Stroke data banks.

To the Editor: Dr. Mohr is to be complimented for a thoughtful, balanced discussion of stroke data banks.' While there are pros and cons to this issue, in our view, the advisability of establishing stroke data banks remains questionable , given the predictably high cost of such endeavors and the predictably low-level, descriptive research that such enterprises can support. The editorial acknowledged some of the problems of quality control in large data banks, where interobserver variability and differing standards between participating centers may play substantial roles. Transcription errors, omissions, and the like also account for serious quality control problems, particularly in large databases that are likely to be computerized. 2 The desire to hoard data is a natural instinct for those of us involved in clinical research. However, unlike money in the bank, data in a data bank do not gain interest merely because they have been saved. On the contrary, data that are routinely collected and salted away for some ill-defined future purpose are likely to harbor many undetected errors. When discovered, many of these errors will be uncorrectable because nobody has looked at the data until long after the patients involved have disappeared from the scene. As Dr. Mohr noted, the referral patterns, practice preferences, and catchment area of an institution or group of institutions that sponsor disease-specific databases may yield data that are difficult to apply in other settings. When it is possible to maintain a database from a well-defined population (e.g., a state tumor registry or mortality statistics), the quality limitations of the stored information are appreciated by knowledgeable investigators. Such records are rarely, if ever, comprehensive ; for purposes of clinical inference, subjects may be identified from such registries, but individual patients or original documents are often located for collection of additional data specific to the studies being undertaken. No database can be all things to all researchers. It often transpires that a critical variable turns out to be the one item that was not collected in the "comprehensive" database. Many quality control problems in stored databases can be overcome with sufficient money and effort, but even when the data collected are complete and accurate, the questions that a data bank can answer may be severely limited. Descriptive studies are widely regarded as the weakest evidence of causation, for example. Unless the database provides information on a control or comparison group — individuals without stroke, …