Risk prediction for heart failure incidence within 1-year using clinical and laboratory factors.

Validated risk scores for heart failure incidence are still lacking, especially for short-term prediction. In this paper we aim at developing a 1-year risk prediction model for heart failure (HF) incidence using both clinical risk factors and laboratory variables. The public MIMIC II clinical database is studied. Two multivariable Cox models are built to assess the 1-year risk of HF, one with conventional clinical risk factors only, another combined with laboratory parameters, including serum creatinine (SCR), blood urea nitrogen (BUN), glucose, prothrombin time (PT), activated partial thromboplstin time (APTT) and total bilirubin (TBIL). The discrimination performances of the different models are internally validated at last with bootstrapping. In addition to known risk factors, more clinical and laboratory indices, including pulmonary circulation diseases, peripheral vascular disease, chronic pulmonary disease, hypothyroidism, electrolyte and fluid disorders, BUN and APTT are identified to be independent predictors of heart failure incidence. Moreover, we found that the long-term risk factor, hypertension, has opposite effect on short-term risk. The C-statistics of 0.712 with internal validation has demonstrated the effectiveness of the prediction model combined clinical and laboratory factors.