Data Science

Data Science

Our main objectives in data science are to ensure legitimate access to data, enhance their discoverability, and strengthen their management. We aim to establish rigorous guidelines that guarantee appropriate and secure access to data. Once the data are accessible, we work to create an improved framework that makes them easier to find, explore, and use across the research community. This new structure will not only support discoverability but also enable more efficient data manipulation, analysis, and long-term reuse.

Optimizing Mental Health Care Pathways with Predictive Artificial Intelligence Models

InterRAI,
CIUSSS-ODIM &
Data Science Team

Website

This project focuses on optimizing the quality and coordination of mental health hospital care by harnessing standardized interRAI clinical data alongside advanced artificial intelligence (AI) techniques.

In collaboration with the interRAI consortium, the initiative leverages well-established psychometric assessment tools designed to support care planning and clinical decision-making. The interRAI instruments enable the consistent collection of high-quality clinical data, allowing for the analysis of patient trajectories, service use, and outcomes across various care settings.

The initial phase of the project focuses on applying machine learning methods to develop predictive models that better understand the hospital readmissions risk following discharge, as well as the length of stay in mental health services. Emphasizing transparency and clinical relevance, the project uses interpretable AI techniques to ensure that insights are meaningful and directly applicable to clinical practice. By combining standardized clinical assessments with explainable AI, this work provides a data-driven approach to understanding mental health care pathways.

In the next phase, the project aims to unlock the full potential of interRAI data by developing clinically meaningful quality-of-care indicators. These indicators, derived from interRAI assessments, will be presented through intuitive dashboards for use by clinicians and administrators. Tracking key patient metrics, such as depression scales and Clinical Assessment Protocols (CAPs) across care episodes, the indicators will highlight areas of strong performance and identify where improvements are needed. Standardizing this monitoring across units or hospitals will enable evidence-informed care planning and decision-making at both the clinical and program levels.

Looking ahead, the project will also leverage historical research data and integrate ISQ data into the initial interRAI dataset to enhance the previously developed machine learning models. This advanced step will refine the model’s ability to predict suicidal events and the use of care within this vulnerable population. By merging clinical, administrative, and research data, our group will offer a more comprehensive and accurate prediction model, ultimately improving patient outcomes, particularly in relation to depression, within mental health services.

Research-Oriented Data Dictionary for the Perfocentre Data Lake

Perfocentre &
Data Science Team

A large volume of clinico-administrative data generated by the Douglas clinical services converges toward the Perfocentre data lake, a CIUSSS entity mainly responsible for producing centralized, aggregate reporting for the Ministry of Health. While this data is already structured to meet reporting and accountability requirements, it remains largely underused for research and for generating insights that can directly inform patient care.

This project, developed through a collaboration between the Perfocentre and the Data Science team, focuses on enhancing the research and clinical value of existing data structures. The primary phase of the initiative aims to organize, harmonize, and document the data from a research-oriented perspective, without altering its role in Ministry reporting. By creating comprehensive, standardized, and accessible data dictionaries, the project seeks to bridge the gap between administrative reporting and meaningful reuse of real-world clinical data.

A primary goal of this foundational work is to ensure the data ecosystem is cohesive, well-documented, and ready to support advanced analyses, while fully adhering to privacy, security, and governance standards. This will involve enhancing metadata quality, refining variable definitions, and making datasets more discoverable and interpretable for both researchers and clinicians. Part of this process will include automating search functions for structured data. Additionally, unstructured data, primarily in the form of medical notes, will be made accessible and discoverable through Large Language Model (LLM) extractions, helping to supplement and complete the structured data by providing additional insights and context to form a full user profile.

Once this groundwork is established, a subsequent phase of the project will explore the integration of artificial intelligence to support and automate parts of the metadata enrichment process in an ongoing fashion. These AI-based tools will be designed to operate within a secure, privacy-preserving environment hosted entirely within the CIUSSS network, and will build on the structured foundation created in the initial phase.

By strengthening data structure and documentation for research purposes, this initiative aims to unlock the full potential of existing clinical data to support knowledge generation, improve understanding of care trajectories, and ultimately contribute to better-informed mental health care.

Perfocentre & Data Science Team

Perfocentre &
Data Science Team