Frozen analysis plan and artifact checksums: sex- and age-related equity in machine-learning detection of myocardial infarction from the ECG (PTB-XL)

Analysis plan and artifact checksums for the study "Why blinding sex is not enough: equitable discrimination, unequal operation in machine-learning detection of myocardial infarction from the ECG" (manuscript JECG-D-26-00486, Journal of Electrocardiology). The study audits sex- and age-related equity in the detection of myocardial infarction from the 12-lead ECG, on the public PTB-XL dataset, across four models sharing a single patient-grouped, checksum-frozen partition: interpretable gradient-boosted trees over morphological features, with and without sex and age, and a raw-signal one-dimensional convolutional network under two normalisation schemes. This deposit contains the analysis plan fixed on 17 June 2026, before any stage of the pipeline was run; SHA-256 checksums for the cohort, the partition, the model outputs and each analysis artifact, including the analyses added at revision; and a record of the software environment and the random seed used throughout. The deposited plan restates in English the analytic content of the internal working document fixed on that date. Sections of that document covering file paths, environment set-up, working conventions and writing guidance are left out, since they carry no analytic decision, and nothing analytic was added, removed or altered. The SHA-256 of the original file is included here, and the original is available from the corresponding author on request. Neither the data nor the analysis code is redistributed. PTB-XL is available from PhysioNet and the interpretable feature set from PTB-XL+/ECGdeli; both should be obtained from source. The analysis code and the frozen artifacts are available from the corresponding author on request, and the checksums listed here allow any copy so obtained to be verified against the one used in the analysis. This deposit was made after the analysis was complete and after the first round of peer review. It is not a pre-registration: the plan was fixed privately, and a timestamp on a local disk is not independent verification of when a document was written. No claim of formal pre-registration is made, here or in the manuscript. Supplementary Table S3 of the article classifies every analytic decision as pre-specified or post hoc against this plan.