Difficulties in diagnosing hypertension: implications and alternatives.

OBJECTIVE: To estimate the magnitude of misclassification rates with commonly used algorithms for the detection of hypertensives and to suggest a sequential approach to screening.
DESIGN: A conventional statistical model was used with several different algorithms to determine the number and types of errors made in categorizing two different populations, a general population sample and a population with a high risk of hypertension.
METHODS: The calculations were made for single-visit screens, similar to those used in epidemiologic studies, for three-visit screens commonly used in clinical practice and clinical trials for cutoff points of 85, 95 and 105 mmHg. A sequential probability ratio screen was proposed and the error rates estimated.
RESULTS: Perhaps only one-third to two-thirds of people whose measured diastolic pressures exceed 95 mmHg actually have average pressures that high. The disparity between a single measured diastolic pressure and the mean of many pressure values also leads to errors in identifying individual subjects with mild hypertension. In a general population, single measurements of diastolic pressure exceed 95 mmHg in approximately equal numbers of normotensive, borderline and hypertensive subjects; moreover, one-third of those who are usually in the hypertensive range are not identified. All commonly used screening algorithms give too many false-positive and/or false-negative results. A sequential screening algorithm averaged 3.8 visits per subject and identified 95% of the hypertensives, with only 2.5% of those identified having usual diastolic pressures below 90 mmHg.
CONCLUSIONS: Population-based surveys like the National Health and Nutrition Examination Survey (NHANES) may markedly overestimate the true prevalence of hypertension. This overestimate is greatest for mild hypertension and could significantly affect the cost/benefit analyses of public health policy. Alternative screening methods, such as the sequential algorithm proposed, may have significant benefits in providing a correct classification.