Défis statistiques dans la méta-analyse des données individuelles avec des applications dans la tuberculose et la dépression

 

Andrea Benedetti

Université McGill

 

Domaine : santé des populations

Programme Chercheurs-boursiers - Junior 2

Concours 2018-2019

Systematic summaries of related literature are considered an integral part of the development of guidelines in healthcare. Meta-analyses attempt to combine evidence from related studies to produce a coherent message based on the totality of evidence. The best meta-analyses are "individual patient data meta-analyses" (IPD-MA) because these collect the actual data from each study.

My research program focusses on the best use of IPD-MA in tuberculosis and depression screening :

  1. In the context of multiple drug resistant tuberculosis, we are proposing methods that can help assess which groups of patients are most helped by certain medications. This would eventually allow for individualized treatment recommendations. We will apply our methods to determine the best treatment strategies for multiple drug resistant tuberculosis.
  2. Depression screening involves using short questionnaires to identify people who may have unrecognized depression. Screening is not currently recommended in Canada's main healthcare guidelines due to concerns that screening tools do not accurately distinguish between patients with and without depression. Another concern with existing screening methods is that they do not do consider patient risk factors, which may be associated with depression.

In our project, we will combine original patient data from published studies that used the Geriatric Depression Scale (GDS) depression screening tool, in order to create a very large database that can be used to determine more accurate ways to identify elderly patients with depression.

In a second project, we will use large datasets (e.g., > 20,000 patients) that we have synthesized for evaluating screening with three commonly used screening tests to develop individual risk models. We will use actual screening questionnaire scores, along with patient demographic and medical characteristics, to estimate risk.