[Pitfalls of searching online: all symptoms, diseases and adverse events appear correlated].

When confronted with an unexpected clinical observation, such as a remarkable symptom in a patient with an unrelated rare disease, clinicians increasingly apply online literature search to support the observed correlation. Against a background of an exponential rise in medical publications and the well-documented problem of publication bias, the easy access to literature carries the risk of suggesting spurious correlations. The current paper expounds on this phenomenon. Queries in medical search engines often provide a number of hits, regardless of the plausibility of the correlation searched for. To quantify this, we recently performed a study involving 30.000 automated queries in PubMed using completely random search terms drawn from lists of diseases, symptoms and medications. This provided a background rate of PubMed hits. The data support that several hits by no means automatically indicate a relevant correlation, and underline need for judicious critical appraisal when searching for a correlation observed in daily practice.