JMIR Publications

ABSTRACT

Background:

Head and neck squamous cell carcinomas (HNSCC) cause considerable morbidity and mortality. Multimodal treatment strategies can cause significant toxicity, and therapy options are limited for recurrent disease. Im-munotherapy has emerged as a promising approach. However, patient response variability underscores the need for better predictive markers.

Objective:

This study aims to use artificial intelligence to develop two predictive models in HNSCC patients to assess 1) progression or recurrence following primary curative treatment and 2) long-term survival after immunotherapy in recurrent and metastatic disease. The study will also describe the characteristics of patients with early/locally advanced and recurrent/metastatic cancers.

Methods:

This is a retrospective, observational study of data captured in electronic health records (EHRs) from participat-ing hospitals between January 1st, 2014, and December 31st, 2021. The study population comprises adults diagnosed with HNSCC at any stage. Study variables including demographics, comorbidities, clinical variables, treatments, and outcomes will be extracted using EHRead®, a technology that applies natural language pro-cessing and machine learning to extract and analyse structured and unstructured clinical information in de-identified EHRs. Predictive models based on dynamic risk stratification for treatment response and recur-rence will be developed using multivariable logistic regressions, decision tree classifiers, and random forest approaches. Descriptive and outcome analyses will be shown for different anatomic subsites and stratified by stage and treatment.

Results:

The study began enrolling sites in July 2021 and is currently ongoing.

Conclusions:

Development of predictive models using artificial intelligence will advance clinical understanding of HNSCC to improve patient outcomes. Clinical Trial: NCT05117775