Indian Pediatrics 2003; 40:204-210 

Triage Score for Severity of Illness

N. Kumar, N. Thomas, D. Singhal, J.M. Puliyel, V. Sreenivas*

Objective: To evolve a triage scoring system for severity of illness based on clinical variables related to systemic inflammatory response syndrome (SIRS). Design: Prospective study in a tertiary-care hospital. Methods: Consecutive pediatric patients admitted to the ward or pediatric intensive care unit (PICU) were studied. The respiratory rate, heart rate, capillary refill time, oxygen saturation (SpO2), systolic blood pressure and temperature were noted, Sensorium level was assessed on AVPU score. Variables were based on SIRS criteria and criteria mentioned in Advanced Pediatric Life Support (APLS). Each study variable was scored as 0 or 1 (normal or abnormal) and total score for each child obtained. The survival at discharge was correlated with the study variables and the total score. Another score based on the magnitudes of the coefficients in multiple logistic regression analysis was computed and the correlation between this score and mortality was also studied. ROC curve analysis was performed to see the overall predictive ability of the score as well as a cut off at which maximum discrimination occurred. Results: Of 1099 children studied, 44 died. Of the seven variables, only five variables were abnormal in the study subjects. Except heart rate and respiratory rate, all other variables and age showed significant association with survival status (P <0.01). The mortality increased with increase in the number of abnormal variables: 0.4%, 2.2%, 6.1%, 15.3%,19.4% and 29.4% for scores of 0,1,2,3,4 and 5 respectively and the linear trend was significant (P <0.01). Mortality also increased with a decrease in age (P <0.01). Children with a score of 2 or more (2 or more abnormal clinical variables) had significantly higher mortality as compared to those with no abnormal clinical variables (score = 0). Based on the regression coefficients, the maximum possible score was 9.8. Regression based score was found to predict survival status well. The area under the ROC curve was 0.887, indicating that overall 88.7% of the subjects could be predicted correctly. Maximum discrimination was observed at a score of 2.5 (sensitivity 84.1%, specificity 82.2%). Conclusion: For triage scoring, any child with 2 or more abnormal clinical variables should be taken as serious that might lead to death. With a more detailed scoring, score of 2.5 can be taken as cut-off to select children who possibly need admission and closer observation.

Key words: Intensive care, Systemic inflammatory response syndrome, Triage score.

Mortality in an intensive care unit (ICU) depends on the severity of illness(1). A good scoring system for identifying the severity of illness can help to prioritize care. More sick children need to be admitted and those at the end of the spectrum would benefit from intensive care manage-ment. Triage is sorting out of patients -the main objective of which is early patient assessment to obviate harmful delay in the management(2). The existing scoring systems have been developed to predict mortality in ICU admissions(3). However, the existing scoring systems depend on both physical and laboratory variables and are inappropriate for primary triage. We hypothesized that a scoring system using physical criteria can be developed to identify severity of illness. This can be used to triage patients for management and predict outcome. Our triage score is related directly or indirectly to the abnormal physical variables of systemic inflammatory response syndrome (SIRS) and its continuum. The SIRS is the host response to various infective and non-infective insults which was proposed by the Consensus Conference of American College of Chest Physicians and the Society of Critical Care Medicine in 1992(4). In the continuum, we also used the physical signs, utilized in the Advanced Pediatric Life Support(5).

Subjects and Methods

This prospective study was done from August 1998 to February 1999 at a tertiary care hospital, having a 50-bed pediatric unit (including 6 pediatric ICU beds). Patients admitted consecutively were included for the study. Patients who left against medical advice and those transferred to another hospital were excluded. Seven clinical variables i.e., heart rate, respiratory rate, systolic blood pressure, oxygen saturation (SpO2), capillary refill time (CFT), tempera-ture and level of consciousness were noted on a pre-designed proforma by the doctor on duty at the time of admission.

Blood Pressure was measured by oscillo-metry (Graseby oscillomats 900; Graseby, Watford). SpO2 was measured by pulse Oximetry (Simed S-100C, Bothell, WA 98011). Axillary temperature was measured using a mercury thermometer. Abnormal values for heart rate, respiratory rate, tempera-ture and blood pressure were according to standard SIRS criteria(6). Consciousness was noted using the AVPU score. Except alert (A) of AVPU, all other states of consciousness were taken as abnormal. AVPU was taken for rapid assessment of sensorium because it requires only 4 observations for its assess-ment. The abnormal value for SpO2, CFT and AVPU were as per Advanced Pediatric Life Support(5) (Table 1). The hospital discharge status (death/survival) was the primary outcome variable.

Table I__Scoring of Abnormal Clinical Variables*

Variable	Abnormal range
Temperature	>38ºC
	<36ºC
Heart rate	Infant >160 per minute Child >150 per minute
Respiratory rate	Infant >60 per minute Child >50 per minute
Systolic blood pressure	Infant <65 mm Hg Child <75 mm Hg
SpO2	<90%
Capillary refill time	≥3 seconds
A Alert	Anyone except A
V  Responds to voice
P  Responds to pain
U  Unresponsive

* Based on SIRS and APLS (references 6,5)

Statistical analysis

The predictors of outcome were studied in 2 ways – association of outcome with the number of abnormal variables, and associa-tion of outcome with a score derived from the magnitude of association of each variable with the outcome. Odds ratios with 95% confi-dence intervals were calculated for each variable. A trend Chi-square test was used when more than 2 ordered groupings were present.

For assessing the magnitude of association, a multiple logistic regression analysis of survival status was carried out with the study variable and age as predictors. The regression coefficients associated with each variable were taken as the respective weight for that variable. If a child had 3 abnormal variables, the total weights of the 3 variables and the weight for the respective age category was taken as the score for that child. A receiver operating characteristic (ROC) curve analysis was also carried out to see the predictive ability of the score as well as a specific score value, which could be taken as a cut-off. All analysis were carried out using BMDP Statistical Software, Release 7.

Results

Of 1133 patients admitted during the study period, 34 were excluded (left against medical advice 10, transferred to another hospital 24). Thus, 1099 children were included for the analysis, of which 109 were neonates. Forty four children died in the hospital. Of the 7 clinical variables considered, only 5 were abnormal in the study subjects.

The distribution of children with each clinical variable (normal/abnormal) along with the discharge status (survived/dead) is shown in Table II. Except heart rate and respiratory rate, all variables were signifi-cantly associated with mortality (P <0.01). It was observed that, the mortality increased as age decreased with the odds ratios being 1.7, 2.8 and 7.7 in the age groups 12-60 months, 1 to 12 months and less than 1 month respectively, as compared to those aged more than or equal to 60 months. The linear trend in relation to age was also significant (P < 0.01). It was observed that mortality increased with increase in the number of abnormal variables, the odds ratios being 5.2, 15.4, 42.6, 57.0 and 98.3 with one, two, three, four and five abnormal clinical variables respectively, at admission. The linear trend with increasing scores was significant (P <0.01).

Table II__Association of Study Variables with Mortality*

Variable		Survived No.     %		Died No.   %		Odds radio	P value
Heart rate	Normal	781	96.7	27	3.3	1.8
	Abnormal	274	94.2	17	5.8	(0.9 – 3.4)	0.10
Respiratory rate	Normal	865	96.4	32	3.6	1.7
	Abnormal	190	94.1	12	5.9	(0.9 – 3.4)	0.18
Blood pressure	Normal	1045	96.5	38	3.5	16.5
	Abnormal	10	62.5	6	37.5	(5.7 – 47.8)	<0.01
Temperature	Normal	826	97.1	25	2.9	2.74
	Abnormal	229	92.3	19	7.7	(1.5 – 5.1)	<0.01
SpO2	Normal	913	98.1	18	1.9	9.3
	Abnormal	142	84.5	26	15.5	(5.0 – 17.4)	<0.01
Capillary refill time	Normal	989	97.2	29	2.8	7.8
	Abnormal	66	81.5	15	18.5	(4.0 – 15.2)	<0.01
AVPU	Normal	951	97.9	20	2.1	11.0
	Abnormal	104	81.2	24	18.8	(5.9 – 20.6)	<0.01

* Based on univariate analysis

Compared to those with no abnormal variables (score = 0), it was noted that those with 2 or more abnormal variables had significantly higher mortality.

A multiple logistic regression on the clinical variables and age was carried out to determine the magnitude of associations of each with mortality (Table III). The total of the regression coefficients (b) for the clinical variables and age was 9.8, which was the maximum possible score for any child. However, in the study subjects, the maximum observed score was 8.0.

Table III__Weight (Regression Coefficient) for Each Variable*

Variable	Weight (b)
Heart rate	0.2
Respiratory rate	0.4
Blood pressure (systolic)	1.2
Temperature	1.2
SpO2	1.4
Capillary filling time	1.2
AVPU	2.0
Age (months)
≥60	0.0
≥12 to <60	0.3
≥1 to <12	1.0
<1	2.2

* Based on multiple logistic regression analysis.

As the total score increased, the mortality increased progressively. The proportion of deaths was 0.4% with score £1 and 75% with a score ³7.1 (Table IV). A child with a score of more than 7.0 had an odds ratio of 724 of dying in the hospital as compared to a child with a score of less than 1.0. Using different cut off points of the score developed, a ROC curve was drawn (Fig. 1). The area under the curve was 88.7% indicating good predictive ability of the score. Maximum discrimination was observed for a score of 2.5 where sensitivity was 84.1% and specificity 82.2%.

Table IV__Outcome at Different Scores

Score	Died No. %		Survived No. %		Odds radio	95% CI
0.0 – 1.0	2	0.4	483	99.6	1.0	–
1.1 – 2.0	2	0.9	220	99.1	2.2	0.3–15.7
2.1 – 3.0	7	3.0	229	97.0	7.4	1.5–35.8
3.1 – 4.0	9	12.9	61	87.1	35.6	7.5–168.7
4.1 – 5.0	11	21.6	40	78.4	66.4	14.2–310.0
5.1 – 6.0	6	28.6	15	71.4	96.6	18.0–518.6
6.1 – 7.0	4	40.0	6	60.0	161.0	24.6–1053.7
³ 7.1	3	75.0	1	25.0	724.5	50.9–10307.3

* Derived from a multiple logistic regression analysis.

Fig. 1. Area under the ROC Curve (Az) for predictive ability of the score is 88.7%

Discussion

The early identification of severity of illness is important for prioritizing treatment to reduce mortality and allow proper utiliza-tion of limited resources in the developing world(6). Various scoring systems have been proposed to assess the severity of illness which predict mortality e.g., PRISM(7). Most of the scoring systems are for ICU patients. Extension of this scoring system is theo-retically possible to less sick children, so as to help in triage. However, these scoring systems rely on a large number of physical and laboratory variables and require prolonged observation. This makes it unsuit-able for practice in developing countries.

WHO developed guidelines for emer-gency triage, assessment and treatment for sick children presenting to hospitals in the developing world. It prioritized the treatment of sick children depending upon the emergency signs related to airway, breathing, circulation, coma, convulsion, confusion and dehydration to decrease the mortality. The limitation of emergency triage, assessment and treatment is that it requires reorganizing of the existing health care system and special training of both staff and doctor(8). In view of the drawbacks of the existing system we developed a score based on physical criteria alone. The SIRS is the host response to presence of an insult regardless of the presence of infection. SIRS is diagnosed when a patient has two or more of the following criteria (i) temperature (ii) heart rate (iii) respiratory rate, and (iv) white blood cell count, as abnormal(9). The children with SIRS may go on to develop multiple organ dysfunction syndrome. We thus took physical variables of SIRS and its continuum and excluded the biochemical and laboratory parameters and tested if the score thus developed could predict mortality. We hypothesized that prediction of mortality based entirely on physical criteria could perhaps be helpful to triage patients.

On univariate analysis, heart rate and respiratory rate were not significantly associated with survival status. However the statistical power of this study to detect a significant difference in mortality between the normal and abnormal values of heart rate and respiratory rate was 40% and 28% respectively. Since both heart rate and respiratory rate may show significant association with mortality if larger samples are studied, we have retained these parameters in our scoring system.

In our study we have found on ROC curve analysis that the scores based on regression could predict 89% subjects correctly. Further, a score of 2.5 showed maximum discrimination with 84.1% sensi-tivity and 82.2% specificity. The results of this study are similar to those reported previously(10). Previous studies have shown that mortality was significantly higher in the SIRS group compared to non-SRS group. It was also observed that mortality was three times higher in the high risk SIRS (which comprises only heart rate, platelet count and C-reactive protein)(4). However, both SIRS and high risk SIRS utilize laboratory parameters and are probably less suitable for triage. We found that two or more abnormal physical variables out of seven were significantly associated with mortality and may be used to assess the severity of illness.

Our study has certain limitations. We only examined patients admitted to the hospital. It is possible that less sick looking children may have been inappropriately sent home and a few of them may have died without coming to our notice. If all children coming into contact with the hospital – both in the OPD and the casualty were studied, the results could be generalized. Since, mortality increased with increase in score among those admitted, it can be assumed that mortality would be less among those sent home because they were apparently well.

Another possible source for bias is the recording of clinical parameters like heart rate, respiratory rate and blood pressure etc. as they have been noted by different pediatri-cians on duty. We have not specifically looked for interpersonal variability among the attending pediatricians.

Contributors: NK was involved in concept, design, manuscript writing and will act as a guarantor of the study. JMP and VS analyzed and interpreted the data and drafted the manuscript. DS and NT were involved in data collection

Funding: None.

Competing interests: None

 References

1. Fery-Lemonnier E, Landais P, Loirat P, Kleinknecht D, Briivet F. Evaluation of severity scoring systems in ICUs-translation, conversion and definition ambiguities as a source of inter-observer variability in Apache II, SAPS and OSF. Intensive Care Med 1995; 21: 356-360.

2. Robertson MA, Molyneux EM. Triage in the developing world -can it be done? Arch Dis Child 2001; 85: 208-213.

3. Shann E, Pearson G, Slater A, Wilkinson K. Pediatric index of mortality (PIM): A mortality prediction model for children in intensive care. Intensive Care Med 1997; 23: 201-207.

5. The structured approach to the seriously ill child. In: Advanced Pediatric Life Support. The practical approach. BMJ Books, 3rd edn. Eds. Mackway- Jones K, Molyneux E, Phillips B, Weeteska S. Bangalore, Panther Publishers, 2001; pp 71-78.

6. Jafri HS, McCraken GH. Sepsis and septic shock: A review for clinician. Pediatric Infect Dis J 1992; 11: 739-748.

7. Pollac MM, Ruttiman UE, Getson PR. The pediatric risk of mortality (PRISM) score. Crit Care Med 1988; 16: 1110-1116.

8. Robertson MA, Molyneux EM. Description of cause of serious illness and outcome in-patients identified using ETAT guidelines in urban Malawi. Arch Dis Child 2001; 85: 214-217.

9. Asayama K, Aikawa N. Evaluation of systemic inflammatory response syndrome criteria as a predictor of mortality in emergency patients transported by ambulance. Keio J Med 1998; 47: 19-27.

10. Singhal D, Kumar N, Puliyel JM, Singh SK, Srinivas V. Prediction of mortality by applica-tion of PRISM score in intensive care unit. Indian Pediatr 2001; 38: 714-719.