Predicting Mortality in Burn Patients: Literature Review of Risk Factors for Burn Mortality and Application in Scoring Systems

SUMMARY

Despite advances in medical technology, mortality due to burn injuries remains significant. Scoring systems are aimed at allowing physicians to effectively and accurately predict the mortality of a given patient. Patients at a higher risk of death from burns include older patients over the age of 65, high-severity burn, presence of co-morbidities, and presence of inhalation injury. Constructing a burn prediction model also needs its own methodological standards. Hence, choosing a prediction model for predicting burn mortality requires careful analysis of its methodology. Attention towards mortality risk factors should be taken when treating burn patients. Tools such as burn prediction models prove helpful in aiding physicians to accurately and effectively predict a patient’s mortality, stratify patient severity, and allocate resources appropriately, especially in settings where resources are scarce, such as natural disasters. Additionally, prediction models are used to monitor patient care and for research purposes.

Introduction

Mortality from burn injuries remains significant worldwide. According to the World Health Organization (WHO), burn injuries cause an estimated 180,000 deaths annually. The majority of burn fatalities come from low- and middle-income countries (LMICs). Non-fatal burns are among the leading causes of disability-adjusted life-years (DALYs), with almost two thirds occuring in African and South-East Asia regions.¹ This is due to the general lack of surgical care resources in LMICs. This is in accordance with The Lancet Commission on Global Surgery, which stated that the major contributor to the burden of surgical diseases is burns and thermal injury.²

Burn prediction models are especially useful in low resource settings such as developing countries and places impacted by natural disasters, where there is a need for effective risk stratification as it will help with clinical decision making, guide patient counselling, and define risk groups.^3,4 Additionally, prediction models can allow physicians to change their treatment approach from prolonging life to enhancing end of life quality care.⁵ From a research point of view, risk stratification is most important in allowing intrahospital comparisons of outcome data in defined patient groups.⁶

Patients at a higher risk of death from burns include patients from a lower socioeconomic class, due to presumed unsafe environments and ignorance of safety precautions.^7,8 Studies show that patients presenting with either old age (>60 years old), extensive burns (>40% TBSA), and/or the presence of inhalation injury had a significantly higher risk of mortality, with the presence of all three risk factors being associated with a mortality rate of 90%.⁹ Whereas, the most common in-patient care cause of death is septic shock.⁸

This review will look at current predictive factors to predict mortality in burn patients immediately at admission. It also addresses the current use of prediction models in admitted burn patients.

Methods

A literature search was conducted from April to August 2021, using the terms “burn”, “prediction models”, “risk factors”, and mortality. Results were limited to the English and Indonesian language. Papers were assessed for relevance initially by title and abstract. Additional papers were identified through hand-searching of reference lists of relevant retrieved articles, and identifying relevant papers that referenced these retrieved articles. Papers were prioritized to scoring systems that are better known, with evidence and methodological standards.

Results and discussion

Burn mortality predictors

Demographic characteristics

A study in dr. Cipto Mangunkusumo General Hospital Indonesia Burn Unit found that burn injuries tend to happen more frequently in ages 16-35 years old, where blue collar workers in Indonesia are speculated to be working under unsafe conditions. Therefore, there is a higher risk of being exposed to burn injuries.⁸ In regards to mortality, patients over the age of 65 years are significantly associated with higher risk of mortality,¹⁰ as patients in old age are associated with a reduction in immune function and thinning of the skin, as well as presence of pre-existing co-morbidities.¹¹

In terms of gender, the trend across papers is that males are more likely to suffer from burn injuries than women. This is thought to be due to males being more likely to be exposed to environmental risks in the workplace compared to females.¹² A study found that only in age groups >70 years old is there a higher proportion of burn injuries in females compared to males. Occurring during retirement, males are less likely to be exposed to workplace burns. However, studies found that gender was not associated with mortality.¹¹

Burn severity

Generally, patients with TBSA >40% have a significantly higher mortality rate compared to patients with lower TBSA%, even in highly specialized burn centres.¹¹ It is evident that the relationship between TBSA% and mortality is directly proportional. TBSA exceeding 20% increases systemic vascular permeability. This results in loss of intravascular fluid to interstitial space, resulting in edema and impaired tissue perfusion.¹³ A paper concluded that >80% TBSA full-thickness burn was uniformly fatal.¹⁴ The direct relationship between TBSA% and mortality was also seen between depth of burn and mortality. Studies found that patients admitted with presence of full-thickness burns were more likely to die, with the combination of a large TBSA% and full-thickness being a significant indicator of mortality.¹⁵ This is thought to be due to the association between burn depth with delay in wound healing, re-epithelialization, and increased risk of systemic infection.¹⁶

Presence of co-morbidities

In patients with burn injuries more than 15% TBSA, there is an acute phase response that ensues,¹⁷ which has the potential to cause systemic inflammation and multiple organ dysfunction. Mortality outcome then depends on the patient’s physiological reserve.¹⁸ Hence, it is important to evaluate presence of comorbidities in burn patients. One measure for co-morbidities is the Charlson Comorbidity Index (CCI) (Table I). CCI was first developed and validated for non-burn patients with various comorbidities in intensive care unit patients. Later, CCI was also validated in burn patients, and, combined with burn specific scores, it is proposed to provide a better prediction of mortality in burn patients.^19,20 Studies then found that an increase in CCI has a directly proportional relationship to the mortality rate of burn patients.²⁰ Comorbidities were especially important in patients with moderate-to-high risk of mortality, and influences outcome in general intensive care.²¹ A study showed that comorbidities that were associated with the highest odds of mortality in CCI were HIV/AIDS, renal disease, liver disease, congestive heart failure, chronic pulmonary disease, and metastatic cancer, compared to patients who did not have previous existing conditions. Patients with other comorbidities such as obesity, alcohol abuse and cardiac arrhythmias were more likely to die compared to patients without said comorbidities.²²

Table I.

Classical and updated Charlson Comorbidity Index (CCI)

Comorbidity	cCCI score	uCCI score
Myocardial infarction	1	0
Congestive heart failure	1	2
Peripheral vascular disease	1	0
Cerebrovascular disease	1	0
Dementia	1	2
Chronic pulmonary disease	1	1
Rheumatologic disease	1	1
Peptic ulcer disease	1	0
Mild liver disease	1	2
Diabetes without chronic complications	1	0
Diabetes with chronic complications	2	1
Hemiplegia or paraplegia	2	2
Renal disease	2	1
Any malignancy without metastasis	2	2
Leukemia	2
Lymphoma	2
Moderate or severe liver disease	3	4
Metastatic solid tumor	6	6
AIDS/HIV	6	4
Maximum comorbidity score	37	24

Presence of inhalation injury

Study concluded with multivariate analysis showed that %TBSA and inhalation trauma (IHT) are the strongest predictors for mortality in burn patients.⁶ One study reported that the odds ratio of mortality with IHT is increased 2.58 times (95% CI interval 2.03-3.29).²³ Another study concluded that the presence of IHT increased the risk of mortality by 8-10 fold.²⁴ Smoke inhalation leads to lung edema, which then results in an increase of pulmonary microvascular pressure and permeability. Inhaled products of combustion and respiratory tract thermal injury are known to activate oxygen radicals and cytokines, which further contribute to pathophysiology of inhalation injury.^25,26 Furthermore, IHT increases the risk of pneumonia by 42% per day, which shows a relationship between susceptibility of the injured airway and lung bacterial infection.

Burn prediction models

Scoring systems to predict mortality of burn trauma patients have gained increased attention and acceptance over recent years. Scoring systems utilise predictive mortality factors to predict the likelihood of death for a given patient.²⁷ The purpose of burn prediction models includes classification of severity of injuries, stratification of patient groups of various treatment modalities, evaluation and monitoring of treatment, multicentre study comparison, and patient allocation to different monitoring protocols.²⁸

In 1961 Baux described the novel burn prediction model, which was:

Mortality rate = age + percentage area burned.

Since then, further research has been conducted, which has led to the modification of the Baux score - prognostic burn index (PBI), which considers burn thickness as a part of the mortality prediction model. Study by Clark et al. found that inhalation injury was then found to be a significant predictor in burn injury, leading to an update of the burn mortality model. Since then, many burn centres have utilized their own data to produce a burn prediction model. However, there have been several criticisms, where the accuracy of prediction models is called into question. A study suggested that this was because there is a lack of adherence to appropriate methodological standards and inappropriate preliminary cross-validation studies.²⁷ Equations for the mentioned prediction models can be seen in Table II.

Table II.

Equations of burn mortality prediction scores

Mortality score	Equation
Revised Baux Score	(Burn TBSA%×100)+(Age)+(17×inhalation injury)
ABSI	S=Bo+(B1) × (summed score) Summed score = age (0–20 = 1; 21–40 = 2; 41–60 = 3; 61–80 = 4; 81–100 = 5) + %TBSA (0–10 = 1; 11–20 = 2; 21–30 = 3; 31–40 = 4; 41–50 = 5; 51–60 = 6; 61–70 = 7; 71–80 = 8; 81–90 = 9; 91–100 = 10) + inhalation injury (yes = 1, no = 0) + full thickness burn (yes = 1, no = 0) + gender (female = 1, male = 0)
Total Burned Surface Index	TBS= face, neck, anterior trunk, posterior trunk, leg, thigh, foot, arm, forearm, hand scored as 3 if bilateral burn, 2 if unilateral and 1 if none. Perineum scored as 2 if burn, otherwise 1.
Coste et al.	Score=2×(Age-50)+TBSA if age>50 Years Score=TBSA<50 years
Ryan et al.	Logit= -5.89+2.58n
McGwin et al.	Logit=-7.3406+(0.0556×age)+(0.0654×%TBSA)+(1.334×inhalation injury) +(0.2052×co-existent trauma)+(0.5177 ×pneumonia)
Galeiras et al. model	X=-7.04+(0.76×Gender)+(1.23×MV)+(0.98×AGE1)+(2.36×AGE2)+(4.55×AGE3)+(1.11×TBSA1)+(2.0×TBSA2)+(2.56×TBSA3)+(3.90×TBSA4) +(1.78×FTBSA1)+(2.19×FTBSA2)+(3.63×FTBSA3) gender = 1 if a woman; MV = 1 if the participant required mechanical ventilation within 72 h after admission; AGE1 = 1 if 40-59 years; AGE2 = 1 if 60-79 years; AGE3 = 1 if 80 years; TBSA1 = 1 if 20-39% burned; TBSA2 = 1 if TBSA 40-59% burned; TBSA3 = 1 if TBSA 60-79% burned; TBSA4 = 1 if 80% burned; FTBSA1 = 1 if 10-19% burned; FTBSA2 = 1 20-59% burned; and FTBSA3 = 1 if 60% burned.
BOBI score	(Age; <50 = 0, 50-64 = 1, 65-79 = 2, 80 = 3) + (% total burn; <20=0, 20-39=1, 40-59=2, 60-79 = 3, 80 = 4) + (inhalation injury; yes = 3, no = 0)

Methodological standards for prediction models

According to a study by Wasson et al., clinical prediction models should be modelled based on a few important points. The first is a clear definition of the outcome. A tool for clinical prediction must have a clearly defined event to be predicted. The outcome should be biological rather than sociological or behavioural. In this case, the outcome is clearly defined as mortality rates of patients with burns. The second criteria for a clinical prediction model should be definition of predictive findings. Another criteria to be considered is blind assessment of outcome when appropriate, and clinical prediction.

Furthermore, it should be considered that predictors must be feasible and relevant. If the clinical tool requires a test that is not readily available at the facility, then it would not be applicable to the burn centre. The clinician should evaluate the patient population and conclude if it might reduce the applicability of the prediction tool.²⁹ In order to validate this, it is also important for the clinical prediction tool to be validated by different clinicians with different patients in a group. Apart from external validation, Laupacis et al. suggested that the clinical prediction rule should be reproducible by the same clinician.²⁹

Accuracy of the prediction rule should be a priority when modelling a clinical prediction tool. A clinical prediction rule can aid in a physician’s decision to direct resources towards patients who are at high risk. It is important for the clinician to know the expected error rate when using a prediction rule in clinical practice. Hence, if the prediction tool places a patient in the wrong risk group, the physician will be able to intervene before the patient is unnecessarily subjected to testing and interventions.²⁹ Another study suggested that it is important that description of the results in terms of sensitivity, specificity, negative and positive predictive values are shown in a study.

The effect of a clinical prediction rule on patient care should try to minimize chance of error in patient care. Even if a prediction rule has met methodological standards, it may still be inappropriate to use because of sociological or behavioural factors. Use of a predictive rule is also useful to reassure the physician about the patient’s status. Lastly, a prediction tool should be transparent as regards the mathematical techniques used in developing the tool.²⁹

Models meeting methodological standards

Modified Baux Score

In 1961, Professor Serge Baux described a simple score for predicting burn mortality using age and percent body burned. Ever since then, there have been developments in medicine that have allowed survival from burn injury to be increased. Additionally, recent studies show that inhalation injury is a significant predictor for mortality. Hence, inhalation injury was added to the Baux score as an updated version in 2010, known as the revised Baux score. The revised Baux score was modelled using the national burn repository (NBR). The author noted that the revised Baux score is a relative measure of injury severity, hence it should be further converted into its logistic regression model. Patients with scores over 140 are considered to be unsurvivable.³⁰ This study assumed that the score was better used in populations with ages between 20 to 80 years old with TBSA% of 30% to 80%.

Several external validation studies showed that the revised Baux score had a high predictive value for mortality in patients with acute burn injury.^31,32,33

One study found that the score resulted in an AUC of 0.96 for the total population, showing that the score had a high specificity and sensitivity in that population.³⁴

A limitation mentioned in the study is that the data was taken from NBR, hence there is a concern that it will not fit all patient populations. Moreover, the score does not take pre-existing comorbidities into the equation. However, taking limitations into account, the revised Baux score continues to be used due to its adherence to methodological standards of constructing a prediction score, its simplicity and applicability.³⁵

The Abbreviated Burn Severity Index (ABSI)

ABSI was derived in 1982 from a retrospective study of two burn centres. The score uses a point system for every risk factor that is present. The risk factors used in this scoring are: female gender (1 point), age (5 points), TBSA (10 points), inhalation injury (1 point), and full thickness burn (1 point), where the survival probability ranges from <10 to 90%.³⁶ The author mentioned that ABSI may be appropriate to use in triage and for evaluation of treatment outcome. Furthermore, the variables used in the score are data that are routinely collected at patient admission, emphasizing the applicability of the scoring system. From the initial study, the model was successful in predicting 82% of deaths. Furthermore, ABSI is useful to accurately categorize patients who are at risk and who are not at risk.³⁶

Multiple centres found that ABSI had a high predictive score, with the AUC found to be between 0.86-0.90.^32,37,38 Another study concluded that despite medical advances over the past 30 years since the ABSI scoring system was first published, it remains an accurate and valuable tool for the prediction of burn mortality.³⁹

Limitations were mentioned in one study conducted in Cairo, which found that awarding one point to patients aged 1-20 years old needed to be modified. This is due to physiological differences between children and adults.⁴⁰ They mentioned that children in developing countries are more prone to malnutrition due to socioeconomic standards, inappropriate first response management at home, and also child abuse present especially in lower socioeconomic areas. Absence of pregnancy in the score was also a limitation mentioned in the study.⁴⁰

Total Burn Surface Index

The authors of the Total Burn Surface (TBS) index proposed that in order to predict mortality, it would be simpler if only extent of burn surface injury was used. Presence of injury was found in 11 body parts, which were; (a) face, (b) neck, (c) front trunk, (d) rear trunk, (e) arm, (f) forearm, (g) hand, (h) thigh, (i) leg, (j) foot, and (k) perineum. All body parts were given a score of 3 if the left and right parts were affected, except the perineum. If the injury was not symmetrical, it would be given a score of 2 and no lesion was given a score of 1. The perineum, if affected, would be given a score of 2 and a score of 1 if the area was unaffected. In the original study, their multiple regression analysis found that TBS index best predicted the surviving group (R²=0,68). They also tested age, marital status, nature and depth of injury against the TBS index but it did not significantly raise the prediction of mortality. Using a cut-off point of >20, the original study was able to identify 74.6% of patients who actually died.⁴¹ The limitation for this index is that it does not have adequate external validation.⁴¹ Furthermore, the author also mentioned that the population may not be representative since they used an older population set. Furthermore, the sample used had mostly severe injuries, which increased the mortality rate. Hence there was further need for external validation.

Coste et al. model

The model proposed by Coste et al. was a validation study of composite measurement scale (CMS) to predict the risk of mortality for burned patients. In the study they mentioned that previous prediction models often came out inconsistent. For example, risk factors that contributed to mortality in one burn centre may not be significant in other burn centres - which they attributed to lack of cross-validation studies. Hence, they addressed this concern by approaching mortality prediction as a continuous phenomenon. Instead of using the prediction model at initial admission, the author proposed that mortality should be evaluated in intervals or ratios - also known as a psychometric scale.

Study found a positive correlated relationship between TBSA, patients aged over 50, and increased risk of mortality. Hence, the score only used those two variables. Furthermore, the study made a nomogram for simpler use to compute the probability of death. The model produced R² of 0.945 in patients age >50 years old, demonstrating that the model accurately predicts mortality. This model can be purposed for stratification of patients into high, moderate and low risk groups.⁴²

Ryan et al. model

The Ryan et al. model was based on a retrospective review of 1665 patients with burn injuries that were admitted to Shriners Burn Institute from 1990-1994. The factors associated with mortality in this study were age >60 years old, burn size >40% TBSA, and inhalation injury. Patients had a significantly high risk of mortality when these three factors were present. Hence, the model proposed by Ryan et al. simply needs to identify the number of risk factors present and include it in the model. Patients with 3 risk factors have approximately 90% risk of mortality with this model and it is applicable to all patients younger than 90 years old.⁴³

External study done in Malang, Indonesia comparing 4 different burn models found that the model had the highest 97.8% specificity and 78.2% sensitivity, but had the lowest accuracy of 75.6% and ROC analysis value (AUC = 0.80).³¹ One of the reasons could be due to the age demographic in this study. As stated previously, in developing countries the productive age groups are more likely to suffer from burns. Since, Ryan et al.’s model includes age as a risk factor when age is over 60 years old, the accuracy of predicting mortality decreased. Similar results were found in another study conducted in Ghana, which found that all models had good discriminative power but Ryan et al.’s performed the least. This could have been attributed to age being a risk factor only when the patient is over 60 years old, whereas the population in the study consisted mostly of children.⁴⁴

McGwin et al. model

This model uses data from the National Burn Repository (NBR) and the National Trauma Data Bank (NTDB). Age, TBSA, inhalation injury, pneumonia and co-existent trauma were used in this model. The addition of pneumonia and co-existent trauma are two variables that are specific to it. The author ran a multivariate test on different models and found that the addition of pneumonia and trauma yielded the best results (AUC 0.94). Even though literature states that inhalation injury is significant in mortality due to an increasing chance of pneumonia, this study found that the two risk factors were independent. However, this model lacks external validation.⁴⁵

Galeiras et al. model

Data used for this study was from the Burn Registry of Hospital Universitario de Getafe from 1992-2005. This study utilised age, gender, inhalation injury and early mechanical ventilation in their model. The population sampled had a high prevalence of flame injury, hence inhalation injury was a significant predictor for mortality. Additionally, mechanical ventilation needed within 72 hours was also mentioned in this model. The author concluded that patients requiring mechanical ventilation soon after admission may indicate the severity of inhalation injury.⁴⁶

Belgian Outcome of Burn Injury (BOBI)

The BOBI score was developed using data from 6 burn centres in Belgium, for a total of 6227 patients. The author mentioned that the goal was to refine the model devised by Ryan et al. by splitting age and TBSA into multiple categories. It was previously mentioned that one of the limitations to Ryan et al.’s model is that it considers age as a risk factor when the patient is over 60 years old. From the sample it was found that age over 50 years old, TBSA and presence of inhalation injury were significant in predicting burn mortality. Hence, a scoring system was formulated to be assigned for these different risk factors. Age was given a score between 0 for ages <50, 1 for 50-64, 2 for 65-79, and 3 for >80. TBSA was divided into <20, 20-39, 40-59, 60-79, and >80 with scores of 0-4 respectively. Lastly, inhalation injury was given a score of 3 if it was present and 0 if it was not present.⁴⁷

The score was then validated with a different sample where there was a strong relationship between the model and observed mortality (OR 2.7 (95% CI 2.2-3.3; p-value <0.001)). Discriminative power was done by assessing area under the curve, which was 0.95 (95% CI 0.90-0.97).⁴⁷ This has been externally validated by multiple studies. A retrospective study carried out in dr. Cipto Mangunkusumo General Hospital, Indonesia showed that the score underestimated the actual deaths with a ratio of (observed:predicted = 1.29). This could be due to higher mortality rates in Cipto Mangunkusumo General Hospital compared to Burn Centres in Belgium (17.7% compared to 4.3%). There is also a concern regarding the different socio-economic backgrounds, since Indonesia is still regarded as a developing country. Additionally, inhalation injury played a large role in mortality compared to the study in Belgium. However, despite the differences in population sample the analysis of ROC showed an area under the curve of 0.964 (95% CI 0.935-0.992),⁴⁸ which shows that the model has high distinguishing power between survivors. This emphasizes that even in a different demographic sample, the model was still able to accurately predict mortality. This is further supported by another study conducted in dr. M. Djamil General Hospital in Padang, Indonesia, which compared BOBI score to revised Baux score, with the BOBI score having a higher predictor rate compared to the latter.⁴⁹

A limitation for the BOBI score lies in the lack of co-morbidities being taken into account. Although in some cases age is an acceptable replacement for specific co-morbidities, there are specific underlying pathologies that could have influenced mortality rates. Also, it does not take into account age groups below 50 years old, especially children. Since children have different physiology compared to adults, it can be useful to differentiate this age group.

Conclusion

Burn prediction models prove to be useful for the physician to predict mortality of a patient, stratify patient severity and allocate resources appropriately, evaluate patient care, and also for research purposes. Risk factors that are most significant include age, TBSAand presence of inhalation injury. Some models also suggest that using pre-existing conditions as a predictor can aid in the prediction of mortality. However, it must be remembered that statistically there will be patients that are misplaced in a category. Hence, burn prediction models should be a guide for physicians to effectively allocate resources. Additionally, clinicians should also keep in mind a patient’s quality of life when using these