- CON laws don’t reduce mortality
- Why don’t people like pharma companies?
- Just give ‘em cash.
- Is value-based purchasing being undermined?
- Cancer drugs don’t live up to the hype?
Tomas Philipson has an interesting post in Forbes describe how value should be incorporated into the U.S. healthcare system but should be done in way suited for the American market.
The American healthcare system is at a crossroads. Shifting to a system that reimburses based on value, rather than volume, of care requires changes in how we evaluate healthcare and make spending decisions. Elsewhere in the world, governmental agencies perform health technology assessment and make coverage recommendations based on cost effectiveness and other metrics. Such an approach runs contrary to fundamental elements of American healthcare, however. To successfully move to value-based healthcare in the U.S., distinctly American approaches to measuring value must be developed that meet several key criteria.
A JAMA Oncology paper estimating the global burden of cancer is getting a lot of attention in the press. The study’s key findings are:
In 2015, there were 17.5 million cancer cases worldwide and 8.7 million deaths. Between 2005 and 2015, cancer cases increased by 33%
A 33% increase in cancer cases!!! There must be an epidemic!!! Why are so many people getting cancer all of a sudden!!!
It’s time for everyone to calm down a bit. Let’s break down these numbers a bit more. Of the 33% increase in cancer cases over the last 10 years, 16% is due to population aging, 13% to population growth, and 4% due to age-specific rates. Let’s take these one by one.
Cancer is a disease that largely occurs in old age. One good way to avoid getting cancer is to die young. Countries with high infant mortality rates, war, contagious disease (HIV, Zika, etc.) are likely to have lower cancer rates. This is not because they are healthier, but because people die before reaching an age when they would get cancer. This concept is known as the the competing risks. Thus, observing that more people are dying of cancer could mean that people are living longer and are less likely to be dying from other diseases.
The second component of cancer growth is population growth. If cancer incidence as a proportion of the population were constant, clearly an increase in the number of people living on the planet would increase the number of cancer cases. Population growth in and of itself, however, is not a bad thing.
Now we get to the bad news: Age-specific mortality rates increased by 4%. That means, for a given person aged 50-54, or aged 55-59, the likelihood they would get cancer increased by 4%. However, this statistic blurs differences between countries. Age-specific mortality rates decreased by 0-10% in most of the Americas and Western Europe, decreased by 10-20% in Russia and Eastern Europe, and decreased by more than 20% in China. Many African countries, in contrast, saw increasing age-specific morality rates.
Further, if you look at the age-standardized absolute years of life loss metric (AS-YLL), for 24 of 31 cancers the AS-YLL decreased (i.e., there were fewer years of life lost) and for 9 of 31 cancers there was no statistically significant change in AS-YLL.
In this post, I do not mean to play down the severity of cancer. It is a serious disease and additional efforts are needed to improve the care and outcomes of cancer patients. What I do want to highlight, however, is that it pays to did a bit deeper into the statistics and not always rely on the headlines.
An article by Ajay Agrawal and Avi Goldfarb (“The Simple Economics of Machine Intelligence“) provide an interesting perspective on how machine learning will affect employment with a nice example from the health care sector.
All human activities can be described by five high-level components: data, prediction, judgment, action, and outcomes. For example, a visit to the doctor in response to pain leads to: 1) x-rays, blood tests, monitoring (data), 2) diagnosis of the problem, such as “if we administer treatment A, then we predict outcome X, but if we administer treatment B, then we predict outcome Y” (prediction), 3) weighing options: “given your age, lifestyle, and family status, I think you might be best with treatment A; let’s discuss how you feel about the risks and side effects” (judgment); 4) administering treatment A (action), and 5) full recovery with minor side effects (outcome).
As machine intelligence improves, the value of human prediction skills will decrease because machine prediction will provide a cheaper and better substitute for human prediction, just as machines did for arithmetic. However, this does not spell doom for human jobs, as many experts suggest. That’s because the value of human judgment skills will increase. Using the language of economics, judgment is a complement to prediction and therefore when the cost of prediction falls demand for judgment rises. We’ll want more human judgment.
How will machine learning affect physician employment? The answer is unclear. The value of physicians for prediction purposes will drop. Why ask a doctor what drug works best when you can ask Watson. However, the value of physicians for judgement will expand. How do physicians take into account different patient preferences? Can physicians identify patient-specific idiosyncratic factors not readily apparent in prediction models that make one treatment better suited for a given patient than another.
One option is that physicians will become more of a luxury good and low-cost, Watson prescribing could be used for less complex patient cases with more straightforward recommendations. Physicians could be used more for more complex cases and also for higher income individuals or less tech savvy individuals who do not want to or are not able to interact with a machine learning technology.
Nurse employment may become more valuable because all the treatment still need to be administered, likely by a person and not a machine.
In short, machine learning is likely to provide huge benefits for patients, but the effect on who will deliver care is unknown.
The annual health spending numbers from CMS’ Office of the Actuary (OACT) show that health care spending is increasing as a share of national income. A study by Martin et al. (2016) estimates that health care spending now makes up 17.8% of the U.S. economy, by far and away the largest percentage in the world.
Total nominal US health care spending increased 5.8 percent and reached $3.2 trillion in 2015. On a per person basis, spending on health care increased 5.0 percent, reaching $9,990. The share of gross domestic product devoted to health care spending was 17.8 percent in 2015, up from 17.4 percent in 2014. Coverage expansions that began in 2014 as a result of the Affordable Care Act continued to affect health spending growth in 2015. In that year, the faster growth in total health care spending was primarily due to accelerated growth in spending for private health insurance (growth of 7.2 percent), hospital care (5.6 percent), and physician and clinical services (6.3 percent). Continued strong growth in Medicaid (9.7 percent) and retail prescription drug spending (9.0 percent), albeit at a slower rate than in 2014, contributed to overall health care spending growth in 2015.
Value-based pricing has become all the rage of late among health policy wonks. Medicare aims to tie 90% of reimbursement to some measure of value by 2018. The AMA has endorsed value-based pricing for pharmaceuticals. Organizations such as IVI and ICER propose different approaches for measuring value as well.
Typically, value is measured as the health benefits patients with a disease receive multiplied the monetary value of a 1-unit increase in health (often the value of a QALY), less the cost of the treatment.
However, a recent working paper by Lakawalla, Malani and Reif claim that this standard approach misses the value healthy individuals receive from a treatment. Further, they claim that the value of a treatment to healthy people is very large.
Why would healthy people value a treatment for a disease they do not have? In short, risk aversion. Patients may not have the disease, but typically have some positive probability of having the disease later in life. Thus, the availability of new technology provides a form of insurance that in the case that the person does have the disease in the future, it will be much less burdensome.
The authors provide the following example:
Think of a healthy consumer facing the risk of developing Parkinson’s disease3 in the years before the discovery of effective dopamine-related treatments that reduce disease symptoms. In the absence of a treatment, contracting Parkinson’s might reduce her quality of life on a 0 to 1 scale from .8 to .4. The consumer could not insure herself against this risk, because real-world insurers did not (and do not) sell pure indemnity insurance contracts that make payments to consumers conditional on the occurrence of illness alone. Moreover, healthcare insurance was a poor substitute, because there were no effective treatments to cover. As a result, this consumer had to bear the full risk of Parkinson’s related symptoms herself.
Now consider the introduction of a new technology like Levodopa. Before Levodopa, Parkinson’s patients stood to lose 50% of their quality of life. After Levodopa, however, developing Parkinson’s becomes less costly, lowering quality of life from 0.8 to only 0.7. Notice how Levodopa’s introduction compresses the variance in the quality of life between the Parkinson’s and non-Parkinson’s states. This compression is valuable to consumers who dislike risk. In addition, the advent of the treatment immediately makes healthcare insurance more valuable, because there are finally effective technologies for insurance to cover. Conventional economic valuations of health technology ignore these two sources of value, but focus exclusively on the 0.3 gain in average quality of life produced.
Lakdawalla and co-authors separate out two forms of insurance value above and beyond the standard approach described above (the authors describe the standard, risk-free approach as the risk free value [RFV]). First, there is self-insurance value (SIV), which is the additional value of a treatment that accrues to uninsured individuals since their risk of a severe decline in health in the even that they contract a disease decreases. The second form of insurance value is the market-insurance value (MIV), which the authors state “represents the incremental value of being able to use health insurance to substitute for the indemnity insurance market. Medical technology is essential to this substitution because health insurance can only be used to fund consumption of medical care.” MIV occurs because typically patients owe a copay or coinsurance. Reducing the copayment or coinsurance when patients are sick is valuable because their baseline utility is lower and thus a marginal increase in utility in a sick state is worth more than in a healthy state due to the decreasing marginal utility.
Overall, this paper takes some standard economic concepts—risk aversion—and applies them to a topical debate about how to measure value for new treatments. Other paper such as recent work by Sanders et al. (2016) have already cited the insurance value components as being useful for inclusion in future cost-effectiveness studies.
This is the title of my latest article in JMCP co-authored with Arijit Ganguli, Yuri Sanchez Gonzalez, Jin Joo Shim, and Seth A. Seabury. The paper’s abstract is below.
BACKGROUND: There is considerable push to improve value in health care by simultaneously increasing quality while lowering or containing costs. However, for diseases that are best treated with comparatively expensive treatments, such as rheumatoid arthritis (RA), there could be tension between these aims. In this study, we measured geographic variation in quality, access, and cost for patients with RA, a disease with effective but costly specialty treatments.
OBJECTIVE: To assess the geographic differences in the quality, access, and cost of care for patients with RA.
METHODS: Using large claims databases covering the period between 2008 and 2014, we measured quality of care metrics by metropolitan statistical areas (MSAs) for patients with RA. Quality measures included use of disease-modifying antirheumatic drugs (DMARDs) and tuberculosis (TB) screening before initiating biologic DMARD therapy. Access to care measures included measured detection and the share of patients with RA who visited a rheumatologist. Regression models were used to control for differences in patient demographics and health status across MSAs.
RESULTS: For the 501,376 patients diagnosed with RA, in the average MSA 64.1% of RA patients received a DMARD, and 29.6% of RA patients initiating a biologic DMARD appropriately received a TB screening. Only 17% (73/430) of MSAs comprised the top 2 Medicare Advantage star ratings for DMARD use. Measured detection was 0.59% (IQR = 0.47%-0.71%; CV = 0.355) on average, and 57.6% (IQR = 48%-69%; CV = 0.341) of RA patients visited a rheumatologist. MSAs with the highest DMARD use spent $26,724 (in 2015 U.S. dollars) annually treating patients with RA, $5,428 more (P < 0.001) than low DMARD-use MSAs, largely because of higher pharmacy cost ($5,090 vs. $7,610, P < 0.001). However, MSAs with higher DMARD use had lower RA-related inpatient cost ($1,890 vs. $2,342, P = 0.024).
CONCLUSIONS: There were significant geographic variations in the quality of care received by patients with RA, although quality was poor in most areas. Fewer than 1 in 5 MSAs could be considered high quality based on patient DMARD use. Access to specialist care may be an issue, since just over half of patients with RA visited a rheumatologist annually. Efforts to incentivize better quality of care holds promise in terms of unlocking value for patients, but for some diseases, this approach may result in higher costs.