Researchers have been exploring advanced statistical methods to more accurately estimate the proportion of patients who are effectively cured following cancer treatments. A study published in Nature investigates the use of flexible parametric cure models (FPCM) in estimating cure proportions within cancer clinical trials, offering a potentially more robust approach than traditional methods. The study leverages data from the CheckMate 141 trial, a phase III study that assessed nivolumab in patients with recurrent squamous cell carcinoma of the head and neck (SCCHN).
Analysis of CheckMate 141 Data
The CheckMate 141 trial was a randomized, open-label, phase III trial designed to evaluate whether nivolumab, an immunotherapy agent, improved overall survival compared to standard therapy in patients with recurrent SCCHN. The trial randomized 361 patients in a 2:1 ratio to receive either nivolumab or standard single-agent systemic therapy. Individual-level data from this trial were reconstructed for the current study.
The researchers focused on estimating the cure proportion within the nivolumab arm of CheckMate 141 using FPCM. This parametric model utilizes a natural cubic spline with user-specified knots to define the baseline log cumulative hazard. The maximum likelihood estimate (MLE) of the cure proportion was then computed. Different knot numbers (4-9) and placements were tested to evaluate their impact on the cure proportion estimates. These estimates were compared against those obtained from the Kaplan-Meier (KM) method and the Weibull non-mixture cure model (WNCM).
The team evaluated the impact of knot number, using FPCMs with 4–9 knots distributed evenly according to centiles of observed event times, with an additional knot at the 95th centile. To evaluate the impact of knot placement, FPCMs with 6 knots were used with varying knot configurations. The cure proportion estimate from the KM method was defined as the KM estimate of survival beyond the last observed event time.
Simulation Study
To further investigate the properties of FPCM, a simulation study was conducted. The study simulated clinical trial data with varying follow-up periods and sample sizes to compare cure proportion estimates obtained from FPCM with different positions of the last knot. Survival times were generated from a mixture cure model:
$$S(t) = p_C + (1 - p_C)S_u(t)$$
where (t) is time in months, (p_C) is the cure proportion, and (S_u(t)) is the survival function for uncured individuals. Two scenarios were considered for (S_u(t)): an exponential distribution and a beta distribution. The exponential distribution represented an asymptotic cure point scenario, while the beta distribution represented a finite cure point scenario, reaching zero at 18 months. Parameters were selected based on the CheckMate 141 trial results.
Performance Measures
The performance of FPCM was evaluated based on convergence frequency, mean, bias, median, and standard error of the estimated cure proportions. The Akaike information criterion (AIC) was used to assess model fit. The study found that the position of the last knot in the FPCM significantly impacted the model fit and cure proportion estimate. The researchers recommend placing the last knot at or after the last observed event time to avoid overestimation of the cure proportion.
This research highlights the potential of flexible parametric cure models to provide more accurate and nuanced estimates of cure proportions in cancer clinical trials. By carefully considering model configuration, particularly knot placement, researchers can leverage FPCM to gain deeper insights into treatment efficacy and long-term patient outcomes.
