Diabetes
Emulation of the PREDIMED Protocol in two US cohorts: Could sample differences explain disagreements between trial and observational results? Octavio Pano Espinola* Octavio Pano Espinola Prof Dalia Stern Arturo Aguilar Prof. Martin Lajous Prof. Deirdre K Tobias Prof. Yu-Han Chiu
The PREDIMED trial concluded that a Mediterranean, compared to a low-fat diet, can effectively reduce the incidence of type 2 diabetes (T2D). However, it remains unclear whether these findings are applicable to other populations, where diet and the distribution of risk factors might differ. Given that conducting randomized trials in diverse populations may not be efficient, we applied the target trial framework (TTF) to emulate the PREDIMED protocol in two US cohorts. Methods We adapted the PREDIMED trial to the Nurses’ Health Study and Health Professionals Follow-up Study data. A similar high-risk population was selected to estimate the observational analogue of the intention-to-treat effect over 4-years on the incident of T2D had everyone initiated a Mediterranean or a Low-fat diet. The interventions were defined using similar foods as the PREDIMED protocol. We estimated the absolute cumulative incidence, risk differences (RD) and risk rations (RR) using the g-formula. Results Applying the protocol we selected a sample of 10,627 participants (77% female). Compared to the PREDIMED, our population had a higher prevalence of smokers (15.6% vs. 7.9%), dyslipidemia (92.3% vs. 51.2%) and hypertension (93.9% vs. 90.0%), but lower prevalence of overweight/obesity (84.1% vs. 94.6%). Overall, adherence to the Mediterranean diet differed between studies, with a mean score of 6.2 (SD 1.9) in the US cohorts and 8.7 (1.9) in the PREDIMED. Moreover, the foods characterizing the dietary patterns were different between studies. The 4-year cumulative incidence of T2D in the Mediterranean arm was 6.9 (5.5–8.6) in the PREDIMED trial and 3.8 (2.8 – 5.1) in the US cohorts. Preliminary results show a RR of 1.4 (95% CI 1.0-2.3) for the ITT comparing a mediterranean diet versus a low-fat diet. Conclusions Using TTF to emulate PREDIMED, we found differences in risk factors, diets and adherence to food groups. Further exploration of these differences, unmeasured confounding and adherence patterns between studies is necessary to understand conflicting results