Primary content

Mining routine radiation oncology clinical datasets within a distributed rapid learning framework: t

Authors

D Thwaites (Lead)

A Dekker

J van Soest

L Holloway

M Bailey

A Miller

S Vinod

G Delaney

M Carolan 

G Goozee

A Ghose

F Hegi-Johnson

D Stirling

S Greenham

D Fraser

R Alvandi

K Foo

Aims

To share and combine routinely-available information from multiple NSW cancer centres in a data-protected manner; to use bioinformatics tools for data-mining and modelling to provide meaningful decision support systems (DSS);  and to integrate  radiomics (imaging-based) and clinical patient features. To demonstrate proof-of-principle using datasets for non-small cell lung cancer (NSCLC) patients receiving radiotherapy.

Method

Treatment, imaging and outcome data for lung patients from one NSW and one European radiotherapy centre have been extracted, standardized and linked, using Semantic Web based tools; keeping each centre’s data secure within its own protected systems. Outcome prediction models have been learned and validated and a model-based DSS produced to predict whether a specific patient may benefit from radical radiotherapy. An extended distributed learning network is currently being developed to include data from 6 more NSW cancer centres.

Results

The single-centre (Liverpool/Macarthur CTC) pilot study initially identified 4000 lung patient datasets,  modelling those meeting eligibility criteria, including having complete data.  The resulting model-based DSS provided new knowledge, capable of helping clinical decisions where standard treatment guidelines are less strong.  Specifically it predicted good prognosis sub-groups in both palliatively and curatively treated patients, enabling prospective DSS use to reconsider palliative treatment decisions for patients where a good prognosis is predicted. The available lung patient dataset numbers will increase to 20,000 in the extended collaborative network, providing stronger models and DSS and widening their use.

Implications

The work will support optimized clinical decisions between radical or palliative radiotherapy for treatment of advanced stage lung cancer patients; and for when to use complex treatment methods. The approaches are generalizable to other cancers and treatment modalities.