Nonlinear modelling for predicting patient presentation rates for mass gatherings

Paul Arbon, Murk Bottema, Kathryn Zeitz, Adam Lund, Sheila Turris, Olga Anikeeva, Malinda Steenkamp

Research output: Contribution to journalArticlepeer-review

7 Citations (Scopus)

Abstract

Introduction Mass gatherings are common in Australia. The interplay of variables, including crowd density and behavior, weather, and the consumption of alcohol and other drugs, can pose a unique set of challenges to attendees' well-being. On-site health services are available at most mass gatherings and reduce the strain on community health facilities. In order to efficiently plan and manage these services, it is important to be able to predict the number and type of presenting problems at mass gatherings.Problem There is a lack of reliable tools to predict patient presentations at mass gatherings. While a number of factors have been identified as having an influence on attendees' health, the exact contribution of these variables to patient load is poorly understood. Furthermore, predicting patient load at mass gatherings is an inherently nonlinear problem, due to the nonlinear relationships previously observed between patient presentations and many event characteristics.Methods Data were collected at 216 Australian mass gatherings and included event type, crowd demographics, and weather. Nonlinear models were constructed using regression trees. The full data set was used to construct each model and the model was then used to predict the response variable for each event. Nine-fold cross validation was used to estimate the error that may be expected when applying the model in practice.Results The mean training errors for total patient presentations were very high; however, the distribution of errors per event was highly skewed, with small errors for the majority of events and a few large errors for a small number of events with a high number of presentations. The error was five or less for 40% of events and 15 or less for 85% of events. The median error was 6.9 presentations per event.Conclusion: This study built on previous research by undertaking nonlinear modeling, which provides a more realistic representation of the interactions between event variables. The developed models were less useful for predicting patient presentation numbers for very large events; however, they were generally useful for more typical, smaller scale community events. Further research is required to confirm this conclusion and develop models suitable for very large international events.

Original languageEnglish
Pages (from-to)362-367
Number of pages6
JournalPrehospital and Disaster Medicine
Volume33
Issue number4
DOIs
Publication statusPublished - 2018

Fingerprint Dive into the research topics of 'Nonlinear modelling for predicting patient presentation rates for mass gatherings'. Together they form a unique fingerprint.

Cite this