Longitudinal Spatial-Nonspatial Decision Support for Competing Outcomes in Head and Neck Cancer Therapy

Project Summary/Abstract Head and neck cancer (HNC) patients survive years after oncologic therapy due to increased efficacy of therapy, increased incidences of human papilloma virus related HNC, and decreased numbers of smoking and tobacco related tumors.

However, the majority of patients are plagued with long lasting or permanent residual effects, whose severity, rate of development and resolution after treatment vary largely between survivors.

At the same time, patient reported outcomes (PROs) offer important information that could be critical for the efficient detection and resolution of long term effects.

However, the interpretation of PRO repositories is plagued by data and analysis issues which so far have prevented their practical use in clinical care, including missing or incomplete data, co-occurence of multiple symptoms, variability across populations and across time, and, in the case of HNC and other spatially- dependent cancers, further symptom dependency on the anatomical location of the tumors and their proximity to organs at risk.

We propose to develop validated, patient-specific models to interpret HNC PROs in order to inform individual treatment and care decisions for patients.

Our data science approach circumvents limitations in the state of the art by accounting longitudinally for PRO symptom clusters and their dynamics over time, while handling incomplete data, by incorporating patient- specific bioimaging markers and spatial dose data pre- and during therapy, by calibrating for inter-patient variability, and by predicting symptom development and computing clinical action signals for a new patient based on cohorts of similar patients.

From a clinical perspective, our integrative data science approach is novel in the field of cancer therapy, through its leveraging of existing patient repositories and similar cohorts, its symptom- cluster analytics, and its integration of heterogeneous data sources, including patient reported outcomes and quantitative bioimaging data.

The resulting methodology will mark a significant advance in biomedical computing because it will be able to identify early specific patients who are at risk for long lasting or permanent treatment-induced residual effects, and will thus enable clinicians to adapt care to the individual patient level.

The proposed supplement application extends the methodological approach of the parent award by incorporating radiation dosimetry data, undertaken within a career development plan designed to enhance and accelerate the capacity of the applicant, who is from a background underrepresented in biomedical sciences, transition to mentored and independent investigator status, thus enhancing cancer research workforce diversity under this program.