Answering Unanswerable Questions in the Clinical Laboratory with Data Warehouses

Abstract

Laboratorians face many “unanswerable” questions in their daily work. What caused a false-positive test in this patient, when all the known interferences have been ruled out? How can I identify mislabeled specimens in the laboratory, when all I have are the specimens themselves? Does this patient really need that test? None of these questions is truly unanswerable, of course; rather, the hapless laboratorian must hazard a guess for each question, based on limited time and information, then move on to the next question. With more data and analysis, perhaps a more satisfying answer would have been possible. The more fundamental question, though, is why don't we have easy access to that data? In many situations, we are tantalizingly close. In this issue of Clinical Chemistry , Hughey and Colby present the results of a study aimed at identifying previously unknown sources of cross-reactivity in urine drug screening immunoassays (1). By leveraging their own institution's electronic health record (EHR)2 system, they obtained 698651 individual assay results from 40741 patients and linked them to medication exposures documented in each patient's chart. Using logistic regression on 2201 assay–ingredient pairs, they were able to identify several new sources of cross-reactivity, which were validated experimentally. These results provide immediately actionable information for future testing, both to help decipher false-positive drug screen results and to add additional medications to a list of potential cross-reacting substances. Such large retrospective observational studies, involving thousands of patients scattered across an institution, could not be possible without access to and effective use of EHR data. Unfortunately, because the primary purpose of EHRs is to facilitate individual patients' care, the means to answer broader questions that affect many patients or groups of patients may be elusive to clinical providers and laboratorians alike. When randomized controlled trials and other well-designed studies …