What Causes Price Differences in Diabetes Medications

PonyWang/E+ via Getty Images

As patients continue to struggle to afford their diabetes medications, the GoodRx Research team, in partnership with researchers at Brigham and Women's Hospital and Vanderbilt University Medical Center, set out to understand why the price for diabetes medications can vary significantly.

This new research, published in PLOS One, found that neighborhood factors, type of medication, and pharmacy type all play a role in the price of a diabetes medication. The piece highlights the importance of shopping around to find the best price for a medication.

The research detailed below looks at diabetes medication price trends based on pharmacy and neighborhood characteristics, and is based on prices from over 45,000 chain and independent pharmacies across the U.S.

Non-insulin generic therapies vary more in price

The difference in price between brand-name and generic medications is well known. In general, brand-name medications tend to cost more. But our new research finds that while generic medications tend to be more affordable, they also tend to see the largest variation in prices.

Non-insulin generic therapies can have vast differences in pricing. This group of medications includes metformin, pioglitazone, glimepiride, and more. In our study, we found the average maximum price of metformin 500 mg, one of the most popular non-insulin generic medications, to be 167% greater than the average lowest price at chain pharmacies.

Our research revealed that the price gap for insulins and novel therapies is not as wide. In fact, when looking at the Lantus 3 mL SoloStar pen and the Jardiance 25 mg tablet, the average maximum price is 7% and 6% greater than the average lowest price at chain pharmacies, respectively.

Pharmacy location affects prices

The income of neighborhoods where pharmacies are located has a surprising link to diabetes medication prices.

Specifically, we found that chain pharmacies tend to have higher cash prices for diabetes medications in neighborhoods with a larger population of high-income residents. They tend to have lower prices in neighborhoods with a larger uninsured population.

So what does this mean for consumers? Shopping around for a medication may mean expanding your search beyond your neighborhood. This is especially true for uninsured individuals that tend to pay cash prices.

Methodology

We used all-payer U.S. pharmacy data of chain and independent pharmacies across all 50 states. These data reflect transactions that were submitted to health insurers and pharmacy benefit managers, with pharmacies reporting the usual and customary (U&C) price for the medication with each transaction. The U&C price is the lowest price a customer might have to pay in the absence of insurance and is also referred to as "cash price." This price is inclusive of all discounts. Pharmacies are required to report the U&C price, which is also what they charge Medicare. Although most customers have prescription medication coverage and do not pay the U&C price, the U&C price is the most visible price at the retail level and impacts those lacking insurance. Pharmacies can respond to their customer needs by adjusting this price. Other prices such as the list price and the average wholesale price are constant across the country, while negotiated prices between pharmacy benefit managers and pharmacies are not publicly available and do not necessarily reflect what the customer ends up paying. The data were de-identified and therefore institutional review was not sought per Department of Health and Human Services HHS regulation 45 CFR 46.101(c).

Transactions between January 1, 2019 to December 31, 2019 were analyzed. We divided diabetes medications into insulins (aspart, degludec, detemir, human, lispro, regular, glargine, glulisine), non-insulin generic therapies (metformin, sulfonylureas, thiazolidinediones), and brand-name non-insulin therapies (glucagon-like peptide-1 receptor agonists (GLP-1RA), sodium glucose cotransporter-2 inhibitors (SGLT2i), and dipeptidyl peptidase-4 inhibitors). Insulins included both generic insulins such as insulin aspart 70/30 and brand-name insulins, which were the vast majority. Non-insulin generics did not include brand-name formulations such as Glucophage. A total of 136 drug formulations were included in the analysis.

To allow for standardized comparisons across medications and dosages, for each drug claim we derived the cash price per unit (PPU) by dividing the reported usual and customary amount by the quantity filled. To aggregate the data from claim to pharmacy level, we averaged the cash PPU of a given drug across all claims in a particular pharmacy, weighted by the quantity filled on that claim. To exclude any extreme outliers, we grouped the data by drug, and determined a cutoff of 10 times the median consistent with a prior published analysis from these data. Because our model results examine percent changes, the interpretation of the results won't change even if we used cost per day as a measure. Our model also includes fixed effects at the medication level, which controls for the absolute level of price across drugs. By using medication-level fixed effects, we can use the model coefficients to estimate the cash PPU of drugs, and then multiply that by different doses and frequencies to estimate cost per day.

We then identified pharmacy and neighborhood factors associated with medication price variation. A generalized linear model with gamma distribution and log link function was constructed for each medication group, and the models were stratified by pharmacy type (independent or chain). The cash price PPU data is a long right-skewed distribution. Log-link represented a natural choice for our cash price data, where residuals increase with x, and we expected errors to be proportional with the conditional mean. Our approach was derived using the methods outlined by Manning and Mullahy in their paper on estimating log models. From the log link, we performed the modified park test, a modification of the heteroscedasticity test that is modified to help determine the family of distributions to use for a log link function. The test indicated that the Gamma family should be used.

Pharmacy-level cash PPU for a medication were regressed on ZIP code tabulation area (ZCTA) 3 level neighborhood variables and drug (fixed effects). Neighborhood variables sourced from the 2019 American Community Survey included mean household income (normalized), insurance coverage (Medicare, Medicaid, other insurance, no insurance coverage), race/ethnicity (Hispanic/Latino), and mean-to-median household income ratio. The proportion of housing units in urban areas was obtained from the 2010 U.S. Decennial Census. To test for pharmacy type interaction, we ran a pooled model, interacting the pharmacy type variable with all neighborhood variables, but without interaction with drug (fixed effects). We cross-walked ZIP codes within the 50 states from 5 (ZCTA5) using the 2019 Uniform Data System Mapper ZIP code to ZCTA Crosswalk and used the first three digits of the ZCTA5 to group data at the ZCTA3 level. To interpret the model results for neighborhood-level variables across ZCTA3s, we multiplied model coefficients for each neighborhood variable by its inter-quartile range and then exponentiated it. The resulting number is the percent change of the cash PPU going from the first to the third quartile for a given neighborhood variable. A cash price per PPU change greater than 5% from first to third quartile was considered noteworthy.

PPU was used to standardize the analysis, and sufficient for calculating percentage variations across ZCTA3. PPUs are insufficient for estimating the cost of a prescription due to difference in form (e.g., a unit of insulin can be used over multiple days), while a tablet, also a unity, may be taken multiple times a day. Monthly estimated costs for a typical prescription for one common medication from each drug group were calculated for interpretability of magnitude of costs and variation across.

Public link

Please use the above public link if you want to share this noodl on another website.