This study aims to examine whether already approved anti-diabetic drugs such as metformin and sulfonylureas can be used to lower the morbidity and mortality of breast cancer, prostate cancer, bowel cancer and lung cancer. We will conduct retrospective longitudinal cohort analyses with data on the effect of commonly prescribed first-line generic anti-diabetic drugs. As the real patient prescription behavior is not observed from EHR data, we will estimate the intention-to-treat effect of drugs on cancer based on Neyman-Rubin framework. For each pair of target drugs, we will adjust for confounding based on propensity score estimated by pre-treatment risk factors selected by physicians with material subject matter knowledge. We would like to adjust for age at initial prescription, initial prescription calendar year, gender, deprivation index, smoking status, body mass index, HbA1c, and comorbidities (e.g., cardiovascular diseases, depression, COPD, chronic kidney disease). Our goal is to balance probabilistic distribution between both intervention groups to ensure that one of our main assumptions called conditional ignorability holds. We can assume other critical causal assumptions including positivity and stable unit treatment value assumption (SUTVA) hold based on our study design. Apart from confounding bias, we will also investigate the impact of missing data. Missing baseline confounders and censoring will be considered based on missing mechanisms such as missing completely at random (MCAR), missing at random (MAR) and missing not at random (MNAR). The impact of confounding bias, missing baseline confounders and survivorship bias will be evaluated jointly by Markov factorization.
The events of interest are morbidity and mortality of breast cancer, prostate cancer, bowel cancer and lung cancer. Readcodes for each type of cancer can be found https://www.phpc.cam.ac.uk/pcu/research/research-groups/crmh/cprd_cam/c…
https://github.com/spiros/chronological-map-phenotypes/tree/master/prim…
https://github.com/rOpenHealth/ClinicalCodes
The event date of onset will be defined as the first cancer diagnosis date.
Ioanna Tzoulaki - Chief Investigator - Imperial College London
Bowen Su - Corresponding Applicant - Imperial College London
Azeem Majeed - Collaborator - Imperial College London
Bang Zheng - Collaborator - Imperial College London
Shenbo Xu - Collaborator - Imperial College London
HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation