In contrast to the suggestion of Bloch and colleagues, we did not recommend any cut-off level for the probability of stenosis to be used in clinical practice. We merely showed the consequences of using different cut-off levels. The probability of 30% was mentioned to illustrate that the accuracy could approximate that of renal scintigraphy in our patient sample. A cut-off level of 10% may be more justifiable because this corresponds to a sensitivity of approximately 90% and a 40% reduction in the number of angiograms. Whether this selection should be followed by a noninvasive test with a high sensitivity or should be followed directly by angiography depends on the patient's probability of stenosis and the clinician's experience with available diagnostics. Regarding the severity of hypertension, we emphasize that the inclusion criterion for our study was a diastolic blood pressure of 95 mm Hg or greater despite antihypertensive therapy. Some patients with only one or two clinical characteristics will be missed, but that is inherent to the variability among patients in their clinical presentation.

We agree with Fine and Blaufox that prospective validation of our prediction rule is important because patient characteristics and their interrelations may differ among populations (for example, regarding ethnicity or referral pattern). We did recognize the risk of "overfitting" the model to our patient data, and we therefore corrected for it in our modeling strategy. Furthermore, assessments of validity according to center were satisfactory. We are planning prospective assessments, and we hope that others will share their experience with the prediction rule with us.

The issue of renovascular hypertension versus anatomic renal artery stenosis, mentioned by Bloch and colleagues and Fine and Blaufox, is often raised in debates over the optimal diagnostic work-up. Unfortunately, there is no useful definition of renovascular hypertension that takes into account events such as technical failure of angioplasty, restenosis, or the problem of third-phase Goldblatt hypertension. The accuracy of scintigraphy in predicting blood pressure response after intervention is controversial (1). This issue will also be addressed in the forthcoming report on the therapeutic part of the DRASTIC study. Even if one accepts that scintigraphy can predict the presence of renovascular hypertension in populations with a prevalence of at least 30% (2),the question of how to reach such a prevalence in an average hypertensive population remains. In our view, selection based on drug-resistant hypertension followed by the prediction rule is a sensible approach.

Recurrent flash pulmonary edema, as mentioned by Rajput and Nammour, is specific but not sensitive because only a minority of patients with renal artery stenosis exhibit this clinical syndrome. Thus, we did not consider it for the prediction rule.

Neggers and van der Meulen's confusion arises from scanty information in the appendix on how to code the risk factors: presence of a risk factor should be coded as 1, absence as 0, male as 1, and female as 0; serum creatinine should be expressed in µmol/L. To enable easy use of this formula, a spreadsheet can be downloaded from our Web site: http://www.eur.nl/fgg/mgz/software.html.