Coding Variation and Adherence to Methodological Standards in Cardiac Research Using the National Inpatient Sample

Background: Code selection is crucial to the accuracy and reproducibility of studies using administrative data, however a comprehensive assessment of coding trends for major cardiac diagnoses and procedures is lacking. We aimed to evaluate trends in administrative code utilization for major cardiac diagnoses and procedures, and adherence to required methodological practices in cardiac research using the National Inpatient Sample (NIS). Methods: In this observational study of 445 articles, ICD-9-CM codes corresponding to acute myocardial infarction (AMI), heart failure, atrial fibrillation, percutaneous coronary intervention, and coronary artery bypass grafting were collected and analyzed. The NIS was used to compare the number of hospitalizations between the most frequently encountered AMI case definitions. Key elements were abstracted from each article to evaluate adherence to required methodological practices. Results: Variation in code utilization was observed for each diagnosis and procedure assessed, and the number of unique case definitions published per year increased throughout the study period ( P < 0.001), driven largely by the significant increase in articles per year ( P < 0.001). Off-target codes were observed in 39 (8.8%) studies. Upon reintroduction into the NIS for 2008–2012, the most commonly encountered case definitions for AMI were found to yield significantly different estimates of AMI hospitalizations and hospitalization trends over time. Three hundred and ninety-nine articles (84%) did not adhere to one or more required research practices. Overall adherence was superior for publications in higher-impact journals ( P = 0.002). Conclusions: Substantial variation in code selection exists for major cardiac diagnoses and procedures, and non-adherence to methodological standards is widespread. These data have important implications for the accuracy and generalizability of analyses using the NIS.

Abstract licence: CC BY

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.