Table of Contents
A new framework for clinical trial analysis emphasizes a careful balance between statistical rigor and pragmatic data handling, particularly regarding interim assessments and missing data. Researchers are outlining detailed protocols for sample size re-estimation, data population definitions, and strategies to address incomplete datasets, aiming to maximize the reliability and validity of study results.
The Importance of Interim Analysis and Sample Size Re-estimation
The cornerstone of this approach lies in the planned interim analysis,designed to re-evaluate the required sample size. This process, expected to occur around month 21, will be triggered once data from at least 29 patients who have completed the primary endpoint is available. According to a senior official involved in the trial, Teare (2014) suggest optimal sample sizes ranging from 20 to 70 patients for two-arm studies with continuous outcomes. The current trial’s plan to proceed with re-estimation after 29 patients falls towards the lower end of these recommendations.
Maintaining data integrity throughout the trial is paramount. The progress of all participants will be meticulously documented using a CONSORT flow diagram, detailing each stage from initial screening to final analysis. This diagram will include information on eligibility, randomization, withdrawals, and reasons for exclusion.
Several distinct analysis populations will be defined before database lock to minimize bias. These include:
- Safety Population: All subjects entered into the study, regardless of treatment received, used for adverse event analysis.
- Interim Analysis Population: Patients who have completed the study,excluding those with insufficient adherence data.
- Intention-to-Treat Population: All randomized subjects, analyzed as if they received their assigned treatment, even if they deviated from the protocol.
- Per-Protocol Population: Notably, a per-protocol population will not be defined due to challenges in accurately assessing adherence to the combined CPAP+MAD treatment.
- Subgroup-Analysis Population: Participants from the Royal Papworth Hospital and Bristol Royal Infirmary.
Addressing the Challenge of Missing Data
Recognizing that missing data is inevitable in clinical trials,researchers have established a clear strategy for handling it. All essential variables are expected to have low missing data percentages (under 5%). Variables exceeding 25% missing data may be excluded from modeling unless deemed clinically crucial, in which case they will be summarized and reported.
For participants who discontinue treatment but continue using the CPAP machine, adherence data will be treated as if they remained on their assigned treatment (intention-to-treat).Conversely, for those who discontinue treatment and stop providing adherence data, adherence will be imputed as zero hours for the remaining study period.
For the primary endpoints, a complete-case analysis will be used if missing data is less than or equal to 5% and considered missing completely at random (MCAR). If missing data exceeds 5% and is not MCAR, multiple imputation by chained equations (MICE) will be employed as a sensitivity analysis. Missingness indicators will be used to assess the mechanism of missing data using statistical tests like Little’s test.
Interim Analysis Specifics & Final Analysis Timeline
Participants included in the sample size re-estimation at the interim analysis must provide at least 14 days of CPAP adherence data for both treatment arms. If more than 10% of participants fail to meet this threshold, researchers will explore methods to include their data with appropriate weighting.
The final analysis will commence after data cleaning and database lock, anticipated to occur once all participants have completed 24 weeks of follow-up. Both statistics and health economics teams will utilize the same datasets for these final analyses, ensuring consistency and transparency.
This detailed approach to trial design and data analysis underscores a commitment to robust and reliable research findings.
