Abstract
BackgroundUsing a real dataset, we highlighted several major methodological issues raised by the estimation of the Minimal Clinically Important Difference (MCID) of a Patient-Reported Outcomes instrument. We especially considered the management of missing data and the use of more than two times of measurement. While inappropriate missing data management and inappropriate use of multiple time points can lead to loss of precision and/or bias in MCID estimation, these issues are almost never dealt with and require cautious considerations in the context of MCID estimation.MethodsWe used the LIGALONGO study (French Randomized Controlled Trial). We estimated MCID on the SF-36 General Health score by comparing many methods (distribution or anchor-based). Different techniques for imputation of missing data were performed (simple and multiple imputations). We also consider all measurement occasions by longitudinal modeling, and the dependence of the score difference on baseline.ResultsThree hundred ninety-three patients were studied. With distribution-based methods, a great variability in MCID was observed (from 3 to 26 points for improvement). Only 0.2 SD and 1/3 SD distribution methods gave MCID values consistent with anchor-based methods (from 4 to 7 points for improvement). The choice of missing data imputation technique clearly had an impact on MCID estimates. Simple imputation by mean score seemed to lead to out-of-range estimate, but as missing not at random mechanism can be hypothesized, even multiple imputations techniques can have led to an slight underestimation of MCID. Using 3 measurement occasions for improvement led to an increase in precision but lowered estimates.ConclusionThis practical example illustrates the substantial impact of some methodological issues that are usually never dealt with for MCID estimation. Simulation studies are needed to investigate those issues.Trial registrationNCT01240772 (ClinicalTrials.gov) registered on November 15, 2010.
Highlights
Using a real dataset, we highlighted several major methodological issues raised by the estimation of the Minimal Clinically Important Difference (MCID) of a Patient-Reported Outcomes instrument
Considering the MCID as the mean General Health (GH) score difference, it was estimated between 4 and 5 in the group who felt little improved, and between − 1 and 2 in the group who felt little worsened
We recommend using a Multiple Imputation by Chained Equation (MICE) procedure for imputation, instead of imputation on the mean, but the modeling of the imputation procedure should be approached with great caution
Summary
We highlighted several major methodological issues raised by the estimation of the Minimal Clinically Important Difference (MCID) of a Patient-Reported Outcomes instrument. Subjective concepts like quality of life or satisfaction are as relevant endpoints as mortality in clinical studies. Since they can be assessed mostly by patients’ speech, these concepts are at best reported directly by the patient himself without interpretation by a clinician with instruments called Patient-Reported Outcomes (PRO) [1]. These instruments are increasingly used in studies and clinical practice, since it gives the patient a central place in his medical care. An improvement of four points on a Health-Related Quality of Life (HRQoL) score after an intervention can be enough to reach statistical significance with appropriate sample size, but it can be hard to tell anyway if it is a meaningful difference
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
