Integrating Patient-Reported Outcomes (PROMs) and Clinical Data to Advance Real-World Evidence in Oncology
Principal Investigator: Sylvie Lambert
Institution: Centre de recherche de l’Hôpital St-Mary’s
Sponsor: PROVEM
Scientific background
Randomized clinical trials remain the cornerstone for evaluating the efficacy of anticancer treatments. However, these studies are conducted under controlled conditions and often involve selected patient populations, which limits their ability to reflect the diversity of trajectories observed in routine clinical practice.
Since 2015, the Department of Oncology at St. Mary’s Hospital has systematically collected patient-reported outcome measures (PROMs) as part of routine care. Integrating these data with clinical and administrative datasets provides a unique opportunity to characterize real-world care trajectories and evaluate treatment impact beyond clinical trial settings.
However, large-scale integration of structured and unstructured data, particularly from clinical notes, remains a major methodological challenge in real-world research and requires advanced analytical approaches.
Objectives
This project aims to develop an integrated approach to analyzing clinical trajectories by combining patient-reported data, clinical data, and information extracted from unstructured sources.
The study also aims to assess the predictive value of early patient-reported measures in anticipating clinical outcomes and treatment tolerability.
Population and data mobilized
- Population:
Adult oncology patients treated at St. Mary’s Hospital.
Approximately 1,500 patients with longitudinal data covering the period 2015–2025, depending on database availability. - Data sources:
- PROM data (ESAS and secondary questionnaires)
- Hospital clinical and administrative data (diagnoses, visits, hospitalizations)
- Longitudinal laboratory results
- Treatment exposure data
- Clinical notes and medical reports in PDF format
The mobilization of tens of thousands of unstructured clinical documents within a real-world cohort represents a unique opportunity to enrich the analysis of care trajectories.
Methodological Approach
This is a retrospective study based on the integration and structured mobilization of multi-source longitudinal data.
Structured and unstructured datasets are linked using hospital identifiers and pseudonymized prior to analysis within a secure environment compliant with applicable regulatory and ethical requirements. Relevant information contained in clinical notes is extracted using large language models (LLMs) deployed locally in a secure environment, enabling large-scale structured analysis of textual clinical data.
Patient trajectories are modeled using:
- Multivariate latent class growth analysis (LCGA / GBTM)
- Unsupervised clustering techniques (k-means, DBSCAN, hierarchical clustering)
- Survival analyses (Kaplan–Meier, Cox proportional hazards models)
- Longitudinal-to-event modeling approaches
This approach enables the integration of clinical dimensions that are typically not captured in structured datasets, thereby enhancing the depth and precision of real-world analyses.
This project serves as a demonstration of PROVEM’s ability to integrate and analyze complex multi-source data, including unstructured data, within a secure and methodologically rigorous framework.
Project status
Current phase: operational setup and preparation for multi-source data extraction and integration.
Next step: database integration and deployment of the clinical text extraction pipeline.
Institutional Signature
PROVEM is a platform dedicated to the mobilization and analysis of real-world oncology data in Quebec, integrating clinical, administrative, and patient-reported data to generate multi-level analyses that inform decision-making.