To the Editor:

Kshirsagar and colleagues1 recently developed a model for predicting 3-, 6-, and 9-year risk of incident hypertension. They have done an incredible attempt to improve hypertension risk prediction. Good calibration and acceptable discrimination capacity has made their work a valuable one in the field. They have also converted their regression models to a risk-scoring algorithm. The algorithm is simple and user-friendly, two characteristics that could promote its utilization in office-based primary care settings. However, the transportability of the model to populations other than the populations from which it has been derived needs to be validated. In external validation, the (logistic regression) parameter estimate for each risk component of the model should be available. Unfortunately, they have not provided constants (logistic regression intercepts) of the regression models. Besides, they have introduced interaction terms between age and diastolic blood pressure (DBP) into their multivariate logistic regression model and observed odds ratios (ORs) of 1 to 1.38 for these interaction terms. Namely, the effect of DBP on risk of incident hypertension is augmented by the effect of age, ie, the older an individual, the greater the magnitude of effect of DBP on the incident hypertension. This is in conspicuous disagreement with previous findings in US populations.2 We recently observed in the Tehran Lipid and Glucose Study that age either does not modify the effect of DBP on the risk of incident hypertension or depreciates its effects. The ORs reported by Kshirsagar and associates for the age × DBP interaction term seems to be too large. For instance, at the age of 50, the OR (6-year model) for 1-mm Hg increase in DBP would be (eln(1.07)*1+ln(1.19)*1*50) more than 6408, which is unlikely to be true. In the risk-scoring algorithm, however, the effect of DBP decreases as age increases as has been observed in other studies. They have included diabetes as a risk factor for incident hypertension, but have not made it clear whether the same criteria have been used to define hypertension among diabetic and nondiabetic participants. Thresholds for defining hypertension are lower for patients with diabetes.3

Disclosure:  None to declare.


  1. Top of page
  2. References