Flexible regression models for estimating postmortem interval (PMI) in forensic medicine.

Area of Science:

• Forensic Medicine
• Biochemistry
• Statistical Modeling

Background:

• Accurate postmortem interval (PMI) estimation is crucial in forensic medicine.
• Existing PMI methods often lack precision and reproducibility.
• Biochemical markers in vitreous humor show promise for improved PMI estimation.

Purpose of the Study:

• To compare advanced statistical models (GAMs, SVMs) with linear regression (LR) for PMI estimation.
• To evaluate the efficacy of vitreous humor potassium ([K+]) and hypoxanthine ([Hx]) levels in PMI determination.
• To provide more flexible and accurate alternatives to current PMI estimation techniques.

Main Methods:

• Analysis of [K+] and [Hx] in over 200 human vitreous humor samples with known PMI.
• Application and comparison of Linear Regression (LR), Generalized Additive Models (GAMs), and Support Vector Machines (SVMs).
• Validation of predictive models using real-world forensic data.

Main Results:

• Both GAMs and SVMs demonstrated superior performance in PMI estimation compared to LR.
• SVM models offered slightly higher precision, while GAMs provided interpretable graphical outputs.
• The study confirmed the utility of [K+] and [Hx] as key biochemical indicators for PMI.

Conclusions:

• Advanced statistical models like GAMs and SVMs significantly enhance PMI estimation accuracy.
• GAMs and SVMs represent valuable, flexible alternatives to traditional LR methods in forensic analysis.
• Accessible R code is provided for practical application of these improved PMI prediction models.