In reply

Abstract

In reply: Dr. Bernardini raises many important points. I will not address those already mentioned in the discussion section of our original publication. Physician's visits or hospitalization were not attributed to the event unless a reference was noted in the medical record. Prolongation of hospitalization was counted if event-related treatment lasted after the primary indication for hospitalization had cleared. Distinction of drug-associated from illness-associated events is very difficult to assess in frail, multiply diseased nursing home residents. Ideally, all etiologic factors should be considered in a study regarding adverse events. Our study, however, was designed to evaluate adverse events and determine the probability that they were drug-related. A validated drug event algorithm and our parallel drug withdrawal algorithm concluded that a substantial proportion of events (68%) related to drug administration, and 40% of those related to drug withdrawal were of low probability, suggesting alternative etiologies. Therefore, multivariate analyses were confined to events of highest probability levels. We did control for function, comorbidity, medication number, hospitalization, and length of stay. Our results are similar to others. Loosely translated, being sick and vulnerable places you at risk of being sick and vulnerable. Why the disparity of event frequency in our two studies? It is possible that our definitions of events differed, producing disparate rates, a reason to develop a “threshold” definition. Overzealous recording may have produced an “upper bound” estimate of events, while the opposite may occur in other settings. Our nursing home is hospital-based, and a large proportion of admissions are recovering from acute illness, enhancing the likelihood of adverse events. No one would disagree with Dr. Bernardini's and his colleagues conclusion that “improvement of care management” is best when “the geriatrician is directly and fully engaged.” I would submit, however, that good care may be easier than care-related research. Even prospective studies using careful methods may be unable to distinguish among causal factors for adverse events. We should identify clusters of risk factors and combinations of interventions which have the greatest chance for success. Is this area of research just too messy? Perhaps, but I don't think so. Policy makers attempting to improve quality of care appear to be moving to large data bases (Medicare), which show patterns of events. Dr. Bernardini, myself, and others have chosen to stick closer to the bedside and the complex, individual patient. In fact, our patient-specific research and others’ data base research might be considered analogs of clinical research and epidemiologic public health research. These two forms of research are both vital and complementary. The findings from one form of research must be validated in the other. Without both tools, real improvement in quality of care is unlikely.