CVD Risk Calculators: Worth the Effort?
Abstract & Commentary
By Michael H. Crawford, MD, Editor
Source: Allan GM, et al. Agreement among cardiovascular disease risk calculations. Circulation 2013;127:1948-1956.
Cardiovascular disease (CVD) risk calculators are frequently used by clinicians to guide patient management. Unfortunately, there are more than 100 risk calculators and smaller studies have shown insistencies between them. Thus, this group of investigators from Edmonton, Alberta, Canada, studied 25 calculators from eight mainly English-speaking countries that used a range of different databases, formats, and variables. All included age, sex, smoking, blood pressure, total cholesterol, and HDL cholesterol. Diabetes was included in 18. Hypothetical patients were used to create 128 unique test subjects to which all the calculators were applied. Since some of the calculators used 5-year risk and others 10-year risk, the subjects were divided into high, medium, and low risk. In nine calculators, 10-year absolute risk was provided and compared for each subject. Each calculator was used by two investigators and agreement in risk calculation was 95%. Half the disagreement was due to errors in the calculators and half from data entry errors. The 128 patients were categorized in a mean of 2.2 risk groups with 41% crossing all three categories. The pooled concordance in risk category assignment for all paired comparisons of calculators was 67%, less for diabetes subjects, 64% vs 73% for non-diabetics. Limiting the analysis to the 10-year risk calculators did not improve concordance, 70% overall. When only the nine Framingham-derived calculators were used, overall concordance was better at 89%. When the absolute 10-year risk calculators were compared, the average calculated risk was 4-5 times higher than the lowest risk calculated for the same subject. This effect was most pronounced in those with the highest calculated risk. The authors concluded that the demonstrated inconsistency between calculators is a clinically relevant limitation to their use.
This study shows that if you compare any two of 25 risk calculators, the assignment of the patient’s risk category (low, medium, high) will be different about one-third of the time. By sticking to calculators estimating 10-year Framingham risk, agreement increases to 89%. However, it is well known that the Framingham population is almost all Caucasians, so this degree of concordance may not be found with other populations. In fact, most of the calculators used in this study originated from English-speaking countries, so their concordance may be worse in other populations. The reasons for these discrepancies are not elucidated in the study, but likely come from different databases, different CVD endpoints, and the mathematical algorithms employed by each calculator. Using calculators that estimate absolute risk didn’t improve the agreement between calculators, nor did selecting the calculators that provided the risk of only hard endpoints such as myocardial infarction and cardiac death. This data is concerning because most guidelines recommend using risk calculators even though there is no evidence that their use improves outcomes.
Surveys of physicians show that 25-50% use risk calculators. Non-use reasons include lack of time, belief that the information is not helpful, belief that calculators oversimplify things, and belief that they can accurately predict risk subjectively. One study evaluating the accuracy of physicians’ subjective assessment vs the Framingham database showed that physicians were 71% accurate; four risk calculators came in at 66-81%, which supports physicians’ belief in themselves. I am one of the non-believers and don’t use risk calculators for all the reasons above, plus some patients fixate on their score.
One limitation of this study is that the hypothetical patients were higher risk than most primary care populations and calculated risk concordance decreased with higher risk patients. One could argue that once a patient is categorized as high risk, wide variability among calculators is moot. However, of the 28 subjects in this study who were assigned the same risk category by all 25 calculators, 79% were high risk. So, it would seem that most of the variability was actually in the low- and moderate-risk subjects. Other studies have suggested that most risk calculators overestimate risk in primary care populations. So if you are going to use risk calculators, the authors suggest that you use one targeted to your patient population and keep in mind that the results are a rough estimate.