Clinical data were grouped into 1-hour blocks for 24 hours ending at the event for case patients or at the end of 12 hours of data collection for control patients. Where there were missing data, the most recent recorded data were used, consistent with the approach used for other scores in critically ill children [21 (link)]. The greatest subscore for each item within each hour was identified and used to calculate the Bedside PEWS score for that hour. We then calculated the maximum PEWS score for the 12 hours ending 1 hour before the clinical deterioration event and in the six 4-hour blocks preceding ICU admission in patients urgently admitted to the ICU.
The primary analysis evaluated the hypothesis that the Bedside PEWS score can identify children at risk for cardiopulmonary arrest with at least one hour's notice. Logistic regression was used to compare the maximum Bedside PEWS score of case and control patients using all 12 hours of data in control patients and the 12 hours of data ending 1 hour before either urgent ICU admission or a code blue event in the case patients. The AUCROC curve was determined from the c-statistic calculated by logistic regression, and the 95% confidence interval (95% CI) was calculated using an accepted algorithm [22 ]. The ROC curve was represented graphically, and the sensitivity and specificity of the score at thresholds of 7 and 8 were calculated based on our previous work [6 (link)].
Repeated measures linear regression was used to evaluate the temporal evolution of scores preceding urgent ICU admission and code blue events in case patients. The dependent variable was the maximum Bedside PEWS score for each of the six 4-hour time periods preceding the clinical deterioration event. The independent variable was the midpoint of the time interval expressed in hours from the time of ICU admission. Linear regression was used to evaluate the relationships between the maximum Bedside PEWS score and the number of risk factors for cardiac arrest. Separate analyses were performed for case and control patients.
The association between the retrospective rating of nurses and the case or control status of patients was evaluated using logistic regression. We used clinical data from the 12 hours ending 1 hour before the clinical deterioration event and for 12 hours in control patients to calculate the maximum Bedside PEWS score. These data were paired with corresponding survey data from frontline nurses. When more than one nurse was surveyed in this time period, we used the data from the nurse who had last cared for the patient. The responses of the frontline nurses were represented on a numerical scale from 1 to 5. We tabulated the maximum Bedside PEWS score for each level of nurse rating in case and control patients. Logistic regression was used to evaluate the performance of nurse rating, the Bedside PEWS score, and the nurse rating with the Bedside PEWS score. We used the c-statistic as a measure of the AUCROC curve and calculated the 95% CI. Comparison of the AUCROC curve for the nurse rating and the maximum Bedside PEWS score was carried out as described by DeLong et al. [23 (link)].
Subgroup analyses described score performance in the following patient categories: urgent ICU patients, code blue patients, those who fell within any of the five age categories of the Bedside PEWS score, across institutions, patients with chronic conditions (bone marrow or organ transplantation, cardiac disease, severe cerebral palsy), patients with medical devices that might have place them at increased risk (tracheostomy, enterostomy feeding device, home oxygen), patients with acute illness (diabetic ketoacidosis, seizures), patients whose conditions had increased complexity (> 3 services involved in care, > 10 medications), patients with an administrative risk (recent transfer of primary service, ICU transfer, postoperative, off-service patient), and patients who had cardiopulmonary arrest. Power calculations based on our previous work suggested that differences between means could be shown with 30 patients per group. Given that our objectives were to evaluate score performance within specified subgroups and at each hospital, we sought to maximise the numbers of cases and controls from participating hospitals. Numbers were thus determined by the duration of the study at each hospital. The protocol was reviewed and approved by the research ethics boards at participating hospitals. All research ethics boards required consent for staff participation in the surveys and waived the need for patient consent.
The primary analysis evaluated the hypothesis that the Bedside PEWS score can identify children at risk for cardiopulmonary arrest with at least one hour's notice. Logistic regression was used to compare the maximum Bedside PEWS score of case and control patients using all 12 hours of data in control patients and the 12 hours of data ending 1 hour before either urgent ICU admission or a code blue event in the case patients. The AUCROC curve was determined from the c-statistic calculated by logistic regression, and the 95% confidence interval (95% CI) was calculated using an accepted algorithm [22 ]. The ROC curve was represented graphically, and the sensitivity and specificity of the score at thresholds of 7 and 8 were calculated based on our previous work [6 (link)].
Repeated measures linear regression was used to evaluate the temporal evolution of scores preceding urgent ICU admission and code blue events in case patients. The dependent variable was the maximum Bedside PEWS score for each of the six 4-hour time periods preceding the clinical deterioration event. The independent variable was the midpoint of the time interval expressed in hours from the time of ICU admission. Linear regression was used to evaluate the relationships between the maximum Bedside PEWS score and the number of risk factors for cardiac arrest. Separate analyses were performed for case and control patients.
The association between the retrospective rating of nurses and the case or control status of patients was evaluated using logistic regression. We used clinical data from the 12 hours ending 1 hour before the clinical deterioration event and for 12 hours in control patients to calculate the maximum Bedside PEWS score. These data were paired with corresponding survey data from frontline nurses. When more than one nurse was surveyed in this time period, we used the data from the nurse who had last cared for the patient. The responses of the frontline nurses were represented on a numerical scale from 1 to 5. We tabulated the maximum Bedside PEWS score for each level of nurse rating in case and control patients. Logistic regression was used to evaluate the performance of nurse rating, the Bedside PEWS score, and the nurse rating with the Bedside PEWS score. We used the c-statistic as a measure of the AUCROC curve and calculated the 95% CI. Comparison of the AUCROC curve for the nurse rating and the maximum Bedside PEWS score was carried out as described by DeLong et al. [23 (link)].
Subgroup analyses described score performance in the following patient categories: urgent ICU patients, code blue patients, those who fell within any of the five age categories of the Bedside PEWS score, across institutions, patients with chronic conditions (bone marrow or organ transplantation, cardiac disease, severe cerebral palsy), patients with medical devices that might have place them at increased risk (tracheostomy, enterostomy feeding device, home oxygen), patients with acute illness (diabetic ketoacidosis, seizures), patients whose conditions had increased complexity (> 3 services involved in care, > 10 medications), patients with an administrative risk (recent transfer of primary service, ICU transfer, postoperative, off-service patient), and patients who had cardiopulmonary arrest. Power calculations based on our previous work suggested that differences between means could be shown with 30 patients per group. Given that our objectives were to evaluate score performance within specified subgroups and at each hospital, we sought to maximise the numbers of cases and controls from participating hospitals. Numbers were thus determined by the duration of the study at each hospital. The protocol was reviewed and approved by the research ethics boards at participating hospitals. All research ethics boards required consent for staff participation in the surveys and waived the need for patient consent.
Free full text: Click here
