Developing and validating a cardiovascular risk score for patients in the community with prior cardiovascular disease

Development and validation of cardiovascular risk prediction equations in 76 000 people with known cardiovascular disease

AIMS

Multiple health administrative databases can be individually linked in Aotearoa New Zealand, using encrypted identifiers. These databases were used to develop cardiovascular risk prediction equations for patients with known cardiovascular disease (CVD).

METHODS AND RESULTS

Administrative health databases were linked to identify all people aged 18-84 years with known CVD, living in Auckland and Northland, Aotearoa New Zealand, on 1 January 2014. The cohort was followed until study outcome, death, or 5 years. The study outcome was death or hospitalization due to ischaemic heart disease, stroke, heart failure, or peripheral vascular disease. Sex-specific 5-year CVD risk prediction equations were developed using multivariable Fine and Gray models. A total of 43 862 men {median age: 67 years [interquartile range (IQR): 59-75]} and 32 724 women [median age: 70 years (IQR: 60-77)] had 14 252 and 9551 cardiovascular events, respectively. Equations were well calibrated with good discrimination. Increasing age and deprivation, recent cardiovascular hospitalization, Mori ethnicity, smoking history, heart failure, diabetes, chronic renal disease, atrial fibrillation, use of blood pressure lowering and anti-thrombotic drugs, haemoglobin A1c, total cholesterol/HDL cholesterol, and creatinine were statistically significant independent predictors of the study outcome. Fourteen per cent of men and 23% of women had predicted 5-year cardiovascular risk <15%, while 28 and 24% had ≥40% risk.

CONCLUSION

Robust cardiovascular risk prediction equations were developed from linked routine health databases, a currently underutilized resource worldwide. The marked heterogeneity demonstrated in predicted risk suggests that preventive therapy in people with known CVD would be better informed by risk stratification beyond a one-size-fits-all high-risk categorization.

Dual versus single long-acting bronchodilator use could raise acute coronary syndrome risk by over 50%: A population-based nested case-control study

BACKGROUND

Coronary heart disease occurs more frequently among patients with chronic obstructive pulmonary disease (COPD) compared to those without COPD. While some research suggests that long-acting bronchodilators might confer an additional risk of acute coronary syndrome (ACS), information from real-world clinical practice about the cardiovascular impact of using two versus one long-acting bronchodilator for COPD is limited. We undertook a population-based nested case-control study to estimate the risk of ACS in users of both a long-acting muscarinic antagonist (LAMA) and a long-acting beta2-agonist (LABA) relative to users of a LAMA.

METHODS

The study was based on the primary care PREDICT Cardiovascular Disease Cohort and linked data from regional laboratories and the New Zealand Ministry of Health's national data collections. The underlying cohort (n = 29,993) comprised patients aged 45-84 years, who initiated treatment with a LAMA and/or LABA for COPD between 1 February 2006 and 11 October 2016. 1490 ACS cases were matched to 13,550 controls by date of birth, sex, date of cohort entry (first long-acting bronchodilator dispensing), and COPD severity.

RESULTS

Relative to current use of LAMA therapy, current use of LAMA and LABA dual therapy was associated with a significantly higher risk of ACS (adjusted OR = 1.72; [95% CI: 1.28-2.31]).

CONCLUSION

Dual long-acting bronchodilator therapy, rather than LAMA mono-therapy, could increase the risk of ACS by more than 50%. This has important implications for decisions about the potential benefit/harm ratio of COPD treatment intensification, given the modest benefits of dual therapy.

Is the SMART risk prediction model ready for real-world implementation? A validation study in a routine care setting of approximately 380 000 individuals

AIMS

Reliably quantifying event rates in secondary prevention could aid clinical decision-making, including quantifying potential risk reductions of novel, and sometimes expensive, add-on therapies. We aimed to assess whether the SMART risk prediction model performs well in a real-world setting.

METHODS AND RESULTS

We conducted a historical open cohort study using UK primary care data from the Clinical Practice Research Datalink (2000-2017) diagnosed with coronary, cerebrovascular, peripheral, and/or aortic atherosclerotic cardiovascular disease (ASCVD). Analyses were undertaken separately for cohorts with established (≥6 months) vs. newly diagnosed ASCVD. The outcome was first post-cohort entry occurrence of myocardial infarction, stroke, or cardiovascular death. Among the cohort with established ASCVD [n = 244 578, 62.1% male, median age 67.3 years, interquartile range (IQR) 59.2-74.0], the calibration and discrimination achieved by the SMART model was not dissimilar to performance at internal validation [Harrell's c-statistic = 0.639, 95% confidence interval (CI) 0.636-0.642, compared with 0.675, 0.642-0.708]. Decision curve analysis indicated that the model outperformed treat all and treat none strategies in the clinically relevant 20-60% predicted risk range. Consistent findings were observed in sensitivity analyses, including complete case analysis (n = 182 482; c = 0.624, 95% CI 0.620-0.627). Among the cohort with newly diagnosed ASCVD (n = 136 445; 61.0% male; median age 66.0 years, IQR 57.7-73.2), model performance was weaker with more exaggerated risk under-prediction and a c-statistic of 0.559, 95% CI 0.556-0.562.

CONCLUSIONS

The performance of the SMART model in this validation cohort demonstrates its potential utility in routine healthcare settings in guiding both population and individual-level decision-making for secondary prevention patients.

The Multi-Ethnic New Zealand Study of Acute Coronary Syndromes (MENZACS): Design and Methodology

Background. Each year, approximately 5000 New Zealanders are admitted to hospital with first-time acute coronary syndrome (ACS). The Multi-Ethnic New Zealand Study of Acute Coronary Syndromes (MENZACS) is a prospective longitudinal cohort study embedded within the All New Zealand Acute Coronary Syndrome Quality Improvement (ANZACS-QI) registry in six hospitals. The objective of MENZACS is to examine the relationship between clinical, genomic, and cardiometabolic markers in relation to presentation and outcomes post-ACS. Methods. Patients with first-time ACS are enrolled and study-specific research data is collected alongside the ANZACS-QI registry. The research blood samples are stored for future genetic/biomarker assays. Dietary information is collected with a food frequency questionnaire and information about physical activity, smoking, and stress is also collected via questionnaire. Detailed family history, ancestry, and ethnicity data are recorded on all participants. Results. During the period between 2015 and 2019, there were 2015 patients enrolled. The mean age was 61 years, with 60% of patients aged <65 years and 21% were female. Ethnicity and cardiovascular (CV) risk factor distribution was similar to ANZACS-QI: 13% Māori, 5% Pacific, 5% Indian, and 74% NZ European. In terms of CV risk factors, 56% were ex-/current smokers, 42% had hypertension, and 19% had diabetes. ACS subtype was ST elevation myocardial infarction (STEMI) in 41%, non-ST elevation myocardial infarction (NSTEM) in 54%, and unstable angina in 5%. Ninety-nine percent of MENZACS participants underwent coronary angiography and 90% had revascularization; there were high rates of prescription of secondary prevention medications upon discharge from hospital. Conclusion. MENZACS represents a cohort with optimal contemporary management and will be a significant epidemiological bioresource for the study of environmental and genetic factors contributing to ACS in New Zealand’s multi-ethnic environment. The study will utilise clinical, nutritional, lifestyle, genomic, and biomarker analyses to explore factors influencing the progression of coronary disease and develop risk prediction models for health outcomes.

Obesity and All Cause Mortality Following Acute Coronary Syndrome (ANZACS QI 53)

BACKGROUND

Some studies have suggested a lower mortality in obese subjects with cardiovascular disease. The aim of this study was to evaluate the relationship between body mass index (BMI) and outcomes in patients with acute coronary syndrome (ACS).

METHODS

The study included 13,742 patients undergoing coronary angiography for ACS between 2012 and 2016 from the All New Zealand Acute Coronary Syndrome-Quality Improvement (ANZACS-QI) registry. Patients were categorised by BMI (kg/m²) as: underweight <18.5, normal 18.5 to <25, overweight 25 to <30, mildly obese 30 to <35, moderately obese 35 to <40, and severely obese ≥40. The primary endpoint of the study was all cause mortality with secondary endpoints of cardiovascular disease (CVD) and non-CVD mortality within 4 years of discharge.

RESULTS

Unadjusted all cause mortality was lowest in the mildy obese but no different to normal or overweight after adjustment for multiple confounders. Adjusted all cause mortality was higher in the moderately (hazard ratio [HR] 1.39, 95% CI: 1.10-1.75) and severely obese (2.06, 95% CI: 1.57-2.70) compared to the mildly obese. Non-CVD mortality (HR 1.58, 95% CI: 1.12-2.23) was the major contributor to higher all cause mortality in moderately obese patients. Both CVD mortality (HR 2.36, 95% CI: 1.67-3.32) and non-CVD mortality (HR 1.67, 95% CI: 1.07-2.61) contributed to higher all cause mortality in the severely obese.

CONCLUSIONS

Moderate and severe obesity is associated with worse survival post ACS influenced by higher non-CVD mortality in moderate/severe obesity and higher CVD mortality in severe obesity.

Use of Electronic Health Data for Disease Prediction: A Comprehensive Literature Review

Disease prediction has the potential to benefit stakeholders such as the government and health insurance companies. It can identify patients at risk of disease or health conditions. Clinicians can then take appropriate measures to avoid or minimize the risk and in turn, improve quality of care and avoid potential hospital admissions. Due to the recent advancement of tools and techniques for data analytics, disease risk prediction can leverage large amounts of semantic information, such as demographics, clinical diagnosis and measurements, health behaviours, laboratory results, prescriptions and care utilisation. In this regard, electronic health data can be a potential choice for developing disease prediction models. A significant number of such disease prediction models have been proposed in the literature over time utilizing large-scale electronic health databases, different methods, and healthcare variables. The goal of this comprehensive literature review was to discuss different risk prediction models that have been proposed based on electronic health data. Search terms were designed to find relevant research articles that utilized electronic health data to predict disease risks. Online scholarly databases were searched to retrieve results, which were then reviewed and compared in terms of the method used, disease type, and prediction accuracy. This paper provides a comprehensive review of the use of electronic health data for risk prediction models. A comparison of the results from different techniques for three frequently modelled diseases using electronic health data was also discussed in this study. In addition, the advantages and disadvantages of different risk prediction models, as well as their performance, were presented. Electronic health data have been widely used for disease prediction. A few modelling approaches show very high accuracy in predicting different diseases using such data. These modelling approaches have been used to inform the clinical decision process to achieve better outcomes.

An Exploratory Study of the Use of the Electronic Health Records of Hypertensive Patients to Support the Primary Prevention of Stroke in Shanghai

Background

The value of identifying and targeting population demographics at high risk of stroke based on patient-reported outcomes (PROs) with electronic health records (EHRs) in Shanghai is largely undiscovered.

Aim

To test the hypothesis that establishing an evidence-based support system composed of PROs integrated with EHRs could be effective at identifying individuals at high risk of suffering from stroke.

Methods

The patients included in this study joined the hypertensive patient management system from 2014 to 2018. We merged the Hypertension Patients Management Database and the Diabetes Mellitus Patients Management Database of Shanghai Jiading district, then kept the hypertension patients with or without diabetes. We subsequently performed a screen analysis utilizing EHRs to target the population with any risk factor for stroke, namely, hypertension, diabetes mellitus, obesity, smoking and physical inactivity. We also calculated the distribution of each risk factor and the combinations of risk factors.

Results

In the Jiading District of Shanghai, 46,580 hypertensive patients with complete baseline information joined the hypertensive patient management system from 2014 to 2018. The majority of the patients were aged above 60 years old. Physical inactivity (83.24%), smoking (24.07%), diabetes (16.87%), and obesity (12.23%) were highly prevalent in hypertensive participants. Approximately 4377 patients were diagnosed with hypertension exclusively, accounting for 9.70% of the total number of patients in this study. Meanwhile, approximately 52.47% of the patients were diagnosed with two concurrent risk factors, and 38.13% of the patients had hypertension, meaning that 17,762 patients could be labeled as the high-risk population for stroke according to the criteria established by the National Stroke Screening Survey.

Conclusion

Our exploratory findings demonstrate the feasibility of pinpointing and targeting populations at high risk of stroke using the EHRs of hypertensive patients.

Predicting cardiovascular disease risk across the atherosclerotic disease continuum

AIMS

Cardiovascular disease (CVD) guidelines dichotomize populations into primary and secondary prevention. We sought to develop a risk equation for secondary prevention of CVD that complements existing equations for primary prevention of CVD, and to describe the distributions of CVD risk across the population.

METHODS AND RESULTS

Adults aged 30-79 years who had routine CVD risk assessment in 2007-16 were identified from a large primary care cohort (PREDICT) with linkage to national and regional datasets. The 5-year risk of developing CVD among people without atherosclerotic CVD (ASCVD) was calculated using published equations (PREDICT-1°). A new risk equation (PREDICT-2°) was developed from Cox regression models to estimate the 5-year risk of CVD event recurrence among patients with known ASCVD. The outcome for both equations was hospitalization for a CVD event or cardiovascular death. Of the 475 161 patients, 12% (57 061) had ASCVD. For those without ASCVD, median (interquartile range) 5-year risks with the PREDICT-1° score were women 2.2% (1.2-4.2%), men 3.5% (2.0-6.6%), and whole group 2.9% (1.6-5.5%). For those with ASCVD, the 5-year risks with the new PREDICT-2° equation were women 21% (15-33%), men 23% (16-35%), and whole group 22% (16-34%).

CONCLUSION

We developed CVD risk scores for people with ASCVD (PREDICT-2°) to complement the PREDICT-1° scores. Median CVD risk is eight-fold higher among those with ASCVD than those without; however, there was overlap and the widest distribution of CVD risk was among people with ASCVD. This study describes a CVD risk continuum and the limitations of a 'one size fits all' approach to assessing risk in people with ASCVD.

Value of Transverse Groove on the Earlobe and Hair Growth on the Ear to Predict the Risk for Coronary Artery Disease and Its Severity among Iranian Population, in Tehran City

Background:

The use of phenotypic parameters along with other noninvasive diagnostic modality can lead to early diagnosis of coronary artery disease (CAD) and prevent its life-threatening outcome. Recently, the application of head and face components for assessing the risk for CAD much attention has been paid. The present study aimed to assess the relationship between ear characteristics (transverse groove on the earlobe and hair growth on the ear) and the risk for CAD and its severity among Iranian patients.

Materials and Methods:

In this cross-sectional study, the study population consisted of 105 consecutive patients with suspected CAD undergoing coronary angiography. The severity of CAD was determined by the number of disease vessels as well as the presence of left main lesions assessed by coronary angiography. All patients were examined to evaluate the appearance of ear regarding the presence of transverse groove on the earlobe and hair growth on the ear.

Results:

Comparing cardiovascular parameters across the groups with and without transverse groove on the earlobe showed a higher rate of CAD as well as the higher number of involved coronary arteries than in the groups without transverse groove on the earlobe. Similarly, the presence of CAD and its higher severity were more revealed in patients with hair growth on the ear as compared to the group without this phenotype. According to multivariable logistic regression analysis and with the presence of baseline parameters, the presence of transverse groove on the earlobe and hair growth on the ear increased the risk for CAD by 2.4 and 4.4 fold, respectively.

Conclusion:

Along with classic cardiovascular risk factors, the role of growing hair on the ear and transverse groove on the ear to predict high risk for CAD should be considered.

Cardiovascular risk in mild to moderately decreased glomerular filtration rate, diabetes and coronary heart disease in a southern European region

INTRODUCTION AND OBJECTIVES

Individuals with mild to moderately decreased estimated glomerular filtration rate (eGFR=30-59 mL/min/1.73 m²) are considered at high risk of cardiovascular disease (CVD). No studies have compared this risk in eGFR=30-59, diabetes mellitus (DM), and coronary heart disease (CHD) in regions with a low incidence of CHD.

METHODS

We performed a retrospective cohort study of 122 443 individuals aged 60-84 years from a region with a low CHD incidence with creatinine measured between January 1, 2010 and December 31, 2011. We identified hospital admissions due to CHD (myocardial infarction, angina) or CVD (CHD, stroke, or transient ischemic attack) from electronic medical records up to December 31, 2013. We estimated incidence rates and Cox regression adjusted subdistribution hazard ratio (sHR) including competing risks in patients with eGFR=30-59, DM and CHD, or combinations, compared with individuals without these diseases.

RESULTS

The median follow-up was 38.3 [IQR, 33.8-42.7] months. Adjusted sHR for CHD in individuals with eGFR=30-59, DM, eGFR=30-59 plus DM, previous CHD, CHD plus DM, and CHD plus eGFR=30-59 plus DM, were 1.34 (95%CI, 1.04-1.74), 1.61 (95%CI, 1.36-1.90), 1.96 (95%CI, 1.42-2.70), 4.33 (95%CI, 3.58-5.25), 7.05 (5.80-8.58) and 7.72 (5.72-10.41), respectively. The corresponding sHR for CVD were 1.25 (95%CI, 1.06-1.46), 1.56 (95%CI, 1.41-1.74), 1.83 (95%CI, 1.50-2.23), 2.86 (95%CI, 2.48-3.29), 4.54 (95%CI, 3.93-5.24), and 5.33 (95%CI, 4.31-6.60).

CONCLUSIONS

In 60- to 84-year-olds with eGFR=30-59, similarly to DM, the likelihood of being admitted to hospital for CHD and CVD was about half that of individuals with established CHD. Thus, eGFR=30-59 does not appear to be a coronary-risk equivalent. Individuals with CHD and DM, or eGFR=30-59 plus DM, should be prioritized for more intensive risk management.

Development and validation of a cardiovascular risk score for patients in the community after acute coronary syndrome

OBJECTIVE

Following acute coronary syndrome (ACS), patients are managed long-term in the community, yet few tools are available to guide patient-clinician communication about risk management in that setting. We developed a score for predicting cardiovascular disease (CVD) risk among patients managed in the community after ACS.

METHODS

Adults aged 30-79 years with prior ACS were identified from a New Zealand primary care CVD risk management database (PREDICT) with linkage to national mortality, hospitalisation, pharmaceutical dispensing and regional laboratory data. A Cox model incorporating clinically relevant factors was developed to estimate the time to a subsequent fatal or non-fatal CVD event and transformed into a 5-year risk score. External validation was performed in patients (Coronary Disease Cohort Study) assessed 4 months post-ACS.

RESULTS

The PREDICT-ACS cohort included 13 703 patients with prior hospitalisation for ACS (median 1.9 years prior), 69% men, 58% European, median age 63 years, who experienced 3142 CVD events in the subsequent 5 years. Median estimated 5 year CVD risk was 24% (IQR 17%-35%). The validation cohort consisted of 2014 patients, 72% men, 92% European, median age 67 years, with 712 CVD events in the subsequent 5 years. Median estimated 5-year risk was 33% (IQR 24%-51%). The risk score was well calibrated in the derivation and validation cohorts, and Harrell's c-statistic was 0.69 and 0.68, respectively.

CONCLUSIONS

The PREDICT-ACS risk score uses data routinely available in community care to predict the risk of recurrent clinical events. It was derived and validated in real-world contemporary populations and can inform management decisions with patients living in the community after experiencing an ACS.

Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes

When designing a study to develop a new prediction model with binary or time-to-event outcomes, researchers should ensure their sample size is adequate in terms of the number of participants (n) and outcome events (E) relative to the number of predictor parameters (p) considered for inclusion. We propose that the minimum values of n and E (and subsequently the minimum number of events per predictor parameter, EPP) should be calculated to meet the following three criteria: (i) small optimism in predictor effect estimates as defined by a global shrinkage factor of ≥0.9, (ii) small absolute difference of ≤ 0.05 in the model's apparent and adjusted Nagelkerke's R2 , and (iii) precise estimation of the overall risk in the population. Criteria (i) and (ii) aim to reduce overfitting conditional on a chosen p, and require prespecification of the model's anticipated Cox-Snell R2 , which we show can be obtained from previous studies. The values of n and E that meet all three criteria provides the minimum sample size required for model development. Upon application of our approach, a new diagnostic model for Chagas disease requires an EPP of at least 4.8 and a new prognostic model for recurrent venous thromboembolism requires an EPP of at least 23. This reinforces why rules of thumb (eg, 10 EPP) should be avoided. Researchers might additionally ensure the sample size gives precise estimates of key predictor effects; this is especially important when key categorical predictors have few events in some categories, as this may substantially increase the numbers required.

The Evolving Cardiovascular Disease Risk Scores for Persons with Diabetes Mellitus

PURPOSE OF REVIEW

We briefly introduce the concept and use of cardiovascular disease (CVD) risk scores and review the methodology for CVD risk score development and validation in patients with diabetes. We also discuss CVD risk scores for diabetic patients that have been developed in different countries.

RECENT FINDINGS

Patients with diabetes have a gradient of CVD risk that needs to be accurately assessed. Numerous CVD risk scores for diabetic patients have been created in various settings. The methods to develop risk scores are highly diverse and each choice has its own pros and cons. A well-constructed risk score for diabetic patients may be advocated by guidelines and adopted by healthcare providers to help determine preventive strategies. New risk factors are being investigated in order to improve the predictive accuracy of current risk scores. A suitable CVD risk score for the diabetes population should be accurate, low-cost, and beneficial to outcome. While the performance (accuracy) has all been internally validated, validation on external populations is still needed. Cost-effectiveness and clinical trials demonstrating improvement in outcomes are limited and should be the target of future research.

Outcome after myocardial infarction without obstructive coronary artery disease

OBJECTIVE

The medium-term outcome and cause of death in patients with myocardial infarction with non-obstructive coronary arteries (MINOCA) is not well characterised. The aim of this study was to compare mortality and rates of recurrent events in post myocardial infarction (MI) patients with obstructive coronary artery disease (CAD) and in patients with MINOCA compared with an age and sex-matched cohort without cardiovascular disease (CVD).

METHODS

We performed a national cohort study of consecutive patients undergoing coronary angiography for MI during 2 years between 2013 and 2015 from the All New Zealand Acute Coronary Syndrome-Quality Improvement (ANZACS QI) registry. MI patient registry data were linked anonymously to national hospitalisation and mortality records. Age and sex matched patients without known CVD formed the comparison group.

RESULTS

Of the 8305 patients with MI, 897 (10.8%) were classified as MINOCA. Compared with those without known CVD, the adjusted HRs for the primary outcome (all-cause death or recurrent non-fatal MI) were 7.81 (95% CI 6.64 to 9.19, p<0.0001) in those with obstructive CAD and 4.64 (95% CI 3.54 to 6.10, p<0.0001) in those with MINOCA. Kaplan-Meier all-cause mortality at 2 years was 7.9% for those with obstructive CAD, with nearly half being CVD deaths (3.6% CVD deaths and 4.5% non-CVD deaths, respectively). In contrast, MINOCA all-cause mortality was 4.9% with non-CVD death (4.5%) predominating.

CONCLUSIONS

MINOCA is common and has an adverse outcome rate approximately half than that of those with obstructive CAD. The predominant contributor to mortality is non-CVD death. The rate of events in MINOCA is significantly greater than the population without CVD.

Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease

Prognostic modelling is important in clinical practice and epidemiology for patient management and research. Electronic health records (EHR) provide large quantities of data for such models, but conventional epidemiological approaches require significant researcher time to implement. Expert selection of variables, fine-tuning of variable transformations and interactions, and imputing missing values are time-consuming and could bias subsequent analysis, particularly given that missingness in EHR is both high, and may carry meaning. Using a cohort of 80,000 patients from the CALIBER programme, we compared traditional modelling and machine-learning approaches in EHR. First, we used Cox models and random survival forests with and without imputation on 27 expert-selected, preprocessed variables to predict all-cause mortality. We then used Cox models, random forests and elastic net regression on an extended dataset with 586 variables to build prognostic models and identify novel prognostic factors without prior expert input. We observed that data-driven models used on an extended dataset can outperform conventional models for prognosis, without data preprocessing or imputing missing values. An elastic net Cox regression based with 586 unimputed variables with continuous values discretised achieved a C-index of 0.801 (bootstrapped 95% CI 0.799 to 0.802), compared to 0.793 (0.791 to 0.794) for a traditional Cox model comprising 27 expert-selected variables with imputation for missing values. We also found that data-driven models allow identification of novel prognostic variables; that the absence of values for particular variables carries meaning, and can have significant implications for prognosis; and that variables often have a nonlinear association with mortality, which discretised Cox models and random forests can elucidate. This demonstrates that machine-learning approaches applied to raw EHR data can be used to build models for use in research and clinical practice, and identify novel predictive variables and their effects to inform future research.

Promises and pitfalls of electronic health record analysis

Routinely collected electronic health records (EHRs) are increasingly used for research. With their use comes the opportunity for large-scale, high-quality studies that can address questions not easily answered by randomised clinical trials or classical cohort studies involving bespoke data collection. However, the use of EHRs generates challenges in terms of ensuring methodological rigour, a potential problem when studying complex chronic diseases such as diabetes. This review describes the promises and potential of EHRs in the context of diabetes research and outlines key areas for caution with examples. We consider the difficulties in identifying and classifying diabetes patients, in distinguishing between prevalent and incident cases and in dealing with the complexities of diabetes progression and treatment. We also discuss the dangers of introducing time-related biases and describe the problems of inconsistent data recording, missing data and confounding. Throughout, we provide practical recommendations for good practice in conducting EHR studies and interpreting their results.