Performance characteristics of a local triage tool and internationally validated tools among under-fives presenting to an urban emergency department in Tanzania

This was a prospective descriptive study of a consecutive sample of paediatric patients below the age of five years with non-traumatic medical complaints presenting to the EMD-MNH from November 2017 to April 2018.

MNH is a national referral government hospital with 1500 beds, and attending 1500 outpatients’ everyday. It is located in Ilala district, Dar es Salaam City, Tanzania. The EMD was opened in 2010, and serves an annual average of 60,000 patients who are referred from across Tanzania. The hospital has no formal admission policy, the process of admission is based on combination of providers clinical gestalt (on severity of illness) and disease specific guidelines. About 25% of patients seen in this department are paediatric age with 75% presenting with non-traumatic medical conditions [16]. There is no validated triage tool that is used for triage of patients in the EMD.

Paediatric patients below the age of five years with non-traumatic medical complaints presenting to the EMD-MNH were eligible. Patients arriving in respiratory or cardiac arrest, returning for follow up, or who had a history of trauma within two weeks were excluded. Consent for participation in the study was requested from the adult accompanying the child.

Our study population was paediatric patients below the age of five years, and we estimated the minimum sample size of 372 assuming a 59% admission rate, based on the local EMD data for the year before this study. This sample size allowed a precise comparison of proportions of triage levels by each triage scale.

A research assistant (RA) was scheduled in the department every other day for 8–12 h during the study period alternating between day and nights. Eligible patients were consecutively enrolled after written informed consent was obtained. The RA recorded initial triage information including chief complaint, vital signs and acuity assigned by the triage nurse. During the triage encounter, the RA calculated triage acuity for each patient, using each of the following- MTS, CTAS with paediatric modifiers, ATS with paediatric modifiers and paediatric SATS (Additional file 1). Subsequent to triage, the RA recorded investigations done, ED diagnosis and disposition from review of the electronic medical record while the patient was in the ED. (Wellsoft Version 11 Corporation, Somerset, NJ, USA). Demographics and other necessary information for triage were also obtained from mother / guardian through interview. All children were followed up. For those discharged, we contacted the parents/ guardians by mobile phone (a call was made 3 times on three different days), and for those admitted; we followed the patient to the specific wards in order to obtain their 24-h outcome.

Outcomes: The outcomes of interest were the sensitivity, specificity, PPV and NPV of the triage acuity on each scale for predicting admission and mortality.

Data were entered into SPSS software (V23) and analysed. Descriptive statistics (counts, percentages, median, quartiles,) were obtained for demographic characteristics, distribution of patients in different triage categories by each triage scale, and the proportions of admissions and deaths within each triage acuity. To assess test characteristics (sensitivity, specificity, PPV, NPV) for prediction of admission and 24-h mortality of each scale, we created 2 groups from each triage scale: For ATS, CTAS and MTS: high acuity Level 1 and 2, low acuity 3–5 (to level 4 in SATS), and for LTS, high acuity level 1 and low acuity level 2 and 3.

We enrolled and followed up 384 out 2030 under fives with non-traumatic complaints who presented to EMD during times when research assistant was available. Figure 1.

Of the 384 patients that were followed up, the median age was 17 months (IQR 7–36 months) 211 (54.9%) were male. In the local triage system, 67 (17.4%) were triaged as emergency, 291 (75.8%) in priority group and 26 (6.8%) in queue group. Overall, 178 (46.4%) of patients were referred from other hospitals. Table 1.

Among the enrolled patients, 229 (59.6%) were admitted, 153 (39.9%) discharged and 2 (0.5%) died at ED. The proportion of patients admitted was highest in the high acuity groups, and decreased with level of acuity in all scales; the proportion of discharges was lowest in the high acuity groups and increased with lower acuity assignments. Five patients (1.45%) died within 24 h, including the 2 ED deaths (both were haemodynamically unstable and they were triaged as level 1 in all triage scales). The MTS, ATS, CTAS and SATS categorized all of these patients as acuity 1 or 2, while LTS categorized 4 patients as emergent and 1 as priority. Table 2.

The area under the ROC (AUROC) curve for the outcome of admission for all triage scales ranged between 0.62–0.63 (Fig. 2). Sensitivity of high acuity assignment for admission ranged between 27.1 and 29.4%, while specificity ranged from 95.4 to 98.0%. Results were similar among all triage scales. Table 3.

With regard to ability to predict 24 h mortality, 41 patients were excluded due to missing data on 24 h outcomes. The AUROC curve for mortality ranged between 0.90–0.91 for all scales except for the LTs, where the AUROC was 0.81 (Fig. 3). The sensitivity for mortality was 80% for LTS and 100% for the other scales; NPV was 99.6%. (98–100%) for LTS and 100% (98–100%) for all other scales. Table 4.