Diagnostic assistance to improve acute burn referral and triage : assessment of routine clinical tools at specialised burn centres and potential for digital health development at point of care

Abstract

Background: Inappropriate referral of patients for specialised care leads to overburdened health systems and improper treatment of patients who are denied transfer due to a scarcity of resources. Burn injuries are a global health problem where specialised care is particularly important for severe cases while minor burns can be treated at point of care. Whether several solutions, existing or in development, could be used to improve the diagnosis, referral and triage of acute burns at admission to specialised burn centres remains to be evaluated.

Aim: The overarching aim of this thesis is to determine the potential of diagnostic support tools for referral and triage of acute burns injuries. More specifically, sub-aims include the assessment of routine and digital health tools utilised in South Africa and Sweden: referral criteria, mortality prediction scores, image-based remote consultation and automated diagnosis.

Methods: Studies I and II were two retrospective studies of patients admitted to the paediatric (I) and the adult (II) specialised burn centres of the Western Cape province in South Africa. Study I examined adherence to referral criteria at admission of 1165 patients. Logistic regression was performed to assess the associations between adherence to the referral criteria and patient management at the centre. Study II assessed mortality prediction at admission of 372 patients. Logistic regression was performed to evaluate associations between patient, injury and admission-related characteristics with mortality. The performance of an existing mortality prediction model (the ABSI score) was measured. Study III and IV were related to two image-based digital-health tools for remote diagnosis. In Study III, 26 burns experts provided a diagnosis in terms of burn size and depth for 51 images of acute burn cases using their smartphone or tablet. Diagnostic accuracy was measured with intraclass correlation coefficient. In Study IV, two deep-learning algorithms were developed using 1105 annotated acute burn images of cases collected in South Africa and Sweden. The first algorithm identifies a burn area from healthy skin, and the second classifies burn depth. Differences in performances by patient Fitzpatrick skin types were also measured.

Results: Study I revealed a 93.4% adherence to the referral criteria at admission. Children older than two years (not fulfilling the age criterion) as well as those fulfilling the severity criterion were more likely to undergo surgery or stay longer than seven days at the centre. At the adult burn centre (Study II), mortality affected one in five patients and was associated with gender, burn size, and referral status after adjustments for all other variables. The ABSI score was a good estimate of mortality prediction. In Study III experts were able to accurately diagnose burn size, and to a lesser extent depth, using handheld devices. A wound identifier and a depth classifier algorithm could be developed with assessments of relatively high accuracy (Study IV). Differences were observed in performances by skin types of the patients.

Conclusions: Altogether the findings inform on the use in clinical practice of four different tools that could improve the accuracy of the diagnosis, referral and triage of patients with acute burns. This would reduce inequities in access to care by improving access for both paediatric and adult patient populations in settings that are resource scarce, geographically distant or under high clinical pressure.