Embedded Files
Drug Price Control Analyses
Motivation: A Freeloading Problem, Answered With Beliefs
On May 12, 2025, the Trump administration claimed to end global freeloading of the U.S. pharmaceutical R&D by going straight with a cutoff: "The Most Favorable Nation Drug Pricing Policy" based on OECD countries whose GDP per capita is higher or equal to 60% of the U.S. standard (in short, most of EU members plus Australia, Canada, and Israel). The executive order frightened many, many, many industry writers, who directly wrote about the potential harms since they are directly suffering as frontline players. Next, from academia, Lakdawalla and Goldman of USC wrote the most persuasive comments on this since the White House announced the baselines, and followed with another after the press release ended this May. However, even with the previous papers cited, what could be the real numbers of the new policy? Afterwards, we can easily find YouTube videos and interviews with professors about the policy, and most answered with beliefs, but nobody answered this question, even with the boldest models or simulations.
As a Ph.D. student here at the University of Toronto, I can claim that there are no conflicts of beliefs to yay or nay the policy (as long as it is not about tariffs), but I do want to build toy models and create the skeletal features which people can easily plug in their real numbers (in the HEOR world, they call this Real World Evidence---RWE) and restate results at any given time.
Goal and Search: Find the Upper Bound and Funnel Down
Goals are easily tweaked based on the political beliefs, whereas the potential scenarios of "a policy will eventually be revoked" may jump out at any split-second. Hence, I am only aiming for the "maximum" benefit as an optimistic approach, where we can check the requirements to reach such aims and whether they make sense or not. Besides, if for any given scenario, we have a negative aggregated effect on the policy implementation, it becomes a strong signal that the policy may face larger pushbacks than the stated benefits.
The net benefit on monetary savings based on a perfect reaction to policy is a simple line of mathematics, so I am diving deeper into the parts of health-related quality-of-life (HRQoL) in the following sections. Therefore, we need to slice and dice the benefits of monetary savings on health, focusing on two key areas: medical compliance (minor) and mental health gains (major). Furthermore, the costs of HRQoL loss are based on the lost potential gains of HRQoL (major) and market access (minor); since the market access in the U.S. is not fully observable under current literature without more transparent ways to evaluate if a firm even tend to quit the market, I am only including the trend tweak at this moment. Aggregating the two above, we can find a net effect through HRQoL alterations.
Also noteworthy, since QALY is banned in federal policies and not eligible for either Medicaid or Medicare analyses, I will include some generalized cost-effectiveness analyses (GCEA). In this project, I primarily apply the generalized risk-adjusted cost-effectiveness (GRACE) model by Lakdawalla and Phelps (2023). Therefore, I will take a discounted HRQoL sum (which does not linearly multiply life-years total) and compare it to the GRACE results in terms of monetary benefits as a primary scope.
BTBA_MFN_Hsu_Lawrence.pdfSome Quick Notes on Q&A Sections (BTBA-based)
[1] From which papers did I get the data?
- Ara & Brazier (2011) for R&D trends based on time (for kink calculation)
- Ayyagari & Shane (2015) for disease-specific HRQoL for all trends
[2] How to Find Cost and Revenue of Firms? Are They Public?
- Including all failed projects, we do not own full-scale R&D costs
- We only know a typical range of $1-3B by the Congressional Office
- Wouters et al. (2022) used STROBE for 2009-2018 to estimate costs
- Sertkaya et al. (2024) worked on 2000-2018 data and showed global revenue
- Caveat: We do not know how the R&D is growing as well as investment and revenue distribution
[3] Why Applying Block Recursive Equilibrium and How?
- Picturing the current input and revenue structure at a macroeconomic level
- To re-estimate the “entry propensity” (λ) after shocks occurred
- To check such work in another format, we need to move the law of motion
- Check for slowed down “meeting rates” or “upshifts” (μ) if fixing λ
- If fixing both μ and λ, check “exit rates” of each step (δ)
- If the other two are fixed, λ drops, or μ drops, or δ increases
- The slice is made by revenue loss, which is only reflected on the top
- Kennedy et al. (2022) aim at oncology ver. of big pharma vs small biotech
- Agility is merged into lower costs but lower potential revenue (Inada Condition on θ)
- Caveat: Too simple and too many restrictions, with the major problem of size affecting potential revenue (θ), might not work on all drugs (e.g., asthma with SABA, SAMA, which we used for long enough)
[4] How Did the BRE Formula Work?
- Fixed cost of R&D every stage of n blocks (I set 30,000)
- The proportion of R&D blocks (α) depends on the maximum of revenue
- R&D cost is set as log-linear where θ → C(θ) : [0 , 1] → [1B , 3B]
- θ is the “HHI × market proportion” linear transformed into [0,1]
- The revenue cut slices off “half of the price cut” in the top section
- The simple calculation ends with a square root difference of 20%-30%
- The cumulative cost is linear, and by including δ, it becomes exponential
- The cumulative revenue is exponentially growing, and δ accelerates it
- We calculate λ loss based on the cut, reflecting the growth trends of HRQoL
- Caveat: Unknown cost functions are the major pain of this entire model
[5] How to Get the Targeted Results?
- Set up 1000 slots with Medicare and Medicaid population and US Census
- Split the directly affected population on prices (including IRA features)
- Note that the projected mental gain next will be universal!
- With the mixed population, I set up another 1000 simulation slots with a linear combination of diseases based on the probability based on age
- With the “sick-generated” and “eligible features,” run discount rates with expo-HARA (Phelps, 2024) for current states
- Toss in compliance, mental, trend differences, simulation lifetime differences, average them for one number and simulate this 1000 times
- For GRACE results, simply run multipliers on the exact same result and multiply it by the benefits from the last step
[6] How Do We Explain Such Results?
- “Feeling richer” is a typical thing of medication willingness-to-pay
- Not fully mentioned but modelled: it triggers a higher rate to visit physicians- Mental health stability is the topic of burnouts, and we save from a societal perspective with the extension of job loss problems
- The transformation is based on the Bondareva-Shapley Theorem, where the Shapley Value crosses the boundaries in a uniform fashion, and thus linearly points to gains in health whenever it crosses through the given border
- The best outcomes are in optimistic hands
[7] Why GRACE? How is it a Game Changer?
- Since health loss and exacerbation of mental health problems are on a growing trend, it weights much more in every step of economic evaluation
- For the very sick population, R&D stagnation could leave severely ill populations untreated, which is potentially 8-11 times worse from my previous stroke work (Hsu, 2024)
- Mulligan et al. (2024) show the standard gamble results, so the general public does not want to pay that much (about 0.71 times only), so the results were shrinking by all means
- With the 40% price cut, benefits shrank to around $2000 per person, which is already fed by the severely ill
- With a 60% price cut, things go negative; we are not going to see it work
[8] Problems I Cannot Solve?
- R&D competition with IO models, need to work deep into this topic
- Specified GRACE results could be different because to standard gamble
- The structural answers are from simulated results, so no great regression nor spans could be estimated in this paper
- The rate of catching each of the diseases is different! The smoothed lines are just working for the older, but we need granular features
What's Next? How to Guide This Model?
As mentioned above, numerous assumptions need to be addressed, and I cannot complete this task alone. If anyone from the industry aims to test the model with their data, I am willing to collaborate. Please contact me at lawrence.hsu@mail.utoronto.ca for further details. I can show the plugged-in results and hear your opinions about what the industry disagrees with.
Besides, my 3 major aims are: (1) The true effect of caregiver burden interventions with cost-savings (with additional emphasis on mental health states); this is a labour economics topic. (2) The reaction and new equilibrium from competition between pharmaceutical companies and biotechnology firms under the policy shock; this is an industrial organization (IO) topic. (3) How can such policies work, or what is the optimal penalty if not following the rules, going with fines and transferrable credits of any government; this is a public economics (public policy) topic. Estimating to update the story somewhere in 2025-26, so keep posted, we will see how things move on!
Disclaimer and Land of Acknowledgement:
Lawrence C.-H. Hsu thanks all who guided him through the process: Thanks to Richard Spady (JHU) for his precious guidance on structural model building, Syedmehdi Rizvi (UMMC) for empirical clinical knowledge, Diego Martinez (JHU, PUCV) for more of an engineering perspective, Ronald Wolthoff (UofT) for the guidance through BRE models in his Macroeconomic Theory class, and the BTBA committee for accepting my preliminary work. All errors are mine.
I (we) wish to acknowledge this land on which the University of Toronto operates. For thousands of years it has been the traditional land of the Huron-Wendat, the Seneca, and the Mississaugas of the Credit. Today, this meeting place is still the home to many Indigenous people from across Turtle Island and we are grateful to have the opportunity to work on this land.
Page updated
Google Sites
Report abuse