Scientific Poster presented on April 2021 at the Osteoarthritis Research Society International Meeting.
Questionnaires assessing patients’ pain level represent the majority of efficacy endpoints in osteoarthritis (OA) randomized clinical trials (RCTs). The intrinsic subjectivity of pain may render the evaluation of these questionnaires difficult. In turn, this could explain the high variability observed in the treatment response in OA RCTs. Mitigation of the evaluation error could therefore help to increase the quality of study data.
The primary objective of this analysis was to estimate the pain evaluation variability and its evolution over time during the course of a clinical trial. In particular, we assessed the learning effect associated with a daily repetition of the pain self-assessment on the evaluation error made by the subjects. In addition, we assessed the relationship between this daily learning effect and other pain questionnaire (Brief Pain Inventory – Severity and its individual questions).Overall, this analysis provides insights into the design of OA RCTs, confirming the importance of a pre-baseline run-in period for the subjects to learn how to rate their own pain. This would allow each subject to determine his/her own interpretation of the pain scale and to be more consistent throughout the study. Furthermore, there would be no longer need to exclude non-consistent subjects at baseline.
Sixty-four OA subjects (hip and knee) were enrolled and given placebo for 12 weeks in a blinded manner. From the initial screening visit through the final end-of-study visit, they were asked to assess their average pain score (APS), worst pain score (WPS) and lowest pain score (LPS) on a daily basis. We analysed the learning effect resulting from the daily repetition of these 3 assessments. At the five clinic visits (screening, baseline, treatment visit 1, treatment visit 2 and end-of-study) they were also asked to complete the short form of the Brief Pain Inventory (BPI).
We assessed the pain evaluation variability by averaging the standard deviation of the daily measures (APS, LPS and WPS) of each subject during a rolling interval of 7 days. To observe the evolution of the evaluation error, we also computed the Pearson’s auto-correlation between two successive days for each daily pain measure. Assuming that the evaluation error would represent the main cause of a lower correlation if the disease was stable, we compared the correlation between the first two consecutive days and the last two consecutive days. As an equivalent error could have less impact with an increase of the variance between the subjects, we adjusted the auto-correlation for the variance. Similarly, a direct estimation of the evaluation error was computed with the formula:
where ei is the evaluation error at day i, and APSi,j are the APS recorded at day i or j. To avoid small daily variations, all the reported standard deviations, correlations and errors were averaged with a Gaussian smoothing with a standard deviation of 5 days.
To observe the relationship between this daily learning effect and other pain-related questionnaires, we computed the adjusted auto-correlation and the associated evaluation error from one time point to another of the BPI-Severity and its questions (BPI APS, BPI WPS, and BPI LPS). We also estimated the evaluation error by computing the adjusted correlations between the different pain questionnaires at each time. Indeed we assumed that a reduction of the evaluation error would help to increase the correlation between these questionnaires.
The weekly variability of the pain evaluation by the subjects decreased as the study progressed. Indeed, the average weekly standard deviation of the APS, LPS and, WPS decreased significantly with a relative reduction of, 34.31% (p-value < 0.001), 45.84% (p-value < 0.001) and 26.49% (p-value = 0.004), respectively.
This reduction of the evaluated pain variability was also observed with a significant increase of the auto-correlations between two successive days for the APS, the WPS and the LPS. After adjustment for the variance, the increase was still significant. The associated estimated evaluation errors were reduced significantly for APS, WPS and LPS. All the correlations, evaluations errors and p-values are reported in Table 1. The evolution of the adjusted correlation of the daily APS is presented in Figure 1.
Table 1: Correlation, Adjusted Correlation and Estimated Evaluation Error Reduction for daily pain scores.
The decrease of the variability of the pain evaluation was also observed for the other pain assessment (BPI-Severity and its individual questions). The increase of the adjusted auto-correlation and the decrease of the associated evaluation error were significant for the BPI-Severity (p-value =0.002) and BPI LPS (p-value <0.001).Accordingly, the correlation between the pain-related questionnaires also increased during the study. The adjusted correlation between weekly APS and BPI APS, weekly WPS and BPI WPS and, weekly LPS and BPI LPS presented a significant increase (with p-values after adjustment of, 0.015, 0.005 and < 0.001, respectively). The adjusted correlation between weekly APS or weekly LPS and BPI-Severity presented also a significant increase (p-values were 0.002 and 0.022, respectively).
The results presented here clearly demonstrate a reduction over time of the evaluation error of daily pain measurements (average, worst and lowest pain scores). The reduction of this evaluation error implied that the subjects were more consistent at the end of the study in their pain evaluation. This growing consistency of the subjects with time could be linked with a learning effect induced by their daily pain assessment.
Furthermore, there was a significant decrease in the evaluation error of the other pain assessments (BPI-Severity and its questions) despite the fact that they were assessed only five times during the study. The correlation between these assessments and the daily repeated measurements also increased significantly.
This would suggest that the learning effect by a daily self-recording of pain would allow an improvement of the way subjects consistently assess their pain from one day to another and from one endpoint to another. It emphazises the importance of a daily pain evaluation, in particular in a run-in pre-baseline period. Based on the results of these analyses (Figure 1), a period of 20-40 days for daily self-recording of pain assessment seemed to achieve a meaningful and significant reduction of pain evaluation error. This increase in the evaluation performance induced by a daily pain self-assessment could allow for much better estimations of the placebo and treatment responses without excluding any subject at baseline.