Abstract
Mass Fatality Incidents (MFIs) present significant challenges for victim identification due to the scale and complexity of genetic relationships. Traditional "one-at-a-time" kinship analysis often leads to false positives in large-scale events because it ignores the global constraint that each recovered body can belong to only one missing person. This paper proposes a robust Bayesian framework that treats MFI identification as a global assignment problem. By evaluating all recovered victims and missing persons simultaneously, the "Full Model" leverages global constraints to improve identification accuracy, even with sparse genetic data or complex familial relationships. Simulation results demonstrate that the Full Model achieves 100% certainty in scenarios where traditional methods fail or provide ambiguous results.
Introduction
In the aftermath of natural disasters or large-scale accidents, authorities must identify a high volume of victims. Traditional forensic methods rely on kinship analysis, calculating a Likelihood Ratio (LR) for individual victim-missing person pairs. However, this "one-at-a-time" approach is fundamentally flawed for MFIs as it treats each case in isolation. In large-scale incidents, genetic profiles may overlap, leading to "false positives" where multiple victims appear to match the same missing person. Previous attempts to mitigate this by grouping families still failed to view the incident as a single, interconnected system. This paper addresses these limitations by proposing a simultaneous evaluation model that considers the entire incident as a global assignment problem.
Methods
The core of the proposed methodology is a Full Model based on a Bayesian framework. It defines a multivariate random variable $H$, representing a "configuration" of all possible identifications across the incident.
Key innovations include:
- Simultaneous Evaluation: Instead of calculating LRs for individuals, the model evaluates the probability of an entire configuration of assignments. This allows "easy" identifications to narrow down the options for "hard" cases.
- Conditional Independence: Assertions are used to decompose the joint probability of genotypes and identification hypotheses, ensuring mathematical tractability.
- Allele Bayesian Networks: Established Bayesian network techniques are utilized to propagate evidence through complex family trees, handling scenarios where direct DNA from a missing person is unavailable.
- Computational Efficiency: A pruning algorithm focuses on configurations with non-zero probability, excluding matches that are genetically impossible under Mendel’s laws.
Results
The model was evaluated using a simulated MFI involving 14 missing individuals from 10 families. The simulation included challenges such as missing direct DNA samples and families searching for different types of relatives (e.g., parents vs. siblings). The Full Model was compared against the Victim Model (standard one-by-one practice) and the Family Model (grouping by family but ignoring the rest of the incident).
As shown in the results table, the Full Model achieved a posterior probability of 1.0 (100% certainty) for all 14 victims. In contrast, the Victim Model struggled significantly with siblings and common genetic profiles. For instance, in Family 5, the Victim Model's probability for the correct match was as low as 0.02, whereas the Full Model maintained 1.0. Even with incomplete information, the Full Model remained highly accurate by using the "closed system" advantage to deduce identities through the process of elimination.
Conclusion
Treating MFI identification as a closed system provides a significant "exclusion power" advantage. By leveraging global constraints, the Full Model successfully identifies victims where traditional methods fail. The model's power comes from the fact that a unique match for one victim reduces the search space for others. However, the author notes that the model's effectiveness is tied to the fraction of victims recovered; as this fraction decreases, the model reverts toward standard kinship analysis. Future work could incorporate additional metadata such as gender, Y-chromosome STRs, or Mitochondrial DNA to further refine identification in extremely large-scale incidents.
Summary updated