Clinical Pharmacists, Generative AI, and InpharmD

Going under the hood of a generative AI product

Looking to hire the best talent in healthcare? Check out the OOP Talent Collective - where vetted candidates are looking for their next gig. Learn more here or check it out yourself.

Hire from the Out-Of-Pocket talent collective

Claims Data 101 Course

Practical introduction to claims. Ideal for all healthcare backgrounds, this course covers what claims are, why they exist, and how to most effectively use them.
Learn more

Featured Jobs

Finance Associate - Spark Advisors

  • Spark Advisors helps seniors enroll in Medicare and understand their benefits by monitoring coverage, figuring out the right benefits, and deal with insurance issues. They're hiring a finance associate.

Data Engineer - firsthand

  • firsthand is building technology and services to dramatically change the lives of those with serious mental illness who have fallen through the gaps in the safety net. They are hiring a data engineer to build first of its kind infrastructure to empower their peer-led care team.

Data Scientist - J2 Health

  • J2 Health brings together best in class data and purpose built software to enable healthcare organizations to optimize provider network performance. They're hiring a data scientist.

Looking for a job in health tech? Check out the other awesome healthcare jobs on the job board + give your preferences to get alerted to new postings.

Check Out The Job Board

TL;DR 

InpharmD answers complex questions about drugs by using a combination of generative AI and a clinical pharmacist. Clinicians can ask questions about whether a drug is right to give a patient, dosage and side effect questions, and what to do if a drug is not on hand. Hospital committees (P&T) that decide which drugs to use can generate reports on cost-effectiveness of drugs.

There’s a lot of go-to-market lessons to for generative AI companies, but the company will also face challenges of competition and trust as it grows.

This is a sponsored post - you can read more about my rules/thoughts on sponsored posts here. If you’re interested in having a sponsored post done, email nikhil@outofpocket.health. Also I’m a small investor in InpharmD, so double disclosures here.

Company Name - InpharmD

InpharmD answers complicated questions about drugs and whether they can be used in different patient contexts.

Prussian Blue sounds like a $19 cocktail that’s too sweet

The company name is a play on “informed” and PharmD, which made me exhale slowly through my nose and blink slowly. And everyone pronounces it differently depending on who you talk to. Kind of like the name Nikhil…

The company was started by Ashish Advani and Tulasee Chintha. They’ve so far raised $8M from 645 Ventures, Atlanta Ventures, Y Combinator, and lil ol’ me!

What pain point does it solve?

Let’s say you end up in the hospital, god-forbid or god-willing depending on your opinion about wearing shoes in the house. In order to take care of you, the docs give you enough drugs to put down Wiz Khalifa.

Which means a few things are happening.

For one, the doctor needs to know what the right drug is for you. In many cases, this is straightforward, there's lots of situations where doctors may not be totally sure. This is because new research about drugs is coming out all the time and because each patient can present a different permutation of issues, some of which the doc hasn’t seen. For example:

  • A patient who’s pregnant wants to take a new drug and the doctor isn’t sure if it can be used. One wrong prescription and the kid might end up being a newsletter writer.
  • A patient taking a medication to treat their cystic fibrosis needs to begin dialysis. How should they dose the medication?
  • A patient is on two other drugs already that cause some changes in the patient’s body that might interfere with the third drug the doc wants to prescribe. Not all drugs are fun and polyamorous. 

A doctor can try and find all the relevant literature to answer their question across PubMed, NEJM, etc. but that’s going to take a long time and they can miss actually relevant studies. The amount of new research comes out is staggering, apparently there are 36M studies on PubMed and ~2 papers are added every minute (from someone that REALLY needs tenure). 

Or the doctor can ask a clinical pharmacist for their input. A clinical pharmacist has experience with direct patient care, and is also an expert in understanding how drugs work. Their job is to understand the mechanism of a drug, stay up to date on research, hate on CVS, and understand when it makes sense to give a drug given the clinical context. 

And that’s it! Oh wait, just kidding, this is healthcare and everything is designed to be the logic equivalent of a Saw game.

Figuring out the right drug and dose to give to a patient is only half the battle. The other half is figuring out what drugs are actually on hand at the hospital, and making the best medication decision based on what’s available. There are lots of things that will affect whether a hospital actually has a drug but we’re going to focus on the two big ones.

The first is whether there are drug shortages. This is surprisingly more common than you think - in certain parts of healthcare, a very small number of suppliers provide materials for everyone. So when there’s a natural disaster, a spike in usage of a drug, or a manufacturer closes down because it’s not economical to produce a drug, it leads to hospitals not having a drug on hand. For example, right now the US is going through a cancer drug shortage that’s so bad the White House had to develop a plan for dealing with it.

The second thing that affects whether a hospital has a drug is their formulary. As you can imagine, it costs money to store a drug. On top of that, for things like inpatient stays the hospital is usually getting paid some fixed amount to care for the patient. This means that everything including the drug costs needs to come below that fixed amount for the hospital to make money. Because of this, the hospital needs to figure out which drugs they should have on hand that are cost-effective while still meeting the clinical need for treating the patients that come in the door. 

[For anyone confused, this is a separate reimbursement system from the pharmacy benefits where you go to your local pharmacy and get drugs.]

Now let’s say you have two drugs with a 5% difference in efficacy for a disease, but one is 20% more expensive to buy. Someone needs to figure out which of those drugs the hospital is going to administer to patients. So hospitals have something called Pharmacy and Therapeutics (P&T) committees that create these formularies. These usually consist of clinicians in the hospital who actively prescribe or care for patients, pharmacists, quality-improvement managers, and sometimes ethicists and finance people.

Depending on the hospital they might make decisions about their formulary based on data, based on the experience of the most senior person in the room that prescribes a given drug, or the position of Mercury relative to Earth.

So by the end of this, a doctor looking at a patient now has to determine:

  1. What are the best drug options for this patient
  2. Based on what’s on the hospital formulary
  3. That the hospital actually has in stock

What does the company do?

InpharmD lets clinicians ask questions about whether a drug is good for a given patient and get an answer quickly based on the literature that’s out there. 

If you’re a clinician, you can put your query as a free-text and say how quickly you need the answer to it. You can also see other questions that people are asking like some sort of strange, pharmacy-specific Venmo feed.

Once you ask your question some fun AI magic starts happening. Natural language processing breaks the question up to understand what you’re asking, then takes that query and pulls from all of the literature that InpharmD has stored in its library of 30M+ pieces of literature from curated sources.

The relevant sources and materials are presented to a clinical pharmacist that InpharmD has on staff, which selects the relevant articles to the query. Then the generative AI model that InpharmD has trained for this specific use case spits out a structured set of information to help answer the question. The clinical pharmacist verifies the information being presented, analyzes it themselves, and answers the initial question with a short summary.

The clinician is given an InpharmD clinical pharmacist summary answer, a grade for how good the evidence is, and tables that summarize the different relevant studies if the clinician wants to dig deeper. The nice part is that if any study is retrospective or non-randomized, InpharmD actually finds the author and breaks their kneecaps for you. Technology is crazy.

You can ask these questions on the InpharmD website, mobile app, or as an integration in the EHR itself at the hospital system. Clinicians tag how quickly they need the query back, and the company gets answers back to clinicians in:

  • Urgent - ~46 minutes
  • Within a day or so - 5-6 hours
  • Not urgent - about 5-6 hours as well

You’d think docs would abuse the urgent button, but apparently it’s only used ~10% of the time. I’m like…not used to people abiding by the social contract.

Those are for the regular clinician users. InpharmD has a more white glove product that’s meant for P&T committees. Those committees will give InpharmD lists of drugs they’re evaluating and a dump of data on how drugs are being used in their hospital and will ask them questions like:

  • A new drug just came out, tell us everything we need to know about it and whether it should be swapped out with one currently on our formulary.
  • For a given drug class (e.g. HIV medications), are we keeping the right ones on our formulary that are cost-effective? How much would we save if we switched?
  • Are the given drugs already on the formulary actually being used correctly?
  • If you give a human mentos and diet Coke at the same time…would they…

InpharmD uses that same technology + library + clinical pharmacist process and combines it with the hospital’s own data to answer these questions. For example- one of their health system clients gave them a dump of all the pharmaceutical products they use and InpharmD showed they were using a few different drugs for gender affirming surgeries. By standardizing behind one, the system would actually save them $6M per year while maintaining clinical efficacy. 

What is the business model and who is the end user?

Members of the P&T committees are usually the first users of InpharmD at a given health system since the company is brought on to help P&T committees make decisions.

However, clinical pharmacists that work at the hospitals themselves are power users of the product. Doctors ask a lot of questions about drugs, so the in-house clinical pharmacists use InpharmD to get the answers faster. On average the company says this workflow saves the clinical pharmacist about 2 hours of work per day. 

Doctors and other clinicians use the product as well, but by nature of the product, they need it for more irregular things that they haven’t seen before so it’s more sporadic usage.

For the core query product, InpharmD’s pricing is pretty simple. They charge per seat, and give discounts for more seats.

  • Up 30 seats costs $500/seat/month
  • 30 - 200 seats costs $400/seat/month
  • 200 - 1000 seats costs $300/seat/month
  • 1000+ seats??? Yo are you sure about that?? That’s like a small city

They charge extra for extra things like integrating with the EHR, custom reports, choreographed dance with each query, etc.

Out-Of-Pocket Take

You know what I like about this business? It’s simple and solves a specific pain point. Drugs are getting more complicated, it’s hard to keep up, and InpharmD will do that for you.
Some other things I like about this:

AI as a Managed Service - One of the big questions everyone has around generative AI is how it will be monetized. One of my beliefs is that for the back-office, this will happen in the form of “managed services”. This is where companies either supplement or outsource departments to third-parties to handle. Did I just learn what this is? Shut up. 

But honestly, it’s not a new thing. You see this pretty typically in IT services and physician offices will outsource tons of tasks like billing to Managed Services Organizations. InpharmD solidifies my belief on why I think this is going to be a more common go-to-market strategy for AI companies.

  • I don’t know many go-to market strategies.
  • Hospitals don’t want to train people on new tools. Have you tried to teach a doc how to use google calendar? It’ll make you lose faith in higher ed. In InpharmD’s case, they just ask a question and get an answer. InpharmD instead trains their own pharmacists on using their AI tool to help arrive at the answer which is maximally efficient. This also avoids the issue of having power users of your tool leave the institution and it stops getting used completely.
  • Using AI as a core part of the workflow enables serious scale which then allows them to price much more competitively. For example, a clinical pharmacist using InpharmD can answer a couple queries in one hour vs a couple hours per search without InpharmD.
  • Labor is consistently an issue. Hospitals are very financially sensitive and would rather not hire more full-time employees if possible. A managed services arrangement is more flexible and can actually match the demand for services. Sometimes queries will spike, like where there’s a widespread drug shortage, and you’re not going to go out and hire full-time just for that.

The AI Workflow -  I think the AI workflow does a few things that are smart.

  1. The user doesn’t have to do anything unnatural in their prompt. The onus is instead pushed to InpharmD to turn the free-text query into something that will get the right output. One reason a lot of people get annoyed with the GPT output is because they don’t know how to prompt it correctly.
  2. There’s a human in the loop. Not only does this provide a level of oversight to catch any issues the large language models might make, but it also provides a feedback system that improves the model in a way that doesn’t force their paying users to start with a subpar product. And I know a human is reading when I ask “I MIX THE MOLLY WITH THE LEAN…contraindications, urgent”.
  3. It provides different levels of depth. Some people just want the answer and don’t want to think too deeply about it. Others want to look at every single referenced study (nerds). The product uses vector embedding to show the source it’s pulling from, gives scores for the confidence of the evidence, and standardizes the format of the studies so that it’s easy to compare them side by side. When people can see how the sausage is made, it’s easier for them to trust the answers coming out.
For anyone interested in how the AI actually works, there’s a pretty good post about it here. It talks about how these queries are actually broken out, how the weighting works, etc.

2Users1Product - InpharmD’s first and core product was essentially “ask a clinical pharmacist anything”. This is very useful for clinical pharmacists, doctors, etc. from a productivity perspective, but it’s actually difficult to demonstrate a hard return-on-investment (ROI). Even though users liked the product, that's usually not enough to get enterprises on board. This is a trap that kills many startups, or leads them to build products that sell well to admin but are hated by the actual clinicians.

At this point, most people give up and go back to fintech or see if their GMAT score is still valid. But InpharmD evolved in an interesting way. The core product was a curated library about drugs and a pharmacy-tuned AI with a human in the loop that could answer questions about them. The company realized that you could use that same product to help P&T committees make formulary decisions. 

This DOES have a hard ROI - using data to inform P&T committees to change which drugs they use can save a lot of money.

Because of this, InpharmD can now create two different products to appeal to two different user types. 

  1. Hospital admins like the P&T product because it helps them make decisions about the drugs they administer in the hospital. 
  2. Hospitals will buy seats for their clinicians so they can make decisions within the constraints of the formulary the P&T committee decides on. Clinicians like the drug query product because the user experience is simple and provides them answers quickly. 

---

As with any company, there are issues that InpharmD has to watch out for as it grows.

Competition - InpharmD has a head start because it’s been working on this specific workflow for a long time and has customized the model weights, human feedback loop, and product itself for the end customer. Plus it already has users, which should make all of the above easier.

However, an open question is whether new generative AI models start getting so good or so easy to customize that new players can enter this space easily. Or could existing companies in the medical knowledge space use their existing distribution and trusted brand to create and roll out their own version?

I think this is an existential question for many generative AI companies. InpharmD is trying to combat this in a few ways:

  1. Utilize the fact that it has users already contributing questions to create a network effect.  This can be people looking at other questions in the feed, quickly recalling answers to questions already asked (34% of questions are repeats), or providing benchmarks/insights to stakeholders like P&T committees.
  2. Provide ancillary services around the tech. If the technology itself does become a commodity, you have to differentiate around everything else. For example, ingesting data and requests from customers to give them answers about how they should change their formulary. This is using services to differentiate yourself.
  3. Utilize channel partners like Group Purchasing Organizations (GPOs) which the hospitals already use for IT, medical devices, etc. Sometimes winning in hospital sales  is just nailing procurement, and GPOs can make that process way easier. InpharmD is already working with Vizient and Premier to act as resellers for them.

Trust and Speed -  The thing about asking your co-worker a question or a consult is that you trust them and give them the benefit of the doubt. Plus you’ve seen them at the holiday party…you have leverage. Vendors do not get that same grace. Corporations are, in fact, not people.

InpharmD has a whole process of checks and balances to make sure the right answer is given. Right now when a query goes in, about one quarter of the literature it pulls is irrelevant and 15% is missed. The role of the clinical pharmacist is to provide the human level filter here, but sometimes things pass through that shouldn’t. When that happens, clinicians lose trust very quickly.

But as a result of having those processes to catch things, sometimes queries will take a longer time to get back to whoever asked it. An urgent request to InpharmD still takes 46 minutes, which might take too long for the context in which they need it and also causes the user to lose trust in a different way.

The company is working to bring down the time to respond while also making sure the quality bar stays high, which is always a challenge especially as the company scales. But either way, the company doesn’t get a lot of wiggle room.

Apathy - Most healthcare companies struggle with their buyers actually realizing this is something they should give a shit about. Many times, the P&T committee feels fine running on vibes. Or they think doing a handful of PubMed searches for the most recent articles is enough to answer their questions. InpharmD’s job is to convince them otherwise. And now that so many health systems are trying to figure out ways to save cash, this might actually be the right time.

Conclusion and parting thoughts

Fun fact: 3.5 years ago when I was just getting started, Ashish the CEO of InpharmD told me that my grammar sucked ass and he refused to forward my email unless I got a copy editor. 

I never forgot that harsh feedback - Ashish inspired me to make changes. Like charging double for a sponsored post if the company was named InpharmD. Revenge sucka!!

Jokes aside, I’m glad to write this piece because I think it gets to show some of the interesting things that go into building a certain kind of generative AI company and the role that clinical pharmacists play. But also because both Ashish and I have gotten a chance to intimately watch each other’s companies grow, culminating in the write up you see today.

Thinkboi out,

Nikhil aka. “T hink boi”

Twitter: @nikillinit

IG: @outofpockethealth

Other posts: outofpocket.health/posts

--

{{sub-form}}

‎If you’re enjoying the newsletter, do me a solid and shoot this over to a friend or healthcare slack channel and tell them to sign up. The line between unemployment and founder of a startup is traction and whether your parents believe you have a job.

Let's Keep In Touch

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
close
search icon
close