Can’t sign in? Forgot your password?
Enter your email address below and we will send you the reset instructions
If the address matches an existing account you will receive an email with instructions to reset your password.
Can’t sign in? Forgot your username?
Enter your email address below and we will send you your username
If the address matches an existing account you will receive an email with instructions to retrieve your username
Your password has been changed
The list price of branded drugs sold in the US market often differs substantially from the “net” prices ultimately paid for them—a phenomenon that has grown more pronounced in recent years. While list pricing data are generally available, net prices have traditionally been proprietary, creating a challenge for researchers and policy makers.
Data on average net prices from SSR Health, LLC, have recently helped fill this gap and have been rapidly incorporated into research. However, these data come with important assumptions and features that need to be recognized to produce accurate analysis with them. In this piece, we clarify how these estimates are produced, what assumptions they include, and highlight best practices for using the data.
As the name suggests, the US Brand Rx Net Pricing Tool from SSR Health includes an estimate of per-unit net prices for branded drugs and some biosimilars in the domestic market. In addition, the data set includes the wholesale acquisition cost (WAC), or list price, for each drug as well as information about manufacturers, marketing dates, and drug classes. The data are quarterly and begin in the first quarter of 2007. The data are available to researchers for a fee and have been used in numerous recent studies of drug pricing and treatment cost.
The net-price estimates for most drugs are produced using two inputs: information on total net revenues to drug manufacturers in each quarter and estimated volume sales of the drug in a quarter. Estimated net prices are then calculated by dividing net revenues by volume. This method is complemented by an approach that makes use of drug manufacturer balance sheet accrual disclosures that can provide information about gross-to-net discounts (the details of this method are proprietary).
Net revenues come from 10-K financial reports from drug makers or other public sources, including earnings calls, press releases, and investor presentations, among others. It has become standard for publicly owned drug makers to release this information disaggregated at the drug level for products with non-trivial sales. These net revenues are made available at the product level (for example, sales for various pill or package sizes are combined); thus, the WAC-to-net percent discount is generally constant for each one of a product’s National Drug Codes (NDCs). (NDCs are the unique identifier the Food and Drug Administration gives to each drug formulation and packaging of a product.) Importantly, revenues are net of all price concessions the manufacturer gives, including rebates, 340B discounts, and coupons provided by manufacturers.
Data on units sold come from a third-party vendor, Symphony Health, which reports commercial sales in both the retail and physician-administered settings. Retail-unit sales are captured from pharmacy claims, meaning they need to be prescribed and distributed to patients to be counted, while physician-administered units reflect shipments to providers. The data reportedly cover more than 90 percent of prescription drugs purchases in the US; however, coverage varies across settings and time. It is most complete for drugs sold through retail settings and less complete for some physician-administered drugs and those delivered through atypical channels (for example, specialty pharmacies); the data are also less complete in earlier years of the sample. (Some of these sampling features are unlikely to be unique to this data set.) Because of this, two recent studies have conservatively restricted analysis to years after 2010 or 2012.
The design features of the SSR Health data have important implications for how these data should be used and interpreted. With that in mind, we highlight a handful of “best practices” that can help researchers using these data, particularly if doing so for the first time. While not an exhaustive list, we focus on features that are most relevant for analysis.
The SSR data include an estimate of branded drug net prices from the perspective of the drug manufacturer that is net of all price concessions. In other words, how much do drug makers earn from sales of a given drug within the US market after all reductions are included? Implicitly, this average reimbursement reflects a mix of different transaction prices paid across settings and insurers. Estimated net prices may or may not be similar to the price paid by any specific payer. In particular, we emphasize that researchers should not conflate this measure with an estimate of net payments made by commercial market insurers. Furthermore, one should not attribute the difference between list and net prices exclusively to rebates.
The data from SSR Health include separate estimates of WAC-to-net discounts for the Medicaid and non-Medicaid markets in addition to aggregate estimates. However, as SSR Health notes, the Medicaid-specific estimates reflect some, but not all, discounts received by the program. They do reflect both the basic rebate (23.1 percent of the average manufacturers price [AMP]) and inflation-based rebate. (Because the AMP is not public, SSR estimates AMP as 97 percent of WAC.) However, they do not reflect the “best price” provision or additional supplemental rebates that state Medicaid agencies may negotiate. (For this reason, these estimates are labeled “statutory Medicaid.”) For a detailed discussion of rebates in the Medicaid program, see Rachel Dolan (2019).
As noted by the Congressional Budget Office, supplemental rebates are smaller than statutory rebates, but the exclusion of the best price provision is likely consequential. For example, among the 176 top-selling Medicare Part D drugs, the best price was 41.0 percent lower than AMP—much lower than the price associated with the basic 23.1 percent discount.
Because SSR Health does not include this provision, estimated Medicaid net prices are overestimated and, thus, net prices received by manufacturers for non-Medicaid sales are likely underestimated. These issues are less pronounced for drugs that have small Medicaid market shares (re-running analysis of non-Medicaid prices on such a sample could represent a plausible robustness check for some analysis using these data). However, in general, we caution against anything more than suggestive use of Medicaid or non-Medicaid specific estimates absent more formal attempts to incorporate the best price provision.
Researchers interested in producing estimates of, say, average WAC-to-net discounts across a basket of drugs must decide about how to weight the analysis. Conceptually, one could imagine two options: weight by an estimate of sales to calculate the average rebate per dollar spent or weight by a standardized measure of units to produce the average discount per, say, prescription sold. The SSR data allow researchers to do the former, but do not allow for the latter without further adjustments.
While SSR Health does not include units as a separate variable, one can theoretically approximate units by dividing total net sales by net price per unit. However, this is inadvisable because it can generate incorrect unit estimates for products with multiple strengths or forms. Furthermore, weighting by this metric is likely to produce estimates without a clear interpretation. Units are reported in different forms for different drugs, varying from grams, to milliliters, to “eaches”—for example, a pill or a single-use vial. (Note that SSR Health uses the National Council for Prescription Drug Programs’ billing units standard to define the units of each product.) Thus, net price estimates across products mean different things—“net price per gram,” “net price per milliliter,” or “net price per each.” There is no reason to expect units to have a sensible and consistent interpretation across these settings; a course of treatment may contain many units for a pill taken daily but a fraction of a unit for an injectable measured in milliliters. Even among drugs with the same units, the number of those units that make up a dose or standardized prescription can vary substantially. This same feature can make it difficult to compare the prices of a course of treatment for two drugs.
Weighting by units will deliver meaningfully different results than if researchers weighted by net sales, but not for reasons that we believe are advantageous for analysis. Consider Eliquis and Humira. In the first quarter of 2019, net sales of Humira were roughly 2.6 times larger than that of Eliquis. However, Eliquis is a tablet that is typically consumed multiple times daily while Humira is a self-injection pen that is typically administered once every other week. Because of this, unit weighting would give Eliquis about 165 times more weight than Humira in that year despite its smaller market share.
Weighting by units does not deliver prescription-weighted analysis, but instead gives more or less importance to drugs based on the units they happen to be expressed in. Weighting by units can be done if care is taken to standardize units appropriately (for example, using external data to define a “standard dose”). For the reasons listed above, SSR Health generally avoids weighting by units. As a default, we suggest weighting all analysis by net sales, although we acknowledge this is imperfect if price is the outcome of interest. Either way, it is best to be clear about the choice of weights and explore the sensitivity of estimates to that decision.
We note that SSR Health does offer estimated prices per year or course of treatment, which can be useful for some comparisons of drugs expressed in different units. However, these estimates require sensible, but substantial, assumptions—namely that all patients are of average body weight or surface area, take the maximum dose associated with the highest-dosed indication of the drug, and that, in cases where duration of treatment is not listed, a course is assumed to be a year. The strength of these assumptions will vary by the nature of each drug. Technically, one could back out the courses of treatment sold per year, but we would suggest some caution given the number of assumptions required.
Ultimately, the SSR Health data rely on the public disclosure of information from drug manufacturers. This fact shapes the sample of drugs included in at least two ways.
First, the data set excludes drugs made by private manufacturers. Fortunately, the vast majority of drug makers are public, with Boehringer Ingelheim as the primary current exception (Purdue is a notable exception in earlier years of data). Second, drug manufacturers typically only report on products with non-trivial sales. This means that many small-market drugs may never appear, and products will drop out of the sample as their market share erodes (for example, after generic entry).
Because they rely on sampling, volume estimates come with some level of imprecision. That imprecision can lead to temporary or systematic inaccuracies in net prices of drugs.
Specifically, net pricing estimates implicitly hinge on two key assumptions: Quarterly volume estimates for each drug reflect all sales of a drug; and inventories of each drug held by intermediaries (most notably, wholesalers) are constant. We consider each in turn.
Volume estimates may not accurately reflect all sales for a drug for multiple reasons. First, sampling-based strategies are more likely to deliver inaccurate volume estimates in cases where drugs have relatively small sales (this is unlikely to be unique to Symphony, or in turn SSR Health). In principle, these small-sample errors could lead to volumes that are too high or too low. Second, volume estimates may systematically only capture a portion of total sales. This is most likely for specialty drugs that are distributed through non-traditional distribution channels such as hospital specialty pharmacies. This is more likely to lead to consistently low than high estimates of volume sales, leading to systematic overestimates of net prices.
The constant inventory assumptions may also be violated if intermediaries are increasing or reducing their supplies. In such cases, volumes sold to intermediaries are not indicative of the number of units actually sold to end users. This can create “jumpy” net price estimates; however, it is often a relatively short-term problem as wholesaler inventories should not diverge from sales to end users in perpetuity.
In general, it is preferable to focus analysis on cases where these systematic errors in net pricing estimates are rare; or, when studying one or a few drugs, to verify that they are unlikely to be problematic. Researchers have a number of options for doing so:
Researchers could reduce some of the sampling-based variability by presenting annual averages for outcomes of interest.
Because volume estimates are less reliable for these products, we suggest avoiding their inclusion in analysis, although we acknowledge that any such rule is likely to be ad hoc to some degree. However, it is worth emphasizing that this is a much smaller concern if analysis is weighted by net sales.
If researchers are concerned that a particular drug or set of drugs may be subject to incomplete volume estimates, they can look for evidence of non-traditional distribution. For example, consider Xtandi and Xospata, both from Astellas Pharma. Published information about Authorized Distributors of Record show that, while the company has agreements with many distributors, both Xtandi and Xospata are excluded from them. Such drugs should likely be excluded.
Intermediaries are more likely to be building or reducing inventories surrounding a product’s launch or loss of exclusivity, raising the potential for erroneous estimates. When analysis focuses on one, or a small number of drugs, one can at least visually inspect pricing trends for evidence of unusual pricing patterns surrounding these dates. For more general analysis of net pricing, one could drop drugs within a few quarters of launch or loss of exclusivity. In cases involving sharp cutoffs—such as analysis surrounding the date of generic entry or similar—one could employ methods akin to regression discontinuity analysis (essentially calculate separate rolling averages on either side of the cutoff and compare values near the cutoff point).
If researchers are interested in the most accurate net pricing data, they could analyze only those drugs typically sold in retail settings where unit estimates are best. Operationally, one could employ the method outlined in Pragya Kakani and colleagues (2020) by excluding injectables, vaccines, devices, and many oncology drugs.
If researchers want to extend analysis beyond just drugs sold in retail settings, they can consider alternative approaches of limiting the sample, although these restrictions are imperfect. Some researchers have opted for exclusion criteria that drop drugs where net price estimates exceed the full list price of the drug—a pattern that should never be true. For example, one could drop drugs that ever have a quarter in which they have estimated net price that exceeds the WAC price. Less conservatively, one could drop drugs until they consistently have net prices that are smaller than WAC prices (say, a full year of meeting that criteria). While less conservative, the latter would avoid excluding drugs that have otherwise reasonable estimates once they pass their earliest, lowest-volume quarters.
While this clearly eliminates drugs with erroneous net price estimates, it raises some questions since it amounts to selecting a sample based on an outcome of interest for many studies. Specifically, dropping cases where net prices are above WAC prices may bias toward finding larger WAC-to-net price discounts for two reasons. First, this could only remove cases where inaccurate volume estimates cause prices to be too high, while ignoring cases where these errors cause them to be too low. That said, we believe it is more common for Symphony to underestimate volume owing to holes in its coverage, rather than overestimate them.
Second, and more problematically, the same level of error in volume estimates is more likely to push a net price above the WAC price if the true level of discounts is small. In other words, if a drug’s true net price is 95 percent of WAC, even a small overestimate will lead to its exclusion, while the same level of error would not do the same one with a net price at 60 percent of WAC. Thus, this exclusion criteria has the potential to overstate the average discount in a basket of drugs.
Data from SSR Health are rapidly changing research on drug pricing in the United States by improving the ability to estimate the net prices paid for branded drugs. For many questions, these data represent a meaningful improvement over publicly reported list prices or prices associated with a single public purchaser. That said, they have important features and underlying assumptions that should be recognized by researchers when using them. With the appropriate methods, and an understanding of their limitations, these data can continue to improve our understanding of the modern domestic drug market.
We thank Anna Anderson-Cook, Jerry Anderson, Josh Feng, Steven Hill, Pragya Kakani, and Luca Maini for very helpful comments on earlier drafts of this piece as well as Richard Evans and Scott Hinds of SSR Health for many helpful conversations. SSR Health was given an opportunity to review this piece for accuracy and to ensure no disclosure of proprietary information. All views expressed are those of the authors.