Decision tree-early warning scores based on common laboratory test results discriminate patients at risk of hospital mortality

Stuart Jarvis, Caroline Kovacs, Tessy Badriyah, Jim Briggs, Mohammed A. Mohammed, Paul Meredith, Paul Schmidt, Peter Featherstone, David Prytherch, Gary Smith

Research output: Contribution to conferencePoster

210 Downloads (Pure)

Abstract

We hypothesised that it might be possible to use decision tree (DT) analysis to build an early warning score (EWS) based exclusively on laboratory data to predict patients at risk of in-hospital death early in their hospital stay. Using an electronic database of 92354 combined routine haematology and biochemistry tests for adult patients for whom the hospital admission specialty was Medicine, we used DT analysis to generate a laboratory DT EWS (LDTEWS) for each gender. DT analysis is a data mining classification technique for building decision trees by recursively splitting or partitioning of datasets into homogenous groups. This partitioning is based on derived associations between the chosen outcome – in our case, in-hospital death – and one or more covariates. Our tree modelling strategy assessed the following covariates individually: haemoglobin, white cell count, serum urea, serum albumin, serum creatinine, serum sodium, and serum potassium results. LDTEWS was developed for a single set (n= 3762) (Q1) and validated in 22 other discrete sets each of three months long (Q2, Q3......Q23) (range of n = 3590 to 4341) by testing its ability to discriminate in-hospital death using the area under the receiver-operating characteristic (AUROC) curve. As expected, because of different reference ranges for laboratory tests for each gender, the data generated slightly different models for males and females. The area under the receiver-operating characteristic curve values (95% CI) for LDTEWS in all patients, irrespective of gender, with in-hospital death as the outcome, ranged from 0.748 (0.723 to 0.772) (Q10) to 0.797 (0.772 to 0.823) (Q9) for the 22 validation sets Q2-Q23. This study provides evidence that the results of commonly measured laboratory tests collected soon after hospital admission can be used in a simple, paper or computer-based early warning score (LDTEWS) to discriminate in-hospital mortality. We hypothesise that, with appropriate modification, it might be possible to extend the use of LDTEWS for use on an ongoing basis throughout the patient’s hospital stay.
Original languageEnglish
Publication statusPublished - 13 May 2013
EventRapid response systems and medical emergency teams - London
Duration: 13 May 201314 May 2013

Conference

ConferenceRapid response systems and medical emergency teams
CityLondon
Period13/05/1314/05/13

Fingerprint

Dive into the research topics of 'Decision tree-early warning scores based on common laboratory test results discriminate patients at risk of hospital mortality'. Together they form a unique fingerprint.

Cite this