'The Score Matters': wide variations in predictive performance of 18 paediatric track and trigger systems

To compare the predictive performance of 18 paediatric early warning systems (PEWS) in predicting critical deterioration.

Retrospective case-controlled study. PEWS values were calculated from existing clinical data, and the area under the receiver operator characteristic curve (AUROC) compared.

UK tertiary referral children's hospital.

Patients without a 'do not attempt resuscitation' order admitted between 1 January 2011 and 31 December 2012. All patients on paediatric wards who suffered a critical deterioration event were designated 'cases' and matched with a control closest in age who was present on the same ward at the same time.

Respiratory and/or cardiac arrest, unplanned transfer to paediatric intensive care and/or unexpected death.

12 'scoring' and 6 'trigger' systems were suitable for comparative analysis. 297 case events in 224 patients were available for analysis. 244 control patients were identified for the 311 events. Three PEWS demonstrated better overall predictive performance with an AUROC of 0.87 or greater. Comparing each system with the highest performing PEWS with Bonferroni's correction for multiple comparisons resulted in statistically significant differences for 13 systems. Trigger systems performed worse than scoring systems, occupying the six lowest places in the AUROC rankings.

There is considerable variation in the performance of published PEWS, and as such the choice of PEWS has the potential to be clinically important. Trigger-based systems performed poorly overall, but it remains unclear what factors determine optimum performance. More complex systems did not necessarily demonstrate improved performance.