Jollis and colleagues [1] found a low sensitivity (but a high specificity) of hospital discharge diagnosis coding for the identification of cardiovascular prognostic factors. The results of this large study were well analyzed and are consistent with previous findings regarding the limitation of hospital discharge diagnosis data.

However, the authors are not justified in making broad statements about "the claims." Reliance on a single set of discharge diagnoses is no more optimal than reliance on one procedure note. Multiple claims data streams, preferably with data from several years, should yield better results.

The hospital discharge data they used are only part of those available in Medicare. In some settings, for example, procedure fields on the hospital claims may be relevant, either in their own right or because they imply certain diagnoses of interest (for example, coronary bypass or angioplasty implies coronary artery disease). Physician claims and outpatient facility use claims, both of which now contain information on diagnoses and procedures, may be linked to hospital claims to build a "claims history" for every patient. Without such complete claims data, any general inference regarding "claims data" is simply unsupported; sensitivity in particular will suffer from incomplete data.

Jollis and colleagues [1] are careful to discuss their results in terms of the implications for prognostic comparisons, but Dr. Dans, in his editorial [2], has generalized the stated conclusions to most uses of claims data except for exploratory "trend" analyses. He is apparently unaware that valid epidemiologic studies using claims are indeed possible, given that procedures are well coded [3] and diagnoses associated with procedures (for example, as fractures are implied by fracture repair) are also reasonably well captured [4]. Claims data do have limitations with regard to chronic diseases and in some settings [5]. However, limitations for one sort of analysis do not necessarily imply trouble in others, and Dr. Dans is not on firm ground in limiting claims research to the areas he set out. Does evidence exist to suggest that most investigations on other topics are done without regard to validity simply because the data are there "to mine"?