OBJECTIVES: As melanoma starts spreading to various parts of the body from where it started, it is known as advanced melanoma. This study compares the results of the mixture and non-mixture cure models with the standard parametric models while estimating the long-term survival probability of the patients treated with nivolumab.

METHODS: The pseudo-individual patient-level data were generated using the Guyot algorithm. Mixture and non-mixture cure models were fitted by using “flexsurvcure” and standard parametric models were fitted by using the “flexsurv” package in R. In mixture and non-mixture cure models, the survival of cured and uncured patients was modeled separately by using different statistical distributions. The Akaike Information Criterion (AIC) and Bayesian information criterion (BIC) were used to compare the models. Lower the AIC and BIC indicate better the model fit. Visual inspection of the survival curves was also considered for comparison. Background mortality was considered for cured patients in estimating long-term survival.

RESULTS: Except for gompertz, all mixture and non-mixture cure models showed lower AIC and BIC than standard parametric models. The log-logistic distribution was best fitted in the mixture, and non-mixture cure models with the lowest AIC of 1786.7 and 1788.7, respectively. In contrast, AIC for the standard parametric model was 1795.44. Also, visual inspection of survival curves suggested that mixture and non-mixture cure models gave a better fit than standard parametric models in the presence of cure fraction in the data. The cure rate estimates from the best fitted model was 30.3%.

CONCLUSIONS: Mixture and non-mixture cure models may reflect life expectancy more accurately than standard parametric models. Hence, cure models are more likely to be used for survival estimation in the case of cancer therapies where survival curves seem to plateau after a particular follow-up time-point.