BMJ Open Is Committed to Open Peer Review. As Part of This Commitment We Make the Peer Review History of Every Article We Publish Publicly Available

BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

BMJ Open is committed to open peer review. As part of this commitment we make the peer review history of every article we publish publicly available.

When an article is published we post the peer reviewers’ comments and the authors’ responses online. We also post the versions of the paper that were used during peer review. These are the versions that the peer review comments apply to.

The versions of the paper that follow are the versions that were submitted during the peer review process. They are not the versions of record or the final published versions. They should not be cited or distributed as the published version of this manuscript.

BMJ Open is an open access journal and the full, final, typeset and author-corrected version of record of the manuscript is available on our site with no access controls, subscription charges or pay-per-view fees (http://bmjopen.bmj.com).

If you have any questions on BMJ Open’s open peer review process please email [email protected] http://bmjopen.bmj.com/ on September 29, 2021 by guest. Protected copyright. BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

10-year Risk Prediction Models of Complications and Mortality of Diabetes Mellitus in Chinese Patients in Primary Care in Hong Kong study protocol

ForJournal: peerBMJ Open review only Manuscript ID bmjopen-2018-023070

Article Type: Protocol

Date Submitted by the Author: 19-Mar-2018

Complete List of Authors: WAN, Eric Yuk Fai; University of Oxford, Nuffield Department of Population Health Yu, Esther Yee Tak; The University of Hong Kong, Department of Family Medicine and Primary Care Chin, Weng Yee; The University of Hong Kong, Department of Family Medicine & Primary Care Fung, Colman; The University of Hong Kong Kwok, Ruby Lai Ping; Hospital Authority, Primary & Community Services Chao, David Vai Kiong; United Christian Hospital and Tseung Kwan O Hospital, Department of Family Medicine and Primary Health Care Chan, King Hong; Hospital Authority Kowloon Central Cluster, Family Medicine & Primary Healthcare Hui, Eric Ming-Tung; Hospital Authority New Territories East Cluster, Family Medicine http://bmjopen.bmj.com/ Tsui, Wendy Wing Sze; Hospital Authority Hong Kong West Cluster, Family Medicine & Primary Healthcare TAN, Kathryn; The University of Hong Kong, Department of Medicine Fong, Daniel Yee Tak; The University of Hong Kong, School of Nursing Lam, Cindy; The University of Hong Kong, Department of Medicine

Diabetes Mellitus, Risk, Complications, Mortality, Cardiovascular Disease, Keywords: Chinese on September 29, 2021 by guest. Protected copyright.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 1 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 10-year Risk Prediction Models of Complications and Mortality of Diabetes Mellitus in 4 5 Chinese Patients in Primary Care in Hong Kong study protocol 6 7 8 9 Authors: Eric Yuk Fai Wan, PhD1, Esther Yee Tak Yu, MBBS1, Weng Yee Chin, MD1, 10 11 Colman Siu Cheung Fung, MBBS1, Ruby Lai Ping Kwok, MHSE2, David Vai Kiong Chao, 12 13 3 4 5 14 MBBS , King Hong Chan, MBBS , Eric Ming-Tung Hui, MBBS , Wendy Wing Sze Tsui, 15 6 7 8 16 MBBS , KathrynFor Choon Bengpeer Tan, MD review, Daniel Yee Tak Fong,only PhD , Cindy Lo Kuen Lam, 17 18 MD1 19 20 21 22 23 1 24 Department of Family Medicine and Primary Care, The University of Hong Kong, Hong 25 26 Kong 27 28 2 29 Department of Primary & Community Services, Hospital Authority Head Office, Hospital 30 31 Authority, Hong Kong 32 33 http://bmjopen.bmj.com/ 3 Department of Family Medicine & Primary Healthcare, Kowloon East Cluster, Hospital 34 35 36 Authority, Hong Kong 37 38 4 Department of Family Medicine & Primary Healthcare, Kowloon Central Cluster, Hospital 39 40 41 Authority, Hong Kong on September 29, 2021 by guest. Protected copyright. 42 43 5 Department of Family Medicine, New Territories East Cluster, Hospital Authority, Hong 44 45 46 Kong 47 48 6 Department of Family Medicine & Primary Healthcare, Hong Kong West Cluster, Hospital 49 50 51 Authority, Hong Kong 52 53 7 Department of Medicine, the University of Hong Kong, Hong Kong 54 55 56 8 School of Nursing, the University of Hong Kong, Hong Kong 57 58 59 1 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 2 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 4 5 6 Corresponding Author: 7 8 Cindy Lo Kuen Lam 9 10 11 Email: [email protected] 12 13 14 Telephone: + (852) 25524690 15 16 For peer review only 17 Fax: + (852) 28147475 18 19 20 21 22 Word Count: 3,935 words 23 24 25 Keywords: Diabetes Mellitus; Risk; Complications; Mortality; Cardiovascular Diseases; 26 27 Chinese 28 29 30 31 32 33 http://bmjopen.bmj.com/ 34 35 36 37 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 2 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 3 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Abstract 4 5 Introduction 6 7 Diabetes mellitus (DM) is a major disease burden worldwide because it is associated with 8 9 disabling and lethal complications. DM complication risk assessment and stratification is key 10 11 to cost-effective management and tertiary prevention for diabetic patients in primary care. 12 13 14 Existing risk prediction functions were found to be inaccurate in Chinese diabetic patients in 15 16 primary care. For This study peer aims to developreview 10-year risk only prediction models for total 17 18 cardiovascular diseases (CVD) and all-cause mortality among Chinese patients with DM in 19 20 primary care. 21 22 23 24 25 Methods and analysis 26 27 A 10-year cohort study on a population-based primary care cohort of Chinese diabetic 28 29 patients, who were receiving care in the Hospital Authority General Out-Patient Clinic on or 30 31 before 1 January 2008, were identified from the clinical management system database of the 32 33 Hospital Authority. All patients with complete baseline risk factors will be included and http://bmjopen.bmj.com/ 34 35 followed from 1 January 2008 to 31 December 2017 for the development and validation of 36 37 prediction models. The analyses will be carried out separately for men and women. Two-third 38 39 40 of subjects will be randomly selected as the training sample for model development. Cox 41 on September 29, 2021 by guest. Protected copyright. 42 regressions will be used to develop 10-year risk prediction models of total CVD and all-cause 43 44 mortality. The validity of models will be tested on the remaining one-third of subjects by 45 46 Harrell's C statistics and calibration plot. Risk prediction models for diabetic complications 47 48 specific to Chinese patients in primary care will enable accurate risk stratification, 49 50 prioritization of resources and more cost-effective interventions for DM patients in primary 51 52 53 care. 54 55 56 57 58 59 3 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 4 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Ethics and dissemination 4 5 The study was approved by the Institutional Review Board of the University of Hong Kong - 6 7 the Hospital Authority Hong Kong West Cluster (reference number: UW 15-258). 8 9 10 11 Trial registration: US Clinical Trial Registry NCT03299010 12 13 14 15 16 Strengths and limitationsFor peer of this study review only 17 18 • This is a 10-year retrospective population-based cohort study cohort of Chinese DM 19 20 patients in primary care which can represent the situation in Hong Kong. 21 22 • Two-third of samples in the cohort will be randomly selected for developing risk 23 24 25 prediction models while remaining one-third would be used for validation to ensure the 26 27 performance of models. 28 29 • Risk prediction nomograms and charts will be established based on the risk prediction 30 31 models for a convenient use in clinical setting. 32 http://bmjopen.bmj.com/ 33 • Multiple imputation will be used to handle missing data to minimize the bias in 34 35 36 developing risk prediction models. 37 38 • Misclassification bias may exist by using diagnosis coding such as ICPC-2 and ICD-9CM 39 40 to identify the outcome events of patients. 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 4 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 5 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Manuscript Text 4 5 6 7 Introduction 8 9 Diabetes Mellitus (DM) is a well-recognized public health issue, affecting 415 million people 10 11 and costing HK$5.2 trillion in global health expenditures worldwide.1 DM can lead to many 12 13 14 complications resulting in morbidity and mortality. According to the International Diabetes 15 16 Federation (IDF),For in 2015, peer diabetes led toreview 5.0 million (14.5% only of all deaths) deaths worldwide 17 18 which translated to one death every six seconds and approximately 70% of DM related deaths 19 20 were attributed to cardiovascular diseases (CVD).1 The development of diabetes-related 21 22 complications significantly increases medical costs.1 23 24

25 26 27 To prevent DM complications, the American Heart Association (AHA) guidelines 28 29 recommend primary care providers to provide regular assessment and management of risk 30 31 factors for patients especially those who are at high risk of developing DM complications. 32 33 Although the National Cholesterol Education Programme (NCEP) in the United States has http://bmjopen.bmj.com/ 34 35 suggested that all diabetic patients be treated as if they had coronary heart disease (CHD), 36 37 however the observed rate of CVD vary vastly among different diabetic patients.2 The 38 39 40 American Diabetes Association (ADA) and the Canadian Diabetes Association guidelines 41 on September 29, 2021 by guest. Protected copyright. 42 both include 10-year overall CVD risk stratification into account to identify high-risk patients 43 44 for more intensive medical and psychosocial interventions.3 The guidance of stain 45 46 prescription from the American College of Cardiology and the American Heart Association, 47 48 which is consistent with the ADA, also takes predicted 10-year overall CVD risk into 49 50 account.4 The ADA recommends aspirin treatment for diabetic patients with a 10-year 51 52 3 53 predicted over CVD risk higher than 10%. Studies in the United States, the United Kingdom, 54 55 Australia, New Zealand and Hong Kong showed that systematic risk assessment and risk- 56 57 58 59 5 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 6 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 stratified management initiatives in primary care settings could improve clinical outcomes 4 5 such as haemoglobin A1c (HbA1c), blood pressure (BP) and low-density lipoprotein 6 7 cholesterol (LDL-C), as well as reduce utilization of health services including accident and 8 9 emergency (A&E) attendance, and hospital admissions.5 In 2009, the Hong Kong Hospital 10 11 Authority (HA) made an initiation to enhance the quality of DM care in all HA primary care 12 13 14 clinics by the introduction of the Multi-disciplinary Risk Assessment and Management 15 16 Programme – DiabetesFor Mellitus peer (RAMP-DM) review to systematically only assess the CVD risk of DM 17 18 patients and then managed according to risk-stratified protocols.6 19 20 21 22 A key to cost-effective management of DM is an accurate risk assessment and stratification 23 24 system that identifies high-risk patients for more intensive medical and psychosocial 25 26 interventions. At the same time, an accurate estimation of risk distribution can inform policy- 27 28 29 makers to allocate appropriate resources and plan services that can maximize population 30 31 health benefit for DM patients. Most of risk-stratified intervention in the guidance were based 32 33 on common prediction functions for 10-years risk including the Framingham,7 8 QRisk 9 10 http://bmjopen.bmj.com/ 34 35 and the European Systematic Coronary Risk Evaluation (SCORE).11 However, most of 36 37 existing prediction models were established and validated for western populations. Our 38 39 previous studies developed a series of models for 5-years DM-related complications 40 on September 29, 2021 by guest. Protected copyright. 41 12-14 42 including CVD, end stage renal disease (ESRD) and all-cause mortality, and showed that 43 44 other 5-years risk prediction model based on non-Chinese populations such as Framingham, 45 46 the Action in Diabetes and Vascular Disease: Preterax and Diamicron-MR Controlled 47 48 Evaluation (ADVANCE), Swedish and New Zealand CVD risk scores either underestimated 49 50 or overestimated the risk for Chinese patients.12-14 Indeed, the prevalence of CVD in Chinese 51 52 populations was only half of that in Caucasian populations.15 16 A recent observational study 53 54 55 also illustrated that CVD risk, even in Asian populations, varied widely among the Malay, 56 57 58 59 6 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 7 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 17 3 Asian Indian, and Chinese populations. In addition, several multiethnic studies showed 4 5 substantial difference in the incidence rate among different racial groups with a generally 6 7 higher incidence rate of renal disease in Chinese than in non-Chinese DM patients.18-20 The 8 9 IDF report revealed that 1.3 million Chinese died because of DM in 2015, which represented 10 11 the highest prevalence of DM-related mortality across various ethnic groups and was two- 12 13 1 14 fold higher than those found in Europeans and Australians. These discrepancies in CVD, 15 16 ESRD, and all-causeFor mortality peer risk may reviewbe related to the differences only in the disease profile and 17 18 other determinants such as genetics, health care policy, and culture.15 16 21-23 Therefore, 19 20 Chinese population specific risk prediction models are necessary. 21 22 23 24 Inaccurate risk stratification may lead to inappropriate risk-stratified interventions. There is a 25 26 need for new robust risk prediction models to predict 10-year CVD risk and mortality for 27 28 29 primary care Chinese patients in order to enable accurate risk stratification of DM patients in 30 31 the HA on-going RAMP-DM or other primary care systematic risk-stratified multi- 32 33 disciplinary management programmes. Furthermore, robust risk prediction models for the http://bmjopen.bmj.com/ 34 35 overall prediction of first CVD and all-cause mortality can inform policy makers in service 36 37 planning and resource allocation. 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 Aims and Objectives 43 44 This study aims to develop 10-year risk prediction models for total CVD and all-cause 45 46 mortality among Chinese diabetic patients in primary care. Risk prediction models for 47 48 individual DM complications including CHD, heart failure, stroke and ESRD will also be 49 50 developed. 51 52

53 54 55 The objectives are to: 56 57 58 59 7 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 8 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 1. Calculate the 10 years incidence of total CVD, all-cause mortality and each major DM 4 5 complication in Chinese DM patients in primary care. 6 7 2. Determine the risk factors that significantly predict total CVD, all-cause mortality and 8 9 each major DM complication for Chinese DM patients in primary care. 10 11 3. Develop and validate risk prediction models for total CVD, all-cause mortality and each 12 13 14 major DM complication for Chinese DM patients in primary care. 15 16 4. Develop a riskFor prediction peer nomogram review and chart for theonly risk of total CVD, all-cause 17 18 mortality for Chinese DM patients in primary care. 19 20 21 22 The following hypotheses will be tested: 23 24 1. Patient socio-demographic, clinical parameters, disease characteristics, and treatment 25 26 27 modalities (these independent variables are described in the Methods/Design section) are 28 29 predictive of 10-year risk of total CVD, all-cause mortality and individual DM 30 31 complication as a dependent variable. 32 33 2. The risk prediction models for total CVD, all-cause mortality and individual DM http://bmjopen.bmj.com/ 34 35 complication developed in this study can have over 70% of discriminating power. 36 37

38 39 40 Methods and analysis 41 on September 29, 2021 by guest. Protected copyright. 42 Study design 43 44 A 10-year retrospective study on a population-based cohort of Chinese DM patients in 45 46 primary care. 47 48 49 50 Subjects 51 52 The cohort will include all patients with a documented clinical diagnosis of DM and were 53 54 55 receiving care in the HA primary care General Out-Patient Clinics (GOPC) on or before 1 56 57 58 59 8 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 9 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 January 2008 identified from the HA clinical management system (CMS) database. 4 5 6 7 The inclusion criteria are patients aged 18 years old or older, had at least 1 GOPC/ Family 8 9 Medicine Clinics (FMC) attendance on or within 1 year before 1 January 2008 and had a 10 11 CMS record of the coding of International Classification of Primary Care, Second edition 12 13 14 (ICPC-2) of T89 (Diabetes insulin dependent) or T90 (Diabetes non-insulin dependent) on or 15 16 before 1 JanuaryFor 2008. peer review only 17 18 19 20 The exclusion criteria are patients who had a diagnosis of any DM complications defined by 21 22 the relevant ICPC-2 or The International Classification of Diseases, Ninth Revision, Clinical 23 24 Modification (ICD-9-CM) (shown in the section below) on or before 1 January 2008 and 25 26 patients exclusively managed by Specialist Out-Patient Clinic (SOPC) on or before 1 January 27 28 29 2008. 30 31 32 33 Sample Size Calculation http://bmjopen.bmj.com/ 34 35 The required sample size is based on the requirements for the development and validation of 36 37 the least common DM complication of ESRD. Specifically, based on our previous study, the 38 39 5-year incidence of ESRD was 1.9%,6 which can be extrapolated to a 10-year incidence of 40 41 on September 29, 2021 by guest. Protected copyright. 42 ESRD was 3.8% by assuming a constant incidence rate over time. To develop the risk 43 44 prediction model for ESRD by multivariable Cox proportional hazard regressions with 45 46 forward stepwise variables selection on 16 potential risks factors, we need 21,053 subjects 47 48 using the 1 in 50 rule that 1 candidate predictor can be studied for every 50 events.24 To 49 50 validate the risk prediction models, the area under the receiver operating characteristic curve 51 52 (AUC) will be used to assess the discriminatory power of the predictive model on a separate 53 54 6 55 sample of subjects. With 10 years incidence rate of ESRD was 3.8%, a total 14,307 subjects 56 57 58 59 9 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 10 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 (544 subjects with ESRD and 13,763 subjects without ESRD) is needed to ensure a precision 4 5 of 0.05 for AUC of 0.7 by a 95% confidence interval. Hence, a total 35,360 male and 35,360 6 7 female are needed for the development (training dataset) and validation of risk prediction 8 9 models stratified by gender. 10 11

12 13 14 Definition of Indicator DM Complications 15 16 The incidence ofFor four major peer DM complications review (CHD, stroke, only heart failure and ESRD), total 17 18 CVD and all-cause mortality will be calculated. The incidence is counted from the earliest 19 20 date of documented diagnosis defined by the relevant ICPC-2 and/or ICD-9-CM coding 21 22 recorded in the HA CMS database from 1 January 2008 to 31 December 2017. The relevant 23 24 ICPC-2 and ICD-9-CM codes of each DM complication and mortality are determined by the 25 26 27 academic and HA clinician co-investigators as listed below: 28 29 30 31 1. CHD (ischaemic heart disease, myocardial infarction (MI), coronary death or sudden 32 33 death) is defined by any of ICPC-2 K74 to K76 and ICD-9-CM 410.x, 411.x to 414.x, http://bmjopen.bmj.com/ 34 35 798.x 36 37 2. Stoke (fatal and non-fatal stroke) is defined by any of ICPC-2 K89 to K91 or ICD-9-CM 38 39 40 430.x to 438.x. 41 on September 29, 2021 by guest. Protected copyright. 42 3. Heart failure is defined by any of ICPC-2 K77 or ICD-9-CM 428.x 43 44 4. CVD is defined as the presence of any of CHD, heart failure and stroke ICPC-2 or ICD-9- 45 46 CM codes listed in 1, 2 and 3 above. 47 48 5. ESRD is defined by any of ICD-9-CM 250.3x, 585.x, 586.x, or an estimated glomerular 49 50 filtration rate (eGFR) < 15mL/min/1.73m2. 51 52 53 6. Mortality is identified from the Hong Kong Death Registry. 54 55 56 57 58 59 10 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 11 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Risk Factors to be Included in the Risk Prediction Models 4 5 Risk factors (independent variables) previously found to be associated with DM 6 7 complications from the literature2 and those that are routinely available in primary care are 8 9 selected to strike a balance between comprehensiveness and feasibility. The potential risk 10 11 factors that will be explored include those related to patient’s socio-demographics, clinical 12 13 14 parameters, disease characteristics and treatment modalities. Patient socio-demographics 15 16 include sex, ageFor and smoking peer status. Clinicalreview parameters includeonly body mass index (BMI), 17 18 HbA1c, systolic and diastolic blood pressure (SBP and DBP), lipid profile (total cholesterol, 19 20 HDL-C, LDL-C, triglyceride), eGFR and albuminuria. Disease characteristics include the 21 22 duration of DM and co-morbidity. Treatment modalities include the use of specific anti- 23 24 hypertensive drugs, insulin, specific oral anti-diabetic drugs and lipid-lowering agents. The 25 26 27 operational definitions of the risk factors are shown in Supplementary File 1. These factors, 28 29 except sex, age, duration of DM and co-morbidity, are modifiable, which have implications 30 31 for practice. 32 33 http://bmjopen.bmj.com/ 34 35 Data Collection 36 37 In the middle of 2018, anonymous data from 1 January 2008 to 31 December 2017 of all DM 38 39 40 patients who satisfy the inclusion criteria and without any exclusion criteria will be extracted 41 on September 29, 2021 by guest. Protected copyright. 42 by the HA statistics team from the HA CMS database. We have successful experience in 43 44 working with the HA in the extraction of similar data from 2009 to 2013 for our extended 45 46 evaluation on quality of care and effectiveness of RAMP-DM study,5 and we have obtained 47 48 preliminary agreement from the HA for the data extraction in the present study. 49 50

51 52 53 Outcome Measures 54 55 1. The incidence of total CVD, all-cause mortality and each of 4 major DM complications 56 57 58 59 11 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 12 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 over 10 years 4 5 2. Factors predictive of total CVD, all-cause mortality and each of 4 major DM 6 7 complications over 10 years 8 9 3. 10-year risk prediction models for total CVD, all-cause mortality and each of 4 major DM 10 11 complications 12 13 14 4. Factors that have sufficient power to classify Chinese DM patients in primary care into 15 16 risk group inFor terms of totalpeer CVD and review all-cause mortality only 17 18 19 20 Data Processing and Analysis 21 22 The cohort will be stratified by gender. Descriptive statistics will be used to calculate the 23 24 incidence of total CVD, all-cause mortality and each of 4 major DM complications will be 25 26 27 analysed annually and cumulatively over 10 years with a 95% confidence interval. The 28 29 distribution of risk factors will be cross-tabulated by complication or mortality events. The 30 31 10-year cumulative incidence of various DM complications and mortalities will be further 32 33 analysed by Kaplan-Meier method. The Kaplan-Meier survival curve will be used to describe http://bmjopen.bmj.com/ 34 35 the survivorship of total CVD, all-cause mortality and each of 4 major DM complications in 36 37 the study cohort over 10 years. Unadjusted associations between the risk factors and odds of 38 39 40 events will be assessed by independent t-test for continuous variables or chi-square test for 41 on September 29, 2021 by guest. Protected copyright. 42 categorical variables. 43 44 45 46 The cohort will be randomly split on a 2:1 basis, with two-third sample used for developing 47 48 the risk prediction models, and the other one-third sample used for validation of the risk 49 50 prediction models. The analyses will be carried out separately for men and women. 51 52 53 54 55 Development of Risk Prediction Models 56 57 58 59 12 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 13 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Cox proportional hazard regressions with forward stepwise method will be used to develop 4 5 the risk prediction models for total CVD, all-cause mortality and each of 4 major DM 6 7 complications. If the main term of a clinical parameter is selected in the models, the quadratic 8 9 term of such clinical parameter will be evaluated. Afterwards, the interaction terms between 10 11 selected predictors and age will be also examined in the risk prediction models. Cox 12 13 14 regression is the method most commonly used in risk prediction models in the Framingham 15 7 8 25 26 16 Heart Study andFor UKPDS peer . It allows review us to estimate theonly risk of disease or death for an 17 18 individual, given their prognostic variables. A positive hazard ratio means a higher likelihood 19 20 of event associated with that specific variable. Conversely, a negative hazard ratio means a 21 22 lower likelihood of the event associated with that specific variable. The key proportional 23 24 hazards assumption will be assessed by examining plots of the scaled Schoenfeld residuals 25 26 27 against time for the covariates. Any non-random pattern indicates a violation of the 28 29 proportional hazards assumptions in which case transformation of covariates may be 30 31 necessary. For example, all continuous variables were naturally logarithmically transformed 32 33 to minimize the influence of extreme values and to improve discrimination and calibration of http://bmjopen.bmj.com/ 34 35 the models. A parametric approach such as exponential or Weibull distribution for the hazard 36 37 function can also be carried out. A total of 6 risk prediction models will be established for 38 39 40 total CVD, all-cause mortality and each of 4 major DM complications. The log of the hazard 41 on September 29, 2021 by guest. Protected copyright. 42 ratio of each selected risk factor in the final model will be used as coefficient weights in the 43 44 prediction model of each relevant outcome. The risk equations for 10 year’s follow-up will be 45 46 established by combining these weights with the survivor function.7 47 48 49 50 51 Validation of Risk Prediction Models 52 53 To validate the risk prediction models for total CVD, all-cause mortality and each of 4 major 54 55 DM complications, the remaining one-third validation sample will be used to estimate the 56 57 58 59 13 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 14 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 risk level of the subjects. The performance of discrimination of the model will be assessed by 4 5 Harrell's C statistic, which is a measure similar to the area under the curve after consideration 6 7 of the censoring pattern of the patients. A Harrell's C statistic less than 0.7 indicates limited 8 9 discriminating power, 0.7 to 0.9 is acceptable, and higher than 0.9 suggests strong 10 11 discrimination of the predictive models.27 The D statistic, R2 statistic and Brier score will be 12 13 14 also calculated for evaluating the predictive power of the model. The D statistic is a measure 15 2 16 of discriminationFor where higherpeer value indicatesreview better discrimination. only The R statistic is a 17 18 measure of explained variation with a higher value indicating better performance. The Brier 19 20 score is a measure of goodness of fit in which a lower value means higher accuracy. 21 22 23 24 Calibration will be used to measure how closely predicted outcomes agree with actual 25 26 27 outcomes. Calibration of the model’s ability to correctly estimate the absolute risks will be 28 29 examined by modified Hosmer-Lemeshow test and calibration plots. The modified Hosmer- 30 31 Lemeshow test for time-to-event data measures how well the predicted probability of the 32 33 expected event rate agrees with the observed event rate, where a p-value higher than 0.05 http://bmjopen.bmj.com/ 34 35 indicates good model calibration. In a calibration plot of the observed incidence of events 36 37 against the predicted risk shows the scatter along the 45° line of perfect fit between predicted 38 39 40 risk and observed incidence of event throughout the entire risk spectrum. 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 Development of a Risk Prediction Nomogram and Chart 45 46 In order to enable the 10-year risk prediction models for total CVD and all-cause mortality to 47 48 be applied in busy clinical setting, risk prediction nomograms and charts will be developed 49 50 for men and women. For the nomogram, the patient’s score for each predictor is plotted on 51 52 53 the appropriate scale and vertical lines are drawn to the line of points to obtain the 54 55 corresponding scores. The score of each predictor will be transformed based on the estimated 56 57 58 59 14 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 15 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 standardized beta coefficient of each predictor from the risk prediction model. For continuous 4 5 predictor such as age, the line with interval depends on its units from minimum to maximum 6 7 values among studies subjects (e.g. 20, 40, 60, 80 years old) will be displayed on the 8 9 nomogram plot and the corresponding scores will be obtained based on the estimated 10 11 standardized beta coefficient of the predictor (e.g. assign age of 20 to 0 point, age of 40 to 2 12 13 14 points, age of 60 to 4 points and age of 80 to 6 points). For categorical predictor such as 15 16 gender, each levelFor of the peer predictor will review be ranked a corresponding only score based on the 17 18 estimated standardized beta coefficient of the predictor (e.g. assign female to 0 point and 19 20 male to 3 points). All scores are summed to obtain a total score. The total score is plotted on 21 22 the total line with corresponding predicted risk of CVD. Moreover, we will develop risk 23 24 prediction charts similar to those developed by the Joint British Society. The most significant 25 26 27 predictors, up to a maximum of five, found in the full Cox regression models will be selected 28 29 to classify subjects into 10-year CVD risk groups of < 10% (low risk), 10%-20% (medium 30 31 risk) and > 20% (high risk). The Kaplan-Meier survival curves of each risk group will be 32 33 developed and compared by log-rank tests to confirm the hazard ratios are significantly http://bmjopen.bmj.com/ 34 35 different among all risk groups. 36 37

38 39 40 STATA software version 13 (STATA Corp, College Station, Texas) will be used for data 41 on September 29, 2021 by guest. Protected copyright. 42 analyses. 5% is used as the level of significance in all statistical tests. 43 44 45 46 Discussion 47 48 Primary care is the entry point of the entire medical system. Therefore, primary care doctors 49 50 need to act as gatekeepers for medical resources. Given the large number and substantial 51 52 heterogeneity of DM patients, the aim of study was to develop several risk prediction models 53 54 55 to facilitate primary care providers in identifying Chinese patients at higher risk of 56 57 58 59 15 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 16 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 developing diabetic complications. This will allow interventions to be implemented to reduce 4 5 the individual, societal, and economic burden. The nomogram and chart can help inform 6 7 clinicians regarding interventions based on overall risk of diabetic complications instead of 8 9 only a single risk factor; in addition, these tools can also be used to educate, motivate, and 10 11 empower patients to prevent future diabetic complications. In terms of policy implications, 12 13 14 misclassification may likely lead to excessive medical treatment, low cost-effectiveness in 15 16 primary prevention,For and potentially peer unnecessary review exposure toonly the risk of adverse drug effects. 17 18 The accurate risk prediction model is particularly useful for screening programs to inform 19 20 decisions concerning service provision for DM primary care to achieve the maximum 21 22 population health benefit. 23 24

25 26 27 We have also identified some potential improvements in the performance of existing risk 28 29 prediction models. Firstly, gender is a factor that is of concern in the analyses of risk factors 30 31 and CVD/mortality because males are typically associated with a higher risk of 32 33 CVD/mortality,28 but statistical adjustment for gender is often insufficient to control for http://bmjopen.bmj.com/ 34 35 varying risk-factor profiles and CVD/mortality incidence.29 Secondly, there are possible 36 37 interaction effects between age and risk factors on the CVD/mortality as the magnitude of the 38 39 30 40 effect of specific risk factors such as LDL-C on the CVD/mortality may decrease with age. 41 on September 29, 2021 by guest. Protected copyright. 42 Thus, the interaction term between age and risk factors should be considered when 43 44 developing the risk prediction models. Thirdly, many studies including our previous studies 45 46 illustrated that there were curvilinear association (J or U shape) between HbA1c/SBP/BMI 47 48 and the risk of CVD/mortality,31-33 and thus the quadratic term of such clinical parameters 49 50 should be evaluated when developing the risk prediction models. Fourthly, recent research 51 52 53 suggested additional clinical parameters such as severity of renal impairment measured by 54 55 eGFR and albumin/creatinine ratio (ACR), and variability of risk factors including HbA1c 56 57 58 59 16 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 17 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 34-37 3 and SBP should be considered to enhance the performance of prediction models. Finally, 4 5 the feasibility of measuring predictors in a routine primary care setting should be considered 6 7 when establishing risk prediction models. For example, some predictors such as serum 8 9 bicarbonate and phosphate included in an ESRD risk prediction model may not be available 10 11 in routine clinical practice, especially in the primary care setting.38 Accordingly, the rationale 12 13 14 for inclusion of such non-standard parameters for risk prediction is often unclear, despite the 15 16 inclusion of suchFor predictors peer improved thereview model prediction onlyaccuracy. However, to develop a 17 18 model with wide applicability in the real world, predictors should be selected to strike a 19 20 balance between comprehensiveness and feasibility. 21 22 23 24 From our previous experience, there were several potential difficulties we may encounter 25 26 27 during the development of risk prediction model. Firstly, as anonymous data are extracted 28 29 from the HA CMS database, there may be some variations of the operation definitions 30 31 throughout this study. To avoid confusion and unnecessary misunderstanding, operational 32 33 definitions (Supplementary File 1) clearly list the definitions of the clinical parameters, http://bmjopen.bmj.com/ 34 35 disease characteristics and treatment modalities according to clinical and professional 36 37 advices. Moreover, it is important to ensure the quality of the data from data collection and 38 39 40 correct use of data. Regular communications between our team and HA will be held to cross- 41 on September 29, 2021 by guest. Protected copyright. 42 validate the method of collecting raw data and the use of the data. Secondly, the outcome 43 44 events in this study will rely on the diagnosis coding such as ICPC-2 and ICD-9CM, which 45 46 may cause misclassification bias. Due to the privacy policy, all extracted data will be 47 48 anonymous, and thus the cross-checking the records in the dataset for auditing purposes will 49 50 not be feasible. However, in routine clinical practice in Hong Kong, clinicians in the clinical 51 52 53 and hospital settings provide ICPC-2 and ICD-9-CM codes, respectively, for each episode of 54 39 40 55 attendance. Moreover, due to the heavily subsidized health care system in Hong Kong, 56 57 58 59 17 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 18 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 patients with chronic diseases and serious complication, e.g. myocardial infarction, mostly 4 5 treated in the HA public health care system. Therefore, our data from HA database should 6 7 have captured nearly all CVD and ESRD outcomes of DM patients who are managed in the 8 9 HA primary care setting. Lastly, the missing data possibly exist in the dataset extracted from 10 11 the database. To preserve the completeness and power of the data, multiple imputation will be 12 13 14 used to deal with the missing data. Comparing with complete-case analysis and single 15 16 imputation, multipleFor imputation peer takes incompletereview data into only account to avoid unnecessary 17 18 biases and producing more reliable and applicable risk prediction models.41 Thus, multiple 19 20 imputation has become one of the most widely-used methods for handling missing data in 21 22 medical research.42 43 This approach was also used in the development of the CVD prediction 23 24 model in the QRisk study. 9 10 25 26 27 28 29 Conclusion 30 31 There is a need for the development of 10-year risk prediction models based on the up-to-date 32 33 population-based cohort for diabetic complications, particularly first total CVD. Chinese http://bmjopen.bmj.com/ 34 35 patients in primary care will enable accurate risk stratification, better prioritization of 36 37 resources and more cost-effective interventions for diabetic patients in primary care. They 38 39 40 can also better inform and empower patients to prevent potential DM complications. At the 41 on September 29, 2021 by guest. Protected copyright. 42 health policy level, the results can inform decisions on service provision in the care of 43 44 diabetes mellitus in primary care to achieve maximum population health benefit. The 45 46 prediction models can also be used as an outcome measure on the potential benefits of 47 48 complication prevention in clinical trials on DM interventions in primary care. 49 50 51 52 53 Ethics and dissemination 54 55 Ethics Approval 56 57 58 59 18 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 19 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Ethics approval of this study was granted by the Institutional Review Board of the University 4 5 of Hong Kong/ Hospital Authority Hong Kong West (UW 15-258). 6 7 8 9 Availability of Data and Materials 10 11 The datasets generated and/or analysed during the current study are not publicly available as 12 13 14 the data is from patient records. 15 16 For peer review only 17 18 Author contribution 19 CLKL initially conceived the study and is the principal investigators of the study. EYFW, 20 21 22 EYTY, WYC, CSCF, DYTF and CLKL helped with the design and implementation of the 23 24 programmes, coordination of the study, drafted and revised the manuscript; EYFW, EYTY, 25 26 WYC, CSCF, RLPK, DVKC, KHC, EMTH, WWST, KCBT, DYTF and CLKL revised the 27 28 manuscript. All authors approved the final version. 29 30 31 32 Acknowledgements 33 http://bmjopen.bmj.com/ 34 The authors are most grateful to the Food and Health Bureau, HKSAR and the Hong Kong 35 36 Hospital Authority, in particular chief of service in primary care in each cluster and Statistics 37 38 and Workforce Planning Department at the Hong Kong Hospital Authority. 39 40

on September 29, 2021 by guest. Protected copyright. 41 42 43 Funding 44 45 This study has been funded by the Health and Medical Research Fund, Food and Health 46 47 Bureau, HKSAR (Project no: 14151181). No funding organization had any role in the design 48 49 and conduct of the study; collection, management, analysis, and interpretation of the data; 50 51 and preparation of the manuscript. All other authors have reported that they have no 52 53 relationships relevant to the contents of this paper to disclose. 54 55 56 57 58 59 19 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 20 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Competing Interest 4 5 The authors declare that they have no competing interests. 6 7 8 9 10 11 12 13 14 15 16 For peer review only 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 http://bmjopen.bmj.com/ 34 35 36 37 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 20 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 21 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 References 4 5 1. International Diabetes Federation. Diabetes Atlas seventh edition, 2015. 6 7 2. Chamnan P, Simmons R, Sharp S, et al. Cardiovascular risk assessment scores for people 8 9 with diabetes: a systematic review. Diabetologia 2009;52(10):2001-14. 10 11 3. American Diabetes Association. Standards of medical care in diabetes—2010. Diabetes 12 13 14 care 2010;33(Supplement 1):S11-S61. 15 16 4. Stone NJ, RobinsonFor JG, peerLichtenstein AH,review et al. ACC/AHA only Guideline on the Treatment of 17 18 Blood Cholesterol to Reduce Atherosclerotic Cardiovascular Risk in AdultsA Report 19 20 of the American College of Cardiology/American Heart Association Task Force on 21 22 Practice Guidelines. Journal of the American College of Cardiology 23 24 2014;63(25_PA):2889-934. doi: 10.1016/j.jacc.2013.11.002 25 26 27 5. Jiao F, Fung CS, Wan YF, et al. Long-term effects of the multidisciplinary risk assessment 28 29 and management program for patients with diabetes mellitus (RAMP-DM): a 30 31 population-based cohort study. Cardiovascular Diabetology 2015;14(1):105. 32 33 6. Fung CS, Chin WY, Dai DS, et al. Evaluation of the quality of care of a multi-disciplinary http://bmjopen.bmj.com/ 34 35 risk factor assessment and management programme (RAMP) for diabetic patients. 36 37 BMC Family Practice 2012;13(1):116. 38 39 40 7. Wilson PW, D’Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk 41 on September 29, 2021 by guest. Protected copyright. 42 factor categories. Circulation 1998;97(18):1837-47. 43 44 8. D'Agostino RB, Wolf PA, Belanger AJ, et al. Stroke risk profile: adjustment for 45 46 antihypertensive medication. The Framingham Study. Stroke 1994;25(1):40-43. 47 48 9. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk 49 50 prediction algorithms to estimate future risk of cardiovascular disease: prospective 51 52 53 cohort study. bmj 2017;357:j2099. 54 55 56 57 58 59 21 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 22 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 10. Hippisley-Cox J, Coupland C, Vinogradova Y, et al. Predicting cardiovascular risk in 4 5 England and Wales: prospective derivation and validation of QRISK2. Bmj 6 7 2008;336(7659):1475-82. 8 9 11. Mancia G, Fagard R, Narkiewicz K, et al. 2013 ESH/ESC Guidelines for the management 10 11 of arterial hypertension: the Task Force for the management of arterial hypertension 12 13 14 of the European Society of Hypertension (ESH) and of the European Society of 15 16 CardiologyFor (ESC). peerBlood pressure review 2013;22(4):193-278. only 17 18 12. Wan EYF, Fong DYT, Fung CSC, et al. Development of a cardiovascular diseases risk 19 20 prediction model and tools for Chinese patients with type 2 diabetes mellitus: A 21 22 population-based retrospective cohort study. Diabetes, Obesity and Metabolism 23 24 2017:n/a-n/a. doi: 10.1111/dom.13066 25 26 27 13. Wan EYF, Fong DYT, Fung CSC, et al. Prediction of new onset of end stage renal 28 29 disease in Chinese patients with type 2 diabetes mellitus–a population-based 30 31 retrospective cohort study. BMC nephrology 2017;18(1):257. 32 33 14. Wan EYF, Fong DYT, Fung CSC, et al. Prediction of five-year all-cause mortality in http://bmjopen.bmj.com/ 34 35 Chinese patients with type 2 diabetes mellitus–A population-based retrospective 36 37 cohort study. Journal of Diabetes and its Complications 2017;31(6):939-44. 38 39 40 15. Diabetes Drafting Group. Prevalence of small vessel and large vessel disease in diabetic 41 on September 29, 2021 by guest. Protected copyright. 42 patients from 14centres: the world health organization multinational study of vascular 43 44 disease in diabetics. Diabetologia 1985;28:615-40. 45 46 16. Chi Z, Lee E, Lu M, et al. Vascular disease prevalence in diabetic patients in China: 47 48 standardised comparison with the 14 centres in the WHO Multinational Study of 49 50 Vascular Disease in Diabetes. Diabetologia 2001;44(2):S82-S86. 51 52 53 54 55 56 57 58 59 22 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 23 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 17. Liu J, Lim S, Yeoh L, et al. Ethnic disparities in risk of cardiovascular disease, endTstage 4 5 renal disease and allTcause mortality: a prospective study among Asian people with 6 7 Type 2 diabetes. Diabetic Medicine 2016;33(3):332-39. 8 9 18. Burrows NR, Li Y, Williams DE. Racial and ethnic differences in trends of end-stage 10 11 renal disease: United States, 1995 to 2005. Advances in chronic kidney disease 12 13 14 2008;15(2):147-52. 15 16 19. McClellan WM,For Warnock peer DG, Judd reviewS, et al. Albuminuria only and racial disparities in the risk 17 18 for ESRD. Journal of the American Society of Nephrology 2011;22(9):1721-28. 19 20 20. Kong AP, Xu G, Brown N, et al. Diabetes and its comorbidities—where East meets West. 21 22 Nature Reviews Endocrinology 2013;9(9):537-47. 23 24 21. Forouhi NG, Sattar N. CVD risk factors and ethnicity—a homogeneous relationship? 25 26 27 Atherosclerosis Supplements 2006;7(1):11-19. 28 29 22. Byrne CD, Wild SH. The metabolic syndrome: John Wiley & Sons 2011. 30 31 23. Zhao W, Chen J. Implications from and for food cultures for cardiovascular disease: diet, 32 33 nutrition and cardiovascular diseases in China. Asia Pacific journal of clinical http://bmjopen.bmj.com/ 34 35 nutrition 2001;10(2):146-52. 36 37 24. Steyerberg EW, Eijkemans MJ, Van Houwelingen JC, et al. Prognostic models based on 38 39 40 literature and individual patient data in logistic regression analysis. Statistics in 41 on September 29, 2021 by guest. Protected copyright. 42 medicine 2000;19(2):141-60. [published Online First: 2000/01/21] 43 44 25. Stevens RJ, Kothari V, Adler AI, et al. The UKPDS risk engine: a model for the risk of 45 46 coronary heart disease in Type II diabetes (UKPDS 56). Clinical Science 47 48 2001;101(6):671-79. 49 50 26. Kothari V, Stevens RJ, Adler AI, et al. UKPDS 60 risk of stroke in type 2 diabetes 51 52 53 estimated by the UK Prospective Diabetes Study risk engine. Stroke 2002;33(7):1776- 54 55 81. 56 57 58 59 23 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 24 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 27. Swets JA. Measuring the accuracy of diagnostic systems. Science 1988;240(4857):1285- 4 5 93. 6 7 28. Mozaffarian D, Benjamin EJ, Go AS, et al. Heart Disease and Stroke Statistics—2015 8 9 Update A Report From the American Heart Association. Circulation 10 11 2015;131(4):e29-e322. 12 13 14 29. Kelly BB, Fuster V. Promoting Cardiovascular Health in the Developing World: A 15 16 Critical ChallengeFor topeer Achieve Global review Health: National only Academies Press 2010. 17 18 30. Prospective Studies Collaboration. Blood cholesterol and vascular mortality by age, sex, 19 20 and blood pressure: a meta-analysis of individual data from 61 prospective studies 21 22 with 55 000 vascular deaths. The Lancet 2007;370(9602):1829-39. 23 24 31. Wan EYF, Fung CSC, Wong CKH, et al. Association of Hemoglobin A1c Levels With 25 26 27 Cardiovascular Disease and Mortality in Chinese Patients With Diabetes. Journal of 28 29 the American College of Cardiology 2016;67(4):456-58. 30 31 32. Wan EYF, Fung CSC, Fong DYT, et al. A curvilinear association of body mass index 32 33 with cardiovascular diseases in Chinese patients with type 2 diabetes mellitus – A http://bmjopen.bmj.com/ 34 35 population-based retrospective cohort study. Journal of Diabetes and its 36 37 Complications 2016;30(7):1261-68. doi: 38 39 40 https://doi.org/10.1016/j.jdiacomp.2016.05.010 41 on September 29, 2021 by guest. Protected copyright. 42 33. Wan EYF, Yu EYT, Fung CSC, et al. Do We Need a Patient-Centered Target for Systolic 43 44 Blood Pressure in Hypertensive Patients With Type 2 Diabetes Mellitus? 45 46 Hypertension 2017:HYPERTENSIONAHA. 117.10034. 47 48 34. Fox CS, Matsushita K, Woodward M, et al. Associations of kidney disease measures with 49 50 mortality and end-stage renal disease in individuals with and without diabetes: a meta- 51 52 53 analysis. The Lancet 2012;380(9854):1662-73. 54 55 56 57 58 59 24 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 25 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 35. Matsushita K, Coresh J, Sang Y, et al. Estimated glomerular filtration rate and 4 5 albuminuria for prediction of cardiovascular outcomes: a collaborative meta-analysis 6 7 of individual participant data. The lancet Diabetes & endocrinology 2015;3(7):514-25. 8 9 36. Wan EYF, Fung CSC, Yu EYT, et al. Association of Visit-to-Visit Variability of Systolic 10 11 Blood Pressure With Cardiovascular Disease and Mortality in Primary Care Chinese 12 13 14 Patients With Type 2 Diabetes Mellitus—A Retrospective Population-Based Cohort 15 16 Study. DiabetesFor care peer 2016:dc161617. review only 17 18 37. Wan EYF, Fung CSC, Fong DYT, et al. Association of variability in hemoglobin A1c 19 20 with cardiovascular diseases and mortality in Chinese patients with type 2 diabetes 21 22 mellitus—A retrospective population-based cohort study. Journal of Diabetes and its 23 24 Complications 2016;30(7):1240-47. 25 26 27 38. Tangri N, Stevens LA, Griffith J, et al. A predictive model for progression of chronic 28 29 kidney disease to kidney failure. Jama 2011;305(15):1553-59. 30 31 39. Wong MC, Jiang JY, Tang J-l, et al. Health services research in the public healthcare 32 33 system in Hong Kong: an analysis of over 1 million antihypertensive prescriptions http://bmjopen.bmj.com/ 34 35 between 2004–2007 as an example of the potential and pitfalls of using routinely 36 37 collected electronic patient data. BMC health services research 2008;8(1):138. 38 39 40 40. Fung V, Cheung N, Szeto K, et al. Hospital Authority Clinical Vocabulary Table: the Past, 41 on September 29, 2021 by guest. Protected copyright. 42 the Present, and the Future. Hospital Authority Clinical Vocabulary Table: the Past, 43 44 the Present, and the Future/AHIMA, American Health Information Management 45 46 Association 2004 47 48 41. Royston P. Multiple imputation of missing values. Stata Journal 2004;4:227-41. 49 50 42. Wan EYF, Fung CSC, Yu EYT, et al. Effect of Multifactorial Treatment Targets and 51 52 53 Relative Importance of Hemoglobin A1c, Blood Pressure, and LowTDensity 54 55 LipoproteinTCholesterol on Cardiovascular Diseases in Chinese Primary Care 56 57 58 59 25 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 26 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Patients With Type 2 Diabetes Mellitus: A PopulationTBased Retrospective Cohort 4 5 Study. Journal of the American Heart Association 2017;6(8):e006400. 6 7 43. Wan EYF, Fong DYT, Fung CSC, et al. Incidence and predictors for cardiovascular 8 9 disease in Chinese patients with type 2 diabetes mellitus–a population-based 10 11 retrospective cohort study. Journal of Diabetes and its Complications 12 13 14 2016;30(3):444-50. 15 16 For peer review only 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 http://bmjopen.bmj.com/ 34 35 36 37 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 26 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 27 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Operation definition of risk factors 4 5 6 The draft of operation definition of the risk factors was summarized as below. The 7 adjustments may be made subject to the clinical or professional advices. 8 9 10 11 Risk factor Unit Operation Definition 12 13 Clinical parameters 14 All BMI readings from HA information systems that are 15 16 available between 1 Jan 2006 and 31 Dec 2017 should be Body Mass IndexFor peer2 review only 17 kg/m provided. Baseline reading will use the latest available (BMI) 18 reading before and 12 months after 1 Jan 2008 19 20 All HbA1c readings from HA information systems that 21 are available between 1 Jan 2006 and 31 Dec 2017 Hemoglobin A1c 22 % should be provided. Baseline reading will use the latest 23 (HbA1c) 24 available reading before and 12 months after 1 Jan 2008 25 26 All systolic BP readings from HA information systems 27 that are available between 1 Jan 2006 and 31 Dec 2017 28 Systolic Blood mmHg should be provided. Baseline reading will use the latest 29 Pressure (BP) 30 available reading before and 12 months after 1 Jan 2008 31 32

All diastolic BP readings from HA information systems http://bmjopen.bmj.com/ 33

34 that are available between 1 Jan 2006 and 31 Dec 2017 35 Diastolic BP mmHg should be provided. Baseline reading will use the latest 36 available reading before and 12 months after 1 Jan 2008 37 38 39 All TC readings from HA information systems that are 40 available between 1 Jan 2006 and 31 Dec 2017 should be

Total cholesterol on September 29, 2021 by guest. Protected copyright. 41 mmol/L provided. Baseline reading will use the latest available (TC) 42 reading before and 12 months after 1 Jan 2008 43 44 45 All TC readings from HA information systems that are 46 Highdensity available between 1 Jan 2006 and 31 Dec 2017 should be 47 lipoprotein mmol/L provided. Baseline reading will use the latest available 48 49 cholesterol reading before and 12 months after 1 Jan 2008 50 51 52 53 54 55 56 57 58 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 28 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Risk factor Unit Operation Definition 4 5 All TC readings from HA information systems that are 6 Lowdensity available between 1 Jan 2006 and 31 Dec 2017 should be mmol/L 7 lipoprotein provided. Baseline reading will use the latest available 8 cholesterol reading before and 12 months after 1 Jan 2008 9 10 11 All TC readings from HA information systems that are 12 available between 1 Jan 2006 and 31 Dec 2017 should be 13 triglyceride mmol/L provided. Baseline reading will use the latest available 14 15 reading before and 12 months after 1 Jan 2008 16 For peer review only 17 If the serum creatinine is known, then the eGFR value 18 can be calculated using a formula1. All eGFR and serum 19 Estimated 15mL/min/ creatinine readings from HA information systems that 20 glomerular 2 1.73m are available between 1 Jan 2006 and 31 Dec 2017 21 filtration rate 22 should be provided. Baseline reading will use the latest (eGFR) 23 available reading before and 12 months after 1 Jan 2008 24 25 26 Patients with microalbuminuria (albuminuria), referring 27 to men with ACR ≥ 2.5 and ≤ 25mg/mmol 28 (ACR>25mg/mmol) and women with ≥ 3.5 and ≤ 29 35/mmol (ACR>35mg/mmol) Also there should be two 30 Albuminuria NA 31 records per patient. The first record should be within 6 32 months on or before baseline. The second record should 33 be within 6 months after baseline. http://bmjopen.bmj.com/ 34 35 Both the first and second records should be the 36 records closest to 1 Jan 2008. 37 Disease characteristics 38 39 Duration of The duration of diabetes will use the selfreported 40 Year

diabetes record in CMS RAMPDM module. on September 29, 2021 by guest. Protected copyright. 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 29 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Risk factor Unit Operation Definition 4 5 Comorbidity includes Chronic lung disease and 6 Cancer. 7 Chronic lung disease is defined by any of ICPC2 8 R79, R95 or ICD9CM 491.x, 492.x, 493.2x ,496 9 Cancer is defined by any of ICPC2 A79, B72B74, 10 D74D77, F74, H75, K72, L71, N74, R84R85, S77, 11 Not 12 Comorbidity T71, T73, U75U77, W72, X75X77, Y78 or ICD9 applicable 13 CM 140.x149.x , 150.x159.x,160.x165.x,170.x 14 176.x, 179.x189.x, 190.x199.x,200.x209.x 15 16 For peer review only 17 The comorbidity will be defined if the patient record 18 contains a diagnosis of any of the above listed 19 comorbidities on or before 1 Jan 2008. 20 21 Treatment modalities 22 Antihypertensive drugs are identified by the 23 following drug item codes:HYDR05, HYDR30, 24 25 INDA01, INDA02, AMIL01, SPIR01, DYAZ01, 26 MODU01, ATEN01, ATEN02, BISO01, BISO02, 27 CARV01, CARV02, CARV03, CARV04, LODO02, 28 LODO03, METO06, METO07, METO10, METO11, 29 30 METO13, METO16, METO17, NADO01, PIND01, 31 PROP04, PROP05, HYDR01, HYDR02, METH22, 32 METH23, PRAZ03, PRAZ04, CAPT01, CAPT02, http://bmjopen.bmj.com/ 33 CAPT03, CAPT06, ENAL01, ENAL02, ENAL03, 34 35 LISI01, LISI02, LISI03, PERI17, PERI28, PERI29, 36 Use of anti RAMI01, RAMI02, RAMI03, CAND01, CAND02, CO Not 37 hypertensive D01, COD02, EXFO01, EXFO02, EXFO03, IRBE01, applicable 38 drugs IRBE02, IRBE03, IRBE04, LOSA01, 39 40 LOSA02, LOSA03, LOSA04, MICA01, S00031, 41 S00069, S00454, S00455, TELM01, TELM02, on September 29, 2021 by guest. Protected copyright. 42 VALS02, VALS03, AMLO01, AMLO02, DILT01, 43 DILT02, DILT05, DILT06, DILT07, DILT08, FELO01, 44

45 FELO02, FELO03, LACI01, LACI02, NIFE03, 46 NIFE04, NIFE05, NIMO01, NIMO02 47

48 The use of antihypertensive drugs will be defined if 49

50 the record of the antihypertensive drug prescription 51 for the patient within 6 months or before 1 Jan 2008. 52 53 54 55 56 57 58 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 30 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Risk factor Unit Operation Definition 4 5 Insulin is identified by British National Formulary 6.1.

6 Not 7 Use of Insulin The use of insulin will be defined if the record of the 8 applicable insulin prescription for the patient within 6 months or 9 before 1 Jan 2008. 10 11 Oral antidiabetic drugs are identified by British National 12 Formulary 6.1.2.1, 6.1.2.2 and 6.1.2.3. 13 Use of oral anti Not 14 diabetic drugs applicable The use of oral antidiabetic drugs will be defined if the 15 16 For peerrecord review of the oral antidiabetic only drug prescription for the 17 patient within 6 months or before 1 Jan 2008. 18 lipidlowering agents includes Atorvastatin, 19 Fluvastatin, Gemfibrozil, Pravastatin, Rosuvastatin and 20 Simvastatin and are identified by the following drug 21 22 item codes ATOR01, ATOR02, ATOR03, ATOR04, 23 Use of lipid Not FLUV02, FLUV03, FLUV05, GEM01, GEM02, 24 lowering agents applicable PRAV01, PRAV02, ROSU01, ROSU02, ROSU03, 25 SIMV01, SIMV02, SIMV04 and SIMV05 26 27 The use of lipidlowering agents will be defined if the 28 record of the lipidlowering agents prescription for the 29 patient within 6 months or before 1 Jan 2008. 30 31 32 33 Reference http://bmjopen.bmj.com/ 34 35 1. Levey AS, Bosch JP, Lewis JB, et al. A more accurate method to estimate glomerular 36 filtration rate from serum creatinine: a new prediction equation. Annals of internal medicine 37 1999;130(6):46170. 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

Ten-year Risk Prediction Models of Complications and Mortality of Diabetes Mellitus in Chinese Patients in Primary Care in Hong Kong: A study protocol

ForJournal: peerBMJ Open review only Manuscript ID bmjopen-2018-023070.R1

Article Type: Protocol

Date Submitted by the Author: 19-Jun-2018

Primary Subject Diabetes and endocrinology Heading: on September 29, 2021 by guest. Protected copyright. Secondary Subject Heading: Diabetes and endocrinology, Public health

Diabetes Mellitus, Risk, Mortality, Cardiovascular Disease, Chinese, Keywords: Prediction

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 1 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 Ten-year Risk Prediction Models of Complications and Mortality of Diabetes Mellitus 4 5 in Chinese Patients in Primary Care in Hong Kong: A study protocol 6 7 8 9 Authors: Eric Yuk Fai Wan, PhD1, Esther Yee Tak Yu, MBBS1, Weng Yee Chin, MD1, 10 11 Colman Siu Cheung Fung, MBBS1, Ruby Lai Ping Kwok, MHSE2, David Vai Kiong Chao, 12 13 3 4 5 14 MBBS , King Hong Chan, MBBS , Eric Ming-Tung Hui, MBBS , Wendy Wing Sze Tsui, 15 6 7 8 16 MBBS , KathrynFor Choon Bengpeer Tan, MD review, Daniel Yee Tak Fong,only PhD , Cindy Lo Kuen Lam, 17 18 MD1 19 20 21 22 23 1 24 Department of Family Medicine and Primary Care, The University of Hong Kong, Hong 25 26 Kong 27 28 2 29 Department of Primary & Community Services, Hospital Authority Head Office, Hospital 30 31 Authority, Hong Kong 32 33 http://bmjopen.bmj.com/ 3 Department of Family Medicine & Primary Healthcare, Kowloon East Cluster, Hospital 34 35 36 Authority, Hong Kong 37 38 4 Department of Family Medicine & Primary Healthcare, Kowloon Central Cluster, Hospital 39 40 41 Authority, Hong Kong on September 29, 2021 by guest. Protected copyright. 42 43 5 Department of Family Medicine, New Territories East Cluster, Hospital Authority, Hong 44 45 46 Kong 47 48 6 Department of Family Medicine & Primary Healthcare, Hong Kong West Cluster, Hospital 49 50 51 Authority, Hong Kong 52 53 7 Department of Medicine, the University of Hong Kong, Hong Kong 54 55 56 8 School of Nursing, the University of Hong Kong, Hong Kong 57 58 59 1 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 2 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 17 3 Asian Indian, and Chinese populations. In addition, several multiethnic studies showed 4 5 substantial difference in the incidence rate among different racial groups with a generally 6 7 higher incidence rate of renal disease in Chinese than in non-Chinese DM patients.18-20 The 8 9 IDF report revealed that 1.3 million Chinese died because of DM in 2015, which represented 10 11 the highest prevalence of DM-related mortality across various ethnic groups and was two- 12 13 1 14 fold higher than those found in Europeans and Australians. These discrepancies in CVD, 15 16 ESRD, and all-causeFor mortality peer risk may reviewbe related to the differences only in the disease profile and 17 18 other determinants such as genetics, health care policy, and culture.15 16 21-23 Therefore, 19 20 Chinese population specific risk prediction models are necessary. 21 22 23 24 Inaccurate risk stratification may lead to inappropriate risk-stratified interventions. There is a 25 26 need for new robust risk prediction models, and thus the aim of this study protocol to develop 27 28 29 the models to predict 10-year CVD risk and mortality for primary care Chinese patients in 30 31 order to enable accurate risk stratification of DM patients in the HA on-going RAMP-DM or 32 33 other primary care systematic risk-stratified multi-disciplinary management programmes. http://bmjopen.bmj.com/ 34 35 Furthermore, robust risk prediction models for the overall prediction of first CVD and all- 36 37 cause mortality can inform policy makers in service planning and resource allocation. 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 Aims and Objectives 43 44 This study protocol aims to develop 10-year risk prediction models for total CVD and all- 45 46 cause mortality among Chinese diabetic patients in primary care. Risk prediction models for 47 48 individual DM complications including CHD, heart failure, stroke and ESRD will also be 49 50 developed. 51 52

1 2 3 January 2008 identified from the HA clinical management system (CMS) database. 4 5 6 7 The inclusion criteria are patients aged 18 years old or older, had at least 1 GOPC/ Family 8 9 Medicine Clinics (FMC) attendance on or within 1 year before 1 January 2008 and had a 10 11 CMS record of the coding of International Classification of Primary Care, Second edition 12 13 14 (ICPC-2) of T89 (Diabetes insulin dependent) or T90 (Diabetes non-insulin dependent) on or 15 16 before 1 JanuaryFor 2008. peer review only 17 18 19 20 The exclusion criteria are patients who had a diagnosis of any DM complications defined by 21 22 the relevant ICPC-2 or The International Classification of Diseases, Ninth Revision, Clinical 23 24 Modification (ICD-9-CM) (shown in the section below) on or before 1 January 2008 and 25 26 patients exclusively managed by Specialist Out-Patient Clinic (SOPC) on or before 1 January 27 28 29 2008. 30 31 32 33 Sample Size Calculation http://bmjopen.bmj.com/ 34 35 The required sample size is based on the requirements for the development and validation of 36 37 the least common DM complication of ESRD. Specifically, based on our previous study, the 38 39 5-year incidence of ESRD was 1.9%,6 which can be extrapolated to a 10-year incidence of 40 41 on September 29, 2021 by guest. Protected copyright. 42 ESRD was 3.8% by assuming a constant incidence rate over time. To develop the risk 43 44 prediction model for ESRD by multivariable Cox proportional hazard regressions with 45 46 forward stepwise variables selection on 16 potential risks factors, we need 21,053 subjects 47 48 using the 1 in 50 rule that 1 candidate predictor can be studied for every 50 events.24 The split 49 50 samples on a 2:1 basis will be applied, and thus 10,527 subjects are needed to validate the 51 52 risk prediction models. Overall, a total 31,580 male and 31,580 female are needed for the 53 54 55 development (training dataset) and validation of risk prediction models stratified by gender. 56 57 58 59 9 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 10 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 4 5 Definition of Indicator DM Complications 6 7 The incidence of four major DM complications (CHD, stroke, heart failure and ESRD), total 8 9 CVD and all-cause mortality will be calculated. The incidence is counted from the earliest 10 11 date of documented diagnosis defined by the relevant ICPC-2 and/or ICD-9-CM coding 12 13 14 recorded in the HA CMS database from 1 January 2008 to 31 December 2017. The relevant 15 16 ICPC-2 and ICD-9-CMFor codes peer of each DMreview complication and only mortality are determined by the 17 18 academic and HA clinician co-investigators as listed below: 19 20 21 22 1. CHD (ischaemic heart disease, myocardial infarction (MI), coronary death or sudden 23 24 death) is defined by any of ICPC-2 K74 to K76 and ICD-9-CM 410.x, 411.x to 414.x, 25 26 27 798.x 28 29 2. Stoke (fatal and non-fatal stroke) is defined by any of ICPC-2 K89 to K91 or ICD-9-CM 30 31 430.x to 438.x. 32 33 3. Heart failure is defined by any of ICPC-2 K77 or ICD-9-CM 428.x http://bmjopen.bmj.com/ 34 35 4. CVD is defined as the presence of any of CHD, heart failure and stroke ICPC-2 or ICD-9- 36 37 CM codes listed in 1, 2 and 3 above. 38 39 40 5. ESRD is defined by any of ICD-9-CM 250.3x, 585.x, 586.x, or an estimated glomerular 41 on September 29, 2021 by guest. Protected copyright. 2 42 filtration rate (eGFR) < 15mL/min/1.73m . 43 44 6. Mortality is identified from the Hong Kong Death Registry. 45 46 47 48 Risk Factors to be Included in the Risk Prediction Models 49 50 Risk factors (independent variables) previously found to be associated with DM 51 52 2 53 complications from the literature and those that are routinely available in primary care are 54 55 selected to strike a balance between comprehensiveness and feasibility. The potential risk 56 57 58 59 10 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 11 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 factors that will be explored include those related to patient’s socio-demographics, clinical 4 5 parameters, disease characteristics and treatment modalities. Patient socio-demographics 6 7 include sex, age and smoking status. Clinical parameters include body mass index (BMI), 8 9 HbA1c, systolic and diastolic blood pressure (SBP and DBP), lipid profile (total cholesterol, 10 11 high-density lipoprotein cholesterol, LDL-C, triglyceride), eGFR and albuminuria. Disease 12 13 14 characteristics include the duration of DM and co-morbidity. Treatment modalities include 15 16 the use of specificFor anti-hypertensive peer drugs, review insulin, specific onlyoral anti-diabetic drugs and lipid- 17 18 lowering agents. The operational definitions of the risk factors are shown in Supplementary 19 20 File 1. These factors, except sex, age, duration of DM and co-morbidity, are modifiable, 21 22 which have implications for practice. 23 24

25 26 27 Data Collection 28 29 In the middle of 2018, anonymous data from 1 January 2008 to 31 December 2017 of all DM 30 31 patients who satisfy the inclusion criteria and without any exclusion criteria will be extracted 32 33 by the HA statistics team from the HA CMS database. We have successful experience in http://bmjopen.bmj.com/ 34 35 working with the HA in the extraction of similar data from 2009 to 2013 for our extended 36 37 evaluation on quality of care and effectiveness of RAMP-DM study,5 and we have obtained 38 39 40 preliminary agreement from the HA for the data extraction in the present study. 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 Outcome Measures 45 46 1. The incidence of total CVD, all-cause mortality and each of 4 major DM complications 47 48 over 10 years 49 50 2. Factors predictive of total CVD, all-cause mortality and each of 4 major DM 51 52 53 complications over 10 years 54 55 3. Ten-year risk prediction models for total CVD, all-cause mortality and each of 4 major 56 57 58 59 11 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 12 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 DM complications 4 5 4. Factors that have sufficient power to classify Chinese DM patients in primary care into 6 7 risk group in terms of total CVD and all-cause mortality 8 9 10 11 12 Data Processing and Analysis 13 14 The cohort will be stratified by gender. Descriptive statistics will be used to calculate the 15 16 incidence of totalFor CVD, all-causepeer mortality review and each of 4 onlymajor DM complications will be 17 18 analysed annually and cumulatively over 10 years with a 95% confidence interval. The 19 20 distribution of risk factors will be cross-tabulated by complication or mortality events. The 21 22 10-year cumulative incidence of various DM complications and mortalities will be further 23 24 analysed by Kaplan-Meier method. The Kaplan-Meier survival curve will be used to describe 25 26 27 the survivorship of total CVD, all-cause mortality and each of 4 major DM complications in 28 29 the study cohort over 10 years. Unadjusted associations between the risk factors and odds of 30 31 events will be assessed by independent t-test for continuous variables or chi-square test for 32 33 categorical variables. http://bmjopen.bmj.com/ 34 35 36 37 The cohort will be randomly split on a 2:1 basis, with two-third sample used for developing 38 39 40 the risk prediction models, and the other one-third sample used for validation of the risk 41 on September 29, 2021 by guest. Protected copyright. 42 prediction models. The analyses will be carried out separately for men and women. 43 44 45 46 Development of Risk Prediction Models 47 48 Cox proportional hazard regressions with forward stepwise method will be used to develop 49 50 the risk prediction models for total CVD, all-cause mortality and each of 4 major DM 51 52 53 complications. If the main term of a clinical parameter is selected in the models, the quadratic 54 55 term of such clinical parameter will be evaluated. Afterwards, the interaction terms between 56 57 58 59 12 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 13 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 selected predictors and age will be also examined in the risk prediction models. Cox 4 5 regression is the method most commonly used in risk prediction models in the Framingham 6 7 Heart Study 7 8 and UKPDS 25 26. It allows us to estimate the risk of disease or death for an 8 9 individual, given their prognostic variables. A positive hazard ratio means a higher likelihood 10 11 of event associated with that specific variable. Conversely, a negative hazard ratio means a 12 13 14 lower likelihood of the event associated with that specific variable. The key proportional 15 16 hazards assumptionFor will bepeer assessed by review examining plots of only the scaled Schoenfeld residuals 17 18 against time for the covariates. Any non-random pattern indicates a violation of the 19 20 proportional hazards assumptions in which case transformation of covariates may be 21 22 necessary. For example, all continuous variables were naturally logarithmically transformed 23 24 to minimize the influence of extreme values and to improve discrimination and calibration of 25 26 27 the models. A parametric approach such as exponential or Weibull distribution for the hazard 28 29 function can also be carried out. A total of 6 risk prediction models will be established for 30 31 total CVD, all-cause mortality and each of 4 major DM complications. The log of the hazard 32 33 ratio of each selected risk factor in the final model will be used as coefficient weights in the http://bmjopen.bmj.com/ 34 35 prediction model of each relevant outcome. The risk equations for 10 year’s follow-up will be 36 37 established by combining these weights with the survivor function.7 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 Validation of Risk Prediction Models 43 44 To validate the risk prediction models for total CVD, all-cause mortality and each of 4 major 45 46 DM complications, the remaining one-third validation sample will be used to estimate the 47 48 risk level of the subjects. The performance of discrimination of the model will be assessed by 49 50 Harrell's C statistic, which is a measure similar to the area under the curve after consideration 51 52 53 of the censoring pattern of the patients. A Harrell's C statistic less than 0.7 indicates limited 54 55 discriminating power, 0.7 to 0.9 is acceptable, and higher than 0.9 suggests strong 56 57 58 59 13 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 14 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 27 2 3 discrimination of the predictive models. The D statistic, R statistic and Brier score will be 4 5 also calculated for evaluating the predictive power of the model. The D statistic is a measure 6 7 of discrimination where higher value indicates better discrimination. The R2 statistic is a 8 9 measure of explained variation with a higher value indicating better performance. The Brier 10 11 score is a measure of goodness of fit in which a lower value means higher accuracy. 12 13 14 15 16 Calibration willFor be used peer to measure howreview closely predicted only outcomes agree with actual 17 18 outcomes. Calibration of the model’s ability to correctly estimate the absolute risks will be 19 20 examined by modified Hosmer-Lemeshow test and calibration plots. The modified Hosmer- 21 22 Lemeshow test for time-to-event data measures how well the predicted probability of the 23 24 expected event rate agrees with the observed event rate, where a p-value higher than 0.05 25 26 27 indicates good model calibration. In a calibration plot of the observed incidence of events 28 29 against the predicted risk shows the scatter along the 45° line of perfect fit between predicted 30 31 risk and observed incidence of event throughout the entire risk spectrum. 32 33 http://bmjopen.bmj.com/ 34 35 Development of a Risk Prediction Nomogram and Chart 36 37 In order to enable the 10-year risk prediction models for total CVD and all-cause mortality to 38 39 40 be applied in busy clinical setting, risk prediction nomograms and charts will be developed 41 on September 29, 2021 by guest. Protected copyright. 42 for men and women. For the nomogram, the patient’s score for each predictor is plotted on 43 44 the appropriate scale and vertical lines are drawn to the line of points to obtain the 45 46 corresponding scores. The score of each predictor will be transformed based on the estimated 47 48 standardized beta coefficient of each predictor from the risk prediction model. For continuous 49 50 predictor such as age, the line with interval depends on its units from minimum to maximum 51 52 53 values among studies subjects (e.g. 20, 40, 60, 80 years old) will be displayed on the 54 55 nomogram plot and the corresponding scores will be obtained based on the estimated 56 57 58 59 14 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 15 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 standardized beta coefficient of the predictor (e.g. assign age of 20 to 0 point, age of 40 to 2 4 5 points, age of 60 to 4 points and age of 80 to 6 points). For categorical predictor such as 6 7 gender, each level of the predictor will be ranked a corresponding score based on the 8 9 estimated standardized beta coefficient of the predictor (e.g. assign female to 0 point and 10 11 male to 3 points). All scores are summed to obtain a total score. The total score is plotted on 12 13 14 the total line with corresponding predicted risk of CVD. Moreover, we will develop risk 15 16 prediction chartsFor similar to peer those developed review by the Joint British only Society. The most significant 17 18 predictors, up to a maximum of five, found in the full Cox regression models will be selected 19 20 to classify subjects into 10-year CVD risk groups of < 10% (low risk), 10%-20% (medium 21 22 risk) and > 20% (high risk). The Kaplan-Meier survival curves of each risk group will be 23 24 developed and compared by log-rank tests to confirm the hazard ratios are significantly 25 26 27 different among all risk groups. 28 29 30 31 STATA software version 13 (STATA Corp, College Station, Texas) will be used for data 32 33 analyses. 5% is used as the level of significance in all statistical tests. http://bmjopen.bmj.com/ 34 35 36 37 Discussion 38 39 Primary care is the entry point of the entire medical system. Therefore, primary care doctors 40 41 on September 29, 2021 by guest. Protected copyright. 42 need to act as gatekeepers for medical resources. Given the large number and substantial 43 44 heterogeneity of DM patients, the aim of study was to develop several risk prediction models 45 46 to facilitate primary care providers in identifying Chinese patients at higher risk of 47 48 developing diabetic complications. This will allow interventions to be implemented to reduce 49 50 the individual, societal, and economic burden. The nomogram and chart can help inform 51 52 clinicians regarding interventions based on overall risk of diabetic complications instead of 53 54 55 only a single risk factor; in addition, these tools can also be used to educate, motivate, and 56 57 58 59 15 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 16 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 empower patients to prevent future diabetic complications. In terms of policy implications, 4 5 misclassification may likely lead to excessive medical treatment, low cost-effectiveness in 6 7 primary prevention, and potentially unnecessary exposure to the risk of adverse drug effects. 8 9 The accurate risk prediction model is particularly useful for screening programs to inform 10 11 decisions concerning service provision for DM primary care to achieve the maximum 12 13 14 population health benefit. 15 16 For peer review only 17 18 We have also identified some potential improvements in the performance of existing risk 19 20 prediction models. Firstly, gender is a factor that is of concern in the analyses of risk factors 21 22 and CVD/mortality because males are typically associated with a higher risk of 23 24 CVD/mortality,28 but statistical adjustment for gender is often insufficient to control for 25 26 29 27 varying risk-factor profiles and CVD/mortality incidence. Secondly, there are possible 28 29 interaction effects between age and risk factors on the CVD/mortality as the magnitude of the 30 31 effect of specific risk factors such as LDL-C on the CVD/mortality may decrease with age.30 32 33 Thus, the interaction term between age and risk factors should be considered when http://bmjopen.bmj.com/ 34 35 developing the risk prediction models. Thirdly, many studies including our previous studies 36 37 illustrated that there were curvilinear association (J or U shape) between HbA1c/SBP/BMI 38 39 31-33 40 and the risk of CVD/mortality, and thus the quadratic term of such clinical parameters 41 on September 29, 2021 by guest. Protected copyright. 42 should be evaluated when developing the risk prediction models. Fourthly, recent research 43 44 suggested additional clinical parameters such as severity of renal impairment measured by 45 46 eGFR and albumin/creatinine ratio (ACR), and variability of risk factors including HbA1c 47 48 and SBP should be considered to enhance the performance of prediction models.34-37 Finally, 49 50 the feasibility of measuring predictors in a routine primary care setting should be considered 51 52 53 when establishing risk prediction models. For example, some predictors such as serum 54 55 bicarbonate and phosphate included in an ESRD risk prediction model may not be available 56 57 58 59 16 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 17 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 38 3 in routine clinical practice, especially in the primary care setting. Accordingly, the rationale 4 5 for inclusion of such non-standard parameters for risk prediction is often unclear, despite the 6 7 inclusion of such predictors improved the model prediction accuracy. However, to develop a 8 9 model with wide applicability in the real world, predictors should be selected to strike a 10 11 balance between comprehensiveness and feasibility. 12 13 14 15 16 From our previousFor experience, peer there werereview several potential only difficulties we may encounter 17 18 during the development of risk prediction model. Firstly, as anonymous data are extracted 19 20 from the HA CMS database, there may be some variations of the operation definitions 21 22 throughout this study. To avoid confusion and unnecessary misunderstanding, operational 23 24 definitions (Supplementary File 1) clearly list the definitions of the clinical parameters, 25 26 27 disease characteristics and treatment modalities according to clinical and professional 28 29 advices. Moreover, it is important to ensure the quality of the data from data collection and 30 31 correct use of data. Regular communications between our team and HA will be held to cross- 32 33 validate the method of collecting raw data and the use of the data. Secondly, the outcome http://bmjopen.bmj.com/ 34 35 events in this study will rely on the diagnosis coding such as ICPC-2 and ICD-9CM, which 36 37 may cause misclassification bias. Due to the privacy policy, all extracted data will be 38 39 40 anonymous, and thus the cross-checking the records in the dataset for auditing purposes will 41 on September 29, 2021 by guest. Protected copyright. 42 not be feasible. However, in routine clinical practice in Hong Kong, clinicians in the clinical 43 44 and hospital settings provide ICPC-2 and ICD-9-CM codes, respectively, for each episode of 45 46 attendance.39 40 Moreover, due to the heavily subsidized health care system in Hong Kong, 47 48 patients with chronic diseases and serious complication, e.g. myocardial infarction, mostly 49 50 treated in the HA public health care system. Therefore, our data from HA database should 51 52 53 have captured nearly all CVD and ESRD outcomes of DM patients who are managed in the 54 55 HA primary care setting. Lastly, the missing data possibly exist in the dataset extracted from 56 57 58 59 17 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 18 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 the database. To preserve the completeness and power of the data, multiple imputation will be 4 5 used to deal with the missing data. Comparing with complete-case analysis and single 6 7 imputation, multiple imputation takes incomplete data into account to avoid unnecessary 8 9 biases and producing more reliable and applicable risk prediction models.41 Thus, multiple 10 11 imputation has become one of the most widely-used methods for handling missing data in 12 13 42 43 14 medical research. This approach was also used in the development of the CVD prediction 15 9 10 16 model in the QRiskFor study. peer review only 17 18 19 20 Conclusion 21 22 There is a need for the development of 10-year risk prediction models based on the up-to-date 23 24 population-based cohort for diabetic complications, particularly first total CVD. Chinese 25 26 27 patients in primary care will enable accurate risk stratification, better prioritization of 28 29 resources and more cost-effective interventions for diabetic patients in primary care. They 30 31 can also better inform and empower patients to prevent potential DM complications. At the 32 33 health policy level, the results can inform decisions on service provision in the care of http://bmjopen.bmj.com/ 34 35 diabetes mellitus in primary care to achieve maximum population health benefit. The 36 37 prediction models can also be used as an outcome measure on the potential benefits of 38 39 40 complication prevention in clinical trials on DM interventions in primary care. 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 Ethics and dissemination 45 46 Ethics Approval 47 48 Ethics approval of this study was granted by the Institutional Review Board of the University 49 50 of Hong Kong/ Hospital Authority Hong Kong West (UW 15-258). 51 52 53 54 55 Patient involvement 56 57 58 59 18 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 19 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 No patients were involved in setting the research question or the outcomes measures, designing the 4 5 investigation or interpreting the data. There are no plans to involve patients in dissemination of the 6 7 results. 8 9 10 11 Data sharing statement 12 13 The dataset generated and/or analysed in the current study is not publicly available as the data 14 15 is from patient records. 16 For peer review only 17 18 19 20 Author contribution 21 22 CLKL is the principal investigators of the study. CLKL and EYFW initially conceived the 23 24 study. EYFW, EYTY, WYC, CSCF, DYTF and CLKL helped with the design and 25 26 implementation of the programmes, coordination of the study, drafted and revised the 27 28 manuscript; EYFW, EYTY, WYC, CSCF, RLPK, DVKC, KHC, EMTH, WWST, KCBT, 29 30 DYTF and CLKL revised the manuscript. All authors approved the final version. 31 32 33 http://bmjopen.bmj.com/ 34 Acknowledgements 35 36 The authors are most grateful to the Food and Health Bureau, HKSAR and the Hong Kong 37 38 Hospital Authority, in particular chief of service in primary care in each cluster and Statistics 39 40

and Workforce Planning Department at the Hong Kong Hospital Authority. on September 29, 2021 by guest. Protected copyright. 41 42 43 44 45 Funding 46 47 This study has been funded by the Health and Medical Research Fund, Food and Health 48 49 Bureau, HKSAR (Project no: 14151181). No funding organization had any role in the design 50 51 and conduct of the study; collection, management, analysis, and interpretation of the data; 52 53 and preparation of the manuscript. All other authors have reported that they have no 54 55 56 relationships relevant to the contents of this paper to disclose. 57 58 59 19 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 20 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 4 5 Competing Interest 6 7 The authors declare that they have no competing interests. 8 9 10 11 12 13 14 15 16 For peer review only 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 http://bmjopen.bmj.com/ 34 35 36 37 38 39 40 41 on September 29, 2021 by guest. Protected copyright. 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 20 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 21 of 30 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2 3 References 4 5 1. International Diabetes Federation. IDF Diabetes Atlas, 7 ed. Brussels, 2015. 6 7 2. Chamnan P, Simmons R, Sharp S, et al. Cardiovascular risk assessment scores for people 8 9 with diabetes: a systematic review. Diabetologia 2009;52(10):2001-14. 10 11 3. American Diabetes Association. Standards of medical care in diabetes—2010. Diabetes 12 13 14 care 2010;33(Supplement 1):S11-S61. 15 16 4. Stone NJ, RobinsonFor JG, peerLichtenstein AH,review et al. ACC/AHA only Guideline on the Treatment of 17 18 Blood Cholesterol to Reduce Atherosclerotic Cardiovascular Risk in AdultsA Report 19 20 of the American College of Cardiology/American Heart Association Task Force on 21 22 Practice Guidelines. Journal of the American College of Cardiology 23 24 2014;63(25_PA):2889-934. doi: 10.1016/j.jacc.2013.11.002 25 26 27 5. Jiao F, Fung CS, Wan YF, et al. Long-term effects of the multidisciplinary risk assessment 28 29 and management program for patients with diabetes mellitus (RAMP-DM): a 30 31 population-based cohort study. Cardiovascular Diabetology 2015;14(1):105. 32 33 6. Fung CS, Chin WY, Dai DS, et al. Evaluation of the quality of care of a multi-disciplinary http://bmjopen.bmj.com/ 34 35 risk factor assessment and management programme (RAMP) for diabetic patients. 36 37 BMC Family Practice 2012;13(1):116. 38 39 40 7. Wilson PW, D’Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk 41 on September 29, 2021 by guest. Protected copyright. 42 factor categories. Circulation 1998;97(18):1837-47. 43 44 8. D'Agostino RB, Wolf PA, Belanger AJ, et al. Stroke risk profile: adjustment for 45 46 antihypertensive medication. The Framingham Study. Stroke 1994;25(1):40-43. 47 48 9. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk 49 50 prediction algorithms to estimate future risk of cardiovascular disease: prospective 51 52 53 cohort study. bmj 2017;357:j2099. 54 55 56 57 58 59 21 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 22 of 30 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from

1 2

3 Operation definition of risk factors BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from 4 5 6 The draft of operation definition of the risk factors was summarized as below. The 7 8 adjustments may be made subject to the clinical or professional advices. 9 10 11 12 Risk factor Unit Operation Definition 13 14 Clinical parameters 15 16 All BMI readings from HA information systems that are 17 available between 1 Jan 2006 and 31 Dec 2017 should be Body Mass Index 2 18 For kg/mpeer provided. review Baseline readingonly will use the latest available 19 (BMI) 20 reading before and 12 months after 1 Jan 2008 21 All HbA1c readings from HA information systems that 22 are available between 1 Jan 2006 and 31 Dec 2017 23 Hemoglobin A1c 24 % should be provided. Baseline reading will use the latest (HbA1c) 25 available reading before and 12 months after 1 Jan 2008 26 27 28 All systolic BP readings from HA information systems 29 that are available between 1 Jan 2006 and 31 Dec 2017 30 Systolic Blood mmHg should be provided. Baseline reading will use the latest 31 Pressure (BP) 32 available reading before and 12 months after 1 Jan 2008 33 34 35 All diastolic BP readings from HA information systems 36 that are available between 1 Jan 2006 and 31 Dec 2017 37 Diastolic BP mmHg should be provided. Baseline reading will use the latest http://bmjopen.bmj.com/ 38 39 available reading before and 12 months after 1 Jan 2008 40 41 All TC readings from HA information systems that are 42 available between 1 Jan 2006 and 31 Dec 2017 should be 43 Total cholesterol 44 mmol/L provided. Baseline reading will use the latest available

(TC) on September 29, 2021 by guest. Protected copyright. 45 reading before and 12 months after 1 Jan 2008 46 47 48 All TC readings from HA information systems that are 49 High-density available between 1 Jan 2006 and 31 Dec 2017 should be 50 mmol/L 51 lipoprotein- provided. Baseline reading will use the latest available 52 cholesterol reading before and 12 months after 1 Jan 2008 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 28 of 30

1 2

3 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from 4 Risk factor Unit Operation Definition 5 All TC readings from HA information systems that are 6 Low-density available between 1 Jan 2006 and 31 Dec 2017 should be 7 mmol/L 8 lipoprotein- provided. Baseline reading will use the latest available 9 cholesterol reading before and 12 months after 1 Jan 2008 10 11 12 All TC readings from HA information systems that are 13 available between 1 Jan 2006 and 31 Dec 2017 should be 14 triglyceride mmol/L provided. Baseline reading will use the latest available 15 16 reading before and 12 months after 1 Jan 2008 17 18 For peerIf reviewthe serum creatinine only is known, then the eGFR value 19 can be calculated using a formula1. All eGFR and serum 20 Estimated 21 mL/min/ creatinine readings from HA information systems that glomerular 2 22 1.73m are available between 1 Jan 2006 and 31 Dec 2017 23 filtration rate should be provided. Baseline reading will use the latest 24 (eGFR) 25 available reading before and 12 months after 1 Jan 2008 26 27 28 Patients with microalbuminuria (albuminuria), referring 29 to men with ACR ≥ 2.5 and ≤ 25mg/mmol 30 (ACR>25mg/mmol) and women with ≥ 3.5 and ≤ 31 35/mmol (ACR>35mg/mmol) Also there should be two 32 Albuminuria NA 33 records per patient. The first record should be within 6 34 months on or before baseline. The second record should 35 36 be within 6 months after baseline.

37 Both the first and second records should be the http://bmjopen.bmj.com/ 38 records closest to 1 Jan 2008. 39 40 Disease characteristics 41 42 Duration of The duration of diabetes will use the self-reported Year 43 diabetes record in CMS RAMP-DM module. 44

45 on September 29, 2021 by guest. Protected copyright. 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 29 of 30 BMJ Open

1 2

3 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from 4 Risk factor Unit Operation Definition 5 Co-morbidity includes Chronic lung disease and 6 Cancer. 7 8 Chronic lung disease is defined by any of ICPC-2 9 R79, R95 or ICD-9-CM 491.x, 492.x, 493.2x ,496 10 Cancer is defined by any of ICPC-2 A79, B72-B74, 11 D74-D77, F74, H75, K72, L71, N74, R84-R85, S77, 12 Not 13 Co-morbidity T71, T73, U75-U77, W72, X75-X77, Y78 or ICD-9- applicable 14 CM 140.x-149.x , 150.x-159.x,160.x-165.x,170.x- 15 16 176.x, 179.x-189.x, 190.x-199.x,200.x-209.x 17 18 For peerThe review co-morbidity will only be defined if the patient record 19 20 contains a diagnosis of any of the above listed 21 comorbidities on or before 1 Jan 2008. 22 23 Treatment modalities 24 Anti-hypertensive drugs are identified by the 25 following drug item codes:HYDR05, HYDR30, 26 27 INDA01, INDA02, AMIL01, SPIR01, DYAZ01, 28 MODU01, ATEN01, ATEN02, BISO01, BISO02, 29 CARV01, CARV02, CARV03, CARV04, LODO02, 30 LODO03, METO06, METO07, METO10, METO11, 31 32 METO13, METO16, METO17, NADO01, PIND01, 33 PROP04, PROP05, HYDR01, HYDR02, METH22, 34 METH23, PRAZ03, PRAZ04, CAPT01, CAPT02, 35 36 CAPT03, CAPT06, ENAL01, ENAL02, ENAL03, 37 LISI01, LISI02, LISI03, PERI17, PERI28, PERI29, http://bmjopen.bmj.com/ 38 Use of anti- RAMI01, RAMI02, RAMI03, CAND01, CAND02, CO- 39 Not hypertensive D01, CO-D02, EXFO01, EXFO02, EXFO03, IRBE01, 40 applicable 41 drugs IRBE02, IRBE03, IRBE04, LOSA01, 42 LOSA02, LOSA03, LOSA04, MICA01, S00031, 43 44 S00069, S00454, S00455, TELM01, TELM02,

45 VALS02, VALS03, AMLO01, AMLO02, DILT01, on September 29, 2021 by guest. Protected copyright. 46 DILT02, DILT05, DILT06, DILT07, DILT08, FELO01, 47 48 FELO02, FELO03, LACI01, LACI02, NIFE03, 49 NIFE04, NIFE05, NIMO01, NIMO02 50 51 52 The use of anti-hypertensive drugs will be defined if 53 the record of the anti-hypertensive drug prescription 54 for the patient within 6 months or before 1 Jan 2008. 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 30 of 30

1 2

3 BMJ Open: first published as 10.1136/bmjopen-2018-023070 on 15 October 2018. Downloaded from 4 Risk factor Unit Operation Definition 5 Insulin is identified by British National Formulary 6.1. 6 7 Use of Insulin Not The use of insulin will be defined if the record of the 8 applicable 9 insulin prescription for the patient within 6 months or 10 before 1 Jan 2008. 11 12 Oral anti-diabetic drugs are identified by British National 13 Formulary 6.1.2.1, 6.1.2.2 and 6.1.2.3. 14 Use of oral anti- Not 15 diabetic drugs applicable 16 The use of oral anti-diabetic drugs will be defined if the 17 record of the oral anti-diabetic drug prescription for the 18 For peerpatient review within 6 months only or before 1 Jan 2008. 19 20 lipid-lowering agents includes Atorvastatin, 21 Fluvastatin, Gemfibrozil, Pravastatin, Rosuvastatin and 22 Simvastatin and are identified by the following drug 23 item codes ATOR01, ATOR02, ATOR03, ATOR04, 24 Use of lipid- Not FLUV02, FLUV03, FLUV05, GEM01, GEM02, 25 26 lowering agents applicable PRAV01, PRAV02, ROSU01, ROSU02, ROSU03, 27 SIMV01, SIMV02, SIMV04 and SIMV05 28 29 The use of lipid-lowering agents will be defined if the 30 record of the lipid-lowering agents prescription for the 31 32 patient within 6 months or before 1 Jan 2008. 33 34 35 Reference 36 37 1. Levey AS, Bosch JP, Lewis JB, et al. A more accurate method to estimate glomerular http://bmjopen.bmj.com/ 38 39 filtration rate from serum creatinine: a new prediction equation. Annals of internal medicine 40 1999.130(6):461-70. 41 42 43 44

45 on September 29, 2021 by guest. Protected copyright. 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml