Fracture incidence and fracture-related mortality decreased with decreases in population mobility during the early days of the COVID-19 pandemic: an epidemiological study

Abstract

Introduction: We investigated the impact of coronavirus disease 2019 (COVID-19) social distancing measures on fracture incidence and fracture-related mortality, as well as associations with population mobility.

Methods: In total, 47 186 fractures were analysed across 43 public hospitals from 22 November 2016 to 26 March 2020. Considering the smartphone penetration of 91.5% in the study population, population mobility was quantified using Apple Inc’s Mobility Trends Report, an index of internet location services usage volume. Fracture incidences were compared between the first 62 days of social distancing measures and corresponding preceding epochs. Primary outcomes were associations between fracture incidence and population mobility, quantified by incidence rate ratios (IRRs). Secondary outcomes included fracture-related mortality rate (death within 30 days of fracture) and associations between emergency orthopaedic healthcare demand and population mobility.

Results: Overall, 1748 fewer fractures than projected were observed during the first 62 days of COVID-19 social distancing (fracture incidence: 321.9 vs 459.1 per 100 000 person-years, P<0.001); the relative risk was 0.690, compared with mean incidences during the same period in the previous 3 years. Population mobility exhibited significant associations with fracture incidence (IRR=1.0055, P<0.001), fracture-related emergency department attendances (IRR=1.0076, P<0.001), hospital admissions (IRR=1.0054, P<0.001), and subsequent surgery (IRR=1.0041, P<0.001). Fracture-related mortality decreased from 4.70 (in prior years) to 3.22 deaths per 100 000 person-years during the COVID-19 social distancing period (P<0.001).

Conclusion: Fracture incidence and fracture-related mortality decreased during the early days of the COVID-19 pandemic; they demonstrated significant temporal associations with daily population mobility, presumably as a collateral effect of social distancing measures.

Introduction

The coronavirus disease 2019 (COVID-19) pandemic, which began in early 2020, has resulted in unprecedented large-scale public health responses. Stringent regional social distancing measures (eg, quarantine, school closures, and restrictions at work and recreation destinations) were rapidly implemented during the early days of the pandemic as forms of non-pharmacological intervention.¹ Although there is evidence that such measures can temporarily contain the spread of severe acute respiratory syndrome coronavirus 2,²collateral effects among non[-]COVID-19[-]related conditions have also been reported.³ Trauma is the leading cause of death and disability among young adults worldwide,⁴ but the effects of the COVID-19 pandemic on injuries and fracture incidence within Hong Kong have not been fully elucidated. This uncertainty has hindered healthcare resource deployment and clinical service demand estimation in times of stringency. We sought to address this problem using ‘big data' sources and regional clinical data repositories, which allow researchers to rapidly delineate epidemiological associations with potential applications in forecasting models, while avoiding resource-intensive collection of conventional epidemiological information and protecting patient anonymity.

We presumed that restrictions on citizen mobility, in concert with social distancing, were associated with reductions in musculoskeletal injuries during the early days of the COVID-19 pandemic. Specifically, we hypothesised that reduced population mobility was associated with reductions in fracture incidence and fracture-related healthcare needs during the early days of the pandemic. We investigated these relationships by analysing daily multicentre hospital data registries in Hong Kong, along with digital population mobility datasets published by a technology company. Our main outcome measurement was skeletal fractures, which served as a specific surrogate for musculoskeletal trauma.

Methods Data collection

This study was conducted in Hong Kong, a highincome region (with gross domestic product per capita of HK$357 667 in 2020⁵) that was among the first areas affected by COVID-19; social distancing measures were implemented during the early days of the pandemic.

Using the Clinical Data Analysis and Reporting System of the Hospital Authority, anonymised patient records were retrieved from all 43 public hospitals in Hong Kong for the period from 22 November 2016 to 20 May 2020. In Hong Kong, up to 90% of hospital bed-days occur in public hospitals, which manage nearly all critical emergencies in the region.⁶ Anonymised clinical data were retrieved, including time of initial injury presentation, emergency department triage, trauma category, hospital admission, diagnosis, and surgical procedures. Diagnoses and procedures were encoded in accordance with the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) by treating physicians based on clinical and radiological investigations, intraoperative findings, and date of hospital discharge. The ICD-9-CM codes that met the inclusion criteria (which included fractures under the purview or commonly admitted under the care of orthopaedic and traumatology service) were all codes from 805 to 829 (inclusive). Duplicate records from fracture reassessment related to follow-up attendances, hospitalisation after emergency department attendance, and elective hospital re-admissions (ie, episodes assigned to the same patient unique identifier with identical diagnostic codes, which occurred within 30 days of the index episode) were regarded as a single event to avoid double counting. Pathological fractures and records with missing diagnosis codes or admission times were excluded from the analysis.

Time intervals

The ‘COVID-19 epoch' was defined as 25 January 2020 (activation of the government's ‘emergency' response and commencement of social distancing policies⁷) to 26 March 2020; this arbitrarily chosen 62-day period included all patients with fractures who presented during that period. This epoch was compared with the 9 weeks preceding the onset of the COVID-19 pandemic (ie, 22 November 2019 to 24 January 2020), as well as the same period over the past 3 years to adjust for seasonality-related variations⁸ (ie, 25 January to 26 March in 2017, 2018, and 2019). Differences between actual and projected daily fracture incidences were calculated based on mean values at the same time of year over the past 3 years. Fracture-related mortality rates, defined as the numbers of deaths within 30 days after initial fracture presentation per 100 000 person-years, were compared. The Chi squared test was used to detect differences in fracture incidence and fracture-related mortality during the COVID-19 pandemic and pre-pandemic epochs.

Quantifying population mobility

Surrogate data concerning population mobility were retrieved from Mobility Trends Reports⁹—an aggregate daily measure of geographical direction requests on Apple Maps, a service established by Apple Inc, which holds the largest market share of electronic mobile devices (including smartphones and tablets) in Hong Kong.¹⁰ Walking index was regarded as an index of population mobility, considering the smartphone penetration of 91.5%¹¹among the 7.50 million residents of Hong Kong.¹²

Data analysis

Associations between daily fracture incidence and population mobility were determined by incidence rate ratios (IRRs) using quasi-Poisson regression. Secondary analysis involved associations between mobility index and fracture repair surgeries, all types of orthopaedic emergency department attendances, orthopaedic hospital admissions, and emergency orthopaedic surgeries.

Because medical records were timestamped in Hong Kong time (8 hours ahead of Greenwich Mean Time), they were converted to Pacific Time to match the time intervals listed in Mobility Trends Reports; this conversion ensured that data were temporally matched for analysis.

To determine whether mobility associations simply reflected health-seeking behaviour, we included analyses of diseases which lacked a physiological basis and were not associated with population mobility—these ‘controls' included appendicitis, cellulitis, and abscess (ICD-9-CM diagnosis codes 540 and 682). Statistical analysis was performed using R software, version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria). Quasi-Poisson regression was used to model the relationship between the population mobility index and the daily incidences of fractures and fracture-related events; the population mobility index was the explanatory variable, whereas the

daily incidences of various events were response variables. A quasi-Poisson distribution was preferred over a Poisson distribution, considering the presence of significant overdispersion among some response variables (in the form of count data) when a dispersion parameter was included. In accordance with standard statistical methods, the natural logarithm was utilised as the link function. Incidence rate ratios were reported and represented by the following formula:

Estimated incidence = IRR^PMI [x] BIR

where IRR represents the incidence rate ratio, PMI represents the population mobility index, and BIR represents the baseline incidence rate. The IRR, which quantifies the relationship between the mobility index and fracture incidence, is multiplicative in nature—for every unit increase in the mobility index, there is a corresponding multiplicative increase in the IRR. If the IRR is [lesser than]1, it is expected to decrease in a multiplicative manner for every unit decrease in the mobility index. Multiple comparisons were adjusted by Bonferroni correction, and the threshold for statistical significance was regarded as P[lesser than]0.00227 (0.05/22).

Results

In total, 59 931 fracture-related medical records from orthopaedic emergency department attendances, hospital admissions, and surgeries were reviewed. After exclusion of 11 498 linked episodes, 284 pathological fractures, 786 follow-up attendances, 175 hospital re-admissions, and two episodes with missing admission times, 47 186 fractures were included in the analysis. Descriptive statistics regarding daily fracture incidences, controls, and fracture-related surgeries during COVID-19 social distancing are shown in Table 1. Intra-year and inter-year comparison cohorts are presented in Table 2.

Table 1. Incidences of fractures and surgeries during the early days of the coronavirus disease 2019 social distancing (25 January to 26 March 2020)

Table 2. Incidences of fractures before and during the early days of the coronavirus disease 2019 pandemic

Fracture incidence during COVID-19 social distancing

A reduction of 1748 fractures in the actual versus projected incidence (321.9 vs 459.1 per 100 000 person-years, P[lesser than]0.001) was observed during the COVID-19 epoch; the relative risk was 0.690 (95% confidence interval [CI]=0.678-0.702), compared with mean incidences in the previous 3 years (ie, inter-year cohort) [Table 2]. Differences in fracture incidence between the pandemic and pre-pandemic epochs are shown in Figure 1.

Figure 1. Daily fracture incidences (triangles) before (22 November 2019 to 24 January 2020) and during (25 January to 26 March 2020) the early days of coronavirus disease 2019 social distancing, with comparison to the same period in the previous 3 years (dots in different shades of grey). There were 1748 fewer fractures than projected

Fracture incidences, population mobilities, and controls are depicted in Figure 2. The first two COVID-19 cases in Hong Kong were reported on 23 January 2020¹³; three additional cases were reported on 24 January 2020. Social distancing measures were implemented on 25 January 2020; these included suspension of schools, initiation of ‘work from home' measures among civil servants, and suspension of hospital visitations. Mandatory border quarantine was enforced on 8 February 2020. The sharpest decrease in mobility was observed on 24 January 2020; population mobility subsequently remained at low levels, in conjunction with cancellations of large-scale social and sporting events, as well as the imposition of travel restrictions with quarantine measures for returning travellers.⁷

Figure 2. Fracture incidences, population mobilities (based on mobility index data from Apple Inc's Mobility Trend Reports), and controls over time in the study population. The largest decrease in population mobility coincided with the first confirmed case of coronavirus disease 2019 in Hong Kong

Associations of fracture incidence with population mobility

Fracture incidence was positively associated with the population mobility index (IRR=1.0055, 95% CI=1.0044-1.0066, P[lesser than]0.001). Analyses of fracture incidence according to anatomical location revealed associations of the population mobility index with upper limb fractures (IRR=1.0073, 95% CI=1.0057-1.0088, P[lesser than]0.001) and lower limb fractures (IRR=1.0045, 95% CI=1.0030-1.0060, P[lesser than]0.001) [Fig 3].

Figure 3. Associations of fracture incidence with population mobility. Fracture incidence was associated with mobility index according to quasi-Poisson regression, with incidence rate ratios of 1.0055 (95% confidence interval [CI]=1.0044-1.0066) for all fractures, 1.0073 (95% CI=1.0057-1.0088) for upper limb fractures, and 1.0045 (95% CI=1.0030-1.0060) for lower limb fractures (all P[lesser than]0.001)

The population mobility index was associated with the incidences of fractures involving the radius and ulna (IRR=1.0079, 95% CI=1.0057-1.0101, P[lesser than]0.001), hand and fingers (IRR=1.0069, 95% CI=1.0039-1.0098, P[lesser than]0.001), femoral neck (IRR=1.0065, 95% CI=1.0035-1.0095, P[lesser than]0.001), and tibia and fibula (IRR=1.0097, 95% CI=1.0044-1.0151, P[lesser than]0.001) [Fig 4]. However, after Bonferroni correction, the population mobility index did not exhibit statistically significant associations with trochanteric hip fractures (IRR=1.0008, P=0.683), spine fractures (IRR=0.996, P=0.183), or pelvic fractures (IRR=1.0064, P=0.00799).

Figure 4. Incidence rate ratios indicating relationships between fracture incidence and population mobility index. Incidence rateratios of fractures are grouped according to anatomical locations with 95% confidence intervals indicated on each bar. Barsin dark grey and asterisks in y-axis labels indicate statistically significant associations (P[lesser than]0.00227). Note ‘control groups' ofdiseases in grey, which were included to investigate possibility of confounding between mobility index and disease incidence byalterations in health-seeking behaviour; no statistically significant associations were present in these groups

Stronger associations were observed among fractures, such that some patients presented at a younger age (eg, patients with tibia, fibula, hand, and finger fractures), whereas other patients presented at an older age (eg, patients with femoral neck fractures). Digital literacy, manual dexterity and visual acuity, and higher internet and smartphone usage among younger residents¹¹ are among the factors that cause the population mobility index to have increased sensitivity for analysis in such age-groups.

The incidences of cellulitis, abscesses, and appendicitis were not associated with the population mobility index (P[greater than]0.00227). These findings support the hypothesis that changes in associations between fracture incidence and population mobility were not solely caused by changes in health-seeking behaviour; if they had been caused by changes in such behaviour, corresponding reductions in those conditions would have been observed.

Secondary exploratory analysis of surgeries, emergency department attendances, and hospital admissions

The daily population mobility index was associated with the number of patients admitted on a particular day who subsequently underwent fracture repair surgeries (IRR=1.0041, 95% CI=1.0020-1.0062, P[lesser than]0.001). The population mobility index was also associated with all types of emergency orthopaedic surgeries (IRR=1.0040, 95% CI=1.0021-1.0058, P[lesser than]0.001), attendances at orthopaedic emergency departments (IRR=1.0076, 95% CI 1.0064-1.0087, P[lesser than]0.001), and emergency orthopaedic hospital admissions (IRR=1.0054, 95% CI=1.0043-1.0064, P[lesser than]0.001). Additionally, the numbers of orthopaedic patients triaged as critical, emergent, and urgent (ie, patients who require physician attention within 30 minutes of attendance) were also associated with the population mobility index (IRR=1.0063, 95% CI=1.0054-1.0073, P[lesser than]0.001). Whereas the numbers of traffic-related and sports-related trauma cases were associated with the population mobility index (IRR=1.008, 95% CI=1.0063-1.0097 and IRR=1.013, 95% CI=1.0092-1.0158, respectively, both P[lesser than]0.001), the number of assault-related trauma cases was not (P=0.238).

Fracture-related mortality rate

Forty-nine patients with fractures died within 30 days of presentation during the COVID-19 epoch. This constituted a mortality rate of 3.22 deaths per 100 000 person-years, which was lower than the rate of 4.70 deaths per 100 000 person-years during the period before the pandemic (P[lesser than]0.001); thus, there were around 19 fewer fracture-related deaths in the Hong Kong population during the 62-day study period. Four patients with fractures had COVID-19 (ie, they had positive results in nasopharyngeal swab reverse transcriptase-polymerase chain reaction tests for severe acute respiratory syndrome coronavirus 2) and survived beyond 30 days after initial fracture presentation. The change in mortality was presumably explained by reduced fracture incidence: 30-day mortality among patients with fractures did not significantly differ between the COVID-19 epoch (1.2%, 49 deaths in 4101 patients) and the preceding period (1.0%, 175 deaths in 17 198 patients) [P=0.305].

Discussion

This study analysed 47 186 fractures in Hong Kong, prior to and during the early days of the COVID-19 pandemic. Population mobility was assessed through aggregate digital footprints using the volume of location service requests as a surrogate marker, considering the high smartphone and internet penetration in Hong Kong¹¹; importantly, datasets of aggregate digital footprints have been published to facilitate efforts to control COVID-19.⁹ The findings support our hypothesis in terms of the relationship between fracture incidence and population mobility.

Fractures incur substantial healthcare costs; for example, fragility fracture-related costs incurred costs of 37.5 billion euros, along with the loss of 1.0 million quality-adjusted life years, among the six largest European countries in 2017.¹⁴ Some fractures (eg, hip fractures) warrant early surgical management to mitigate the morbidity and mortality associated with surgical delays.¹⁵ Guidance regarding early surgical management remained in effect, even during the early days of the COVID-19 pandemic.¹⁶ Despite the best available tools, fracture prediction remains difficult; there are additional challenges associated with epidemiological projections of specific time points when such fractures occur. Accordingly, hospitals and public health entities experience difficulties in terms of estimating emergency trauma service load and allocating limited healthcare resources. Our findings suggest that population mobility indices, which are freely and publicly accessible, can provide insights regarding fracture incidence. Population mobility may be useful in quantitative modelling of fracture-related inpatient and surgical theatre service demand, using the IRRs described in this study.

Although there is evidence to support the efficacy of social distancing measures with respect to COVID-19 transmission,² our findings emphasise the collateral impacts of pandemic-related interventions on non-communicable diseases. We found that fracture incidence decreased when population mobility was hindered by social distancing measures; the relative reduction in overall fractures appeared to be similar to the effect of established pharmacological interventions on fragility fractures.¹⁷ Although this relationship appears to contradict the common notion that physical activity confers a protective effect against fractures in both young and old age-groups,^{18 19} associations of increased fracture risk with specific types of exercises (eg, bicycling), or regular participation in other exercise and sports activities, have been described.²⁰ Thus, long-term benefits (eg, increased bone mineral density) may be accrued at the expense of increased exposure to fracture risk when engaging in physical activity. Although the long-term impact of reduced population mobility on fracture incidence remains unclear, vitamin D deficiency caused by prolonged time indoors (ie, without sunlight exposure) is an established risk factor for future fractures.²¹

The strengths of our study include its inclusion of data from all public hospitals in Hong Kong, which allowed extensive analysis of rare events such as fractures. Our database has a high ([greater than]96%) positive predictive value for fractures,²² presumably because data entry is conducted by impartial registered medical practitioners. Furthermore, high internet and smartphone penetration increased the sensitivity of the population mobility analysis, such that the mobility index was geographically specific to the study population. Pedestrian and road traffic densities, which are indirectly represented by the population mobility index, could also precipitate accidents, falls, and subsequent fracture risk. Additionally, potential confounding based on health-seeking behaviour was partially mitigated by the inclusion of ‘control' groups. Fortunately, all hospitals involved in the study maintained full emergency service during the early days of the COVID-19 pandemic²³; this maintenance of emergency service minimised potential confounding by hospitals that were unable to provide service to patients with fractures.

Limitations of the study involved deficiencies in the population mobility index. For example, travel between familiar places and travel where navigation guidance is unnecessary, as well as the usage of alternative electronic service providers, were not considered. Therefore, the population mobility index served as a more specific (rather than sensitive) tool for assessment of population mobility. Global positioning system (GPS)[-]based mobility tracking would theoretically allow more extensive data collection, thus providing greater detection sensitivity; however, such mobility tracking would cause substantial privacy issues, resulting in legal and ethical challenges.

Notably, older adults are less adept in smartphone usage (62.2% of residents aged [greater than or equal to]65 years reported internet usage in 2020¹¹), and the digital population mobility index does not adequately illustrate this division in the population. Furthermore, fractures in older adults are largely caused by osteoporosis, whereas high-energy injury mechanisms are observed in younger individuals.²⁴Therefore, social distancing may have a negligible effect on the incidences of osteoporotic fractures sustained indoors. We caution against using population mobility data as the sole source of estimates for health service planning because that approach could underestimate fragility fracture service demand.

Additionally, the use of fracture incidence data from a public healthcare database only included approximately 90% of the population health demand. During the early days of the COVID-19 pandemic, instances of diversion to the private sector, attendances in private clinics, and visits to alternative practitioners were not coded; the lack of these data may have led to underestimation of total fracture incidence. Finally, we caution against generalising these findings to regions with less internet and smartphone penetration.

Conclusion

During the early days of the COVID-19 pandemic, fracture incidence and fracture-related mortality considerably decreased with the implementation of government social distancing measures that targeted population mobility. This unique opportunity enabled the identification of collateral associations and revealed that population mobility could be used (along with many other factors) to estimate clinical service demand.

Author contributions

Concept or design: JSH Wong, DKH Yee.

Acquisition of data: JSH Wong.

Analysis or interpretation of data: JSH Wong, ALH Lee, DKH Yee, CX Fang.

Drafting of the manuscript: JSH Wong, ALH Lee, CX Fang.

Critical revision of the manuscript for important intellectual content: CX Fang, DKH Yee, FKL Leung, KMC Cheung.

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.

Conflicts of interest

All authors have disclosed no conflicts of interest.

Funding/support

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

Ethics approval

Ethics approval was granted by the Institutional Review Board of The University of Hong Kong/ Hospital Authority Hong Kong West Cluster (HKU/HA HKW IRB Ref No.: UW 20-275), and investigations were carried out in accordance with the Declaration of Helsinki. The requirement for patient informed consent was waived by the Board because the study used anonymised data and the risk of identification was low.

References

1. Leung K, Wu JT, Liu D, Leung GM. First-wave COVID-19 transmissibility and severity in China outside Hubei after control measures, and second-wave scenario planning: a modelling impact assessment. Lancet 2020;395:1382-93.

2. Cousins S. New Zealand eliminates COVID-19. Lancet 2020;395:1474.

3. De Filippo O, D’Ascenzo F, Angelini F, et al. Reduced rate of hospital admissions for ACS during Covid-19 outbreak in Northern Italy. N Engl J Med 2020;383:88-9.

4. Krug EG, Sharma GK, Lozano R. The global burden of injuries. Am J Public Health 2000;90:523-6.

5. Census and Statistics Department, Hong Kong SAR Government. Table 31: Gross Domestic Product (GDP), implicit price deflator of GDP and per capita GDP. 2020. Available from: https://www.censtatd.gov.hk/en/web_table.html?id=31. Accessed 26 Apr 2020.

6. Food and Health Bureau, Hong Kong SAR Government. Report of the Strategic Review on Healthcare Manpower Planning and Professional Development. 2017. Available from: https://www.fhb.gov.hk/en/press_and_publications/otherinfo/180500_sr/srreport.html. Accessed 20 May 2020.

7. Leung GM, Cowling BJ, Wu JT. From a sprint to a marathon in Hong Kong. N Engl J Med 2020;382:e45.

8. Yee DK, Fang C, Lau TW, Pun T, Wong TM, Leung F. Seasonal variation in hip fracture mortality. Geriatr Orthop Surg Rehabil 2017;8:49-53.

9. Apple Inc. COVID-19–mobility trends reports. Available from: https://www.apple.com/covid19/mobility. Accessed 26 Apr 2020.

10. Statcounter GlobalStats. Mobile & tablet vendor market share Hong Kong. Jan – Mar 2020. Available from: https://gs.statcounter.com/vendor-market-share/mobile-tablet/hong-kong/#monthly-202001-202003. Accessed 26 Apr 2020.

11. Census and Statistics Department, Hong Kong SAR Government. Thematic Household Survey Report No. 69: Personal Computer and Internet Penetration. 2020. Available from: https://www.ogcio.gov.hk/en/about_us/facts/doc/householdreport2020_69.pdf. Accessed 5 May 2020.

12. Census and Statistics Department, Hong Kong SAR Government. Population– Overview. 2020. Available from: https://www.censtatd.gov.hk/hkstat/sub/so20.jsp. Accessed 12 Apr 2020.

13. Centre for Health Protection, Department of Health, Hong Kong SAR Government. Latest situation of novel coronavirus infection in Hong Kong. Available from: https://chp-dashboard.geodata.gov.hk/covid-19/en.html. Accessed 12 Apr 2022.

14. Borgström F, Karlsson L, Ortsäter G, et al. Fragility fractures in Europe: burden, management and opportunities. Arch Osteoporos 2020;15:59.

15. Leung F, Lau TW, Kwan K, Chow SP, Kung AW. Does timing of surgery matter in fragility hip fractures? Osteoporos Int 2010;21 Suppl 4:S529-34.

16. British Orthopaedic Association. COVID BOAST-Management of patients with urgent orthopaedic conditions and trauma during the coronavirus pandemic. Available from: https://www.boa.ac.uk/resources/covid-19-boasts-combined1.html. Accessed 13 Feb 2023.

17. Tsuda T, Hashimoto Y, Okamoto Y, Ando W, Ebina K. Meta-analysis for the efficacy of bisphosphonates on hip fracture prevention. J Bone Miner Metab 2020;38:678-86.

18. Fritz J, Cöster ME, Nilsson JA, Rosengren BE, Dencker M, Karlsson MK. The associations of physical activity with fracture risk—a 7-year prospective controlled intervention study in 3534 children. Osteoporos Int 2016;27:915-22.

19. Morseth B, Ahmed LA, Bjørnerem Å, et al. Leisure time physical activity and risk of non-vertebral fracture in men and women aged 55 years and older: the Tromsø study. Eur J Epidemiol 2012;27:463-71.

20. Appleby PN, Allen NE, Roddam AW, Key TJ. Physical activity and fracture risk: a prospective study of 1898 incident fractures among 34,696 British men and women. J Bone Miner Metab 2008;26:191-8.

21. Nilson F, Moniruzzaman S, Andersson R. A comparison of hip fracture incidence rates among elderly in Sweden by latitude and sunlight exposure. Scand J Public Health 2014;42:201-6.

22. Sing CW, Woo YC, Lee AC, et al. Validity of major osteoporotic fracture diagnosis codes in the Clinical Data Analysis and Reporting System in Hong Kong. Pharmacoepidemiol Drug Saf 2017;26:973-6.

23. Hospital Authority, Hong Kong SAR Government. HA adjusts service provision to focus on combatting epidemic. 2020. Press Release. Available from: https://www.info.gov.hk/gia/general/202002/10/P2020021000711.htm. Accessed 20 May 2020.

24. Bergh C, Wennergren D, Möller M, Brisby H. Fracture incidence in adults in relation to age and gender: a study of 27,169 fractures in the Swedish Fracture Register in a well-defined catchment area. PLoS One 2020;15:e0244291.