Abstract
Objective
To develop a clinical prediction score for predicting mortality in children following return of spontaneous circulation after in-hospital cardiac arrest.
Design
Observational study using prospectively collected data.
Setting
This was an analysis using data from the Get With The Guidelines®-Resuscitation (GWTG-R) registry between January 2000 and December 2015.
Patients
Pediatric patients (< 18 years of age) who achieved return of spontaneous circulation.
Interventions
None.
Measurements and Main Results
The primary outcome was in-hospital mortality. Patients were divided into a derivation (3/4) and validation (1/4) cohort. A prediction score was developed using a multivariable logistic regression model with backwards selection. Patient and event characteristics for the derivation cohort (n = 3893) and validation cohort (n = 1297) were similar. Seventeen variables associated with the outcome remained in the final reduced model after backwards elimination. Predictors of in-hospital mortality included age, illness category, pre-event characteristics, arrest location, day of the week, non-shockable pulseless rhythm, duration of chest compressions, and interventions in place at time of arrest. The c-statistic for the final score was 0.77 (95%CI: 0.75, 0.78) in the derivation cohort and 0.77 (95%CI: 0.74, 0.79) in the validation cohort. The expected vs. observed mortality plot indicated good calibration in both the derivation and validation cohort. The score showed a stepwise increase in mortality with an observed mortality of < 15% for scores 0 – 9 and > 80% for scores ≥ 25. The model also performed well for neurological outcome and in sensitivity analyses for events within the past five years, and for patients with or without a pulse at the onset of chest compressions.
Conclusions
We developed and internally validated a prediction score for initial survivors of pediatric in-hospital cardiac arrest. This prediction score may be useful for prognostication following cardiac arrest, stratifying patients for research, and guiding quality improvement initiatives.
Keywords: Pediatrics, Prediction model, Heart arrest, Prognosis, Mortality, Neurological deficit
INTRODUCTION
Cardiac arrest occurs in an estimated 16,000 children annually in the United States, with a higher rate of cases in the in-hospital setting compared to the out-of-hospital setting.(1,2) While a number of clinical models exist for predicting mortality following in-hospital cardiac arrest in adult patients(3–7), there is a paucity of clinically useful models to predict outcomes for children with return of spontaneous circulation (ROSC) after in-hospital cardiac arrest.
Using a United States-based national registry, Chan et al. recently developed a model to predict favorable neurological outcome in adult patients successfully resuscitated from in-hospital cardiac arrest.(4) The Cardiac Arrest Survival Post-Resuscitation In-Hospital (CASPRI) score joined other models as ways to predict outcomes after adult in-hospital cardiac arrest, including the Good Outcomes Following Attempted Resuscitation (GO-FAR) score(5), the Pre-Arrest Morbidity (PAM) score(6), and the Prognosis After Resuscitation (PAR) score.(7) Prediction models provide outcome estimates for individual patients, help physicians provide prognostic information for family members, allow for stratification across hospitals to compare outcomes for quality improvement initiatives, and might be useful for patient stratification in clinical trials.
The aim of the current study was to develop a clinical prediction score for predicting mortality in children following ROSC after in-hospital cardiac arrest.
MATERIALS AND METHODS
Study design and data source
This study is an analysis using data from the Get With The Guidelines®-Resuscitation (GWTG-R) registry. The GWTG-R is a national, prospective, quality-improvement registry of in-hospital cardiac arrest patients, sponsored by the American Heart Association. In the GWTG-R registry, cardiac arrest is defined as no pulse or a pulse with inadequate perfusion requiring chest compressions, defibrillation or both, with a hospital-wide or unit-based emergency response by acute care personnel. Data are collected on all patients with a cardiac arrest without a do-not-resuscitate order. The design, data collection, and reliability of the GWTG-R registry have been described in detail elsewhere.(8,9) In this study, we included data from January 1, 2000 to December 31, 2015.
All participating hospitals are required to comply with local regulatory guidelines. Because data are used primarily at the local site for quality improvement, sites are granted a waiver of informed consent. The GWTG-R registry is de-identified and the use in research in the absence of any other identifiable subject data does not meet the definition of human subject research.
Study population
We included pediatric patients (< 18 years of age) with an index cardiac arrest and ≥ 1 minute of documented chest compressions who achieved ROSC, defined as a palpable pulse without chest compressions for at least 20 minutes. We included patients with a loss of pulse and without a loss of pulse at the onset of chest compressions (e.g., patients receiving chest compressions for poor perfusion). We excluded patients for whom the event occurred in the delivery room or in the neonatal intensive care unit, and patients in whom cardiopulmonary bypass (e.g., E-CPR) was employed during the event, as these events differ substantially from other cardiac arrests.(10) Furthermore, patients with missing data on included covariates and in-hospital mortality were excluded for the primary analysis.
Outcome and predictor variables
The primary outcome was in-hospital mortality. The secondary outcome was neurological outcome at hospital discharge measured by the Pediatric Cerebral Performance Category (PCPC) score(11), as recommended by the Utstein guidelines.(12) A PCPC score of 1 indicates no neurological deficit; 2, mild cerebral disability; 3, moderate cerebral disability; 4, severe cerebral disability; 5, coma or vegetative state; and 6, brain death. A PCPC score of 1 or 2 was considered a favorable neurological outcome, and a PCPC score from 3 to 6 or death was considered an unfavorable neurological outcome. To account for alternative definitions of unfavorable neurological outcome in pediatric cardiac arrest patients, we performed sensitivity analyses using the following definitions: (1) a PCPC score of 3, 4, 5, 6 or worse than baseline, (2) a PCPC score of 4, 5 or 6, and (3) a PCPC score of 4, 5, 6 or worse than baseline.(13,14)
Predictor variables for the study outcome were selected a priori for model inclusion based on their potential relevance to in-hospital mortality and based on availability in the registry.(9,15–18) The full list of variables is presented in Table 1. See Table S1 in the Supplement for a detailed description of pre-event characteristics, illness category, and interventions in place at the time of cardiac arrest.
Table 1.
Patient characteristicsa
| Characteristics | Derivation cohort (N = 3893) | Validation cohort (N = 1297) |
|---|---|---|
| Demographics | ||
| Sex | ||
| Female | 1741 (45) | 535 (41) |
| Male | 2152 (55) | 762 (59) |
| Age group | ||
| Neonate (< 1 month) | 963 (25) | 353 (27) |
| Infant (1 month to < 1 year) | 1208 (31) | 386 (30) |
| Child (1 year to < 12 years) | 1211 (31) | 387 (30) |
| Adolescent (≥ 12 years) | 511 (13) | 171 (13) |
| Illness categoryb | ||
| Medical cardiac | 613 (16) | 196 (15) |
| Medical non-cardiac | 1628 (42) | 527 (41) |
| Surgical cardiac | 817 (21) | 274 (21) |
| Surgical non-cardiacc | 619 (16) | 213 (16) |
| Newbornd | 216 (5) | 87 (7) |
| Pre-event characteristicsb | ||
| Heart failure this admission | 366 (9) | 126 (10) |
| Heart failure prior to admission | 281 (7) | 91 (7) |
| Hypotension | 1055 (27) | 399 (31) |
| Respiratory insufficiency | 2381 (61) | 797 (61) |
| Hepatic insufficiency | 163 (4) | 44 (3) |
| Renal insufficiency | 359 (9) | 127 (10) |
| Metabolic or electrolyte abnormality | 553 (14) | 185 (14) |
| Metastatic or hematological cancer | 146 (4) | 35 (3) |
| Acute non-stroke CNS event | 256 (7) | 84 (6) |
| Baseline Depression in CNS function | 590 (15) | 191 (15) |
| Pneumonia | 325 (8) | 125 (10) |
| Septicemia | 478 (12) | 155 (12) |
| Major trauma | 274 (7) | 83 (6) |
| Location and Time of Arrest | ||
| Location | ||
| Emergency department | 322 (8) | 99 (8) |
| Intensive care unit | 2669 (69) | 926 (71) |
| Monitored floor | 78 (2) | 13 (1) |
| Unmonitored floor | 247 (6) | 75 (6) |
| Othere | 577 (15) | 184 (14) |
| Time of the day | ||
| Nighttimef | 1049 (27) | 372 (29) |
| Daytime | 2844 (73) | 925 (71) |
| Day of the week | ||
| Weekendg | 1105 (28) | 354 (27) |
| Weekday | 2788 (72) | 943 (73) |
| Arrest characteristics | ||
| Witnessed | ||
| Yes | 3709 (95) | 1229 (95) |
| No | 184 (5) | 68 (5) |
| Monitored | ||
| Yes | 3601 (93) | 1195 (92) |
| No | 292 (8) | 102 (8) |
| Pulseless rhythm | ||
| Ventricular fibrillation | 170 (4) | 50 (4) |
| Pulseless ventricular tachycardia | 141 (4) | 48 (4) |
| Pulseless electrical activity | 985 (25) | 336 (26) |
| Asystole | 760 (20) | 212 (16) |
| Never pulseless | 1837 (47) | 651 (50) |
| Duration of chest compressions (minutes)h | ||
| 1 | 427 (11) | 145 (11) |
| 2 – 4 | 1053 (27) | 324 (25) |
| 5 – 9 | 831 (21) | 294 (23) |
| 10 – 14 | 448 (12) | 143 (11) |
| 15 – 19 | 255 (7) | 88 (7) |
| 20 – 24 | 233 (6) | 85 (7) |
| 25 – 29 | 181 (5) | 58 (4) |
| 30 – 39 | 181 (5) | 61 (5) |
| 40 – 49 | 101 (3) | 36 (3) |
| ≥ 50 | 183 (5) | 63 (5) |
| Interventions in place at the time of cardiac arrestb | ||
| Mechanical ventilation | 2435 (63) | 838 (65) |
| Intravenous antiarrhythmics | 109 (3) | 34 (3) |
| Intravenous vasopressors | 1095 (28) | 366 (28) |
| Arterial line | 1044 (27) | 387 (30) |
All variables are reported as counts (frequencies). Abbreviations: CNS, central nervous system; ROSC, return of spontaneous circulation.
See Table S1 in the Supplement for definitions.
Including non-cardiac surgical patients (60%), trauma patients (39%), and obstetric patients (2%).
Newborn defined as born on the current admission.
Including ambulatory or outpatient clinics, diagnostic or interventional areas, operating room, post-anesthesia recovery room, rehabilitation units, same-day surgical areas, and other.
Night defined as the time between 11:00 pm and 6:59 am.
Friday 11:00 pm to Monday 7:00 am.
Duration of chest compression was defined as the time from when chest compressions was first required to ROSC. The categories for this variable were based on visual inspection of the unadjusted associated between time with chest compressions and mortality in the derivation cohort.
Statistical analysis
The cohort was divided at random into a derivation set (3/4 of the overall cohort) and validation set (1/4 of the overall cohort). Categorical variables are reported as counts with relative frequencies.
We created a list of potentially relevant clinical variables, which was entered in a multivariable logistic regression model in the derivation cohort. To account for correlations between patients within the same hospital we used generalized estimation equations (GEE) with an exchangeable covariance structure. Backwards selection was used to sequentially (based on the highest p-value) eliminate variables not significantly associated with the primary outcome (at a p < 0.05) or that did not improve overall model fit as determined by the Quasi-likelihood under the Independence-model Criterion (QIC) statistic.(19) A clinical prediction score was then created by assigning weighted points proportional to the B-coefficient values from the remaining variables in the parsimonious model, such that a higher point represented an increased risk of in-hospital mortality. Categories with similar B-coefficients (i.e., categories receiving the same score in the final model) within the same variable were collapsed for simplification.
Discrimination (c-statistic) and calibration (comparison of predicted and actual mortality in deciles, the Hosmer-Lemeshow test) were calculated for the full model and the parsimonious model in the derivation cohort, and for the clinical score in both the derivation and validation cohorts. For the GEE models, the 95% confidence interval for the c-statistic was estimated by bootstrapping 1000 data sets using unrestricted random sampling.
Multiple imputation was conducted as a predefined sensitivity analysis to account for missing data. This analysis was conducted in the complete cohort with patients meeting all inclusion criteria and no exclusion criteria (i.e., combining the derivation and validation cohorts). Missing data on covariates and in-hospital mortality were imputed using the fully conditional specification method.(20) A total of 20 imputed data sets were created.(21) For each data set, we calculated the c-statistic and Hosmer-Lemeshow statistic for the final prediction score. Given that our measure of discrimination (c-statistic) is not symmetrically distributed, does not follow a specific distribution, and is bounded by zero and one, we report the median c-statistic with the 1st and 3rd quartiles to illustrate the distribution of the values over the imputed datasets.(22) A similar approach was used for the results of the Hosmer-Lemeshow test.
As a predefined sensitivity analysis, the discrimination and calibration of the final prediction score was also assessed for neurological outcome. Sensitivity analyses were conducted with alternative definitions of neurological outcome. To test the robustness of the model, we performed a number of post-hoc analyses in the validation cohort including (1) only events within the past five years (2011 – 2015) to assess score performance in a more contemporary cohort, (2) only patients with a loss of pulse, (3) only patients without a loss of pulse, and (4) patients based on age (neonate, infant, child, or adolescent) to verify the applicability of the score in specific age groups.
The newborn illness category (i.e., being born on the current admission) was removed from the GWTG-R registry in 2015 leaving the categories medical-cardiac, medical non-cardiac, surgical-cardiac, and surgical non-cardiac. To confirm the future value of the score, we conducted two post-hoc sensitivity analyses. First, we excluded patients classified as newborn from the validation cohort and reassessed the score. Second, we reclassified newborn as missing for the illness category variable and reran the multiple imputation analysis for the complete cohort. Using this approach, we essentially reclassified newborn into medical-cardiac, medical non-cardiac, surgical-cardiac, or surgical non-cardiac.
SAS version 9.4 (SAS Institute, Cary, NC, USA) was used for all analyses. No adjustments were made for multiple comparisons.
RESULTS
Patient Characteristics
The final cohort included 5190 patients (Figure 1). Patient and event characteristics for the derivation cohort (n = 3893) and validation cohort (n = 1297) were similar (Table 1). The median age was 7 months (quartiles: 28 days, 4 years) and 44% were female. Of those with a pulseless rhythm, 85% had an initial non-shockable rhythm. The median duration of chest compressions was 7 minutes (quartiles: 3, 17). In-hospital mortality was 39%.
Figure 1. Flow diagram for inclusion and exclusion criteria.
Out of 17767 pediatric patients with an in-hospital cardiac arrest, 5190 patients were included in the final cohort. Missing data on individual variables exceed 1315, as patients may have missing data on more than one variable. ROSC indicates return of spontaneous circulation.
Derivation Cohort
We included 27 variables in the full non-parsimonious model. The c-statistic for the full model was 0.77 (95%CI: 0.76, 0.79). Backwards selection eliminated 10 variables. The c-statistic for the reduced model remained unchanged (c-statistic: 0.77, 95%CI: 0.76, 0.79). The expected vs. observed mortality plot and the Hosmer-Lemeshow goodness-of-fit test (p = 0.07) indicated good calibration of the model.
The coefficients from the reduced model were assigned weighted points for the 17 variables independently associated with mortality (Table 2). The scores ranged from 2 to 40 (theoretically from 0 to 54) (Figure 2A) with a median score of 13 (quartiles: 10, 18). Predictors of in-hospital mortality following ROSC included age, illness category, pre-event characteristics, arrest location, day of the week, non-shockable pulseless rhythm, duration of chest compressions, and interventions in place at time of arrest. The odds ratio for in-hospital mortality per unit increase in the final score was 1.22 (95%: 1.20, 1.24, p < 0.001). The c-statistic for the final score was 0.77 (95%CI: 0.75, 0.78, Figure 3A). The Hosmer-Lemeshow goodness-of-fit test (p = 0.04) was nominally significant but the expected vs. observed mortality plot indicated good calibration (Figure 4A). The score was separated into 5 categories (0 – 9, 10 – 14, 15 – 19, 20 – 24, and ≥ 25) and plotted against in-hospital mortality (Figure 5A). All scores showed a stepwise increase in mortality with an observed mortality of 12% (95%CI: 9%, 14%) for scores 0 – 9 and 84% (95%CI: 79%, 88%) for scores ≥ 25. See Figure S1A in the Supplement for the mortality distribution in the raw score.
Table 2.
Final model for the derivation cohorta
| Characteristics | OR (95%CI) | P-value | Score |
|---|---|---|---|
| Age group | |||
| Neonate/Infant/Child (< 12 years) (reference) | – | – | 0 |
| Adolescent (≥ 12 years) | 1.46 (1.16, 1.83) | 0.001 | 2 |
| Illness category | |||
| Surgical (reference) | – | – | 0 |
| Medical | 1.53 (1.28, 1.81) | < 0.001 | 2 |
| Newborn | 1.82 (1.23, 2.71) | 0.003 | 3 |
| Pre-event characteristics | |||
| Metabolic or electrolyte abnormality | 1.40 (1.14, 1.74) | 0.002 | 2 |
| Hypotension | 1.53 (1.22, 1.90) | < 0.001 | 2 |
| Absence of pneumonia | 1.53 (1.19, 1.97) | 0.001 | 2 |
| Acute non-stroke CNS event | 1.54 (1.16, 2.05) | 0.003 | 2 |
| Septicemia | 1.76 (1.35, 2.29) | < 0.001 | 3 |
| Hepatic insufficiency | 1.84 (1.29, 2.64) | < 0.001 | 3 |
| Renal insufficiency | 2.08 (1.60, 2.70) | < 0.001 | 4 |
| Metastatic or hematological cancer | 3.96 (2.55, 6.15) | < 0.001 | 7 |
| Major trauma | 4.03 (3.03, 5.35) | < 0.001 | 7 |
| Location | |||
| Floor/Other (reference) | – | – | 0 |
| ICU/ED | 1.61 (1.28, 2.04) | < 0.001 | 2 |
| Day of the week | |||
| Weekday (reference) | – | – | 0 |
| Weekend | 1.26 (1.12, 1.43) | < 0.001 | 1 |
| Pulseless rhythm | |||
| Shockable/Never (reference) | – | – | 0 |
| Non-shockable | 1.34 (1.15, 1.56) | 0.001 | 1 |
| Duration of chest compressions (minutes) | |||
| 1 (reference) | – | – | 0 |
| 2 – 4 | 1.42 (1.15, 1.76) | 0.001 | 2 |
| 5 – 29 | 2.19 (1.68, 2.85) | < 0.001 | 4 |
| ≥ 30 | 3.93 (2.59, 5.98) | < 0.001 | 7 |
| Interventions in place at time of arrest | |||
| Mechanical ventilation | 1.51 (1.28, 1.78) | < 0.001 | 2 |
| Intravenous vasopressors | 2.05 (1.59, 2.63) | < 0.001 | 4 |
Abbreviations: ICU, intensive care unit; ED, emergency department
Figure 2. Histogram for the score in the derivation and validation cohort.
In the derivation (A) cohort, scores ranged from 2 – 40 with a median of 13 (quartiles: 10, 18). In the validation (B) cohort, scores ranged from 2 – 37 with a median of 14 (quartiles: 11, 19).
Figure 3. Area Under the Receiver Operating Characteristics curve for the score in the derivation and validation cohort.
The c-statistic was 0.77 (95%CI: 0.75, 0.78) for the derivation (A) cohort, and 0.77 (95%CI: 0.74, 0.79) for the validation (B) cohort.
Figure 4. Model calibration for the score in the derivation and validation cohort.
The expected vs. observed mortality plot indicated good calibration for both the derivation (A) and the validation (B) cohort.
Figure 5. Mortality per score category in the derivation and validation cohort.
We found a stepwise increase for in-hospital mortality per increasing score in both the derivation (A) and validation (B) cohort. The vertical lines represent 95% confidence intervals.
Validation Cohort
The score ranged from 2 to 37 with a median score of 14 (quartiles: 11, 19) (Figure 2B) in the validation cohort. The c-statistic was 0.77 (95%CI: 0.74, 0.79, Figure 3B). The Hosmer-Lemeshow goodness-of-fit test (p = 0.14) and expected vs. observed mortality plot (Figure 4B) remained indicative of good calibration with no meaningful discrepancy between the expected and observed outcomes. The scores in the validation cohort showed a similar stepwise increase in mortality as in the derivation cohort with an observed mortality of 14% (95%CI: 9%, 18%) for scores 0 – 9 and 86% (95%CI: 78%, 94%) for scores ≥ 25 (Figure 5B). See Figure S1B in the Supplement for the mortality distribution in the raw score.
Neurological Outcome
1008 patients from the validation cohort were included in the secondary analysis regarding neurological outcome. 638 (63%) patients had an unfavorable neurological outcome defined as PCPC 3, 4, 5, 6 or death. The c-statistic for this outcome was 0.72 (95%CI: 0.69, 0.75) with a Hosmer-Lemeshow goodness-of-fit p-value of 0.25 indicating good calibration. The sensitivity analyses accounting for alternative definitions of neurological outcome were similar and demonstrated discriminative power and calibration comparable to the primary definition. See Table S2 in the Supplement for more details.
Missing Data
After performing multiple imputations for missing data, 6506 patients were available for analysis. The results for the imputed data sets were similar to the primary analysis. The median c-statistic was 0.76 (quartiles: 0.76, 0.76) and the median p-value from the Hosmer-Lemeshow goodness-of-fit test was 0.15 (quartiles: 0.08, 0.17).
Post-Hoc Sensitivity Analyses
The sensitivity analyses for different categories of patients (events within the past five years, patients with a loss of pulse, patients without a loss of pulse, patients in pre-defined age groups, and patients without a newborn illness category) in the validation cohort demonstrated overall good discrimination and moderate calibration. See Table S3 and Figure S2 in the Supplement for more details. The multiple imputation analysis reclassifying newborn as missing data was comparable to the primary imputed data sets. The c-statistic was 0.76 (quartiles: 0.76, 0.76) and the median Hosmer-Lemeshow goodness-of-fit p-value was 0.09 (quartiles: 0.07, 0.15).
DISCUSSION
We developed and internally validated a prediction score for initial survivors of pediatric in-hospital cardiac arrest. The final score demonstrated good discrimination and calibration in both the derivation and validation cohort. With higher scores, there was a stepwise increase in mortality with an observed mortality of < 15% for scores 0 – 9 and > 80% for scores ≥ 25.
To our knowledge, this is the first clinical prediction score for in-hospital mortality in children following ROSC after in-hospital cardiac arrest. Jayaram et al. recently leveraged the GWTG-R registry to develop a model for predicting survival to hospital discharge in pediatric in-hospital cardiac arrest patients from 2000 to 2009.(23) Their model provides institutions an opportunity to compare risk-adjusted survival outcomes within and between institutions. However, for several reasons the model may have limited clinical applicability at the patient level. First, since the primary aim of their study was to assess hospital variation in cardiac arrest survival, they included both patients with and without ROSC. Successfully resuscitated cardiac arrest patients have significantly different patient characteristics and mortality rates compared to those who do not survive the initial event. Second, their model did not include patients without a loss of pulse, which is a common presentation in pediatric cardiac arrests.(18,24) Third, they included patients who were rescued with E-CPR (cardiopulmonary bypass).(23)
In our model, variables independently associated with pediatric in-hospital mortality were largely consistent with prior studies looking at individual risk factors. For example, pediatric in-hospital mortality vary depending on age with higher survival rates in lower age-groups.(16) A first rhythm of pulseless electrical activity or asystole has been associated with higher mortality rates compared to ventricular fibrillation or pulseless ventricular tachycardia.(9) Cardiac arrest secondary to trauma and malignancy has particularly high mortality rates, reflecting the overall poor prognosis associated with these conditions.(24,25) A history of trauma and malignancy was relatively uncommon in our study (7% and 3%) and the high prediction score for these variables are therefore only applicable to a relatively small proportion of patients. Lastly, patients who arrest following cardiac surgery are known to have more favorable outcomes compared to those with a medical illness.(15) Other factors that might be associated with outcomes did not improve the predictive performance of our score, such as time of day of the arrest, certain pre-event characteristics, whether the event was witnessed or monitored, and whether antiarrhythmics were in place at the time of the cardiac arrest. Of note, we found pneumonia at the time of the event to be associated with decreased mortality. While this might seem counterintuitive, cardiac arrests in pediatric patients frequently result from respiratory failure, which has been shown to have a better prognosis compared to other etiologies.(26) There was no difference in the prognosis for patients with shockable rhythms and patients without a loss of pulse in our model, although these estimates depend on what variables are included when developing the model and may be different in other settings.(18)
The major distinction between our pediatric model and the model developed by Chan et al. for adult patients using the GWTG-R registry(4) were that certain pre-event characteristics (metabolic or electrolyte abnormality, pneumonia, acute non-stroke CNS event, and major trauma), the use of intravenous vasopressors, the day of the week, and the inclusion of patients without a loss of pulse were associated with mortality and included in the final score. Pediatric patients have different outcomes compared to adult patients, likely due to differences in etiology and pathophysiology.(2,27) For example, pediatric events are frequently caused by respiratory failure and circulatory shock(28) and the initial rhythm in pediatric patients are usually asystole or bradycardia(9), rather than asystole or pulseless electrical activity as seen in adult patients(29). Both the pediatric and adult model found that higher age, non-shockable rhythms, duration of chest compressions, the use of mechanical ventilation, and certain pre-event characteristics (renal insufficiency, hepatic insufficiency, sepsis, malignancy, and hypotension) were negatively associated with outcomes.
Our study cohort included both patients with a loss of pulse and without a loss of pulse when chest compressions were initiated. Bradycardia with poor perfusion requiring chest compressions was the most frequent presentation, accounting for 50% of all rhythms at the time chest compressions were initially provided. Sensitivity analyses indicated that our score performed well in both populations. The positive but relatively weak relationship between a non-shockable rhythm and mortality might be explained by: (1) our inclusion only of patients who survived the initial event and (2) because other variables potentially related to the initial rhythm, such as duration of cardiopulmonary resuscitation, were also included.
There are multiple post-cardiac arrest variables, not collected by the GWTG registry, that may improve the clinician’s ability to predict outcomes among children with ROSC. For example, multiple studies have found early electroencephalographic (EEG) findings after successful resuscitation to be predictive of outcomes at hospital discharge.(30–32) Neurological examination for motor and pupil response can be predictive of outcomes in pediatric patients undergoing therapeutic hypothermia post-cardiac arrest.(33) Likewise, biomarkers including NSE and S-100B can be predictive of both survival and neurological outcomes.(34) Due to the nature of the data registry, our score did not include post-cardiac arrest variables or quality of CPR metrics. Future models should aim to include these factors to potentially improve predictive performance.
We chose mortality as the primary outcome since this is the standard outcome measure for in-hospital cardiac arrest research(35) and there was little missing data for this variable. Our final score also demonstrated good discrimination and calibration for unfavorable neurological outcome, and performed well for alternative definitions of unfavorable neurological outcome.
Prediction scores may be useful for prognostication, research, and quality improvement. For example, a 1-month old patient who arrests in the pediatric intensive care unit following cardiac surgery and survives after receiving chest compressions for 5 minutes will have a score of 8. On average, approximately 13% of patients with this score died prior to hospital discharge. This information may be useful for health care providers in providing prognostic information for patients and family members. The prediction score may also be used for stratifying patients in clinical trials by identifying low-, medium-, or high-risk patients who might be more or less likely to benefit from a given intervention. Lastly, the score could be utilized to compare outcomes among children with ROSC after in-hospital cardiac arrest across institutions while adjusting for baseline risk, as previously done by Jayaram et al. for all children with in-hospital cardiac arrest (both those with and without ROSC).(23)
Our results should be interpreted in the context of some study limitations. First, we had only access to in-hospital outcomes and the predictive performance after hospital discharge remains unknown. Second, there were some missing data on neurological outcome and analyses related to this outcome should be interpreted with caution. Third, although the GWTG-R data is derived from a large number of hospitals within the United States, the performance of the score may be different for non-participating hospitals. Prospective external validation should be performed before use in the clinical setting. Fourth, given the constraints of the database, the score could not be compared to existing pediatric severity of illness scores. Finally, although the score showed good discrimination and calibration, we were not able to identify a group of patients with 100% mortality and the score should not be used for individual withdrawal of care decisions.
CONCLUSIONS
We developed and internally validated a prediction score for initial survivors of pediatric in-hospital cardiac arrest. This prediction score may be useful for prognosticating following cardiac arrest, stratifying patients for research, and guiding quality improvement initiatives.
Supplementary Material
Acknowledgments
SOURCES OF FUNDING
The remaining authors have no conflicts of interest relevant to this article to disclose.
Mathias J. Holmberg and Lars W. Andersen conceived the idea, were responsible for data acquisition, performed the statistical analyses, and drafted the manuscript. All authors contributed substantially to the design of the study, interpreted the results, critically revised the manuscript, and approved the final manuscript prior to submission.
Footnotes
No reprints will be ordered.
CONFLICTS OF INTEREST
There was no specific funding for this study. Dr. Andersen serves as a compensated statistical reviewer for JAMA. Dr. Donnino is supported by grant 1K24HL127101-01 from the National Heart, Lung, and Blood Institute.
Copyright form disclosure: Drs. Moskowitz, Topjian, and Donnino received support for article research from the National Institutes of Health (NIH). Dr. Berg disclosed that he is an American Heart Association Get With the Guidelines-Resuscitation volunteer committee member. Drs. Topjian and Donnino’s institutions received funding from the NIH. Dr. Topjian received funding from expert testimony and legal consulting. The remaining authors have disclosed that they do not have any potential conflicts of interest.
References
- 1.Topjian AA, Berg RA, Nadkarni VM. Pediatric cardiopulmonary resuscitation: advances in science, techniques, and outcomes. Pediatrics. 2008;122(5):1086–1098. doi: 10.1542/peds.2007-3313. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Berg MD, Nadkarni VM, Zuercher M, Berg RA. In-hospital pediatric cardiac arrest. Pediatr Clin North Am. 2008;55(3):589–604. doi: 10.1016/j.pcl.2008.02.005. [DOI] [PubMed] [Google Scholar]
- 3.Cohn EB, Lefevre F, Yarnold PR, Arron MJ, Martin GJ. Predicting survival from in-hospital CPR: meta-analysis and validation of a prediction model. J Gen Intern Med. 1993;8(7):347–353. doi: 10.1007/BF02600069. [DOI] [PubMed] [Google Scholar]
- 4.Chan PS, Spertus JA, Krumholz HM, et al. A validated prediction tool for initial survivors of in-hospital cardiac arrest. Arch Intern Med. 2012;172(12):947–953. doi: 10.1001/archinternmed.2012.2050. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ebell MH, Jang W, Shen Y, Geocadin RG Investigators GWtGR. Development and validation of the Good Outcome Following Attempted Resuscitation (GO-FAR) score to predict neurologically intact survival after in-hospital cardiopulmonary resuscitation. JAMA Intern Med. 2013;173(20):1872–1878. doi: 10.1001/jamainternmed.2013.10037. [DOI] [PubMed] [Google Scholar]
- 6.George AL, Folk BP, Crecelius PL, Campbell WB. Pre-arrest morbidity and other correlates of survival after in-hospital cardiopulmonary arrest. Am J Med. 1989;87(1):28–34. doi: 10.1016/s0002-9343(89)80479-6. [DOI] [PubMed] [Google Scholar]
- 7.Ebell MH. Prearrest predictors of survival following in-hospital cardiopulmonary resuscitation: a meta-analysis. J Fam Pract. 1992;34(5):551–558. [PubMed] [Google Scholar]
- 8.Peberdy MA, Kaye W, Ornato JP, et al. Cardiopulmonary resuscitation of adults in the hospital: a report of 14720 cardiac arrests from the National Registry of Cardiopulmonary Resuscitation. Resuscitation. 2003;58(3):297–308. doi: 10.1016/s0300-9572(03)00215-6. [DOI] [PubMed] [Google Scholar]
- 9.Nadkarni VM, Larkin GL, Peberdy MA, et al. First documented rhythm and clinical outcome from in-hospital cardiac arrest among children and adults. JAMA. 2006;295(1):50–57. doi: 10.1001/jama.295.1.50. [DOI] [PubMed] [Google Scholar]
- 10.Lasa JJ, Rogers RS, Localio R, et al. Extracorporeal Cardiopulmonary Resuscitation (E-CPR) During Pediatric In-Hospital Cardiopulmonary Arrest Is Associated With Improved Survival to Discharge: A Report from the American Heart Association’s Get With The Guidelines-Resuscitation (GWTG-R) Registry. Circulation. 2016;133(2):165–176. doi: 10.1161/CIRCULATIONAHA.115.016082. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Fiser DH. Assessing the outcome of pediatric intensive care. J Pediatr. 1992;121(1):68–74. doi: 10.1016/s0022-3476(05)82544-2. [DOI] [PubMed] [Google Scholar]
- 12.Perkins GD, Jacobs IG, Nadkarni VM, et al. Cardiac Arrest and Cardiopulmonary Resuscitation Outcome Reports: Update of the Utstein Resuscitation Registry Templates for Out-of-Hospital Cardiac Arrest: A Statement for Healthcare Professionals From a Task Force of the International Liaison Committee on Resuscitation (American Heart Association, European Resuscitation Council, Australian and New Zealand Council on Resuscitation, Heart and Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Council of Southern Africa, Resuscitation Council of Asia); and the American Heart Association Emergency Cardiovascular Care Committee and the Council on Cardiopulmonary, Critical Care, Perioperative and Resuscitation. Resuscitation. 2015;96:328–340. doi: 10.1016/j.resuscitation.2014.11.002. [DOI] [PubMed] [Google Scholar]
- 13.Andersen LW, Berg KM, Saindon BZ, et al. Time to Epinephrine and Survival After Pediatric In-Hospital Cardiac Arrest. JAMA. 2015;314(8):802–810. doi: 10.1001/jama.2015.9678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Andersen LW, Raymond TT, Berg RA, et al. Association Between Tracheal Intubation During Pediatric In-Hospital Cardiac Arrest and Survival. JAMA. 2016;316(17):1786–1797. doi: 10.1001/jama.2016.14486. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Ortmann L, Prodhan P, Gossett J, et al. Outcomes after in-hospital cardiac arrest in children with cardiac disease: a report from Get With the Guidelines--Resuscitation. Circulation. 2011;124(21):2329–2337. doi: 10.1161/CIRCULATIONAHA.110.013466. [DOI] [PubMed] [Google Scholar]
- 16.Meaney PA, Nadkarni VM, Cook EF, et al. Higher survival rates among younger patients after pediatric intensive care unit cardiac arrests. Pediatrics. 2006;118(6):2424–2433. doi: 10.1542/peds.2006-1724. [DOI] [PubMed] [Google Scholar]
- 17.Matos RI, Watson RS, Nadkarni VM, et al. Duration of cardiopulmonary resuscitation and illness category impact survival and neurologic outcomes for in-hospital pediatric cardiac arrests. Circulation. 2013;127(4):442–451. doi: 10.1161/CIRCULATIONAHA.112.125625. [DOI] [PubMed] [Google Scholar]
- 18.Donoghue A, Berg RA, Hazinski MF, et al. Cardiopulmonary resuscitation for bradycardia with poor perfusion versus pulseless cardiac arrest. Pediatrics. 2009;124(6):1541–1548. doi: 10.1542/peds.2009-0727. [DOI] [PubMed] [Google Scholar]
- 19.Pan W. Akaike’s information criterion in generalized estimating equations. Biometrics. 2001;57(1):120–125. doi: 10.1111/j.0006-341x.2001.00120.x. [DOI] [PubMed] [Google Scholar]
- 20.van Buuren S. Multiple imputation of discrete and continuous data by fully conditional specification. Stat Methods Med Res. 2007;16(3):219–242. doi: 10.1177/0962280206074463. [DOI] [PubMed] [Google Scholar]
- 21.Graham JW, Olchowski AE, Gilreath TD. How many imputations are really needed? Some practical clarifications of multiple imputation theory. Prev Sci. 2007;8(3):206–213. doi: 10.1007/s11121-007-0070-9. [DOI] [PubMed] [Google Scholar]
- 22.Marshall A, Altman DG, Holder RL, Royston P. Combining estimates of interest in prognostic modelling studies after multiple imputation: current practice and guidelines. BMC Med Res Methodol. 2009;9:57. doi: 10.1186/1471-2288-9-57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Jayaram N, Spertus JA, Nadkarni V, et al. Hospital variation in survival after pediatric in-hospital cardiac arrest. Circ Cardiovasc Qual Outcomes. 2014;7(4):517–523. doi: 10.1161/CIRCOUTCOMES.113.000691. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Meert KL, Donaldson A, Nadkarni V, et al. Multicenter cohort study of in-hospital pediatric cardiac arrest. Pediatr Crit Care Med. 2009;10(5):544–553. doi: 10.1097/PCC.0b013e3181a7045c. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Suominen P, Räsänen J, Kivioja A. Efficacy of cardiopulmonary resuscitation in pulseless paediatric trauma patients. Resuscitation. 1998;36(1):9–13. doi: 10.1016/s0300-9572(97)00088-9. [DOI] [PubMed] [Google Scholar]
- 26.Berens RJ, Cassidy LD, Matchey J, et al. Probability of survival based on etiology of cardiopulmonary arrest in pediatric patients. Paediatr Anaesth. 2011;21(8):834–840. doi: 10.1111/j.1460-9592.2010.03479.x. [DOI] [PubMed] [Google Scholar]
- 27.Tress EE, Kochanek PM, Saladino RA, Manole MD. Cardiac arrest in children. J Emerg Trauma Shock. 2010;3(3):267–272. doi: 10.4103/0974-2700.66528. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.de Caen AR, Berg MD, Chameides L, et al. Part 12: Pediatric Advanced Life Support: 2015 American Heart Association Guidelines Update for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation. 2015;132(18 Suppl 2):S526–542. doi: 10.1161/CIR.0000000000000266. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Girotra S, Nallamothu BK, Spertus JA, et al. Trends in survival after in-hospital cardiac arrest. N Engl J Med. 2012;367(20):1912–1920. doi: 10.1056/NEJMoa1109148. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Kessler SK, Topjian AA, Gutierrez-Colina AM, et al. Short-term outcome prediction by electroencephalographic features in children treated with therapeutic hypothermia after cardiac arrest. Neurocrit Care. 2011;14(1):37–43. doi: 10.1007/s12028-010-9450-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Topjian AA, Sánchez SM, Shults J, Berg RA, Dlugos DJ, Abend NS. Early Electroencephalographic Background Features Predict Outcomes in Children Resuscitated From Cardiac Arrest. Pediatr Crit Care Med. 2016;17(6):547–557. doi: 10.1097/PCC.0000000000000740. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Nishisaki A, Sullivan J, Steger B, et al. Retrospective analysis of the prognostic value of electroencephalography patterns obtained in pediatric in-hospital cardiac arrest survivors during three years. Pediatr Crit Care Med. 2007;8(1):10–17. doi: 10.1097/01.pcc.0000256621.63135.4b. [DOI] [PubMed] [Google Scholar]
- 33.Abend NS, Topjian AA, Kessler SK, et al. Outcome prediction by motor and pupillary responses in children treated with therapeutic hypothermia after cardiac arrest. Pediatr Crit Care Med. 2012;13(1):32–38. doi: 10.1097/PCC.0b013e3182196a7b. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Topjian AA, Lin R, Morris MC, et al. Neuron-specific enolase and S-100B are associated with neurologic outcome after pediatric cardiac arrest. Pediatr Crit Care Med. 2009;10(4):479–490. doi: 10.1097/PCC.0b013e318198bdb5. [DOI] [PubMed] [Google Scholar]
- 35.Cummins RO, Chamberlain D, Hazinski MF, et al. Recommended guidelines for reviewing, reporting, and conducting research on in-hospital resuscitation: the in-hospital ‘Utstein style’. A statement for healthcare professionals from the American Heart Association, the European Resuscitation Council, the Heart and Stroke Foundation of Canada, the Australian Resuscitation Council, and the Resuscitation Councils of Southern Africa. Resuscitation. 1997;34(2):151–183. doi: 10.1016/s0300-9572(97)01112-x. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.