Project abstract (1000 Characters): Precision medicine is expected to improve healthcare through the integration and personalized analysis of clinical and genomic data. However, new approaches are needed to transform the current landscape of fragmented EHR data and complicated genetic results. Mobile health technologies that leverage FHIR and artificial intelligence can be a key part of the solution.
We created and released an iPhone app that searches all of a patient’s aggregated health records pulled via Apple’s FHIR API, finds and prioritizes (via a machine learning algorithm trained on the NIH Genetic Testing Registry) the most relevant gene-related laboratory tests for precision medicine, and displays results as a simple “snapshot” to improve doctor-patient communication of precision medicine results.
Project rationale, impact and innovation (3500 Characters): Precision Medicine is expected to improve healthcare through the integration and personalized analysis of clinical and genomic data. A patient’s Electronic Health Record (EHR) contains many different laboratory test results, including some about particular genes relevant to patient health. However, many patients currently have clinical and genetic data fragmented across many different EHR systems, requiring novel approaches to integrating, organizing, and analyzing this data. Moreover, there are estimated to be roughly 20,000 to 40,000 genes in the human genome, with many genes affecting patient health in different ways. As a result, manually searching ALL health records for useful gene-related laboratory results would be very time-consuming for patients or their healthcare providers.
Apple recently created and released a FHIR-based Health Records framework to encourage development of smartphone solutions that integrate fragmented EHR data. Furthermore, the increased availability of 1) public healthcare-related datasets and 2) new machine learning frameworks have catalyzed the development of AI-based tools which could help make widespread precision medicine a practical reality.
Leveraging the above innovations, we created and released an iPhone app available on the App Store (“Precision Medicine Genes + AI”) that automates the search process for gene-related laboratory test results. In a nutshell, our iPhone app is an informatics tool that searches all of a patient’s aggregated Health Records pulled via Apple’s FHIR API, finds and prioritizes (via a machine learning algorithm trained on the NIH Genetic Testing Registry [GTR]) the most relevant gene-related laboratory tests for precision medicine, and displays results as a clean, simple “snapshot” which can be used to improve doctor-patient communication of precision medicine results.
Project design and implementation (7000 characters): Informatics Filter for Genetic Lab Results
The Logical Observation Identifiers Names and Codes (LOINC) database contains a comprehensive collection of standardized codes used to describe laboratory results stored in the electronic health record (https://loinc.org/). We downloaded a recent version of the LOINC database (version 2.65), and developed an informatics pipeline (in Python) that used natural language processing to search the LOINC database and identify laboratory codes dealing with clinical results involving specific genes. The results of the final informatics filter were then verified using the Human Genome Organisation (HUGO) Gene Nomenclature Committee (HGNC) database (https://www.genenames.org/).
Machine Learning Model Development
We created an informatics pipeline in Python that applied natural language processing to identify clinically-relevant genes from a recent download of the NIH Genetic Testing Registry, a comprehensive database of genetic testing information supported by organizations throughout the world (https://www.ncbi.nlm.nih.gov/gtr/). The informatics pipeline helped build the labeled dataset used to train our model, which was developed using Apple’s Turi Create (version 4.3.2) machine learning framework (https://github.com/apple/turicreate).
iPhone App using HealthKit and FHIR Health Records
Our iPhone app was designed to use Apple’s HealthKit and FHIR-based Health Record APIs to handle the diverse clinical and genetic data that can be aggregated and integrated into a patient’s iPhone. It does this by leveraging Apple’s APIs with the 1) lab results informatics filter and 2) NIH GTR-based machine learning model described above to streamline the health record search/prioritization process for gene-related laboratory results for precision medicine.
We addressed patient privacy by using Apple’s HealthKit functionality to allow patients to enable/disable access and control the frequency of Health Record data access, which only occurs inside the app and is not shared with any third party.
The app workflow is as follows: after the user/patient downloads and opens the app on their iPhone, the app searches all of a patient’s aggregated Health Records pulled via Apple’s FHIR API, finds and prioritizes the most relevant gene-related laboratory test results as they relate to precision medicine (using AI/ML to analyze the content of each lab record and calculate a clinical relevance score), and displays results as a clean, simple “snapshot” which can be used to improve doctor-patient communication of precision medicine results.
The app was built in Swift Version 4 using XCode 10, and with the HealthKit API enabled along with the FHIR-based Health Records feature activated.
Project evaluation and sustainability (3500 characters): We recently published a paper that assessed the readiness of precision medicine interoperability using the NIH Genetic Testing Registry (https://www.ncbi.nlm.nih.gov/pubmed/29334348). This paper served as a guide for both developing the app and the machine learning model used by the app to prioritize laboratory results.
Twitter project summary (140 characters): Precision medicine iPhone app w/AI to organize EHR genetic labs: https://itunes.apple.com/us/app/precision-medicine-genes-ai/id1451470406
How is FHIR used in the App being demonstrated (500 characters)? : Our submission is an end-user facing iPhone app currently available on Apple’s App Store.
1. What FHIR release does your application use? (500 characters)?: Our app leverages the Apple Health Records FHIR API that was recently integrated into Apple’s HealthKit API, which (per Apple documentation) leverages FHIR DSTU 2 (v1.0.2) and the Argonaut Data Query Implementation Guide 1.0.0 (https://developer.apple.com/documentation/healthkit/samples/accessing_health_records).
What is the data source for the FHIR resources and how are the FHIR resources accessed? (500 characters): Our app leverages the Apple Health Records FHIR API that was recently integrated into Apple’s HealthKit API, which (per Apple documentation) leverages FHIR DSTU 2 (v1.0.2) and the Argonaut Data Query Implementation Guide 1.0.0 (https://developer.apple.com/documentation/healthkit/samples/accessing_health_records).
Jay Ronquillo (Presenter)