Amino: Taking the Guesswork out of Healthcare
Amino uses big data and analytics to make cost-of-care more transparent, and physician recommendations more meaningful
Two of my childhood friends graduated magna cum laude from top American medical institutions. They both just so happen to have the same insurance carriers and plans. But when I asked them how much their out of pocket costs would be for 3 very common procedures – an MRI, an ACL surgery, and a hip replacement – their estimates were wildly different… by thousands of dollars. For an industry that makes up 20% of US GDP, the opacity of healthcare costs in America is, very frankly, a joke. Meanwhile, deductibles are growing at a rate that is 6 times the growth of wages, intensifying the need for a solution.  That’s why David Vivero (HBS 2008) founded Amino.
Amino is a free web resource that uses objective healthcare data to give ailing users personalized recommendations on:
- a doctor/specialist
- estimated healthcare costs based on his or her insurance plan
- free appointment booking services
On the backend, Amino has aggregated over 9 billion insurance claims (the records that help your provider request payments from your insurer) that represent over 220 million people in the US and 950,000 doctors and facilities.  At a high level, Amino uses experience as one of its main proxies for quality. That is, they will rank doctors nationwide by the number of specific patients they treated for a specific procedure over a set time period. The comparable patient set has the same age, gender, and condition as you. 
Although billions of these records have been created annually for years, various companies have only used this dataset to improve marketing efforts or research service effectiveness. This is the first time that same data is being used to proactively help users make informed decisions. 
So the obvious question – why hasn’t anyone else been doing this? Insurance claims data are generated by thousands of different medical practices that all enter information in different ways, making the information extremely heterogenous. It took a specialized team of Amino engineers and 2 years to compile and serve the data in a way that makes sense. 
As a result, competitive offerings rely on survey data to estimate healthcare costs. However, surveys have inherent biases and misreporting issues that can lead to wide ranges and errors. Additionally, companies like Zocdoc and HealthGrades use subjective patient reviews as a proxy for quality, but anecdotal quality measures don’t tell a whole and consistent story nor do they make pricing accurate and transparent. [3,4,5]
Another huge difference between Amino and its competitors is Amino’s pledge to never have ads on its site or accept agreements from providers and organizations to favor them in search results. Their patient investor base is by itself a huge asset that accommodates a business model that can monetize much later. 
Data Analysis and Processing
For Amino, data analysis is an obvious and necessary core competency and they perform it well in several ways.
First, Amino asks the user to enter some basic information like age, gender, and condition.
Second, Amino helps match doctors well suited to an ailing user’s needs based on the necessary specialty/procedure, as well as quality metrics such as revision rates – the rate at which patients need to go back for treatments related to the same condition.
Third, Amino spits out how much different doctors charge regionally (think AirBnB price comparisons that you’ve seen in a map), ranked by the system described earlier. Here, you can toggle for whether or not a physician is in or out of your network, temporal differences, experience, education, and other things that may matter to you. 
Data Analysis Nuances
It is important to note that Amino doesn’t compare individual doctor rates to the overall average rate for all doctors, since this wouldn’t account for specific patient characteristics. Therefore, doctors aren’t unfairly penalized for higher-risk or sicker patients. 
Another interesting metric Amino reports are risk-adjusted decision factors. Essentially, Amino compares the predicted rate of a procedure to the actual rate measured for a specialist. Their model incorporates data about every patient in the database who received a procedure, including patients’ demographics, diagnostics, and places of care. As a result they can provide a risk-adjusted decision factor analysis per provider. 
While price transparency and physician recommendations are obvious applications of their huge dataset, other future applications including predictive outcomes and preventative care innovations could be interesting to explore. Both of these latter solutions have been attempted for decades in a very crowded space. Will Amino’s current machine learning capabilities hold the key to better solutions? Only time will tell.
Student comments on Amino: Taking the Guesswork out of Healthcare
Nice post Felix.
Few questions –
a) “Amino has aggregated over 9 billion insurance claims” – How has Amino achieved this? Have they tried to form arrangements with every insurance provider? As a startup did they not face data privacy concerns in getting this data?
b) Do you visualize a situation in which they would try to make the data entry format consistent and remove the heterogeneity ?
c) How does Amino make money?
Thanks for the question Sidharth!
(b) I certainly do. As the first company to receive national claims data from the CMS (Centers for Medicare and Medicaid Services), Amino has also received a stamp of approval from CMS to create new healthcare quality measures. This could be a huge move towards standardizing homogeneous HC data.
(c) It currently seems like they are partnering with a variety of institutions that want access to their de-identified data/trends/analyses. They also offer their service to employers as a benefit – in this version the service is customized to their specific plans, and may offer additional features.
Cool Felix! It’s a tremendous feat to have cleaned up healthcare data sufficiently such that it is useful for analytics.
I wonder how shifting in the data sets over time would affect the validity of historical data. For example, due to factors like Affordable Care Act, medical practices are now incentivized to cut costs. For those that cut costs and prices effectively, historical data might be less reflective.