Limitations of inferences from observational databases in amyotrophic lateral sclerosis: all that glitters is not...

Data from three observational databases have suggested that survival in patients with ALS who take riluzole is far greater than that reported in randomized controlled studies. This editorial discusses why therapeutic efficacy cannot be inferred from observational databases. Data in these databases cannot control for biases in treatment assignment or for differences in intensity of follow-up or supportive care. The retrospective riluzole data, as presented so far, have not demonstrated comparability between the treated and untreated groups across all known prognostic factors, including vital capacity at the start of the observation period. Furthermore, the similarity of untreated patients to historical cohorts likely reflects adverse selection. Optimization of analysis in retrospective studies may be accomplished by allowing full access to data to all interested parties.