Methods for Incorporating Nonrandomized Study Data for Policy-making
Objectives: Undesirable effects, or harms, of an intervention are often not adequately reported in randomized controlled trials. We attempted to devise a cost and time effective method, consisting of three approaches, to collect and present this data for a panel of experts to consider when developing clinical practice guidelines (CPG). We applied this method during the development of an American Dental Association’s CPG on the use of antibiotics to treat pulpal and periapical conditions.
Methods: We used the following approaches to search for evidence reporting harms: 1) an exhaustive search for systematic reviews (SRs) of observational studies on patients exposed to the intervention of interest; 2) a non-exhaustive search for national or registry reports presenting surveillance data on patients exposed to the intervention of interest and/or containing data on relevant harms; and 3) a non-exhaustive search for individual observational studies presenting data on patients exposed to the intervention of interest and/or containing data on relevant harms. We then populated GRADE summary-of-findings and evidence-to-decision tables with the data obtained from these approaches. Theses tables were presented to the guideline panel to consider when formulating recommendations.
Results: During our exhaustive search and screening process for SRs using the first approach, we found none that were useful for the panel’s decision-making needs. We did however find four registry reports using the second approach, and one SR and three individual observational studies using the third approach. These studies informed twelve harm outcomes including mortality due to antibiotic resistant infections and anaphylaxis. The panel found this information useful and incorporated it into clinical recommendations.
Conclusions: This approach appears to be an efficient method to provide a broader array of harm outcomes to panels when formulating recommendations. While this method may offer utility, it needs to be compared to other approaches to assess its validity.