Over the weekend of 6-8 July 2018, a team from UNSW, in collaboration with local clinicians and data scientists, participated in the National University of Singapore (NUS) - National University Health System (NUHS) - Massachusetts Institute of Technology (MIT) Healthcare Artificial Intelligence (AI) Datathon, held at NUS. More than 200 participants, in 23 teams, competed in three tracks (critical care data, surgical EMR data and imaging data).

The joint UNSW-NUS team won 1^st prize (worth over AU$4000 in cash and Google cloud computing credits) in the critical care track in which eight Singaporean and international teams competed. Using the eICU Collaborative Research Database, a large multi-centre critical care database made available by Philips Healthcare in partnership with the MIT Laboratory for Computational Physiology, they investigated whether the importance of prognostic factors varied in the course of each patient’s stay in ICU. Currently-used predictive models assume that they don’t vary. Using cutting-edge machine learning models trained on millions of rows of patient demographic, physiological, treatment and laboratory data, they found that the importance of prognostic factors in predicting mortality does indeed change over the course of a stay in ICU. For example, abnormal mean arterial blood pressure assumes greater importance in predicting a poor outcome with each successive day of a stay in ICU. Based on these novel results, the team will continue to collaborate on the development of models for critical care outcome prediction which take into account such temporal variations in variable importance in various types of ICU patient.

The team comprised Oluwadamisola Sotade, Mark Hanly and Oisin Fitzgerald (UNSW Centre for Big Data Research in Health), Tim Churches (Ingham Institute for Applied Medical Research/UNSW South Western Sydney Clinical School), Peter Straka (UNSW Mathematics and Statistics), Stella Ang and Pak Ling Lui (Singapore National University Health System), Siqi Liu  (NUS Saw Swee Hock School of Public Health) and Peng Shen  (Tan Tock Seng Hospital)

Originally the brainchild of Silicon Valley, CA, hackathons have proven to be successful models for innovation in business settings and are typically organized as intense, short-duration, competitions in which teams generate innovative solutions. The hackathon model integrates collaboration, idea generation, and group learning by joining various stakeholders in a mutually supportive setting for a limited period of time.

For health data analysis, the goal of the hackathon is to assemble clinical experts, data scientists, statisticians, and those with domain-specific knowledge to create ideas and produce clinically relevant research that reduces or eliminates biases, relies on sound statistical methods and adequate data samples, and aims to produce replicable, scientifically valid results. The term “datathon” was coined as a portmanteau of data + hackathon, encapsulating the application of the hackathon model to health data analytics. For example, a critical-care datathon is an event in which participants are brought together to form interdisciplinary teams and answer research questions in the field of critical care (that is, the treatment of severely-ill patients in intensive care units).