BD.20.006 – Developing automated anomaly detection in healthcare (Acronym: deTECT)

Route: Creating Value through responsible access to and use of big data

Cluster question: 105 How can Big Data and technological innovation (e-health) contribute to health care?

The aim of the deTECT project is to use (big) data science methodology to identify unusual diseases or events to enable timely detection of patients afflicted with less common problems. Subsequent adjustment of their diagnostic and treatment pathways will improve such patient life quality, minimize their healthcare costs, and increase their societal participation. For that purpose, deTECT will 1. combine information from hospital, general practitioner and public health databases, while observing privacy and other regulations, to enable detection of anomalous patterns using current as well as historic patient data; 2. tackle the false-positive paradox with state-of-the-art data science techniques on these combined data, 3. combine multi-domain expertise and involve stakeholders, including patients and their carers, to design an alert prototype for electronic health record systems. The work will be performed on the basis of three use-cases with a declining a priori chance of being different from the majority from relatively high (frailty in the elderly), to occasional (severe infection as adverse event of biologicals), to rare (primary antibody deficiency as typical example of a difficult-to-recognize rare disease). Preliminary studies have already been executed for these three use-cases. Leveraging on the combined expertise from the data science as well as medical domains, deTECT will address this complicated problem combining insights from both fields, such as model selection on optimal false-positive instead of false-negative rates, incorporating accumulating data in health records with elapsing time, oversampling the minority class while undersampling the majority class, and labeling true-cases by using medical as well as non-medical information obtained from all stakeholders including the patients themselves. With that, deTECT will deliver a prototype solution that can be further developed into applications that communicate with or are integrated in electronic health record systems while special attention will be paid to user-friendliness and feasibility for everyday use.

Keywords

adverse events of biologicals, anomaly detection, combining databases, frailty in the elderly, healthcare, searching for patterns, timely detection of rare disease

Other organisations

CentERdata, Jeroen Bosch Ziekenhuis, NIVEL

Submitter

Organisation Tilburg University (TiU)
Name Prof. dr. E. de Vries
E-mail e.devries@tilburguniversity.edu
Website https://www.tilburguniversity.edu/staff/e-devries