History-Dependence in Drug Demand
Patients Keep Taking the Same Drug Over Time
My research on prescription drug demand starts from the fact that patients generally keep taking the same drug from year-to-year to manage chronic conditions.
The table on the right shows that year-on-year, patients on anti-cholesterol medication continue on the same drug at an 80% rate, with few patients switching to other drugs. This pattern holds for a variety of chronic drug markets, including diabetes, HIV/AIDS, high blood pressure, and asthma.
History-Dependence?
Persistence in choice could occur for two reasons:
The drug is a great fit for the patient
History-dependence: assigning a patient to any drug means that they will keep taking the drug
To separate these out, I construct a quasi-experiment around new drug introductions. I use MarketScan claims data (1996-2013) throughout the study.
The graph on the right shows the 2007 market share of Crestor, which was introduced in August 2003, by the quarter in which a patient starts on anti-cholesterol medication (their "cohort"). The cohorts to the left of the red line were forced to start on other anti-cholesterol drugs, whereas the cohorts to the right could start on Crestor.
The graph shows that cohorts to the right are persistently more likely to choose Crestor, even four years after Crestor is introduced. In other words, some people in the cohorts to the left of the line would be good fits for Crestor, but end up not switching over because of history-dependence.
Additional Findings and Implications for Demand Modeling
Additional Findings
The average effects of history-dependence are large and sustained. Quasi-randomly assigning a patient to a new drug leads to a 54 percent chance they are still taking the drug 5 years later.
Generics and line extension drugs exhibit lower degrees of history-dependence. Generics benefit from substitution laws that help it attract over 80% of the patients taking the reference brand after entry. Line extensions also attract some share of the original branded drug (about 20% of existing patients). Both types attract very few switchers from other brands.
Towards a General Switching Cost Model
I argue that a pair-specific switching cost model of demand is required to accurately capture the patterns in the quasi-experimental data. In other words, while patients generally switch at very low rates, some switches are more frequent than others.
I use an SMLE approach to incorporate switching costs and unobserved heterogeneity and estimate it using MarketScan data.
I show that the pair-specific model (vs. drug-specific switching costs typically used in the literature) is particularly useful for modeling drug choice when there are generic drugs on the market.
Implications for Health and Markets
Health implications
A first-order concern with history-dependence is that patients are not switching to better medication.
Patients who are older or take multiple medications exhibit greater history-dependence in their drug choice. Taking multiple medications increases the chances of negative effects of interactions.
Using data on incremental quality-adjusted life years (QALYs) for a broader set of drugs approved in the period 1999-2012, I also find that entering drugs that are clearly better than existing options are less history-dependent. In other words, patients are more willing to switch to them.
Overall, it appears that the degree of history-dependence varies in a way that mitigates concerns about the impact of history-dependence on health
Implications for Prescription Drug Markets
Another concern is how history-dependence affects competition in drug markets. Klemperer's seminal work on competition in markets with switching costs suggests that history-dependent demand could lead to higher prices and encourage entry deterrence.
The structure of drug markets is more complex than the stylized structures used in theoretical modeling, so I leave a rigorous investigation of pricing to a different paper. I also discuss the use of line extensions as a way to deter generic entry, because the traditional approach of investing in market share is neutralized by generic substitution laws.
Instead, I provide suggestive evidence that history-dependence alters drug manufacturers' entry decisions in the context of priority review vouchers. Priority review vouchers are given to companies that successfully commercialize drugs treating rare diseases, and allow companies to shorten FDA review by four months. Crucially, vouchers are transferable.
I find that, on average, PRVs are purchased for $177 million on average, but the first four months' worth of sales averages to $60 million. This suggests that manufacturers take history dependence into account when applying vouchers. Vouchers have also been applied to chronic drugs facing subsequent competitor entry, suggesting that manufacturers are using vouchers to leverage history-dependence to gain a first-mover advantage.