Defining Patient Populations Using Analytical Tools

Defining patient populations is an important first step when identifying opportunities for clinical improvement, but it can be a daunting one. How can a clinician easily find a specific patient population? Then, once found, how does that list turn into actionable steps that improve outcomes?

Cohort Builder for Patient Populations

Define Patient PopulationsIn the past, when a physician wanted to view a specific patient population, he or she would need to email a data analyst who would then conduct a database query…or add it to the long list of report requests. Usually, the requests were iterative and required refinement resulting in multiple back-and-forth emails. Now, this physician (or anyone with access to the application) can use an application called Cohort Builder.

Cohort Builder is built on the Health Catalyst data warehouse and allows people who are not data analysts (nor experts on querying databases) to conduct complex queries to find patient cohorts using a self-service, simple interface. Users can input clinical and demographic criteria to identify specific populations of patients, share population definitions with other users, and download detailed information on these populations. Additionally, the application can be deployed in a patient de-identified configuration to facilitate pre-IRB analysis.

When identifying a cohort, it is important to consider the level of detail or “grain” needed first. There are three fundamental cohort levels exposed in Cohort Builder:

  • Patient-centric (i.e. Asthma)
  • Episode-centric (i.e. Pregnancy)
  • Encounter-centric (i.e. Appendectomy)

It is easier to identify cohorts using procedure-based criteria (like appendectomies). Pulling cohorts based on chronic conditions (like asthma) is more difficult due to complexities in identifying patients who may not be presenting with a primary complaint related to that chronic condition Also, there are cases where patients would leave a cohort after extended periods of time in remission and re-enter if the condition re-presents itself. Episode-centric criteria (like pregnancies) become challenging because patients may have multiple encounters over many months with several providers. Determining which of these is really part of the episode and which are unrelated can be difficult.

Also, there are times it is challenging to decide whether a patient falls into a patient-centric cohort or an episode-centric cohort. Take depression as an example. If you have a traumatic event in your life that causes you to seek temporary help from a mental health specialist, you may meet criteria that would include you in a depression patient-centric cohort. However, you probably belong in a depression episodic-centric cohort. In this case, the patient-centric cohort would include you only if you have suffered and received treatment for depression since childhood.

There are many places to look when creating a cohort. Cohort Finder can leverage any discrete field in the enterprise data warehouse, but out of the box it contains many of the most valuable clinical and administrative data including:

  • Medications: pharmaceutical class, pharmaceutical subclass, generic and brand name, route, dosage, etc.
  • Labs: orders, results, reference ranges, etc.
  • Diagnosis Codes: ICD9 codes found in EMRs, professional bills, facility bills, problem lists, etc.
  • Procedure Codes: ICD and CPT codes found in EMRs, professional bills, facility bills, etc.
  • Patient Demographics: birth date, sex, race, marital status, smoking history, problem list, social history, etc.
  • And many other filters

Cohorts can be further refined through iterations with the organization’s knowledge manager, clinician input and by utilizing evidence-based standards.

Patient Population Risk Stratification

After a cohort has been defined, you will begin to tie metrics and stratifications to the cohort. Each distinct cohort may have its own unique set. (We will dive into the types of common metrics and stratifications in another post.) For now, let’s tackle one of the most important stratifications found with many, if not all, cohorts: risk stratification.

Risk stratification is an exciting methodology that assists in identifying outcomes for specific patient populations. Once a population has been identified, clinicians can intervene appropriately to promote a positive result. The ability to risk stratify saves healthcare systems a great deal of time, energy, and money.

An example of the applicability of this tool (and of current concern for hospital administrators) is heart failure readmission rates. Heart failure (HF) currently impacts 6.5 million adults in this country (Roger et al., 2012). Up to $39 billion is spent annually on HF hospitalizations (Dunlay et al., 2011), and one in four patients who have been hospitalized with HF will be readmitted within 30 days (Bernheimer et al., 2010). To make matters more challenging, the Centers for Medicare & Medicaid Services (CMS) are requiring hospitals report 30-day readmission rates for HF. If a HF patient is readmitted to the hospital within a 30 day period, that hospital will be penalized financially.

A recent national study by Bradley and others (2012) provides good evidence that some practices implemented by hospitals to reduce readmissions, such as follow-up appointments within 48 to 72 hours of hospital discharge, medication reconciliation by a nurse, involving social workers or case managers, and ensuring outpatient physicians are provided with a discharge summary, really does help prevent readmissions.

So what do HF readmissions have to do with risk stratification? Everything! By using a risk stratification tool, a provider is able to determine if a patient is at a high risk for re-admission and can provide intense focus on providing interventions like those suggested above. However, these interventions require finite resources and precious clinician time.  By stratifying for a specific group of patients, these limited resources can be directed to the patients where they provide the most value. It makes so much sense!

One of my clients came to the same conclusion. They set a goal to reduce 30-day HF readmissions. One of their Aim statements was to develop a risk stratification tool targeting those patients at greatest risk of returning within 30 days. They researched various risk stratification tools currently in use and found that, although there is a great deal of interest in this topic, there are relatively few examples available. Many of those examples are in various stages of development and testing. A tool developed by Amarasingham and others (2010) was helpful. It demonstrated that pertinent data to help risk stratify HF patients could be extracted from the EMR, as well as from census data.

With a great deal of thought and input from clinicians, my client developed  a tool that has begun to identify patients at high risk of being readmitted. This tool analyzes a number of variables including socioeconomic issues, abnormal lab values, the percentage of left ventricle ejection, a patient’s history of previous readmissions, etc. Patients that need interventions will be flagged as high risk and appropriately treated by clinicians.

Risk stratification takes the guesswork out of helping our patients and helps us preserve fixed resources. It will be an important part of controlling healthcare costs in the future. But most important, in the case of heart failure, it provides our patients the additional assistance needed for a better outcome.

Amarasingham, R., Moore, B.J., Tabuk, Y.P., Drazner, M.H., Clark, C.A., Zhang, S., Reed, G.W., Swanson, T.S., Ma, Y., Halm, E.A. 2010.  An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data. Medical Care 48(11), 981-988.
Bernheim, S.M., Grady, J.N., Lin, Z., Wang, Y., Savage, S.V., Bhat, K.R., Ross, J.S., Desai, M.M.,Merrill, A.R., Han, L.F., Rapp, M.T., Drye, E.E., Normand, S.L., Krumholtz, H.M.    2010.  National patterns of risk-standardized mortality and readmission for acute  myocardial infarction and heart failure. Circulation: Cardiovascular Quality Outcomes, 3, 459-467.
Bradley, E.H., Curry, L., Horwitz, L.I., Sipsma, H., Thompson, J.W., Elma, M., Walsh, M.N. & Krumholz, H.M. 2012. Contemporary evidence about hospital strategies for reducing   30-day readmissions.  Journal of the American College of Cardiology, 60(7), 607-614.
Dunlay, S.M., Shah, N.D., Shi,Q., Morlan, B., VanHouten, H., Long, K.H. & Roger, V.L. 2011. Lifetime costs of medical care after heart failure diagnosis.  Circulation: Cardiovascular Quality Outcomes, 123, 933-944.
Roger, V.L., Go, A.S., Lloyd-Jones, D.M., Benjamin, E.J., Berry, J.D., Borden, W.B., Bravata, D.N., Dai, S., Ford, E.S., Fox, C.S., Fullerton, J.H., Gillespie, C., et al. 2012. Heart disease and stroke statistics -2012 update:  a report from the American Heart Association. Circulation, 125, e200-220.
Loading next article...