{"title":"Identifying factors that help improve existing decomposition-based PMI estimation methods.","authors":"Anna-Maria Nau, Phillip Ditto, Dawnie Wolfe Steadman, Audris Mockus","doi":"10.1111/1556-4029.70046","DOIUrl":null,"url":null,"abstract":"<p><p>Accurately assessing the postmortem interval (PMI) remains a challenging task in forensic science. Existing regression models use the decomposition score to predict the PMI or accumulated degree days (ADD) but are often imprecise and rely on small sample sizes. This study explores if we can construct more accurate outdoor PMI estimation models using (a) a larger sample, (b) more sophisticated statistical models, and (c) additional predictors derived from demographic and environmental factors. Using a sample of 213 human subjects from a human decomposition photographic dataset collected at the [removed for double-blind review], we evaluated existing outdoor PMI and ADD formulae developed by Gelderman et al. [Int J Legal Med, 2018, 132, 863] and also developed more sophisticated models that incorporate additional predictors. Models using the total decomposition score (TDS), demographic factors (age, sex, and BMI), and weather-related factors (season and humidity history) reduced PMI and ADD prediction errors by over 50%. The best PMI model, incorporating TDS, demographic, and weather predictors, achieved an adjusted R-squared of 0.42 and an RMSE of 0.76. It had a 15% lower RMSE than the TDS-only model to predict PMI and a 54% lower RMSE than the pre-existing PMI formula. Similarly, the best ADD model, using the same predictors, achieved an adjusted R-squared of 0.54 and an RMSE of 0.73. It had a 10% lower RMSE than the TDS-only model to predict the ADD and a 55% lower RMSE than the pre-existing ADD formula. These results demonstrate that significant improvements in accuracy can be achieved using readily available predictors.</p>","PeriodicalId":94080,"journal":{"name":"Journal of forensic sciences","volume":" ","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of forensic sciences","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1111/1556-4029.70046","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Accurately assessing the postmortem interval (PMI) remains a challenging task in forensic science. Existing regression models use the decomposition score to predict the PMI or accumulated degree days (ADD) but are often imprecise and rely on small sample sizes. This study explores if we can construct more accurate outdoor PMI estimation models using (a) a larger sample, (b) more sophisticated statistical models, and (c) additional predictors derived from demographic and environmental factors. Using a sample of 213 human subjects from a human decomposition photographic dataset collected at the [removed for double-blind review], we evaluated existing outdoor PMI and ADD formulae developed by Gelderman et al. [Int J Legal Med, 2018, 132, 863] and also developed more sophisticated models that incorporate additional predictors. Models using the total decomposition score (TDS), demographic factors (age, sex, and BMI), and weather-related factors (season and humidity history) reduced PMI and ADD prediction errors by over 50%. The best PMI model, incorporating TDS, demographic, and weather predictors, achieved an adjusted R-squared of 0.42 and an RMSE of 0.76. It had a 15% lower RMSE than the TDS-only model to predict PMI and a 54% lower RMSE than the pre-existing PMI formula. Similarly, the best ADD model, using the same predictors, achieved an adjusted R-squared of 0.54 and an RMSE of 0.73. It had a 10% lower RMSE than the TDS-only model to predict the ADD and a 55% lower RMSE than the pre-existing ADD formula. These results demonstrate that significant improvements in accuracy can be achieved using readily available predictors.