Healthcare: Measuring medical policy changes with O.R.

An application to the diagnosis and treatment of depression

Insurers are increasingly calling for screening, and where appropriate, treatment of mental health issues by primary care clinicians

Insurers are increasingly calling for screening, and where appropriate, treatment of mental health issues by primary care clinicians. Image © lightwise | 123rf.com

By Sumana Reddy and Harrison Schramm

How can advanced statistical methods be made accessible and gain insights for small- to mid-sized medical practices?

For the co-authors, the answer began with a series of conversations about how questions arising from current medical practice could be addressed using analytics. The hope is that readers will see the ensuing project and examples presented here as motivation to consider using data analytics – and the professionals who perform analysis – in their own practices to gain insights improving both management and patient care. The co-authors also hope to demonstrate that this type of analysis is not just a “major medical center” or “large practice” activity, but also is within reach for smaller practices.

Methodology

The Acacia Family Medical Group (AFMG) data used in this analysis consists of two tables containing appointment data and diagnosis codes. The appointment data consists of approximately 165,000 rows to include “no-shows” and cancellations. The diagnosis data consists of dates, ID and diagnosis codes. This data spans the time period 2012-2017; the full, clinically relevant patient history is not available.

Confidentiality of patients is preserved by using a randomly generated identifier for each patient, such as “ZZZZZ0002D” (see Table 1). Confidentiality of doctors is maintained by randomly recoding their name to a letter, i.e., Dr. “A.” The data analysis was performed in the statistical programming language R, with heavy use of the packages that make up the so-called “tidyverse” [1].

Table 1. Sample data from analysis. This raw data was grouped by ID to determine a “pattern of life” for patients with specific conditions.

Table 1. Sample data from analysis. This raw data was grouped by ID to determine a “pattern of life” for patients with specific conditions.

Depression

The use case of this study focuses around depression, as denoted by the following two International Classification of Diseases (ICD-9) codes [2]:

  • 296.2: major depressive disorder, single episode
  • 311: depressive disorder not elsewhere classified

Our effort focuses on ICD-9 codes; to maintain consistent methodology across the timespan of data, we do not consider current (ICD-10) codes. The switchover between systems occurred in October 2015.

One of the objectives of this work was to consider the shift in diagnosis and coding of depression over time. Legislation, particularly the 2008 Mental Health Parity Act [3] and the 2016 21st Century Cures Act [4], allowed for increased access to mental health benefits. Due to the shortage of mental health professionals, insurers are increasingly calling for screening, and where appropriate, treatment of mental health issues by primary care clinicians. We expected to see a gradual change in prevalence of diagnoses, as clinicians adapted from an environment pre-2008 in which a diagnosis of depression was not reimbursable, to one in which recognition and treatment is encouraged.

Acacia Family Medical Group

AFMG is a medical practice serving the full spectrum of clients in California’s Salinas Valley. It has offices in the cities of Salinas and Prunedale. Salinas and the surrounding catchment area are rural and semi-urban, and patients are largely in city government and services such as teaching, as well as agri-business employees from executives to farm laborers. It is a younger than average population when compared against the region, with a high proportion of working-class people. The practice generally employs four physicians and a physician assistant at any given time; more doctors are present in the data set due to individuals joining and leaving the practice. It has participated in all available quality metrics and programs through Medicare and private insurers.

Results

As a first step, we considered how many patients in AFMG have been diagnosed with depression, by year (see Figure 1). This chart shows an overall increase in the treatment of depression across a stable patient population, with particular growth in the 296.2 code.

Figure 1: Appointments for depression by year and type.

Figure 1: Appointments for depression by year and type.

A deeper analysis considers the first time that a patient is diagnosed with depression, the average time between visits (for those who had more than one visit) and the total number of visits (see Figure 2). These are broken out for readability between the 296.2 and 311 ICD diagnoses.

Figure 2: Patient history by date of first diagnosis. The dates on the X axis are the date first diagnosed, the Y axis represents the average time (in day) between visits, and the size of the point represents the total number of visits.

Figure 2: Patient history by date of first diagnosis. The dates on the X axis are the date first diagnosed, the Y axis represents the average time (in day) between visits, and the size of the point represents the total number of visits.

This approach is useful in several contexts because it allows practitioners to create a “pattern of life” for patients. Once a “pattern of life” is determined, individual cases may be cross-compared, and outliers – doctors or patients that are “off trend” – may be identified. This approach may have broad applicability in extracting other insights from data.

Co-morbidity at First Diagnosis

We can also consider the difference in co-morbidity, specifically what other conditions are frequently diagnosed when a person is first diagnosed with depression, and how does this change in time? Our hypothesis, based on our practice, is that patients rarely make appointments for visits expecting a diagnosis of depression. It is our further hypothesis that the character of first appointments changed around 2014, because it took some time for the change in policy to “trickle down” to individual practitioners (see Figure 3).

Figure 3: Co-morbidity with depression diagnoses, grouped by epoch.

Figure 3: Co-morbidity with depression diagnoses, grouped by epoch.

Finally, we can consider which doctors are performing initial diagnoses for depression (see Figure 4). A chart like this needs to be read with care; one conclusion is that the doctors with the highest number of diagnoses of depression are more apt to make this diagnosis. A competing hypothesis is that in a family practice, individual doctors may tend to treat groups of patients that are at higher risk for depression. Evaluating these hypotheses is beyond the scope of data and effort for this study.

Figure 4: Initial diagnoses of depression by doctor (doctor names obfuscated).

Figure 4: Initial diagnoses of depression by doctor (doctor names obfuscated).

Conclusion and Future Work

This article is a demonstration of the use of alternate statistical tools on a data set from a family practice to gain insights into that practice. Our sense is that we have barely scratched the surface as to the insights to be gleaned. Specifically, this or similar analysis could be performed on any ICD code or collection thereof.

Our major takeaway for readers is that this sort of analysis is within the technical reach of many practices with a modest amount of external help from professionals in statistics and/or data science.

With a small adjustment, these approaches and data sets like this could be applied to finding insights about prescriptions and the “pattern of life” for patients across a variety of questions. Specifically, they could also be used to detect emerging trends in patient needs, as well as internal mechanisms for practices to perform their own quality audits.

Sumana Reddy, M.D., is a family practice specialist in Salinas, Calif. A fellow of the American Academy of Family Physicians (AAFP), Dr. Reddy is board certified in Family Medicine (U.C. San Francisco Medical School) and is affiliated with the Acacia Family Medical Group.

Harrison Schramm (Harrison.schramm@gmail.com), CAP, PStat, is a principal operations research analyst at CANA Advisors. He earned a master’s degree in operations research at the U.S. Naval Postgraduate School.

Acknowledgements

The authors thank Daniel Von Forell of SVMH for his assistance obtaining raw data, as well as the CANA Foundation for its support of this effort.

References

  1. Wickham, H., 2014, “Advanced R,” CRC Press, Boca Raton, Fla.
  2. Centers for Disease Control and Prevention, 2017, International Classification of Diseases – 10th Revision – Clinical Modification; https://www.cdc.gov/nchs/icd/icd10cm.htm
  3. Centers for Medicare and Medicaid Services (CMS), 2008, “The Mental Health Parity and Addiction Equity Act (MHPAEA); https://www.cms.gov/cciio/programs-and-initiatives/other-insurance-protections/mhpaea_factsheet.html
  4. Food and Drug Administration, 2016, “21st Century Cures Act”; https://www.fda.gov/RegulatoryInformation/LawsEnforcedbyFDA/SignificantAmendmentstotheFDCAct/21stCenturyCuresAct/default.htm