Patients treated with anti-CD19 chimeric antigen receptor (CAR) T-cell therapies have shown either sustained remission or rapid progression. Traditional survival modeling may underestimate outcomes in these situations, by assuming the same mortality rate for all patients. To illustrate this issue, we compare standard parametric models to mixture cure models for estimating long-term overall survival in patients with relapsed or refractory large B-cell lymphoma treated with axicabtagene ciloleucel (axi-cel). Compared to standard models without cure proportions, mixture cure models have similar fit, but substantially different extrapolated survival. Standard models (Weibull and generalized gamma) estimate mean survival of 2.0 years (95% CI (1.5, 3.0)) and 3.0 years (95% CI (1.7, 5.6)), respectively, compared to 15.7 years (95% CI (9.3, 21.1)) and 17.5 yrs (12.0, 22.8) from mixture cure models (using Weibull and generalized gamme distributions). For cancer therapies where substantial fractions achieve long term remission, our results suggest that assumptions of the modeling approach should be considered. Given sufficient follow-up, mixture cure models may provide a more accurate estimate of long-term overall survival compared with standard models.
Keywords: cost-effectiveness analysis; economics; oncology; outcomes research; statistical methods: survival analysis.