Predictive validity of different definitions of hypertension for type 2 diabetes.

Background: Models to predict diabetes or pre-diabetes often incorporate the assessment of hypertension, but proposed definitions for 'hypertension' are inconsistent. We compared the classifications obtained using different definitions for 'hypertension'.

Methods: We compared records for 5158 cases from 181 family practices, who were later diagnosed with diabetes and prescribed oral hypoglycaemic drugs, with 5158 controls, matched for age, sex and family practice, who were never diagnosed with diabetes. We compared classifications obtained using definitions of hypertension based on medical diagnoses, prescription of blood pressure lowering drugs or both. We compared family practices where diagnosis or prescribing varied systematically.

Results: Classification of hypertension based on recorded medical diagnoses gave a sensitivity of 32.2% for diabetes (95% confidence interval from 30.4 to 34.1%). Prescription of blood pressure lowering drugs in the 12 months before diagnosis gave a sensitivity of 47.2% (45.7 to 48.7%). Combining either a medical diagnosis or a blood pressure lowering prescription gave a sensitivity of 52.8% (51.3 to 54.3%). In family practices where hypertension was least frequently recorded, a diagnosis of hypertension gave a sensitivity of 19.5% for diabetes (17.4 to 21.6%) compared with 50.8% (46.3 to 55.3%) in the highest quintile. Prescription of blood pressure lowering drugs gave a sensitivity of 36.1% (33.1 to 39.0%) in the lowest prescribing practices but 58.2% (55.5 to 61.0%) in the highest quintile.

Conclusions: Misclassification errors depend on the definition of hypertension and its implementation in practice. Definitions of hypertension that depend on access or quality in health care should be avoided.