Objective. Predictive efforts using individual factors or scoring systems do not adequately identify all intact survivors, and therefore all drowning victims are aggressively resuscitated in most emergency departments. More reliable outcome prediction is needed to guide early treatment decisions.
Methods. The charts of 274 near drowning patients admitted to Loma Linda University Children's Hospital were retrospectively reviewed. Patient outcome was categorized into good (near normal function), and poor (vegetative or dead) categories. Discriminant analysis was used to identify combinations of variables most able to predict outcome and a clinical classification system was constructed. The acute care hospital costs for each group were compared.
Results. Discriminant analysis classification achieved 95% accuracy, predicting death in 6 intact survivors. No combination of variables could accurately separate all intact survivors from the vegetative and dead groups. The clinical classification method achieved 93% overall accuracy, predicting death in 5 intact survivors. Of patients predicted to have a poor outcome, 5 (6.3%) survived intact. Children may experience an unpredictable, prolonged vegetative state followed by full recovery. Vegetative patients are the most expensive to care for (consuming 53% of total costs) while intact survivors are the least expensive. The majority of costs were spent on patients with poor outcome.
Conclusions. Individual outcome cannot be reliably predicted in the emergency department; therefore, aggressive resuscitation of near drowning victims should be performed. Decisions to subsequently withdraw life support should be made based on integration of likelihood of survival, high (but not absolute) certainty, and parental/societal issues. The vegetative patients are the most expensive to care for, while intact survivors are least expensive. Reduction of expenditures on patients likely to have vegetative or dead outcome would result in substantial savings, but loss of normal survivors.
Despite preventative measures, drowning is the third most common cause of childhood death in the United States. Numerous studies have attempted to categorize near drowning victims into prognostic groups using data available in the emergency department (ED). However, predictive efforts using these factors or various scoring systems still do not adequately identify all intact survivors, and anecdotal reports of spectacular recoveries are common. In most hospitals, all drowning victims are aggressively resuscitated to avoid loss of potentially intact survivors. To determine whether survivors can be reliably identified early in their treatment course, we reviewed the characteristics and outcome of pediatric near drowning victims admitted to Loma Linda University Medical Center.
The medical records of all children admitted to Loma Linda University Medical Center for treatment of near drowning between January 1985 and June 1994 were identified by computer search of hospital Diagnosis Related Group codes and 39 variables were retrospectively recorded. Historic variables recorded included age, sex, admission and discharge dates, race, ethnicity, weight, estimated submersion interval, whether drowning was witnessed, performance of bystander cardiopulmonary resuscitation (CPR), and paramedic arrival interval (defined as the interval from receipt of 911 call until arrival of an emergency response team). To facilitate data analysis, a numeric physical examination categorization system was devised (Fig1). Examination category was assigned based on the best level of function documented for 3 time points: at pool side, paramedic arrival, and ED arrival. Treatment variables recorded included the pool side, paramedic, and initial ED physical examination category (Fig 1); ED Glasgow Coma Score; duration of CPR; prehospital or ED tracheal intubation, vascular access, and epinephrine or sodium bicarbonate administration; and initial temperature, blood glucose, pH, PaCO2 and base excess. Outcome variables included time to first purposeful movement, outcome category (Fig2), discharge disposition, and length of stay in the intensive care unit, step-down, or basic pediatric units. Follow-up was by review of hospital discharge and subsequent clinic and hospital visits recorded in the hospital chart.
Patients with functional neurologic recovery were classified as good outcome. All others were classified as poor outcome. Good outcome was later subdivided into intact (if return to near normal function was predicted and occurred), and unpredictable good outcome (UGO) groups. UGO patients are defined as intact survivors who were predicted to be in the vegetative or dead groups. Poor outcome was categorized as vegetative if survival without functional cognitive recovery occurred, and as dead if death occurred. Subdivisions within each group are shown in Fig 2.
Discriminant Analysis: Multivariate
Stepwise multivariate discriminant analysis to minimize Wilk's Lambda, calculating prior probabilities based on group size was used to identify variable combinations which most accurately predicted good, vegetative, or dead patient outcome. Patients were excluded if the variable used was not documented. The most predictive variable was recorded and eliminated from further use in this portion of the analysis. The process was repeated until all variables with a probability of F <.05 were recorded in order of predictive ability. An identical procedure was used to identify the variables which most accurately categorized patients into good versus poor outcome. Final discriminate analysis was performed entering all variables together to find the combination achieving optimal classification. Optimal classification was defined as highest overall accuracy after minimizing errors predicting poor outcome in intact survivors.
Discriminant Analysis: Clinical Classification
Using variables identified as most predictive, a simplified clinical classification system was then constructed and applied to the patient database. Variables were selected based on predictive utility in discriminant analysis evaluation, ability to minimize UGO patients, and the ease and accuracy with which they could be obtained. Initial paramedic examination was not used in the clinical classification because it was frequently unavailable. UGO group characteristics were compared to the intact and poor outcome groups using Student'st test. Acute care hospital costs, defined as hospital costs incurred during the patient's initial hospitalization (excluding professional fees and referring ED costs) were calculated for each outcome group. The resulting categorizations and hospital costs are described.
Two hundred seventy-four patients were identified. Patient age ranged from 6 months to 15 years (mean 32 months, median 24 months) and 63% were males. The submersion was witnessed in only 33 cases. Submersion site data was collected in 126 patients; 101 events occurred in a backyard pool or spa, 14 in a bathtub, 7 in a lake or pond, and 4 in other sites. Bystander resuscitation was documented in 218 patients. Submersion events occurred in a large geographic area, including urban, rural, mountain, and desert areas. Average paramedic arrival time was 6.8 minutes (143 patients). Upon paramedic arrival, 78 children were in cardiac arrest and 13 had pulseless electrical activity. Paramedic CPR was documented in 87 children, 18 of whom did not require CPR in the ED. Paramedics intubated 19 children and administered epinephrine to 30 patients.
Upon arrival in the ED, 71 children were in cardiac arrest, 13 had pulseless electrical activity, 7 were apneic, and 26 were breathing but comatose. Five additional patients deteriorated in the ED and required CPR. In the ED, 125 patients were intubated and 84 patients received epinephrine. CPR was administered to 89 patients in the ED (average duration 8.9 minutes, range 2 to 105 minutes), of whom 41 survived (8 intact, 33 vegetative). The longest CPR duration in an intact survivor was 47 minutes. Persistent deficits were recorded in 15 of the 185 functionally intact survivors. An initial ED Glasgow Coma Score of 3 was recorded in 100 patients, 14 of whom survived intact. Of 165 patients having a documented Glasgow Coma Score ≥4 upon arrival in the ED, two survived in a vegetative state and all others survived intact. Of the 51 patients who died, parents chose to withdraw support in 22 and brain death occurred in 23. The time interval from submersion to first documented purposeful movement is shown in Fig 3. All intact survivors demonstrated functional recovery within 48 hours except UGO patient 5 (Fig 3). UGO patient 5 was lost to follow up, but was neurologically intact when seen 7 years after injury. Mean values for selected variables in patients with intact, UGO, and combined vegetative/dead outcome are shown in Table 1.
The classification accuracy of individual variables is shown in Table 2. Physical examination classifies patients most accurately, followed by the need for CPR in the ED, and then initial blood pH. All variables collected but not listed in Table 2 are not predictive. The best outcome classifications achieved by discriminant analysis entering all variables together are shown in Table3. The most accurate discriminant analysis categorization into intact, vegetative, and dead outcome used ED physical examination, performance of CPR in the ED, and initial pH achieving an overall accuracy of 87%. Categorization into only good vs poor outcome improved overall accuracy to 96% by eliminating errors between vegetative and dead groups. Addition or substitution of variables afforded no improvement in accuracy.
A clinical classification system was constructed by selecting one variable from each of the physical examination, ED treatment, and laboratory variable types shown in Table 2. A combination of 3 variables minimized the number of UGO patients: 1) need for CPR in the ED, 2) apnea and coma in the ED, and 3) pH <7.00. Patients meeting all 3 criteria were predicted to have poor outcome and all others were predicted to have good outcome. Classification using this method resulted in an overall accuracy of 93% (Table 4).
The actual outcome of patients classified by the clinical system into good and poor outcome categories is shown in Fig 4. Of those whose outcome was accurately predicted to be good, 8 had pH <7.00, 4 had apnea and coma in the ED, and 2 required CPR in the ED (both with pH >7.1, prehospital cardiac arrest, total CPR duration 15, 18 minutes). Of those predicted to have good outcome who survived in a vegetative state, 4 received CPR in the ED (all had pH ≥7.00), 2 had pH <7.00 but did not require CPR in the ED, and 3 met none of the poor outcome criteria. Of those predicted to have good outcome who died, 3 required CPR but had pH ≥7.00 and the remaining 3 did not require CPR in the ED. The 5 patients predicted to have poor outcome who survived intact (UGO patients) represent 6.3% of all patients predicted to have a poor outcome (Table 4). By 24 hours after injury, UGO patients 1 and 4 (Table 5) exhibited purposeful movements. UGO patients 2 and 3 demonstrated posturing to stimuli (1 decorticate, 1 decerebrate) at 24 hours, but exhibited semi-purposeful movements by 48 hours. Patient 5 was treated for 72 hours with barbiturate coma, after which she was noted to have a grimace, flexion withdrawal to painful stimuli, and absence of spasticity but no clearly purposeful movement. Although her neurologic examination was unchanged at the time of hospital discharge 1 month after injury, she subsequently recovered with attention deficit disorder and hyperactivity as her only deficit.
Mean acute care hospital costs per patient are shown in Fig5. Patients with vegetative outcome are the most expensive to care for while intact survivors predicted to have good outcome are the least expensive. Total acute care hospital costs by outcome group are shown in Fig 6. Intact survivors (two-thirds of all patients) account for only one-third of the total hospital costs, including 6% of total costs incurred by UGO patients. The vegetative survivors (14% of all patients) account for 53% of total costs.
Prediction of outcome in near drowning victims is important to guide triage decisions, counsel families, reduce unnecessary interventions, and guide withdrawal of support. Scoring and classification systems1,2 as well as individual prognostic factors have been used to prognosticate victims of submersion accidents. Factors shown to correlate with outcome include historic variables such as duration of submersion,2,3 water or patient temperature,4-6 and intervention at the scene;4 treatment variables such as need for CPR in the ED,4,7-9 apnea and pulselessness,10,11 depth of coma,1,12,13 and neurologic response to therapy;12,14 and laboratory variables such as pH2 and serum glucose.15 In our study, optimal categorization into intact vegetative and dead outcome occurred with combined use of only 3 variables. This, plus our finding that discriminant analysis provides no advantage over the simple clinical system in minimizing incorrect poor outcome prediction supports published criteria for differentiating most near drowning survivors from nonsurvivors using simple classification systems.
When evaluating classification systems, two types of categorization errors should be considered: misclassification between good and poor outcome, and misclassification between vegetative and dead categories. Two therapeutic endpoints are commonly endorsed: 1) to preserve life regardless of functional outcome, and 2) to preserve only functional survivors. Attempts to prolong life regardless of functional outcome require that all possible therapies be employed for all patients, making prediction of outcome unimportant for guiding therapy. Alternatively, when functional outcome is the desired endpoint, prediction of good or poor outcome is the clinically important distinction; misclassification between vegetative and dead categories is not important because both vegetative survival and death are highly undesirable. Hence, accurate prediction of good vs poor outcome is essential for guiding clinical decisions, while preoccupation with prediction of vegetative vs dead outcome has little clinical utility. Treatment decisions based on incorrect good outcome prediction (predicted good outcome for patients with actual vegetative survival or death) lead to futile therapy. Conversely, decisions based on prediction of poor outcome for actual intact survivors (incorrect poor outcome prediction) may lead to failure to provide life saving therapy. Because our society places a high value on preservation of functional life, incorrect poor outcome prediction must be minimal in any clinically useful classification system.
Although published scoring systems and prognostic factors have been shown to correlate with survival and outcome, no combination of available variables has been able to prospectively differentiate all intact survivors from those with poor outcome. For example, the need for CPR in the ED is commonly considered to portend death or vegetative outcome. However, 6 studies reported functional recovery in 0%,7 12%,4 15%,618%,13 21%,3 and 24%9 (17% overall average) of drowning victims who required CPR in the ED. Despite these reports, withholding or withdrawal of therapy from patients who have low probability of functional survival after warm water submersion injury has been suggested6,8,12,14,16 and criteria have been informally proposed for limitation of support. Such criteria include failure to respond to advanced life support within 25 minutes,3 and lack of purposeful movements and normal brain stem function 24 hours after near drowning.14 Such criteria are not widely employed because anecdotal experience with spectacular recoveries and the small numbers of severely injured patients in most studies raises uncertainty about their predictive accuracy. Findings shown in Tables 1 and 5 indicate that use of either of these criteria in our patients would have resulted in loss of intact survivors. Similarly, Graf et al. recently suggested that outcome for pediatric submersion victims can be predicted with four simple measures: coma, absence of pupillary light reflex, admission blood glucose concentration and sex.17 Of our 5 UGO patients, all had coma and 4 had absent pupillary light reflexes on arrival in the ED, 4 had initial blood sugar >400 mg/dL, and 4 were females. Using Graf's logistic-regression model, 4 of these 5 patients would have a probability of poor outcome >0.92, with 3 patients >.99. Mean values for patients with intact, UGO, and combined vegetative/dead outcome shown in Table 1 demonstrate that UGO patients differ from other patients with intact survival but are similar to those who die or have vegetative outcome. Table 5 illustrates UGO survivor characteristics which contrast with published series of drowning victims including prolonged CPR in patient 1,3 a temperature of 34°C in patient 3,6 and a prolonged vegetative state in patient 5.14 Our finding that addition or substitution of other variables was unable to improve accuracy suggests that no index or score applied in the ED will predict outcome with complete accuracy.12,18 Because hypoxic ischemic insult is the primary mechanism of postsubmersion neurologic injury, serial neurologic examinations over time would be expected to provide additional prognostic information. The rapid recovery of neurologic function demonstrated in Fig 3 suggests that 48 hours is a reasonable observation period for patients with predicted poor outcome. It is evident that limits beyond which intact survival is excluded cannot be established until other measures of neuronal injury become available.
There are several ways to respond to uncertainty in critical care medicine.19,20 One approach is to make decisions based on statistical likelihood of good or bad outcome; near drowning victims predicted to have poor outcome should not receive therapy. If this approach had been applied to our patients, the 5 UGO patients who represent 6.3% of submersion victims predicted to have poor outcome (Fig 4) would not have received life sustaining therapy. Most clinicians consider this error rate unacceptable. A contrasting approach is to provide treatment for all near drowning victims and to continue therapy until it is certain that a particular patient will either die or be severely impaired. This has been the traditional approach. Adherence to this approach assures aggressive treatment of all potential intact survivors but also results in many vegetative survivors. The personal, social, and financial costs of vegetative survival are enormous. Direct hospital costs shown in Figs 5 and 6represent an objective measure of the minimum cost associated with near drowning. Although the vegetative group accounts for more than 50% of all acute care hospital costs (Fig 6), these costs are only a small fraction of the total financial, emotional, and social costs incurred by this group. Financial costs to society should have a limited influence on individual care decisions until a social consensus is achieved. Instead, clinicians should focus on potential benefits of therapy as the primary factors in individual patient care decisions. Potential personal costs of debilitating neurologic injury may also influence decisions if vegetative survival is likely. We suggest an intermediate approach which begins with aggressive treatment of all near drowning victims, but supports termination of therapy before absolute prognostic certainty is established. The intermediate approach avoids the undesirable outcomes of both the statistical and traditional approaches. The possibility of error should be a reason for caution, not an excuse for mandating aggressive treatment to the bitter end.19,20 Although observation and therapy for 48 hours improves certainty (Fig 3), it does not eliminate the possibility of error. Insistent focus on small degrees of uncertainty prevents open exploration of therapeutic options such as discontinuation of life support, of intravenous fluids, and of nutritional support.20 True patient advocacy often requires that caretakers and family consider withdrawal of support early in the intensive care unit course.
Strategies suggested by our data to maximize the number of intact survivors while minimizing personal, hospital, and social costs include full therapy for patients in the good prognosis group and the following steps for all patients in the poor prognosis group.
Prehospital resuscitation, including early intubation, ventilation, vascular access, and administration of advanced life support medications.
Continued resuscitation and stabilization in the ED.
Full supportive care in the pediatric intensive care unit for a minimum of 48 hours.
Consider withdrawal of support if no neurologic improvement is detected after 48 hours. Ancillary testing such as brainstem evoked responses, electroencephalography, and magnetic resonance spectroscopy may prove helpful to corroborate the neurologic examination.
Aggressive pursuit of potential new therapies for hypoxic ischemic brain injury.
In summary, our results support published criteria for separating most near drowning survivors from non survivors using simple classification systems. Available variables cannot prospectively differentiate UGO from vegetative or dead groups. Infrequently, children may experience a prolonged vegetative state followed by unexpected good outcome. Because outcome cannot be accurately predicted in the ED, all near drowning victims should be treated aggressively for at least 48 hours. Reduction of expenditures on patients likely to have vegetative or dead outcome would result in substantial savings, but loss of normal survivors. Withdrawing therapy from patients who show no improvement 48 hours after injury may minimize loss of normal survivors and reduce costs. Community education, and implementation of effective prevention measures should receive continued emphasis.
- ED =
- emergency department •
- CPR =
- cardiopulmonary resuscitation •
- UGO =
- unpredictable good outcome
- Quan L,
- Kinder D
- Kyriacou DN,
- Arcinue EL,
- Peek C,
- Kraus JF
- Peterson B
- Ashwal S,
- Schneider S,
- Tomasi L,
- Thompson J
- Copyright © 1997 American Academy of Pediatrics