People experiencing depression and those who care for them are often faced with many treatment options and a lack of information for how to choose the best treatment strategy for their individual, evolving characteristics. The variability of medications in terms of effectiveness both across individuals and within an individual over time is, for the most part, poorly understood. This research project proposes to use electronic health records to discover adaptive treatment strategies that tailor treatment decisions to patient characteristics such as demographic characteristics, concurrent health problems, and response to previous treatments. Our work will serve to identify patient characteristics that can be used to tailor therapies so that patients receive the treatments most likely to confer the greatest benefit. Furthermore, we will compare tailored treatment strategies estimated from two different sources: one which is drawn from a general population and thus is likely to be applicable to a wider patient group and another drawn from a more selected population of Americans with health insurance, to see whether the estimated strategies are similar. If they are indeed similar, it would allow us to draw conclusions for a wider population based on the second dataset, which has more detailed information available and thus might permit even more individualized treatment strategies to be formed.
People experiencing depression and those who care for them are often faced with many treatment options and insufficient information for how to choose the best treatment. This research project proposes to discover adaptive treatment strategies (ATSs) that tailor treatment decisions to patient characteristics such as demographic characteristics, comorbidities, and response to previous treatments. The aims of this retrospective cohort study are to (1) estimate a tailored pharmaco-therapeutic treatment strategy for depression that adapts to patient-specific variables, and (2) compare the estimated strategy with one derived from the Mental Health Research Network, which includes longitudinal, patient-reported depression symptoms but is less representative sample to assess the validity and generalizability of the latter sample.
The analyses for aims (1) and (2) will estimate the ATS that reduces the risk of treatment failure, defined as the composite of self-harm, suicide, or hospitalization for depression. First-line treatments considered will be an antidepressant that is either a selective serotonin reuptake inhibitor, a third-generation antidepressant, or a tricyclic. For individuals not responding to the initial treatment, second-line treatments considered will be switching to a different medication in the same class, or in a different class. Estimation will be by Q-learning and its more robust counterpart, dynamic weighted ordinary least squares.
To undertake aim (2), comparisons between estimated rules will be taken both qualitatively (examining whether the same tailoring variables are used in both cohorts, evaluating the similarity of treatment decision thresholds) and quantitatively, by simple statistical metrics such as Cochran's Q test, a non-parametric test of concordance to assess whether the recommended treatments from each estimated rule differs within the sample. Our work will serve to identify patient characteristics that can be used to tailor therapies so that patients receive the treatments most likely confer the greatest benefit.
Primary outcome: Composite of self-harm, suicide attempt, hospitalization for depression;
Secondary outcome: BMI
Samy Suissa - Chief Investigator - Sir Mortimer B Davis Jewish General Hospital
Erica Moodie - Corresponding Applicant - McGill University
Christel Renoux - Collaborator - McGill University
Janie Coulombe - Collaborator - McGill University
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation