Biomarkers are increasingly seen as a cornerstone of evidence-based medicine, but they also represent one of medicine’s thorniest challenges. That’s because biomarker data comes in so many forms and from myriad sources — from standard numerical measurements (such as blood glucose readings) to handwritten notes by physicians and other care givers, to data from wearable devices and fitness apps. This complexity and diversity makes it difficult to realize the full potential of biomarkers.
In a pioneering collaboration, Austria’s K1 Competence Center for Biomarker Research in Medicine (CBmed) and SAP Health are using the state-of-the-art SAP Health platform, to collect, integrate and analyze patient data from routine documentation for biomarker discovery. The goal is to enhance patient outcomes while optimizing workflows in clinical and research environments. This platform has been deployed at the general hospital of the Medical University of Graz, although eventually CBmed hopes to roll it out many other hospitals in Europe.
“We’re making clinical data interoperable and comparable,” said Dr. Stefan Schulz, a physician, researcher and professor in medical informatics at the Medical University of Graz. “Most content in routine documentation is in the form of unstructured text. That suits the communication needs of clinicians, but it is a big obstacle for data analysis.”
Although this project is in its early stages, it has already relieved some major bottlenecks, especially for data export. Traditionally, electronic health record data used to be confined to relatively closed information systems, while data analysis required a completely separate platform. Exporting and rehashing data from clinical information systems into separate data warehouses caused considerable delays in analysis. SAP HANA (the core of the SAP Health platform) handles data transactions and analytics more efficiently, on the same machine.
“Very complex analytics on large datasets used to take hours. Now SAP’s customers can do them in seconds,” said Dr. Clemens Suter-Crazzolara, Vice President for Product Management for Health and Precision Medicine at SAP Health.
The project’s initial hurdle was to transform existing clinical data to a useful format: plain text. The hospital’s clinical information system provides a vast amount of data as PDF documents. After extraction of plain text from it, SAP Health platform uses advanced text mining software to refine unstructured source data into structured and semantically explicit data, which can be used across multiple systems.
“The architecture of most hospital information systems goes back to the 1990s. Back then, nobody thought of secondary users of data,” said Schulz. “Now, we can pick out important data items, bring them into context, and standardize them so that they can be stored and queried.”
Patient privacy is a top consideration and requirement in the context of medical data. CBmed’s SAP Health platform-based system includes stringent consent management. Patients must give consent for how their data can be added to the system and used.
“Regarding privacy, our internal regulations support efficient cooperation between the medical university and the hospital provider,” said Schulz. “In addition, everything stored on the SAP Health Platform is pseudonymized. We follow the U.S. HIPAA ‘safe harbor’ criteria by removing names, professions, geographic subdivisions, ages over 89, and so on.”
Professionals directly involved in patient care can view their patient’s information. However, other authorized personnel can access a “de-identified” version of patient data, in order to discover relevant cases — helpful for physicians seeking context about treatment options and outcomes, or for recruiting for clinical trials.
Natural Language Processing (NLP) for text mining is a key technology included in the platform. However, to perform well, NLP software must understand the local language, including local versions of specialized terminology. SAP and CBmed are working together with the Medical University of Graz to build or extend several German-language resources that are tailored to the terms and acronyms that clinicians use.
CBmed also had to develop enabling methods to pre-process clinical data before it is fed into the system. For example, Schulz noted that most hospitals lack fully structured digital medication records. Thus, the history of which medications were prescribed for, or taken by, a patient must be pieced together from several sources.
“It’s an iterative process,” Schulz said. “We create dictionaries and extraction routines related to medications. Then, through text mining, we extract data, find errors, and adjust our tools and dictionaries to get more precise results.”
SAP Health facilitates this iterative process. It tracks how and when data was processed. Thus, as text mining improves, data can be re-processed with newer, better methods.
In addition to these data mining approaches, SAP is working with CBmed to build applications of high clinical value that serve as interfaces for the SAP Health platform. For instance, the Patient QuickView app displays decision-relevant patient information.
“A doctor at the point of care must quickly assess a patient. In the past, this meant flipping through a folder full of papers, or consulting several different computer systems,” said Suter-Crazzolara. “The QuickView app delivers a single report, with hyperlinks to jump off to more information about the patient. It also supports communication and collaboration between doctors and other clinicians. And it supports different user types: a cardiologist would see a different overview than a hospital administrator.”
Additional SAP Health apps being developed through this project with CBmed may eventually help clinicians recruit patients for clinical trials, optimize standard coding for diagnoses and procedures, and help predict patient outcomes through next-generation biomarkers.
Ultimately, CBmed plans to roll out this system to most hospitals in Austria. However, once data quality- and access rights-aspects are resolved, subsets of this data could be of interest to many other parties, such as pharmaceutical companies, insurers and public health agencies. However, that is much further in the future.
In the nearer term, CBmed will be busy overhauling clinical data collections, processes and tools to ensure that the system becomes highly efficient. “That’s an ambitious yet exciting goal,” said Schulz.
A decade ago, an Austrian hospital pioneered the introduction of computerized drug prescriptions. “Doctors didn’t like it at first, but now everyone uses them,” Schulz noted. “We’ll need to train a new generation of medical professionals to produce more consistently structured data. Until that happens, processing of medical narratives through text mining is a high priority.”
“In the end, CBmed’s strategy is to drive innovative topics such as biomarker-based preventive, predictive, and personalized precision medicine, and optimized clinical trial execution and recruitment,” said Schulz. “For this, real-time analytics across disparate data sources is important, enabled through the SAP Health platform. Together this will tremendously benefit each individual patient.”