The modified south African triage scale system for mortality prediction in resource-constrained emergency surgical centers: a retrospective cohort study

Emergency triage is the systematic process of determining the priority for treatment based on the severity of the patients’ conditions. The principal aim of triage is to ensure that patients receive the most appropriate level of care according to their clinical status, while focusing attention on patients at a higher risk of death [1]. Identifying patients at a high risk of death is important in the emergency department (ED) to offer adequate treatment and to recognize patients in need of more intensive management and possible admission to an intensive care unit (ICU). Triage is recognized as a central component of the ED and was first introduced in the 1950s in the USA [2, 3]. More recently, the need for triage systems was also identified in low-resource settings with reports showing that the process of triage can improve patient flow, reduce patient waiting times, and decrease mortality in these contexts [4, 5]. These contexts can be characterized by few health resources including a limited number of physicians or qualified nurses and/or limited drugs or health materials.¹ As one of the only validated tools that exists for the triage of patients in low-resource settings, Médecins Sans Frontières (MSF) has chosen the South African Triage Scale (SATS) [6‐8] as a standard tool for its EDs such as Nap Kenbe Surgical Center in Tabarre, Port-au-Prince (Haiti) and in Kunduz (Afghanistan) [9, 10]. The SATS uses a physiology-based scoring system, the Triage Early Warning Score (TEWS), and a list of discriminators designed to triage patients into one of four color-coded priority groups for medical attention. Physiological variables used to compute the TEWS are mobility, temperature, systolic blood pressure, heart rate, respiratory rate, and neurological status. The TEWS also depends of the presence of trauma. Patients with a TEWS >6 are distributed into the red group, a TEWS of 5 to 6 into the orange group, a TEWS of 3 to 4 into the yellow group, and a TEWS of 0 to 2 into the green group. A fifth group (blue) allows to deal with the patients who looks obviously dead at admission. Discriminant parameters can move a patient into a higher priority group. Convulsions, burns on the face, and hypoglycemia < 55 g/dL were criteria to include patients in the red group; high-energy transfer, non-controlled hemorrhage, acute dyspnea, hemoptysis, thoracic pain, open fracture, member dislocation, member ischemia, post-epilepsy coma, focal neurological deficit, alteration of consciousness, burns over 20% of the body, burns from electricity or chemicals, circumferential burns, intoxication, and overdose were criteria for inclusion in the orange group; and controlled hemorrhage, closed fracture, burns over less than 20% of the body, finger or toe dislocation, and abdominal pain were criteria for inclusion in the yellow group. More specific discriminants are used under the age of 13 years. Following triage, patients in the red group must be managed immediately before patients from the other groups according to their priority. Gottschalk reported a detailed description of the SATS first know as the Cape triage score that contains more details and a SATS score table [8].

The aim of our study was to verify if data from the SATS system combined with other easily available patient characteristics can facilitate the identification of patients at high risk of mortality. Such patients could then receive more focused supportive care during their inpatient stay. For this purpose, we constructed and validated a prognostic model based on information available from the ED, including the reason for admission according to the MSF classification (Table 1) and data from the SATS system, and compared the model’s discriminative power with that of a model based on the four categories of the SATS alone. We hypothesized that the combined model would allow for better identification of patients at higher risk of in-hospital mortality, as it would take into account other independent factors linked to mortality. That model could be useful for audit purpose but also for clinical purpose as it would focuses attention on underestimated factors linked to mortality.

At the Nap Kenbe Hospital from January 1, 2013 to June 30, 2015, there were 7618 patients hospitalized from the ED. The male sex predominated (75%) and the mean age of the patients was 29 ± 17 years, with ages ranging from 0 to 102 years. The most frequent reasons for admission were trauma-related lesions due to traffic and road accidents (3070, 40%), work, sport, and domestic accidents (2356, 31%), gunshot wounds (991, 13%), wounds from knives (510, 7%), and non-traumatic causes (e.g. acute surgical conditions such as peritonitis or obstructions) (551, 7%). Of the 7618 admitted patients, 171 (2.22%) died during hospitalization (Table 1). Thirty-seven of these patients died in the ED, 25 in the operating room, and 104 (56% of all deaths) in the ICU. The classification of all hospitalized patients according to the SATS, including those who died in the ED or during hospitalization, was 617 (8%) in the red category, 3106 (40%) in the orange category, 3601 (47%) in the yellow category, and 294 (4%) in the green category. Multivariate analysis identified the red and orange categories of the SATS, non-traumatic reasons for admission, age between 45 and 65 years and age over 65 years as independent factors associated with mortality, and these were retained in the final combined model (Table 2). No trauma category was retained in the multivariate regression analysis as an independent factor linked to mortality. The score of the combined model can be computed using the coefficients resulting from the logistic regression:where R is 1 if the patient is in the red category and 0 otherwise, O is 1 if the patient is in the orange category and 0 otherwise, NT is 1 for non-trauma patients and 0 otherwise, Age45_65 is 1 if the patient is older than 45 and lower or equal to 65 years and 0 otherwise, Age65 is 1 if the patient is older than 65 and 0 otherwise. The probability of death can be computed using the following formula:

The combined model divided the population into 18 groups with increasing probabilities of death according to the combination of parameters in the final combined model (Table 3).

In this retrospective cohort study, we illustrated how routine ED data, such as information from the SATS system combined with the reason for admission and age of the patient can be used to predict in-hospital mortality. The described model showed good discrimination in identifying patients at a higher risk of in-hospital death. The SATS is a triage system that is easy to use and implement, and was demonstrated to have been successfully implemented and used in resource-constrained settings outside South Africa [9, 10]. Our combined model allows for the identification of patients meriting enhanced monitoring and supportive care during their inpatient stay because of an increased risk of death.

The use of routinely available parameters to assess mortality risk is one of the main strengths of our study, leading to low rates of missing data. The sample size of the cohort of analyzed patients was large enough to offer good statistical power. The combined model was internally validated using the bootstrapping technique, which is the most widely recommended for internal validation as it allows derivation of the final model from the full derivation sample and does not waste precious information by developing the model from a random part of the original data-set, as with the splitting method [11, 12]. The combined model was also applied on an external data-set, accurately predicted in-hospital mortality in the external population, and had better prognostic performance than the model restricted to the four SATS categories for both populations. However, the entire model could not be validated on that external population of only trauma patients.

Different risk scoring systems for the triage of emergency patients have been described [13‐19], but their performances outside the contexts and settings used in their validations cannot be warranted [20]. Furthermore, several of these models were designed to be used in a non-surgical ED and cannot be used in our surgical settings. The discriminative power of the combined model presented in this study, with an AUROC corrected for optimism of 0.8601, compares well with the powers of models described in the previous literature. The AUROCs of these models ranged from 0.65 to 0.90 for the best performing model, which has an optimism-corrected AUROC of 0.90 [19], or with the inclusion of blood test results [13]. Our model is more adapted to resource-constrained settings and does not rely only on complex computations to predict the risk of death for every patient, as the data from Table 3 can simply be used for this purpose.

The surgical characteristics of the patients in the cohort limit the usefulness of the model to surgical facilities. Another limitation concerns the proportion of trauma versus non-trauma cases in the studied population, where trauma cases were more frequent. Caution is needed when applying the model to populations of other centers, as the results from the study may depend on the nature of the emergency surgical center and the proportion of trauma versus non-trauma cases in the treated population. The exclusion of patients discharged from the ED or transferred to others institutions in our study may also have influenced the results in some extend.

In conclusion, in this cohort study of 7618 surgical patients, we found that a prediction model derived from SATS system data, combined with the reason for admission and age of the patient, all readily available after the triage of patients at arrival in the ED, was a good predictor of the subsequent risk of death. Such a model is well adapted to use in resource-constrained surgical settings, predicted well mortality when applied to an external data-set of trauma patients, and performed better than a model based only on the SATS categories.