Quantifying the impact of novel cancer treatments on Lung Cancer Control and Normal Tissue Complication Probability

Background: Lung Cancer (LC) has has seen little improvement in survival in the past several decades.

While Stereotactic ablative radiotherapy (SABR), proton radiotherapy and immunotherapy have individually shown promise, their relative impact on toxicity free patient survival is not yet well understood or quantified.

Ultimately, to probe the inner mechanisms of treatment impacts on tumour control probability (TCP) and normal tissue complication probability (NTCP), validated models must be built and analyzed in order to determine which treatment features can maximize clinical outcomes for patients. Aims: We aim to generate validated TCP/ NTCP models via machine learning (ML).

We subsequently aim to analyze predictive ML models to determine the relationship between treatment features (e.g. dose distribution, immunotherapy interval, radiotherapy particle) and TCP/NTCP to gain insight into which features are most predictive of toxicity free tumour control.

This project also aims to quantify the relative degree to which these features (and their interactions) have an impact on patient outcomes. Methods: We have arranged to acquire a dataset of 1607 LC patients treated with curative intent.

This dataset contains information on patient demographics (age, gender, etc.), diagnosis (tumour type, location, stage, etc.), treatment (radiotherapy dose distribution and particle, immunotherapy, concurrent chemotherapy, etc.). The dataset also contains patient follow-ups on incidence of various self-reported toxicities and follow-up CT scans.

We seek to fund clinician time to sift through these patients and determine progression free survival (PFS) to use as a definition for tumour control.

From this data we aim to identify and collate key patient dosimetric and non-dosimetric features for input into ML based TCP/NTCP models.

Lastly, we seek to determine which treatment features (and interaction effects between features) have the largest relative impacts on patient TCP.

How the results will be used: Our results could provide insight into which treatment characteristics most affect TCP/NTCP.

Some key questions that we wish to answer are 1) which treatment parameters in this large dataset are most predictive of TCP/NTCP? 2) Are there any important interactions between treatment features that have not noticed in previous studies? 3) Can we answer lingering questions that can help clinicians gain insight on how best to plan LC treatments (e.g. in cases where patients receive sequential immunotherapy following radiotherapy (RT), there is an open question about the the time interval between these two therapies as it affects progression free survival).

Therefore the results should help clinicians best optimize LC patient outcomes.