Data Analytics

Read this page in French: Analyse de données

Combine modern statistical approaches, machine learning, and multiplatform data visualization in order to better analyze data.

Goal: To build interactive and user-friendly web applications using the programming languages R and Python.

The solutions provided represent online dashboards and allow end-users to modify pre-specified parameters to obtain real-time update of the analysis output such as graphs, summary tables, or reports. This enables fast data review and interactive data analysis for various end-users, including health care providers, researchers, decision makers and patients.

We present three showcases that illustrate the range of topics for methodological developments and solutions.

Showcase 1: DSQ Dashboard

Illustrating DSQ usage time series data through electronic medical record systems

Authors: Motulsky (UdM), Schuster (McGill) et al.

Daily access to drug suspense data by clinicians, pharmacists and other health stuff is essential to safeguard treatment and drug prescriptions in the wider patient population. Increasingly available electronic medical record (EMR) systems aim to incorporate province wide data from the Dossier Santé Québec (DSQ) to facilitate fast and real-time access to this important information through the various local EMR systems. However, usability and acceptance of DSQ interfaces within the EMR systems are expected to vary across sites, systems and end-user group. In order to better understand differences in usage patterns and to identify possibly less optimal EMR systems for incorporating DSQ data, we implemented a secure online dashboard solution that allows for customized analysis of DSQ usage data across relevant end-user subgroups, regions and time horizons.

DSQ Usage Data

Link:

Showcase 2: BayesiCALLY

Facilitating guided content validation of questionnaires using Bayesian Confirmatory Factor analysis

Auteurs: Zhang & Schuster (McGill)

Careful instrument development and questionnaire validation are essential to ensuring the validity and reliability of self-reported outcomes in clinical practice as they inform day-by-day medical decision making. The proper validation of questionnaires requires relatively large sample sizes (between 100 to 300 questionnaire responses) and potentially multiple iterative rounds of validation. This high demand in resources calls for more efficient statistical approaches to establish validity of a questionnaire. We have implemented modern Bayesian methods to increase parameter estimation efficiency and developed a user-friendly application for the validation of questionnaires using smaller sample sizes by employing expert knowledge as prior input.

screenshot of questionnaire validation

Link: https://github.com/HZ888/Bayesian-Questionnaire-Validation

Showcase 3: Clinical Trial Simulator

Design and simulation of adaptive clinical trials for the comparison of multiple treatment alternatives

Auteurs: Aboutalebi & Schuster (McGill)

Development of a new Reinforcement Learning approach for adaptive trial designs (Platform Trials) that will be used to plan a study on patient-reported barriers to treatment adherence within the Montreal HIV population. Dynamic (event-driven) randomisation is a key element of modern adaptive trial designs. We complement existing Bayesian methods with modern machine learning methods in the context of the exploration-exploitation trade-off.

Clinical Trial Simulator

Link:

logos de Unité de soutien, SRAP, SPOR, IRSC, CIHR, McGill département de médecine de famille,

Développements méthodologiques
de l’Unité de soutien SRAP | Québec


Département de médecine de famille
5858, chemin de la Côte-des-Neiges
3e étage
Montréal, Québec H3S 1Z1
Tél.: (514) 399-9134
supportunit.fammed@mcgill.ca

Liens rapides
Unité de soutien SRAP | Quebec
Instituts de recherche en santé du Canada (IRSC)
McGill | Département de médecine de famille

© 2019 Développements méthodologiques de l’Unité de soutien SRAP du Québec