A recent study published in BMC Medical Research Methodology has shed light on the reporting practices of sample size calculations in adaptive clinical trials. The study, which analyzed 265 eligible trials, reveals significant inconsistencies in how sample sizes and operating characteristics are reported, potentially impacting the interpretability and reproducibility of trial results.
Trial Design and Sponsorship
The analysis of 265 adaptive trials showed that 61.9% were sponsored by industry, while 38.1% received public sector funding. A majority (56.6%) were Phase III trials. Industry-sponsored trials predominantly evaluated pharmacological treatments (drugs) or drug-involved combinations (over 90%), compared to 69.3% in public sector trials. The most common trial design features included superiority primary hypotheses, comparisons of one treatment against one comparator, and the use of frequentist methods for interim and final analyses.
Adaptive Features and Interim Analysis
The most frequently used adaptive design was a group sequential design (59.8% of industry-sponsored trials and 70.3% of public-sponsored trials), allowing for early stopping of the entire trial. Early stopping for either efficacy or futility was planned in 89.8% of trials. Notably, 23.0% of trials considered multiple adaptations simultaneously, and 23.8% planned for sample size re-estimation. The majority (61.1%) had one interim analysis, typically planned at the halfway point of the trial. Public-sponsored trials tended to have more interim analyses compared to industry-sponsored trials.
Sample Size Reporting and Enrolment
The study found that 86.4% of trials performed sample size calculations analytically, and 81.5% described the methods in detail, enabling reproducibility. However, over two-thirds of trials did not disclose the statistical software used for the calculation, and a substantial proportion lacked justification for the chosen parameters (48.2% of industry-sponsored and 36.5% of public-sponsored trials). The maximum sample size was the most frequently reported metric, while the minimum sample size was directly stated in only 6.0% of trials, although it could be inferred from the interim analyses plan in 92.3% of trials.
Actual vs. Planned Enrolment
Of the 154 completed trials, 44.7% enrolled the planned sample size without significant deviation (defined as >5% change). However, 29.6% had fewer participants than planned, and 25.7% had more. In trials with inflated sample sizes, 38.5% of the increases were due to trial adaptations, primarily sample size re-estimation. Trials with reduced sample sizes were often due to pre-planned early stopping for futility (28.9%) or efficacy (44.4%).
Reporting of Operating Characteristics
Less than half of the trials provided explicit information on the operating characteristics of the adaptive design. The stagewise type I error rate and cumulative power at the interim analysis were the most frequently reported terms. In the 'ROSE' study (NCT02509078), the researchers provided information regarding sample size, nuisance parameters, targeted effect size and justification. The maximum required total sample size was 1408 subjects.
Implications for Trial Interpretation
The inconsistencies in reporting sample size calculations and operating characteristics in adaptive clinical trials raise concerns about the transparency and reproducibility of trial results. Clear and comprehensive reporting is essential for accurate interpretation and application of trial findings, highlighting the need for improved guidelines and adherence to best practices in adaptive trial design and reporting.
