[Disagreement among hospital and community studies evaluating the same research question]

The goal of this review is to delineate some of reasons that justify the lack of consistency between hospital-based and community research. The main reasons for the differences are the selection of the hospital population and information based on clinical chart (its lack of uniformity and the treatment of the not available data). The reasons for lack of consistency are divided according to the type of research question: frequency, diagnosis, etiology, prognosis and treatment-prevention. The way a hospital population is selected justifies discrepancies regarding frequency and prognosis. As regards diagnosis, differences are mainly due the prevalence of disease. In the ascertainment of causality several biases are more common in hospital-based research, such as detection bias, protopathic (both producing an away-from-null estimate), and inclusion bias (diminishing the strength of association). Examples taken from the medical literature are offered to illustrate each bias. Regarding treatment-prevention problems arise from external validity, as clinical trials are less prone to bias; this latter situation is exemplified with an assessment of vaccine efficacy in both patients and healthy population. The frequency of citation of bias was assessed by a Medline search; in hospital studies detection bias and confounding by indication were more often quoted than in non-hospital research (RR = 2.71; 95% CI; 1.69-4.37; RR = 1.76; 95% CI, 0,90-3,42, respectively). Lastly, several recommendations are given to increase the validity of hospital-based research.