My collaborators and I combine observational data, mostly untestable assumptions, and statistical methods to emulate randomized experiments. We emphasize the need to formulate well defined causal questions, and use analytic approaches whose validity does not require assumptions that conflict with subject-matter knowledge. For example, when time-dependent confounders affected by prior treatment are expected, we do not use methods (e.g., conventional regression analysis) that require the absence of such confounders. Our work is applied to the following areas:
- Cardiovascular disease: We study pharmacological, nutritional and lifestyle interventions to reduce the incidence of coronary heart disease and related conditions. This work is conducted in collaboration with investigators from the Nurses’ Health Study and the Health Professionals Follow-up Study at Harvard, and from the Spanish Centre for Pharmacoepidemiologic Research in Madrid. Funding: NIH R01-HL080644.
- Cancer: We compare the effectiveness of diagnostic and therapeutic interventions in cancer patients, as well as the effectiveness of various screening strategies for cancer. This work is conducted in collaboration with investigators from the Dana Farber Cancer Institute in Boston, the University of California San Francisco, and the University of Oslo. Funding: NIH P01-CA134294.
- Kidney disease: We study interventions using erythropoiesis-stimulating agents to reduce the mortality of patients undergoing hemodialysis. This work, based on data from the United States Renal Data System, is conducted jointly with investigators from MTPPI in Washington, DC, and the VA Harbor Healthcare System in New York. Funding: AHRQ R21-HS19513.
- HIV disease: We investigate the optimal use of antiretroviral therapy in persons infected with HIV. This work is conducted jointly with investigators from the HIV-CAUSAL Collaboration and PHACS. Funding: NIH R01-AI102643, R37-AI32475, and U01-HD052102.