Article Text

Download PDFPDF

The civilian validation of the Modified Physiological Triage Tool (MPTT): an evidence-based approach to primary major incident triage
  1. James Vassallo1,2,
  2. Jason Smith3,4,
  3. Omar Bouamra5,
  4. Fiona Lecky5,6,
  5. Lee A Wallis1
  1. 1 Division of Emergency Medicine, University of Cape Town, Cape Town, South Africa
  2. 2 Institute of Naval Medicine, Gosport, UK
  3. 3 Department of Emergency, Derriford Hospital, Plymouth, UK
  4. 4 Department of Military Emergency Medicine, Royal Centre for Defence Medicine (Research and Academia), Medical Directorate, Joint Medical Command, Birmingham, UK
  5. 5 Trauma Audit and Research Network, Hope Hospital, Manchester, UK
  6. 6 Health Services Research Section, School of Health and Related Research, University of Sheffield, Sheffield, UK
  1. Correspondence to Dr James Vassallo, Institute of Naval Medicine, Alverstoke, Gosport PO12 2DL, UK; vassallo{at}


Introduction Triage is a key principle in the effective management of a major incident. Existing triage tools have demonstrated limited performance at predicting need for life-saving intervention (LSI). Derived on a military cohort, the Modified Physiological Triage Tool (MPTT) has demonstrated improved performance. Using a civilian trauma registry, this study aimed to validate the MPTT in a civilian environment.

Methods Retrospective database review of the Trauma Audit and Research Network (TARN) database for all adult patients (>18 years) between 2006 and 2014. Patients were defined as Priority One if they received one or more LSIs from a previously defined list. Only patients with complete physiological data were included. Patients were categorised by the MPTT and existing triage tools using first recorded hospital physiology. Performance characteristics were evaluated using sensitivity, specificity and area under receiver operating characteristic (AUROC).

Results During the study period, 218 985 adult patients were included in the TARN database. 127 233 (58.1%) had complete data: 55.6% male, aged 61.4 (IQR 43.1–80.0) years, Injury Severity Score 9 (IQR 9–16), 96.5% suffered blunt trauma and 24 791 (19.5%) were Priority One. The MPTT (sensitivity 57.6%, specificity 71.5%) outperformed all existing triage methods with a 44.7% absolute reduction in undertriage compared with existing UK civilian methods. AUROC comparison supported the use of the MPTT over other tools (P<0.001.)

Conclusion Within a civilian trauma registry population, the MPTT demonstrates improved performance at predicting need for LSI, with the lowest rates of undertriage and an appropriate level of overtriage. We suggest the MPTT be considered as an alternative to existing triage tools.

  • triage
  • major incidents
  • life-saving interventions
  • physiological parameters

Statistics from

Key messages

What is already known on this subject?

  • There is increasing evidence to suggest that existing primary major incident triage tools perform poorly at predicting the need for a life-saving intervention.

  • Derived using logistic regression, the Modified Physiological Triage Tool (MPTT) demonstrated improved performance at predicting need for life-saving intervention with the lowest rates of undertriage and acceptable rates of overtriage within a military population. It is the first physiological triage tool to be derived specifically to predict need for life-saving intervention.

What this study adds?

  • This study demonstrates that existing methods of primary major incident triage have unacceptably high rates of undertriage, with the existing MIMMS Triage Sieve performing the worst. The MPTT outperforms all existing triage tools with the lowest rates of undertriage, while maintaining an acceptable level of overtriage, comparable to that observed following the London 7/7 bombings.


Major incidents occur worldwide on an almost daily basis, ranging from natural disasters to transport incidents to terrorist atrocities.1 For the health services, they are defined as incidents requiring ‘extraordinary resources’ in order to manage the number or severity of casualties.2 Over the last decade, we have seen an increase in terrorism-related incidents directed towards civilians worldwide, and notably the 2015 Paris marauding terrorist firearms attacks, which produced patterns of injuries in civilian casualties that are more akin to that seen in the military setting than had previously been observed.3

Triage is the process of determining a patient’s clinical priority and is a key step for the effective management of major incidents. Its origins as a clinical sorting process date back to 1846, when Wilson, a Royal Naval Surgeon described sorting patients into groups corresponding to slight, serious or fatal.4 A key tenet of major incident triage is that it must be rapid, reliable and reproducible, irrespective of the provider performing it.2 Most methods of major incident triage use physiology to guide allocation to a particular triage category, with Priority One or Immediate being the most acute.5 This process is used to either prioritise patient evacuation or to predict patient need for a life-saving intervention.6 7

There is limited evidence to support the use of the three commonly used civilian major incident triage tools (Triage Sieve,2 START8 and Careflight,5 see table 1) with a number of studies demonstrating poor accuracy at predicting need for life-saving intervention.5 9 10 Following the London 7/7 bombings, a study of the patients treated at the Royal London Hospital found that all three triage tools had an undertriage rate of 50%.11

The Modified Physiological Triage Tool (MPTT) was derived on a military cohort, using logistic regression models for each individual physiological variable.12

Within a military population, the MPTT significantly outperformed existing triage tools at predicting need for life-saving intervention, with the lowest rate of undertriage (30.1%) while minimising rates of overtriage (35.2%).12 However, this was a young population (median 24 years, IQR 21–29 years) sustaining predominantly blast injuries (55% were injured by explosion).

Table 1

Comparison of existing triage tools5 12

Despite outperforming existing triage tools in a military population, no studies have evaluated the performance of the MPTT in a civilian population. Before the MPTT can be suggested as a replacement to the Triage Sieve in a civilian major incident setting, an evaluation of its performance in this population needs to be undertaken.

Ideally, this should be in the major incident setting, under the circumstances in which the MPTT is expected to operate. However, owing to the unpredictable nature of major incidents, prospective research into the development of novel triage algorithms is impractical. Instead, we analyse major incidents retrospectively or use trauma databases as a source of injured patients; while the retrospective analysis of major incidents conveys the advantage of utilising a genuine scenario, previous attempts to use real major incidents have been hampered by small numbers of seriously injured patients. With small sample sizes, the ability to draw reliable conclusions on a triage tools’ ability to predict need for life-saving intervention is therefore limited. By contrast, the use of a trauma database allows for the comparison of triage tools using large numbers of injured patients, testing their performance at predicting those in need of life-saving intervention following a variety of trauma mechanisms. We therefore aimed to validate the use of the MPTT on a civilian population using the UK Trauma Audit and Research Network (TARN) database.


A retrospective database review was undertaken using the TARN database from 1 January 2006 to 31 December 2014. All adult (≥18 years) patients with trauma meeting TARN inclusion criteria presenting to hospitals in England and Wales were eligible (

Established in 1988, TARN is the largest trauma database in Europe collecting data on patients sustaining moderate to major traumatic injuries from all trauma receiving hospitals in the England and Wales. Data are submitted electronically by trained clerical staff from the receiving hospital to TARN and the data follow the patient pathway from injury to discharge. TARN eligibility includes patients with trauma admitted to hospital ≥3 days, critical care unit admission or who die in hospital.13 Only direct admissions from scene of injury were included and patients with incomplete physiological data were excluded. Patients declared dead at scene and not conveyed to hospital are not included in the TARN database and therefore were not included in our analysis. Due to the nature of the TARN database and its inclusion criteria, patients were assumed to be non-ambulant.

In keeping with the derivation study, outliers, defined as HR >170 beats per minute, respiratory rate >45 breathes per minute and systolic blood pressure >206 mm Hg were removed.12 Patients were defined as Priority One (P1) if they received one or more life-saving interventions from a previously defined list, derived through international consensus of experts involved in major incident management (see table S1 in the online supplementary file 1).7 Using first recorded ED physiology, patients were categorised using existing triage tools (START, Careflight, Military Sieve, Triage Sieve).5 8 14

Supplementary file 1

Not all life-saving interventions are recorded as variables on the TARN database, requiring surrogates to be used (see table S1 in the online supplementary file 1). These were determined prior to the database analysis and represent the closest approximation to the interventions required. Additionally, in keeping with previous work, a systolic blood pressure surrogate of 90 mm Hg was used to represent the presence of a radial pulse for the purposes of prioritisation using START and Careflight, as it is not a recorded variable on the TARN database.

Our primary outcome was a comparative analysis of the test performance of the MPTT with existing major incident triage tools at predicting need for life-saving intervention. Secondary outcomes were to evaluate the performance of the MPTT using a subgroup analysis split by gender, age and mode of injury. For all triage tools sensitivity, specificity, undertriage (1-sensitivity) and overtriage (1-positive predictive value) with 95% CIs were calculated.15 Using a McNemar test, tools with similar performance characteristics were evaluated for any statistically significant difference in performance.16

SPSS V.23.0 (SPSS, Chicago, Illinois, USA) and STATA V.12.0 (StataCorp, College Station, Texas, USA) were used for data processing, multiple imputation and analysis.

Missing data

A comparison was made between the complete-data and missing-data patient groups to evaluate for a systematic difference with respect to age, Injury Severity Score (ISS), outcome and requirement for life-saving intervention. Performing a list-wise deletion on patients without complete data can introduce systematic errors. Missing data were investigated using multiple imputations under a missing at random assumption using chained equations.17 A comparative analysis was performed on the imputed dataset. The imputation modelling strategy consisted of the following variables: ISS, age, 30-day outcome, gender, mechanism of injury and P1 status. The missing data method was utilised using the ice procedure in STATA with five sets of imputed data generated.


During the study period, 218 985 adult patients met TARN inclusion criteria with 127 233 included in our analysis (figure 1 breakdown). Median age was 61.4 years (IQR 43.1–80.0 years) with males accounting for 55.6% cases (n=70 747).

Overall, 30-day mortality was 5.7% (n=7266). Injury secondary to falls from low height (<2 m) accounted for the majority of cases (n=68 354; 53.7%) with limbs the most frequently injured body region (n=73 755; 38.9%). ISS was recorded for all patients, with a median and mean of 9 and 11.9, respectively. Additional study characteristics are presented in table 2. A 24 791 (19.5%) patients received one or more life-saving interventions and were considered Priority One. Intubation and ventilation were the most frequent life-saving intervention (n=8813, 20.7%).

Table 2

Characteristics of study population

A summary of triage tool performance is shown in table 3. The MPTT demonstrated the highest sensitivity of all existing triage tools (57.6%; 95% CIs 56.9% to 58.2%) with an absolute increase of 44.7% over the existing UK civilian Triage Sieve (12.9%; 95% CIs 12.5% to 13.4%). Full test characteristics are shown in table 4.

Table 3

Triage tool summary of results

Table 4

Test characteristics with 95% Confidence Intervals

Using a McNemar test with Bonferroni correction (α=0.05/4=0.0125), a statistically significant difference in performance was again observed between the MPTT and the Military Sieve (χ2=30,405, P<0.001) and the MPTT and the Triage Sieve (χ2= 36,804, P<0.001).

Missing data

Statistical significance was observed for both age and gender (P<0.001) between the missing and complete data groups; however, observationally, the relative frequencies were similar for missing versus complete (55 vs 61 years and 62.2% vs 55.6% male).

The 30-day mortality was significantly higher in the missing data group (10.1% vs 5.7%, P<0.001) and was associated with a greater proportion requiring life-saving intervention (34.7% vs 19.5%, P<0.001). A statistical significance (P<0.001) was observed in median ISS between the missing data group (10 (IQR 9–24)) and complete data group (9 (IQR 9–16)).

Performance was largely unchanged following multiple imputation to account for missing data under a missing at random analysis. The performance of the MPTT remained superior to existing triage tools with 60.2% sensitivity and 71.3% specificity. Full test characteristics following multiple imputation are provided in table S2 in the online supplementary file 2.

Supplementary file 2

Subgroup analysis

Injury type

Patients sustaining penetrating trauma received a greater number of life-saving interventions when compared with blunt trauma (62.7% vs 17.9%). Rates of undertriage were lower for all triage tools with a penetrating mechanism, but this must be interpreted with caution due to the low numbers (3.5%). For blunt trauma, in keeping with the main data analysis, the MPTT was seen to have the lowest rate of undertriage, with the highest overtriage rate.


The study population was split into age ranges 18–25 years, 26–49 years, 50–74 years and 75+ years in keeping with previous TARN publications.13 Falls<2 m increased dramatically throughout the age ranges, accounting for 10% of injuries in the under 25s through to 85% in those over 75 years of age. For all triage tools, there was a trend of increasing undertriage and overtriage throughout all age groups, with the MPTT having the lowest rate of undertriage across all age groups, although at the expense of overtriage (see figure S1 in the online supplementary file 3: Relationship between triage tool performance and age range.)


Large differences in overtriage rates were observed for all triage tools, ranging from an additional 15.5% (MPTT) to 19.2% (Careflight). By comparison, undertriage rates were similar irrespective of gender for all triage tools.


A key limitation of our work is the use of a retrospective trauma database in which to validate the MPTT, the injury pattern observed following a major incident may not reflect that on the database. Ideally, any validation should be conducted in the environment where the tool is to be used in practice. Owing to the unpredictable nature of major incidents, this is largely unpractical and frequently results in the use of trauma databases as a surrogate. We acknowledge that by conducting our study in this way, we are unable to recreate the environment in which the MPTT would be used in real life. However, by performing our analysis on the TARN trauma database, we are able to reliably test individual triage tools’ performance at predicting the need for life-saving intervention on a large number of seriously injured patients.

While the proportion of patients not receiving a life-saving intervention in our study was 80.5%, the presence of inclusion criteria for the TARN database is likely to skew the study population towards those sustaining a higher mean severity of injury. Therefore, it can be expected that the actual population frequency of patients not receiving a life-saving intervention will be higher than observed in our study. We recognise this as a limitation of our study and therefore relative caution must be taken when interpreting the specificity of all triage tools in our comparison.

Thirdly, not all life-saving interventions are recorded as variables in the TARN database, requiring us to use a number of surrogates in order to conduct the study (see table S1 in the online supplementary file 1). These surrogates were chosen to represent the closest approximation to the life-saving interventions required. While our final study population is large (127 233 patients), we acknowledge that an additional limitation is the exclusion of those with incomplete physiological data. While the demographics of the missing data population are comparable to the complete data set, we observed significant differences in outcome and need for life-saving intervention between the two groups. In order to explore and mitigate the effect of excluding missing data, we performed an additional performance analysis, employing multiple imputation for missing values. Little difference was observed between the two datasets with the MPTT continuing to demonstrate superior performance characteristics to existing triage tools.


There is a paucity of evidence examining the performance of existing adult major incident triage tools, with a number of contradictory studies in the literature.

Despite using retrospective major incident cohort’s in which to perform their analyses, both Challen’s and Kahn’s studies are limited largely by the small numbers of genuine P1 patients (eight and two, respectively). Additionally, Kahn’s study is limited by the evaluation of START in isolation and is not a triage tool comparison.8 9 Similar to our study, both Garner and Cicero used trauma registries in which to perform a comparative analysis. Despite being a large study, the applicability of the work by Cicero is largely by the use of ISS and mortality as the outcome measure; the ISS is a retrospective measure of injury and bears little correlation to clinical acuity and the resource needs of a patient. This precludes direct comparison with our study.12 18 19

While the specificities reported by Garner are similar to those in the literature, the sensitivities differ considerably.5 The definition of the P1 patient was the same for both Challen and Garner, whereas a more comprehensive definition, derived through consensus to represent current methods in trauma management was used for the purpose of this validation.5 7 9 This is likely to explain the differences in sensitivity, with a comparison by study shown in the table 5.

Table 5

Comparative analysis by study with 95% Confidence Intervals5 9 12 27

There are a number of challenges associated with major incident research, not limited solely to the practical conduct of such studies. One such challenge is determining what is successful triage. In an ideal world, the methods we use for triage will correctly identify all patients with high levels of sensitivity and specificity, without incorrectly triaging patients to higher (overtriage) or lower (undertriage) categories. Studies to date have shown that with simple physiological triage this is not possible; with high sensitivity comes low specificity and so the performance of the optimum triage tool is a balance of accepting overtriage and undertriage. An additional challenge is how to define overtriage and undertriage. In keeping with previous studies measuring triage tool accuracy, we have used 1-positive predictive value and 1-sensitivity to calculate over and undertriage, respectively.15 20 21 Alternative measures such as 1-specificity for overtriage22 and 1-negative predictive value21 for undertriage have been described elsewhere.

The MPTT, derived using individual logistical regression models for each physiological parameter, had the lowest rate of undertriage and approximately equal rates of overtriage and undertriage (35.2% vs 30.1%). The methodology behind its derivation is likely to suggest that this represents the limit of the capability of physiological triage at predicting need for life-saving intervention.12

Overall success of triage is not based solely on sensitivity or the identification of those in need of life-saving intervention. As with any diagnostic test, increasing triage tool sensitivity comes at the expense of lower specificity and there will be a number of patients who are incorrectly classified. A successful primary major incident triage tool needs to provide not only high sensitivity, but a compromise between those incorrectly classified (undertriage/overtriage). While the effects of undertriage are clearly apparent (failing to identify a patient in need of a life-saving intervention), overtriage in itself can be harmful as well. Previous studies have shown that a consequence of overtriage is the potential to overwhelm hospital resources, with a direct association between overtriage and critical mortality.23 24 This is a key difference between major incidents and routine clinical practice, where a form of triage occurs for every patient in the ED (using systems such as the Manchester Triage System), but the key feature of these tools is to correctly identify those in need of urgent treatment (at the expense of overtriage).

Current guidance for major incident triage simply states that rates of undertriage and overtriage should be kept as low as possible.25 By contrast, for the triage of individual patients to major trauma centres, a threshold of 35% overtriage and 5% undertriage is recommended.25 Here, in addition to an assessment of physiological instability, the field triage process includes an anatomical and mechanistic assessment to aid in the decision-making. It is a more time-consuming process and is inappropriate for the purposes of primary major incident triage. While the rate of undertriage demonstrated by the MPTT is the lowest of all existing triage tools, it does come at the expense of increased overtriage. Although the highest of all triage tools (67.1%), the MPTT’s overtriage rate is comparable to that encountered overall following the London 7/7 bombings (64%).23 However, while this level of overtriage was tolerated following this incident, we acknowledge that this may not be transferable to all major incidents, especially in rural areas with limited surrounding healthcare facilities and in those settings with a less developed emergency medical service response.26

The MPTT showed the highest sensitivity 57.6% (95% CI 56.9% to 58.2%) at predicting the need for life-saving intervention with an absolute increase of 44.7% over the existing Triage Sieve 12.9% (95% CI 12.5% to 13.4%). Throughout the subgroup analysis, the performance of the MPTT was superior to all existing triage tools in terms of minimising undertriage. A reduction in MPTT sensitivity is observed when compared with the derivation study (42.4% vs 35.1%). This is likely to be multifactorial, including the differing population age (median 62 years vs 24 years), the predominating mechanism of injury (falls <2 m vs explosive) and the proportion of P1 patients (19.5% vs 47.6%).

In summary, we present a civilian validation of the MPTT, the first example of an evidence-based physiological triage tool for use in the major incident setting. Our findings demonstrate that the MPTT outperforms existing triage tools with respect to rates of undertriage, while maintaining an acceptable level of overtriage. We suggest that the MPTT should be considered as an alternative to existing systems for the purposes of major incident primary triage. Ideally, the MPTT should be tested in the major incident environment, but in the absence of this, simulation or computer modelling may represent an alternative.

Supplementary file 3



  • Twitter @jamievassallo

  • Contributors JV, JES, FL and LAW conceived the study. JV conducted the analysis, supervised by JES. OB provided statistical advice and assisted with data analysis. JV drafted the manuscript and all authors contributed substantially to its revision. JV takes responsibility for the paper as a whole.

  • Competing interests None declared.

  • Ethics approval University of Cape Town. Human Research Ethics Committee. Reference 285/2013

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.