New Understanding from Analytics: Today and in the Future

An interesting paper caught my attention describing the prescribing patterns of short-term corticosteroids and the incidence of adverse events in the 30 days after their prescription. The results were surprising, and I wondered why such a large number of patients, roughly 20% of this insurer’s beneficiaries between 18 and 64 years of age in a 3 year period of time, received this therapy. They found that it was usually prescribed by primary care (family medicine and internal medicine), although emergency medicine, otolaryngology and orthopedics specialists also prescribed it less frequently. This retrospective study based on claims data showed that those who received these prescriptions experienced significantly more adverse events, including sepsis, venous thromboembolism and fracture, than those who did not receive these prescriptions. What should prescribers do with this information? Does it mean that patient risks outweighed benefits for short-term corticosteroids? If it provides benefit, can the risk of an adverse event be accurately predicted in individual patients?

Data needed to understand risks and benefits. Researchers are careful to compare apples to apples in prospective studies, recruiting subjects in test and control groups that are free of confounding characteristics. These “gold standard” studies are expensive and usually look at a narrow set of outcomes, such as a response to a drug for a specific condition. After a therapy is proven effective, it is often used for patients in populations who don’t necessarily match those in the study. Provider organizations use retrospective data using quality improvement to study selected issues, but these activities are limited to issues that come to their attention for some reason (such as a cluster of infections), and are more common in hospital settings where data is more easily aggregated and studied. Connecting the dots for outcomes outside hospital settings is more difficult, as this study exemplifies. But other data could be helpful identifying factors associated with poorer outcomes and perhaps even ways to identify patients who would benefit from the therapy. By gathering this additional clinical and other data, more personalized care can be given to future patients.

Population Health. Organizations, including providers and payers, utilize population aggregation to monitor and manage groups with common chronic conditions. Different conditions are monitored by different metrics, and clinicians can use dashboards to monitor those metrics across populations. Individual patients can also get a view of their metrics in a personal “dashboard” view. For example, patients with heart failure have specific measurements (lab and otherwise) monitored by clinicians that indicate disease status and progression. Clinicians can also use that data to manage communication with patients and other clinicians.

Population health provides efficiencies for groups of patients with common issues, but it doesn’t provide individualization that’s needed for optimal care. Even within populations with the same/similar therapies, patients differ in significant ways. Health behaviors, social and family environments, response to and tolerance of medications, co-morbidities all add to variation among patients within the population. Each patient in the population therefore also needs an individualized plan for reaching optimal health.

Personalized Medicine. Many think that personalized medicine refers to genomics, but more data beyond genomics is involved. Beyond a patient’s conditions, physical and social characteristics will impact outcome so need to be considered when creating a plan. Poor outcomes are not just associated with a patient’s diagnosis and medical treatments. If a patient is frail or has other limitations, how does that impact their ability to manage their own care? How are these issues identified, and how do those conditions impact plans for managing their conditions?

A newly-described factor that can dramatically influence outcomes no matter what the patient’s diagnosis is patient workload as compared to patient capacity. If the workload overwhelms capacity, nothing clinicians can do will lead to better outcomes. Clinicians often respond to such imbalances by intensifying interventions, which is usually counterproductive. Nathan Shippee and colleagues define cumulative complexity as something that can be measured and mitigated by reducing workload by simplifying therapies, increasing capacity with better support, or both. This potentially critical factor can be important in many patients who have complex conditions and/or complex treatment regimens and is independent of most population health metrics.

Merging Population Health and Personalized Medicine. A paper written by the Personalized Medicine Coalition describes the difficulty implementing personalized medicine practices in today’s health systems. They report challenges in 5 major areas, which include clinician and consumer awareness, patient empowerment and information management. There will be changes in how clinical care is delivered, in both problem diagnosis and care plan development. As new discoveries are made that create more testing and treatment options, there will need to be ways that patients can interact with their own data to manage their ongoing health conditions. For example, as genomic data is used more to predict tolerance and optimal dosing of medications, more information will be needed before a prescription is written. Will patients be screened for relevant genotype information that will be available at the time a prescription is written, or will patients have to wait until after the test results return, thereby delaying therapy? As new expectations grow, data will need to be available by multiple clinicians at their various points of care, and the data will need to follow the patient to various clinical settings.

Population Health practices will be used to help clinicians understand how outcomes can be influenced by monitoring metrics. Clinicians will also use Population Health tools to notify patients of new educational resources, necessary tests and changes in treatment recommendations. More information will be made available to clinicians and patients, however, to better understand what helps and what doesn’t help individual patients manage their health. Even though a patient is part of a population, variation will mean that some things work for one while not for another with the same condition. Care can become more personalized, even while patients still are monitored and offered resources based on their inclusion in a population. This way patients and populations most likely to benefit from short-term corticosteroids can be identified, as well as patients and populations most likely to suffer serious complications from them.


1. Waljee AK, Rogers MAM, Lin P, et al. Short term use of oral corticosteroids and related harms among adults in the United States: population based cohort study. BMJ 357:j1415 (2017).

2. Shippee ND, Shah ND, May CR, et al. Cumulative complexity: a functional, patient-centered model of patient complexity can improve research and practice. J Clin Epidemiol 65:1041-51 (2012).

3. Pritchard DE, Moeckel F, Villa MS, et al. Strategies for integrating personalized medicine into healthcare practice. Per Med 14(2):141-52 (2017).