Background

Accurately characterizing placebo response is essential for improving sensitivity in osteoarthritis (OA) trials. This analysis investigates how baseline patient characteristics and psychological factors – leveraged through historical data models – contribute to placebo response. It is based on the ATLAS study, an online, double-blind, randomized, placebo-controlled study that assesses the efficacy and safety of an oral supplement combination in people with symptomatic knee osteoarthritis. Characterizing the placebo response enables more precise estimation of treatment effect and improved interpretation of endpoints.

Methods

A Multidimensional Participant-specific Questionnaire was used to characterize the subjects’ profiles and behaviors. The following 5 characteristics were extracted from the questionnaire. These were used as baseline covariates for adjusting the treatment effect analysis.

• Expectations captures how much a patient expects improvement in their general condition by participating in a study.
• Psychological Profile combines stable dispositional traits that are linked to placebo responsiveness and to non-specific improvements commonly observed in clinical trials.
• Therapeutic Alliance characterizes a study participant’s perception of their relationship with the study staff, as well as their general perception of the clinical trial. This is expected to be less relevant in the case of this remote study.
• Psychological Distress quantifies a patient’s negative psychological state at baseline.
• Health Engagement characterizes the way a patient takes an active role in their health care.

In addition to these 5 psychological variables, a disease profile variable based on classical baseline characteristics (e.g., baseline disease intensity) is also considered.

These six baseline covariates were integrated into ANCOVA models already adjusted for baseline to assess their additional explanatory power across primary and secondary endpoints. R‑squared values quantified the proportion of variance in patient-reported outcomes attributable to these covariates.

Results

The baseline covariates jointly explained between 7% and 15% of additional variance across endpoints.

• Average Pain Score (APS): 14.0% variance explained, mainly due to Expectations and Psychological Distress.
• AQoL‑8D: 15.4% variance explained, primarily driven by Disease Profile and Psychological Distress.
• Patient Global Assessment (PGA): 11.4% variance explained, with Expectations and Disease Profile as major contributors.
• Patient Global Impression of Change (PGIC): 11.1% variance explained, with Expectations as the strongest predictor.
• KOOS Pain: 7.3% variance explained, influenced by Psychological Profile and Health Engagement.

Table 1 reports the individual contributions of each variable in terms of type II ANOVA R‑squared.

The directions of the associations between the preponderant psychological factors and a larger placebo response are consistent across all endpoints. Indeed, Expectations and Psychological Profile are positively associated with a better response, and Psychological Distress is negatively associated with it.

Conclusions

In this remote OA trial, baseline psychological and behavioral factors significantly influenced placebo response. In particular, the results confirm the important role of patients’ Expectations for several endpoints. In addition, the results highlight the interest of jointly adjusting for several psychological covariates. Indeed, while their relative importances appear to vary from one endpoint to the other, the directions of their associations with placebo response are consistent across endpoints. Adjusting for them, therefore, improves the precision of the estimation of the treatment effect.

This confirms that leveraging models based on historical data and psychological profiling enhances analytical precision and supports more robust trial designs. These insights may inform future strategies for optimizing endpoint analysis and identifying patient characteristics to consider in trial planning.

TABLE

Table 1 – Type II ANOVA R² of individual covariates