Blood pressure variability, likelihood ratios, and other necessary considerations when assessing self‐reported hypertension status

To the Editor, Self-reported data on medical conditions like hypertension drive many research studies and related public health efforts. Although it is known that self-reported data are subject to recall bias, the true complexity of the issue is only now unfolding. We read with great pleasure the article by Wellman et al, titled “Accuracy of self-reported hypertension: Effect of age, gender, and history of alcohol dependence”, which found that self-reported hypertension diagnosis was highly specific in identifying those without hypertension, but not sensitive in identifying those with hypertension.1 We would like to provide key statistical and methodological considerations that place conclusions into context as well as offer insight into future studies to fully uncover this issue. Firstly, although sensitivity and specificity are important measures of the discriminatory power of a test (eg, self-report), they do not provide useful information to a clinician who seeks to determine from self-report whether a patient truly has hypertension. Predictive values are also not ideal or generalizable as they depend highly on the prevalence of disease (eg, hypertension) in the study population. As this study's hypertension prevalence was low at 9.9%, use of predictive values may underestimate positive predictive value (PPV) and overestimate negative predictive value (NPV). Instead, clinicians use positive (LR+) and negative likelihood ratios (LR-) which provide information on post-test probability of disease. In this study, the LR+ (sensitivity/[1-specificity]) was approximately 7 while the LR- was 0.66, suggesting that for a person with a pre-test probability of, for example, 10%, and self-reporting as having hypertension, the post-test probability would be 44%. For someone with a pre-test probability equivalent to the prevalence of hypertension in the United States (29%2), the post-test probability would be 74%. These results may or may not be clinically useful as there is still a sizeable chance that the patient may not have hypertension. Additionally, calculations of sensitivity and specificity were done using all participants, including those taking anti-hypertensive medications. Those with self-reported hypertension who are appropriately taking medications may have normotensive BP; this would incorrectly be considered a discordance and potentially underestimate the LR+. Secondly, while blood pressure (BP) was measured in a standardized manner, there is risk for misclassification of one's BP due to physiological BP variability3 (eg, situational variability, emotional state,4 and bladder fullness), pathological BP variability (eg, autonomic dysregulation,5 kidney disease,6 and cardiovascular disease7), artificial BP variability (eg, measurement technique,3 time between self-reported status and BP reading), and/or presence of white coat hypertension or masked hypertension.8 It may be worthwhile to determine whether sensitivity and specificity change upon adjusting for the second and third BP readings and the standard deviation of all BP readings. Jaubert et al found that increased alcohol consumption was associated with higher daytime BP variability,9 which may be reflected in the reported differences in BP readings in this study. Thirdly, 27 (63%) of the 43 hypertensive participants said “No” to self-report, suggesting that they have either uncontrolled or undiagnosed hypertension. This is similar to the 52% prevalence of uncontrolled hypertension (among hypertensive participants) most recently reported by Fryar et al2 Prevalence of uncontrolled/undiagnosed hypertension in alcohol dependent vs. not alcohol dependent hypertensive participants was similar (63.6% vs. 62.7%), leading us to believe that alcohol dependence is not correlated with BP control, despite it being a risk factor. Overall, we believe the authors make a strong case for the accuracy of self-reported hypertension status in normotensive participants. Future studies with larger sample size of hypertensive participants and control for sources of BP variability are needed to fully characterize use of self-reported hypertension. The authors declare that they have no competing interests. This study was exempt from Rutgers’s Institutional Review Board approval.