Key result
Adding risk-enhancing factors to the pooled cohort equation may paradoxically produce a loss of accuracy and lack of fidelity across the risk prediction process.
Adding risk-enhancing factors to standard risk prediction models like the pooled cohort equation may paradoxically reduce accuracy and primarily results in clinically meaningless reclassification.
College of Cardiology/American Heart Association cholesterol guidelines introduced the pooled cohort equation (PCE) for risk assessment.Although a major advancement, the accuracy of the PCE and the omission of emerging risk factors (eg, hsCRP [high-sensitivity C-reactive protein]) have been concerns.The recent American Heart Association/American College of Cardiology Special Report 1 recommends initiating risk assessment using the PCE, but then considering risk-enhancing factors (REFs) and coronary artery calcium (CAC) in some individuals, as introduced in the latest American College of Cardiology/American Heart Association cholesterol 2 and primary prevention guidelines.3 The REFs include family history, metabolic syndrome, chronic kidney disease, chronic inflammatory conditions, premature menopause, preeclampsia, high-risk race/ethnicities, and hypertriglyceridemia. Several blood tests were also listed: hsCRP, lipoprotein(a), and apoprotein-B and ankle brachial index.The report explains that the PCE provides estimates of group means and that individual's risks are distributed around these means, but it is difficult to determine how much a REF or comorbidity may change the 10-year risk estimate quantitatively for an individual patient.It reviews factors leading to miscalibration, either underprediction or overprediction, and concludes the presence of REFs can help to overcome most issues of miscalibration for any risk prediction equations and help to reclassify risk appropriately.Here, we provide a context for better understanding risk models and show that REFs introduce new concerns including a lack of fidelity across the risk prediction process and can (paradoxically) produce a loss of accuracy.Epidemiological methods, such as the PCE, relate outcomes to risk factors and allow the identification of subgroups differing in risk (ie, population risk stratification).The only reasonable criterion of accuracy/calibration for any such model is whether the observed outcomes equal the predicted outcomes for a risk subgroup.If so, the model is accurate/calibrated and appropriate for clinical use (at least in similar populations).The potential benefit of adding risk factors to the PCE is not a correction of miscalibration, but improved discrimination (ie, a more disperse risk distribution curve with a higher receiver operator curve area-under-the-curve or C statistic).Also, models including different risk factors commonly place an individual in a very different risk subgroup.4 Regardless, both models may be accurate and have a similar receiver operator curve area-under-the-curve.If the continuous distribution of predicted risk is divided into categories, these differences may result in a reclassification.This merely indicates that different risk factors produce different population risk stratifications, an example of the reference class problem.Reclassification of a patient is not a clinically meaningful event, even if one model includes additional risk factors.It simply represents a re-shuffling of individuals (typically those near the cut points) up and down risk categories, whereas the overall population risk distribution typically differs little among models.Since we are
No takes yet. Share an insight, caveat, or question.
Stern et al. (2019) conducted a review in Cardiovascular risk assessment. Risk-enhancing factors (REFs) vs. Pooled cohort equation (PCE) alone was evaluated on Risk prediction accuracy and calibration. Adding risk-enhancing factors to the pooled cohort equation may paradoxically produce a loss of accuracy and lack of fidelity across the risk prediction process.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: