Sensitivity analysis in multiple imputation in effectiveness studies of psychotherapy

Area of Science:

• Psychotherapy research
• Statistical methodology
• Health outcomes research

Background:

• Effectiveness studies increasingly require methods to handle incomplete data.
• Multiple Imputation (MI) is a common technique but assumes data are missing at random (MAR).
• Sensitivity analysis is crucial to assess the impact of departures from the MAR assumption.

Purpose of the Study:

• To present a novel sensitivity analysis technique for missing data in effectiveness studies.
• To evaluate the robustness of statistical inference against missing not at random (MNAR) data.
• To assess the impact of missing data on clinical significance in longitudinal outcome data.

Main Methods:

• Developed a sensitivity analysis technique based on posterior predictive checking.
• Incorporated the concept of clinical significance for intra-individual changes.
• Applied the method to irregular longitudinal data from the Outcome Questionnaire-45 (OQ-45) and Helping Alliance Questionnaire (HAQ) in 260 outpatients.

Main Results:

• The sensitivity analysis can quantify bias from MNAR data in worst-case scenarios.
• It allows comparison of different methods for handling missing data.
• The analysis can detect violations of model assumptions, such as non-normality.
• Patient and therapist ratings on the HAQ improved predictive value for OQ-45 based monitoring.

Conclusions:

• The proposed sensitivity analysis enhances the reliability of effectiveness studies with missing data.
• Multiple Imputation (MI) with repeated OQ-45 and HAQ measurements improves accuracy in quality assurance and non-randomized studies.
• This approach is valuable for outpatient psychotherapy research where dropouts are common.