Health Plans

Optimizing drug treatments using real-world evidence

By Colin Russi, and Daniel Viray

Jan. 21, 2020 | Article | 7-minute read

Optimizing drug treatments using real-world evidence

In our previous blog post, we discussed how health plans can use real-world evidence to enable higher-value healthcare. Plans now have access to enriched data sets that when used correctly can improve patient outcomes and reduce healthcare costs. One such way is through the optimization of pharmaceutical prescribing behaviors.


Pursuant to this, we conducted a study to investigate how real-world evidence can provide insight into optimal prescribing patterns for patients with Type 2 diabetes. Given that clinical trials are often limited in scope—especially when they involve experimenting with multiple drug combinations—and the burgeoning need for individualized patient treatment, we looked to real-world evidence for deeper answers. We leveraged the scale of real-world data, and all of its concomitant features, to search for hidden patterns unbeknownst to the human eye.


Specifically, we sought to investigate:

  • After a metformin monotherapy regimen, what is the next best drug regimen to prescribe?
  • How can the effectiveness of the drug regimen differ by patient profile?

A background to Type 2 diabetes

Type 2 diabetes affects nearly 30 million patients in the U.S. (of which 25% are undiagnosed) and costs $327 billion yearly in medical costs and lost wages, according to the Centers for Disease Control and Prevention. These figures reiterate the need to search for optimal treatments to better control patients’ medical complications and simultaneously reduce healthcare costs. While there are existing medical algorithms that attempt to clarify the best diabetes treatment pathways, physicians are too often left with guidelines that are too general to fit the specific needs of their patients. The existence of “real-world” variables can add layers of uncertainty to the recommended pathways, rending it unclear which drugs ought to be prescribed. For example, how should a 60-year-old Caucasian female with other medical diagnoses be treated differently than a 30-year-old Asian male with no other medical complications? Real-world data may highlight systemic patterns to suggest the best ways to treat these patients of different profiles.

After metformin: What’s next?

While there is general agreement that metformin should be the first-line therapy to treat Type 2 diabetes, it’s unclear as to what the next-line therapy should be. For example, the only direction provided by the American Diabetes Association’s Standards of Medical Care in Diabetes is to use an SGTL2, DPP4, GLP-1 or TZD therapy.


To provide a more prescriptive perspective, we partnered with Symphony Health to procure an anonymized, HIPAA-compliant data set containing data for around 20 million Americans who are treated for diabetes. This data set contained three years’ worth of records and covered information for all drug classes across a wide range of characteristics. The information provided both prescription claims as well as linked lab values, which we used to measure disease progression over time.


In our study, we narrowed the data to a sample set of around 20,000 anonymized patients—all whom had sufficiently robust lab data and underwent a transition from a metformin monotherapy to a second-line therapy (and met criteria for adherence). To assess the effectiveness of the drug regimens, we analyzed HbA1c levels (a three-month blood sugar assessment) in the data before and after a six-month time period on a given therapy treatment. Given that lower HbA1c levels are associated with slower disease progressions and reduced medical complications, the larger the decrease in patients’ HbA1c levels, the more effective the regimens.


Overall, the top three most effective next-line therapies after metformin were GLP + metformin, DPP4 + SFU + metformin, and SGLT2 + metformin, respectively. The bottom three most effective (that is, the least effective) next-line therapies after metformin were DPP4 + metformin, DPP4, and SFU, respectively.

Notwithstanding, the real-world data also included patient attributes such as age, sex and geography, which allowed us to explore how certain patient characteristics affected the effectiveness of the drug treatments.

Drug effectiveness by patient profile

The real-world evidence highlighted how drug regimen effectiveness can differ by patient profile—findings that galvanize the need for more in-depth studies and births ideas for wider exploration. In our study, for example, we found that:

  • After a metformin monotherapy, a GLP + SFU + metformin polytherapy regimen was most effective in females (-1.56% decrease in HbA1c) whereas a DDP4 + SFU + metformin polytherapy regimen was most effective in males (-1.24% decrease in HbA1c).
  • Additionally, consider that DPP4 + SFU + metformin and TZD + metformin regimens have similar effectiveness but appreciable differences in costs. DPP4 + SFU + metformin has an overall average HbA1c change of -1.07%, while TZD + metformin has as an overall average HbA1c change of -0.92% (Note: Not a statistically significant difference between the therapies.) However, the annual cost per patient of DPP4 + SFU is around $5,000, while that of TZD is around $200, resulting in an approximately 96% savings, according to GoodRx.
  • While the DDP4 + SFU + metformin polytherapy regimen was most effective in males (first of 14 drug regimens), it ranked only eighth of 16 drug regimens in females, with a -0.85% decrease in HbA1c.

While real-world data highlighted potential areas for optimized treatment based on certain patient characteristics, other attributes such as age, co-morbidities and other determinants of health could be investigated to better understand optimal treatments for patients at the individual level.

Population impact

Through our study, we estimated that costs for select Type 2 diabetes patients can be reduced by $345 per member per month. This can be achieved by substituting patient drug regimens with those that are cheaper—and are at least as effective—than the drugs that were prescribed historically. For example, we found that DPP4 + metformin has an average HbA1c decrease of -0.82%, while TZD + metformin has an average decrease of -0.92%. Nevertheless, TZD is $345 (per member per month) cheaper than DPP4.


Furthermore, we created a model that extrapolated these savings across the entire U.S. healthcare system. Assuming that 15% of treatments are optimized for one year in select drug combination categories, savings related to diabetes spend can total nearly $100 million per year. If we project these small-scale optimizations and assume that all treatments are optimized for two years (the typical length of a treatment cycle), the cumulative savings would increase by at least tenfold.

Real-world evidence: Enriching the medical frontiers

With the tectonic shift from intuition-based to evidence-based medicine, real-world data is redefining the term “evidence.” Healthcare stakeholders no longer need to depend primarily on outputs of clinical trials: They are now equipped with resources to conduct their own studies, using data that captures variables beyond those controlled for in idealized environments.


In our study, we mined real-world data to understand how drug regimen effectiveness can differ based on aspects of patients’ profiles. Nonetheless, there are more variables (and interactions between them) that can be studied to glean further insight. To conduct such studies at a large scale, health plans and providers would likely need the correct data capabilities: data management platforms, data analysts and scientists, and subject matter experts with deep industry knowledge. They could research many topics such as optimal pharmaceutical prescribing patterns, side effects of drug combinations, or the interrelation of co-morbidities. The investment in these capabilities would beget substantial information gain, allowing health plans and providers to better spearhead initiatives aimed at improving patient outcomes and reducing healthcare spend.

About the author(s)