Criteria abstracted from The Users' Guide to Medical Literature, from the Health Information Research Unit and Clinical Epidemiology and Biostatistics, McMaster University

Highlighted lines and questions below provide links to the pertinent description of criteria in The EBM User's Guide, now available at the Canadian Centres for Health Evidence

Article Reviewed:

Please visit the new Evidence Based Journal Club Reviews

Predicting outcome in pediatric submersion victims

Graf WD, Cummings P, Quan L, Brutocao D.

Ann Emerg Med. 1995 Sep;26(3):312-9 [abstract]

Reviewed by Aaron Godshall MD, Children's Hospital Los Angeles

Review posted July 12, 2005

I. What is being studied?

Study objective:

Attempt to formulate a positive or negative prediction model to predict neurologic outcomes of pediatric submersion victims in non-icy water.

Study design

Retrospective cohort chart review.

The patients included:

All patients (n=194, age 5mo to 18 yr, mean 2.6 yr) who were identified by the hospital registry as presenting with a submersion event were included and their charts were reviewed for demographics (age, sex), episode related (month, year, high temperature of the day, urban or rural location, estimated submersion time, transfer from another hospital), clinical findings in the ED (mental status, pulse +/-, blood pressure, respiration, pupillary response, body temp, +/- seizures), laboratory and radiographic studies (pH, PO2, initial blood glucose, CXR), and pre-hospital admission treatments (epinephrine, glucose, intubation, CPR, duration of CPR).

The patients excluded:

Submersion victims who died in the Emergency Room.

Outcome measures:

Outcome was defined by the best neurological recovery before hospital discharge (via Glasgow Outcome Score). A favorable outcome was survival with any level of neurologic impairment associated with consciousness. An unfavorable outcome consisted of death or a vegetative state, defined as wakefulness without awareness of self or environment.

II. Are the results of the study valid?

Note: These questions follow from Randolph AG et al. Understanding articles describing clinical prediction tools. Crit Care Med 1998;26:1603-1612. [abstract]

1. Was a representative group of patients completely followed up? Was follow-up sufficiently long and complete?

Yes, the study format was retrospective in design and thus complete follow-up was obtained. All patients were reviewed until death or hospital discharge. This group is appropriate for designing a model to predict outcome to discharge, however predictors may not correlate with long-term, post-discharge outcomes.

2. Were all potential predictors included?

The authors search many prior studies and included each of the factors from those protocols to determine the best to include in their present study. There are omissions, for instance, a formal Glasgow Coma Score on admission. As typical with retrospective study design many variables may have been missing.

3. Did the investigators test the independent contribution of each predictor variable?

First the authors determined that comatose state in the ER was almost exclusively associated with an unfavorable outcome. Therefore, their subsequent analysis was looking at the independent contribution of the variables in comatose patients only (n=72), rather than of the 194. Table 2 (pg. 315) analyzed the individual (univariate) relative risks for each of the candidate predictive variables. It was from this table that the authors determined which variables to include in their final predictive model using logistic regression.

4. Were outcome variables clearly and objectively defined?

Yes, the authors used a modified Glasgow outcome scale with 5 outcomes. Each outcome is defined. There are some subjective variables such as how much awareness the patient possesses. Otherwise, ability is judged on ability to perform activities of daily living. An unfavorable outcome was defined as death or persistent vegetative state. A favorable outcome was defined as retaining any level of consciousness regardless of level of neurologic impairment.

III. What are the results?

1. What is(are) the prediction tool(s)?

The prediction tool is a mathematical formula (1/(1+e^{6.38-4.23(pupillary reflex) - 0.01(Blood Glucose) - 2.3 (Gender)} , where Absent pupillary response = 1, Present pupillary response = 0, Male gender = 1 and Female gender = 0 ) that the authors claim has 100% specificity of death or a persistent vegetative state. Patients who presented to the ED in a non-comatose state had a nearly universal favorable outcome and therefore, the above formula was applied only to patients presenting in coma.

Absence of a papillary reflex was the most strongly associated predictor value. Scores calculated with the above formula will range from 0 to 1. A higher score denotes a more unfavorable outcome. The authors determined that a score of 0.9 or above correlated with an unfavorable outcome as defined by the outcome criteria. This cutoff gave a sensitivity of 81% and specificity of 100%.

Strongest associations with death/vegetative state	Unfavorable outcomes
Absence of pupillary light reflex High initial blood glucose Absence of seizures History of CPR Duration of CPR > 25 minutes	Male sex Rural location Longer submersion time Lack of a pulse, recordable BP, or respirations in the ED PO2 < 60 mm Hg Lower blood pH Abnormal CXR History of epinephrine use

2. How well does the model categorize patients into different levels of risk?

Specificity is very high with the predictive model. Sensitivity seems relatively low (65%), but if one is going to error when reporting likelihood of an individual recovering, one would like to error on the side of having a favorable outcome. In other words, they chose values to be sure not to misclassify an individual (as having a bad outcome) who might actually have a favorable outcome. Therefore the authors decided to set their model with 100% specificity and avoid falsely classifying patients to having a unfavorable outcome. An ROC was not constructed because they did not desire a balance between sensitivity and specificity. Instead they chose to maximize specificity and see how much sensitivity they could obtain.

3. How confident are you in the estimates of risk?

The 95% confidence intervals in the univariate analysis are very large. The one reported CI is for patients who present in coma and equals a relative risk for an unfavorable outcome of 88.11 and a 95% CI = 12.45 to 623.78. Many of the predictor variables also have confidence intervals that span 1.0, including two of the three variables included in the final prediction tool. Pupillary response was the strongest discriminator of all variables tested and was used as the core of the formulation of the prediction formula. In order to allow for a range of sensitivities and specificities the continuous variable of blood glucose was added. Although univariate analysis revealed a CI which crosses 1, this variable had the highest relative risk toward unfavorable outcome. The choice of male sex being chosen as the third variable is unclear. One could assume that it is because of ease in determining the variable assignment value.

IV. Will the results help me in caring for my patients?

1. Does the tool maintain its prediction power in a new sample of patients?

There is no validation in this report. There were not enough patients to have a population of patients to derive the tool and then a subsequent subset of patients to test it within the same publication.

2. Are your patients similar to those patients used in deriving and validating the tool(s)?

Yes, our hospital and the center in the study are both are pediatric referral centers who receive both primary as well as referred cases from outlying facilities.

3. Does this tool improve your clinical decisions?

Submersion injuries have often been shown to be a prognostic challenge. Most studies have shown that early successful resuscitation leads to the best overall outcomes. The absence of cognitive functioning after 72 hours has also been shown to correlate with a very poor outcome (1). If validated, this tool could be used to assist physicians in making decisions regarding allocation of medical resources more rapidly and with more confidence. It may also help in guiding the family through the difficult decisions that may lay ahead. It does not specify between children who may have very serious neurologic impairment requiring extensive support from medical services.

Submersion injuries associated with hypothermia has been correlated with more favorable outcomes than cases absent of hypothermia on presentation (2). Body temperature was not included in the regression analysis of this study. Therefore, it is unclear how body temperature on presentation would affect the prognostic prediction offered by the authors.

References

1. Sarnaik, A.P., et al., Intracranial pressure and cerebral perfusion pressure in near-drowning. Crit Care Med, 1985. 13(4): p. 224-7. [abstract]
2. Biggart, M.J. and D.J. Bohn, Effect of hypothermia and cardiac arrest on outcome of near-drowning accidents in children. J Pediatr, 1990. 117(2 Pt 1): p. 179-83. [abstract]

 

Comments

Back to the EB Journal Club Index