Ethical clearance for the study was secured from Turfloop Research and Ethics Committee (TREC/268/2017: PG). Consent was not applicable as we were not working on human subjects but case reports, permission for which was obtained from the Limpopo Provincial Department of Health. The case files were coded with numbers and names of the deceased were not recorded.

This quantitative descriptive study retrospectively analysed 85 cases of deceased individuals across three years from 01 January 2014 to 31 December 2016. Cases sustained head injuries, underwent ACT imaging and had no surgical intervention after being referred to an academic hospital. Deceased cases were subjected to autopsy procedure as per legal requirement at the attached forensic pathology facility in Limpopo, a rural province of South Africa.

Data were obtained from post-mortem reports from the forensic pathology laboratory while CT scan reports, clinical data and human resource records were obtained from the Pietersburg hospital records section and radiology department. Autopsy findings served as a reference point because studies have demonstrated that it remains superior to ACT scan for the detection of brain injuries.^1,8 The majority (n = 80) of the cases had only one ACT scan done and the rest a second; only the first scan reports were therefore used in the study. The following intracranial haemorrhages were assessed from these reports: epidural, subdural and subarachnoid haemorrhages.

Data were captured and analysed using Microsoft Excel (Microsoft Office Professional Plus 2013; Microsoft Corp., Redmond, Washington, United States) and International Business Machines Corporation Statistical Package for the Social Sciences version 23 (Armonk, New York, United States) running under Microsoft Windows (Microsoft Corp., Redmond, Washington, United States). Cross-tabulations were used to establish the percentage agreement between ACT scan and autopsy findings of extradural haemorrhage (EDH), subdural haemorrhage (SDH) and subarachnoid haemorrhage (SAH) single or combination occurrence. If there is agreement between the ACT scan and autopsy findings the individual case scored 1, but if there is a disagreement between the ACT scan and autopsy findings, the case scored 0. Levels of agreement and resultant discrepancy rates were determined using Cohen’s kappa statistics. The pattern of discrepancies for identified factors such as the level of consciousness by Glasgow Coma Scale, the length of hospital stay, the experience of the clinician (radiologist and pathologist) and the site of haemorrhage was also evaluated. Results were considered statistically significant when p was less than 0.05.

In 75.29% (64/85) of cases, the ACT scan and autopsy agreed on the presence or absence of haemorrhage with a kappa coefficient of 0.3834 (Table 1). The remaining 24.71% represents the overall discrepancy rate between the ACT scan and autopsy detection of haemorrhage.

The highest agreement between the ACT scan and autopsy finding was recorded in the diagnosis of EDH (agreement 90.59%, kappa coefficient 0.5823 and discrepancy rate 9.41%) (Table 1). With regard to SDH, the agreement was 74.12% with a kappa coefficient of 0.4857 and a discrepancy rate of 25.88% (Table 1). The lowest agreement of 65.88% with a kappa coefficient of 0.3219 and the highest discrepancy rate of 34.12% was found for SAH (Table 1).

In 38 of the 85 cases (44.7%; 25 with and 13 without haemorrhage), both methods agreed in the diagnostic intracranial haemorrhage category (i.e. single, binary or ternary combinations of haemorrhages), denoting a diagnostic category discrepancy of 55.3% (Table 2 and Table 3). ACT agreed in 7 of the 11 cases categorised as singular SAH by autopsy (bolded) but reclassified three cases as having binary haemorrhage (SAH and SDH) and one case as absent (no haemorrhage). For 25 cases classified as ‘Absent’ (having no haemorrhage) by autopsy, ACT agreed in 13 (bolded) cases but categorised the remaining 12 as having different haemorrhages.

The analysis revealed that most discrepant intracranial haemorrhage diagnoses (5 out of 8 cases for EDH; 15 out of 22 cases for SDH; 19 out of 29 cases for SAH) were seen in patients with markedly low levels of consciousness, denoting severe traumatic head injury (Figure 1).

Half (4/8) of the discrepant cases for EDH were found to have been admitted on average for a day or less (Figure 2). On the other hand, most discrepancies for SAH were seen in cases that were admitted for two weeks or less (9/29) and four weeks or more (10/29) on average, while for SDH the majority (9/22) stayed in the hospital for more than a month.

In addition, more discrepant case diagnoses were seen with a radiologist or pathologist who had less than three years of working experience (Figure 3).

The major limitation of this study is the small sample size. Moreover, the study addressed only three types of brain haemorrhages (EDH, SDH and SAH); therefore, the findings cannot be generalised to forensic pathology practice, or any other type of intracranial haemorrhage. Further research with a larger sample size and broader scope is recommended.

The overall detection discrepancy rate of 24.74% and CT diagnostic accuracy of 44.7% implies that ACT scans may not be used as an altervative to reduce the number of autopsies performed at the mentioned facility but can only be used as an adjunct to autopsy. However, the low discrepancy rate in EDH, especially after a day of admission, implies that ACT may be useful for the diagnosis of this haemorrhage in the forensic setting. A markedly reduced level of consciousness, length of hospital stay depending on the type of haemorrhage and three or fewer years of doctors’ experience all contributed to discrepancies observed between ACT and autopsy findings. The study employed a limited sample and thus calls for more similar studies in both high and low- and middle-income countries.