Using machine learning to predict paediatric 30-day unplanned hospital readmissions: a case-control retrospective analysis of medical records, including written discharge documentation

Abstract

Objectives. To assess whether adding clinical information and written discharge documentation variables improves prediction of paediatric 30-day same-hospital unplanned readmission compared with predictions based on administrative information alone.Methods. A retrospective matched case-control study audited the medical records of patients discharged from a tertiary paediatric hospital in Western Australia (WA) between January 2010 and December 2014. A random selection of 470 patients with unplanned readmissions (out of 3330) were matched to 470 patients without readmissions based on age, sex, and principal diagnosis at the index admission. Prediction utility of three groups of variables (administrative, administrative and clinical, and administrative, clinical and written discharge documentation) were assessed using standard logistic regression and machine learning.Results. Inclusion of written discharge documentation variables significantly improved prediction of readmission compared with models that used only administrative and/or clinical variables in standard logistic regression analysis (x 2 17 ¼ 29.4, P ¼ 0.03). Highest prediction accuracy was obtained using a gradient boosted tree model (C-statistic ¼ 0.654), followed closely by random forest and elastic net modelling approaches. Variables highlighted as important for prediction included patients' social history (legal custody or patient was under the care of the Department for Child Protection), languages spoken other than English, completeness of nursing admission and discharge planning documentation, and timing of issuing discharge summary.Conclusions. The variables of significant social history, low English language proficiency, incomplete discharge documentation, and delay in issuing the discharge summary add value to prediction models.What is known about the topic? Despite written discharge documentation playing a critical role in the continuity of care for paediatric patients, limited research has examined its association with, and ability to predict, unplanned hospital readmissions. Machine learning approaches have been applied to various health conditions and demonstrated improved predictive accuracy. However, few published studies have used machine learning to predict paediatric readmissions.

Applicability of predictive models for 30-day unplanned hospital readmission risk in paediatrics: a systematic review

“…All studies were based on retrospective data, with 9 studies based on tertiary or paediatric hospital data, 22 55 61–67 and 19 studies based on centralised databases 31 53 54 56–60 68–78 . Four of 28 studies additionally included census data in the analysis.…”

“…10 of 28 studies developed or validated more than one predictive model for UHRs, 22 58 59 65–70 75 which were in part excluded due to non-agreement with the inclusion criteria. The models included were grouped into three health conditions: (1) all-cause UHR (n=13), 22 61 63–65 68 69 (2) surgical condition-related UHR (n=17) 31 53 54 56–60 67 70 73–75 77 78 and (3) general medical condition-related UHR (n=7) 55 62 66 71 72 76 . The 30-day UHR rates varies from 1.5% 53 to 41.2% 71 .…”

“…Among the 37 predictive models included, 32 (87%) used a development design 22 31 53–61 63–67 70–78 ; 3 (8%) used an external validation design 62 65 69 ; and 2 (5%) used an incremental value design 66 68 . All external validated models were based on existing predictive models that had been previously used in the adult population 65 69 or for different outcomes 62 .…”

“…This inclusion criterion enables us to identify significant risk factors and to apply the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) adherence form, which was originally developed for regression models. 20 This implies that predictive models using machine-learning (ML) techniques (eg, least absolute selection and shrinkage operator 21 or random forest 22 ) are excluded and coded as non-regression models. Studies that aimed to identify 30-day UHR predictors and did not provide a discrimination measure are classified as prognostic factor studies and are thus excluded from the analysis (so as not to bias them adversely in TRIPOD adherence).…”

“… x=risk factor (OR/hazard ratio>1). *The six predictive models of Zhou et al 22 are not included in this analysis due to missing information about ORs. See online supplemental table A6 in the online supplemental material for a list of included variables.…”