By Ted James
November 29, 2018
“Uncontrolled variation is the enemy of quality.”
– Edwards Deming
Widespread variations in clinical practice that cannot be explained by differences in medical condition or patient preference are routinely observed across multiple specialties and clinical settings. The problem with unexplained variation in health care is that it negatively impacts clinical outcomes, increases the cost of care, and diminishes value for our patients.
Although every patient is different and has individual needs that must be addressed, specific processes that are known to have better outcomes should be implemented uniformly to improve care. Interventions that reduce unintended departures from ‘best practice’ will result in greater consistency in quality and help us to deliver on the promise of safe and effective care.
Is Centralization the Answer?
Attempting to improve quality by centralizing care to select centers of excellence has pros and cons. We have data demonstrating that high-volume specialty centers outperform low-volume centers in several measures of performance, including mortality, across a broad spectrum of medical conditions and procedures. We also do not necessarily benefit from having multiple iterations of the same services in every hospital. Centralizing care to organizations with the necessary infrastructure and expertise to manage complex patient care needs can improve coordination of limited resources, reduce costs, facilitate economies of scale, and improve outcomes. However, the problem with centralization is that it displaces patients from the supportive structure of their families and community. Having to travel to a ‘center of excellence’ can create financial and logistical problems. Furthermore, it is just not practical. There are not enough high-volume centers and specialists to adequately accommodate the volume of patient demand in the United States.
Moving Beyond Standardization of Care
Another approach to improving care is to implement evidence-based best practices to corral variation. Commonly referred to as ‘standardization of care’, this approach can lead to more reliable outcomes, enhanced efficiency, and improved safety. For example, Luther Midlefort, a Mayo Health System, reduced variation by creating a system-wide protocol for insulin use. After six weeks they achieved a significant reduction in the number of hypoglycemic events as a result of standardized practices.
Although standardization to reduce unintended variation is helpful, we must be careful not to adopt a “cookbook” approach that fails to acknowledge the differences in our patients, organizational cultures, and resources. For example, one of the reasons cited for the safe surgery checklist being successfully adopted globally is that organizations were given some degree of flexibility on how to implement the checklist in a manner that best aligned with the culture and characteristics of their organization. Furthermore, patients with the same condition may need different treatment depending on other variables. One-size-fits-all clinical pathways may hinder physicians’ ability to use their training and experience to appropriately personalize care. Standardization of care alone is not sufficient for attaining the highest levels of quality in our health care systems. Significant progress in health care improvement will require more than a set of guidelines.
A Data-Analytic Approach to Ideal Care
We can make progress ensuring that patients receive the highest quality of care by rigorously measuring outcomes using meaningful metrics developed in conjunction with the health care team, and then sharing this data transparently and in a timely fashion. This facilitates engagement and accountability, enabling a data-driven approach to establishing best practices and reducing unintended variation. Using robust analytics and risk stratification tools to monitor and evaluate the effectiveness of care allows organizations to understand variation, personalize care, and leverage opportunities for improvement and innovation.
Once robust measures of performance are in place, health systems need to identify and understand the causes of variation in outcomes. Unfortunately, when leaders don’t know how to analyze performance data, they are more vulnerable to miss opportunities for improvement, act on misinterpreted data, incorrectly assign credit and blame, and fail to predict future outcomes.
The interpretation of variation observed in data is greatly facilitated by the use of dynamic displays, where the data is presented as a function of time. This enables decision-makers to assess the process being measured appropriately and to analyze the results of improvement initiatives over time. As espoused by quality pioneers Walter A. Shewhart and W. Edwards Deming, relying on static, summary statistics (e.g., mean, median, standard deviation, etc.) provides a limited understanding of the process and may lead to conclusions not supported by a richer, dynamic display of data. Implementing control charts with probability-based interpretation rules allows leaders to determine whether any observed variation is a function of common causes (natural or chance fluctuations) versus special causes (unusual and assignable signals of change), each of which requires a different management response. If only common cause variation is present, any effort made to improve the process should be comprehensive, system-based, and not focused on particular “outliers,” as there are none. Conversely, if special causes are present, efforts should be focused on either removing these outliers (if they are detrimental to the process) or replicating them (if they enhance the process). Displaying data over time and using control charts is essential for evaluating the effectiveness of change.
Sample Chart Comparison
Figure 1: Summary Statistics
Here we see summary data describing wait times, but not much can be inferred to help with quality improvement or process redesign:
Figure 2: Dynamic Display
Now here the data points are displayed over time. We see that Mondays often have higher wait times, Wednesdays have lower wait times, and the Monday after July 4 had the highest wait time. This helps in making decisions that can improve the process being measured:
The addition of defined rules to determine ‘common cause’ versus ‘special cause’ variation provides further detail for the analysis and interpretation of the data.
If we are to be successful at transforming health care, we must recognize that leading quality improvement is a science. Health care leaders require specific education and training to hone skills in understanding and managing variation. Practical solutions will come from keeping data as the common denominator as we establish best-practices, and ultimately overcome unintended variation in health care through strong analytic systems.
Learn more about these postgraduate programs from Harvard Medical School:
Halm EA, Lee C, Chassin MR. (2002) Is volume related to outcome in health care? A systematic review and methodologic critique of the literature. Ann Intern Med. Sep 17;137(6):511-20. Review.
Lee TH. (2010) Turning doctors into leaders. Harv Bus Rev. Apr;88(4):50-8.
Institute for Healthcare Improvement: Static Vs. Dynamic Data: http://www.ihi.org/education/IHIOpenSchool/resources/Pages/AudioandVideo/Whiteboard10.aspx
Mohammed MA et al. (2004) Using statistical process control to improve the quality of health care [PDF]. Qual Saf Health Care.
Dr. Ted James is a medical director and vice chair at BIDMC/Harvard Medical School. He is an alumnus of the Harvard Health Care Management program and is involved internationally in leadership development and health care transformation. He also teaches through the HMS Office of Executive Education.
Dr. James blogs about health care transformation. To see more of his posts, click on his name in the tags below.
*OPINIONS EXPRESSED BY OUR GUEST AUTHORS ARE VALUABLE TO US AT LEAN FORWARD, BUT DO NOT REPRESENT OFFICIAL POSITIONS OR STATEMENTS FROM HARVARD MEDICAL SCHOOL.