generalised from one population to another. However, calibration measures how well the absolute values translate. Thus, a score that underestimates everyone’s risk by 50% would have good discrimination (all those who have an event will have a higher score than those who don’t have an event), but poor calibration, as the annual predicted event rate would be 50% of the real rate. Recalibration can be used to update an old score to contemporary event rates and risk factor levels.
National Institute for Health and Care Excellence guidelines previously recommended use of the UKPDS risk engine but now suggest QRISK2 [29]. The UKPDS risk engine was developed in the trial cohort which enrolled patients with recently diagnosed diabetes. The QRISK2 score is not diabetes-specific having been developed using a large primary care database utilising individual level medical records linked to hospital admissions data. The model has recently been updated (QRISK3) using data from almost 8 million adult patients in the derivation cohort and almost 3 million in the validation cohort [42]. Three models are available for use depending on the patient data available, with the most comprehensive including 18 predictor variables. The outcome is 10-year risk of coronary heart disease, ischaemic stroke or transient ischaemic attack. The score is designed to be built into the electronic medical record and give a computer-generated result. It performed well in the validation cohort with Harrell’s C statistics (similar to AUC) of 0.858–0.880 across the 3 models in men and women, suggesting good discrimination. On average, the score explains almost 60% of the variability in time to diagnosis of cardiovascular disease (R2). However, among patients with known type 2 diabetes, it only explained 25.2% of the variability in women and 22.9% in men. Independent external validation studies have been performed of the earlier QRISK2 score, showing similar results to the validation studies performed within the GP database [42]. There hasn’t been an intervention trial to confirm the score’s clinical impact.
Novel Biomarkers for Cardiovascular Disease Risk
Overall, risk prediction models based on traditional cardiovascular risk factors have limited ability to predict outcomes in patients with type 2 diabetes. These risk scores typically include minimal biomarker data, often just a lipid marker. More comprehensive scores may include albuminuria, HbA1c or eGFR. It has been hypothesised that novel biomarkers may exist which could improve cardiovascular risk prediction. Selection of candidate biomarkers has largely been hypothesis-driven, based on known pathophysiological pathways such as cardiac stress, inflammation, matrix remodelling and advanced glycation end products [43]. Numerous individual biomarkers have been shown to be associated with cardiovascular outcomes in patients with diabetes, although the strength of the association and degree of risk inferred is often limited. It has been proposed than inclusion of multiple biomarkers, relating to various pathophysiological mechanisms, may improve risk prediction models. The benefit of including biomarkers in a risk prediction tool can be evaluated by assessing discrimination, calibration and reclassification. Discrimination and calibration are discussed earlier in the chapter. Reclassification is the ability of a particular variable to improve an individuals’ calculated levels of risk beyond the defined thresholds such that their determined risk categories change, thus having an impact on the recommended clinical management.
Aside from established biomarkers, such as lipids, eGFR, urine albumin-creatinine ratio and HbA1c, current guidelines do not recommend the routine clinical use of any novel biomarkers [27, 28]. Some of the most well investigated novel biomarkers include N-terminal pro-B-type natriuretic peptide, high-sensitivity troponin T and high-sensitivity C-reactive protein (hsCRP) [43]. Independent associations with cardiovascular outcomes in patients with diabetes have been established for each of these [44–47]. N-terminal pro-B-type natriuretic peptide and high-sensitivity troponin T are both markers of cardiac stress and high-sensitivity C-reactive protein is a marker of inflammation.
A few studies have investigated the predicative capabilities of multiple novel biomarkers beyond established risk factors among patients with diabetes [43]. These studies have investigated between 23 and 237 different biomarkers with final models including between 3 and 10 biomarkers [48–50]. Each of these studies showed modest improvements in discrimination and reclassification with the addition of the novel biomarkers to a model based on traditional cardiovascular risk factors. However, it is not known whether these small differences in predictive capability would lead to any meaningful changes in clinical management or to prevention of subsequent cardiovascular outcomes, let alone whether the measurement of multiple biomarkers could be cost-effective.
Most studies to date have looked at biomarkers that relate to known pathophysiological mechanisms of cardiovascular disease. Thus there is a reasonable likelihood that these biomarkers will correlate with known risk factors and provide limited additional predictive capability. Recent developments in approaches to the discovery of novel biomarkers, such as proteomics and metabolomics, can assess hundreds or thousands of potential biomarkers simultaneously. It remains to be seen whether such studies could yield useful biomarkers for predicting cardiovascular disease in patients with diabetes.
Genetic Risk Scores for Cardiovascular Disease Prediction
The heritability of diabetes and cardiovascular disease risk is polygenic in the vast majority of cases. In recent years, genome-wide association studies conducted in extremely large cohorts have enabled the discovery of single nucleotide polymorphisms (SNPs) that are associated with diabetes and cardiovascular disease. Nevertheless, the degree of risk inferred by each of these polymorphisms in isolation is minimal. Genetic risk scores have been developed, combining multiple SNPs, and have been shown to have strong associations with cardiovascular disease [51]. These associations have been confirmed in a diabetes cohort for a 13-SNP and a 30-SNP risk score [52]. However, adding the genetic risk score results to models incorporating traditional cardiovascular risk factors did not have a significant impact on discrimination or reclassification. It may be that known clinical risk factors are on the causal pathway to cardiovascular disease and therefore the genetic risk scores add little in terms of predictive capability in those with established risk factors, such as most patients with diabetes.
Risk Prediction in the Era of Big Data and Machine-Learning
Increasingly large datasets are being utilised for the development of risk prediction tools due to advancing technology in a number of fields, such as electronic health records, genomics, proteomics and metabolomics. Such data have posed challenges to traditional regression-based methods for creating predictive models. Analytic challenges include assumptions of linear relationships between risk factors and outcomes, interactions between covariates and large numbers of predictor variables. Machine learning methods have been developed to aid in harnessing the potential predictive capabilities of so-called big data. Machine learning algorithms perform automated stochastic (random) or deterministic searches for the predictive model with optimal fit within certain specified constraints [53]. Typically, the resulting models are less useful for understanding associations or causal relationships than traditional regression analyses,
