Improving mortality prediction using acid/base parameters in PIM2

Abstract

Aims & Objectives: Paediatric Index of Mortality 2 (PIM2) is used as a mortality predictor in PICU and includes absolute base excess (BE) along with other clinical and physiological variables. Small scale studies have shown that lactate is a better predictor than base excess (BE). We investigated the effect on PIM2 of a) utilising venous samples, which have historically been excluded, b) adding lactate to the model, with or without BE, c) utilising separate terms for positive and negative BE, and d) the treatment of missing values. Methods Data were abstracted from the UK and Republic of Ireland Paediatric Intensive Care Audit Network (PICANet) and the Australian and New Zealand Paediatric Intensive Care Registry (ANZPICR). All admissions between 1st January 2012 and 31st December 2015 were analysed (N= 123,252) using a series of logistic regression models. Area under the Receiver Operating Characteristic (ROC) Curve (AUROC) and the Akaike Information Criterion (AIC) were used to assess model fit. Results Addition of venous BE measurements improved the fit of PIM2 (AUROC improved by 0.0016). Addition of positive and negative BE as two separate terms didn't improve the model fit, but showed that negative BE is a stronger mortality predictor. A model with missing lactate values set to 1 gave a better fit than replacing with 0. Table shows results of the different models. Conclusions Lactate improves the fit of PIM2. Addition of two variables for BE, inclusion of lactate, and using only lactate measurements if both lactate and BE are obtained should be considered.L6238154752018-09-13 <br />