A recent Wall Street Journal article highlights the work of Dr. Thomas McGinn in embedding clinical decision support tools in the electronic medical record (EMR). Integrating clinical decision rules (CDRs) for various conditions into the EMR assists physicians in determining the probability of a particular diagnosis (e.g. pneumonia vs. acute bronchitis). CDRs are presented in the EMR based on the patient's symptoms (e.g., cough) and typically consist of a set of evidenced based questions which categorize the likelihood of a diagnosis (e.g. pneumonia) from low probability to high. Based on the predicitive results, tests and treatment may be deemed appropriate or inappropriate.
Clearly, implementation of such clinical practice guidelines based on the best medical evidence available in clinical workflows is central to the value-based model of care. By reducing inappropriate variation in care, the expectation is for improved patient outcomes (quality) at reduced cost, i.e., increased value. Integration of clinical decision support (CDS) based on best medical evidence into the EMR at the point of care is fundamental to this strategy. As the WSJ article points out, however, many physicians are resistant to adoption of CDS. Important concerns include loss of autonomy, EMR-related "click-fatigue," and dehumanization of medical care. Moving forward, these issues must be addressed. I've recently written about a hybrid approach - "customized evidence-based medicine" which merges the best of standardized medicine and craft-based personalized care. Of equal urgency is the need to evaluate the effectiveness of EMR-enabled clinical decision support in controlling costs and improving patient-centered clinical outcomes. CMS and health care organizations are betting heavily on this model as the lynchpin of the transformation to value-based care. The evidence for efficacy at this point is promising but surprisingly thin and spotty. While I think the "glass is half full", much work needs to be done.
How Effective is EMR-enabled CDS? - The first 25 years.
Clinical practice guidelines and decision support are not new concepts in medicine. The most comprehensive systematic review of studies evaluating the effects of clinical decision support systems covered 25 years of research from 1976 through 2011. This landmark report published in the Annals of Internal Medicine included 148 randomized controlled trials (RCTs) - the gold standard of medical research. The report included evaluation of "electronic" CDS systems implemented in real clinical settings and used by clinicians for decision making at the point of care or for a specific care situation. Studies were ranked in quality as high, moderate, or low based on rigorous criteria. Not surprisingly, most of the studies (128, 86%) focused on health care process measures. Only 29 (20%) assessed patient clinical outcomes. Even fewer, 22 (15%) measured economic impacts including costs and efficiency. It's worth taking a closer look at the analyses in these three primary areas:
1. Health Care Process Measures
The data is clear. Multiple studies of moderate to high quality found solid evidence that CDS effectively improves ordering of preventive services, recommended clinical studies and treatments. Providing clinicians with point of care evidence-informed information improves performance on these metrics. We expect improvement in process measures to translate to improved patient outcomes.
2. Clinical Outcomes
Here's where we begin to see gaps in the data. Expectations aside, studies assessing the impact of CDS on clinical outcomes are fewer and generally of only low-moderate quality. They do demonstrate modest improvement in morbidity metrics such as hospitalizations, wound infections, deep venous thrombosis (DVT or blood clot) and episodes of hypoglycemia. The results are inconclusive, however, on whether decision support affects mortality rates. Finally, on an important metric of health care status - Health Related Quality of Life (HRQOL) scores - evidence for improvement was rated as poor and not statistically significant. Similarly, evidence for reduction in adverse effects of treatment was lacking. Obviously, much more work needs to be done to validate the role of CDS in the march to value based care.3. Economic Outcomes
This category encompasses measures of clinician workload, cost of care and cost-effectiveness of care. Clearly, we would expect that a reduction in unnecessary testing or treatment would save health care dollars. Indeed, moderate quality studies considered in the review found that CDS leads to a trend toward lower treatment costs, total costs and cost savings. Whether this is also a result of improved clinical outcomes - such as decreased hospitalizations - isn't clear. At least as of 2012, the evidence was not sufficient to evaluate whether clinician efficiency or workload improved. While evidence rated of moderate quality found higher provider satisfaction, most included studies found relatively low use of CDS (< 50%) by clinicians.
What we know from this review then, at least as of 2012, there was solid evidence that EMR-embedded CDS improves health care process measures across diverse settings. However, strong evidence for improvement in other important domains including clinical outcomes, efficiency and clinician workloads was lacking.
Updating the Evidence: 2013-2016
Since the Annals review, the evidence that CDS reduces unnecessary and inappropriate testing and treatment continues to mount. A recent report, for example, described implementation of a clinical decision rule in the EMR at the point of ordering a lab test (ceruloplasmin) used in the detection of a rare liver condition, Wilson's disease. The test has a false positive rate of 98.1%! Implementation of the best practice alert (BPA) reduced ordering of ceruloplasmin by 82% in the outpatient clinics. Another recent systematic review found integration of CDS in the EMR improves appropriate use of diagnostic radiology by a moderate amount and decreases use by a small amount. This finding is very consistent with our experience at Cleveland Clinic when we implemented clinical guidelines for imaging orders in patients with low back pain. In our experience, the change in imaging utilization resulted in significant cost savings.
On another front, since inappropriate antibiotic use is a major concern in the era of emerging "superbugs", trials evaluating clinical decision support for antibiotic use at the point of care for common respiratory symptoms are of great interest. A recent study of the outpatient management of uncomplicated acute bronchitis evaluated the potential of CDS to reduce inappropriate antibiotic use. While clinical practice guidelines don't recommend antibiotic treatment of uncomplicated acute bronchitis, approximately 70% of patients seen with the diagnosis receive a prescription. Disappointingly, this study found CDS support provided at the point of care in the EMR had only a very modest effect, reducing antibiotic prescribing from 74% to 61%. Whether it's too many clicks in an antiquated EMR, poorly built one-size-fits-all CDS, or perceived loss of provider autonomy, we need to better understand the reasons for this surprisingly poor result.
Another trial evaluating the use of clinical prediction tools in the EMR to assess respiratory tract infections including pneumonia and strep pharyngitis did report a significant and appropriate reduction in the use of antibiotics for suspected strep by about 25%. However, as in the earlier study, the improvement in inappropriate use was only very modest. Though greater than 96% of patients were found to have a low probability of pneumonia using the clinical prediction tool, more than 50% still received antibiotics! Certainly the fact that fewer than 60% of physicians used the prediction tool contributed to this result.
Investigation of the economic outcome of point of care decision support remain poorly studied. Despite the centrality of evidenced based medicine and CDS to the emerging value based transformation of U.S. health care, none of these studies provide data on the actual economic impact , i.e., the greatly anticipated potential cost savings.
What we need now:
- Data on CDS effects on patient-centered clinical outcomes
- Data on CDS effects on costs
- Improved provider engagement with CDS in the EMR
Clearly the published evidence for the effectiveness of evidence-based clinical decision support integrated into the EMR is uneven. It's reasonably clear that inappropriate and unnecessary lab and radiology testing can be reduced and preventive care measures augmented. However, as we quicken the pace of health care transformation, we still have major gaps in our understanding of the impact of the model on our patients and the economic impact to the health care system. While we expect more widespread adoption of evidence-informed best practice guidelines for clinical care to improve patient outcomes, the evidence is surprisingly thin. Similarly, though we expect guideline driven care to be more cost-effective, we need more real data to validate this concept.
The most recent studies continue to report low physician utilization of decision support. Understanding and addressing barriers to adoption of CDS is a critical step. An urgent task is to fully engage these front line clinicians in the design of the best process and tools to facilitate best practice, evidence-informed patient care which recognizes the complexity and nuances of individual patient encounters. Building a user-friendly EMR interface which seamlessly incorporates point of care decision support within real clinical workflows is mandatory. Smart EMR encounter documentation that flows "like the clinician thinks" is a first step.
It's a mistake to underestimate the complexity of medical decision making. Evidence-informed clinical practice guidelines are a fundamental starting point. Patient-centered values and physician experience cannot be discounted. Customized evidence-based medicine is the best approach.
As always, I welcome your feedback and comments.