Maybe the 340B Drug Pricing Program Costs Employers $3.40B

We’re constantly making predictions. Will it rain? How will they respond if we do or say X? How much do I need to save for retirement? Some of us make better predictions than others. When explaining his success, Wayne Gretzky said, “I skate to where the puck is going to be, not to where it has been.”

“All models are wrong, but some are useful,” observed statistician George Box. At best, a model is a simplification that may yield useful estimates or predictions. At worst, they can lead us astray.

IQVIA recently estimated that the 340B drug pricing program costs self-insured employers and their workers $5.2B per year in lost rebates. Perhaps because of confirmation bias, many quickly gave that number unabashed support. Not so fast.

How wrong is this estimate? How useful is this estimate?

How wrong

To estimate rebates for branded, self-administered products, IQVIA’s model was based on 14 products spanning four therapeutic areas — diabetes, HIV, immunology, and oncology — to “represent different mixes of product types and different degrees of competition.” We don’t know which products IQVIA included in its analysis, but Humira is a highly rebated product and, presumably, was one of the 14. Which were the other 13?

Exclusions are important, too. Were some products and therapeutic areas considered or initially included but ultimately excluded from the final analysis? If so, why? We might have greater confidence in the results if the inclusion and exclusion criteria were transparent.

The degree of competition and the extent to which it impacts rebates is a function of several factors:

• The number of treatment options
• Market shares of the products considered
• Whether the mechanism of action is the same or different
• Whether clinical performance (i.e., safety and efficacy) is comparable among products
• Recommendations by guidelines and key opinion leaders
• Technology assessments (e.g., ICER)
• Availability and number of generics or biosimilars, and whether a biosimilar is interchangeable with the reference product

Can 14 products spanning four therapeutic areas adequately represent all relevant rebate scenarios? No. So, are results from a model based on so few products and therapeutic areas likely generalizable? No.

More products could’ve/should’ve been included. Ceteris paribus, 10x (i.e., 140 products) would likely have yielded more stable estimates. And 100x (i.e., 1,400 products) would have been even better. Does that sound crazy? It has been done. An analysis by USC’s Leonard Shaeffer Center for Health Policy & Economics calculated the difference between list and estimated net prices for 1,335 brand drugs from 2015 to 2018.

And while it made sense for IQVIA to include diabetes, HIV, immunology, and oncology products, the model would have been more representative had it included other high-rebate drugs such as:

• PCSK9 inhibitors
• Hepatitis C antivirals
• Direct oral anticoagulants (DOACs)
• Disease-modifying therapies (DMTs) for multiple sclerosis
• Vascular endothelial growth factor (VEGF) inhibitors
• GLP-1 inhibitors

IQVIA noted the “uncertainty in how representative this sample of disease areas was” and tested values at 5% above and below the weighted means for estimated rebates. It is common to test values surrounding point estimates, but the rationale to test values at 5% above and below the means was not provided. A sensitivity analysis should have been based on the expected or observed distributions underlying the means. Even better, IQVIA could have used a Monte Carlo simulation to sample the probability distributions for each input variable and modeled hundreds or thousands of possible outcomes for each scenario of interest.

IQVIA’s model estimated:

• WAC, manufacturer rebates, and 340B claim conversion rates (the estimated percentage of 340B-eligible product that was purchased at 340B prices) each for branded self-administered drugs, generic self-administered drugs, branded physician-administered drugs, and generic physician-administered drugs
• Annual healthcare costs per worker for self-administered drugs, physician-administered drugs, and other physician-delivered medical services
• The PBM pass-through (i.e., the percentage of rebates the PBM passes back to the plan sponsor)
• The percentage of workers receiving care from 340B providers in some unspecified way based on estimates of the percentage of 340B sales and 340B conversion rates

That’s a whole lot of estimation!

Among the explicit limitations, IQVIA notes that, for simplicity, “the model did not take into account the plan’s benefit design, plan member demographics, or any change in behavior in reaction to the tested scenarios.” That’s quite a simplification!

So, given the estimates, limitations, and simplifications, does the 340B program cost self-insured employers $5.2B in lost rebates? Not likely. Maybe it is higher — or maybe it is lower. Maybe the 340B program costs employers, well, $3.40B.

Then there is the elephant in the room: the variable not included in the model. What IQVIA does not mention, even in passing, is that 340B savings offset some of these costs. IQVIA’s analysis estimates only the cost of lost rebates, not the net cost of the 340B program to self-insured employers. That’s like a cost-benefit analysis that includes the cost but excludes the benefit.

Here’s the logic for savings:

1. Some 340B providers pass on some drug discounts to uninsured and underinsured patients
2. Some of these discounts improve access to drugs
3. Some of this improved access increases drug utilization and adherence
4. Some of this increased drug utilization and adherence yields medical-cost offsets (a common argument by drug manufacturers)
5. Some of these offsets reduce hospitals’ uncompensated care (e.g., charity care and bad debt) and reduce insurance premiums
6. Some of the reduced uncompensated care lowers hospital costs
7. Some of the lower hospital costs reduce insurance premiums
8. Lower premiums reduce employers’ costs

Further, the 340B program revenue is used to fund other medical services. It is reasonable to assume that some of these services also lowers employers’ costs.

While 340B savings have not been rigorously measured (unfortunately, 340B providers are not required to report how they use 340B revenue), they could be estimated and assumed to offset some of the estimated cost of lost rebates. Though the current evidence of savings is not strong, an argument that “drugs yield cost offsets except when provided through the 340B program” is weak. Remember, absence of evidence is not evidence of absence.

How useful

So, even if IQVIA’s model is wrong, is it useful? Yes.

IQVIA’s analysis surfaces a potentially important issue for self-insured employers in aggregate. In response, some employers may consider the extent to which the 340B program negatively impacts their own rebates. Most employers will likely experience some lost rebates, but the actual leakage will vary and range from 0% to 100%. For some employers, IQVIA’s “counterfactual” scenario where no drugs were 340B-elgible is, in fact, their actual experience. For example, some employers have employees that live and work in locations that fall outside a 340B provider’s service area. On the flip side, a self-insured employer may itself be a 340B provider (e.g., a hospital) experiencing a total loss of rebates, but simultaneously enjoying the discounts. Self-insured employers at both ends of this continuum likely care less about the lost rebate issue.

Loss of rebates due to the 340B program is less of a problem for self-insured employers that:

• Adopt a low-net-cost formulary that prefers low-WAC drugs without rebates over products with the largest rebates, or exclude a small number of high-rebate products like Humira from the formulary to encourage biosimilar competition
• Take an aggressive utilization management approach that reduces inappropriate utilization and/or utilization not supported by evidence
• Leverage alternative funding programs to exclude some specialty drugs from their formulary
• Strike a deal with a 340B provider to give up rebates in exchange for discounted drugs
• Avoid 340B utilization by excluding 340B providers from their provider networks

For self-insured employers that newly become aware of lost rebates due to the 340B program, IQVIA’s analysis may encourage them to consider their options. And some solutions may not be good for drug manufacturers. For example, this analysis could result in an uptick in agreements between employers and 340B providers. Perhaps unintended consequences like this were not considered by the National Pharmaceutical Council, which funded IQVIA’s analysis and reviewed the white paper’s manuscript and provided feedback.

The analysis may also encourage the American Hospital Association, America’s Essential Hospitals, 340B Health or others to more rigorously measure the savings that discounted drugs and other services provided to the uninsured and underinsured have on self-insured employers.

IQVIA’s model is fragile, but it can encourage evidence generation and inform the design of a stronger model that gets us closer to the truth. We should not let perfect be the enemy of good. But we also should not assume an initial model is good enough. In sum, IQVIA’s model is wrong but useful.