How to understand the “Why” in Healthcare Risks

Increasing costs in US healthcare is estimated to reach $4.7 trillion by 2020 (Zamosky, 2014). At the center of this are we, the patient. We seek quality outcome from our healthcare visits. But even with the bourgeoning costs, there is no established direct association between cost and quality (P. S. Hussey, 2013). Fee for service that is independent of the outcome still dominate 60% of provider payments (Verel, 2014). So the question is, what can be done to improve the quality of healthcare results?

One important step in that direction is to understand the “why” in healthcare risk. Understanding the reason for the risks will help predict and at times, preemptive prevent healthcare problems. The NIH Director has argued that to improve healthcare, medicine must follow the 4 P’s: Predictive, Personalized, Preemptive and Participatory. Future research will enable physicians to predict how, when and in whom a disease will develop (Zerhouni, 2008). Consequently, there are numerous such predictive models for healthcare risks. For example, the Center for Medicare and Medicaid Services Hierarchical Condition Category (CMS-HCC) Risk Adjustment Model uses demographic information and profile of major medical conditions in a base year to predict Medicare expenditures in the next year (G. C. Pope, 2011). However, such models have a predictive accuracy (R-squared) of no more than 18% (J. Chen, 2013). One reason for this is that there is no clarity in understanding the “why” in the manifestations of healthcare risks. As long as that gap remains, prediction accuracies will be curtailed.

The reason for the lack of complete understanding of healthcare risks is that current data-driven decisions are (primarily) based on claims, enrolment and clinical data, which is not enough. For example, the average patient visits a physician 4 times a year, and each visit is about 15 minutes. So in total the patient spends about 60 minutes with a physician in a year. Consequently, the data generated is not enough to modelor understand the patient’s healthcare risks, let along predict anything.

To construct a complete view of a patient’s healthcare risks we need additional data sources, dimensions and attributes like demographics, psychographics, financial, socio-economic, and social and reviews, among others. Which is a tall order of tasks, doable but not trivial.

To begin with small steps, it is beneficial to understand a patient’s daily routine (Figure 1). Daily behaviors serve as a proxy for some of the aforementioned attributes. For example, they gives insights into the habits, activities and stress levels of a person that help to understand future potential risks and why existing treatments are working or not working.

article picture

One way to engage with a patient is through care management where healthcare professionals use patient intervention protocols and take extensive notes. Subsequently, a lot of insightful patient information is contained in these care management notes. Such patient notes are unstructured data. Not surprising, given that unstructured clinical data make up for approximately 50% of the 1.2 billion clinical documents produced in the US each year (Amorosano, 2012). To extract meaningful clinical information from such unstructured data, there exists a variety of text distillation tools.

Patient clinical information extracted from such care management notes is used to understand factors that contribute to patient’s healthcare risks. For example, notes on engaging patients with diabetes to get their hemoglobin A1C levels under 8.0%, blood pressures less than 130/80 and LDL cholesterols of under 100. Understand what factors are contributing to the patient’s hemoglobin, blood pressure, cholesterols and other statistics. Another example is notes on engaging a patient to get a mammogram that is due. Understand the effect of medications given that mammogram is due and steps to take to ensure sooner rather than later service date.

All such clinical notes are key evidences to understanding why certain risks are manifested and why existing risks are being aggravated or mitigated. Also, these notes help determine strategies that work or do not work in terms of physician visits, diet, and medicine intake, among others.

Furthermore, notes on patients that demonstrate new symptoms help detect factors that contribute to the symptoms. As with any predictive modeling, the assumption is that there will be factor in every risk that will have a predictive power in the manifestation of the risk (Burke, 2013). Once there is an understanding on why symptoms are happening, it is manageable to control and in some cases, predict healthcare risks for otherwise unsuspecting patients.

It is important to note that the information from unstructured data is not used in isolation. Instead, the distilled information is combined with the structured information in current risk model workflows to improve prediction accuracy.

Specifically, unstructured care management notes help to:

  1. For symptomatic patients, identify and understand why some factors aggravate existing healthcare risks and some factors have no bearing on the patient health. Use the understanding to focus on the patient’s health treatment.
  2. For symptomatic patients, identify and understand why certain factors trigger additional health risks. Use the understanding to proactively help other at-risk patients.
  3. For given DRG and patient demographic, identify and understand why certain care management protocols such as Patient Centered Medical Homes (PCMH), Patient Education, Wellness programs have worked vs not-worked in the past. Use this information to increase the probability of success for a new patient.
  4. For at-risk patients, identify factors that have historically contributed to increasing or decreasing the risk and accordingly implement their care management protocols.

In conclusion, understanding the “why” in healthcare risk translates into better quality outcomes without driving up the costs.



[1] Anderson, M. (2007). Artificial Intuition. Retrieved from Artificial Intuition:
[2] Hoffe, O., & Salazar, C. (2003). Aristotle. State University of New York Press.
[3] Kurzweil, R. (2013). How to Create a Mind: The Secret of Human Thought Revealed. Penguin Books.
[4] Lama, T. D. (2001). An Open Heart - Practicing Compassion in Everyday Life. New York: Back Bay Books.
[5] Wikipedia. (2014). Moore's Law. Retrieved from Wikipedia:'s_law