Perioperative mortality in pediatric patients: a systematic review of risk assessment tools for use in the preoperative setting

There are multiple preoperative risk scores for pediatric mortality. The aim of this study was to systematically describe and compare the existing studies of patient-specific multispecialty risk prediction scores for perioperative mortality in pediatric populations, with the goal of guiding clinicians on which may be most appropriate for use in the preoperative setting.

We conducted a systematic literature review of published journal articles that presented the development, extension/updating, and/or validation of a risk score which predicted all-cause mortality (up to 30 days postoperatively) in pediatric patients undergoing a procedure in which anesthesia was used. Scores needed to be applicable to surgeries in more than one non-cardiac surgical specialty and had to be able to be calculated by the anesthesiologist at the time of the pre-anesthetic assessment. Two investigators independently screened studies for inclusion and assessed study quality in the domains of clinical applicability, feasibility/ease of use in the clinical setting, and risk of bias.

A total of 1,681 titles were retrieved. Of these, 10 studies met inclusion criteria: nine reported the development and validation of scores, and one was an external validation of an existing score. Seven studies used varying years of multicenter data from the National Surgical Quality Improvement Program – Pediatric Participant Use File for development and/or validation. The unadjusted rate of mortality in the studies ranged from 0.3% to 3.6%. The preoperative predictors of mortality used in score development included patient demographics, preoperative therapies, and chronic conditions, among others. All models showed good discrimination upon validation (AUC > 0.8). Most risk scores had high or unclear risks of bias.

There are numerous scores available for the prediction of mortality in pediatric populations, all of which exhibited good performance. However, many have high or unclear risks of bias, and most have not undergone external validation.