Observability of Fifth Universal Definition of Myocardial Infarction components in linked electronic health records: a retrospective MIMIC-IV study

Background Clinical definitions can be operationalised from routine data only when the required evidence is represented, linked, and available at the intended decision time. We assessed the fitness of linked electronic health record (EHR) data for retrospective evaluation of Fifth Universal Definition of Myocardial Infarction (UDMI) components. Methods This retrospective data-field audit included 6189 adults in MIMIC-IV with a pragmatic dynamic troponin T pattern above an available local upper reference limit during an estimated 2008–2019 care period. The primary analysis mapped required definition components to linked sources and quantified component-specific operational observability. Symptom and machine-electrocardiogram (ECG) evidence were anchored to 1 hour before to < 2 hours after emergency-department arrival, and ECG status was represented as positive, negative, or unavailable. Mortality models were secondary and descriptive; eICU was used only for a portability audit. Results The pragmatic troponin rule and chief-complaint field were observable for all 6189 patients but were only partial proxies for the corresponding definition components. A linked ECG report was available in the strict window for 4911 patients (79.4%); 1278 (20.6%) had no linked report. Sex-specific upper reference limits, systematic time-locked imaging, thrombus evidence, mechanism adjudication, and procedure-related myocardial infarction adjudication were unavailable. Captured symptom-or-machine-ECG evidence was present in 1479 patients, whereas 4710 had neither captured source. ECG status was positive in 208 patients and negative in 4703. Standardised mortality was 14.72% and 10.54%, respectively (risk difference + 4.18 percentage points, 95% confidence interval − 0.80 to + 9.16). Conclusions Historical EHR fields supported only partial retrospective evaluation of Fifth-UDMI components. Source- and time-aware observability auditing exposed missing inputs and distinguished an unavailable record from a negative finding. The audit did not establish diagnosis, clinician action, or care quality. Local prospective source verification and human review are required before clinical implementation.