Professor and Director Center for Pharmacoepidemiology and Treatment Science, Rutgers University New Brunswick, United States
Background: The validity of International Classification of Diseases, 10th Revision (ICD-10) codes for myocardial infarction (MI) in US Medicare is uncertain.
Objectives: To validate various ICD-10-based definitions for MI against medical chart review among older adults in US Medicare.
Methods: ICD-10 algorithms were constructed based on established definitions and clinical input (JBS, SS, BLS). Hospital discharge codes for MI (I21.0-I21.4) were evaluated in 1st position, in 2nd position, and in other positions (3rd through 25th). We also evaluated codes for unspecified (I21.9) and other (I21.A) MI in 1st position, as it is unclear whether these codes should be included in claims-based MI algorithms. Using these definitions, new MI events were identified in a 50% national sample of fee-for-service US Medicare older adults (≥age 66) with Part D prescription drug benefits during 2019. Medical charts were retrieved and redacted by a contract research organization and then reviewed and adjudicated against the 4th Universal Definition of MI (2018) by 3 trained physicians (MK, BL, AT), supervised by a faculty cardiologist (JBS). PPVs and 95% confidence intervals (CIs) were calculated.
Results: We identified all patients meeting the standard MI definition (I21.0-I21.4) in 1st (N=27,040; adjudication target=100 charts), 2nd (N=23,141; adjudication target=100 charts), or other (N=33,391; adjudication target=150 charts) position. An additional 6,579 patients had codes I21.9 and I21.A in 1st position (adjudication target=100 charts). A stratified random sample of 1,574 patients across these 4 groups was sent for retrieval and 468 redacted charts were returned (retrieval rate 30%). The respective PPVs to date are 76% (62%-90%) for I21.0-I21.4 in 1st position (based on 37 adjudicated charts), 86% (71%-100%) for I21.0-I21.4 in 2nd position (based on 21 adjudicated charts), 42% (26%-58%) for I21.0-I21.4 in other positions (based on 36 adjudicated charts), and 87% (75%-99%) for I21.9 or I21.A in 1st position (based on 31 adjudicated charts). PPVs for I21.0-I21.4 in 1st or 2nd position combined and for I21.0-I21.4, I21.9, or I21.A in 1st position combined (weighted to account for the population distribution of the component events) were 80% (66%-95%) and 78% (65%-91%), respectively. Adjudication is ongoing for the remaining charts retrieved.
Conclusions: ICD-10 based algorithms for MI in US Medicare perform adequately. Inclusion of diagnosis codes I21.0-I21.4 in 2nd position and diagnosis codes I21.9 and I21.A in 1st position captures additional cases without reducing PPV.
