Project abstract (1000 Characters): The confluence of increasing medical data and easier access to such information has led to "data overload" for busy clinicians. Graphical data visualization is broadly employed across industries and disciplines as a tool for aiding interpretation of voluminous, complex data. However, electronic health record (EHR) vendors have been slow to incorporate meaningful data visualization tools into their products. Our project sought to address the shortcomings of current, vendor-supplied data visualization tools by providing clinicians with a timeline-based, customizable data visualization that incorporates multiple types of clinical data. We created a SMART on FHIR app called MedTimeLine to display a timeline-based, customizable data visualization that incorporates multiple types of clinical data. Several technical challenges emerged during the development process, but our team was able to address these, implement the app, and now is assessing impact.

Project rationale, impact and innovation (3500 Characters): The current era of medicine is being defined by ever-increasing data generation and advanced information systems placing raw data streams at clinicians' fingertips. The confluence of these two trends has led to "data overload" for busy clinicians taking care of patients in increasingly complex, multi-disciplinary care settings. Graphical data visualization is broadly employed across industries and disciplines as a tool for aiding interpretation of voluminous, complex data. However, electronic health record (EHR) vendors have been slow to incorporate meaningful data visualization tools into their products. Stock visualization tools from vendors often fail to incorporate widely accepted principles of good data presentation, locate data in disparate areas making correlations between different data streams difficult, and do not allow for customization of views. Often clinicians may have to look at multiple different screens and write data points on paper to correlate different types of data. Consequently, any clinician dealing with more than a handful of clinical data points suffers from cognitive overload when trying to interpret patient data.
Our project sought to address the shortcomings of current, vendor-supplied data visualization tools by providing clinicians with a timeline-based, customizable data visualization that incorporates multiple types of clinical data. We focused our efforts on infectious diseases clinicians as representative of many different types of clinicians that regularly deal with complex clinical data. The immediate impact of this work is that it reduces the cognitive burden for day-to-day clinical care, allowing clinicians to focus on their patients. Ultimately, our goal is to spur innovation in clinical data visualization and demonstrate portability of such tools across institutions through use of the SMART on FHIR app paradigm.

Project design and implementation (7000 characters): Asim Ahmed, an infectious disease clinician at Boston Children's Hospital (BCH), realized for complex patients clinicians often made hand-drawn patient care timelines, plotting their patients' progress over time along multiple data streams spanning sections across the chart, including vital signs, labs, microbiology results, procedures, and medications. Asim submitted his idea to make an EHR application to automatically and electronically plot this timeline to the Innovation and Digital Health Accelerator (IDHA) program at BCH. Through a partnership with Verily Life Sciences, we designed, tested, and implemented the application as a SMART on FHIR app within BCH’s instance of Cerner PowerChart.
To inform our initial design, we shadowed clinicians as they reviewed patient charts for clinical care, attended conference presentations of infectious disease cases, ran through paper prototypes with several clinicians, and studied clinicians’ own hand-drawn timelines. Over the next nine months, we worked through iterative, participatory design for each timeline type and interaction paradigm, doing informal usability testing as we implemented various components with clinicians and laypeople along the way.
We implemented our project's frontend using Angular and chart.js and embedded it into BCH's Cerner PowerChart as a SMART on FHIR app. It retrieves data using FHIR API calls and renders it into interactive graphs. Our key requirements were patient safety and clinician usability. Patient safety is our top priority. We worked to eliminate any ambiguities in the app that could lead to faulty assumptions by the user, addressing details as specific as the number of decimal places on a graph axis. We also implemented several safety-critical paradigms such as always defaulting to flag data points as abnormal when its normality status is unclear or defaulting to show an error in the place of a graph if an unexpected LOINC code mapping is returned from a FHIR call. We assessed our code with hundreds of unit tests and hundreds of automated and manual integration tests which required a 100% pass rate before product release. We also provided a user Instructions for Use and a technical administrator Instructions for Use to ensure clear expectations about what the product was and was not capable of.
For our second requirement, usability, we particularly targeted the attributes of learnability, predictability, and speed. To this end, we maintained consistency throughout the app, utilizing standardized legends, colors, and mouse events. We designed the system with accessibility in mind, ensuring that color was not the only distinguishing factor between elements in the legends. To promote speed of use, we provide clinicians with intelligent presets such as automatically loading in the last patient encounter and the data streams that are useful in most cases.
We encountered several technical challenges in our implementation. Our primary challenges stem from the differences in the state of the FHIR standard and the Cerner Millennium implementation of the DSTU2 standard. One key resource for infectious disease clinicians is DiagnosticReports, which should hold the results of microbiology tests that they use to determine which infections the patients suffer from. Unfortunately, Cerner currently only supports returning DiagnosticReports that are radiology reports, compelling us to develop our own implementation of microbiology. The IDHA team implemented their own FHIR server that retrieved microbiology data from the Cerner data store and returned it as DiagnosticReports. Another challenge arose from medication-related resources: MedicationOrder and MedicationAdministration. We wanted to show all MedicationAdministrations of specific antibiotics within a given time frame. In the Cerner implementation, you cannot search for a MedicationAdministration by a specific medication code or by MedicationOrder ID. So, we searched for all MedicationAdministrations within a specific date range, and then, in our frontend, filtered by RxNorm code. Additionally, we wanted to show which MedicationAdministrations belong to which MedicationOrder. Because you cannot search MedicationAdministrations by order and there may not be a date associated with the MedicationOrder, we need to get all MedicationAdministrations for a patient then filter down by order. This excessive number of calls to FHIR (due to paging) caused our application to take major performance hits. To overcome this, we implemented a few caching strategies to cut down on the number of FHIR calls necessary after the initial page load.
Another set of challenges revolved around integrating the application into the EHR. Because there is no debugging console available in the EHR embedded browser, and it was unclear what version of Internet Explorer was running in the browser, we had to take a guess-and-check approach to selecting the correct polyfills. The Cerner sandboxes did not have patients with enough data coverage to measure our functionality, so another challenge was generating and recording enough test data to ensure accurate functionality within the medical record. To that end, two informaticians created five test patients with a wide range of values that we could test against, both as flat files we could load in with our application running externally, and in the development environment of PowerChart within BCH.

Project evaluation and sustainability (3500 characters): We have already spent time with several pilot users to gather their initial, informal feedback on the application. We've used that to prioritize future features and to generate a list of bugs to fix.

We have an evaluation plan approved by the Boston Children's Hospital IRB to test the application for usability through asking clinicians to complete several tasks using MedTimeLine, then collecting information such as: answers to usability questionnaires (such as the System Usability Scale, the NASA Task Load Index), time to task completion, and verbal feedback about several usability aspects of the system. We plan to run that evaluation for 5-10 participants (including medical students, attending physicians, and fellows) ahead of full-system release in October.

After the full-site release, we plan to conduct a comparative study, asking users to complete a series of tasks in standard PowerChart and also in MedTimeLine. We plan to measure similar variables as in the study above, in addition to using eyetracking to better understand how users interact with both systems. We are submitting an IRB amendment to cover that work.

We have released our product as open source, so the SMART on FHIR community can contribute to its maintenance. Additionally, the IDHA at Boston Children's Hospital as well as the Information Systems Department has committed to maintain their existing deployment and potentially add new features based on clinician input.

Twitter project summary (140 characters): MedTimeLine allows clinicians to see how their patient's conditions change over time through a customizable data visualization application.

How is FHIR used in the App being demonstrated (500 characters)? : MedTimeLine uses Cerner’s implementation of the FHIR specification to query information about a patient from the EHR so that it can display this information in a clinician-facing application. The application uses the FHIR API’s search functionality to gather relevant data about the patient within a time period.

1. What FHIR release does your application use? (500 characters)?: We use the DSTU 2 (v1.0.2) version of FHIR.

What is the data source for the FHIR resources and how are the FHIR resources accessed? (500 characters): We are using FHIR bundles retrieved from a FHIR server that runs on premises at Boston Children's Hospital. We use the SMART on FHIR Javascript API to make HTTP calls to the FHIR server.

Any other information about the project we should know about (1500 characters)?: We also use these FHIR resources that are unlisted below:
MedicationOrder
Encounter

Authors:

Asim Ahmed, Boston Children's Hospital
John Brownstein, Boston Children's Hospital
Sam Chennuri, Boston Children's Hospital
Emily Cohn, Boston Children's Hospital
Lauren Cairco Dukes, Verily Life Sciences
Haiwen Gui, Verily Life Sciences
Nitin Gujral, Boston Children's Hospital
Joshua Herigon (Presenter)
Boston Children's Hospital

Mei-Mei Hollenstein, Verily Life Sciences
Shilpa Kumar, University of Washington
Hilary Mulholland, Verily Life Sciences
Melissa Van Cain, Boston Children's Hospital
Paul Varghese, Verily Life Sciences