Homicide or suicide? A probabilistic approach for the evaluation of the manner of death in sharp force fatalities

The role of forensic science can be defined as providing relevant opinions to assist investigators and courts of law in answering questions. The Likelihood Ratio (LR) provides a quantitative and logical approach to communicating the strength of expert evidence. We reviewed existing forensic literature on sharp force fatalities, focusing on studies reporting the manner of death and the frequency of some characteristics that are traditionally assessed. Four studies were included, resulting in a database of 173 suicides and 354 homicides. The LR of each of the characteristic under both hypotheses (suicide and homicide) was obtained. Subsequently, the LR was computed in six fatalities with known manner of death, three suicides and three homicides, by multiplying the corresponding LR of each individual characteristic. LR ranged from 115 to 140,250 in suicidal cases and from 9 to 2728 in homicidal cases. Compared to other fields of forensic science where LR is used extensively, the values obtained in our cases of sharp force fatalities is low. However, in forensic pathology there is evidence that is outside the expert's opinion, and it is for the trier of fact, such as the judge or jury, to draw conclusions. Nevertheless, the LR serves as a tool for interpreting and weighing evidence while maintaining the distinct roles of the trier of fact and the expert. To comprehensively apply the LR in the field of sharp force deaths, it will be necessary to standardize the methodology of investigation and data collection in descriptive studies.


| INTRODUC TI ON
The role of the forensic expert is to provide a relevant opinion to help answer questions to investigators and to courts of law [1,2].
Even if Bayesian decision theory is not necessarily suited for presentation of evidence in court [3], many fields of forensic sciences have adopted this method of evidence interpretation, which provides an approach to logically convey the strength of the expert's evidence via the so-called Likelihood Ratio (LR).The essential challenge is whether and to what extent the expert's observations may be used to distinguish between two mutually exclusive hypotheses.These hypotheses, referred to as propositions, usually represent the positions of the prosecution and the defense [4].Forensic pathology seems slow to adopt these principles [4], and few examples of application of the LR have been reported in forensic literature [5,6], none of them in the field of sharp force fatalities.
During investigative proceedings of sharp force trauma, as well as at trial, the forensic expert can provide key elements for decision makers, enabling them to reach an opinion regarding the manner of death.In such context, a forensic pathologist can be asked to define if a sharp force fatality is the consequence of a suicide or of a homicide, as accidental sharp force fatalities are very rare and often have an unequivocal mode of death [7] based on the evidence observed during death/crime scene investigation, autopsy, toxicological examinations, and other ancillary investigations.
Criteria for the differentiation of self-inflicted injuries from injuries inflicted by another person are described in classic and modern textbooks of forensic pathology, and include anatomical site, number of injuries, hesitation wounds, type and localization of the weapon, clothing analysis, psychiatric history, and scene/autopsy findings [8].It is impossible for the forensic expert to quantitatively express these criteria and to provide the court with a number that expresses the strength of the evidence [9].However, the collection of case series published in the forensic literature could ideally be used to determine whether and to what extent the evidence supports the hypothesis of homicide as opposed to suicide or vice versa, based on the frequency of observations in both circumstances [10,11].The probative value of the evidence, without having to consider the prior probability of the hypothesis, can be measured using the LR [12].
The aim of this study was to review the existing forensic literature on sharp-force deaths and to assess the frequency of characteristics related to the manner of death.The probative value of each of the characteristics assessed, expressed in terms of LR, was then reported and calculated for a series of sharp force deaths with known manner of death.

| Systematic review
A systematic review of the literature for cross-sectional studies involving sharp force fatalities was performed on PubMed (All fields) and Scopus (Article title/Abstract/Keywords).The search terms were "sharp force" and "fatalities," and the research was kept intentionally broad to be as sensitive as possible.The research was performed on 19 October 2022 and a temporal restriction was applied to articles published after January 1986.In addition, bibliographies of published review articles as well as studies relevant to the research question were manually searched, and potentially eligible references were included in the full-text review.A comprehensive database of the retrieved articles was built and checked to remove duplicates.Only original English-language articles that clearly reported the manner of death and the frequency of relevant recurrent forensic features commonly described in the forensic literature were included.Studies with small sample sizes (<30 cases) or samples focused on a single mode of death (suicide only or homicide only) were excluded.The following characteristics were extracted from the database: gender of the victim, number of sharp force injuries and anatomical wound distribution, alcohol detection, history of psychiatric illness, damage to clothing, place where the body was found, and type and location of the sharp force object used.Studies reporting the frequency of two or more characteristics identified in relation to the manner of death were included.Titles and abstracts were independently screened against inclusion criteria by two reviewers (MI, MD).
Disagreements were resolved by discussion or referred for arbitration to a third reviewer (GP).According to the results obtained from retrieved studies, which may show heterogeneities in categories, forensic characteristics have been classified.

| Likelihood ratio
Once we identified the recurrent relevant characteristics reported in the included papers, a dichotomic classification for each of them was performed.This classification was achieved in analogy to the classification reported in the original studies, considering all characteristics universally relevant.The LR was calculated as the ratio of frequencies of the evidence under both hypotheses, as follows: • LR in suicides (LRs) = (number of suicides where X was found/ Total Suicides)/(number of homicides where X was found/Total homicides); • Calculating the LR can be useful in distinguishing between suicide and homicide.
• LR can be used for weighing evidence, but further research to standardize data collection is needed.
• LR in homicides (LRh) = (number of homicides where X was found/ Total homicides)/(number of suicides where X was found/Total Suicides); Where X is the characteristic found.The resulting LR were rounded to the third digit after the decimal point.

| Application of the LR on real cases
The LR was calculated in six cases of sharp fatalities selected among forensic casework in the time frame 2020-2022.Only cases with a known manner of death were included, specifically those in which all circumstantial data confirmed one of the hypotheses and where the criminal legal proceedings have concluded.We considered each characteristic as independent given the hypothesis, implying that the probability of observing one characteristic is not influenced by the probability of observing any other characteristic.The overall LR for each of the real forensic cases was calculated by multiplying the respective LR of the individual characteristics, as previously suggested [12], and rounded to the nearest unit.The LR was also expressed by a verbal equivalent according to a scale of conclusions as proposed by ENFSI Guideline for Evaluative Reporting in Forensic Science [13].

| Systematic review
The search was conducted on 19 October 2022 in the Scopus and PubMed databases, yielding 57 and 84 articles respectively, for a total of 141 articles, of which 35 were duplicates and 12 were excluded because they were not published in English.Of the 94 articles screened, 79 were excluded from the title and abstract screening and 16 full-texts were reviewed.Twelve articles were excluded for the reasons listed in Figure 1.Four studies met all inclusion criteria [7,11,14,15].
The database included a total of 527 sharp force fatalities, 173 suicides, and 354 homicides (Table 1).

| Relevant forensic characteristics in suicides and in homicides and likelihood ratio
Due to the heterogeneity of the characteristics reported, for data extrapolation a dichotomic differentiation for each characteristic was used, whenever feasible in analogy to the classification reported in the original studies, as follows.
Two papers [14,15] provided information about the number of wounds, reporting them as "single injury" and "multiple injuries."The same classification was used for the purpose of the study.
The following locations were investigated in the studies: neck, thorax, abdomen, crook of the arm, wrist, upper limbs (hand and arm, excluding the two previous areas), and lower limbs.Only the article by Karlsson [15] described the number of sharp force homicides and suicides that reported injuries in specific anatomical regions, allowing LR to be calculated for each of them.Other articles [7,11,14] only reported the overall distribution of injuries in suicidal and homicidal cases and cannot be used for the purpose of this study, as frequencies cannot be extrapolated.
The result of toxicological analysis regarding the blood-alcohol presence was reported in three papers [7,14,15] as "under the influence of" or as "alcohol detected."To ensure a broader inclusion, the presence of blood-alcohol was divided into "detected" and "not detected," even if a cut-off could not be identified due to the absence of quantitative data.
The history of psychiatric illness was reported in two papers [7,11] as "present" and "absent," and the same classification was used.
Damage to clothing was reported in three papers [7,11,15] as "present," "absent," and "undetermined."For the purpose of this study, only the presence or absence of clothing damage was considered.
All of the papers [7,11,14,15], included the place where the body was found, differentiated as "victims or other person's home" versus "other place indoors" or "outdoors" [11,14,15], or "residential unit vs outside a residential unit" [7].Therefore, for the purpose of this study, the place where the body was found was classified as "home environment" and "outside a residential unit."Vassalini et al. [14] reported some cases in which the place where the body was found was "not known," so these cases were not taken into account.
Due to the high number of existing sharp objects used in fatalities in the included studies [7,11,14,15] (e.g., axe, razor blade, scissor, metal chips, splinters of glass) and considering that the most common sharp object used was a knife, for the purpose of the study the characteristic was divided into "knife" and "sharp object other than knives."The knife category included kitchen knife, weapon knife, pocketknife, sheath knife, and tool knife.Two articles [7,11] reported the location of the sharp object in relation to the body's position as "near the body or inside the body" and "away from body or missing," and the same classification was used.
Then the absolute frequency and the LR of each characteristic in suicides (LRs) and in homicides (LRh) was calculated (Table 2).

| Forensic casework presentation and application of the LR
All the included characteristics were identified in our casework and the overall LRh and LRs were calculated in all cases by multiplying the LRs/LRh of each relevant characteristic, as previously reported [6,12].

| Case 1 -Suicide
A 43-year-old male (LRs = 1.121;LRh = 0.892) was found in his bed (LRs = 2.689; LRh = 0.372).She was suffering from major depression (LRs = 16.133;LRh = 0.062).Toxicological analyses were negative for alcohol (LRs = 1.950;LRh = 0.513) and other drugs.All data obtained from police investigation and bloodstain patter analysis supported the hypothesis of suicide [17] and the case was filed by the public prosecutor.
The overall LRs and LRh are 115 and 9 × 10 −3 respectively, indicating that the identified characteristics provide moderately strong support for the hypothesis of suicide rather than homicide.The total number of suicides and homicides corresponds to the overall number of cases included in the papers that specifically address each characteristic.

| Case 3 -Suicide
b Likelihood ratios for all characteristics included.
c For practical reasons for thorax wounds only penetrating wounds were considered.
This paper shows the use of LR in the complex task of distinguishing between suicide and homicide in cases of sharp force fatalities, which often require careful evaluation and the consideration of evidentiary weight.In recent years, scores and mathematical models have been utilized to quantify opinions regarding the manner of death in sharp force fatalities.Karlsson [18] was the first to propose a model to make predictions regarding whether a certain fatality appears more likely to be a homicide or suicide through logistic regression analysis.More recently, Visentin et al. [19] applied a scoring system [20] for the correct framing of suicide caseworks based on statistical frequency of the suicidal method adopted, the victim's history of mental illness, circumstantial data, number of means, and compatibility of means and injuries with suicidal dynamics, to be applied during death scene investigation.Although these papers also presented methods to evaluate the manner of death to be applied in sharp force fatalities, an approach based on an aggregate analysis may provide, using LR, an accessible and comprehensive structure to deal with data uncertainty and logically convey the strength of evidence [21].
After extracting data from included studies and evaluating frequencies in suicidal and homicidal cases, some characteristics that are traditionally associated with manner of death, such as the gender of the victim or the number of injuries [22][23][24] showed similar frequencies in homicides and in suicides.As expected, other characteristics were strongly associated with manner of death, such as the presence of psychiatric illness, the alcohol detection, or the type and location of the sharp object used [7].Also, we found that injuries occurring at the thorax and at lower limbs significantly more frequently observed in homicides, while injuries at crook of the arm and wrist were more frequently observed in suicides.Some characteristics, namely the place where the body was found, the type or location of the weapon, may be defined as "context information" [4].The question about whether context information may be used to influence the expert's decisions has long been addressed in many fields of forensics [25][26][27][28].In forensic pathology, Dror et al. [29] observed that relying on contextual information can be risky, as its objectivity, susceptibility to bias, and relevance can vary.Consequently, errors in determining the manner of death may arise when contextual information is either overemphasized or disregarded too easily.On the other hand, H.H.
de Boer et al. [4], stated that only context information that influences the LR is relevant for the expert, as long as all characteristics used to calculate the LR are explicated to the trier-of-fact, not only because they are important to understand the basis of the expert's opinion, but because they help to consider the effect of context information on the expert's opinion, preventing the so-called "double counting of evidence" [4].
When calculating the LR in a real case, we found that the total LRs ranged from 115 to 140,250 in suicidal cases, whereas the total LRh ranged from 9 to 2728 in homicidal cases.Expressing the LR by a verbal equivalent [13], we obtained "very strong support" (cases 1 and 3) and "moderately strong support" (case 2) for the suicidal death hypothesis, and "weak support" (case 4) and "strong support" (cases 5 and 6) for the homicidal death hypothesis.The LRs obtained are low compared to other fields of forensic science where this approach is widely used.Nevertheless, in forensic pathology, there is a great deal of information that should be evaluated by the trier of fact that is outside the scope of expert evidence.Indeed, in the six cases presented, evidence independent of the physical findings at the scene and on the body clearly indicated that some of the deaths were homicides, regardless of the LR.A correct understanding and interpretation of the case requires consideration of these contextual factors.The proposed approach helps the forensic pathologist to form and support an opinion based on characteristics gathered mainly during the death scene investigation, the autopsy and other post-mortem data, and serves as a tool to interpret the evidence in a logically correct way.

| Limitations and future perspectives
The literature review revealed that, although numerous studies have examined cases involving sharp force fatalities, only a limited number of papers have provided comprehensive descriptions of forensic characteristics that can be gathered and analyzed independently in a larger dataset.
Moreover, a significant level of descriptive heterogeneity was found among the reviewed articles.For instance, many of them failed to specifically mention important forensic aspects such as the presence of defense injuries, hesitation marks, and psychiatric history.The lack of key forensic details in most cases was previously emphasized by certain authors [22,23], who observed that the localization of the lesion is sometimes described only in vague or general terms, which hampers comprehensive analysis across studies.
In this study, as the frequencies of single characteristics were derived from descriptive retrospective studies, it was not feasible to separate data from each individual case or examine injuries and other characteristics as dependent variables.This led to the infeasibility of applying a Bayesian network, ideal to visualizes dependencies between variables and the flow of information between these variables while estimating the joint probability distribution of data, as each relevant characteristic has been assessed independently [21].As a result, the method presented does not enable a thorough assessment of the evidence through a comprehensive and integrated logical analysis.Such an evaluation is a crucial task that should be entrusted to a forensic expert who possesses the necessary expertise to conduct a comprehensive analysis.The main challenges of research in forensic pathology relies in the identification of new strategies in descriptive studies, as one of the major Learning Objectives of the American Academy of Forensic Science (AAFS) Continuing Education Program [30].

| CON CLUS IONS
The LR helps to maintain the separate roles of the trier-of-facts and the expert and could be implemented in forensic casework to assess its robustness also in forensic pathology.To be able to extensively apply the LR in the field of sharp force fatalities, as in other violent fatalities, it will be necessary to standardize the methodology of investigation and data collection in descriptive studies, and to validate the LR approach on a larger casework.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors have no conflicts of interest to declare.