OBJECTIVES: Long-term management of patients with aNSCLC treated with nivolumab (anti-PD1 immunotherapy) is poorly documented. The heterogeneity of practices makes the analysis of treatment sequences complex and therefore appropriate for machine learning and innovative algorithmic approaches. The objective of this analysis is to identify treatment patterns of long-term surviving patients after immunotherapy initiation.

METHODS: Patients initiating nivolumab in 2015-2016 and alive two years later were identified from French hospital databases (PMSI). A K-clustering algorithm automatically classified and grouped patients (cluster) with similar sequences across the first 2 years. Therapeutic options were nivolumab, chemotherapy (incl. pemetrexed, bevacizumab and other chemotherapies) and treatment-free interval. For each cluster, cumulative median durations (CMD) of therapeutic options were calculated, and patients’ characteristics were described (sociodemographic, previous management care and comorbidities).

RESULTS: Among 10,452 patients who initiated nivolumab over the inclusion period, 2212 (21.2%) had at least 2 years of survival. Four clusters have been identified: Cluster 1) nivolumab was the main treatment over the first 24 months, administered almost continuously with a CMD of 21.0 months; Cluster 2) nivolumab was the main treatment (CMD: 16.5 months) but was discontinued and followed by chemotherapy (CMD: 2.5 months) and/or treatment-free interval (CMD: 5.3 months); Cluster 3) nivolumab treatment was short (CMD: 6.4 months) and followed by a long treatment-free interval (CMD: 14.4 months); Cluster 4) nivolumab treatment was short (CMD: 5.5 months) and followed by one or several lines of chemotherapy (CMD: 9.5 months). Patient characteristics differed among the clusters in age, comorbidities and previous management.

CONCLUSIONS: Analyzing a large sample of patients surviving at least two years after nivolumab initiation, the K-clustering approach classified treatment sequences and identified four homogeneous groups of patients presenting different long-term management care. An in-depth study of patients’ clinical profiles could provide a better understanding of their management.