailty assessment results, and physical function. Demographic characteristics included age, sex, body mass index, living arrangement (with family members or alone), New York Heart Association class, medical history (eg, heart failure, coronary artery disease, valvular disease, hypertension, diabetes mellitus, dyslipidaemia, atrial fibrillation, chronic renal dysfunction, and/or stroke), cognitive function assessed using the revised Hasegawa’s Dementia Scale, Life-Space Assessment score, interval from admission to initiation of cardiac rehabilitation, interval from admission to rehabilitation room entry, and length of hospital stay. Hasegawa’s Dementia Scale scores of 21-30, 15-20, 10-14, and ≤9 were regarded as normal, suspected dementia, mild to moderate dementia, and severe dementia, respectively.16 The Life-Space Assessment developed by Baker et al17 was used to evaluate life-space mobility. Up to 120 points were assigned based on the degree of independence in each life-space level during the month prior to the assessment; higher scores were considered indicative of broader life-space and/or greater independence.

Clinical characteristics included blood data, cardiac function, and pharmacotherapy. Blood data included the Geriatric Nutritional Risk Index, brain natriuretic peptide level, estimated glomerular filtration rate (eGFR), and haemoglobin (Hb) level. Cardiac function was evaluated using left ventricular ejection fraction (LVEF), as determined by echocardiography. Pharmacotherapy data included whether patients were receiving dopamine, dobutamine, noradrenaline, phosphodiesterase III inhibitor, or diuretics.

Frailty was assessed using the following five conditions based on the Cardiovascular Health Study (CHS) Index: slow gait speed, weakness, exhaustion, low activity, and weight loss.18 Slow gait speed was defined as <1.0 m/s. Weakness was assessed using maximum grip strength according to sex-specific cut-offs (<26 and <18 kg for men and women, respectively). Exhaustion was assessed using the question “During the past 2 weeks, have you felt tired without a specific reason?” A positive response to this question (ie, “yes”) was considered indicative of exhaustion. Physical activity was evaluated using the question “Do you engage in low levels of physical activity to improve your health?” A negative response to this question (ie, “no”) was considered indicative of a low activity level. Weight loss was assessed using the question “Have you lost 2 kg or more in the past 6 months?” A positive answer to this question was considered indicative of weight loss. There are various criteria for assessing frailty; Fried’s frailty phenotype model18 and the accumulated deficit model established by Mitnitski et al19 are well known. The CHS criteria and the Frailty Index were developed based on the above frailty phenotype model and accumulated deficit model, respectively. Furthermore, a Japanese version of the CHS criteria (J-CHS) has been established,20 and its validity has been confirmed.21 Thus, we selected the J-CHS to assess frailty. Patients with none of the above conditions were considered non-frail (robust), patients with one to two conditions were considered pre-frail, and patients with three conditions or more were considered frail.22

Physical function was assessed using the Short Physical Performance Battery score, 10-metre walk time, single-leg standing time, and hand grip strength. Patients performed the Short Physical Performance Battery test in the following sequence, in accordance with the National Institute on Aging protocol: standing balance tests, gait test (4 m), and chair stand test (five repetitions). The sum of the three test components comprised the final Short Physical Performance Battery score, which ranged from 0 to 12; a score of 12 indicated optimal lower extremity function.23 24 Moreover, 10-m walk time was measured at a comfortable walking speed to assess walking ability. The 10-m walk test has demonstrated high validity and reliability in multiple populations, including healthy older individuals and patients with stroke, neurological disorders, or orthopaedic dysfunction.25 26 Measurements were performed twice at an interval of 30 s; the smaller value was used to indicate walking ability. Standing balance was evaluated by measuring single-leg standing time, which reflects the ability to maintain the body’s centre of gravity within its base of support. A stopwatch was used to measure the duration that a patient could stand on one leg with their eyes open and hands on their waist, without any assistance or falling. A second trial was performed if the result of the first trial was <60 s.27 Hand grip strength (kg) was measured using a digital hand grip strength dynamometer (TKK-5101; Takei Scientific Instruments, Tokyo, Japan or 12B3X00030; Tsutsumi Works, Chiba, Japan). Accordingly, patients were asked to squeeze the dynamometer with maximum effort during two trials for each hand. The maximum value (rounded to the nearest 0.1 kg) for either the left or right hand was used for subsequent analyses.28

Demographic and clinical characteristics were assessed upon admission or each time, whereas cognitive function and Life-Space Assessment were assessed during the first physical therapy session. Frailty and physical function were assessed upon initial entry into the rehabilitation room and at discharge. The amount of change in physical function (discharge−initial) was calculated.

To reduce selection bias, outcomes were selected based on the methods of previous studies.16 17 18 19 20 21 22 23 24 25 26 27 28 To reduce measurement bias, the first author (T Umehara) was not involved in participant enrolment or data collection. Patients received explanations about the purpose of the study but did not receive information concerning the hypothesis tested.

Classification and regression tree analysis29 was used to predict the primary outcomes. Patients with pre-frailty at admission were categorised into an improved group and a non-improved group: patients who were non-frail at discharge were assigned to the improved group, whereas patients who were pre-frail or frail were assigned to the non-improved group. Similarly, patients with frailty at admission were categorised into an improved group and a non-improved group: patients who were non-frail or pre-frail at discharge were assigned to the improved group, whereas patients who were frail were assigned to the non-improved group. Binary trees were used to recursively split predictor variables based on answers to yes/no questions for each variable. All statistical distributions were considered without limitation to linear relationships between outcome variables and predictor variables. These algorithms have been used to develop prediction models in various fields.30 31 32 The CART method with the Gini index was used for the following models: A, which predicted the presence or absence of pre-frailty improvement during hospitalisation; and B, which predicted the presence or absence of frailty improvement during hospitalisation. Pre-frailty or frailty improvement was the dependent variable, whereas demographic and clinical characteristics, pharmacotherapy, and amount of change in physical function were the independent variables. The accuracies of the CART models were evaluated using the area under the receiver operating characteristic curve (AUROC). The method proposed by Delong et al33 was used to identify optimal cut-off points. Subsequently, the sensitivity, specificity, positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were calculated. To assess model validity, cross-validation was performed as follows: the sample was divided into 10 subgroups and the model developed from nine subgroups was used to test the 10th subgroup; this was repeated for all 10 combinations and the rates of misclassification were averaged. Model validity was considered high when the misclassification rates were similar before and after cross-validation. The accuracies of the CART models were evaluated using the AUROC developed from each method. The maximum Youden index (sensitivity + specificity − 1) was defined as the optimal cut-off point. All statistical analyses were performed using SPSS (Windows version 23.0; IBM Corp, Armonk [NY], United States) and the significance level was set at 5%.

Sample size calculations were conducted using MedCalc statistical software, version 19.2 (MedCalc Software bvba, Ostend, Belgium). Before plotting an AUROC, the following values were established: statistical significance (P<0.05), alpha (0.05), statistical power (0.80), and the AUROC value to be included in the null hypothesis (0.5). The AUROC could distinguish between non-predictive (AUROC<0.5), less predictive (0.5<AUROC^lt;0.7), moderately predictive (0.7<AUROC<0.9), highly predictive (0.9<AUROC<1), and perfectly predictive (AUROC=1).34 In this study, an AUROC value of 0.7 was considered indicative of superior statistical discrimination. The frailty improvement ratio (ie, positive/negative ratio) considerably varied among previous studies.35 36 37 Therefore, the positive/negative ratio used here was set at 1:1-5. Moreover, a large sample size was needed to allow for the possibility of stratified analysis. Thus, 62 to 120 patients with frailty were required: 20-31 and 31-100 in the improved and non-improved groups, respectively.

Figure 1 shows the flowchart of patient recruitment. Among the 30 patients with pre-frailty, two with severe dementia were excluded; 28 patients were included in the analysis. Among the 161 patients with frailty, five with severe dementia were excluded; 156 patients were included in the analysis. The patient characteristics are summarised in Tables 1 and 2 (patients with pre-frailty and patients with frailty, respectively).

Figure 2 shows CART model A, which predicted the presence or absence of pre-frailty improvement during hospitalisation. Among the 28 patients with pre-frailty, seven experienced improvements. Single-leg standing time at admission was identified as the best single discriminator for pre-frailty improvement (≤4.4 or >4.4 s). Among patients with single-leg standing time of >4.4 s at admission, the next predictor was LVEF (≤36.8% or >36.8%). Our CART analysis resulted in the establishment of three terminal nodes. The terminal node with the highest probability of a favourable outcome (pre-frailty improvement) was defined as rank 1, whereas the terminal node with the lowest probability of a favourable outcome was defined as rank 3. Based on the AUROC (95% confidence interval [CI]), this CART model had an accuracy of 0.96 (95% CI=0.89-1.00), with an optimal cut-off point of rank 1 (sensitivity: 85.7%, specificity: 95.2%, PLR: 18.0, and NLR: 0.15). The misclassification rates before and after cross-validation were 7.1% and 28.6%, respectively.

Figure 3 shows CART model B, which predicted the presence or absence of frailty improvement during hospitalisation. Among the 156 patients with frailty, 57 experienced improvements. Hand grip strength at admission was identified as the best single discriminator for frailty improvement (≤16.8 or >16.8 kg). Among patients with hand grip strength of >16.8 kg at admission, the next predictor was eGFR (≤27.0 or >27.0 mL/min/1.73 m²). Among patients with hand grip strength of ≤16.8 kg at admission, the next predictor was eGFR (≤83.5 or >83.5 mL/min/1.73 m²). Among patients with eGFR ≤83.5 mL/min/1.73 m², the next predictor was Hb level (≤14.3 or >14.3 g/dL). Among patients with Hb level ≤14.3 g/dL, the next predictor was change in single-leg standing time (≤4.9 or >4.9 s). Our CART analysis resulted in the establishment of six terminal nodes. The terminal node with the highest probability of a favourable outcome (frailty improvement) was defined as rank 1, whereas the terminal node with the lowest probability of a favourable outcome was defined as rank 6. Based on the AUROC (95% CI), this CART model had an accuracy of 0.84 (95% CI=0.78-0.91), with an optimal cut-off point of rank 4 (sensitivity: 70.1%, specificity: 86.9%, PLR: 5.3, and NLR: 0.3). The misclassification rates before and after cross-validation were 19.2% and 35.3%, respectively.

Model A predicted pre-frailty improvement with high accuracy; it identified single-leg standing time at admission as the best predictor, followed by LVEF. Notably, changes in physical function during hospitalisation were not identified as predictors. These results suggest that conditions at admission strongly influence pre-frailty improvement during hospitalisation; moreover, improvement can be expected among patients with good physical function (single-leg standing time >4.4 s) and cardiac function (LVEF >36.8%) at admission. To our knowledge, no study has examined the factors that influence pre-frailty improvement among patients with heart failure. However, one study found that physical function, nutrition, and quality of life were factors that influenced pre-frailty improvement among community-dwelling older individuals.12 The above findings suggest that although physical function is a common factor that influences pre-frailty improvement among older patients with heart failure and community-dwelling older individuals, cardiac function specifically influences pre-frailty improvement among patients with heart failure.

Model B predicted frailty improvement with moderate accuracy; it identified hand grip strength at admission as the best predictor, followed by eGFR, Hb level, and change in single-leg standing time during hospitalisation. Thus, frailty improvement can be expected among patients with good hand grip strength and/or renal function at admission. However, the cut-off values for eGFR, an index of renal function, considerably differed between patients with good (>16.8 kg) and poor (≤16.8 kg) hand grip strength (27 and 83.5 mL/min/1.73 m², respectively). A previous study also showed that older patients with heart failure who had higher hand grip strength were more likely to experience frailty improvement.10 Although no association has been identified between renal function and frailty, renal function is known to influence improvements in exercise capacity among patients with heart failure.38 Moreover, our results showed that, despite the presence of poor physical and renal function at admission, patients with a high Hb level (>14.3 g/dL) were likely to improve from frailty. Although there is no published literature concerning an association between Hb level and frailty, a low Hb level has been shown to cause fatigability,39 40 which is one of the criteria for assessing frailty using the CHS. Furthermore, the Hb level reportedly influences improvements in exercise capacity among patients with heart failure.38

Despite poor hand grip strength, weak renal function, and a low Hb level at admission, frailty improvement was observed in most patients whose single-leg standing time during hospitalisation was >4.9 s. In previous studies that sought to predict frailty improvement among older patients with heart failure, investigators mostly focused on conditions at admission without considering changes during hospitalisation.41 The present results indicate that, despite the presence of poor conditions at admission, patients can recover from frailty by improving physical function during hospitalisation; therefore, rehabilitation is essential during hospitalisation. Resistance training is known to improve single-leg standing time among older individuals.42 Thus, we recommend cardiac rehabilitation, including resistance training, to improve frailty among older patients with heart failure.

The clinical implications of our findings are as follows. Thus far, no algorithms have been established concerning pre-frailty and frailty improvements in older patients with heart failure. Using the CART method, we developed models that could predict the prognosis of older patients with heart failure complicated by pre-frailty and frailty. These models will provide useful information for patients and caregivers. Many of the factors extracted in this study were only assessed at the time of initial rehabilitation. Thus, the prognoses of patients with pre-frailty and patients with frailty can be inferred (to some extent) at the time of initial rehabilitation. Additionally, an increase in single-leg standing time during hospitalisation was associated with frailty improvement. For older patients with heart failure who show signs of frailty, interventions to increase single-leg standing time may help to improve frailty. The PLR and NLR were used to assess the diagnostic performances of the CART models; these parameters revealed that both model A (pre-frailty) and model B (frailty) had good performance. However, our data should be interpreted cautiously because of the small number of patients with pre-frailty.

There were several notable weaknesses and limitations in this study. First, the sample size was limited; the numbers of patients with pre-frailty and frailty were 28 and 156, respectively. Therefore, this study should be regarded as a pilot prospective cohort study. Despite the moderate to high accuracies of models A and B, more large-scale studies are needed to enhance the generalisability of our results. Second, three aspects of frailty exist, namely physical, social, and mental frailty; mental frailty was not considered in this study. A previous study reported that patients with multifaceted frailty, including physical, mental, and social frailty, had worse prognoses compared with patients who had physical frailty alone.5 Therefore, additional studies are needed to develop models that predict improvements in multifaceted frailty, including mental frailty. Third, interventions were customised for each patient; they were not uniform. However, physical therapists in both study hospitals received 2 weeks of training in cardiac rehabilitation, and the methods of cardiac rehabilitation were standardised as much as possible. Fourth, we used the J-CHS to measure frailty; the use of this tool to assess patients with heart failure is potentially controversial. The J-CHS was developed for older adults and has been used in multiple studies.7 21 Additionally, the reliability and validity of this tool have been confirmed.21 We thus consider this use of the J-CHS to be appropriate. Fifth, this study was performed in an unblinded manner, and we could not completely rule out the potential for bias. However, to reduce the measurement bias, the first author (T Umehara) was not involved in participant enrolment or data collection. Sixth, a selection bias might have been present because patients were only recruited at two hospitals in Japan. Caution is needed when generalising our results, particularly to patients in other countries.

By using the CART method, we developed moderately to highly accurate prediction models for pre-frailty and frailty improvement among older patients with heart failure. Model A, which predicted pre-frailty improvement, showed that patients with good single-leg standing time and cardiac function at admission are likely to experience pre-frailty improvement. Furthermore, Model B, which predicted frailty improvement, showed that patients with good hand grip strength, excellent renal function, and/or a high Hb level at admission are likely to experience frailty improvement. Notably, despite the presence of poor conditions at admission, frailty improvement may occur in patients who show improvement in single-leg standing time during hospitalisation. Overall, our results suggest that cardiac rehabilitation to prolong single-leg standing time is necessary to improve frailty, particularly when conditions at admission are poor.

Concept or design: All authors.
Acquisition of data: T Umehara and N Katayama.
Analysis or interpretation of data: T Umehara, M Tsunematsu, M Kakehashi.
Drafting of the manuscript: T Umehara, A Kaneguchi and Y Iwamoto.
Critical revision of the manuscript for important intellectual content: T Umehara, A Kaneguchi, Y Iwamoto.

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.

We thank Y Yamamoto, M Ota, Y Nakashima, W Kawakami, and M Iwanaga (Kure Kyosai Hospital, Kure, Japan), as well as S Nagao, Y Kawabata, and D Tomiyama (Saiseikai Kure Hospital, Kure, Japan) for their support in data collection.

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

The study was approved by the Ethics Committees of Saiseikai Kure Hospital (Ref 127) and Kure Kyosai Hospital (Ref 31-17). Patients provided written informed consent to participate.

1. Bleumink GS, Knetsch AM, Sturkenboom MC, et al. Quantifying the heart failure epidemic: prevalence, incidence rate, lifetime risk and prognosis of heart failure The Rotterdam Study. Eur Heart J 2004;25:1614-9. Crossref

2. Smart N, Marwick TH. Exercise training for patients with heart failure: a systematic review of factors that improve mortality and morbidity. Am J Med 2004;116:693-706. Crossref

3. Joyce E. Frailty in advanced heart failure. Heart Fail Clin 2016;12:363-74. Crossref

4. Newman AB, Gottdiener JS, Mcburnie MA, et al. Associations of subclinical cardiovascular disease with frailty. J Gerontol A Biol Sci Med Sci 2001;56:158-66. Crossref

5. Rodríguez-Pascual C, Paredes-Galán E, Ferrero-Martínez AI, et al. The frailty syndrome is associated with adverse health outcomes in very old patients with stable heart failure: A prospective study in six Spanish hospitals. Int J Cardiol 2017;236:296-303. Crossref

6. Denfeld QE, Winters-Stone K, Mudd JO, Hiatt SO, Chien CV, Lee CS. Frequency of and significance of physical frailty in patients with heart failure. Am J Cardiol 2017;8:1243-9. Crossref

7. Kojima G, Taniguchi Y, Iliffe S, Jivraj S, Walters K. Transitions between frailty states among community-dwelling older people: A systematic review and meta-analysis. Ageing Res Rev 2019;50:81-8. Crossref

8. Takabayashi K, Ikuta A, Okazaki Y, et al. Clinical characteristics and social frailty of super-elderly patients with heart failure—the Kitakawachi clinical background and outcome of heart failure registry. Circ J 2016;81:69-76. Crossref

9. Sze S, Zhang J, Pellicori P, Morgan D, Hoye A, Clark AL. Prognostic value of simple frailty and malnutrition screening tools in patients with acute heart failure due to left ventricular systolic dysfunction. Clin Res Cardiol 2017;106:533-41. Crossref

10. Bekfani T, Pellicori P, Morris DA, et al. Sarcopenia in patients with heart failure with preserved ejection fraction: impact on muscle strength, exercise capacity and quality of life. Int J Cardiol 2016;222:41-6. Crossref

11. Umehara T, Tanaka R, Tsunematsu M, et al. Can the amount of interventions during the convalescent phase predict the achievement of independence in activities of daily living in patients with stroke? A retrospective cohort study. J Stroke Cerebrovasc Dis 2018; 27:2436-44. Crossref

12. Gené Huguet L, Navarro González M, Kostov B, et al. Pre Frail 80: multifactorial intervention to prevent progression of pre-frailty to frailty in the elderly. J Nutr Health Aging 2018;22:1266-74. Crossref

13. McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Stiell IG, Richardson WS. Users’ guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group. JAMA 2000;284:79-84. Crossref

14. Fonarow GC, Adams KF Jr, Abraham WT, Yancy CW, Boscardin WJ, ADHERE Scientific Advisory Committee, Study Group, and Investigators. Risk stratification for in-hospital mortality in acutely decompensated heart failure: classification and regression tree analysis. JAMA 2005;293:572-80. Crossref

15. Arenja N, Breidthardt T, Socrates T, et al. Risk stratification for 1-year mortality in acute heart failure: classification and regression tree analysis. Swiss Med Wkly 2011;141:w13259. Crossref

16. Imai Y, Hasegawa K. The revised Hasegawa’s Dementia Scale (HDS-R)—Evaluation of its usefulness as a screening test for dementia. J Hong Kong Coll Psychiatr 1994;4:20-4.

17. Baker PS, Bodner EV, Allman RM. Measuring life-space mobility in community-dwelling older adults. J Am Geriatr Soc 2003;51:1610-4. Crossref

18. Fried LP, Tangen CM, Walston J, et al. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci 2001;56:146-56. Crossref

19. Mitnitski AB, Mogilner AJ, Rockwood K. Accumulation of deficits as a proxy measure of aging. ScientificWorldJournal 2001;1:323-36. Crossref

20. Satake S, Arai H. The revised Japanese version of the Cardiovascular Health Study criteria (revised J-CHS criteria). Geriatr Gerontol Int 2020;20:992-3. Crossref

21. Makizako H, Shimada H, Doi T, Tsutsumimoto K, Suzuki T. Impact of physical frailty on disability in community-dwelling older adults: a prospective cohort study. BMJ Open 2015;5:e008462. Crossref

22. Satake S, Shimada H, Yamada M, et al. Prevalence of frailty among community-dwellers and outpatients in Japan as defined by the Japanese version of the Cardiovascular Health Study criteria. Geriatr Gerontol Int 2017;17:2629-34. Crossref

23. Bernabeu-Mora R, Medina-Mirapeix F, Llamazares-Herrán E, García-Guillamón G, Giménez-Giménez LM, Sánchez-Nieto JM. The Short Physical Performance Battery is a discriminative tool for identifying patients with COPD at risk of disability. Int J Chron Obstruct Pulmon Dis 2015;10:2619-26. Crossref

24. Guralnik JM, Simonsick EM, Ferrucci L, et al. A short physical performance battery assessing lower extremity function: association with self-reported disability and prediction of mortality and nursing home admission. J Gerontol 1994;49:85-94. Crossref

25. Bohannon RW. Comfortable and maximum walking speed of adults aged 20-79 years: reference values and determinants. Age Ageing 1997;26:15-9. Crossref

26. van Hedel HJ, Wirz M, Dietz V. Assessing walking ability in subjects with spinal cord injury: validity and reliability of 3 walking tests. Arch Phys Med Rehabil 2005;86:190-6. Crossref

27. Michikawa T, Nishiwaki Y, Takebayashi T, Toyama Y. One-leg standing test for elderly populations. J Orthop Sci 2009;14:675-85. Crossref

28. Huang C, Niu K, Kobayashi Y et al. An inverted J-shaped association of serum uric acid with muscle strength among Japanese adult men: a cross-sectional study. BMC Musculoskelet Disord 2013;14:258. Crossref

29. Breiman L, Friedman J, Olshen R, Stone C. Classification and Regression Tree Analysis. Boston University; 2014: 1-16.

30. Aguiar FS, Almeida LL, Ruffino-Netto A, Kritski AL, Mello FC, Werneck GL. Classification and regression tree (CART) model to predict pulmonary tuberculosis in hospitalized patients. BMC Pulm Med 2012;12:40.Crossref

31. Takahashi O, Cook EF, Nakamura T, Saito J, Ikawa F, Fukui T. Risk stratification for in-hospital mortality in spontaneous intracerebral haemorrhage: a Classification and Regression Tree analysis. QJM 2006;99:743-50. Crossref

32. Westreich D, Lessler J, Funk MJ. Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression. J Clin Epidemiol 2010;63:826-33. Crossref

33. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44:837-45. Crossref

34. Swets JA. Measuring the accuracy of diagnostic systems. Science 1988;240:1285-93. Crossref

35. Espinoza SE, Jung I, Hazuda H. Lower frailty incidence in older Mexican Americans than in older European Americans: the San Antonio Longitudinal Study of Aging. J Am Geriatr Soc 2010;58:2142-8. Crossref

36. Schuurmans H, Steverink N, Lindenberg S, Frieswijk N, Slaets JP. Old or frail: what tells us more? J Gerontol A Biol Sci Med Sci 2004;59:962-5. Crossref

37. Collard RM, Boter H, Schoevers RA, Oude Voshaar RC. Prevalence of frailty in community-dwelling older persons: a systematic review. J Am Geriatr Soc 2012;60:1487-92. Crossref

38. Silverberg DS, Wexler D, Blum M, Iaina A. The cardio renal anemia syndrome: correcting anemia in patients with resistant congestive heart failure can improve both cardiac and renal function and reduce hospitalizations. Clin Nephrol 2003;60 Suppl 1:S93-102.

39. Roy CN. Anemia in frailty. Clin Geriatr Med 2011;27:67-78. Crossref

40. Chaves PH, Ashar B, Guralnik JM, Fried LP. Looking at the relationship between hemoglobin concentration and prevalent mobility difficulty in older women. Should the criteria currently used to define anemia in older people be reevaluated? J Am Geriatr Soc 2002;50:1257-64. Crossref

41. Reeves GR, Whellan DJ, O’Connor CM, et al. A novel rehabilitation intervention for older patients with acute decompensated heart failure: The REHAB-HF pilot study. JACC Heart Fail 2017;5:359-66. Crossref

42. Gonzalez AM, Mangine GT, Fragala MS, et al. Resistance training improves single leg stance performance in older adults. Aging Clin Exp Res 2014;26:89-92. Crossref