Clover Health, an insurance startup trying to use data to keep customers healthy
bloomberg.com
bloomberg.com
So it sounds like the company's biz model is to send nurses out in the field to get senior citizens to enroll in Medicare, check up on them and collect data, and sell them healthy lifestyle products (gym memberships, cooking classes, etc.). As a 23 year old I'm not sure if this is any different than what big insurance co's do for their Medicare enrollees.
The actual tech part seems to be the web portal/CMS (Django) and the storage and management of this tangential customer data that is generated with each visit:
Clover uses Python for all of our backend coding so using a pipeline orchestration tool that is itself written in Python is valuable for us. Airflow meets that requirement and it’s the tool we’ve chosen to manage all of our pipelines.
Here's the actual coverage sheet if you want to compare with your/your parent's plans: https://cdn.cloverhealth.com/filer_public/c4/a8/c4a894d0-000...
Deductibles (excluding prescription drugs) are $6,700 annually. Someone maxing out prescription charges would spend an additional $5,000 per year. I'm not sure how that compares to a traditional provider.
Tangentially: I think we've almost stretched the definition of the word "startup" into oblivion at this point. It doesn't mean anything to me anymore.
As a 23 year old I'm not sure if this is any different than what big insurance co's do for their Medicare enrollees.
The difference is most insurers would say something like "OK, out of X thousand Medicare-age people, Y of them will have this expensive health problem over the next year, are we profitable when that happens?". We say "OK, how can we identify the people most at risk of that and use preventive care to head it off?" So while most insurers just take the statistics as a given and budget for it, we're trying to change the statistics.
storage and management of this tangential customer data
I think you vastly underestimate how much low-hanging fruit there is in this field, and how difficult getting and maintaining good data (so you can find the things you can do) really is :)
I processed insurance claims for over five years and that was not my observation. Granted, it was one company, I processed accident claims and it was supplemental coverage. But when I was there, I was initially encouraged by the trend towards adding "wellness benefits" because I had naïve fantasies this meant they were working towards getting people healthier.
Nope. It was entirely about "If we offer specific tests and call them wellness benefits, we have higher retention."
That was it. There was zero interest in actually making customers healthier. And I was horrified and I am pretty cynical these days.
So, I guess, I am wondering if you have supporting evidence? (Links to articles, stats, etc?)
Also most insurance companies aren't really set up to do preventive work to the extent we do (which involves home visits from nurses and other clinical staff, staying in touch and proactively reaching out when there seem to be problems, etc. etc.), largely because they consist of essentially competing divisions which each own a subset of the data and responsibilities and jealously guard against every other division. I worked in a company like that, once upon a time -- everyone had very clearly-defined limits to their responsibilities (well beyond what privacy rules required) and even just contacting someone in a different department to make them aware of a problem or pass along a suggestion would trigger turf-war consequences.
I'm very happy to now be working with the opposite of that.
Finally, it's worth noting that although there are a lot of dismissive comments about tech in this thread (and some of it I can understand -- most of what we do is internal-facing and harder to promote as a result), the perspective inside is pretty different. It's true there's not a ton we're doing that's cutting edge and "sexy" tech (though we have open-sourced some neat tools for temporalizing SQLAlchemy's ORM), but that's because health care hasn't yet caught up to just the boring everyday things we take for granted.
There's an example that really drove it home for me, from about six months ago. Claims are obviously a big part of what we're about, and so we have an internal team that reviews processed claims for accuracy. They have to check against a pretty complex set of Medicare claim-processing rules, part of which involves rules about combinations of medical procedures (i.e., Procedure A might be allowed but only if alongside Procedure B, or A might only be allowed without B, or A and B might just be independent and can both appear or not appear). The canonical source for that data is... gigantic Excel spreadsheets of all the procedure codes and rules. And so the claims-review team would open up Excel, type in a code, wait five minutes while it churned to find the code, then repeat for the next code, etc.
An engineer saw this and spun up a two-day project to ingest those spreadsheets (on an ongoing basis, since they change over time) into a database and wrap a search interface around them which allowed punching in all the codes and getting back a list of rules and an indicator of whether they'd been violated.
To tech people that's so mundane it almost puts you to sleep, and makes you say "wait, that's it?". But to healthcare people, it was a gigantic step forward. So many people from health backgrounds came from places that never let them work side-by-side with tech staff and figure out what was possible (or where all tech was outsourced to vendors with opaque planning and feature-request processes), which is why so much of healthcare tech feels decades out of date.
So just having tech and medical/insurance people in the same room, working and talking together every day, acts as a pretty impressive multiplier compared to traditional insurance companies.
What you are doing is far more interesting. It was pretty obvious that there is a lot one can do with data, assuming you can deal with the HIPAA limitations. One ill person is hard to get much from, but 10000 is different story.
I wouldn't be surprised to see you guys getting acquired by a Bigco insurance. I'm guessing that's an exit your investors are hoping for and part of a reason for the Bloomberg article, to start getting the company on the biz radar.
Some of the mid-size companies are okay at this; but the low-end guys are very, very unsophisticated (there are a lot of local / state plans as mandated by various state laws) and the super-big guys just have so much stuff they have no idea how to integrate it all. Think 15-20 different, incompatible claims systems with overlapping numbering schemes, different versions of schema, different versions of underlying DB required, etc. Then do the same with provider management systems, contracting systems, etc. You get the point.
By the time you get halfway through scoping out a project, the architecture will be obsolete. Oh and also the federal government keeps changing the rules about ICD-10, ACA, Medicare reimbursement, and a whole bunch of other stuff, so make sure you add in that capability to all existing systems and don't get sanctioned. A lot of companies are having a hard time keeping their heads above water.
The big guys are going to eat the data-hungry guys. Or rather, the big guys will buy the best of the new data-first health plans (Clover, Oscar, etc.) for their platforms and migrate everything onto them.
I agree there needs to be better health care data science going on towards improving wellness and health care (assuming data is collected, stored and analyzed honestly and appropriately - whatever that means these days is whole other debate).
But I think it would be helpful to everyone if there was a bit more transparency and clarity from health care related startups about their plans and what they are doing. For example:
It sounds like a couple very separate things are going on here:
1) You are a health insurance company
2) You are an organization doing medical patient data collection and data analyses and pushing real-world actions based on this analyses to try to improve wellness and prevent/reduce expensive medical problems.
The first is relatively straightforward, the insurance business is really just a math equation with a marketing department. Health insurance and Medicare are a bit of a special case because the government has said health insurance companies must pay out most of the premiums to medical care costs or give the money back. Though this government contract component creates a numbers game with a single digit profit % built in once you scale (you can also do a decent back of the envelope valuation just discount your member growth as other inputs and comps are ~standardized).
The second thing, medical data science is where it gets complicated. This is serious work that needs to involve doctors and medical research professionals and studies before anything major can happen. I think it's great if an insurance company wants to take some of its revenue and put it towards data science focused on researching better wellness and health care. (Let's ignore for a moment that the incentives from a business model and profit potential are not clearly aligned by doing this, eg you don't get to keep medical care cost savings, you give it back).
More of an issue is the ambiguity around stuff like the "promise of using technology to solve such a large and socially significant problem as health care is one of Clover’s most powerful recruiting tools" and then goes on to say a guy from Yammer and engineers from FB and Google are mining health data to "suggest ways to prevent them before they happen." What is that suggesting process and who is involved? Recommendations of brown rice and gym memberships to a couple of octogenerians is funny and relatively mild (and kinda pathetic tbh). But it's a risky slippery slope when potentially unqualified and premature medical data analyses conclusions start to be implemented in the real world, it could be even dangerous to people's health, the opposite effect, even with the best of intentions. The example from the article could go bad: have you studied a potential rise in injuries when you recommend gym memberships to 83 year olds? If this has been properly studied and implemented make it public.
We are not a tech-led company: we're a company that makes tech -- in the form of tools, data science, etc. -- available in service to our clinical, operations, and insurance staff. My project right now involves building out a social-work assessment app, for example; I don't decide what goes in it or what actions to take based on the responses entered. No engineer does. The LCSW who's in charge of the assessment tells us what goes in it and what the workflow rules for it are and how data from it should propagate into our systems and be used. My job is to bring technology to bear to make it A) easier for the social workers to use than their current paper forms, B) easier to process the data so that they can use data-science frameworks on top of it to inform their decisions about how to care for someone.
I also didn't assume random engineers are making care decisions. I said this is ambiguous and I asked a question: What is that suggesting process and who is involved?
The article says you are "developing software to organize the trove of data coming from workers on the ground in New Jersey, mine it for likely health problems and suggest ways to prevent them before they happen." Your website says "we’re also partners in care, using patient-centered analytics and a dedicated care management team to identify potential risks and directly provide preventive care."
Well nurses generally don't know much about data science software and data scientists generally don't know much about practicing medicine or preventive care.
And of course you are not a tech-led company. You are a marketing company and government contracted middle man connecting health care services to patients and handling some of their paperwork and payments. Still this is an industry entirely living off regulatory capture.
I'm not accusing you of denying someone's valid insurance claims because they didn't take your eat brown rice suggestion.
But you yourself claim to be an insurer "trying to change the statistics". I get the preventive care and data science shtick looks good when marketing to potential members and to potential investors. But it's not clear how you are taking on the "change the statistics" part you claim to be, which is the most interesting and important part of all this to the extent it is real as improving peoples' health and wellness, and preventing serious health problems would be amazing!
So I asked a question about how the process works and who is involved as getting data and actionable data analyses to and from health care professionals in the field is not straightforward, nor is it trivial, nor is it explained anywhere.
The way I see it, the premise of Clover is asking what would happen if you removed the corporate-organization barriers and the organizational disinclination to do much beyond just processing claims. All those easy preventive-care wins suddenly become real options, for example. And as the insurer you have all the data to identify gaps and potential problems. For example: you know someone has a prescription because you see the pharmacy claims. And you know if they stop filling it because you stop getting claims. That's a potential problem that's easy to identify, and you have the staff to get in touch and find out what happened; maybe it's as simple as finding a pharmacy that's easier for them to get to, or a service that will pick up and deliver without requiring regular trips to the pharmacy. Sometimes it's the complexity of a particular medication regime (especially since often it just comes in the form of written instructions with little context), and clinical staff or pharmacists pharmacists take a look and come up with better explanations and adherence plans.
And really, just the easy obvious stuff is so copious that I suspect it's going to be years before we run out of it and have to worry about trying to come up with new ideas -- it's just that most insurers are not structured to take advantage even of stuff that's known to be helpful for people, so nobody's ever seen what happens when you try.
Still this is an industry entirely living off regulatory capture.
I'm not really sure what you mean by this. I'm not privy to exactly why Clover's in the Medicare space, but my inclination is that it's because Medicare provides a stable, existing and relatively well-known space to operate in, and precisely because of how it's structured and regulated it provides a few things already packaged up that you don't have to bootstrap or which aren't quite as hard to bootstrap. And although it is heavily regulated, I've seen no evidence that the regulatory structure is designed to exclude competitors or preserve existing players; for example, in January of this year, Cigna (hardly a disruptive new player) was hit with sanctions that temporarily remove their ability to enroll new members into their Medicare Advantage plans. Everything I've seen of the regulatory environment is that it's designed to ensure the safety of Medicare enrollees and to push Medicare Advantage insurers to take good care of them.
100% agree. Health insurance data is about as garbage of data as you can get.
Many providers still fill out claims by hand and fax them in. A good many more have an EMR system that just automatically faxes it for them. All this faxing and handwriting and manual data entry means numbers get transposed and data is wrong all the time. Add to this the fact that the people at the doctor's offices filing the claims often have no idea what they're doing and will often file things completely incorrectly (i.e. under a provider code instead of a facility code), leading to a rejection.
I cannot overstate how shitty this data is. Across the board. EMR systems are getting better, but generally only younger doctors in independent practices use the more cloud-based/big data ones. Big hospitals and the like are still generally on proprietary EMR systems (and that doesn't mean it's any better; different departments can have different billing practices that still screw things up).
It will take AI to clean up this mess. It's just too complex for humans to tackle.
For example, if the deductible was $6700, a 6 day hospital stay would cost $6700 out of pocket. Because it is a $290 per day co-pay, it will cost $1740 out of pocket. There is a tidy difference there.
from my experience, a lot of it depends on having the right family, and perhaps community, support structure in place.
I was hoping there would be insights or code on the "inputs/outcomes" seen on the title
I'm doing http://outcomereference.com/ so i'm esp. interested in this
I'm a fairly healthy person (at least right now), and I'm not sure how I feel about it.
Let's take the Type 2 diabetes example given in the article.
No mention is made of putting a Continuous Glucose Monitor on the 83-year old patient, to graphically show in real time the impact white rice has upon her blood sugar.
No mention is made of five additional antibodies tests to identify Latent Autoimmune Diabetes in Adults (LADA), which looks a lot like Type 2 in its early stages, but likely needs to be treated as Type 1 as early as feasible according to some very recent research.
No mention is made of regular tracking of her HbA1c, fructosamine (gives a 14-day instead of ~115 day window into long-term blood sugar management), and 1,5-anhydroglucitol (indicates time spent with hyperglycemia).
Not to speak of doing those three areas on a regular, monthly/quarterly basis. It is very rare in the US to do all of that even on an annual basis; at best, the common practice is HbA1c once a year. While metabolic syndrome leading to diabetes is a huge drain on US insurers, aggressive data-led syndrome reversion is de facto not recognized by either the mainstream medical field nor insurance industry.
There is lots more data they could gather than what I mentioned here, but all they did instead was suggest "try to eat brown instead of white rice". She should try not eat rice (and other carb-heavy food) at all for a month, and see if that improves her blood sugar numbers. While the fitness community has already empirically identified that diet is 90% of success unless you are a genetic anomaly, the food, medical and insurance industries still push the idea upon patients they need to go out and exercise more, and only tinker around the edges of their diets (unless you are fasting/caloric-restricting, going very low-carb, going vegetarian, going raw food, etc., most Americans idea of a diet change by adding a couple salads a week, picking "low fat" packaged food and so on is just tinkering).
This doesn't even get into monitoring controlled changes in habits, measuring the impact upon the data, and using those results to guide further changes. Nor have we even touched upon the lack of incentives in the US for companies seeking to "disrupt" medicine through holistic, integrated process changes instead of "product" like pharma or medical devices: if company X can provably intake Y number of customers with say, metabolic syndrome or even Type 2 diabetes, and achieve a reversion rate of Z% with negligible (<5%) backsliding 5+ years out from date of reversion, monetizing that proprietary process is limited to either being a provider (in which case you're squeezed by the insurance companies with take-it-or-leave-it negotiations) or an insurer (in which case you're too spread out covering other ailments to really capitalize upon the proprietary knowledge). I suspect the lack of incentive structures is similar in nations with national health plans.
Are there any papers / data-sets that demonstrate this? If so, could you share here or at my email (see profile)? Would be greatly appreciated it. It's been my hunch this is true, but haven't seen the evidence for it.
Most of the tests you mention above are wasteful for your typical Medicare Advantage enrollee. There's a ton of low hanging fruit for a start-up like Clover to make a meaningful impact. Trying to change human behavior through diet or exercise is incredibly difficult, especially for those from disadvantaged communities or of lower socioeconomic status. Kudos to Clover to trying to make a marginal impact on that front; most insurance plans wouldn't do anything.
A good point to discuss. Perhaps I'm overly optimistic about people in general, but how do we know, a priori to a data-centric healthcare model applied to a patient, about their participation? These tests are only wasteful in the context of the current care model, which stipulates that once you have Type 2, there is a lockstep progression of increasingly invasive and expensive medication and interventions, graduation to insulin injection, and culminating in early death from complications? In the face of that kind of prognosis, it is not at all surprising that any additional tests are considered futile by both medical practitioners and patients alike. But if the patient was offered via data-centric care a clearer window to their condition, amplified proactive participation in care management with negative feedback loops tamping down undesirable fluctuations, and in the case of Type 2 or metabolic syndrome the clear goal of reversion (though not a cure) and drastically improved eventual outcomes, why do we assume that even elderly patients on Medicare Advantage would not statistically respond well as a whole population? I dunno, this isn't my domain expertise, so I'm honestly asking these questions; if there is the equivalent of behavioral economics studying patient behavior, then perhaps that field has the answers I'm seeking.
> There's an entire cottage industry devoted to solving this problem for insurers.
I'd like nothing more than to see these efforts succeed, look forward to following the progress.
You might want to take care here -- to be honest I don't know all the things we do in home visits for diabetes assessment (though I am working a few desks over from the person building the next version of our diabetes assessment, so I could find out but then I'm unsure whether I can publicly detail the whole thing), but I'm pretty sure no matter how many things there are, an article like this one isn't going to go into enough deep medical detail to satisfy you :)
The business model of a US health insurer in this situation frankly sucks, however. If Clover successfully manages a reversion outcome with aggressive short-term techniques for a chronic condition, then nothing prevents the patient from switching away to a cheaper insurer afterwards. Clover puts in all the hard work, and another insurer reaps the actuarial benefits. That has to be fixed, because all the big killers of patients, insurer profits and medical budgets in the developed nations are the biological analogs of Advanced Persistent Threats, and not amenable to fixes developed for acute situations (which as a civilization we're really good at addressing).
That's a big "if" for a Medicare Advantage insurer, since we can't enroll someone unless they're eligible for Medicare. Although it's possible to become Medicare-eligible prior to age 65, it's much less common, and some of the ways it happens make someone ineligible to enroll in a Medicare Advantage plan (for example, someone with end-stage renal disease is Medicare-eligible regardless of age, but unless already enrolled in one cannot join a Medicare Advantage plan -- they must opt for traditional Medicare instead).
So by the time a typical diabetic becomes eligible for a Medicare Advantage plan, the intense regimes you're talking about don't (as far as I know) offer any realistic chance of reverting. Which means most of the effort instead goes into proper management techniques to prevent progression and complications.