OBJECTIVES

Kaplan-Meier (KM) overall survival (OS) curves in immuno-oncology (I-O) therapy often exhibits a plateau effect not well-described by single parametric functions; such heterogeneous survivorship profiles lend themselves to mixture distribution modeling. This study aimed to empirically quantify the degree of bias associated with failing to account for cure fractions in heterogeneous survivorship profiles.

METHODS

OS data from digitized KM curves reported in a published trial of nivolumab-based regimens in recurrent small-cell lung cancer (CheckMate 032 the pre-expansion cohort)[1] were utilised. We then [1] developed a piece-wise exponential model to estimate OS; [2] developed mixture models to estimate OS for cured fractions (Weibull form); [3] defined a cure fraction bias (CFB) as the difference between mean OS from [1] and [2]; [4] calculated the CFB for a range of assumed cure fractions. Cured patients were assigned UK general population age-adjusted life expectancy.

RESULTS

The piece-wise exponential model produced mean OS of 1.16 years; when assuming a 100% cure fraction for patients surviving to 12 months, the mixture model produced mean OS of 7.78 years. Varying the cure fraction from 10% to 100% resulted in CFB estimates of 0.12 to 6.62 years. Thus, failing to account for the cure fraction, may result in a greater than five-fold underestimate of OS.

CONCLUSIONS

Mixture models may be a more appropriate method for describing and reflecting underlying mechanisms and survival extrapolations in I-O. The degree of bias associated with failing to account for cure fractions was shown to increase as a function of the size of the cure fraction. Standard parametric models applied to heterogeneous survivorship data may lead to unduly pessimistic survival projections associated with I-O therapy.

[1] Antonia et al. Nivolumab alone and nivolumab plus ipilimumab in recurrent small-cell lung cancer (CheckMate 032): a multicentre, open-label, phase 1/2 trial. Lancet Oncol 2016; 17: 883–95.