OBJECTIVES: Recently reported 5-year update of CROWN trial (ASCO 2024) in patients with ALK positive advanced non-small-cell lung cancer (aNSCLC) showed median PFS was not reached in the lorlatinib arm, which is confirmed as the longest PFS with any monotherapy of targeted agents in aNSCLC and across all metastatic solid tumors (Solomon et al, JCO 2024), with HR 0.19 [95% CI, 013-0.27). 5-year PFS was 60% in lorlatinib arm and 8% in crizotinib arm. These unprecedented results, which are encouraging for patients, create a methodological challenge for survival modeling. Previous standard parametric models have not provided the best fit and extrapolation of PFS, thus more flexible approaches are explored.

METHODS: Patient-level data from 5Y CROWN were used to extrapolate PFS for lorlatinib via traditional parametric models, spline models, and piecewise models.

RESULTS: For the standard parametric models, generalized gamma predicted median PFS of 12.9 years and 42.4% PFS at 30 years. The exponential distribution predicted median PFS of 6.1 years and 3.2% PFS at 30 years. UK clinical experts reported that generalized gamma was too optimistic and exponential was too pessimistic. Flexible models were implemented: splines and piecewise models. Splines did not improve on the standard models. Piecewise models were implemented using the Kaplan-Meier to 23 months then applying the standard distributions after. The piecewise gamma distribution predicted median PFS at 8.5 years and 15.8% PFS at 30 years, long-term outcomes deemed to be more plausible based on UK survival data for persons of similar age and gender.

CONCLUSIONS: Given the unprecedented impact of lorlatinb on PFS duration and trend after 2 years, fitting survival curves to PFS poses unique challenges. Piecewise models have provided better long-term estimates but uncertainties remain. Flexible methods are expected to be required for OS, and additional methods such as mixture cure should be explored.