Main Article Content
Abstract
Background. Molecular changes in protected ocular compartments may complement conventional post-mortem interval (PMI) estimation, but association does not guarantee accurate individual prediction.
Objective. To internally validate a ridge-regression model that estimates PMI from nuclear magnetic resonance (NMR)–quantified metabolites in human aqueous humor.
Methods. This secondary analysis used 33 human samples and 41 metabolites from Animal Metabolite Database (AMDB) Experiment 242 (human samples). Age, sex, cause of death, temperature, and sampling details were unavailable. Concentrations were log(1 + x)-transformed. Ridge regression was tuned by four-fold inner cross-validation and evaluated by five-fold outer cross-validation repeated 20 times; metrics used the mean of 20 out-of-fold predictions per sample. Spearman associations used Benjamini–Hochberg false discovery rate (FDR) control.
Results. PMI ranged from 5.0 to 58.6 hours. Mean absolute error was 7.04 hours (95% CI 5.06–9.33), root mean squared error (RMSE) 9.48 hours (6.63–12.45), R² 0.233 (interval extending below zero), and Pearson r 0.485. The mean-only benchmark RMSE was 10.94 hours, so the improvement of 1.46 hours (13.4%) has no interval. Six metabolites passed FDR control (hypoxanthine, ρ = 0.707). Predictions were compressed toward the center: samples under 12 hours were overestimated by 8.2 hours on average and the 58.6-hour sample underestimated by 30.6 hours.
Conclusion. These data suggest a detectable PMI-related signal in human aqueous humor, but individual precision is inadequate and the model remains a proof of concept. Locked external validation with standardized sampling and clinical and environmental covariates is required before medicolegal use.
Keywords
Article Details
Authors retain copyright and publishing rights in articles published in Sriwijaya Journal of Forensic and Medicolegal (SJFM), without transfer of copyright to the journal or publisher.
The authors remain the holders of commercial rights in their work. Third-party reuse is subject to the applicable article license. The journal’s standard license is CC BY-NC-SA 4.0.
Copyright in the journal website’s presentation is separate from copyright in individual articles. Website branding and footer notices do not change the authors’ rights.
