The FAIR-ICI investigators assembled a multicenter retrospective cohort of 3,231 adults treated with immune checkpoint inhibitors across four Belgian hospitals. Data sources included both structured electronic health record fields and unstructured clinical notes. The study team mapped those records to the Observational Medical Outcomes Partnership Common Data Model (OMOP-CDM) and applied natural language processing to extract treatment events and outcomes. Ethical approval was granted by the Ethics Committee of Antwerp University Hospital, and the requirement for individual informed consent was waived, the authors reported.
Clinical composition skewed toward advanced disease: 2,514 patients (77.8%) had metastatic cancer at the time of ICI therapy. The manuscript did not provide further granular demographic breakdowns in the abstract, and specific dosing schemas or ICI agents beyond example cohorts (eg, pembrolizumab plus chemotherapy in NSCLC) were not enumerated in detail in the source text. The authors created an aggregated, public interactive dashboard to allow exploration of cohort characteristics and treatment patterns; patient-level datasets remain subject to institutional governance and are not openly shared without authorization.
Investigators characterized several longitudinal endpoints: overall survival (OS), time on treatment (ToT), time to next treatment (TTNT), a discrete definition for ICI discontinuation (a treatment gap exceeding 12 weeks without re-initiation), and survival after discontinuation. Time on treatment and time to next treatment were generally shorter than OS across malignancies, indicating that many patients lived substantially longer after stopping active ICI therapy. The authors highlighted melanoma as a clear example: the cohort-level median OS was reported at 78 months, while ToT and TTNT were substantially shorter (7 months and 26 months, respectively).
Among the 2,514 patients with metastatic disease, 568 (22.6%) discontinued ICIs and did not receive further systemic therapy. Median post-discontinuation OS for that subgroup was 21.4 months (95% CI 16.5–37.2), and roughly 49.1% remained alive at 24 months. The study team interpreted these observations as evidence that real-world ICI care may produce prolonged survival after cessation of active treatment for a meaningful subset of patients. The source did not provide explicit reasons for discontinuation (for example, toxicity versus clinical decision) in the abstract, nor did it present detailed post-discontinuation management strategies.
FAIR-ICI investigators compared observed real-world survival descriptively with outcomes reported in pivotal clinical trials. They reported that survival in the real world was often less favorable than trial results. Patient selection and baseline status drove much of this divergence: performance status emerged as a major prognostic factor.
Multivariable Cox regression indicated that an ECOG performance status ≥2 was associated with worse OS than ECOG 0–1 (adjusted hazard ratio 2.18, 95% CI 1.82–2.60), the authors reported. In a subgroup analysis of patients with non–small cell lung cancer receiving pembrolizumab plus chemotherapy, survival for those with ECOG 0–1 more closely matched trial populations, while outcomes for ECOG ≥2 were substantially poorer. The manuscript emphasized that differences in baseline patient fitness and other real-world factors likely explain at least part of the discrepancy between trial and routine-care outcomes.
The abstract did not list other covariates in the multivariable model within its summary; therefore, the source did not report further adjusted predictors or model specifications in the abstract text.
The study team developed a publicly accessible interactive dashboard to explore aggregated cohort results, treatment patterns, and outcomes. The dashboard URL was provided in the article; it displays summary-level visualizations but does not expose identifiable patient-level records. The authors stated that patient-level data are restricted for privacy and governance reasons, and that requests for access would need to follow institutional and ethical approvals.
FAIR-ICI was presented as a scalable framework to evaluate real-world immunotherapy effectiveness and treatment trajectories. The authors specifically mentioned that the resource could support future investigations into treatment-related toxicity, though the present report focused on utilization and survival measures rather than detailed toxicity outcomes.
Strengths of the FAIR-ICI effort include multicenter EHR harmonization using OMOP-CDM, the integration of natural language processing to capture unstructured data, and the creation of a public dashboard to disseminate aggregated results. These elements support reproducible, scalable evaluation of ICI use outside of clinical trials.
Limitations were explicit in the source: this report is a preprint and has not undergone peer review. Patient-level data are not publicly available due to governance restrictions, limiting independent validation without formal data access. The abstract did not specify certain methodological details in full—such as the full set of covariates in adjusted models, granular dosing or agent-level breakdowns across all indications, or explicit reasons for treatment discontinuation—so readers should consult the full manuscript and supplementary materials for those items.
Unanswered clinical questions noted by the authors include the drivers of post-discontinuation survival, the role of toxicity in shaping treatment trajectories, and how to best reconcile trial-derived efficacy with heterogeneous real-world populations. The FAIR-ICI dataset and dashboard were presented as resources to address these questions, subject to appropriate governance, further peer review, and prespecified analytic plans.