RESUMO
PURPOSE: To propose and validate an algorithm to analyze the reports of suspected therapeutic failure (TF) received in pharmacovigilance programs. METHODS: A Delphi consensus method with a group of 12 international experts was used to identify the different causes that prompt TF and to propose an algorithm to analyze reports of suspected lack of efficacy of medicines. A pilot analysis of 50 reports was the basis to evaluate the interrater and intrarater validity of the algorithm. RESULTS: A 10-question algorithm was proposed. The evaluation of 50 reports of suspected TF showed that only 8% could be actually attributed to a manufacturing quality problem, whereas the real reason underlying the alleged TF was the inappropriate use of the prescribed drug in 28%. Minimum information to attribute the cause to a TF was lacking in 31% of these reports. The interrater reliability was "moderate" (kappa coefficient = 0.55), and the intrarater reliability ranged from 0.732 to 0.908 ("substantial" to "almost perfect"). CONCLUSIONS: The proposed TF algorithm is a valid, reliable, and reproducible analysis tool that can help to disentangle the frequent and complex reports of suspected TF.