A Comparison of Methods for Modeling Factors Associated with Hospital Admissions in Incident Cases of Heart Failure

Background: Heart Failure (HF) is at important public health problem due to its increasing prevalence, and the decompensation associated with hospital admission represents an increased risk of death. The objective of this study was to compare several methods to model the variable hospitalizations and to determine the effect of factors associated with hospital admissions in incident cases of HF. Methods: Study of a retrospective cohort of patients with information extracted front electronic medical records of PC was performed. Patients 24 year and older with at least 1 visit to PC in 2006 were included. Registered hospital admissions of HF incident oases between 2006 and 2010 or until death were analyzed and comparison of Poisson, Negative Binontial (NB). zero-inflated and Hurdle regression models were conducted to identify factors associated con hospitalizations. Results: 3,061 patients were identified ill a cohort of 227,984. Regarding the factors associated with hospitalizations and according to the zero inflated NB regression model, patients who presented valvular disease (OR-2.01; CI95% 1.22-3.30), or were being treated with antiduombotics (OR=3.45; CI95%: 1.61-7.42) or diuretics (OR=2.28; CI95% 1.13-4.58) had a lower likelihood of hospitalization. Factors associated with a higher rate of hospital admissions were having valvular disnise (IRR=1.37; CI95% 1.03-1.81) or diabetes mellitus (IRR=1.38; 1.07-1.78), and being treated with calcium antagonists (IRR=1.35; CI95% 1.05- 1.73) or ACE in (IRR=1.43; CI95% 1.06- 1.92). Having being referred to a cardiologist bad a protective effect (IRR-0.86; CI95% 0.760.97). Conclusion: The regression model that obtained the best adjustmon was the zero inflated NB. According to this model, the factors associated with an increase in hospital admissions were valvulopathies. diabetes and treatment with calcium antagonists.