Association Between Evening Admissions and Higher Mortality Rates in the Pediatric Intensive Care Unit

Patients

Data for analysis were obtained from the most current available PICUEs database, Research Dataset 20–02.⁹ This database contains patient-level data from 34 993 admissions to 15 PICUs in the United States. Details of the site-selection procedures for the 15 study sites, general data-collection methods, and other analyses of this data set have been published.^9,10 Briefly, a minimum of 67 demographic, diagnostic physiologic, laboratory, and outcome data points were collected for all patients within 24 hours after admission to the PICU. Data for these analyses were collected for consecutive admissions between May 1995 and December 2001. Patients excluded from the database were patients admitted to the PICU for recovery from procedures usually cared for at other hospital locations, patients staying in the ICU <2 hours, and patients admitted in a state of continuous cardiopulmonary resuscitation. All data included in the database met quality standards, as assessed by data reabstraction and reliability testing.

The database includes data on 20 547 emergency admissions and 14 446 elective admissions, as defined by the Pediatric Risk of Mortality (PRISM)/PICUEs criteria. We limited our analyses, a priori, to patients admitted on an emergency basis, because we thought that their outcomes would be most sensitive to time variations in structure (eg, staffing levels) and processes (ie, the specific clinical activities performed) of care.¹¹

Study Design

The primary outcome variable for these analyses was death within 48 hours after admission to the PICU. Only deaths that occurred within 48 hours after admission were considered, because such deaths were more likely to be attributable to differences during the initial hours after admission (eg, death resulting from shock is more sensitive to care during the initial 24 hours after admission, whereas death occurring days to weeks after admission is more likely to be attributable to complications of care or hospital-acquired infections).^1,2,12 The primary independent variables were admission to the PICU on a weekend and admission to the PICU during the evening hours. Weekend admissions were defined as admissions to the PICU between 7 pm Friday and 7 am Monday. Evening admissions were defined as admissions between 5 pm and 7 am.

Other independent variables included as covariates were previous PICU admission during the current hospitalization, postoperative status, transfer from the inpatient ward, and use of vasoactive medications or ventilation within 24 hours after admission. The severity of illness was adjusted by using the PRISM III probability of death.^9,10 PRISM III is a third-generation mortality-prediction algorithm based on individual patient demographic, physiologic, and laboratory data^9,13 and is used to adjust for differences in mortality risk when comparing different populations of patients. The natural logarithm of the PRISM III probability of death was entered into the analyses as a quadratic transformation because of its curvilinear relationship with the observed logarithm of the odds of death.

Subgroup analyses were conducted on primary admission diagnoses according to the PRISM/PICUEs admission categories, including asthma, cancer, cardiac arrest, diabetic ketoacidosis, drug overdose, congenital cardiovascular disease, acquired cardiovascular disease, acute hypoxic ischemic encephalopathy, medical device malfunction, meningitis, pneumonia/bronchiolitis, seizure, sepsis, shock, suicide attempt, and trauma. These analyses were conducted to assess differences in mortality rates among individual diagnoses and were performed only for diagnoses with at least 100 patients and ≥10 deaths.^14,15 Primary admission diagnoses from PRISM/PICUEs that were not included were gastroesophageal reflux, human immunodeficiency virus, chromosomal abnormality, solid-organ transplantation, and “other.”

Analyses

Univariate analyses of continuous variables were conducted with Student's t test or the Mann-Whitney U test, when appropriate. Univariate analyses of dichotomous variables were conducted with the χ² test or Fisher's exact test, when appropriate. The odds of death within 48 hours were analyzed by using mixed-effects, multivariate, logistic regression, with clustering at the hospital level.¹⁶ Although hospital names were anonymous, each patient in the database was identified with a unique hospital code to allow statistical clustering of patients from the same institution. The adjusted odds ratios (ORs) are reported with 95% confidence intervals (CIs). Statistical significance was defined as P < .05. All calculations were performed by using Stata version 7 software (Stata, College Station, TX).

Human Subject Protection

This study was reviewed and approved by the University of California, Davis, Office for Human Subjects Protection.

Univariate Analyses

There were 20 547 pediatric patients admitted to the 15 PICUs on an emergency basis, and 7052 (34.3%) were admitted on weekends (Table 1). Compared with patients admitted on weekdays, patients admitted on weekends were slightly older (6.1 vs 5.8 years, P = .01), less likely to be admitted postoperatively (8.4% vs 10.2%, P = .01), less likely to have experienced a previous PICU admission (3.4% vs 4.0%, P = .04), and less likely to have received vasoactive medications within 24 hours after admission (16.0% vs 17.1%, P = .04). There was no difference between weekend and weekday admissions regarding the severity of illness (mean PRISM III scores of 5.7 vs 5.9, P = .12).

Patients admitted during the evening (n = 12 746, 62%), compared with those admitted during the daytime (n = 7801, 38%), were more likely to be male (57.8% vs 56.4%, P = .04) and were older (6.1 vs 5.6 years, P < .01) but were less likely to have experienced a previous PICU admission (3.0% vs 5.2%, P < .01), received vasoactive medications within 24 hours after admission (15.1% vs 19.5%, P < .01), or received mechanical ventilation within 24 hours after admission (34.5% vs 38.4%, P < .01). Patients admitted during the evening demonstrated a lower mean PRISM III score, compared with those admitted in the daytime (mean PRISM III scores of 5.5 vs 6.3, P < .01) (Table 1).

Table 1 demonstrates that the unadjusted overall PICU mortality rates were not statistically different for patients admitted on weekends and patients admitted on weekdays (4.71% vs 4.71%, P = .99). Similarly, there was no statistical difference in the rates of death within 48 hours after admission between the groups (2.28% vs 2.17%, P = .61). As shown in Table 2, the primary diagnoses for the weekend and weekday groups were trauma (20% vs 16%, P < .01) and asthma (10.0% vs 8.8%, P = .05). Overall, there were proportionately more admissions because of trauma, asthma, seizures, and drug overdoses on weekends. On weekdays, proportionately more patients were admitted with the diagnoses of cancer, congenital cardiovascular disease, acquired cardiovascular disease, and sepsis.

In a comparison of evening admissions and daytime admissions, the unadjusted overall PICU mortality rates (4.59% vs 4.91%, P = .29) and rates of death within 48 hours (2.24% vs 2.15%, P = .67) were not statistically different. As shown in Table 2, the most common diagnoses for evening and daytime admissions were trauma (20% vs 12%, P < .01) and pneumonia/bronchiolitis (13% vs 15%, P < .01). In addition, there were proportionately more admissions because of asthma and congenital cardiovascular disease during the daytime, compared with the evening, but proportionately more admissions because of seizures and drug overdoses during the evening, compared with the daytime.

Multivariate Analyses

Mixed-effects, multivariate, hierarchical, logistic regression was used to adjust for age, gender, a quadratic transformation of the logarithm of the PRISM III probability of death, previous PICU admission, postoperative care status, patient transfer status, and the requirement for vasoactive medications or ventilation within 24 hours after admission. The odds of death within 48 hours after admission for patients admitted on weekends was not higher than that for patients admitted on weekdays, as shown in Table 3 (OR: 1.00; 95% CI: 0.78–1.28). Furthermore, in the subgroup analyses there was no diagnosis with a different odds of death when weekend admissions were compared with weekday admissions (Table 4).

However, similar analyses demonstrated that pediatric patients admitted during evening hours had a higher odds of death than did patients admitted during the daytime hours (OR: 1.28; 95% CI: 1.00–1.62), after adjustment for the severity of illness and other covariates (Table 3). In the subgroup analyses (Table 4), we found higher odds of death among patients admitted with shock (OR: 4.09; 95% CI: 1.65–10.1), with congenital cardiovascular disease (OR: 3.90; 95% CI: 1.37–11.1), or after cardiac arrest (OR: 1.80; 95% CI: 1.04–3.13). Other diagnosis-specific analyses did not demonstrate significantly different mortality rates when evening admissions were compared with daytime admissions.