Advertisement
Canadian Journal of Cardiology

Using Big Data for Cardiovascular Health Surveillance: Insights From 10.3 Million Individuals in the CANHEART Cohort

      Abstract

      Background

      The increasing availability of large electronic population-based databases offers unique opportunities to conduct cardiovascular health surveillance traditionally done using surveys. We aimed to examine cardiovascular risk-factor burden, preventive care, and disease incidence among adults in Ontario, Canada—using routinely collected data—and compare estimates with health survey data.

      Methods

      In the Cardiovascular Health in Ambulatory Care Research Team (CANHEART) initiative, multiple health administrative databases were linked to create a population-based cohort of 10.3 million adults without histories of cardiovascular disease. We examined cardiovascular risk-factor burden and screening and outcomes between 2016 and 2020. Risk- factor burden was also compared with cycles 3 to 5 (2012 to 2017) of the Canadian Health Measures Survey (CMHS), which included 9473 participants across Canada.

      Results

      Mean age of our study cohort was 47.9 ± 17.0 years, and 52.0% were women. Lipid and diabetes assessment rates among individuals 40 to 79 years were 76.6% and 78.2%, respectively, and lowest among men 40 to 49 years of age. Total cholesterol levels and diabetes and hypertension rates among men and women 20 to 79 years were similar to Canadian Health Measures Survey (CHMS) findings (total cholesterol: 4.80/4.98 vs 4.94/5.25 mmol/L; diabetes: 8.2%/7.1% vs 8.1%/6.0%; hypertension: 21.4%/21.6% vs 23.9%/23.1%, respectively); however, patients in the CANHEART study had slightly higher mean glucose (men: 5.79 vs 5.44; women: 5.39 vs 5.09 mmol/L) and systolic blood pressures (men: 126.2 vs 118.3; women: 120.6 vs 115.7 mm Hg).

      Conclusions

      Cardiovascular health surveillance is possible through linkage of routinely collected electronic population-based datasets. However, further investigation is needed to understand differences between health administrative and survey measures cross-sectionally and over time.

      Résumé

      Contexte

      La disponibilité croissante de vastes bases de données électroniques populationnelles offre des possibilités uniques d’effectuer une surveillance de la santé cardiovasculaire qui aurait été traditionnellement réalisée par des enquêtes. Notre objectif était d’examiner le fardeau des facteurs de risque cardiovasculaire, la prestation de soins de prévention et l’incidence des maladies cardiovasculaires chez des adultes de l’Ontario (Canada) en utilisant les données recueillies systématiquement, et de comparer ces estimations avec celles obtenues avec des données provenant d’enquêtes sur la santé.

      Méthodologie

      Dans le cadre de l’initiative de la Cardiovascular Health in Ambulatory Care Research Team (CANHEART), différentes bases de données de santé de nature administrative ont été liées pour créer une cohorte populationnelle de 10,3 millions d’adultes sans antécédents de maladies cardiovasculaires. Nous avons examiné le fardeau des facteurs de risque cardiovasculaire, ainsi que le dépistage et les résultats de santé cardiovasculaire entre 2016 et 2020. Le fardeau des facteurs de risque a également été comparé aux données des cycles 3 à 5 (de 2012 à 2017) de l’Enquête canadienne sur les mesures de la santé (ECMS), menée auprès de 9 473 personnes au Canada.

      Résultats

      L’âge moyen des personnes faisant partie de la cohorte à l’étude était de 47,9 ± 17,0 ans, et 52,0 % étaient des femmes. Les taux d’évaluation des lipides et du statut du diabète chez les personnes âgées de 40 à 79 ans étaient respectivement de 76,6 % et 78,2 %, et ces taux étaient les plus faibles chez les hommes de 40 à 49 ans. Les taux de cholestérol total, de diabète et d’hypertension chez les hommes et les femmes de 20 à 79 ans étaient comparables à ceux rapportés par l’ECMS (cholestérol total : 4,80/4,98 vs 4,94/5,25 mmol/l; diabète : 8,2 %/7,1 % vs 8,1 %/6,0 %; hypertension : 21,4 %/21,6 % vs 23,9 %/23,1 %, respectivement). Par contre, les patients de l’étude CANHEART présentaient des valeurs moyennes légèrement plus élevées pour la glycémie (hommes : 5,79 vs 5,44; femmes : 5,39 vs 5,09 mmol/l) et la pression artérielle systolique (hommes : 126,2 vs 118,3; femmes : 120,6 vs 115,7 mm Hg).

      Conclusions

      Il est possible d’effectuer une surveillance de la santé cardiovasculaire par l’association d’ensembles de données électroniques recueillies systématiquement à l’échelle des populations. Une investigation plus approfondie reste néanmoins nécessaire pour comprendre les différences entre les mesures provenant des bases de données de santé administratives et celles provenant d’enquêtes, sur le plan transversal et au fil du temps.
      Public health surveillance has been described by the World Health Organization as “the continuous, systematic collection, analysis, and interpretation of health-related data needed for the planning, implementation, and evaluation of public health practice.”
      World Health Organization
      Surveillance.
      Notable platforms of chronic disease surveillance using representative surveys of the general population, such as the National Health and Nutrition Examination Survey (NHANES) in the United States and its counterpart in Canada, the Canadian Health Measures Survey (CHMS), have played a major role in surveillance of cardiovascular disease (CVD) in North America.
      National Health and Nutrition Examination Survey
      ,
      Statistics Canada
      Canadian Health Measures Survey (CHMS).
      Traditionally, each study participant undergoes an interview and physical examination in a mobile clinic that travels across the country. The participant is then given a sampling weight, which enables the calculation of national estimates of the population with risk factors such as high cholesterol, diabetes, and hypertension. These surveys are used to estimate national trends in population health by comparing findings from different waves of the surveys over time. Although these traditional surveillance systems have yielded many important insights, they suffer important limitations. Owing to substantial costs per participant, sample sizes are modest (∼5000 participants per year in NHANES and ∼6000 participants in CHMS every 2-year cycle), thus limiting geographic generalizability and analyses at subgroup levels. In addition, they often exclude participants aged 80 and older, the most rapidly growing segment of the population.
      National Health and Nutrition Examination Survey
      Statistics Canada
      Canadian Health Measures Survey (CHMS).
      • Roger V.L.
      • Sidney S.
      • Fairchild A.L.
      • et al.
      Recommendations for cardiovascular health and disease surveillance for 2030 and beyond: a policy statement from the American Heart Association.
      As increasing amounts of health and population data are being collected and stored in digital form, a potential alternative is the secondary analysis of population cohorts assembled using information routinely collected on an ongoing basis in health care systems.
      Institute of Medicine (US) Committee on a National Surveillance System for Cardiovascular and Select Chronic Diseases
      A Nationwide Framework for Surveillance of Cardiovascular and Chronic Lung Diseases.
      The Cardiovascular Health in Ambulatory Care Research Team (CANHEART) "big data" initiative involves linkage of multiple population-based electronic health databases to create cohorts of the population in Ontario, Canada.
      • Tu J.V.
      • Chu A.
      • Donovan L.R.
      • et al.
      The Cardiovascular Health in Ambulatory Care Research Team (CANHEART): using big data to measure and improve cardiovascular health and health care services.
      ,
      • Tu J.V.
      • Chu A.
      • Maclagan L.
      • et al.
      Regional variations in ambulatory care and incidence of cardiovascular events.
      In this study, our primary objective was to demonstrate the potential utility of a CANHEART cohort for cardiovascular health surveillance by examining cardiovascular preventive health care utilization, risk-factor prevalence, and incidence of CVD. In addition to surveillance, this information can be a means to identify gaps in current screening practices (eg, subpopulations over- and underscreened) and opportunities to evaluate and enhance the effectiveness of cardiovascular prevention guidelines. To triangulate our results, we also compared risk factor estimates with those from the CHMS.

      Methods

      The research reported in this paper adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline (Supplemental Table S1).

      Study population

      The CANHEART 2016 cohort was created by linking together information on community-dwelling Ontario residents, aged 0 to 105 years, as of January 2016, from 21 routinely collected population-based health databases (Supplemental Fig. S1).
      • Tu J.V.
      • Chu A.
      • Donovan L.R.
      • et al.
      The Cardiovascular Health in Ambulatory Care Research Team (CANHEART): using big data to measure and improve cardiovascular health and health care services.
      Data sources included survey, physician services, hospitalization, laboratory, and electronic medical record databases. For this study, we identified a primary prevention cohort of individuals 20 to 105 years without known histories of hospitalization between 1988 (start of data availability) and cohort inception for any of myocardial infarction, stroke, heart failure, or revascularization with percutaneous coronary intervention or coronary artery bypass graft surgery.

      Cardiovascular indicators

      Our preventive health care utilization indicators included rates of primary care physician visits, lipid and diabetes assessments, and periodic health examinations during a 5-year follow-up period (2016 to 2020). A lipid assessment was defined as having had total cholesterol, high-density lipoprotein (HDL) and triglycerides tests all on the same day. Diabetes testing was defined as tests of fasting blood glucose, oral glucose tolerance, and HbA1c and excluded testing during pregnancy and among individuals with diabetes as of January 1, 2016. Cardiovascular risk factors studied were baseline blood lipid and glucose levels; prevalence of diabetes, hypertension, obesity and smoking; and systolic and diastolic blood pressure. Finally, 3-year (2016 to 2018) CVD incidence was examined using 2 definitions: major CVD events, defined as a composite of hospitalization for myocardial infarction or stroke or death caused by ischemic heart disease or stroke and general CVD events, which additionally included hospitalization for heart failure, coronary revascularization, or death caused by all diseases of the circulatory system as listed in Chapter IX of the International Classification of Diseases 10th Revision codebook.
      World Health Organization
      International Statistical Classification of Diseases and Related Health Problems. Tenth Revision, Volumes 1 to 3.
      CVD incidence was restricted to 3 years of follow-up owing to lags in the availability of cause of death data beyond 2018.

      Data sources

      To study preventive care utilization indicators, we used the Ontario Health Insurance Plan (OHIP) Physician Claims Database, which captures information from all physician claims and outpatient laboratory visits for the Ontario population (see Supplemental Table S2 for codes). Lipid and HbA1c glucose test results were available for a subpopulation of approximately 6.7 million individuals (65.1% of population), with results available from the Ontario Laboratories Information System (OLIS). OLIS is a repository of laboratory test information from all of Ontario’s hospital, community, and public health laboratories introduced in 2007, with gradual uptake since then.
      Linkages to information on 294,922 primary care adult patients captured in the Electronic Medical Records Primary Care database (EMRPC) between 2011 and 2016 provided information on systolic and diastolic blood pressure and weight and height measurements for determination of obesity (body mass index ≥ 30 kg/m2).
      • Tu K.
      • Mitiku T.F.
      • Ivers N.M.
      • et al.
      Evaluation of electronic medical record administrative data linked database (EMRALD).
      EMRPC is an electronic medical record database with data from more than 350 family physicians practicing across Ontario, although rural communities are under-represented.
      • Tu K.
      • Mitiku T.F.
      • Ivers N.M.
      • et al.
      Evaluation of electronic medical record administrative data linked database (EMRALD).
      ,
      • Jaakimanian L.
      • Crampton N.
      • Pinzaru V.B.
      • DelGiudice L.
      • Tu K.
      Using family physician electronic medical record data to measure the pathways of cancer care.
      The prevalence of physician-diagnosed diabetes and hypertension in the overall study population was determined using information from the Ontario Diabetes Database and Ontario Hypertension Database. Both are validated population-based registries of individuals with these conditions identified from the Canadian Institute for Health Information Discharge Abstract Database (CIHI DAD) or the OHIP Physician Claims Database.
      • Hux J.E.
      • Ivis F.
      • Flintoft V.
      • Bica A.
      Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm.
      ,
      • Tu K.
      • Campbell N.R.
      • Chen Z.L.
      • Cauch-Dudek K.J.
      • McAlister F.A.
      Accuracy of administrative databases in identifying patients with hypertension.
      Estimates of smoking prevalence was determined from representative and weighted samples of more than 100,000 Ontario residents participating in the 2005 to 2014 cycles of the Canadian Community Health Survey, a telephone survey designed to collect information on health status, health determinants, and health care utilization for the Canadian population 12 years and older.
      Canadian Community Health Survey Annual Component
      Statistics Canada. 2019.
      To determine the incidence of CVD events, we performed record linkages to the CIHI DAD and the Office of the Registrar General of Ontario Vital Statistics Database.
      These datasets were linked using unique encoded identifiers (ie, encrypted health card numbers) and analyzed at ICES (formerly Institute for Clinical Evaluative Sciences) in Toronto, Ontario, Canada. ICES is an independent nonprofit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. The use of the data in this project is authorized under section 45 and approved by the ICES Privacy and Legal Office. Datasets are held securely in coded form at ICES. Although legal data sharing agreements between ICES and data providers prohibit ICES from making the dataset publicly available, access may be granted to those who meet prespecified criteria for confidential access, available at www.ices.on.ca/DAS.

      Canadian Health Measures Survey

      We also examined how risk-factor burden in the CANHEART cohort compares with estimates from the CHMS to assist with interpretation of results when using either data source for surveillance. The CHMS is an ongoing national survey of community-dwelling Canadians initiated in 2007, being used for national-level health surveillance.
      Statistics Canada
      Canadian Health Measures Survey (CHMS).
      Surveys are collected in 2-year cycles and aim to measure the health of community-dwelling Canadians 3 to 79 years of age through household interviews and direct physical measures of a representative sample of the Canadian population. Persons living on reserves or other Aboriginal settlements, institutional residents, certain remote areas or areas with a low population density, and full-time members of the Canadian Forces are excluded. Household interviews gather demographic, socioeconomic, health, nutrition, and lifestyle information, whereas visits to mobile examination centres are used to obtain physical measures including blood pressure, height, weight, and a blood sample. Estimates of risk factor burden from the CANHEART cohort were compared with 9473 participants aged 20 to 79 years in Cycles 3 to 5 (2012 to 2017) of the CHMS.
      Statistics Canada
      Canadian Health Measures Survey (CHMS).
      ,
      • Bushnik T.
      • Hennessy D.
      • McAlister F.A.
      • Manuel D.G.
      Factors associated with hypertension control among older Canadians. Statistics Canada, Catalogue no. 82-003-X.
      ,
      • Leung A.A.
      • Bushnik T.
      • Hennessy D.
      • McAlister F.A.
      • Manuel D.G.
      Risk factors for hypertension in Canada. Statistics Canada, Catalogue no. 82-003-X.

      Data analyses

      Analyses of linked health care utilization and risk-factor data were stratified by sex and 10-year age groups. Health care utilization was calculated as the proportion receiving services at least once over the 5-year follow-up period, except for mean annual visits to primary care physicians. Prevalence of dichotomous risk factors was calculated as proportions, and for continuous measures, as means and standard deviations or medians and quartiles 1 and 3 to illustrate the spread of the data. Laboratory-based measures were restricted to results available between 2012 and 2017, with preference for results closest and before the cohort inception date of January 1, 2016, to be consistent with CHMS data. For comparisons with the CHMS, we restricted estimates to persons 20 to 79 years of age, as this is the age range for which data were available from both CANHEART and CHMS. To account for the complex sampling design of CHMS, analyses of CHMS data were weighted using the combined survey weights from Cycles 3,4, and 5, except for analyses of fasting glucose, which came from Cycles 3 and 4 only. Replicate weights generated by Statistics Canada were used to perform the variance estimation. We were unable to perform formal tests of statistical significance of differences because analyses of CHMS data were performed external to ICES, and sharing of individual-level data was prohibited by privacy laws. Incidence of CVD events was calculated as events per 1000 person-years of follow-up, with individuals censored at death or loss of OHIP eligibility.
      In supplementary analyses to compare changes over time, we also performed similar analyses using a CANHEART cohort with an inception date of January 1, 2008, including 9.8 million individuals aged 20 to 105 years without histories of CVD. Estimates were compared with those from 3517 participants in Cycle 1 (2007 to 2009) of the CHMS.
      Canadian Health Measures Survey: Cycle 1 Data Tables 2007 to 2009
      Catalogue no. 82-623-X.
      Analyses of CANHEART data were conducted using SAS version 9.4 (SAS Institute, Cary, North Carolina USA). Analyses of CHMS data were conducted at Statistics Canada in Ottawa, Ontario, Canada using SAS 9.4 and Stata 14 statistical software (StatCorp LP, College Station, Texas, USA) or obtained from a previously published report.
      Canadian Health Measures Survey: Cycle 1 Data Tables 2007 to 2009
      Catalogue no. 82-623-X.

      Results

      Of 10,793,909 individuals aged 20 to 105 years in the CANHEART cohort, 10,311,810 (95.5%) were free of a history of CVD at cohort inception and included in this study. Mean age of our study cohort was 47.9 ± 17.0 years (men: 47.0 ± 16.6 years and women: 48.7 ± 17.4 years), and 52.0% were women.

      Preventive health care utilization

      Figure 1 and Supplemental Table S3 show the rates of primary care physician visits and cardiac risk-factor assessments between 2016 and 2020 among individuals aged 20 to 105 years in Ontario. Overall, 88.1% of individuals (85.5% of men and 90.4% of women) made at least 1 visit to primary care physicians during the 5 years. On average, women made more frequent visits than men (3.9 vs 2.8 visits per year, P < 0.001) and had higher testing rates for all cardiac risk factors.
      Figure thumbnail gr1
      Figure 1Primary care visits and cardiac risk-factor assessments, 2016 to 2020. Diabetes testing rates are among patients without diabetes at baseline and exclude screening for gestational diabetes.
      By age, lipid and diabetes screening rates among individuals 40 to 79 years were 76.6% and 78.2%, respectively; however, men aged 40 to 49 years had lipid and diabetes testing rates of only 63.9% and 65.2%, respectively. Among individuals 20 to 39 years, for whom Canadian guidelines do not generally recommend testing, 38.0% and 44.5% of men and women received lipid testing, whereas 45.6% and 68.4% received diabetes testing, respectively.
      • Ekoe J.-M.
      • Goldenberg R.
      • Katz P.
      Diabetes Canada Expert Committee
      Screening for diabetes in adults.
      ,
      • Anderson T.J.
      • Grégoire J.
      • Pearson G.J.
      • et al.
      2016 Canadian Cardiovascular Society guidelines for the management of dyslipidemia for the prevention of cardiovascular disease in the adult.
      Rates of periodic health examinations were highest among 50- to 79-year-old men and 40- to 69-year-old women.

      Cardiac risk factors

      Prevalence of traditional cardiac risk factors and results from risk-factor assessments are shown in Table 1. Cholesterol levels were greatest among men 40 to 49 years of age, with the exception of HDL, and women aged 50 to 59 years. Prevalence of diabetes increased with age from ∼1% among 20- to 29-year-old men and women to 24.8% and 19.8% among 70- to 79-year-old men and women, respectively. Prevalence of hypertension also increased with age, reaching 74.4% in men and 80.6% in women aged 80 to 105 years. Obesity rates were high in all age and sex groups and particularly among middle-aged groups, in which rates were above 30%. Men and women 30 to 39 years of age had the highest smoking rates (30.3% and 21.3%, respectively).
      Table 1Cardiac risk factors, 2016
      CharacteristicAge group (years)
      20-2930-3940-4950-5960-6970-7980-105Overall
      Men
       Population size, n911,262884,617960,0681,018,352682,726338,404156,7704,952,199
       Total cholesterol, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      4.41 ± 0.924.90 ± 1.005.06 ± 1.054.97 ± 1.084.66 ± 1.094.32 ± 1.044.10 ± 0.984.77 ± 1.08
       HDL, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      1.28 ± 0.331.23 ± 0.331.23 ± 0.331.27 ± 0.361.29 ± 0.371.30 ± 0.371.31 ± 0.371.27 ± 0.35
       Total cholesterol/HDL ratio, mean ± SD
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      3.67 ± 1.264.23 ± 1.394.34 ± 1.354.13 ± 1.273.80 ± 1.173.50 ± 1.083.30 ± 1.023.99 ± 1.29
       LDL, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      2.55 ± 0.792.91 ± 0.853.02 ± 0.892.91 ± 0.932.64 ± 0.952.35± 0.902.18 ± 0.842.77 ± 0.93
       LDL ≥ 3.5 mmol/L, %
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      11.422.828.025.918.711.47.321.1
       Diabetes, %0.81.85.310.618.324.824.38.7
       Fasting serum glucose, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      5.01 ± 1.325.29 ± 1.555.65 ± 1.865.96 ± 1.976.15 ± 1.926.18 ± 1.776.08 ± 1.685.80 ± 1.85
       Impaired fasting glucose (≥ 6.1 mmol/L), %
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      2.35.19.815.118.919.318.312.6
       Hypertension, %1.05.214.728.547.264.374.423.1
       Systolic BP, mean ± SD, mm Hg
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      119.9 ± 13.0122.2 ± 13.1124.9 ± 14.4128.1 ± 15.5131.3 ± 16.2132.3 ± 16.7132.2 ± 17.7126.4 ± 15.5
       Diastolic BP, mean ± SD, mm Hg
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      74.3 ± 9.577.7 ± 9.780.2 ± 10.180.9 ± 9.979.2 ± 9.875.8 ± 9.872.4 ± 10.778.3 ± 10.2
       Obese (BMI ≥ 30 kg/m2), %
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      19.728.635.737.738.134.023.732.9
       Current smoker, %
      Among 46,899 men and 60,510 women linked to the 2005 to 2014 Canadian Community Health Surveys.
      21.330.325.726.618.610.95.923.1
      Women
       Population size, n883,238934,1101,000,5321,059,233788,208436,917257,3735,359,611
       Total cholesterol, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      4.40 ± 0.834.60 ± 0.864.90 ± 0.925.31 ± 1.045.22 ± 1.124.92 ± 1.124.70 ± 1.104.97 ± 1.05
       HDL, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      1.53 ± 0.401.50 ± 0.401.54 ± 0.411.61 ± 0.451.60 ± 0.451.59 ± 0.451.58 ± 0.441.57 ± 0.43
       Total cholesterol/HDL ratio, mean ± SD
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      3.05 ± 0.953.25 ± 1.033.39 ± 1.063.50 ± 1.083.44 ± 1.043.25 ± 0.983.14 ± 0.973.35 ± 1.04
       LDL, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      2.38 ± 0.712.56 ± 0.752.77 ± 0.803.04 ± 0.912.93 ± 0.982.65 ± 0.972.46 ± 0.942.77 ± 0.90
       LDL ≥ 3.5 mmol/L, %
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      6.710.417.028.927.619.314.120.2
       Diabetes, %0.82.04.68.514.619.819.87.7
       Fasting serum glucose, mean ± SD, mmol/L
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      4.77 ± 1.074.93 ± 1.165.23 ± 1.435.54 ± 1.655.77 ± 1.675.88 ± 1.645.87 ± 1.635.42 ± 1.55
       Impaired fasting glucose (≥ 6.1 mmol/L), %
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      1.62.75.08.511.813.414.57.5
       Hypertension, %0.84.112.026.146.467.080.624.4
       Systolic BP, mean ± SD, mm Hg
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      112.4 ± 12.1113.4 ± 13.0117.8 ± 15.0123.9 ± 16.3129.2 ± 17.0133.5 ± 17.4136.3 ± 18.8121.3 ± 17.1
       Diastolic BP, mean ± SD, mm Hg
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      71.6 ± 9.272.7 ± 9.775.2 ± 10.177.3 ± 10.076.9 ± 9.675.1 ± 9.873.3 ± 10.474.8 ± 10.0
       Obese (BMI ≥ 30 kg/m2), %
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      19.625.829.533.135.532.624.229.3
       Current smoker, %
      Among 46,899 men and 60,510 women linked to the 2005 to 2014 Canadian Community Health Surveys.
      15.421.316.218.215.28.95.016.0
      BMI, body mass index; BP, blood pressure; HDL, high-density lipoprotein; LDL, low-density lipoprotein; SD, standard deviation.
      Among 3,043,241 men and 3,668,192 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017. Estimates for impaired fasting glucose are among patients without diabetes at baseline.
      Among 122,041 men and 172,881 women linked to the Electronic Medical Records Primary Care database.
      Among 46,899 men and 60,510 women linked to the 2005 to 2014 Canadian Community Health Surveys.

      Comparison with CHMS

      Comparisons between CANHEART and the CHMS are shown in Table 2 and Supplemental Fig. S2. Blood cholesterol levels and diabetes and hypertension rates determined from those 20 to 79 years of age in the CANHEART cohort were similar to CHMS findings (total cholesterol, 4.80/4.98 vs 4.94/5.25 mmol/L; diabetes, 8.2%/7.1% vs 8.1%/6.0%; hypertension, 21.4%/21.6% vs 23.9/23.1 among men and women, respectively) as were obesity rates among women (29.5% vs 28.9%). However, the CANHEART cohort had slightly higher mean glucose levels (men: 5.79 vs 5.44; women: 5.39 vs 5.09 mmol/L), blood pressure levels (men: systolic 126.2 vs 118.3; women: 120.6 vs 115.7 mm Hg), and obesity rates among men (33.3% vs 28.2%). Comparing trends by sex and age (Supplemental Fig. S2), patterns were similar with women in both populations having better risk factor profiles than men, and in both the CANHEART and CHMS populations, triglycerides (among women), glucose levels, and systolic blood pressure all increased with age, whereas triglycerides (among men), low-density lipoprotein (LDL) and total HDL ratio peaked in the middle-age group (40 to 59 years).
      Table 2Baseline cardiac risk factors among CANHEART vs Canadian Health Measures Survey (CHMS) study populations, 20 to 79 years, 2012 to 2017
      CharacteristicCANHEART
      Among 2,917,846 men and 3,464,067 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017.
      CHMS
      From Cycles 3 to 5 of the Canadian Health Measures Survey (2012 to 2017), except fasting glucose, which is from Cycle 3 to 4 (2012 to 2015).
      Mean (95% CI) unless otherwise specified
      Men
       Total cholesterol, mmol/L4.80 (4.80-4.81)4.94 (4.80-5.08)
       HDL, mmol/L1.26 (1.26-1.26)1.23 (1.21-1.25)
       Total cholesterol/HDL ratio4.02 (4.02-4.02)4.25 (4.08-4.41)
       LDL, mmol/L2.79 (2.79-2.79)2.88 (2.83-2.94)
       Non-HDL, mmol/L3.54 (3.54-3.54)3.56 (3.51-3.62)
       Triglycerides, mmol/L1.69 (1.69-1.69)1.54 (1.46-1.62)
       Diabetes8.2 (8.1-8.2)8.1 (7.1-9.1)
       Fasting glucose, mmol/L5.79 (5.79-5.79)5.44 (5.34-5.55)
       Hypertension21.4 (21.4-21.4)23.87 (22.07-25.77)
       Systolic BP, mm Hg
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      126.2 (126.1-126.3)118.3 (117.7-119.0)
       Diastolic BP, mm Hg
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      78.6 (78.5-78.6)77.0 (76.5-77.5)
       Obese (BMI ≥ 30 kg/m2)
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      33.3 (32.9-33.6)28.2 (25.5-31.1)
      Women
       Total cholesterol, mean, mmol/L4.98 (4.98-4.98)5.25 (4.94-5.57)
       HDL, mean, mmol/L1.57 (1.57-1.57)1.54 (1.51-1.57)
       Total cholesterol/HDL ratio3.36 (3.36-3.36)3.41 (3.33-3.49)
       LDL, mean, mmol/L2.79 (2.79-2.79)2.73 (2.65-2.80)
       Non-HDL, mean, mmol/L3.41 (3.41-3.41)3.36 (3.28-3.43)
       Triglycerides, mean, mmol/L1.37 (1.37-1.37)1.29 (1.24-1.35)
       Diabetes7.1 (7.1-7.1)6.0 (5.2-7.0)
       Fasting glucose, mean, mmol/L5.39 (5.39-5.40)5.09 (4.99-5.19)
       Hypertension21.6 (21.5-21.6)23.1 (21.0-25.4)
       Systolic BP, mm Hg
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      120.6 (120.5-120.6)115.7 (114.8-116.6)
       Diastolic BP, mm Hg
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      74.8 (74.8-74.9)74.2 (73.7-74.6)
       Obese (BMI ≥ 30 kg/m2)
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.
      29.5 (29.2-29.7)28.9 (25.9-32.2)
      BMI, body mass index; BP, blood pressure; CANHEART, Cardiovascular Health in Ambulatory Care Research Team; CI, confidence interval; HDL, high-density lipoprotein; LDL, low density lipoprotein.
      Among 2,917,846 men and 3,464,067 women with test results available from the Ontario Laboratories Information System (OLIS) between 2012 and 2017.
      From Cycles 3 to 5 of the Canadian Health Measures Survey (2012 to 2017), except fasting glucose, which is from Cycle 3 to 4 (2012 to 2015).
      Among 117,394 men and 164,479 women linked to the Electronic Medical Records Primary Care database.

      Clinical event rates

      The 3-year incidence rates of CVD events among the CANHEART cohort are shown by sex and individual age in Figure 2. Overall incidence of major CVD events was 2.89 events per 1000 person-years, increasing with age from less than 1 per 1000 person-years in the population ≤ 40 years for both sexes to more than 20 per 1000 person-years (or a 10-year risk of 20%) among men 85 years and older and women 87 years and older. Men and women reached the 7.5% 10-year major CVD risk threshold for consideration of initiation of statin therapy at age 66 years and 76 years, respectively. The overall general CVD event rate was 4.68 events per 1000 person-years of follow-up and showed similar patterns to major CVD event rates but higher absolute rates, with men and women reaching the widely used intermediate-risk threshold of 10% at 62 and 73 years, respectively, and the high-risk threshold of 20% at 76 and 81 years, respectively.
      Figure thumbnail gr2
      Figure 2Incidence rate of cardiovascular disease events by age (2016 to 2018). A major cardiovascular event is defined as a composite of hospitalization for myocardial infarction, stroke, or death caused by ischemic heart disease or cerebrovascular diseases. A general cardiovascular event is defined as a major cardiovascular event or a hospitalization for heart failure, revascularization with either percutaneous coronary intervention or coronary artery bypass graft surgery, or death due to all diseases of the circulatory system as listed in Chapter IX of the International Classification of Diseases 10th Revision codebook. A 10-year risk of 7.5% for a major cardiovascular event is the threshold for consideration of initiation of statin therapy by US clinical practice guidelines. A 10-year risk of 10% and 20% for a general cardiovascular event are the intermediate and high-risk thresholds in Canadian clinical practice guidelines.

      Comparison with 2008

      Overall, trends in health and health care by age group and sex in 2016 were similar to results in 2008. However, compared with preventive care in 2008 to 2012, care in 2016 was lower among younger age groups, in which screening is generally not recommended, and higher among older age groups (Supplemental Fig. S3). Among cardiac risk factors, lipid profiles improved from 2008 in both the CANHEART and CHMS populations, whereas rates and levels of other risk factors were similar (Supplemental Tables S3 and S4). Along with improved lipid profiles, incidence of major and general CV event incidence also improved from 3.26 and 5.12 events per 1000 person-years, respectively, between 2008 and 2012, to 2.89 and 4.68 events per 1000 person-years, respectively, between 2016 and 2018. In addition, risk thresholds were reached at older ages among both male and female subjects during the earlier period (Supplemental Fig. S3).

      Discussion

      This large population-based study of 10.3 million primary-prevention patients from Ontario shows the potential of a big data linkage approach for population health surveillance and health system evaluation. We observed that almost 90% of Ontarians visited family doctors over a 5- year study period and that subsequent testing rates for cardiac risk factors were generally high in the middle-aged 40- to 79-year-old demographic groups for whom universal risk-factor screening is recommended in Canada; however, testing rates were relatively low among men below 50 years of age. We also found significant variations in the relationship between age and the traditional cardiac risk factors. Although improved since 2008, lipid levels peaked among middle-aged men (40 to 49 years of age) and women (50 to 59 years of age), whereas diabetes and hypertension rates generally increased with age, and smoking rates were highest in young adults. Comparing risk-factor prevalence with the CHMS, lipid levels were most similar, whereas other risk factors—particularly glucose and blood pressure values and obesity rates—were more favourable among CHMS participants. Incidence of major and general CVD events decreased approximately 11% and 9%, respectively, between 2008 and 2016, although men continue to reach high-risk thresholds several years earlier than women.
      Our analyses demonstrate how big data cohorts using population-based health administrative databases linked to other data sources contain rich individual-level information. These cohorts can be used to enhance health surveillance and evaluate health system performance across the spectrum of risk-factor assessment and prevalence to population health and clinical outcomes. We show consistently high rates of adherence to Canadian cardiovascular-related clinical practice guidelines such as from the Canadian Cardiovascular Society, Diabetes Canada, and Canadian Hypertension Education Program.
      • Ekoe J.-M.
      • Goldenberg R.
      • Katz P.
      Diabetes Canada Expert Committee
      Screening for diabetes in adults.
      • Anderson T.J.
      • Grégoire J.
      • Pearson G.J.
      • et al.
      2016 Canadian Cardiovascular Society guidelines for the management of dyslipidemia for the prevention of cardiovascular disease in the adult.
      • Daskalopoulou S.S.
      • Rabi D.M.
      • Zarnke K.B.
      • et al.
      The 2015 Canadian hypertension education program recommendations for blood pressure measurement, diagnosis, assessment of risk, prevention, and treatment of hypertension.
      Yet, our results also indicate that risk-factor assessments of men 40 to 49 years of age could be improved, as this demographic group has the least favourable lipid profiles and high rates of obesity. Greater testing could potentially identify more individuals in need of behavioural modifications or initiation of medical therapy in an effort to lower risk of CVD. As men in this age group averaged 2.5 visits to primary care physicians per year vs 3.5 for women of the same age group, whether more frequent visits or other interventions to reach men would improve assessment rates deserves further exploration. The contribution of improvements in lipid profiles to reductions in cardiovascular event rates and the age of reaching cardiovascular risk thresholds between 2008 and 2012 vs 2016 to 2018 is also worth additional study. Nonetheless, the testing and cardiovascular health patterns detected highlight surveillance opportunities using big data from administrative databases to identify care gaps and areas for targeted action. Tracking testing rates through such a big data cohort could also provide a means for evaluating the impact of health system improvement initiatives and identifying opportunities for additional education to improve the efficiency of health care resources.
      Also noteworthy are the similarities we found in cholesterol levels and risk-factor prevalence alongside differences in glucose and blood pressure values and obesity rates (in men) when compared with the CHMS. Although estimates from the Ontario population may indeed be different from the Canadian population represented in the CHMS, the differences observed may also be attributable to selection bias. In the CANHEART population, blood pressure and obesity were obtained from a subpopulation of primary care patients, and it is possible that patients selected for measurement are at greater cardiovascular risk. On the other hand, survey participants represent general community dwelling and possibly healthier Canadians willing to participate in surveys and undergo testing. Thus, although direct comparisons of absolute results are not our intention, caution must be exercised in using blood pressure and weight metrics from electronic medical records for general population health surveillance. Further investigations are also warranted to identify the reasons for these differences.
      Querying of electronic health databases has been proposed as an emerging approach for cardiovascular care, health, and disease surveillance that may serve as a potential alternative or complement to prospective epidemiologic studies.
      • Roger V.L.
      • Sidney S.
      • Fairchild A.L.
      • et al.
      Recommendations for cardiovascular health and disease surveillance for 2030 and beyond: a policy statement from the American Heart Association.
      Although health indicators such as risk-factor burden are easier to capture than physiological measures such as blood pressure and weight, population-based electronic health databases have several advantages. The large sample sizes facilitate stratified analyses on subpopulations, such as by various sociodemographic characteristics, geography, risk factor, and disease and also enable the use of novel modelling methods for improved prediction of cardiovascular health metrics.
      • Roger V.L.
      • Sidney S.
      • Fairchild A.L.
      • et al.
      Recommendations for cardiovascular health and disease surveillance for 2030 and beyond: a policy statement from the American Heart Association.
      ,
      • Tu J.V.
      • Chu A.
      • Maclagan L.
      • et al.
      Regional variations in ambulatory care and incidence of cardiovascular events.
      ,
      • Di Giuseppe G.
      • Chu A.
      • Tu J.V.
      • Shanmugasegaram S.
      • Liu P.
      • Lee D.S.
      Incidence of heart failure among immigrants to Ontario, Canada: a CANHEART immigrant study.
      • Howell N.A.
      • Tu J.V.
      • Moineddin R.
      • et al.
      Interaction between neighborhood walkability and traffic-related air pollution on hypertension and diabetes: the CANHEART cohort.
      • Kapral M.K.
      • Austin P.C.
      • Jeyakumar G.
      • et al.
      Rural-urban differences in stroke risk factors, incidence, and mortality in people with and without prior stroke.
      With lower costs, such information is also a cost-effective tool for evaluating practice, monitoring health inequities, and guiding health policies to promote population health.

      Limitations

      Several limitations in this study warrant discussion. First, although we were able to capture several major cardiac risk factors, electronic health administrative databases are limited in information about lifestyle and behaviours (eg, diet and physical activity), which we were thus unable to examine. Second, with incomplete information on medication use (eg, over the counter or paid for by private insurers), our estimates of some risk factors (eg, lipid and blood-pressure measures) represent untreated and treated patients combined. Nonetheless, our results still provide insights into population level risk factor burden. Third, our cohort is created from linking several sources of routinely collected administrative data in Ontario and thus reflects health and health care in Ontario specifically. Although such linkages may not be possible in all jurisdictions, other Canadian provinces are increasingly performing secondary analyses of routinely collected data for health services research, and our hope is this proof-of-concept spurs extended use of such data for similar health surveillance nationally. Finally, administrative data are subject to variation in use of codes or coding errors. However, most indicators included in our study have been validated against external gold standards in previous studies.

      Conclusions

      Surveillance of cardiovascular health services use, risk factors, and incidence of disease is possible through linkage of routinely collected electronic population-based datasets. As a complement to traditional health surveillance, cohorts created from such linkages also have value for evaluating health system performance and identifying care gaps in select subpopulations not traditionally possible with smaller data sets. However, further investigation is needed fully to understand differences in findings between measures from health administrative vs survey data cross-sectionally and over time.

      Acknowledgements

      Parts of this material are based on data and information compiled and provided by the Canadian Institute for Health Information (CIHI) and Ontario Registrar General (ORG) information on deaths, the original source of which is ServiceOntario. The analyses, conclusions, opinions, statements, and views expressed herein are solely those of the authors and do not reflect those of CIHI, ORG, the Ministry of Government and Consumer Services, the funding, or data sources; no endorsement is intended or should be inferred. We thank Dynacare Medical Laboratories for providing access to the laboratory data being used in the CANHEART initiative.

      Funding Sources

      This study was supported by ICES , which is funded by an annual grant from the Ontario Ministry of Health ( MOH ) and the Ministry of Long-Term Care (MLTC). This study and the CANHEART initiative also received funding from an Institute of Circulatory and Respiratory Health ( ICRH, Canadian Institutes of Health Research ( CIHR ) Chronic Diseases Team operating grant (TCA 118349) and is currently funded by CIHR foundation grants (FDN-143313 and FDN-154333) and a CIHR Strategy for Patient-Oriented Research Innovative Clinical Trial Multi-Year Grant (MYG-151211).
      Dr Udell is supported by a Heart and Stroke Foundation National New Investigator-Ontario Clinician Scientist Award and the Ontario Ministry of Research, Innovation and Science Early Researcher Award. Dr Lee is the Ted Rogers Chair in Heart Function Outcomes, University Health Network, University of Toronto. Dr Tu was supported by a Tier 1 Canada Research Chair in Health Services Research and an Eaton Scholar award from the Department of Medicine, University of Toronto . Dr Ko is supported by the Jack Tu Chair in Cardiovascular Outcomes Research.

      Disclosures

      Dr Udell has received consulting fees or honoraria from Amgen, Boehringer Ingelheim, Janssen, Merck, Novartis and Sanofi and grant support from AstraZeneca, Novartis , and Sanofi. The other authors have no conflicts of interest to disclose.

      Supplementary Material

      References

        • World Health Organization
        Surveillance.
        (Available at:)
        • National Health and Nutrition Examination Survey
        (Available at:)
        https://www.cdc.gov/nchs/nhanes/index.htm
        Date accessed: February 14, 2020
        • Statistics Canada
        Canadian Health Measures Survey (CHMS).
        (Available at:)
        • Roger V.L.
        • Sidney S.
        • Fairchild A.L.
        • et al.
        Recommendations for cardiovascular health and disease surveillance for 2030 and beyond: a policy statement from the American Heart Association.
        Circulation. 2020; 141: e104-e119
        • Institute of Medicine (US) Committee on a National Surveillance System for Cardiovascular and Select Chronic Diseases
        A Nationwide Framework for Surveillance of Cardiovascular and Chronic Lung Diseases.
        National Academies Press, Washington, DC2011
        • Tu J.V.
        • Chu A.
        • Donovan L.R.
        • et al.
        The Cardiovascular Health in Ambulatory Care Research Team (CANHEART): using big data to measure and improve cardiovascular health and health care services.
        Circ Cardiovasc Qual Outcomes. 2015; 8: 204-212
        • Tu J.V.
        • Chu A.
        • Maclagan L.
        • et al.
        Regional variations in ambulatory care and incidence of cardiovascular events.
        CMAJ. 2017; 189: E494-E501
        • World Health Organization
        International Statistical Classification of Diseases and Related Health Problems. Tenth Revision, Volumes 1 to 3.
        World Health Organization, Geneva1994
        • eHealth Ontario. Lab Results
        (Available at:) (Published 2008. Accessed February 14, 2020)
        • Tu K.
        • Mitiku T.F.
        • Ivers N.M.
        • et al.
        Evaluation of electronic medical record administrative data linked database (EMRALD).
        Am J Manag Care. 2014; 20: e15-e21
        • Jaakimanian L.
        • Crampton N.
        • Pinzaru V.B.
        • DelGiudice L.
        • Tu K.
        Using family physician electronic medical record data to measure the pathways of cancer care.
        Int J Population Data Sci. 2018; 3
        • Hux J.E.
        • Ivis F.
        • Flintoft V.
        • Bica A.
        Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm.
        Diabetes Care. 2002; 25: 512-516
        • Tu K.
        • Campbell N.R.
        • Chen Z.L.
        • Cauch-Dudek K.J.
        • McAlister F.A.
        Accuracy of administrative databases in identifying patients with hypertension.
        Open Med. 2007; 1: e18-e26
        • Canadian Community Health Survey Annual Component
        Statistics Canada. 2019.
        2012 (Available at:)
        • Bushnik T.
        • Hennessy D.
        • McAlister F.A.
        • Manuel D.G.
        Factors associated with hypertension control among older Canadians. Statistics Canada, Catalogue no. 82-003-X.
        Health Rep. 2018; 29: 3-10
        • Leung A.A.
        • Bushnik T.
        • Hennessy D.
        • McAlister F.A.
        • Manuel D.G.
        Risk factors for hypertension in Canada. Statistics Canada, Catalogue no. 82-003-X.
        Health Rep. 2019; 30: 3-13
        • Canadian Health Measures Survey: Cycle 1 Data Tables 2007 to 2009
        Catalogue no. 82-623-X.
        Statistics Canada, 2010: 52-57
        • Ekoe J.-M.
        • Goldenberg R.
        • Katz P.
        • Diabetes Canada Expert Committee
        Screening for diabetes in adults.
        Can J Diabetes. 2018; 42: S16-S19
        • Anderson T.J.
        • Grégoire J.
        • Pearson G.J.
        • et al.
        2016 Canadian Cardiovascular Society guidelines for the management of dyslipidemia for the prevention of cardiovascular disease in the adult.
        Can J Cardiol. 2016; 32: 1263-1282
        • Daskalopoulou S.S.
        • Rabi D.M.
        • Zarnke K.B.
        • et al.
        The 2015 Canadian hypertension education program recommendations for blood pressure measurement, diagnosis, assessment of risk, prevention, and treatment of hypertension.
        Can J Cardiol. 2015; 31: 549-568
        • Di Giuseppe G.
        • Chu A.
        • Tu J.V.
        • Shanmugasegaram S.
        • Liu P.
        • Lee D.S.
        Incidence of heart failure among immigrants to Ontario, Canada: a CANHEART immigrant study.
        J Card Fail. 2019; 25: 425-435
        • Howell N.A.
        • Tu J.V.
        • Moineddin R.
        • et al.
        Interaction between neighborhood walkability and traffic-related air pollution on hypertension and diabetes: the CANHEART cohort.
        Environ Int. 2019; 132104799
        • Kapral M.K.
        • Austin P.C.
        • Jeyakumar G.
        • et al.
        Rural-urban differences in stroke risk factors, incidence, and mortality in people with and without prior stroke.
        Circ Cardiovasc Qual Outcomes. 2019; 12e004973

      Linked Article