By Dr. Darryl W. Roberts and Dr. David Friedenberg
Will all of tomorrow’s healthcare decisions be made by computers? Not quite, but Clinical Decision Support (CDS) tools can make doctors, nurses and other healthcare workers smarter and more effective. Thanks to advances in data analytics, this machine-assisted medical future is closer than you may think.
Beyond the Pop-Up Alert
Technology-enabled healthcare decision making is not new. Alarms on ventilators and IVs, pop-up drug interaction alerts and other basic decision support tools have been in use for decades. However, these simple alerts and alarms do not rise to the level of CDS.
The Office of the National Coordinator for Health Information Technology (ONC; HealthIT.gov) defines CDS as a system that “provides clinicians, staff, patients or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care.” In other words, it is the ability to get the right information to the right people at the right time in order to inform diagnosis, treatment or follow-up. CDS uses data from a variety of sources (e.g. medical sensors, electronic health records (EHRs), physicians’ and nurses’ observations and clinical knowledge bases) to provide real-time guidance to clinicians using established and vetted guidelines. Many examples are already in use today:
- An obstetrician inputs data and observations into a computerized decision tool to predict the viability of a fetus and make informed decisions in a high-risk pregnancy.
- A pacemaker uses wireless technology and an mHealth app to provide real-time alerts, as well as analysis of stress conditions leading to spikes, to inform patient care.
CDS has the potential to improve patient outcomes and healthcare efficiencies significantly. CDS tools can help doctors and nurses avoid errors and adverse events and increase quality of care while reducing overall costs. That’s why the Centers for Medicare and Medicaid Services (CMS) gave CDS an important role in the original Medicare and Medicaid EHR incentive program. That worked so well that under the Medicare Access and CHIP Reauthorization Act (MACRA) Final Rule, CMS does not require CDS reporting, because it has “topped out” or achieved its goal. This change suggests that hospitals and healthcare providers will increasingly rely on CDS tools to help them meet patient care and financial goals under CMS’ Alternative Payment Models (APMs)[BND1] and similar models preferred by private insurers.
The Smart, Connected Future of Clinical Decision Support
While healthcare providers already rely on a multitude of tools to support decision making, we have only begun to scratch the surface of what will soon be possible with CDS. Imagine:
- A hospital physician faced with a patient presenting unusual symptoms inputs data and observations into a system that combs through millions of health records across the country to identify a handful of similar cases in other regional hospitals, suggesting the emergence of a new infectious disease.
- A rural primary care physician accesses an expert knowledge base and decision support tools to properly diagnose and make effective treatment decisions for a patient with a rare genetic disorder who is unable to easily travel to a specialty hospital for treatment.
- An epidemiologist tracks the spread of flu in real-time using data mined from millions of EHRs or personal health records (PHRs).
- A pharmaceutical company mines data from patients receiving a recently approved cancer therapy to monitor for rare adverse effects that may not show up in clinical trials and identify variables in dosage, timing, drug co-administration, patient characteristics, lifestyle and cancer type that impact treatment efficacy.
These examples are already within reach using the technology we have today, thanks to advances in health IT and data analytics. EHRs improve accessibility to patient data, but limitations to usability and interoperability still present challenges. In its current state, EHRs accessible to clinicians, patients and (with appropriate HIPAA protections) researchers make it possible to employ advances in “big data” analytics tools that can put that data to work. Instead of simply making recommendations for individual patients based on standardized guidelines, we can now use data mining, text mining and machine learning methods to extract subtle patterns in massive data sets and create new recommendations informed by cumulative expert knowledge and statistical analysis.
Sophisticated analytical programs are able to collect and analyze data from disparate sources—for example, body-worn medical sensors, EHRs, PHRs, administrative and billing records, imaging and diagnostic data and more—in order to answer complex medical questions[BND2] . Using these methods, hospitals and care providers will be able to make treatment decisions matched to patient and disease characteristics, quantify the added value of a treatment, or even automate some aspects of patient care. Other programs enable providers to mine large knowledge bases (such as PubMed) to inform evidence-based practice and identify hidden connections. For example, Battelle Sematrix™ uses natural language processing to sort through large corpora of unstructured scientific or technical information, allowing users to answer complex questions. Unlike simple keyword queries, these programs are able to make inferences by combining knowledge contained in different documents within the corpus.
Removing the Barriers to CDS Implementation
While basic technologies for CDS are already in place (and according to CMS “topped out” in their use), that doesn’t mean the industry is maximizing their utility just yet. Before that can happen, the industry needs to resolve significant market, technical and human constraints.
One of the most significant technical constraints is EHR interoperability. Each of the commercially available EHR systems tags and stores data differently. This severely limits our ability to combine records from different hospital systems for large-scale data mining. Interoperability minimizes the differentiators among EHR companies and even healthcare systems. There are currently not enough incentives in place to make data interoperability a priority for them in the current market. In fact, there are disincentives, even in the presence of government-sourced incentive programs. This means that researchers and clinicians who depend on easy exchange of health information for improved care coordination, reduced costs of care, and research to enhance practice are forced to use existing, inefficient workarounds. These workarounds include using traditional resource, such as questionnaires, that obtain needed information at the cost of additional clinician burden.
Another significant factor for the acceptance of CDS is usability. Decision support tools must fit within the clinical and administrative workflows and must be easy to use and understand in a complex, fast-paced environment. This is especially true of decision support built into “RN-ware,”1 or devices used by nurses to make on-the-fly patient care decisions. Already, nurses working in an ER or ICU may have dozens or even hundreds of devices that they use over the course of their day, each with its own unique alert sounds and interface. The resulting confusion makes medical errors more likely as nurses may miss critical alerts buried in the noise or misinterpret recommendations due to data overload. As more of these devices begin to incorporate sophisticated decision support tools, developers need to incorporate human factors to ensure that the systems help clinicians manage the volumes of data they are presented with and enable them draw fast, effective conclusions.
CDS and the Evolution of Medical Care
Clinical decision support is not about supplanting human thought. In an ideal world, CDS frees up doctors, nurses, caregivers and patients themselves to access wider knowledge bases and datasets and to think more deeply about what the collected data and knowledge is telling them. CDS tools are already making healthcare providers more effective and efficient and improving the quality of care for patients.
The data revolution in healthcare is moving fast. According to the MACRA Final Rule, CMS expects 80 to 90 percent of eligible clinicians to participate in APMs and other value-based payment models by 2020, thus shifting away from the current method of payment for quantity. This will result in a similar payment model shift by private insurers, as well as increased competitive pressure for improved quality of care. This will serve to push healthcare systems towards wider adoption of data-driven decision making. Data analytics and CDS tools will be critical components in helping the industry meet CMS targets and improve patient outcomes and financial returns. We can expect to see increasingly sophisticated CDS tools baked into tomorrow’s smart medical devices and programs.
Most of the components necessary to achieve the full promise of CDS already exist in some form. To bring the vision to fruition, medical device manufacturers, EHR and other software developers, health care systems, clinicians and patients will need to work together to solve issues around interoperability, data accessibility and human factors. As we make data more accessible, usable and understandable, CDS and data analytics are poised to take healthcare to a whole new level.
1 Dr. Roberts coined the term “RN-ware” in a 2015 MS Health Blog to explain how nurses pull together knowledge from a multiplex of disparate information sources (e.g., EHR, IV pump, ventilator) to draw the clinical picture of a patient for herself and other clinicians—oftentimes in an instant.
About the Authors
Dr. Darryl W. Roberts is a Registered Nurse and a Healthcare Quality Research Leader at Battelle. He has more than 25 years of experience in patient care, informatics, management, research and education. His recent work includes developing economic metrics for public health interventions, supporting the quality Measures Management System for CMS, developing quality measures for SAMHSA and bending qualitative research methodologies to make their use possible in natural language processing and big data analytics.
Dr. David Friedenberg is an Applied Statistician and Data Scientist whose work focuses on extracting meaningful information and structures from large, high-dimensional datasets. At Battelle, he analyzes complex high-dimensional data in fields such as Neuroscience, Chemical Forensics, Medical Devices and Astronomy as well as large health care and insurance databases.