Mortality prognostic scores in burns

Overview

Burn mortality is among the most predictable outcomes in surgery. From a handful of variables available at admission, clinicians can estimate the probability of death soon after injury on the basis of simple, objective clinical criteria. The earliest formal comparison, applied to more than three thousand consecutive admissions in Milwaukee, set Baux's rule against probit, discriminant, and logistic methods ^[1]. Decades of work since have refined the same core inputs into named composite scores used for triage, prognostic counseling, quality benchmarking, and case-mix adjustment in research [3, 18].

These tools share a common architecture. Each combines age and burn extent, with most adding inhalation injury and some adding physiologic or comorbidity data, then maps the combination to a mortality probability. They differ in how variables are weighted and whether the output is a simple sum, a points table, or a logistic transform. The practical consequence is that scores discriminate survivors from non-survivors well but calibrate poorly when applied outside their derivation population, a pattern that recurs across modern, pediatric, and resource-limited cohorts [10, 16, 40].

Pathophysiology

The predictive power of these scores reflects the biology of burn death. Larger burns, deeper full-thickness injury, inhalation injury, and older age each independently raise the risk of death [6, 27]. In a logistic model of post-burn death, age and percent burn contributed almost equally to mortality, and the presence of inhalation injury added risk equivalent to roughly 17 additional years of age or 17 percentage points of burn ^[6]. Full-thickness depth carries independent weight: overall burn size, inhalation injury, and full-thickness extent were the leading independent predictors of death from sepsis, in that order of importance ^[27]. Because these mechanisms are captured by variables measurable on the day of injury, mortality can be predicted soon after admission from simple, objective clinical criteria ^[3].

Classification

The composite scores in routine use are built on overlapping variable sets and validated head-to-head across many cohorts ^[12].

Baux and revised Baux scores

The classic Baux score sums age in years and percent body surface area burned; the original assertion was that values over 75 indicated a very poor prognosis ^[1]. Modern care has rendered the original predictions too pessimistic, prompting a revision ^[6]. The revised Baux score adds 17 points for the presence of inhalation injury, and its inverse logit transformation, provided by calculator or nomogram, yields a precise mortality probability while remaining simple enough for mental calculation [6, 7]. In a 27-year regional cohort the Baux score remained a valid indicator of mortality risk, with a Baux50 (the score at which predicted mortality is 50%) of 109.6 and a point of futility of 160 ^[17].

Abbreviated Burn Severity Index (ABSI)

The Abbreviated Burn Severity Index was an early quantitative index that proved superior to the Baux and modified Baux rules of thumb in predicting fatalities, while remaining nearly as easy to use ^[2]. It has functioned as a robust survival indicator in validation work: in one series, 11 of 12 patients with an ABSI of 7 or less survived and 9 of 11 with an ABSI of 9 or more died ^[36]. Borderline ABSI groups (scores 7-10) are where additional risk factors shift mortality predictions most ^[37].

Ryan score

The Ryan score reduces prognostication to three dichotomous risk factors: age greater than 60 years, more than 40% TBSA burned, and inhalation injury ^[3]. The derivation cohort showed that mortality rose stepwise with the number of factors present, predicting approximately 0.3%, 3%, 33%, and 90% mortality for zero, one, two, or three factors respectively ^[3].

BOBI (Belgian Outcome in Burn Injury)

The BOBI model assigns 0-4 points for burned surface area band, 0-3 points for age band, and 3 points for inhalation injury, for a 0-10 scale ^[4]. In its development and validation work the model predicted 40 deaths against 42 observed and achieved an area under the curve of 0.94 ^[4].

FLAMES and APACHE-based models

The FLAMES score (Fatality by Longevity, APACHE II score, Measured Extent of burn, and Sex) was derived from age, day-1 APACHE II score, partial- and full-thickness burn extent, and sex ^[5]. In its derivation cohort FLAMES reached an AUC of 0.97, exceeding APACHE II alone (0.91), and retained an AUC of 0.93 in validation ^[5]. APACHE II itself performs well even without burn-specific parameters in critically ill patients with moderate-to-severe burns ^[12]. The Sequential Organ Failure Assessment, a general dynamic organ-dysfunction score, was a valid instrument for predicting 30-day mortality in critically ill burn patients, with very good discrimination (AUC 96.4%) in one developing-country burn-ICU cohort ^[39].

Methodological landscape

A systematic review of composite prediction models published between 1949 and 2010 identified 45 studies but found that only 8 models met published methodological standards for construction and validation, including the modified Baux score, ABSI, the Ryan model, and BOBI ^[8].

Assessment

The variables that populate these scores are consistent across derivation studies. Age, total burn surface area, and inhalation injury are the dominant independent predictors of in-hospital mortality, with full-thickness extent and premorbid conditions adding further risk [38, 3]. A logistic model built on full-thickness burn size, age, age-squared, and inhalation injury captured the nonlinear effect of advancing age ^[26]. In nationwide and registry validations, sex, age, total burn area, full-thickness area, and inhalation injury were the recurring risk factors ^[10].

Discrimination is consistently strong. In a Japanese nationwide registry the Baux score, revised Baux, and ABSI all achieved AUROCs of approximately 0.94, and the Baux score emerged as an optimal model for that population ^[11]. A meta-analysis of 34 studies and 98,610 patients found the classic Baux, revised Baux, and FLAMES scores carried the highest discriminative power, with FLAMES exhibiting the single highest discriminative ability among the risk-assessment models studied ^[9]. Even so, a large validation of four burn-specific models found good discrimination (all AUC >0.89) but poor fit to observed mortality, with the models requiring a newly derived logistic equation to restore calibration ^[10]. External validation in an Indian tertiary cohort placed ABSI, revised Baux, and FLAMES in the fair-discrimination range (AUROC 0.71-0.75), with ABSI and revised Baux fitting the local population but FLAMES failing the Hosmer-Lemeshow test ^[13].

Machine-learning and biomarker-augmented models

Machine-learning approaches have a long history in this field, beginning with artificial neural networks that predicted survival with more than 98% accuracy in one early series ^[21], with a separate cohort reaching 90% training accuracy ^[22]. Contemporary models identify the same dominant features: full-thickness burns, age, and total burn surface area remain the most important predictors in machine-learning analyses, with red-cell distribution width emerging as an additional signal [24, 25]. A random-forest model for 90-day mortality after burn surgery reached an AUC of 0.922 ^[25]. In patients with burns of at least 50% TBSA, FLAMES retained the largest AUC (0.875) and a new nomogram built on age, %TBSA, full-thickness area, and blood lactate discriminated well across training and external-validation sets ^[14]. Biomarker-augmented prediction is an active area; in pediatric burns, the blood urea nitrogen-to-albumin ratio predicted in-hospital mortality with power similar to ABSI and the Pediatric Baux score ^[23]. Interpretive machine-learning models are now being developed for outcomes such as 60-day mortality in burn patients with suspected infection ^[33].

Special Considerations

In children, standard adult formulae behave differently. The ABSI and Pediatric Baux showed high discrimination (AUROC 0.83 and 0.85) in severely burned children but exceedingly overestimated mortality, indicating poor calibration ^[16]. Pediatric-specific scoring such as the Pediatric Risk of Mortality (PRISM) tracks observed mortality closely, and both PRISM and ABSI are predictive of mortality in severely burned children ^[29]. Total burn surface area is the most important predictor in pediatric burns, with female sex, deeper burns, positive wound cultures, and inhalation injury adding risk ^[28].

In the elderly, age and frailty carry independent prognostic weight. In a multicenter cohort, age was associated with three-month mortality independently of TBSA, APACHE II, and the Charlson Comorbidity Index, and patients aged 80 and older had significantly poorer outcomes irrespective of injury and critical-illness severity ^[32]. A frailty score on admission independently increased the risk of mortality (odds ratio 1.67) and of discharge to a skilled nursing facility (odds ratio 2.5) in elderly burn patients ^[20]. A frailty scoring system distinguished survivors from non-survivors in elderly burns admitted to high-dependency or intensive care, where survivors more often underwent surgical debridement ^[19].

In resource-limited settings, the scores require local recalibration. In a developing-country cohort the revised Baux score was more accurate than the classic Baux but was characterized as best reserved for adult and elderly prognostication in that setting ^[30]. Mortality rates in developing-country burn populations approach 34%, far above high-income benchmarks ^[39].

Outcomes

Score-based prognostication also frames population-level survival trends. The LA50, the burn size lethal to 50% of patients, has risen markedly with improved care and falls steeply with age: in one institutional series the LA50 was 76.4% TBSA in the 15-44 age cohort but only 30.8% in patients older than 65 ^[17]. In a contemporary multicenter elderly analysis the lethal dose for half of patients aged 80 and older was just 20.5% TBSA ^[32]. The standardized mortality ratio, computed by dividing observed deaths by deaths expected from ABSI, lets centers benchmark performance against case-mix; one national analysis reported a global SMR of 0.99 ^[18].

The scores predict death well but do not capture survivorship. Mortality prognostication scores did not predict long-term, health-related quality of life after burn injury: higher revised Baux and Ryan scores correlated weakly with worse one-year physical health but not mental health, and all models explained little of the variance in SF-12 scores ^[15].

Controversies and Evidence Gaps

Which score to use remains unsettled because, in head-to-head validation, the leading models discriminate similarly. In a Japanese registry the Baux, revised Baux, and ABSI scores all clustered around an AUROC of 0.94 with no clinically decisive separation ^[11]. Across multicenter validations the older models have delivered predictive performance comparable to the newer ones ^[12], and a six-model comparison in a developing-country burn population evaluated APACHE II, ABSI, BOBI, the Ryan model, revised Baux, and FLAMES head-to-head ^[35].

The dominant problem is calibration drift rather than discrimination. Modern care has rendered the original Baux predictions too pessimistic, and several scores overestimate mortality when applied to current populations, prompting repeated recalibration and modified-score derivations [6, 31]. The ABSI in particular has been shown to overestimate mortality in severely burned patients, leading to a modified, more accurate version that dropped sex as a predictor ^[31]. Some authors argue that former prediction scores are losing accuracy over time as survival improves ^[24].

The age-versus-burn-size weighting is also contested. The revised Baux derivation found age and percent burn contributed almost equally, but elderly-specific data show age and frailty predict death independently of TBSA and comorbidity indices, suggesting the linear additive weighting understates age in older patients [6, 32]. Finally, the promise of machine learning is unproven for this task: although some machine-learning methods performed marginally better than logistic regression, the differences were seldom statistically significant or clinically substantial, and established logistic models perform well against more complex alternatives ^[34].

References

[1] Stern et al. "Comparison of methods of predicting burn mortality." Scandinavian journal of plastic and reconstructive surgery 1979. PMID: 451472 https://pubmed.ncbi.nlm.nih.gov/451472/

[2] Tobiasen et al. "Prediction of burn mortality." Surgery, gynecology & obstetrics 1982. PMID: 7071708 https://pubmed.ncbi.nlm.nih.gov/7071708/

[3] Ryan et al. "Objective estimates of the probability of death from burn injuries." The New England journal of medicine 1998. PMID: 9449729 https://pubmed.ncbi.nlm.nih.gov/9449729/

[4] Belgian Outcome in Burn Injury Study Group. "Development and validation of a model for prediction of mortality in patients with acute burn injury." The British journal of surgery 2009. PMID: 19109825 https://pubmed.ncbi.nlm.nih.gov/19109825/

[5] Gomez et al. "The FLAMES score accurately predicts mortality risk in burn patients." The Journal of trauma 2008. PMID: 18784578 https://pubmed.ncbi.nlm.nih.gov/18784578/

[6] Osler et al. "Simplified estimates of the probability of death after burn injuries: extending and updating the baux score." The Journal of trauma 2010. PMID: 20038856 https://pubmed.ncbi.nlm.nih.gov/20038856/

[7] Williams et al. "A nomogram for calculation of the Revised Baux Score." Burns : journal of the International Society for Burn Injuries 2015. PMID: 24934520 https://pubmed.ncbi.nlm.nih.gov/24934520/

[8] Hussain et al. "Predicting survival in thermal injury: a systematic review of methodology of composite prediction models." Burns : journal of the International Society for Burn Injuries 2013. PMID: 23384617 https://pubmed.ncbi.nlm.nih.gov/23384617/

[9] Mrad et al. "Risk Models to Predict Mortality in Burn Patients: A Systematic Review and Meta-analysis." Plastic and reconstructive surgery. Global open 2022. PMID: 36569241 https://pubmed.ncbi.nlm.nih.gov/36569241/

[10] Zhou et al. "Validation of four burn-specific prognostic models in a cohort of 9625 cases, and a novel model for prediction of mortality in burn patients." Burns : journal of the International Society for Burn Injuries 2020. PMID: 32900549 https://pubmed.ncbi.nlm.nih.gov/32900549/

[11] Yoshimura et al. "Comparison of prognostic models for burn patients: A retrospective nationwide registry study." Burns : journal of the International Society for Burn Injuries 2020. PMID: 33148486 https://pubmed.ncbi.nlm.nih.gov/33148486/

[12] Halgas et al. "A comparison of injury scoring systems in predicting burn mortality." Annals of burns and fire disasters 2018. PMID: 30374258 https://pubmed.ncbi.nlm.nih.gov/30374258/

[13] Sharma et al. "External Validation of Three Burn-Specific Mortality Prediction Models in Adult Burn Patients at a Tertiary Care Hospital in India." Indian journal of plastic surgery : official publication of the Association of Plastic Surgeons of India 2023. PMID: 36998930 https://pubmed.ncbi.nlm.nih.gov/36998930/

[14] Wang et al. "Models predicting mortality risk of patients with burns to ≥ 50% of the total body surface." Burns : journal of the International Society for Burn Injuries 2024. PMID: 38490836 https://pubmed.ncbi.nlm.nih.gov/38490836/

[15] Stewart et al. "Mortality prognostication scores do not predict long-term, health-related quality of life after burn: A burn model system national database study." Burns : journal of the International Society for Burn Injuries 2021. PMID: 33092898 https://pubmed.ncbi.nlm.nih.gov/33092898/

[16] Tsurumi et al. "Do standard burn mortality formulae work on a population of severely burned children and adults?" Burns : journal of the International Society for Burn Injuries 2015. PMID: 25922299 https://pubmed.ncbi.nlm.nih.gov/25922299/

[17] Roberts et al. "The Baux score is dead. Long live the Baux score: a 27-year retrospective cohort study of mortality at a regional burns service." The journal of trauma and acute care surgery 2012. PMID: 22310134 https://pubmed.ncbi.nlm.nih.gov/22310134/

[18] Angulo et al. "Mortality analysis of adult burn patients in Uruguay." Revista Brasileira de terapia intensiva 2020. PMID: 32401983 https://pubmed.ncbi.nlm.nih.gov/32401983/

[19] Masud et al. "The use of a frailty scoring system for burns in the elderly." Burns : journal of the International Society for Burn Injuries 2013. PMID: 22575336 https://pubmed.ncbi.nlm.nih.gov/22575336/

[20] Romanowski et al. "Frailty score on admission predicts outcomes in elderly burn injury." Journal of burn care & research : official publication of the American Burn Association 2015. PMID: 25383979 https://pubmed.ncbi.nlm.nih.gov/25383979/

[21] Frye et al. "Simulated biologic intelligence used to predict length of stay and survival of burns." The Journal of burn care & rehabilitation 1996. PMID: 8951542 https://pubmed.ncbi.nlm.nih.gov/8951542/

[22] Estahbanati et al. "Role of artificial neural networks in prediction of survival of burn patients-a new approach." Burns : journal of the International Society for Burn Injuries 2002. PMID: 12220917 https://pubmed.ncbi.nlm.nih.gov/12220917/

[23] Payza et al. "The effect of blood urea nitrogen (BUN)/albumin ratio in evaluating mortality in pediatric burn patients with 20% or greater total body surface area involvement." Ulusal travma ve acil cerrahi dergisi = Turkish journal of trauma & emergency surgery : TJTES 2026. PMID: 41789750 https://pubmed.ncbi.nlm.nih.gov/41789750/

[24] Schmidt et al. "Improvement of Predictive Scores in Burn Medicine through Different Machine Learning Approaches." Healthcare (Basel, Switzerland) 2023. PMID: 37685472 https://pubmed.ncbi.nlm.nih.gov/37685472/

[25] Park et al. "Prediction of Mortality after Burn Surgery in Critically Ill Burn Patients Using Machine Learning Models." Journal of personalized medicine 2022. PMID: 36013242 https://pubmed.ncbi.nlm.nih.gov/36013242/

[26] Clark et al. "Burn mortality. Experience at a regional burn unit. Literature review." Acta chirurgica Scandinavica. Supplementum 1987. PMID: 3475887 https://pubmed.ncbi.nlm.nih.gov/3475887/

[27] Merrell et al. "The declining incidence of fatal sepsis following thermal injury." The Journal of trauma 1989. PMID: 2810411 https://pubmed.ncbi.nlm.nih.gov/2810411/

[28] Dhopte et al. "A prospective analysis of risk factors for pediatric burn mortality at a tertiary burn center in North India." Burns & trauma 2017. PMID: 28944226 https://pubmed.ncbi.nlm.nih.gov/28944226/

[29] Berndtson et al. "Estimating severity of burn in children: Pediatric Risk of Mortality (PRISM) score versus Abbreviated Burn Severity Index (ABSI)." Burns : journal of the International Society for Burn Injuries 2013. PMID: 23768709 https://pubmed.ncbi.nlm.nih.gov/23768709/

[30] Lam et al. "Prognosis value of revised Baux score among burn patients in a developing country." International journal of burns and trauma 2021. PMID: 34336385 https://pubmed.ncbi.nlm.nih.gov/34336385/

[31] Bartels et al. "The ABSI is dead, long live the ABSI - reliable prediction of survival in burns with a modified Abbreviated Burn Severity Index." Burns : journal of the International Society for Burn Injuries 2020. PMID: 32534892 https://pubmed.ncbi.nlm.nih.gov/32534892/

[32] Dempsey et al. "Redefining the concept of the elderly burn patient: Analysis of a multicentre international dataset." Burns : journal of the International Society for Burn Injuries 2025. PMID: 40288004 https://pubmed.ncbi.nlm.nih.gov/40288004/

[33] Ren et al. "Interpretative machine learning for predicting 60-day mortality in burn patients with suspected infection." World journal of emergency medicine 2026. PMID: 42199764 https://pubmed.ncbi.nlm.nih.gov/42199764/

[34] Stylianou et al. "Mortality risk prediction in burn injury: Comparison of logistic regression with machine learning approaches." Burns : journal of the International Society for Burn Injuries 2015. PMID: 25931158 https://pubmed.ncbi.nlm.nih.gov/25931158/

[35] Salehi et al. "Comparison of six outcome prediction models in an adult burn population in a developing country." Annals of burns and fire disasters 2017. PMID: 28592928 https://pubmed.ncbi.nlm.nih.gov/28592928/

[36] Hammond et al. "Burns in octogenarians." Southern medical journal 1991. PMID: 1948214 https://pubmed.ncbi.nlm.nih.gov/1948214/

[37] Germann et al. "The impact of risk factors and pre-existing conditions on the mortality of burn patients and the precision of predictive admission-scoring systems." Burns : journal of the International Society for Burn Injuries 1997. PMID: 9232278 https://pubmed.ncbi.nlm.nih.gov/9232278/

[38] Dokter et al. "External validation of the revised Baux score for the prediction of mortality in patients with acute burn injury." The journal of trauma and acute care surgery 2014. PMID: 24553558 https://pubmed.ncbi.nlm.nih.gov/24553558/

[39] Aditianingsih et al. "Respiratory and coagulation dysfunctions on admission as independent predictors of in-hospital mortality in critically ill burn patients." Annals of burns and fire disasters 2019. PMID: 31528149 https://pubmed.ncbi.nlm.nih.gov/31528149/

[40] Brusselaers et al. "Assessment of mortality prediction models in a Ghanaian burn population." Burns : journal of the International Society for Burn Injuries 2013. PMID: 23146574 https://pubmed.ncbi.nlm.nih.gov/23146574/