New Study Exposes Reliability Gap in Traditional Time-of-Death Methods
by Jo Ellen Nott
For decades, forensic pathologists have relied on body temperature, rigor mortis, and potassium levels in the vitreous humor of the eye to estimate time of death. These methods have underpinned countless prosecutions, but their accuracy degrades sharply beyond the early post-mortem window. Rectal temperature becomes unreliable once a body reaches ambient temperature, often within 24 hours, while vitreous potassium measurements lose precision after roughly 48 hours. Beyond that point, estimates become increasingly subjective, raising serious questions about the weight such testimony has been given in courtrooms.
A new study, “The Human Metabolome and Machine Learning Improves Predictions of the Post-Mortem Interval,” published in Nature Communications in February 2026, offers both a technological leap forward and an implicit indictment of the status quo. Researchers at Linköping University and the Swedish National Board of Forensic Medicine trained a neural network on metabolomic data – the chemical profiles of small molecules in the blood that break down in a predictable sequence after death – drawn from 4,876 autopsy blood samples with known post-mortem intervals, selected from a broader collection of more than 45,000 samples gathered over a decade.
The AI model predicted the interval between death and autopsy with a mean absolute error of 1.45 days and a median error of just over one day, maintaining that accuracy for remains up to 13 days post-mortem. When tested on an independent dataset of 512 individuals collected in a different year and analyzed on a separate mass spectrometry platform, the model still performed well, with a mean error of 1.78 days. By comparison, traditional methods offer no reliable quantitative estimate at all once the initial 24–48-hour window closes.
The implications for criminal defense practitioners are significant. Time-of-death estimates have long been presented to juries with an air of scientific certainty that the underlying methods may not support, particularly in cases where a body is discovered days after death. This study quantifies what defense experts have argued for years: that conventional PMI testimony resting on methods reliable only in the first day or two can be deeply misleading when applied to longer intervals. Defense attorneys handling cases in which a prosecution timeline depends on a medical examiner’s PMI estimate should take note of this research as potential grounds for challenging that testimony.
Equally noteworthy is the study’s scalability finding. Lead author Rasmus Magnusson noted that a workable model can be built with just a few hundred samples, meaning regional forensic laboratories could implement this approach without requiring massive data infrastructure. For defense teams, the accessibility of the technology matters: it is not confined to elite research institutions and could, in principle, be deployed by independent experts retained by the defense.
The study also produced a secondary benefit. In training on such a large sample set, the researchers created what Magnusson described as a unique database of post-mortem blood chemistry that tracks drugs, toxins, pharmaceuticals, and disease markers – a resource that could prove valuable for toxicological analysis in both prosecution and defense contexts.
Important limitations remain. The model currently predicts the date of death, not the specific hour, and the research team is working toward time-of-day specificity. Still, even in its current form, the technology narrows the post-mortem interval far beyond what traditional methods can achieve outside the first 48 hours. As forensic science continues to evolve, defense attorneys and their experts should be prepared to reference these findings, both to challenge outdated PMI testimony and to demand that the prosecution’s forensic evidence meet the standard of the best available science.
Source: Rasmus Magnusson et al., The Human Metabolome and Machine Learning Improves Predictions of the Post-Mortem Interval, 17 Nature Communications 1504 (2026).
As a digital subscriber to Criminal Legal News, you can access full text and downloads for this and other premium content.
Already a subscriber? Login
