Project abstract (1000 Characters): Worldwide, more than 50,000 patients undergo Hematopoietic Stem Cell Transplantation (HSCT) each year, and many of these patients will develop transfusion-dependent thrombocytopenia. To support the management of these patients, we have developed a visualization application to improve the workflow around platelet transfusion for patients undergoing HSCT. The application is designed to support SMART on FHIR and works embedded directly within the EHR or as a standalone application. The application provides a unified view into several essential data elements relevant for patients requiring platelet transfusion, including platelet count trend, prior transfusions, patient HLA data, and current blood bank platelet inventory. Our application implements a custom sorting algorithm for the blood bank platelet inventory based on available HLA data, which has created an entirely new workflow and standard operating procedure for how platelets are assigned at our institution

Project rationale, impact and innovation (3500 Characters): Platelet transfusions are an essential component of the care for patients undergoing hematopoietic stem-cell transplantation (HSCT), and many of the 2 million platelet units transfused annually in the United States are used for patients who develop thrombocytopenia while undergoing HSCT. Thrombocytopenia, particularly at the levels experienced by HSCT patients, is a life-threatening condition, and can lead to spontaneous and catastrophic bleeding. Platelet refractoriness—a consistent failure to respond to platelet transfusions—is a common and challenging clinical problem, often caused by allo-immunization to Human Leukocyte Antigen (HLA) antibodies. Platelet refractoriness can lead to adverse outcomes, including increased bleeding risk, decreased survival, and longer hospitalizations.

Platelet refractoriness due to suspected allo-immunization is often managed by using HLA-matched platelets, but identifying which patients may benefit from HLA—matched platelets can be a clinical challenge, and today is currently done in two ways: (1) Observation based on refractoriness to several unmatched bags. This is problematic because many bags of platelets are wasted before a patient is identified as needing HLA-matched platelets, and (2) based on likelihood of rejection, for example Calculated Panel Reactive Antibody (cPRA) scoring. This is also challenging because it requires HLA data for the recipients, and must be calculated and presented to clinicians, so requires a new EHR workflow.

Optimizing decision-making and workflow around platelet transfusion is complex and involves numerous pieces of data that often live in different locations—for example, platelet count, tissue typing (HLA) data, blood bank inventory, and donor inventory. Synthesizing this data was previously done manually, and required paper lists, faxes, and phone calls. We built a point-of-care platelet inventory and transfusion visualization application aimed at improving the provider experience and platelet matching workflow, decreasing overall platelet utilization, and most importantly, decreasing the amount of patient-time spent at dangerous levels of thrombocytopenia. Users of our application include blood bank technicians, supervisors, physician trainees, and attendings. Our application is directly embedded in our EHR (Epic), launches with patient and provider context, and utilizes FHIR for data transfer. Our innovative approach to transfusion management has resulted in significant changes to the blood bank’s standard operating procedures.

Project design and implementation (7000 characters): The initial idea for our project came from a discussion with a clinical pathologist at our institution. Frustrated with data silos and paper-based processes, this physician recognized the potential of a software application to improve the workflow around platelet transfusions at our institution. This led to an extensive design phase where we worked to define the problem, understand requirements, and propose a solution.

As we began defining the problem space, it became clear that access to data was an important challenge to overcome. We identified several data silos at our institution that we would need to interface with. First, the tissue typing lab was responsible for our HLA antigen and antibody data about HSCT patients as well as platelet donors who were typed. This data was currently stored entirely outside the EHR, so we built a data extract, transform and load (ETL) process to bring this data into one of our systems. Importantly, we used the FHIR DiagnosticReport resource, as well as the FHIR Genomics Implementation Guidance, to help drive the data model. We also worked with our blood bank and donor center to ETL relevant data from those systems into our application. Finally, we had extensive discussions with our EHR integration team to understand how to retrieve certain data elements from our EHR. While some data, like platelet count, was straightforward, other data elements—most notably, transfusion events—was more challenging to reliably identify. Identifying all of the relevant data sources for our application, and building requisite interfaces to those data sources, was a major challenge that we were able to overcome.

Surfacing previously-siloed data was a critical component of our application. Additionally, calculating and displaying Panel Reactive Antibody scores (cPRA) was also essential for understanding platelet-refractoriness risk. Previous processes required manual calculation of this data, and was available only “as needed” for clinicians. To enable real-time cPRA calculation, we built a custom RESTful web-service. We have subsequently open-sourced this web service (https://github.com/persistentsystems/open-cpra-calculator) so that it can be used by the broader community. The web service has also enabled research, and we have published a novel calculation using custom population frequencies (Gordon, et al. Development of a calculated panel reactive antibody web service with local frequencies for platelet transfusion refractoriness risk stratification. J Pathol Inform 2019;10:26).

While we were securing access to data, we began to build the actual application. The application is a web application that brings in various data sources and allows a user to visualize each data source in a browser window. During this phase, we did extensive user testing with our potential clinical users to make sure that the application experience was optimal. We also did standard software-development quality assessments (automated and manual). During this phase, we often had to generate “dummy” data since many of our required data interfaces were not yet available.

Finally, one of the biggest challenges in implementing our solution was related to the standard operating procedures (SOP) of the blood bank. Blood banks are understandably heavily regulated, and the technology that is used to support and manage human product is under heavy scrutiny. In addition to normal software QA efforts, we worked with our blood bank colleagues to do extensive testing and validation (largely manual) once our application was complete. We also worked with the blood bank to update the SOPs for how they managed platelet transfusion to include our application as a standard component of their workflow. This took several months and speaks to the importance of ensuring high-quality software, particularly for mission-critical, operational programs like stem-cell transplantation and platelet transfusion.

Project evaluation and sustainability (3500 characters): We are evaluating our project through several mechanisms:

(1) Usage and retention. Usage of the application has been better than expected, and consistent. We average ~30-40 users per month, and this has been steady since implementation.

(2) Platelet utilization. We are tracking overall platelet utilization pre- and post- application implementation. Our institution has a donor center, but we also purchase platelets from the American Red Cross, so reduction in platelet utilization has important financial consequences for our institution. We hope to report our initial results after the first year of usage this winter.

(3) Thrombocytopenia. One of the goals of our project is to reduce patient-time spent at dangerous levels of thrombocytopenia, by more appropriately matching platelet units to patient-specific data like HLA. We are tracking this data currently and hope to report it after the first year of usage this winter.

Our project is currently part of the standard operating procedure of our blood bank and is supported by hospital operations.

Twitter project summary (140 characters): Improving platelet transfusion workflow through a SMART on FHIR application

How is FHIR used in the App being demonstrated (500 characters)? : Our application leverages FHIR primarily as a machine-to-machine interface for exchanging HLA data. As part of the ETL process for extracting data from the HLA source system, data is transformed into FHIR and then loaded into an oracle database, which is then used for our production application. This FHIR-based ETL process is kicked off every time a new HLA report is finalized by the tissue typing laboratory.

1. What FHIR release does your application use? (500 characters)?: Our implementation is based on FHIR R3 and primarily uses the DiagnosticReport resource as well as Observation resource. Our implementation was guided by FHIR Genomics Implementation Guidance. While we have not tested/validated these resources in R4, we expect that they would work in that release as well.

What is the data source for the FHIR resources and how are the FHIR resources accessed? (500 characters): Our FHIR resources are generated from a 4D database and ETL’ed into an oracle database and primarily used in machine-to-machine communication. Our application is built on the igia platform, an open-source platform for healthcare software development (https://igia.github.io/) which includes an embedded FHIR server.

Authors:

William Gordon (Presenter)
Brigham and Women's Hospital

Samuel Aronson, Partners HealthCare
Rory Dela Paz, Partners HealthCare
Michael Oates, Partners HealthCare
William Lane, Brigham and Women's Hospital