Prescription drugs are important for preventing and treating health problems, but are not as beneficial as they could be. This is because there is often little information on a drug’s safety and effectiveness after it is approved for sale. It is also difficult to personalize treatment to a patient’s particular situation and guide physician prescribing. Personalizing treatment is important because patients with the same conditions do not always benefit equally from the same medications. Patient experiences differ based on many factors, including demographic, lifestyle, and clinical factors. A patient of a certain age, sex, ethnicity, and medical history may benefit from a medication that causes harm to a patient with different traits. In this study, we aim to develop predictive algorithms to personalize drug treatment. We have selected three medication classes with typically unexplained differences between patients in their experiences with the medication. These include blood thinning medications for patients with atrial fibrillation (a heart rhythm disorder) and venous thromboembolism (blood clots), blood sugar-lowering medications for patients with type 2 diabetes (condition characterized by high blood sugar), and antidepressant medications for patients with depression, anxiety, and insomnia. We will develop predictive algorithms that predict, for a single patient, the risks and benefits of different medications within each therapeutic class. Such algorithms have the potential to improve experiences and health outcomes for thousands of patients who take these drugs in the UK. This, in turn, could reduce health system costs due to reduced hospital and emergency department use by these patients.
Despite the considerable benefits of modern drug therapy, the full potential of medications to improve population health is currently not realized due to limited capacity to assess real-world safety and effectiveness, personalize treatment, and influence use in practice. The main objective of this study is to develop and validate algorithms that predict individual treatment response for three therapeutic drug classes and treatment indications: anticoagulant drugs for atrial fibrillation and venous thromboembolism (VTE), antidiabetic drugs for type 2 diabetes, and antidepressant drugs for depression, anxiety, and insomnia. We will create each cohort by selecting individuals with a diagnosis for the condition of interest in CPRD AURUM and HES APC data, and subsequently prescribed a drug of interest between 2018 and 2022, as observed in CPRD AURUM data. Individuals will be followed from the date of their first prescription of a study drug until the first occurrence of the outcome being modelled, their departure from the cohort, or the end of 2022, whichever occurs first. Exposures will be modelled as time-varying use and time-varying dose of anticoagulant, antidiabetic, and antidepressant medications. For each drug class and condition of interest, we have identified the top 5-10 outcomes that patients most value and prioritize. For every outcome being modelled, we have also identified time-fixed and time-varying predictors, including a range of patient demographic, lifestyle, and clinical characteristics and history. Both standard statistical approaches and machine learning will be used to generate algorithms that predict the occurrence of these outcomes. We will use measures such as the c-statistic, ROC curve, and the Brier score to measure model performance. The best-performing models will then be validated externally in comparable cohorts assembled from partner institutions. Validated algorithms will be implemented in the real-word and reinforcement learning methods will then be employed to improve algorithm accuracy.
Anticoagulant drug cohort (for atrial fibrillation): ischemic stroke, bleeding, systemic embolism, myocardial infarction (MI), heart failure, death, hospital admission, side effects of anticoagulant drugs
Anticoagulant drug cohort (for VTE): recurrent venous thromboembolism (VTE), bleeding, post-thrombotic syndrome, chronic thromboembolic pulmonary hypertension, chronic thromboembolic pulmonary disease, death, hospital admission, side effects of anticoagulant drugs
Antidiabetic drug cohort: glycaemic control, stroke, MI, microvascular and cardiovascular complications of diabetes, composite of MI, stroke and cardiovascular death, hospital admission, death, side effects of antidiabetic drugs
Antidepressant medication cohort: antidepressant treatment discontinuation or treatment change, suicide attempts, side effects of antidepressant drugs.
Samy Suissa - Chief Investigator - Sir Mortimer B Davis Jewish General Hospital
Robyn Tamblyn - Corresponding Applicant - McGill University
Bettina Habib - Collaborator - McGill University
David Buckeridge - Collaborator - McGill University
Emily McDonald - Collaborator - McGill University
Kristian Filion - Collaborator - McGill University
Nadyne Girard - Collaborator - McGill University
pauline reynier - Collaborator - Sir Mortimer B Davis Jewish General Hospital
Robert Platt - Collaborator - McGill University
Teresa Moraga - Collaborator - McGill University
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation