1.gif (1892 bytes)

Original Articles

Indian Pediatrics 2001; 38: 714-719  

Prediction of mortality by application of prism score in intensive care unit


D. Singhal, N. Kumar, J.M. Puliyel, S.K. Singh, V. Srinivas*

From the St. Stephen’s Hospital, Tis Hazari, Delhi 110 054, India and *Division of Non Communicable Diseases, Indian Council of Medical Research, Ansari Nagar, New Delhi 110 029, India.
Correspondence to: Dr. Nirmal Kumar, 4 Rajpur Road, Quarter No. B-2, Tis Hazari, Delhi 110 054, India.
E-mail:nsk9-2000@yahoo.com

Manuscript received: July 12, 2000, Initial review completed: August 24, 2000,
Revision accepted: January 15, 2001.

Objective: Prediction of mortality by application of Pediatric Risk of Mortality (PRISM) score in Pediatric Intensive Care Unit (PICU) patients under Indian circumstances. Design: Prospective study. Setting: PICU of a tertiary care multi-speciality hospital. Methods: 100 sick pediatric patients admitted consecutively in PICU were taken for this stduy. PRISM score was calculated. Hospital outcome was recorded as (died/survived). The predicted death was calculated by the formula: Results: Of 100 patients, 18 died and 82 survived. By PRISM score 49 children had the score of 1-9. The expected death in this group was 10.3% (n = 5.03) and the observed death was 8.2% (n = 4). Among 45 children with the score of 10-19, the expected mortality was 21.2% (n = 9.6) and observed was 24.4% (n = 11). There were 3 pateints with the score of 20-29, the expected mortality in this group was 39.3% (n = 1.18) and observed mortality 33.3% (n = 1). There were 3 patients with score ³30, observed death 66.3% (n = 2) and expected mortality was 74.7% (n = 2.24). There was no significant difference between expected and observed mortality in any group. (p >0.5). ROC analysis showed area under the curve of 72%. Conclusion: PRISM score has good predictive value in assessing the probability of mortality in relation to children admitted to a PICU under Indian circumstances.

Key words: Intensive care, Mortality, PRISM score.

THE outcome of intensive care in India has not been widely reported(1). The need for sophisticated equipment and aggressive treatment of critically ill infants and children is well recognized. Evaluation of the results of such therapy requires the use of accurate and easily applied methods for describing the patients as well as their outcome. Various patient classification systems developed for use in the adult population have served as guides for prognosis, cost analysis, staffing and interinstitution comparisons(2). Scoring systems are aimed at quantifying case mix and using the resulting score to predict the outcome(3). The Pediatric Risk of Mortality (PRISM) Score has been devised to predict which helps the physician to predict outcome and risk of mortality. It provides medical staff with epidemiological criteria and may help in decision making for Pediatric Intensive Care Unit (PICU) admissions and correct identification of patients who might benefit from such care(5). The purpose of this study was to see if PRISM score can predict mortality in children admitted to a PICU under Indian circumstances.

Subjects and Methods

This study was carried out at a multi-speciality, tertiary care hospital having 80 bed Pediatric department and a 4-bed PICU. One hundred consecutive pediatric cases admitted directly to the PICU were studied prospect-ively. Cases aged between 1 month and 12 years admitted during the 6 months from August 1988 to February 1999 were enrolled into the study. All the subjects had their PRISM score evaluated immediately after arrival to the ICU. The PRISM score evalua-tion was done as per recommendation of Pollack et al.(4). Those children who had congenital malformation and babies less than 1 month of age were excluded from the study.

Non invasive blood pressure was recorded using blood pressure monitor and oxygen saturation measured with pulse oximeter at admission. The FiO2 required to maintain oxygen saturation above 90% was noted with oxygen monitor. Arterialized capillary heel prick blood was used for determining PaO2, PaCO2 and bicarbonate. Standard laboratory techniques were utilized to measure blood levels of total bilirubin, potassium, calcium and glucose. Prothrombin time and partial thromboplastin time were measured. The clinical assessment of heart rate, respiratory rate and pupillary reaction was made by a pediatric resident doctor at the time of admission. The Glasgow Coma score and modification of Glasgow Coma score was utilized(6). The children were followed up during hospital stay and the outcome measures were recorded as died or survived at the end of the hospital stay.

Statistical Methods

The association between the study variables, namely, age, sex, PRISM Score and diagnosed cause of illness with the ICU mortality was tested using the contingency/trend chi-square, as appropriate. Further, the association in terms of Odds Ratio was assessed by logistic regression analysis. The appropriateness or the aptness of the model is assessed by Hosmer-Lemeshow summary chi-square test and also by the Receiver Operating Characteristics (ROC) Curve analysis. The statistical software BMDP (Biomedical Data Program dynamics; University of California, Berkley) was used.

Results

A total of 100 admitted ICU children were studied, in which 18 died and 82 survived. All the admitted PICU children received appro-priate management which was required. Fifteen children required ventilator care, out of which 12 died. The maximum duration of ventilator care was 7 days. In most of the cases it was required for <24 hours. Male children formed the majority (78%). The average age (± SE) of the subjects was 25.5 ± 3.2 months (range 1 to 32 months). Respiratory conditions (40%), CNS morbidity (27%), sepsis (9%) followed by other conditions (24%) were the causes of illness diagnosed on admission. They were grouped according to the primary system involved. The others include cardiovascular disorders (n = 6), infection (n = 6), gastro-intestinal and liver disorders, (hepatitis, fulmi-nant hepatic failure (n = 4), gastroenteritis with severe dehydration (n = 2), surgical patients (n = 2), poisoning (n = 1), leukemia (n = 1), drowning (n = 1) and diabetes (n = 1). The average PRISM score of the subjects was 10.9 ± 0.66 (range 1 to 42). The duration of stay between admission and discharge (dead/alive) varied from minimum of 4 hours to a maximum of 624 hours with a mean of 173.78 hours.

Table I summarizes the association of the study variables with the ICU mortality. It can be seen that except PRISM score, the remaining variables showed no significant association with the mortality (p >0.11). PRISM score showed a significant association with the mortality. The proportion of deaths which was only 8.2% among children with the scores 1-9 showed a gradual increase with higher scores, reaching 66.7% among the children with a score of > 30.

Table I Association of Study Variables with ICU Mortality

Variable
Number(%)
Died(%)
Alive
P value
Sex
Male
72
16.7
83.3
Female
28
21.4
78.6
0.79
Age (mo)
< 12
53
17.0
83.0
12-36
24
20.8
79.2
> 36
23
17.4
82.6
0.90
Cause of illness
Respiratory
40
10.0
90.0
CNS conditions
27
18.5
81.5
Sepsis
9
44.4
55.6
Other conditions
24
20.8
79.2
0.11
PRISM score
1-9
49
8.2
91.8
10-19
45
24.4
75.6
20-29
3
33.3
66.7
>=30
3
66.7
33.3
0.01

Since the association between mortality and the PRISM score only turned out to be significant in the initial analysis, a logistic regression analysis was done on the discharge status (died/alive), taking PRISM score as a predictor for mortality. The analysis yielded a logit r = (0.1138* PRISM) – 2.902. The probability of death was calculated by formula: Probability of death = er/(1 + er), where r = (0.1138 * PRISM) – 2.902 (In our study). Here –2.902 is constant. This implies that the probability of death is 0.1463, for a child with a score of 10. The Odds Ratio corresponding to the model is 1.12 with the 95% confidence interval (1.03 - 1.22). In other words, for an increase of 1 in the PRISM score, a child’s odds of death increases by 12%. The predicted probability of ICU mortality from our data is shown in Table II. A child with a PRISM score of 5 had a 9% chance of dying in ICU and a child with a score of 35 had 75% probability of dying in the ICU. A score of above 26 yielded 50% probability of death in ICU.

Table II - Prediction of Probability of Death in ICU According to PRISM Score

Score
Probability of death (%)
5
9
10
15
15
23
20
35
25
49
30
63
35
75

The results on goodness of the prediction model as seen by the Hosmer-Lemeshow goodness of fit chi-square are presented in Table III. We can see that the discrepancies between the observed and the expected across the 4 score strata are not significant (p = 0.88). Taking the probability of death at different levels the plot of false positives vs true positives i.e., ROC curve analysis (Fig. 1) shows that 72% of the subjects could be predicted correctly. Thus, the prediction of ICU mortality based on PRISM Scores is valid and moderately reliable.

Table III - Goodness of the Predictive Model

PRISMScore
Total Number
Deaths
Alive
   
Observed
Expected
Observed
Expected
1-9
49
4(8.2)
5.03(10.3)
45
43.97
10-19
45
11(24.4)
9.60(21.2)
34
35.40
20-29
3
1(33.3)
1.18(39.3)
2
1.82
>=30
3 6
2(66.7)
2.24(74.7)
1
0.7

Hosmer-Lemeshow Chi-square = 0.66 3 d.f. p = 0.88.
Figure in parentheses indicate percentages.

Discussion

The prediction of patient outcome is always associated with uncertainty but is important because patients do not respond to the same insult in the same way. Their clinica manifestations and their individual potentials to recover differ. The clinical judgement of the severity of a disease process is not uniform and is related to the experience and clinical ability of physician. Prediction of patient outcome is important for the patients and family and is relevant for policy formulation and resource allocation; the optimum usage of ICU beds will obviously allow maximum utilization of limited resources(7).

The first scoring system in pediatric intensive care unit was the Therapeutic Intervention Scoring System (TISS). The basis of this system is that therapeutic intensity defines severity of illness. A score of one to four points is awarded to each of 70 nursing and medical procedure. However, the system does not take into account the variability in clinical practice which may occur between ICUs and between different countries. It is however a useful system for assessing expenditure and has been extensively used for this purpose(8).

Fig. 1. ROC cure analysis indicates that the area under the curve (Az) is 72% which is moderately good

The first physiology based scoring system to assess severity of acute illness in the total population of infants and children admitted to the pediatric ICU was Physiology Stability Index (PSI). It was developed to aid evaluation of severity of illness for acute disease. It has a total of 34 routinely or frequently measured variables from seven systems(9). The PRISM Score was developed from the Physiologic Stability Index (PSI), a pediatric severity of illness measure used to predict mortality(10).

The PRISM Score was developed and validated in intensive care setting by Pollack et al.(4). The score describes the severity of illness according to physiological derangement detected on clinical examination and standard laboratory tests. It has been observed that accuracy or reliability of PRISM score is unlikely to be influenced by measurement frequency to the overall variability(5). Our observation that increase in PRISM

Score is associated with an increase in the mortality was similar to previous workers(4,11).

Tilford et al.(10) compared the difference in PICU mortality risk over a period between 8 to 12 years. They observed 15% reduction (p <0.001) in mortality risk based on predicted mortality probability from the original PRISM score. A number of medical services and technology innovations may be responsible for improvement in mortality risk. In our study the both observed and predicted death were low at lower score and it was not similar to other studies(1,7,12).

In a study from South Africa, there was discrepancy between observed and the pre-dicted mortality rates. There was under prediction of mortality at lower PRISM scores and over prediction at higher scores. The authors suggested that this might be related to their “lead time bias”. Late presentation to the hospital and delay in admission to the PICU might be responsible. The PRISM, score at admission to the PICU may have been masked by their initial treatment causing a falsely low PRISM score and under estimation of mortality(7).

In one study in India among adult patients, deaths at lower score were more than that predicted by Logistic Organ Dysfunction System (LODS) score. It was speculated by the authors that this was probably due to early shift of patients from ICU to the ward, a practice in their hospital because of the pressure on ICU beds(1).

Manten Radovan et al.(13) observed that mortality was significantly higher than predicted among lower risk patients by application of PRISM score. Tracheal intuba-tion, central catheters, pneumonia, sepsis, and non-surgical status were associated with poor outcome for low risk group and it was probably because of treatment associated with central venous cannulation and tracheal intubation at lower risk which caused this.

Proulx et al.(14) observed multiple organ failure during PICU stage, age <12 months and PRISM score on the day of admission to be independent risk factors of death. But in our study, age did not emerge as a risk factor for death, probably because of less numbers of deaths in the age group of children less than 12 months in comparison to other age groups.

Since mortality rises with increase in PRISM score at the time of admission, PRISM score can be taken as an indicator of the initial severity of illness. As intensive care facilities are limited in India, clinicians need a more objective criteria as to which patient needs intensive care. PRISM score is one such indicator for PICU admission. Also, since PRISM score may reflect the severity of illness initially, it has a great relevance in clinical epidemiology. For example the PRISM score can be used as an inclusion/exclusion criteria, or as a stratification variable in clinical trials.

The probability of death in the ICU from our study can serve as a basis for comparison of the experience of other ICUs/hospitals, beside our own center over a period of time.

We studied 100 PICU admissions, of which 18 died before discharge. This sample is expected to provide a power of 65% (for a prevalence of PRISM score ³10 in alive cases of 45%) to detect an odds ratio of 4.25 or more in two sided test at 0.05 level of significance. Our study sample had only 6% of the subjects with a score greater than or equal to 30 and had only 18 events (deaths). Since the predictions for children with high scores are based on a few numbers, they should be viewed conti-nuously, keeping the numbers in mind. A larger study with more events is warranted to have better understanding of PRISM scores vis-a-vis ICU mortality.

Contributors: NK and NMP were responsible for concept, design and drafting of manuscript; they will act as guarantors for the paper. DS and VS conducted, analysis and interpreted the data. SKS helped in drafting the article.

Competing interests: None stated.
Funding: St. Stephen’s Hospital, Delhi-110 054.

Key Messages

  • PRISM score is a good predictor of mortality in PICU patients under Indian circumstances.

  • PRISM score can help to concentrate efforts on those who can benefit more in PICU.

  • PRISM can help to select sick children for PICU admission and optimal utilization of limited PICU resources.


 References

1. Sampath S. Fay MP, Pias P. Use of the logistic organ dysfunction system to study mortality in an Indian intensive care unit. Natnl Med J India 1999; 12: 258-261.

2. Ye TS, Pollack MM, Holbrook PR, Fields AI, Ruttiman U. Assessment of pediatric intensive care – application of the therapeutic intervention scorting system. Crit Care Med 1982; 10: 497-500.

3. Giunning K, Rowan K. Outcome data and scoring system, BMJ 1999; 319: 241-244.

4. Pollack MM, Ruttiman UE, Getson PR. The Pediatric Risk of Mortality (PRISM) Score. Crit Care Med 1988; 16: 1110-1116.

5. Pollack MM, Patel KM, Ruttimann U, Cuordon, T. Frequency of variable measure-ment in 16 pediatric intensive care units: Influence on accuracy and potential for bais in severity of illness assessment. Crit Care Med 1996; 24: 74-77.

6. Allen EM, Boyer R, Cherney WB, Brockmayer D, Tait VF. Head and spinal Cord Injury. In: Textbook of Pediatric Intensive Care, 3rd edn. Ed Rogers MC. Baltimore, Williams and Wilkins, 1996; pp 809-857.

7. Wells M, Riera Fanego JF, Luyt DK, Dance N, Lipman J. Poor discriminatory performance of Predictors Risk of Mortality (PRISM) score in a South African intensive care unit. Crit Care Med 1996; 24: 1507-1513.

8. Bion J. Severity scoring: Principles methods, and applications. Rec Adv Anesth Analg 1992; 17: 173-195.

9. Yeh TS, Pollak MM, Ruttimann UE, Holbrook PR, Fields A. Validation of a Physiologic Stability Index for Use in Critically III, infants and children. Pediatric Res 1984; 18: 445-451.

10. Tilford JM, Roberson PK, Lensing S, Fiser DH. Difference in Pediatric ICU mortality risk over time. Crit Care Med 1998; 26: 1737-1743.

11. Kanter RK, Edge WE, Caldwell CR, Nocera MA, Orr RA. Pediatric mortality probability estimated from PICU severity illness. Pediatrics 1997; 99: 59-63.

12. Earle M Jr., Martinez Natera O, Zaslavsky A, Quinones E, Carrillo H, Garcia Gonzalz E, et al. Outcome of Pediatric Intensive Care at Six Centers in Mexico and Ecuador. Crit Care Med 1997; 25: 1462-1467.

13. Manten Radovan I, Gutierrez Castrellon P, Zaldo Rodriguez R, Martinez Natera O. PRISM score evaluation to predict outcome in pediatric patients on admission at an emergency department. Arch Med Res 1996; 27: 553-558.

14. Proulx F, Gautheir M, Nadeau D, Lacroix J, Farrell CA. Timing and predictors of death in pediatric patients with multiple organ system failure. Crit Care Med 1994; 22: 1025-1031.

Home

Past Issue

About IP

About IAP

Feedback

Links

 Author Info.

  Subscription