(299) Identifying Hospitalizations for Heart Failure in Healthcare Databases: Algorithm Development and Validation

Publication Number: 630

Background: While hospitalization for heart failure (HHF) is an important outcome in clinical and epidemiologic studies, the appropriate definition to identify HHF in database research is still undetermined.

Objectives: To develop and validate algorithms to identify HHF in healthcare databases.

Methods: This validation study was conducted in adults with gout or urate-lowering drugs who were admitted to a medical center in Taiwan. A random sample of 300 hospitalization events with HF-related discharge diagnosis in the first five coding positions between 2014 and 2017 was selected as the validation cohort (150 each for the ICD-9-CM and ICD-10-CM eras). Algorithms to identify HHF were developed with a combination of different lists of HF-related diagnosis codes, diagnostic positions, use of intravenous (IV) loop diuretics, IV inotropes, augmentation of oral loop diuretics, renal replacement therapy (RRT), and elevation of N-terminal pro-B-type natriuretic peptide (NT-proBNP) level (if tested). Manual chart review by two independent cardiologists was used as the reference standard for evaluating developed algorithms. HHF case ascertainment was based on the consensus definitions for HHF (as endpoint events in clinical trials) and the Atherosclerosis Risk in Communities criteria. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were assessed.

Results: Of all the sampled hospitalizations, 62 (41%) in the ICD-9-CM era and 47 (31%) in the ICD-10-CM era were adjudicated as definite HHF. The algorithms solely with diagnosis codes in the first five coding positions showed the worst PPVs (0.40 and 0.31 respectively). The algorithm which contained “diagnosis code of ICD-9-CM 428.x or ICD-10-CM I50.x in the first two coding positions” and “IV loop diuretics or IV inotropes” (with or without the criterion of “augmentation of oral loop diuretics”) had the best performance. This algorithm exhibited a sensitivity, specificity, PPV, and NPV of 0.79 (95% CI 0.69-0.90), 0.88 (0.81-0.95), 0.81 (0.71-0.91), and 0.87 (0.80-0.94) respectively in the ICD-9-CM era, and 0.74 (0.62-0.87), 0.91 (0.86-0.96), 0.78 (0.66-0.90), and 0.90 (0.84-0.95) respectively in the ICD-10-CM era. Adding RRT or elevation of NT-proBNP did not significantly enhance the sensitivity or PPV of the algorithms.

Conclusions: Our results suggest that adopting an explicit code definition of HF in the first two coding positions, combined with the use of IV loop diuretics and IV inotropes, attained an acceptable level of sensitivity and PPV for identifying HHF in both the ICD-9-CM and ICD-10-CM eras in healthcare databases.