AHRQ-funded research aims to improve the delivery of health services at the health systems or organizational level; this investment was $41.8 million over the duration of projects that were ongoing in 2020. The use of machine learning and efforts to share health information across technologies and healthcare environments are key aspects of research projects focused on health systems. Dissemination of successful data strategies and technology to strengthen healthcare delivery are also key.
Machine Learning
Machine learning is a type of artificial intelligence that programs computers to learn information without human intervention. In machine learning, development of the underlying algorithms relies on computational statistics. Computers are provided data, which they then “learn from.” The data actually “teach" the computer by revealing their complex patterns and underlying algorithms. The larger the sample of data the “machine" is provided, the more precise the machine's output becomes. Machine learning and algorithm use in healthcare are becoming more widely used, and AHRQ is funding multiple research projects to understand how they can help patients and clinicians.
Dr. Michael Avidan and team at Washington University developed and evaluated an air traffic control-like command center to monitor hospital operating rooms and forecast adverse patient outcomes. The Anesthesiology Control Tower (ACT) used data mining strategies and machine learning algorithms to analyze data from perioperative EHRs and real-time physiological data from the operating suite. The research team found that the ACT algorithms were able to predict postoperative adverse outcomes with a high degree of accuracy, leading to better outcomes for the highest risk patients.
Researchers at the University of Michigan, led by Dr. Charles Friedman, applied learning health system methodology with the goal of advancing individualized precision prevention (IPP) for Grade A and B USPSTF recommendations. Using realistic patient scenarios, the researchers tested concordance between the IPP algorithm’s ranking of preventive service recommendations and primary care provider rankings of those same services. They found an intermediate level of concordance, suggesting that IPP algorithms can help tailor selection and enactment of preventive services in practice.
Dr. Gondy Leroy and team at the University of Arizona developed and evaluated natural language processing (NLP) and machine learning algorithms to identify autism spectrum disorder (ASD) behaviors according to Diagnostic and Statistical Manual of Mental Disorders criteria. The research team created a prototype user interface that reviews and automatically annotates the often-overlooked free-text notes in EHRs with a high degree of precision, revealing that the use of ASD-specific NLP algorithms in clinical practice have the potential to facilitate earlier diagnoses of ASD in children, leading to earlier treatment and support for these individuals.
A Brown University team led by Dr. Thomas Trikalinos used machine learning technologies, including natural language processing, information retrieval, and text mining methods, to optimize the efficiency of the systematic review process and mitigate the challenges associated with information overload in the literature identification process. The team developed a literature identification process that unifies the query formulation and citation screening steps and uses modern approaches for text encoding to represent the text of the citations in a form that can be used by information retrieval and machine learning algorithms.
Patient-Reported Outcomes (PROs)
Scaling the use of patient-reported outcomes (PROs) to improve care is a priority of the AHRQ Digital Healthcare Research Program. PROs can yield insights into health status, function, symptom burden, adherence, health behaviors, and quality of life, since they come directly from the patient. Yet, PRO data are not routinely collected or used in clinical practice for several reasons, including patient and provider usability and lack of integration with EHRs. Recent efforts to advance the use of proven PRO apps include the following AHRQ-funded research:
Drs. Deliya Wesley and Raj. M. Ratwani of MedStar Health Research Institute investigated the functionality and usability of user-friendly electronic applications to collect diverse PRO measures in a standardized manner in a mix of ambulatory care settings. The research team pilot-tested two apps—the PROMIS Reporting and Insight System from Minnesota (PRISM) app (the winner of AHRQ’s 2018 Step Up App Challenge) and an existing PRO data collection app that was modified by incorporating FHIR technical specifications. The findings identified various factors crucial for the successful adoption, potential scaling, and sustained use of such technologies, including the availability of technical assistance, additional staffing, and supportive institutional policies.
Electronic Health Information Exchange (HIE)
Electronic health information exchange (HIE) allows physicians, nurses, pharmacists, and other healthcare providers to appropriately access and securely share a patient’s vital medical information electronically—improving the speed, quality, safety, and cost of care. While HIE is common, its utility is often hampered by the lack of seamless integration into clinical workflow. AHRQ-funded research is identifying the best ways for providers to fully leverage this technology, including the following study:
Dr. Brian Dixon of Indiana University–Purdue University at Indianapolis investigated the use and impact of a HIE system among multiple hospital EDs across Indiana. The HIE, which linked provider access data to clinical data about each ED encounter, was used more frequently once a single sign-on (SSO) feature was implemented, highlighting the importance of integrating the HIE into clinical workflows seamlessly. The findings suggest that future research should examine the effect of standardized HIE training for providers and whether additional functionalities such as customized user profiles can increase HIE use and impact.
Medication Safety
Medication safety has improved significantly over the last two decades with the use of digital health and other health IT tools. While errors may happen at all stages of the medication process, different tools have been developed to support the prescribing process (e.g., computerized prescribing with decision support), the dispensing process (e.g., barcoding or automated dispensing and unit-dose systems), and the administration process (e.g., electronic medication administration records and smart pumps). Digital healthcare tools can reduce medication dosing errors and preventable adverse events such as drug-drug and drug-allergy interaction rates by increasing documentation quality and transparency, enhancing accuracy and correctness within the medication process, and supporting information exchange by interlinking different stages of the medication process.
Dr. Joanna Abraham and team at the University of Illinois at Chicago examined the use of a void alert tool (VAT), an error alert function embedded in a hospital inpatient computerized provider order entry (CPOE) system, to identify medication ordering errors. The research team found that all reported ordering errors were due to a combination of associated risk factors, including communication gaps in the patient care team and workflow interruptions. Although no adverse drug events occurred, the findings point to the need for multifaceted strategies such as VAT in CPOE systems; this will likely ensure medication ordering safety and prevent potential harm to patients.
While the implementation and enhancement of digital healthcare tools has made significant strides in addressing medication safety, patients taking prescription medication still experience adverse drug events, with miscommunication in medication discontinuation playing a critical role. Dr. Michelle Anne Chui and a University of Wisconsin–Madison-based team evaluated the use of CancelRx, an e-prescribing tool to communicate medication discontinuation orders between EHRs and pharmacies. The research team conducted an interrupted time series analysis with outpatient clinics and community pharmacies, finding an immediate and significant increase in successful medication cancellations. Through interviews and observations with pharmacists and clinic staff, the need for a standardized workflow emerged, highlighting the utility of CancelRx to streamline the medication cancellation process, improve medication safety, and reduce potential adverse drug events.