In the current study, we attempted to develop tools for identifying individuals with an elevated probability of UMI as determined by ECG among those without a self-reported history of MI. Such a tool would allow targeting of ECG screening for UMI to higher-risk individuals. However, neither the tool built using potential correlates routinely collected nor the expanded tool which considered a large battery of variables had a c-statistic greater than 0.7, a widely used guideline for adequate discrimination [14
]. Furthermore, the test characteristics were such that a reasonably high sensitivity required a very low specificity.
Previous studies have found that coronary heart disease risk factors such as older age, male sex, diabetes, hypertension, dyslipidemia, and smoking are associated with UMI [15
]. While the traditional coronary heart disease risk factors were associated with prevalent UMI in the REGARDS study population, they did not discriminate well between participants with and without UMI. For incident coronary heart disease, the Framingham risk score, which incorporates many of the risk factors and is recommended for use in the general United States population without known cardiovascular disease, has been shown to have a c-statistic greater than 0.7 in many, but not all, populations [16
]. In the current study, the c-statistic for UMI associated with the Framingham risk score was 0.587. The Framingham Risk Score is a validated risk score for incident coronary heart disease. Its use for prevalent UMI is unvalidated. The AUC of FRS in lower than expected but this should be interpreted with caution. The difference in the ability to discriminate between people who will develop coronary heart disease and those who will not and the ability to discriminate between people with prevalent UMI and those without prevalent UMI may be because of the variety of processes that are involved in having UMI. An ischemic injury causing Q-waves must occur but go unrecognized by the individual or the individual’s health care providers or both. Then factors related to survival with MI come into play in order to become a member of the REGARDS study. Finally, ECG abnormalities are known to resolve over time in some people [17
]; the participants in this study considered to have UMI were only those with persistent ECG indications of MI. The complicated pathway to UMI may hinder identification of populations with high prevalence of UMI.
The current study focused on the c-statistic as the metric of discrimination for the UMI assessment tools. C-statistics are known to be higher in the population in which they are derived than in independent populations, and therefore it is customary to validate the assessment tools in external populations. In the current study, this step was not taken because of the poor performance of the tools in the derivation cohort. The difference in c-statistics between the UMI assessment tools, which were developed in the REGARDS study, and the Framingham risk score for coronary heart disease would likely have been smaller in an independent population. We did not examine calibration metrics in this study because, in the absence of acceptable discrimination, even a very well-calibrated assessment tool would not add important information to the decision of whether or not to screen for UMI.
The major strengths of this study include the large, diverse REGARDS study population and the wide variety of potential correlates which have been measured. There are also limitations of this study. We used ECG criteria for defining UMI, though Q-waves on ECG are absent in some patients with documented MI [17
], may be due to conditions other than MI [18
], and do not agree well with myocardial scars detected on cardiac magnetic resonance imaging with gadolinium contrast [19
]. The inability to detect UMI not associated with Q-waves is an inherent limitation of the ECG method for detection of UMI; however, the use of the MC scheme reduces the concern about Q-waves due to other conditions which take precedence over MI in the coding system [12
]. We believe that ECG would be more appropriate for screening in a general population than cardiac magnetic resonance imaging with gadolinium contrast because of the greater availability, lower cost, and time needed for ECGs, and the potential for serious adverse reactions to gadolinium contrast.
We relied on self-report to identify individuals with recognized MI and history of coronary heart disease. Previous studies have shown that some participants report MI when there is no clinical record of MI, and others may report no history of MI when a diagnosis of MI has been documented [20
]. Some of the participants in the REGARDS study population who we considered to have UMI may have a known MI from the perspective of their healthcare providers. While the patient-centered approach we took was designed to simulate patient encounters with physicians, it is possible that lack of accuracy in self-report of MI makes an UMI assessment tool impractical in the clinical setting. Many of the potential correlates of UMI were also self-reported. Some of the correlates are by definition self-reported, such as self-rated health, perceived stress, and depression symptoms, but different approaches may be more reliable for other covariates, such as prior medical history. In these cases, reliance on self-reported data could lead to increased error and an underestimation of the ability of these factors to discriminate between individuals with and without UMI. Finally, ECG, blood pressure, and biomarkers were measured during a single in-home visit and are subject to measurement error which could also lead to underestimation of the ability to discriminate between individuals with and without UMI. However, these measurements were made using a research protocol, which may be more reliable than measures obtained in routine clinical practice; this may result in worse performance of the tool in clinical care settings than we report here.