Launch HN: HealthWiz (YC S17) – Navigating health benefits to lower costs | Hacker News Reader

Launch HN: HealthWiz (YC S17) – Navigating health benefits to lower costs

55 pointsYertis36 comments

Hi HN — My name is Nate Fox. Along with Nate Maslak (Nate^2), I am one of the cofounders of HealthWiz (https://www.myhealthwiz.com) in the current YC batch.

HealthWiz guides employees to convenient and cost-effective healthcare decisions and lowers healthcare costs by eliminating wasted spend. We help users understand what’s wrong and how to get better quickly and cost-effectively.

While lowering costs frequently means taking away benefits, we do the opposite — we bring transparency and information to a messy healthcare system that allows employees to better navigate their benefits, resulting in less wasted spend and a faster return to health.

This is personal for us. When Nate Maslak's mom needed to find a doctor for joint pain, she went to one physician, then another, then another. Six months and thousands of dollars later, she still wasn’t better and gave up on looking. Nate and I found it ludicrous that despite all the information available to us, healthcare remained opaque and intimidating.

Without HealthWiz, we have to scour the web for (often biased) information for what’s wrong and take our best guess at how to find treatment, only to be shocked by the bill. Our goal is to help you triage your symptoms with AI and access the most convenient ways to get better while knowing the cost in advance.

We’d love to get your feedback on the product (i.e. would you want your employer to offer this?) and look forward to discussing the nitty-gritty of just how we do all this, if people are interested!

Really like the product and idea. Tools like this are the future of the internet. For problems/questions where a decent answer requires multiple inputs from the user, Google doesn't cut it. I think you'll start to see this AI driven, decision tree logic applied to a ton of complicated questions in the near future.

Your go to market strategy is exciting also. Health care can be a terribly frustrating industry to start a business. Best of luck!

Thank you for your kind words!

And agreed on the trickiness of starting a business in healthcare. Many factors that drive economics/incentives/needs, and requires a lot of thought toward thinking how to best align many different interests. Definitely helps to be mission driven!

Hey everyone - Nate Maslak (Nate #2) here! Excited to get your feedback on the product and answer any questions you might have.

Definitely not #2!

Cool idea! Went through the "check symptoms" channel and came across a few things...

(1) Generally slightly more scrolling than I'd like. I'm on a MBPro 13" and if I have more than 1 symptom then I have to scroll down to see what they are (should I decide to remove some).

What does help is that the "Possible Matches" button is just high enough for me to not have to scroll to notice it pop up. Something to keep an eye on if more things get added to this page since it'll be important for users to know when the questions they're answering are leading to possible matches.

(2) Clicking "X" to remove my single symptom brings me back to the check symptoms starting point. This makes sense, but I noticed there's a message on that page that says "Something went wrong in the browser -- please try again."

Would suggest only having that error message appear if user got there because of an actual error (seems to be a result of reusing a page for multiple purposes).

https://www.myhealthwiz.com/symptoms?status=restart

(3) Got to this question: "Do your symptoms begin during your period or a few days before?"

But the options were Yes, No, and Don't Know. Doesn't really match what the question was asking.

https://www.myhealthwiz.com/healthwiz_symptom_question?gende...

----

Otherwise, great work! Here are a few things I really liked:

- Being able to see the original symptoms I added and the option to remove them

- Seeing potential matches appear as I answer clarifying questions

- Estimates of options (Online Doctor, Visit, etc.)

Thank you for going through the product and for the comments, incredibly helpful! (1) Agree with your feedback (amazing how obvious it is now that you point it out). We'll make the containers around the inputted symptoms shorter to give more room at the bottom of the screen; (2) That's just bad UX, sorry about that! We'll get right on updating that; (3) You're 100% right, that is a weird question for the tool to ask for Back Pain... we're looking into what triggered that question and what condition the system was confirming / ruling out.

Have you considered working with https://watsi.org to offer this directly to patients in the third world who might have no other alternatives (check symptoms/research conditions features)?

We just digged a bit to learn more about them; WOW, what an amazing organization and mission!

A lot of the tech we've built is focused on integrating with US insurance plans, but we have been starting to get asks from some of our clients about offering our solution in other markets as well, so we can definitely see efforts like ours expanding internationally. In the short term we're focused on serving US companies who are struggling with rising healthcare costs, but perhaps there's an opportunity to partner on the tech related to things like our symptom checker and condition content.

From a mission standpoint of helping everyone get access to better healthcare, that's something where we definitely align strongly. We'll reach out to see if they might find any of our technology useful -- would be an amazing partner!

[edited for grammar + expanding on a point]

Appreciate it!

You're working in an area that has tremendous room for improvement. The giant in this space is Castlight Health.

How do you guys differentiate?

Are you targeting a specific market? (Geographic, specific health plans, procedures, etc.)

Who do you need? (Payor partners, employer types to test with, ...?)

Thank you for the smart questions. We are focused on a specific market of midsize employers (frequently self-insured). Our goal is to help employees of our clients navigate the full point of healthcare in a way that is scalable.

Many companies in the navigation ecosystem today provide call centers that members/employees can call to get help. Unfortunately, these models have a hard time scaling to the midsize market -- our tech solution is able to scale for this market.

This seems like a cool product, but I'm not sure if I'm either misunderstanding how to use it or it is wildly inaccurate. I started off the questionnaire for shoulder joint pain, went through about 20 questions. the only one I answered yes to was "Does the stiffness in your joint feel better after exercise". The condition results were "Common Cold, Match: Very Low". I expected that maybe it might give me a list of chiropractors or physical therapists that specialize in sports medicine in my area, but I was definitely surprised by this outcome.

That's a poor result on our end -- we're building the feature that you described so that we're more on point when we have a low match like that. And the sad part is we actually do have national coverage of physical therapists, so it was a lost opportunity to get you to that next step.

If you're interested, would love if you tried the next version that we're working on. Our goal is that it will be able to better handle cases like these (which happen about 10% of the time right now -- which is a lot and in our minds leaves a lot of room for improvement.)

Cool product! I have two bits of UX feedback:

1) in step 2, after you click on Yes/No/Don't Know, there isn't enough feedback in the product that indicates that you successfully answered the question. I originally thought that the web request failed to send and didn't notice that the two questions presented were different.

2) When does a person ever reach step 3? It doesn't appear to be the case for me, you can circumvent reaching step 3 thru the possible matches. I think this is misleading and confusing.

Thank you for the kind words and the helpful feedback! On your points:

1) That's a great point -- we should make that more clear that the user answered the question (and not seem like its broken if the questions are similar)

2) The way it currently works is we wait until the symptom checker has reached high enough confidence to render the results. Sometimes this can take 15+ questions, which we've found to be frustrating to the user. So we've designed the tool to give the user lots of ways to exit the symptom checker toward taking action (learning more about a relevant condition, or getting help to find a doctor, etc). To your point, I think we definitely need to make it clear that step 3 isn't the goal per say -- that there are multiple ways to exit the tool, and it's not just only answering enough questions to get to a high enough level of confidence

Looks pretty cool! It's hard to know what to trust on WebMD and all of the other links at the top of search results.

Quick heads up. I tried filling out the example you listed "strep throat" and got the error "Sorry, we do not have any results for "strep throat"" (https://www.myhealthwiz.com/condition_search?query=strep%20t...)

Thank you for flagging that! We'll get onto fixing that UI/UX issue.

In the short term, if you click on the autocomplete results or select any of them, it'll bring you to the content page. (See here: http://imgur.com/a/XFdOg)

Also in the meantime, the link to the Strep Throat article is: http://www.myhealthwiz.com/condition_overview/101558

Again, thank you so much for flagging!

You guys are going to save so many people from misdiagnosis on webmd! Love the "check symptoms" feature and a user interface. Would love a little graphic that shows how far I've gone in the self-diagnosis (so I don't think it's going to go on forever). Love the product and the UI! Simple/beautiful/sleek.

I agree with the idea of feedback on how far you've progressed - though a linear graphic might not be possible. I assume we are navigating a tree whose depth varies greatly with your answers.

After a particular selection you might have tree of max depth 10, but with the other selection you land on a tree of max depth 5, for example.

Personally, I've love to see a list of the conditions whose probabilities are increasing and decreasing with each selection. This may not be suitable for the average consumer, but it would be very educational! And also fun.

We've gone back and for the with the idea of showing the probabilities across the matches.

Ultimately, to your point, we learned that at times it can scare users more than it can help them (e.g., if a condition goes from 0.5% to 1%)... not to say that we can't do that above a certain threshold.

We'll keep testing this. Is there a certain minimum probability you'd want to see before seeing a condition (e.g., if something is 1% matching vs 10% vs 70%)?

That's wonderful that you are open to feedback from geeks with fringe interests.

If I were in charge of this project, I would strive for simplicity and for appeal to the widest audience, and not indulge the geeks, students, professionals, and hypochondriacs who might want this feature.

The casual user should get fuzzy names for probability values, as you already do. I wouldn't even let them see the numeric values - that seems like bad psychology and an invitation to ascribe the wrong meaning.

Maybe there is a way to make everyone happy though? Is there already a "debug mode" for developers? Making it possible (not easy, but possible) for users to activate debug mode might be a way to expose functionality which would provide the kind of information.

> Is there a certain minimum probability you'd want to see before seeing a condition (e.g., if something is 1% matching vs 10% vs 70%)?

For people like me, I'd like to be able to dynamically set the value myself.

For the general public - the way you do it now seems good.

Thank you!

As a geek with a few of my own "fringe interests" ;), I love this idea of making it so hacker types can see the adjusting rankings behind the hood.

Sometimes websites do this unintentionally by the things they leave in the html and what not, but would be fun to think of the easter eggs we could leave behind!

It's a question of risk vs downside. Pulmonary embolism for example is deadly and can look like a lot of more common things. Which is why it kills a lot of people every year.

This is a great point -- weighing false positives against false negatives/true positives/false positives is something we think about a lot. It's really important to get this right.

This is why, at this stage, we have a strong bias towards guiding people towards the right care provider rather than trying to replace them; it's more of an aid rather than substitute. One example of this is how some of our users have used the symptom tool with joint pain, and its helped them figure out that they should see a rheumatologist, who was then able to guide them towards the right care to treat their particular version of arthritis.

It's smart care providers like that that can really deal with the nuanced cases and make sure people minimize the risk that they miss things like a pulmonary embolism.

Attention deficit hyperactivity disorder

Bladder infections, also known as Urinary tract infections (UTIs), are common infections that can affect the bladder, the kidneys and the tubes connected to them.

https://www.myhealthwiz.com/condition_overview/100156

:)

Hah! Thank you -- Condition summary string got mis-mapped we're fixing that now.

Fixed!

"Attention deficit hyperactivity disorder (ADHD) is a group of behavioural symptoms that include inattentiveness, hyperactivity and impulsiveness."

Just a little bit more on point ;)

I haven't started a company that needs to provide employee healthcare--does this have a short-term improvement to insurance rates? What incentivizes a company to employ HealthWiz--are there insurance providers that reduce rates as a result?

How do you convince employers that this service is in their best interest?

In the US, employers ultimately shoulder much of the burden of health care costs, so a tool that helps employees make cost-saving choices eventually saves money for the employer.

However, that's a very indirect path. Most employers use a separate insurance carrier (Humana, Blue Cross, Cigna, etc) and adjust costs on a ~yearly basis by shopping around. HealthWiz would probably love to go to market by integrating with these insurers, and offering an incentive to employers whose members make use of the HealthWiz product to save money.

That's a tough way to go to market because you have to convince an insurance company to trial your thing with their customers. Insurance companies have deeply risk-averse cultures. Plus you have to actually achieve significant cost savings for individuals who use HealthWiz, and wide enough usage throughout the employer, to make the insurer comfortable with offering a discount for that population.

Two other potential markets are self-insured employers and accountable care organizations. Self-insured employers are companies (or often, government agencies) huge enough that rather than paying an insurance carrier, they just operate their own, and pay for healthcare directly. These folks feel the costs a little more urgently and may be easier to pilot with.

Accountable care orgs (ACOs) are a relatively newer model, where the ACO operator (usually a hospital/health system) receives a fixed per-person amount to address all the health needs of a population. These orgs are probably the most informed about both cost and medical needs, and are making real strides in long-term cost saving measures like improving preventive care. But because they're so different from insurers and employers, they'd probably need a somewhat different product, likely one that maintains the primacy of their brand (and maybe doesn't do things like suggest that cheaper hospital a few miles away).

That's a great assessment of the space! We're focused on the midsize self-insured market for that very reason. The incentives are closely aligned.

For example, ERs frequently take in patients hoping to get a UTI treated. Much of the time, the right treatment is a prescription. But, going to the ER to get the diagnosis can cost ~$2K (split between the employer and the employee) and take as long as 12 hours of waiting in the waiting room.

In this case, if a patient uses a telemedicine provider for the same service, they're likely to get the same diagnosis and prescription but the visit will cost ~$50.

The nearly $2K in savings would go to both the employee and employer, the employee saves money and is able to get treatment faster.

I have a neck fracture and after answering 25 questions it said I have a 'thoracic aortic aneurysm'

Unfortunately our tool doesn't get it right 100% of the time. The cases like this where we miss, we try to encourage the user to find a healthcare provider when the tool falls short. Fortunately, the models always learning -- but it's still got a long way to go.

We're releasing a new and improved version of the symptom tool next month -- if you're interested, would love to have you try it once it's ready!

And sincerely sorry to hear about your neck fracture. Best wishes from us that you're getting great care for an as quick and painless recovery as possible.

[edited]

Hmm. Doesn't seem to work very well in iOS.

Have you tested this page for mobile users?

Sorry to hear it didn't work well for you on mobile! We do test on mobile / iOS but seems like we didn't do a good job on this side - we'd love to learn more about what wasn't working well so we can work on that.

Can you talk more about what data you used to build you AI

Absolutely! We partner with a few API providers who are experts at building the logic that drives the symptom checker, and then build our own layer on top of that to make it as user friendly as possible.

As for how the AI itself specifically works, given the data it has on the stated symptoms (Fever, Sore Throat, Chills, etc), it has based on prior data a best guess of the top conditions correlated with those symptoms.

Then, it asks a question that can help inform its calculations, which has simple boolean answers (True this symptom exists, False it does not exist). Then, the ranking of conditions updates, and the API responds with the next symptom it is curious about to drive towards a smarter diagnosis.

[edited for grammar]

Reply on news.ycombinator.com