JMIR Publications

ABSTRACT

Background:

Precision oncology has the potential to leverage clinical and genomic data in advancing disease prevention, diagnose, and treatments. A key research area focuses on early detection of primary cancers and the potential prediction of cancers of unknown primary in order to facilitate optimal treatment decisions.

Objective:

This study presents a methodology to harmonize phenotypic and genetic data features to classify primary cancer types and predict unknown primaries.

Methods:

We extracted the genetic data elements from a collection of oncology genetic reports of 1,011 cancer patients, and corresponding phenotypical data from the Mayo Clinic electronic health records (EHRs). We modeled both genetic and EHR data with HL7 Fast Healthcare Interoperability Resources (FHIR). The semantic web Resource Description Framework (RDF) was employed to generate the network-based data representation (i.e., patient-phenotypic-genetic network). Based on RDF data graph, graph embedding algorithm Node2vec was applied to generate features, and then multiple machine learning and deep learning backbone models were adopted for cancer prediction.

Results:

With six machine-learning tasks designed in the experiment, we demonstrated the proposed method achieved favorable results in classifying primary cancer types and predicting unknown primaries. To demonstrate the interpretability, phenotypic and genetic features that contributed the most to the prediction of each cancer were identified and validated based on a literature review.

Conclusions:

Accurate prediction of cancer types can be achieved with existing EHR data with satisfactory precision. The integration of genetic reports improves prediction, illustrating the translational values of incorporating genetic tests early at the diagnose stage for cancer patients.