114P - Real-World Insights on Pan-Cancer Immune Checkpoint Inhibitor Treatment: Initial Findings of a Belgian Multicenter Study

Background

To bridge the gap between clinical trial patients and real-world populations, we conducted a comprehensive study in Belgium to characterize cancer patients treated with immune checkpoint inhibitors (ICIs), which have demonstrated survival advantages in various cancer types.

Methods

Retrospective multicenter study in three Belgian hospitals, processing anonymized electronic health records including 10 data sources, using natural language processing (NLP) and machine learning. NLP model validation compared algorithm outputs to a physician-generated standard. The algorithm mapped 597 variables to SNOMED-CT, generating OMOP CDM databases, validated per hospital (federated), ensuring patient privacy. Cancer patients (≥18 years old, regardless of stage) receiving ICIs between March 2017 and August 2022 were included.

Results

Our study included 1,659 patients (median age [IQR]: 67 [60-74]); age categories: 2.4% (18-40 years), 6.1% (41-50), 17.8% (51-60), 35.1% (61-70), and 38.6% (>71). Most patients were male (65.9%). Current/former smokers totaled 43.2%, 18.6% were never smokers, and 38.2% had unknown smoking status. Alcohol abuse was observed in 10.8% of patients and negative/unknown in 89.2%. The most common ICI treatment (monotherapy or with other ICIs or antineoplastic drugs) was pembrolizumab (43.6%), followed by nivolumab (29.7%), atezolizumab (10.7%), ipilimumab (8.7%), durvalumab (4.2%), avelumab (2.1%), cemiplimab (0.8%) and dostarlimab (0.1%). Lung cancer was the most frequent cancer type (54.94%), followed by renal cancer (9.23%), bladder cancer (7.27%), melanoma (6.08%), and head and neck cancer (5.97%). Median overall survival by ICI ranged from 8 to 38 months.

Conclusions

These initial findings highlight the feasibility of automatically extracting and locally validating federated hospital databases- an invaluable tool for real-world data on ICI-treated patients. Comparable datasets would require linking government databases, which cannot be done in a federated manner nor enriched with hospital-level data. Ongoing analyses will explore immune-related adverse events, comorbidities, tumor stage, anatomical pathology, and outcomes in various cancer types and treatment lines.

Legal entity responsible for the study

LynxCare Clinical Informatics NV.

Funding

LynxCare Clinical Informatics NV.

Disclosure

All authors have declared no conflicts of interest.