(1) To identify and test the best methods for incorporating potential confounders of patient-reported outcomes that researchers did not measure at the same time as treatment initiation into the analysis of clinical registry data; (2) To assess patient preferences regarding clinical registry participation
|Simulation study and empirical analysis
|Data Sources and Data Sets
- Objective 1: computer data simulations based on the Brigham and Women’s Rheumatoid Arthritis Sequential Study (BRASS) patient registry
- Objective 2: survey responses from 150 BRASS patients and 169 patients from Brigham and Women’s Inflammatory Bowel Disease Tissue Repository (BrITR)
- Objective 1: multivariate logistic regression modeling comparing adjustments to a confounder variable of patient global assessment scores in estimating risk ratio
- Objective 2: focus groups, survey questions about registry participation
- Objective 1: descriptive, inferential statistics to determine the least biased confounder adjustment method
- Objective 2: patient preferences for patient registry participation
Using patient registry data in comparative effectiveness research can help researchers examine changes in patient outcomes after starting a new treatment. Registries include patient-reported outcomes, such as measures of overall health and fatigue, to monitor patients’ well-being over time. Baseline measures capture confounders, which are variables that may affect the treatment results. Patients often report these outcomes at regular intervals, but these intervals rarely coincide with treatment initiation. Researchers must adjust for baseline confounders that they did not measure at true baseline, prior to treatment initiation.
This simulation study compared the ability of computational methods to incorporate patient-reported outcomes as confounders into the analysis of comparative drug safety research. The research team then applied these adjustment methods to rheumatoid arthritis (RA) patient registry data to validate results. The simulation study compared 13 confounder adjustment methods to model the total comparative effect of drug treatments on risk ratios for infection. The adjustment methods included, for example, using the most recent measurement prior to treatment initiation and using the arithmetic mean of the two nearest measurements. Simulation scenarios assumed a 5,000-patient sample, used different distributions of patient global assessment scores, and employed varying time intervals between patient global assessment measurement and treatment initiation.
The team empirically validated the tested confounder adjustment methods using data from 294 eligible patients with RA who began treatment with a tumor necrosis factor-α inhibitor or other medicine. Researchers assessed comparative safety based on infection risk following treatment.
To obtain patient preferences for participating in registry research, the research team conducted focus groups and administered surveys to 150 patients with RA and 169 patients with inflammatory bowel disease (IBD). In each survey group, 95% of respondents were non-Hispanic white. The mean age was 62 for the RA respondents and 43 for the IBD respondents, and 83% of RA respondents and 62% of IBD respondents were women.
Patients, clinicians, and patient registry researchers helped develop focus group questions and study design elements.
- For the simulation studies, the confounder-adjustment method that produced the least biased risk ratio used patient-reported outcomes measured at the timepoint closest to—either before or after—treatment initiation.
- All tested approaches using empirical data gave similar estimates of treatment effect. Using patient-reported outcomes measured at the timepoint closest to treatment initiation—either before or after—yielded the most consistent and least biased estimates.
Patient preferences about registry participation
- The top three motivating factors for registry participation across all focus group and survey respondents were altruism, possible benefits for one’s own care, and convenience.
- Spanish-speaking focus group participants expressed less interest in registry participation, due to worries about being guinea pigs and distrust of translators.
- Survey respondents under 45 years preferred surveys via e-mail, internet, or phone apps, and those over 65 years preferred mailed surveys. Respondents aged 45–65 had mixed preferences.
The simulation study assessed a single continuous confounder; results may not apply to studies with multiple confounders. Incomplete, missing, or differentially reported data may alter the accuracy of the confounder adjustments.
Most survey respondents were non-Hispanic white; preferences may be different for other populations or for those who chose not to participate.
Conclusions and Relevance
Confounder measurement timing relative to treatment initiation may influence estimates of associations between treatment and clinical outcomes in comparative effectiveness research. Using patient-reported outcomes measured at the timepoint nearest an intervention’s initiation, even if measured after treatment begins, may control for confounders most effectively.
Because registry participants prefer different methods for completing surveys, researchers should consider tailoring their approach for delivering surveys.
Future Research Needs
Future research could examine multiple potential confounders and other health conditions relevant to registry data sets to determine whether these results generalize. Studies could also examine preferences for registry participation among racially and ethnically diverse patients.