Personalized healthcare means embracing variation. Yet the reduction of variation around proven care pathways has been shown to improve clinical outcomes. What is the balance between a minimum standard of assuring that proven effective care is provided before personalizing care based on patient values and other care options? Tomson summarizes the concern that minimum standards of effective implementation of proven care strategies are not being met: “Over the past few decades, the practice of physicians has been transformed by a growing evidence base quantifying the benefits and risks of interventions in disease states, and the development of clinical practice guidelines summarizing this evidence base. Although this has eroded the traditional autonomy of physicians to choose the treatment that they considered right for ‘their’ patient, considerable variation in the quality of care delivered to patients across the National Health Service (NHS) seems to persist. This implies that many patients leave healthcare encounters having not been offered treatments for their condition for which there is high-quality evidence.”1
The concept of variation reduction comes from industrial process improvement efforts where proven processes produced precision outcomes and reduced variation in those processes increased the quality of the product. In the 1970s John Wennberg of Dartmouth applied these methods in healthcare and found significant geographic variations in healthcare service delivery. He questioned whether appropriate services were being denied or other services over utilized.2 When the Triple AIM of healthcare was described, these analyses of utilization and cost variations were combined with mortality outcomes, so that a value of care calculation could be made.3 At KenSci we investigate geographic variations in following up for annual wellness visits by looking at the impacts of ethnicity, spoken language and a social determinant index on this health promotion behavior.
Small area variation analysis has been applied to healthcare systems, looking at variation between hospitals, provider groups and individual providers with a common goal of lowering costs of care by reducing low-value care.4 To understand these variations in practice patterns, a variety of strategies are used to compare like groups of patients by risk adjusting similar disease groups with severity of illness and social determinant indices.5
Strategies to reduce clinical variation are depicted in the graphic below and include standardizing care pathways, standardizing fixed cost purchasing decisions and influencing physician and patient decision-making. 6 7
Reducing Care Variation
Data analytics that can be used by administrators, managers, individual physicians and patients is key to implementing successful strategies to reduce variation. Developing evidence-based care pathways, creating order sets or value contract for high-cost items requires multidisciplinary teams. At KenSci we offer a Power BI based tool to analyze data from a variety of perspectives and present predictions and insights tailored to different team members. Once common goals are established, the Power BI dashboards can create standard score cards for individual physicians or teams to be compared to peers. Adjustments in care processes are supported by automating cues, reporting of insights at the point-of-care and sharing predictions with all team members, such as with the real-time COVID command center.
Focusing on Clinical Outcomes
Improving outcomes is the primary goal. Therefore, a proven relationship between specific care processes and improved outcomes is required for variation in care reduction to become the secondary goal. If there is no variation in a quadruple AIM outcome, then there may be NO need to reduce variation in the processes required to achieve that outcome. On the other hand, if there is wide variation in an outcome measure, then process variation assessment is needed. Are there subgroups of patients and/or care processes that achieve the best or worst outcomes?
Predicting outcomes for a group of patients and then surfacing the factors that influence that prediction is one strategy for understanding and narrowing variation. Patients may comply with established care pathway or a doctor select surgical approach with the highest likelihood of success.8 9 The prediction of patient-reported and clinical outcomes for an individual receiving anatomic or reverse total shoulder arthroplasty surgery provided insights that facilitated these care decisions, but they also supported shared decision-making where the patient insistent on surgery not to have the procedure.10
Managing Multiple Outcomes
If there are equally important competing outcomes, for example patient satisfaction or function and the cost of care. An analytic approach needs to optimize for both. One technique is to create a ratio of the clinical outcome divided by the cost and optimize for this index. This strategy has been used was used to understand variation in total shoulder arthroplasty care value.11 They observed the wide variation in the value of shoulder arthroplasty was most strongly associated with procedure type and certain preoperative characteristics.
When one outcome, such as mortality, is deemed more important than a second outcome, such as the costs, the former is first optimized in the development of an evidence-based care pathway. Once established, variation analytics can then identify subgroups based on cost differences within this optimal care pathway cohort Targeted cost containment strategies can then be individualized to the subgroup institution, provider group, patient group or individual provider.
Optimizing Outcomes without Reducing Variation
An example of how the drum beat of “variation reduction” might be misleading if we don’t fully consider the context and the outcome goals is when mortality outcomes are considered. In his New Yorker article “The Bell Curve“, Atul Gawande wrote about variation in mortality outcomes as viewed through a bell curve graph of mortality outcomes for Cystic Fibrosis clinics across the country. Those to the right of curve had the lowest mortality, and those on the left the highest. Since both ends may be considered outliers from a statistical perspective, clinicians might not consider them.
But Gawande recognized one private practice clinic that always appeared as an outlier to the right, and he wondered what an independent clinic was doing that allowed them to achieve lower mortality rates than the university clinics year after year. Rather than focusing on the central part of the bell curve, mean or median clinics, and trying to bring the outliers towards the center, Gawande studied the outliers to the right to discover that why and how they were different. He learned that this doctor and team cared for individual patients’ emotional and social issues in addition to their medical concerns, and how this improved compliance with their treatment program. By dealing with the social determinant of health (SDoH) and the emotional, or affective determinants of health (ADoH), they were able to personalize care and continually achieve better clinical outcomes.
Regressing to the Mean or Aspiring to Be The Best
KenSci’s machine learning, variation analytics approach to healthcare system assessment allows each customer to evaluate associations between many system features and the improvement in specific outcomes - from provider types to institutions to care pathways. Identifying subpopulations with the best outcomes and the care pathways that achieve those outcomes may be an important step before predicting adverse outcomes. Identifying groups of patients and associated care pathways that are already achieving the desired outcomes and then scaling best practices is frequently easier and more successful than identifying deficient ones and then attempting to fix them.
The Art of Medicine
Reducing variation around proven care pathways is an important strategy for improving outcomes where proven therapies and care pathways exist. New analytic strategies that include focus on emotional and social factors can help optimize and standardize alternative care pathways. Optimizing for these social and emotional factors is the “art of medicine”.
REFERENCES Tomson CR, van der Veer SN. Learning from practice variation to improve the quality of care. Clin Med (Lond). 2013;13(1):19-23. doi:10.7861/clinmedicine.13-1-19
 Wennberg J, Gittelsohn null. Small area variations in health care delivery. Science. 1973;182(4117):1102-1108. doi:10.1126/science.182.4117.1102
 Wennberg JE, Fisher ES. Finding high quality, efficient providers for value purchasing: cohort methods better than methods based on events. Med Care. 2002;40(10):853-855. doi:10.1097/00005650-200210000-00003
 Colla CH, Morden NE, Sequist TD, Schpero WL, Rosenthal MB. Choosing Wisely: Prevalence and Correlates of Low-Value Health Care Services in the United States. J Gen Intern Med. 2015;30(2):221-228. doi:10.1007/s11606-014-3070-z
 Standardizing Social Determinants Of Health Assessments | Health Affairs. Accessed January 2, 2020. https://www.healthaffairs.org/do/10.1377/hblog20190311.823116/full/?utm_source=Newsletter&utm_medium=email&utm_content=The+Most-Read+Health+Affairs+Blog+Posts+Of+2019&utm_campaign=HAT+1-2-20
 Ardoin D, MD, MBA, Malone J. Reducing Clinical Variation to Drive Success in Value-Based Care (Part 1). hfma. Accessed August 5, 2021. https://www.hfma.org/topics/operations-management/article/reducing-clinical-variation-to-drive-success-in-value-based-care0.html
 Ardoin D, MD, MBA, Malone J. Reducing clinical variation to drive success in value-based care (part 2). hfma. Accessed August 5, 2021. https://www.hfma.org/topics/operations-management/article/reducing-clinical-variation-to-drive-success-in-value-based-care.html
 Kumar V, Roche C, Overman S, et al. Use of machine learning to assess the predictive value of 3 commonly used clinical measures to quantify outcomes after total shoulder arthroplasty. Seminars in Arthroplasty: JSES. 2021;0(0). doi:10.1053/j.sart.2020.12.003
 Kumar V, Roche C, Overman S, et al. Using machine learning to predict clinical outcomes after shoulder arthroplasty with a minimal feature set. Journal of Shoulder and Elbow Surgery. Published online August 19, 2020. doi:10.1016/j.jse.2020.07.042
 Predict+ Patient-Specific Outcome Predictor | Exactech. www.exac.com. Accessed August 11, 2021. https://www.exac.com/extremities/predict-plus/
 Menendez ME, Mahendraraj KA, Grubhofer F, Muniz AR, Warner JJP, Jawa A. Variation in the value of total shoulder arthroplasty. Journal of Shoulder and Elbow Surgery. 2021;30(8):1924-1930. doi:10.1016/j.jse.2020.10.039