Data Resource Profile: The Hong Kong Diabetes Surveillance Database (HKDSD)

Abstract

The Hong Kong Diabetes Surveillance Database (HKDSD) is a territory-wide data resource set up to investigate the epidemiology of gestational diabetes, prediabetes and diabetes in Hong Kong. It includes all people who have ever had a measurement of blood glucose since 2000 in the Hospital Authority (HA) electronic medical record (EMR) system, which captures clinical information on all people attending public hospitals and clinics. The HKDSD contains routinely collected longitudinal data on demographic information, hospital admissions and discharges, diagnoses and procedures, laboratory measurements, drug prescriptions and mortality. In 2000, the HA implemented a diabetes risk-assessment and management programme at 18 hospital-based diabetes centres adapted using the protocol of the Hong Kong Diabetes Register (HKDR). In 2009, the programme was extended to the primary care setting including the risk-stratification component of the Joint Asia Diabetes Evaluation (JADE) Programme, which evolved from the HKDR. Data of the risk assessments were captured in the Risk Assessment and Management Programme for Diabetes Mellitus (RAMP-DM) module in the HA EMR system. Between 2000 and 2019, the HKDSD curated data from 4 089 903 people with at least one glucose measurement, among whom 581 811 underwent structured assessment with data in the RAMP-DM module. Analysis of the database has generated important findings which contributed to the current body of knowledge related to diabetes epidemiology in Asia. Diabetes has become one of the most challenging health problems, which affects around 1 in 10 adults worldwide.1,2 The last three decades have witnessed a significant change in the global epidemiological patterns of diabetes.3,4 However, the existing large epidemiological studies are mainly from North America and Europe. Population-based representative data are lacking in Asia,3–5 where 60% of the world’s population with diabetes are living.2 Asians have a different diabetes phenotype compared with White European ancestry, with different countries in Asia undergoing epidemiological transition at various stages.6 Large diabetes databases, such as population-based diabetes registries and those generated from a national electronic medical record (EMR), are important data sources for diabetes epidemiological research. These resources are representative of the study population with findings that are generalizable.7 The Hong Kong Diabetes Surveillance Database (HKDSD) includes all people who have ever had a measurement of blood glucose since 2000 in the public healthcare sector in Hong Kong. The HKDSD was established with the major aims of: (i) integrating EMR data for surveillance of gestational diabetes (GDM), prediabetes, diabetes and diabetes-related complications in Hong Kong; (ii) evaluating the quality of care for diabetes, exploring and quantifying the effect of clinical factors and interventions on health outcomes of diabetes; (iii) identifying priorities for prevention and treatment of diabetes, and curating evidence to inform clinical practice and policy-making. Hong Kong is a special administrative region of the People’s Republic of China with a population of ∼7.5 million. Since the transfer of sovereignty from the UK in 1997, Hong Kong has had a high degree of autonomy with a healthcare system that is separate from that of mainland China. The healthcare system of Hong Kong runs on a dual-track basis encompassing both public and private sectors (Supplementary Figure S1, available as Supplementary data at IJE online). The public healthcare system in Hong Kong is adapted from the UK National Health Service, with a single public healthcare provider that serves as a safety net through universal health coverage.8 The public healthcare system, heavily subsidized by the Hong Kong government, provides ∼90% of all health services for chronic health conditions.9 In 1990, the Hong Kong government established the Hospital Authority (HA) to manage all hospitals and the majority of clinics in the public sector (Figure 1).10 As of 31 March 2020, there were 43 hospitals/institutions, 49 specialist outpatient clinics (SOPCs) mainly based in hospitals and 73 community-based general outpatient clinics (GOPCs). These health institutions are organized into seven clusters based on locations with a population of ∼1 million within each cluster (Figure 2). Timeline of public healthcare reform and evolution of diabetes care in Hong Kong Distribution of hospitals/institutions, specialist outpatient clinics and general outpatient clinics in the public sector governed by the Hong Kong Hospital Authority in Hong Kong In 2000, the Hong Kong HA developed a territory-wide EMR system to routinely collect clinical information for all people attending public hospitals/institutions, SOPCs and GOPCs (Figure 1). A unique identification number, the Hong Kong Identity Card number (HKID), which is compulsory for all residents in Hong Kong, is used in the HA EMR. The HKID enables linkages between the HA EMR and other data sets, such as the Hong Kong Death Registry. In addition, the database includes a unique clinical identifier number for each person, generated by the HA EMR, which allows anonymous analysis of the data. In 1995, the Prince of Wales Hospital (PWH), the teaching hospital of the Chinese University of Hong Kong (CUHK), initiated a structured diabetes assessment and education programme as a research-driven quality-improvement programme (Figure 1). Dedicated clinic sessions were created at the CUHK-PWH Diabetes Centre where trained nurses performed protocol-guided assessment including blood, urine, eye and feet to establish the Hong Kong Diabetes Register (HKDR). The data structure of the HKDR was adapted from the International Diabetes Federation St. Vincent’s Declaration, which advocated the establishment of diabetes registers to improve the quality of diabetes care.11 In 2000, when the HA EMR was launched, the HA adopted the HKDR protocol and implemented a diabetes risk-assessment programme by establishing 18 hospital-based diabetes centres to provide education, supporting services and regular comprehensive assessment of metabolic control, micro- and macro-vascular complications of diabetes to people with diabetes referred from both public SOPCs and GOPCs (Figure 1).8 All referrals were at the discretion of treating physicians with no specific referral criteria. All data were captured in the Risk Assessment and Management Programme for Diabetes Mellitus (RAMP-DM) module in the HA EMR. In 2007, the HKDR evolved to become the Joint Asia Diabetes Evaluation (JADE) Programme, which includes risk categories based on HKDR risk equations to facilitate care triage and personalized management.8 In 2009, the HA adapted this risk-stratification programme and extended the RAMP-DM from the 18 diabetes centres in the secondary care setting to all GOPCs in the primary care setting. Since then, all patients who were enrolled in the RAMP-DM underwent structured assessments by trained nurses at the GOPCs 12 or hospital-based diabetes centres. With full implementation of the RAMP-DM programme, ∼60% of Hong Kong people with diabetes received care in the primary care setting.13 The RAMP-DM module in the HA EMR consists of a structured database and standardized reporting system, and has improved communication and patient flow between the primary and secondary care settings, as well as between clinics, especially for patients attending multiple clinics. The HKDSD includes all people who have ever had a measurement of either fasting plasma glucose (FPG), random plasma glucose, 2-hour oral glucose tolerance test (2-h OGTT) plasma glucose or haemoglobin A1c (HbA1c) in the HA EMR since 1 January 2000. Therefore, the HKDSD consists of a cohort of people with normal blood glucose levels, women with GDM, people with prediabetes, people with diabetes and a subgroup of people with diabetes with structured data collection in the RAMP-DM module through regular metabolic risk-assessment and diabetes-complication screening. The definitions for GDM, prediabetes and diabetes in the HKDSD are shown in the Supplementary Table S1 (available as Supplementary data at IJE online). Between 1 January 2000 and 31 December 2019, the HKDSD curated data from 4 089 903 people (46.2% men) with at least one glucose measurement in the public sector, among whom 76 191 were women who had ever had GDM, 1 028 931 people (49.0% men) who had ever had prediabetes, 964 950 people (50.7% men) who had ever had diabetes and 581 811 people (50.6% men) who had structured data captured by the RAMP-DM module (Supplementary Figures S2 and S3, available as Supplementary data at IJE online). Table 1 lists the variables for people included in the HKDSD and enrolled in the RAMP-DM module. Individual-level longitudinal data in the HKDSD including demographics, hospital admissions and discharges, diagnoses and procedures, laboratory measurements and drug prescriptions are obtained from the HA EMR. Diagnoses and procedures at the SOPCs and during the hospitalization are coded using ICD-9. Previous studies have demonstrated high coding accuracy in the HA EMR.14,15 Laboratory measurements including glycaemic indexes, lipid profile, liver function, kidney function, hepatitis markers, hematologic markers, electrolytes and hormones are captured from both outpatients and inpatients. Drug data including insulin, oral glucose-lowering drugs (OGLDs), renin–angiotensin–aldosterone system (RAAS) inhibitors, blood-pressure-lowering drugs, lipid-lowering drugs, anti-platelet drugs and anticoagulants, uric-acid-lowering drugs and psychotropic drugs are captured from both outpatient and inpatient prescriptions. Drug classes are recorded following the British National Formulary Classification.16 The HA EMR has been linked to the Hong Kong Death Registry using HKID, allowing complete follow-up until time of death. The underlying cause of death is coded using ICD-9 until the end of 2000 and ICD-10 thereafter. All people in the HKDSD are followed up using the HA EMR until death, except those who have emigrated out of Hong Kong or were treated in the private sector. People’s emigration status and date are unknown to allow for censoring of follow-up. Summary of measurements for people in the HKDSD and enrolled in the RAMP-DM module between 2000 and 2019 HKDSD, Hong Kong Diabetes Surveillance Database; RAMP-DM, Risk Assessment and Management Programme for Diabetes Mellitus; GOPC, general outpatient clinic; SOPC, specialist outpatient clinic; FPG, fasting plasma glucose; 2-h OGTT, 2-hour oral glucose tolerance test; OGLDs, oral glucose-lowering drugs; RAAS, renin–angiotensin–aldosterone system; CABG, coronary artery bypass graft; anti-VEGF Rx, anti-vascular endothelial growth factor therapy. The RAMP-DM module includes a standardized clinical protocol adopted from the HKDR and JADE Programme8 to guide comprehensive assessment of risk factors and diabetes-related complications. The assessment includes basic parameters (e.g. age at diabetes diagnosis), lifestyles (e.g. use of tobacco and alcohol), physical examination [e.g. blood pressure and body mass index (BMI)], laboratory measurements (e.g. HbA1c and blood cholesterol), medical history and examinations of eyes and feet (Table 1). Based on the results of the risk assessment, people are given appropriate care by a team of multidisciplinary healthcare professionals in the primary and secondary settings. As the programme evolves, there are reminders in the EMR to encourage physicians to refer people to undergo structured risk assessment every 2–3 years for quality assurance either in the primary or secondary care setting. Between 2000 and 2019, 581 811 people were enrolled in the RAMP-DM module with 2 154 498 assessments. Of these, 72.0% people had undergone assessment at least twice (Supplementary Figure S4, available as Supplementary data at IJE online). The data in the RAMP-DM module are available in the EMR platform accessible by the multidisciplinary team for decision-making and referrals. In the HKDSD, other clinical data during the follow-up of people in the RAMP-DM module are captured using the HA EMR with linkage to the Hong Kong Death Registry. Table 2 shows the key characteristics of people during their first risk-assessment and complication screening in the RAMP-DM module between 2000 and 2019. At enrolment, 50.6% (n = 294 473) were men. The mean [standard deviation (SD)] age was 57.1 (12.4) years at diabetes diagnosis and 61.9 (12.2) years at assessment. Overall, 99.0% (n = 546 682) had type 2 diabetes and 0.6% (n = 3311) had type 1 diabetes. The mean (SD) BMI was 26.0 kg/m2 and the mean (SD) value of HbA1c was 7.5% (1.7%). Cardiovascular disease, chronic kidney disease and cancer were present in 14.6%, 12.6% and 3.9% of the people, respectively, at enrolment. Selected characteristics of people at first risk-assessment and complication screening who were enrolled in the RAMP-DM module between 2000 and 2019 Proportion among people with complete data. Definition for disease history is shown in Supplementary Table S2 (available as Supplementary data at IJE online). RAMP-DM, Risk Assessment and Management Programme for Diabetes Mellitus; SD, standard deviation; BMI, body mass index; FPG, fasting plasma glucose; eGFR, estimated glomerular filtration rate; LDL, low-density lipoprotein; HDL, high-density lipoprotein; IQR, interquartile range; ACR, albumin-to-creatinine ratio; OGLDs, oral glucose-lowering drugs; RAAS, renin–angiotensin–aldosterone system. A key feature of the HKDSD is the comprehensive nature of the longitudinal clinical data that enables the investigation of long-term trends in incidence, control of risk factors, treatment and outcome of people with GDM, prediabetes and diabetes. A series of papers from analyses based on an earlier version of the HKDSD has been published. The earlier version of the HKDSD only included people with diabetes in 2000–2016, and people who had blood glucose measured but not meeting the criteria for diabetes were not included. A selection of key papers is listed as follows. Ke et al. developed and validated an algorithm using diagnosis codes and drug prescriptions to classify incident type 1 and type 2 diabetes in the HKDSD.17 Luk et al. investigated the secular trends in the incidence of type 1 and type 2 diabetes and found an increasing burden of young-onset type 2 diabetes (defined as age of diagnosis <40 years) in Hong Kong from 2002 to 2015.18 By linking clinical profiles to hospital-admission records and the Hong Kong Death Registry, Wu et al. investigated the secular trends in all-cause and cause-specific mortality rates and the rates of major diabetes-related complications among people with diabetes.19–22 Overall, both mortality rates in people with diabetes and standardized mortality ratios comparing people with and without diabetes have declined from 2001 to 2016 in Hong Kong.19 However, the decline was not observed amongst young people aged <45 years, who had a 5- to 9-fold higher risk of mortality compared with those without diabetes in the same age group. Rates of hospitalization for major diabetes-related complications, including cardiovascular disease, hyperglycaemic crisis and lower-extremity amputation, have also decreased but with less improvement in young people during the same study period.20,21 Rates of hospitalization for most infections had been stagnant, with similar trends in people with and without diabetes.22 Yang et al. used drug-prescription and laboratory data, and reported the trends in the use of different classes of OGLDs and glycaemic control among people with diabetes.23 Between 2002 and 2016, we observed an increase in the use of metformin and newer OGLDs, and a decrease in the use of sulfonylureas. Mean HbA1c levels declined after 2007, although persistently poor glycaemic control was observed in young people aged 20–44 years. Ke et al. reported an excess burden-of-hospitalization rate among people with young-onset diabetes compared with those with diabetes diagnosed at or after middle age by any attained age.24 Ke et al. also used longitudinal HbA1c measurements and drug-prescription data to study the association between age at diabetes diagnosis, glycaemic trajectories and responses to OGLDs.25 Compared with those with usual-onset diabetes, people with young-onset diabetes had faster glycaemic deterioration and smaller responses to most OGLDs. Furthermore, using data from the RAMP-DM module, Luk et al. reported declining trends in the risk of cardiovascular–renal complications and improvement in metabolic risk control between 2001 and 2013.13 In the primary care setting, using a propensity score-matching method, Wan et al. reported that participation in the RAMP-DM programme supplemented by a patient-empowerment programme26 was associated with 57%, 12% and 66% decreased risk of cardiovascular disease, microvascular complications and all-cause mortality, respectively, compared with non-participants.12 The HKDSD has a number of major strengths. First, the territory-wide population coverage ensures a high level of generalizability with great applicability for informing healthcare policy-making regarding disease trends, care gaps and emerging patterns. Second, the large volume of routinely collected clinical data and long surveillance period (since 2000) provide a unique opportunity to assess the long-term epidemiological trends and evolution of hyperglycaemia including normal glucose tolerance, GDM, prediabetes and diabetes in Hong Kong. Third, Hong Kong has developed one of the world’s largest diabetes risk-assessment and complication-screening programmes, which covers all people with diabetes receiving care in public primary and secondary care settings, with structured data collection captured by the RAMP-DM module in the EMR system. This ongoing data collection makes it possible to study longitudinal associations between changes in a wide range of clinical factors, lifestyles, treatments and health outcomes of diabetes with enhanced power for causal inference in an Asian population. Last, the very large study population allows the investigation of rare outcomes of diabetes with sufficient statistical power for reliable results. The HKDSD also has several limitations. First, GDM, prediabetes and early-stage diabetes are often asymptomatic. People with undiagnosed GDM, prediabetes or diabetes who do not have any measurement of blood glucose would not be included in the HKDSD. Around 10% of people treated in the private sector are not captured in the HKDSD. Characteristics may differ between people treated in the public and private sectors. It remains unclear whether findings in the HKDSD can be broadly applicable to people receiving care in the private sector. Second, given the nature of administrative data, misclassification bias in the HKDSD is possible. Third, missing data are inevitable even for those who underwent risk-assessment and complication screening with data captured in the RAMP-DM module (Table 2). Fourth, due to the structured risk assessment that prompts intervention and promotes self-management, people with data in the RAMP-DM module have better health outcomes than those who have not been exposed to these assessments12 and thus may not be fully representative of all people with diabetes in Hong Kong. All data of HKDSD are stored in a data repository in the PWH, the teaching hospital of the CUHK, where the HKDR and JADE Programme were first conceptualized, implemented and evaluated.8 According to the rules and regulations of the HA, the individual-level data of HKDSD can only be analysed at the PWH. External parties cannot access the database through the CUHK Diabetes and Endocrine Research Team. Request for access can be made through the HA at https://www3.ha.org.hk/data/Home/Index/. Ethical approval was obtained from the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (CREC Ref. No. 2017.298/2020.032) for analysis of anonymized data. H.W. drafted the first version of the manuscript. H.W., E.S.H.L., A.Y. and X.Z. contributed to data harmonization and analysis. J.C.N.C. and A.O.Y.L. contributed to interpretation of results and drafting of the manuscript. A.O.Y.L. is the guarantor of this work and is the principal investigator of the HKDSD. All authors contributed to critical revision of the manuscript and approval of the final draft. Supplementary data are available at IJE online. The EMR system and RAMP-DM are continuously supported by the Hong Kong Hospital Authority. The curation of the Hong Kong Diabetes Surveillance Database received no external funding. We acknowledge the Hong Kong Hospital Authority for providing anonymized data. We declare there is no conflict of interest.

Topics & Concepts