Purpose

Disparities among sites are frequently observed in the results of Randomized Controlled Trials (RCTs).  As a result, adjusting for site effects is a common strategy in RCT analyses to account for these disparities. However, such adjustments introduce additional parameters into the analysis, which can be especially detrimental in Osteoarthritis (OA ), with an average number of subjects per site relatively small. This trade-off raises questions about whether the benefits of site adjustment outweigh the potential loss in precision.

This analysis evaluates the utility of site adjustment in RCTs, focusing on the special relationship between site disparities and patient expectations, a key factor influencing outcomes in chronic pain conditions. By examining these interactions, we aim to optimize RCT analyses and better understand contextual factors shaping trial outcomes.

Methods

This post-hoc analysis was conducted on data from a study evaluating the effects of an intra-articular injection in patients with OA. The study included 173 subjects across 18 sites, with the number of subjects per site ranging from 2 to 18 (median = 10). To better assess the influence of site variability, analyses were performed both on all sites and on a subset of larger sites, having at least 10 subjects.

The primary outcome was the response to treatment, measured using the WOMAC Pain subscale. Baseline expectations of improvement were assessed using the MPsQ Questionnaire. Statistical analyses evaluated the extent to which differences between sites explained variability in baseline expectations and study outcomes. Pearson’s correlation was used to investigate the relationship between site-level expectations and response variability, as well as to compare the impact of adjusting for sites versus baseline expectations on the precision of treatment effect estimation.

Results

Site differences explained 17.4% of the variance in baseline expectations (p = 0.02), but only 12.5% (p = 0.052) for larger sites. Thus, while site differences influenced expectations, inter-patient variability was much greater than site-induced variability.

Regarding the study outcomes, site differences explained 16.2% of the variance in WOMAC Pain response (p = 0.043), decreasing to 12.7% (p = 0.055) for larger sites. However, after adjusting for baseline expectations, these differences were no longer statistically significant (p=0.126 for all sites and p=0.168 for larger sites).

This means that a large part of the differences in response observed between sites is explained by the patient’s expectations. Indeed, the average Womac-Pain response in each site was highly correlated with the associated average expectation for the site (correlation=0.54, p=0.02). This correlation was even stronger when considering only the sites with more than 10 patients (correlation=0.73, p=0.02).

While sites play a small role in shaping patients’ expectations, differences in expectations largely explain the variation in responses between sites. Measuring and adjusting for the patients’ expectations should improve the estimation of the treatment effect and decrease the variability coming from the differences between the sites. To test this idea, we compared the precision of the estimated treatment effect when using the expectation or the sites as covariates. Used as a covariate, the sites was not able to increase the precision of the estimated treatment effect (-6.3% of precision). At the opposite, the adjustment for expectation increase the precision by 15.0%.

Conclusions

In this study, we explored the role of clinical sites in shaping patients’ expectations and response. Our findings suggest that the influence of sites on baseline expectations is relatively small. The observed site-related differences in expectations were only marginally significant, indicating that they could primarily result from random variation. If these differences are not random, they may reflect heterogeneity in patient populations across sites rather than being attributable to site-specific practices or personnel.

However, differences in patients’ expectations between sites were strongly correlated with the variation in treatment responses across sites. When baseline expectations were accounted for, the differences in WOMAC Pain response between sites were no longer statistically significant.

These results highlight the value of measuring and adjusting for patients’ baseline expectations as a more effective approach to improving study power and reducing variability than relying on site adjustments alone. Furthermore, they call into question the utility of adjusting for site disparities, as sites have no direct statistical impact on precision of treatment response, in particular once expectations are considered.

Overall, this emphasizes the importance of incorporating patient-centric measures to enhance the precision of RCT analyses.