To analyze whether the trends in NT-proBNP level were associated with 6-month survival we used the extended Cox proportional hazards regression model. NT-proBNP was treated as a time varying co-variate. For each NT-proBNP measurement that was obtained the nearest available albumin value, creatinine, SBP and DBP values were used. Charlson Co-morbidity scores were based on all the ICD-CM codes available for all the encounters during the study period. The results of the multivariate Cox regression analysis are in Table 2. For binary and categorical variables, dummy variables were used. Observations with missing values were not used in model construction. Variables were selected for inclusion in the model based on clinical relevance in prior HF literature and consensus amongst the authors. Log(2) of the NT-proBNP was used because the distribution of NT-proBNP levels was skewed, in addition this allowed us to estimate the hazard ratio (HR) for each doubling of the NT-proBNP level (this formulation may be easier for clinicians to intuit).

6-month survival rates amongst groups of patients with improving, stable and worsening NT-proBNP values. There was progressively a lower rate of 6-month survival within the three categories of temporal trends in NT-proBNP (6-month survival in HF: down-trending (23.2%), stable (13.4%) and up-trending (7.2%), Table 1).