• Number of subjects required To assess the effectiveness of the MSProgress quality approach support tool on the evolution of the indicators selected by the MSPs, the estimation of the number of subjects required (NSR) will be based on the comparison between groups, with and without the MSProgress quality approach support tool, of the percentage of indicators selected at inclusion, satisfied at 12 months. For a criterion met at 20% at baseline, an absolute difference of 10% (i.e., 20% vs. 30%) could be demonstrated with 1,850 patients per group (with and without the MSProgress quality approach support tool), according to the stepped-wedge cluster randomization design described above (with 3 randomization sequences and 3 study periods (years), for a two-sided Type I error risk of 5%, a power greater than 90%, an intraclass correlation coefficient of 0.025 (to account for intra- and inter-cluster variability), and a mean number of 100 patients affected by the indicator, per MSP (steppedwedge command, Stata software, version 15, StataCorp, College Station).
If we consider 4,500 patients per MSP based on the data provided by the French National Health Insurance, the inclusion of 20 MSPs per group (with and without the MSProgress quality approach support tool) could involve 90,000 patients. With 2,000 patients per group (with and without the MSProgress quality approach support tool) (i.e., 20 MSPs with an average number of 100 patients affected per indicator for a calculated minimum population of around 1,850), the selected indicator should cover 2 to 3% of an MSP's patient base, which seems entirely consistent and realistic in terms of feasibility.
For a criterion met at 30% at inclusion, 800 patients are required per group (with and without the MSProgress quality approach support tool) under the assumptions described above; For a criterion met at 15%, 2,000 patients are required per group (with and without the MSProgress quality approach support tool), which is compatible with the inclusion of 20 MSPs and an average of 100 patients affected by each indicator.
• Statistical Methods Statistical analysis will be performed using Stata software (version 15, StataCorp, College Station).
All statistical tests will be performed at a 5% risk of Type I error (α). As discussed by Feise in 2002, adjustment for Type I error will not be proposed systematically, but rather based on clinical and not solely statistical considerations. No correction will be proposed for secondary objectives except for comparisons between the three sequences of the stepped-wedge cluster randomization design.
The statistical individual will be the patient. The data used will be statistical and epidemiological data from the French National Health Insurance databases, which do not allow for patient identification (apart from sex and age), but simply for the identification of a number of patients affected by the indicator chosen by the MSPs.
A description of the MSPs will be provided according to the three sequences of the stepped-wedge cluster randomization design.
Continuous variables will be presented as mean and standard deviation, subject to normal distribution (Shapiro-Wilk test if necessary). In cases of non-normality, they will be presented as median, quartiles, and extreme values. Qualitative variables will be expressed as numbers and associated percentages.
Graphical representations will be included with these analyses whenever possible.
The initial comparability of the groups with and without the MSProgress quality approach support tool will be assessed based on the participants' main characteristics (age and gender).
A description of protocol deviations, patients distributed according to these deviations, and the reasons for discontinuation will also be provided. The number of patients included and the inclusion curve will be presented by groups with and without the MSProgress quality assurance tool.
The primary analysis will be conducted on an intention-to-treat basis as a first-line method.
In a second phase, MSPs who discontinue their participation during the study may be removed as part of a per-protocol analysis. As the data used will be statistical and epidemiological data from the French National Health Insurance databases, which do not allow patient identification, it will not be possible to consider any other per-protocol analysis.
To assess the impact of the MSProgress quality approach support tool on the change in the indicators chosen by MSPs compared to standard care (primary objective), the comparison between groups, with and without the MSProgress quality approach support tool, regarding the primary outcome measure will be carried out using a mixed model to account for intra-cluster variability as well as period and duration effects and their interactions.
Indeed, in addition to the precautions required for all types of cluster-randomized studies, particularly the consideration of intra-cluster correlation, a significant bias specific to the stepped-wedge design must be sought, studied, and taken into account. In this type of study, there are more clusters (MSPs) exposed at the end of the study than at the beginning; the effect of the intervention can therefore be confused with any other underlying time trend (such as an overall improvement in the quality of care or an improvement in the quality of life of the patients included in the study). This effect is likely to reduce the precision of the study. It must be discussed and the results statistically adjusted for this "time effect".
Given the statistical nature of the variable studied (percentage of indicators chosen at inclusion, satisfied at 12 months), the statistical model will be of the generalized linear type (logit function for binary dependent variable). These regression models consist of an extension of regression models applied to hierarchically structured data by the introduction of a so-called random effect (MSP cluster effect); in order to model the differences, not only between micro-units (patients), but also between macro-units (MSPs).
The results will be expressed in terms of odds ratio and 95% confidence interval. The intra-class correlation coefficients will be presented with a 95% confidence interval for the entire population and by group, with and without the MSProgress quality approach support tool.
The secondary analyses, which will aim to (1) compare the impact of the MSProgress quality support tool on the evolution of the indicators chosen by MSPs according to the number of years of use of the tool, (2) evaluate the impact of the MSProgress quality support tool on the evolution of all the indicators made available to MSPs, and compare the impact of the MSProgress quality support tool on the evolution of all the indicators between the MSPs that participated in the pilot study and the MSPs included in the randomized controlled trial, will be based on the models described for the main analysis.
For longitudinal analyses, the time effect (period) will be considered as a fixed effect; the group (with and without the MSProgress quality support tool) x time interaction will be studied. The statistical analyses will focus on the evolution of the three indicators selected by the MSPs between 2021 and 2027, with the years 2021 to 2023 allowing for the natural evolution of the results.
To study the "having participated in the pilot study yes/no," this variable will be considered as a covariate in the models described above.
The results will be presented as described above, in terms of odds ratios and 95% confidence intervals.
MPSs who have already participated in the tool development phase and/or the MSProgress pilot study will not be considered for randomization but will be followed in a parallel open-label arm. The comparative analyses described above for the MSPs included in the cluster-randomized study will then be replicated considering these MSPs who have already participated in the tool development phase and/or the MSProgress pilot study.
* Interim Analyses Not applicable.
* Method for Accounting for Missing Data No missing data is expected; the data used is exhaustive as it comes from the National French Health Insurance databases.
* Management of Changes to the Analysis Plan A detailed statistical analysis plan will be prepared before the database is frozen. It will take into account any protocol changes or unexpected events occurring during the study that impact the analyses presented above. Planned analyses may be completed in accordance with the study objectives.
Any subsequent changes to the statistical analysis plan must be justified and will result in a new version of the document. These deviations from the analysis plan will be reported in the final study report. All documents will be kept in the study file.
