The utility of histology in assessing the cause of death in medico‐legal autopsies in selected trauma deaths: Suspension‐, immersion‐, fire‐, and traffic‐related

The goal of a medico‐legal autopsy is primarily to determine the cause and manner of death. To this end, the pathologist often uses auxiliary analyses, including histology. However, the utility of routine histology in all medico‐legal autopsies is unknown. Earlier studies on the utility of routine histology have shown inconsistent effects, with some studies recommending it and others rejecting it. To study the degree to which histology informs on the underlying cause of death, we sent autopsy reports from suspension‐, immersion‐, fire‐, and traffic‐related deaths to senior board‐certified forensic pathologists and had them assess the cause of death, first without knowledge of the histological findings and then with knowledge thereof. Fifty cases were identified in each of four subgroups: fire‐, immersion‐, suspension‐, and traffic‐related deaths. The autopsy reports were anonymized, and the histological findings and conclusions were removed. Two board‐certified forensic pathologists independently reviewed the reports in each subgroup and assessed the manner and underlying cause of death (including their certainty of this assessment on a five‐level scale) with and without access to histological findings. The probability of changing the underlying cause of death posthistology was low in all study groups. There was a slight increase in the degree of certainty posthistology in cases where the underlying cause of death was not changed, but only when the antehistology certainty was low. Our results suggest that histology does not meaningfully inform on the underlying cause of death in suspension‐, immersion‐, fire‐, and traffic‐related deaths except when antehistology certainty is low.


| INTRODUC TI ON
The main goal of a medico-legal autopsy is to determine the cause and manner of death. To this end, the pathologist performing the autopsy uses macroscopic findings, as well as auxiliary analyses, most commonly histology and toxicology.
Histology usage in medico-legal autopsies is uneven. In Sweden, rates vary between the six medico-legal offices, with samples being taken in ~60%-90% of autopsies (internal National Board of Forensic Medicine data) [1]. However, according to a recommendation from the Council of the European Union [2], all medico-legal autopsies should include histological sampling. This recommendation is, however, ambiguous as to whether samples should always be examined or only "taken." Further, the recommendation does not seem to be evidencebased (no references are given in the recommendations) and stands in contrast to Swedish law which allows sampling "necessary to fulfill the purpose of the autopsy," which, as defined in Swedish law, is primarily to determine the cause and manner of death [3].
In a prospective study of 428 decedents subjected to a medico-legal autopsy, histology changed the assessment of the "mechanism of death" in 40% of cases and the assessment of the cause of death in 8% of cases [6]. However, in a retrospective study of 287 decedents, the assessed cause of death was changed by histology in only 2% of cases [8].
Hence, it remains an open question whether histological sampling should be used in all medico-legal autopsies, even when there is a macroscopically probable or certain cause of death. The aim of the present investigation was to determine whether or not histological analyses contribute to the assessment of cause and manner of death, and/or to an increased degree of certainty in diagnoses.

| Materials
We searched the autopsy database of the National Board of Forensic Medicine (NBFM) including only case information available before an autopsy, that is, not related to the cause of death, using relevant search terms to find fire-, immersion-, suspension-, and trafficrelated deaths ( Table 1). We included all cases where histological samples were taken and reviewed. A board-certified forensic pathologist then reviewed the cases and excluded those where the circumstances were obviously irrelevant, for instance, when the search term "smoke" identified a decedent who was a "smoker" but died in circumstances unrelated to the fire. We also excluded suspected homicides, since such cases do not represent "routine handling." In each group, we selected 50 cases from 2015 to 2018. The study was initially conducted as four separate studies with the data later combined. Regrettably, the selection process was not entirely uniform for all four subgroups ( Table 1). In the fire and traffic groups, cases from the study period were selected randomly, while in the immersion and suspension groups, cases were selected consecutively, starting from the most recent. It is possible that the consecutive selection process might induce inclusion bias, for example, if there were more cases where hypothermia was considered due to the consecutive cases oversampling cases from cold seasons. Still, we believe that this aberration would have minimal impact on the study results, as the (non-)randomness of these cases should not severely affect statistics.
In each selected case, the complete autopsy protocol, including histological and toxicological analyses, was acquired. All cases were anonymized by masking any information connected to the decedent, the forensic pathologist, and/or the regional medico-legal office.

Highlights
• The value of histology in determining the cause of death (COD) in certain traumatic deaths was assessed.
• Histology was unlikely to be informative when certainty of COD after the autopsy was at least moderate.
• Histology information did not increase pathologist certainty about COD for these deaths.
• Probability of change in COD or its certainty after histology showed significant interobserver variability.

Fire Traffic Suspension
Search terms a Drown b , water b , lake b , ocean b , The originally assigned cause and manner of death and information regarding histological analyses were removed.

| Methods
Two independent board-certified forensic pathologists were recruited for each group. Our study included a total of four reviewers, referred to as Reviewers A-D, who were asked to study different subgroups ( Table 1).
The reviewers studied the autopsy protocols, which included a short background summary, but no histological findings, and recorded their assessment of the cause and manner of death, as well as their degree of certainty in this assessment. In Sweden, medico-legal affidavits include a degree of certainty on a five-level scale from the findings being "compatible with" [X being the cause/manner of death] to that the findings "show" [that X is the cause/manner of death] ( After the reviewers had submitted their initial assessments, they were sent the histology findings and asked to again assess the cause and manner of death for each case, now based on all the information.
We analyzed the resulting data in a Bayesian statistical model

| Descriptive statistics
There were only minor changes to the assessed underlying cause of death posthistology ( Table 3). The assessment of the underlying cause of death was changed by at least one reviewer in 8 (4%) out of the total 200 cases. However, there was only 1 (0.5%) case, where both reviewers changed their assessment of the underlying cause of death. This was drowning case 47, in which there were histological signs of acute myocardial infarction.
In cases, where the assessment of the underlying cause of death was not changed, there were some changes in the degree of certainty (Table 3). Histology did not impact the interreviewer agreement ( Table 3).
In the antehistology assessment, there was agreement on the under- The manner of death was changed from unnatural to natural by at least one reviewer in 5/200 (2.5%) cases, with agreement on this specific change in only 1 (0.5%) case ( Table 3). Reviewer B accounted for 4/4 of cases in which only one reviewer changed the manner of death to natural.

| Changes in the assessed underlying cause of death
The median probability of changing the assessed underlying cause of death posthistology in immersion-related deaths was 3.1% (0.1-8.1%, 95% HPDI). As antehistology certainty increased, the probability of changing the underlying cause of death decreased ( Table 4).
The probability of changing the assessed underlying cause of death was lower in the fire-related and suspension-related subgroups and higher in the traffic-related subgroup (Table 4). When the degree TA B L E 2 Degrees of certainty in Swedish medico-legal autopsies, translated from Swedish

Compatible with
Possibly implies Implies Strongly implies Shows/is TA B L E 3 Underlying and contributing cause of death, manner of death, and degree of certainty without and with histology in cases where the cause of death and/or the degree of certainty was changed by one or both reviewers after access to the histology findings. Accidental Implies of certainty was at the "implies" level or higher, the 95% HPDI was below 10%, except for traffic-related cases.

| Changes in the degree of certainty
The degree of certainty increased posthistology when antehistology certainty was low, but not when certainty was moderate to high (i.e., "implies" or higher, Figure 1). The effect on the degree of certainty was almost identical in all four subgroups. There was a slightly higher chance of increasing the degree of certainty when the underlying cause of death was unchanged.

| DISCUSS ION
Our results suggest that there is only a slim chance (in the groups included in this study) that histology will inform the cause of death when the ante-histology degree of certainty is high, suggesting that the value of routine histological sampling is low. A common objection is that even if the underlying cause of death does not change, the pathologist may become more certain in their cause of death assessment posthistology. However, this objection is not supported by our study. There were only minor increases in the degree of certainty and only when the ante-histology certainty was low.
Interreviewer disagreement was larger for traffic-related deaths than for other subgroups. We believe that this was due to there being several possible competing causes of death and variability in how the reviewers valued histological findings ( Table 3). While it is possible that one or more reviewers felt the need to prove the utility of histology and changed their assessment for this reason, we consider this unlikely.
Further, many changes posthistology were unlikely to be the effect of histology itself, for example, in case 30 in the drowning group, where drowning was added as a contributing cause of death posthistology ( Table 3)  are not the results of actual histological findings. We have not reassessed the utility of histology in each case as we believe the results are best presented as is and without change, such that readers can themselves assess whether the included changes are significant.
It is notable that reviewer B seems to be an outlier regarding changes in posthistology ( Table 3). We have not been able to come up with a reasonable explanation for why this might be. Reviewer B was included in the traffic accident group, where there were many competing causes of death, as discussed above. However, reviewer A made changes far less often with the same underlying data.
Reviewer B is also far from the least experienced of the included reviewers. Regardless, the results of reviewer B's variability should be somewhat accounted for in the varying intercept model which was used.
Two earlier studies have found large effects of histology [6,7], whereas others have not [8][9][10]. The two studies with large effects [6,7] and one of the studies with a small effect [8] used unselected medico-legal populations. However, the paper with a small effect had a population where natural causes of death constituted only 23% of cases. The remaining two papers with slim to no effect used hanging deaths [10] and deaths where the pathologist would not routinely have reviewed histology [9], respectively. This difference in study populations probably explains the differences in results, as a higher number of cases with slim autopsy findings and/or pathological findings, as opposed to traumatic findings, naturally increases the utility of histological analyses.
Aside from utility for the autopsy at hand, a common justification of routine nondirected histology is that when fewer histological samples are taken, the pathologist becomes less skilled in the assessment of histological samples. While assessing fewer samples might theoretically reduce a pathologist's skill, we do not think that forgoing routine histology would result in less than 50%-60% of medico-legal autopsy cases having samples taken (roughly the proportion of deaths due to non-traumatic causes in our national medico-legal population, internal NBFM data). Whether such a decrease would reduce a pathologist's skill is questionable. Furthermore, considering the often considerable workload in forensic pathology worldwide, the time spent on routine histology could be spent more meaningfully on more complicated cases. Limited financial resources also mean that the process must be as efficient and evidence-based as possible. Either way, taking histology samples when they are not needed for an autopsy is not compatible with Swedish law [3], which means that not even specialist training in forensic pathology is an acceptable justification for histological sampling. Resident training in histology can be performed more systematically and efficiently in other ways.

| Limitations
As stated above, the study was initially planned and performed as four separate studies and as such did not use a fully uniform selection process for all four subgroups. However, this aberration should have minimal impact on the study results. Further, we thought it better to include the results and be transparent about the issue regarding case selection. Due to the amount of work involved in eliciting responses from board-certified forensic pathologists, we considered it too costly to redo the substudies with aberrant selection processes.
Further, the reviewers were unevenly split among studies; reviewer A was included in three studies, whereas reviewers C and B were included in only one study each. We tried to alleviate the uneven inclusion of reviewers by modeling reviewers as varying intercepts, but it is still possible that reviewer A might have had a slightly stronger impact due to overinclusion.

U COD C COD MOD Certainty
Skull base fx with brain injuries BP Accidental Strongly implies BP with underlying traumatic asphyxia -Accidental Strongly Implies The reviewers had access only to the briefcase summary included in the standard autopsy report, which would limit their capacity to correctly assess the manner of death. However, this should not impact the assessment of the cause of death, as any perceived lack of background information would have resulted in increased uncertainty. The opposite was true in our data; the reviewers were generally quite certain of their assessments in the selected study groups.

| Conclusions
Among fire-, immersion-, suspension-, and traffic-related deaths, the assessed underlying cause of death was changed in only a small proportion of cases after access to the histology findings. Histological analyses in these cases did not seem to inform the cause of death meaningfully unless the antehistology certainty was low. This is not to say that histology may not be useful in cases with slim autopsy findings or with macro-pathological findings of unknown or undetermined nature, for example, in abuse-related and natural deaths.
However, the utility of routinely collected nondirected histology in all cases, regardless of macroscopic findings, is not supported in suspension-, immersion-, fire-, and traffic-related deaths. Hence, we suggest that the recommendations from the Council of the European Union should be updated.

ACK N OWLED G M ENTS
It is with great thanks that we acknowledge the expert assistance of Adam Berkowicz, as well as Anton Myhrman Larsson, Peter Lydig, and Staffan Finn, for their assistance in data collection and management.

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

O RCI D
Torfinn Beer https://orcid.org/0000-0003-3509-1654 F I G U R E 1 The relative probability of posthistology certainty regarding the underlying cause of death in firerelated deaths, assuming the cause of death was not changed. (A) When antehistology certainty was "compatible with," (B) when antehistology certainty was "possibly implies," (C) when antehistology certainty was "implies," (D) when antehistology certainty was "strongly implies," (E) when antehistology certainty was "shows/is." The line represents the posterior median and the shaded areas cover 95%, 85%, 75%, and 65% highest posterior density intervals, respectively. [Colour figure can be viewed at wileyonlinelibrary.com]