You’re a risk pool of one

A new way to think about genetic testing, based on the economics

Can I tell you something surprising I learned recently?

An estimated 5% of women have a genetic variant that causes them to metabolize the hormones in hormone-based birth control more quickly than average. For about a third of those women, this causes their hormone level to fall below the threshold required to suppress ovulation (and prevent pregnancy). In other words, it can render hormonal birth control methods ineffective (study here).

This was one study of 350 women, and the authors acknowledge that their findings need to be replicated with a larger sample. But assuming they’re correct, this completely changes how most women understand the effectiveness of birth control. We’ve thus far believed that birth control is 90–95% effective for all women who take it. Instead, these findings suggest that birth control (when used correctly) may be closer to 99% effective for 95% of women who take it, and closer to 0% effective for some of the 5% of women with this variant. And the only way to know, in advance, whether you’re the 1-in-20 woman for whom hormonal birth control could be ineffective is through a genetic test.

Birth control isn’t the only medication where the effectiveness can be determined by genetic factors. Antidepressants, statins, blood thinners, pain killers, and anesthesia are just a few examples where drug response and/or likelihood of severe side effects can be predicted (in part) based on genetics.

Some data points:

• An estimated 10–30% of hospitalizations due to adverse drug response can be attributed to genetic factors (citation)
• 7–29% of patients experience statin-associated muscle pain and weakness due to poor uptake of the drug by the liver and over-accumulation of it in the bloodstream, which can be predicted based on genetics
• 5–8% of patients experience hypersensitivity to the antiretroviral, abacavir, commonly used in HIV treatment. This can be life-threatening and testing for genetic predisposition has become standard of care (citation)
• Nearly 50% of dosing variability (from 0.5–20mg) of Warfarin, the most widely used anticoagulant in the world (used about 1% of the population) is explained by genetic factors along with age and body surface area. This finding has been widely replicated and included on FDA labels, yet is inconsistently applied in clinical practice (citation)
• An estimated 25% of the variation in antidepressant and antipsychotic drug response is due to genetics, primarily related to two genes — CYP2C19 and CYP2D6 — that impact drug metabolism. Allele frequencies for poor, intermediate, and ultra metabolizers (all requiring different dosing and treatment choice) are as high as 58% for some ethnicities (citation)

Unfortunately, with a few exceptions, many patients only find out that these drugs are ineffective or have intolerable side effects after experiencing them. The term of art here is “trial and error.”

As patients, we expect our doctors to protect us from these situations. And sometimes they do — they ask about our past medical history, known allergies, and family history. They have software to flag drug-drug interactions and sometimes voice-over added warnings of signs and symptoms to watch out for when starting a new drug regimen. And sometimes they recommend a genetic test to predict drug response. But outside of a few specialties like oncology, this is rare. Human genetics and genomics represent less than a semester of the medical school curriculum and few clinicians outside of academic medical centers are able to keep up to date with the rapidly changing pharmacogenomics literature and landscape. Furthermore, integrating these findings into clinical practice requires large-scale, prospective studies which are expensive and time-consuming. Their impact is typically to limit who gets a drug so are unlikely to be funded by drug companies. This leaves slower-moving institutions like academia, government, and the military.

One stakeholder I’ve always expected to be an eager adopter of pharmacogenomic testing is insurance companies. After all, why would they accept to pay for drugs that don’t work? The answer is economics. The economics of pharmacogenomic testing for insurance companies suffers from two problems:

Problem #1: Genetic tests today are designed to answer approximately one question. This is an artifact of what insurance is willing to pay for. We could do comprehensive sequencing of individuals, ideally at birth, and consult that dataset to answer many questions over one’s lifetime. That would very likely be more cost-effective than running multiple single gene or multi-gene panels. But insurance time horizons are short and they are generally only willing to pay for things that have clinical utility and ROI over a 2–4 year period (mapping to their average customer lifetime because, after all, people change jobs, change insurance providers, etc. about once every 2–4 years).

Problem #2: Assuming one test answers one question, the economics of pharmacogenomic tests that individually have a big impact on a small % of patients often don’t make sense. Here’s an example: let’s say I created a $100 test (which is unfathomably cheap for any clinical test) for the variant I described above that impacts birth control effectiveness. I would give this test to 100 people to find 5 with the variant of interest. The effective cost for each person I “found” would be $100 / .05 = $2000. So is it worth $1000 per year (assuming a 2-year customer lifetime) to an insurance company to avoid birth control maybe not working? Probably not.

Below is a quick model of this concept. Insurance (or anyone) would pay for a test (of any kind, including pharmacogenomic tests) when the below is true:

Cost of Test / Prevalence < Annual Savings * Customer Lifetime

So when do these economics make sense?

1. Cheap test — definitely not provided by the vast majority of clinical labs who charge >$1,000 for most tests, largely anchored around what insurance will reimburse for a small subset of patients.
2. High prevalence — includes tests that identify a lot of people who may be non-responders or have adverse effects, or alternatively, tests that answer more than one question. In the latter case, the prevalence of any one problematic variant may be low, but across many variants and disease areas the likelihood of uncovering something clinically relevant grows. Employer-based insurance probably doesn’t want to pay for these given that much of their value will accrue beyond their 2–4 year time horizon, but a stakeholder with a longer customer lifetime could.
3. Significant cost savings — insurers do already pay for genetic tests when patients might need expensive drugs (like oncology drugs) or drugs with very costly side effects (like abacavir). And as new drugs, such as for Alzheimer’s and dementia, are approved, insurers may see a positive ROI for pharmacogenomics tests that identify patients most likely to benefit from these treatments. Unfortunately, hormonal birth control, SSRIs and many other medications that could be tested for response to are generics. This means they’re quite cheap and giving them to people, even if they don’t work for several months, isn’t that big of an expense. Side effects and drugs not working are, of course, a big deal, but don’t necessarily factor into insurance calculations.
4. Long customer lifetime — As discussed above, most employer-based insurance does not have a long customer lifetime. But Medicare and other groups like the VA that keep people in their risk pools for a very long time do. That said, patients have the longest “customer lifetime” of any stakeholder. I am a risk pool of one, and I hold that risk forever.

Where does this leave us? We’re in a predicament where, in aggregate, a lot of people don’t respond to a lot of drugs. Many suffer for months, not getting better and/or enduring difficult side effects until the trial and error approach confirms that the drugs don’t work. And then they switch and start the process over again. And sometimes again. And again.

Can these economics ever make sense?

So how do we get out of this predicament? I see a few possible solutions.

Solution #1: Consumers pay to answer one question (cheaply). As a patient, it’s hard to agree to pay out of pocket for a diagnostic test. We’re accustomed to taking the tests that our doctors prescribe and our insurance plans reimburse. But without hesitation, if I were 18 years old going on birth control for the first time, and was told that there was a 1 in 20 chance it could be completely ineffective, I would pay $100 to tell me if I was that one. This solution requires a high volume-low cost approach. Birth control and SSRIs are definitely high volume. Now we need a low-cost clinical testing solution. This is likely to come from a new provider that doesn’t already have a reimbursement strategy that could be jeopardized by a low consumer price point.

Solution #2: Consumers pay to answer many questions (more expensively). While the odds of any one test being impactful for an individual at a given point in time are relatively low, the odds of every test we could take yielding impactful results for each individual over the course of their lifetime are high. The cost of this type of genome sequencing today is around $600, down from about $1,000 a few years ago, and continuing to fall.

Solution #3: Consumer-oriented healthcare providers subsidize the costs. While insurance companies retain patients for 2–4 years, healthcare providers often retain them for much longer. Increasingly, providers are becoming more consumer-oriented and taking steps to differentiate themselves based on high-quality care and better patient experience. Genome sequencing is one differentiator that enables these practices to deliver higher-quality patient care over a much longer span of time. I’m particularly bullish on this solution and can see it playing out in several verticals from women’s health to mental health to pediatrics.

It’s time for patients to take the reins

Every patient, to an insurance company, is part of a large risk pool. But individually, we’re each in a risk pool of one. And we fully bear the costs, economic and non-economic, of delayed treatment, choosing the wrong drug, preventable side effects, etc.

This means it’s on us to demand that healthcare decisions incorporate all information that’s available, including pharmacogenomic data. We, the patients, have the longest time horizon against which to value the information that’s gleaned from genetic tests. We also bear the costs of a trial-and-error approach. So it shouldn’t be any surprise that we’re the ones with the greatest incentive to push for more personalized medicine.

The birth control example is one instance where learning about a genetic variant can have a huge impact on an individual’s life, yet not necessarily make sense for an insurance company or even a healthcare provider to investigate. This means patients have to take the reins.

But it’s not just a matter of demanding relevant tests. We need to demand that larger-scale pharmacogenomics studies take place and are replicated such that their findings can be integrated into mainstream clinical practice. We must also demand that our providers stay up to date with the latest pharmacogenomics research by “voting with our feet.”

We should each be doing this individually because we’re the greatest beneficiaries of this information. But to really change the system, we need to also do this collectively in communities with other patients who have things in common with us. I’ll write more about this opportunity next time.

Working in this sector? I’m an interested investor — get in touch via sara@spero.vc.