Willy Wonka and the Medical Software Factory

The fact that it's a paid for trip is why it is worth mentioning, not necessarily the length. Five weeks of vacation, while not very common, is common enough that it's not really worth writing about otherwise. Edit: Agreed that 5 weeks of contiguous vacation is rare

I didn't realize that companies in Europe were requires to actually pay for the vacation, on top of normal pay, by providing airfare, meals, and lodging.

I get 10 month vacation fully paid as any normal work month.

I work in ad tech.

It just depends on what you know, workplace is flexible for people who can't be replaced easily, so we might as well flex.

Honestly that doesn’t sound like that great of a deal. I’d rather have continuous work life balance than kill myself for 5 years so someone can buy me a 5 week vacation.

> The offices are shared by up to 3 people in offices designed to have 1 or 2.

Yes and no. The campus is rapidly expanding to accommodate new hires.

> The vast majority of hires are new grads and turnover is very high.

Turnover is only high in implementation, other positions have veterans of 15 and 20 years.

> ... it seems like a large facade.

The look of the place does not translate well to the intensity and energy that many people bring to their work. There isn't time for most people to 'have fun'. They're working.

Source: a family member works at Epic.

Epic caught up on building new buildings and I don't think there is a single 3 person office on campus. The general trend is two to a window office, one to an indoor office. Higher tenure often can get solo window office now if they want it.

> Epic has nearly every person in the US's medical history in their databases (since their usage is so widespread) so if you sign up to use Epic you get the medical history with it

Why is this legal?

Epic EHR is typically deployed as a separate instance on customer premises (not cloud SaaS). Epic customers don't automatically share patient charts with each other. However they can do so if desired through the Epic Everywhere product. And they can also share the same patient charts with other third-party EHRs through standards based APIs, although that requires configuration.

Check out Atul Gawande’s article in the New Yorker on the topic: https://www.newyorker.com/magazine/2018/11/12/why-doctors-ha...

Enterprise sales into a heavily regulated industry are really, really hard. On top of that, the primary purpose of an EMR in the US is to provide accurate billing information. So the usability of the system really only matters to the degree that it affects the efficacy of billing.

I went into medical school with the same goal. Intended to do an informatics fellowship after residency. After witnessing the Epic rollout at my hospital, I want no part in EMR development. The beuracracy is disheartening.

I’ve found other places in medicine to make software and have an impact.

That said, I truly hope you succeed. We need more hackers in medicine.

Healthcare tech is notoriously bad, but over the past 5 ish years there has been a wave of healthcare tech companies who focus on making their products consumer-friendly (ZocDoc, DrChrono, Oscar, PillPack, Capsule etc). NYC in particular has a large healthcare tech industry.

So why are EMRs so bad in the US? 1) Most EMRs are on-premise software with that being the preference due to regulatory/compliance/security risk. Because the EMR companies don't have the ability to deploy changes or updates on their schedule, the rate of change is much slower and updates made for one client aren't made for all clients. 1a)Many of these EMRs are heavily customized per client (think Epic in particular). These customizations aren't often being brought back into the core product but are one-offs. Think of it like starting from scratch with each client. 1b) Unlike the Googles/Facebooks of the world which are built on web software, there's no easy way to deploy. This means that where tech first companies are probably deploying multiple times an hour, Epic/Cerner installs are probably being updated once a month or less. Some hospital systems are running on software that's 10+ years old. 2) Large initial barrier to entry. The sales cycle for most healthcare software is close to a year minimum. For EHRs, which affect how the entire hospital runs, it's almost certainly a multi-year effort. Most EHRs are signing 5 year contracts, so providers can only switch every 5 years. And for the large providers it can take 5+ years to implement a switchover from one EHR to another [1]. This means that any new entrant needs a lot of startup cash and the ability to sell to provider systems. 3) You're building regulated software, which means that most of your third-party components have to be secure and HIPAA-compliant. Also your developers probably don't have access to production data or error reporting, so fixing issues can take a long time. 4) You're building enterprise systems. Whatever you build has to support EVERY department in a provider from the ER, to the pharmacy, to billing. What makes sense for one department won't make sense to another so you end up with products built by committee.

PS You may be interested in looking into some of the modern EMRs that have come out. They seem to focus on either small practices or specialties. DrChrono is one, Flatiron Health is another.

[1]: https://www.epic.cumc.columbia.edu/timeline/implementation-p...

> Epic is a steaming pile of shit.

I'd love to hear more about this.

I know of this industry, in the hardware side, that if someone "has a way" to get a device certified to selling in a country, that someone is sitting on a lot of money. It's ridiculous. The laws made to protect user's from low quality products became a corruption-bureacracy user's money exploiting shady horrible industry. Not to mention the ethics on the Patents related side of the medical tech.

i tried for 3 years to create an emr. we had some success (its being used in the real world) but not enough to be sustainable against the monoliths like epic

my 0.02 on why they arw so bad:

1. the number one goal of emr software is not quality of care. its not user experience. its suppprt and integration for billing. ie, they are accounting systems

2. this partly follows from the fact that emrs are not purchased and implemented by the same people who actually use them. a government mandate and high level health execs make the decision and their priority is the bottom line, in terms of $$$.

3. its not an encouraging field to get into. there is lot of regulation and standards to deal with and not much money to go around to pay for it

> why the EMRs are all terrible.

Because the primary contribution to a patient's health is factors beyond the healthcare provider's control, like the patient's wealth, education and genes, followed by a mix of real and placebo effects in evidence-based medicine.

Generously, if IT is number three, it's just not going to work as much per dollar as say, transferring money directly to the patient (in the form of free healthcare) or R&D funding (in institutions like the NIH).

You might enjoy this video then: https://www.youtube.com/watch?v=xB_tSFJsjsw

Now on to your question. A lot of the EHR systems have grown organically and in the end they are made by people with software experience but not per se medical experience.

Disclaimer first: I work on a competitor for Epic (Well, they only run in 3-4 hospitals in my country and we run in 30, so they're not really a threat). And I'm not saying we are doing it better than Epic, but these are a few of my thoughts (loosely collected).

1) Most of us don't have medical training. Our campus is shared by a university / hospital, so we do have access to a lot of doctors and when they work with us, their sides of the program tend to run quite well. But, not all doctors want to spend time on this (understandable) so one system might be well tailored for doctor A, but not to Doctor B.

2) The systems are _old_. They grow organically, but Epic is a few decades old and our system started in 1992, so it's by no means 'young' either. The Medical field evolves and so does the software field - we are playing catchup on both most of the time.

3) We can't afford downtime. We can't just decide "upgrade tonight and go offline for a bit". So we have to engineer around this. We release a new system of our software every week (one week testing by a dedicated team with medical knowledge, then production). But since we can't do a bulk update and force people to restart, we need to deal with different versions which impacts the code quality.

4) We run on multiple hospitals. Hospitals have their own way of working, but to tailor to each hospital directly becomes impossible at some point. In the end you get a system that doctor Y from hospital A really likes, but doctor B from hospital A doesn't. Here, sadly, money will factor in. If you want the software more tailored you will have to pay more. And this is not just to the doctors to decide.

5) These systems are _huge_. (SVN reports a linecount of just over 37 million at the moment). Because they do a lot, to give an idea: * Patient scheduling (find optimal time for a patient to get X therapies) * Decision support for doctors (suggested medication, ..) * AI System to monitor patients in the hospital * Scheduling for the devices (MRI, CT, ..) * Communication system between specialists (GDPR compliant nowadays, yay) * Import/Export to share with all other hospitals * Billing system * Social insurance integration * Government integration * Medication system * ER System * ....

This is a pretty loose collection of some things that come to mind. :)

I own a company that is attempting to modernize electronic medical records, and there are plenty of legitimate reasons why medical software is bad. The decision makers at large hospitals are even more conservative than other large enterprises. Nobody is willing to risk their career/hosptial by writing an eight or nine figure cheque to an unproven enterprise to run their critical infrastructure. So the market is dominated by companies that were earliest to market (Meditech, Epic, Cerner, Allscripts) or have big enough pockets to compete (GE's Centricity, McKesson). Nobody ever got fired for choosing IBM.

Hospitals were some of the earliest adopters of computers and the systems that were developed have been in place for decades. Most of these systems carry a lot of baggage from their development history and are too risky or brittle to modify without good reason. It's the same story that most big banks find themselves in. Everything is clunky, confusing, and old but the job gets done and that is good enough.

The people making purchasing decisions do not care about the usability of the product, as they are not going to be the ones using it. They mostly care about billing, compliance, data integrity, security, and managing costs. Electronic medical records primarily exist for billing an insurance company or the government. Every other reason given for their existence is secondary to that need. If you don't believe me, check how many staff members are dedicated to billing at your hospital. For reference, the average staff ratio is 2.7 billing professionals for each physician across the industry.

Medicine is both an art and a science. Even physicians in the same discipline vary greatly in the way that they practice their craft. There is no way to turn a hospital visit into an assembly line. Nothing is standard, everything is specialized. This is at odds with the ideal software project, where everything is defined up front and can be turned into an efficient repeatable process. If it were easier I think you would have greater competition which would lead to better products.

Can we do better? Absolutely. The limiting factor is not software development talent. You need to understand how the industry operates to be able to find a foothold. Attacking the incumbents directly will end in certain doom. If you want to learn more about how we're approaching the problem my email is in my profile. We built a product to run our own clinic and are almost ready for wide release.

> I’m upvoting this in hopes that someone with more knowledge than me can explain why the EMRs are all terrible.

As another poster said: EMRs are billing and compliance tools, paid for by administrators that are intending to buy billing and compliance tools. They support that function very well.

The people using it clinically (a) aren't the ones paying for it, and (b) tend to be inconvenienced by the very same billing and compliance elements that make it worth paying for.

EMRs don't suck for clinicians because it's impossible to have a nice clinical interface. They suck because they're a non-clinical tool being shoe-horned into a clinical setting and forcing clinicians to do the data entry work.

The interoperability standards are actually pretty much there, at least for the more common types of patient records. The situation has improved tremendously in just the past few years. Epic and their major competitors are doing a decent job of supporting the lastest HL7 CDA and FHIR standards, including specific implementation guides.

The bigger problem now is that many provider organizations haven't upgraded to current EHR release versions, or haven't opened up their systems to access by other providers.

> - Epic makes it easier for medical directors to track population health and impose standard protocols of care. Individual practitioners don't always like that! I am not expert enough in those fields to say who's right, and I suspect there's not always an obviously correct answer.

This is more complex than it looks. Picture this: "I only have 12 minutes to spend with this patient, the initiative wants me to spend 7 minutes talking about things x,y, and z but the patient has come in with an urgent pain and rash that have nothing to do with x, y, and z. I must fail on this initiative in order to do what the patient has actually come to me for."

There are a lot of things that intersect here. One is creating initiatives that really belong to public health and creating an unfunded mandate to make healthcare organizations provide that (e.g., active outreach to get people to come in and get their flu shots). No specialist in X - whether it be tech or healthcare - wants to get tangential work piled on them that has nothing to do with their actual work, and not get paid for it to boot. But if you don't do it, your reimbursement gets dinged. Essentially all healthcare policy in the US is built around sticks, not carrots: do X or we will withhold your pay.

Two, is that these initiatives often diverge from what the patient wants. A patient comes in for an acute rash and pain, and they really don't want you using up the time you need to examine them to discuss something else. "What about general check-ups?" One of the discoveries early on in the ACO program roll-out was that regulations built on the assumption that patients had annual check-ups went belly-up. Patients tend to go to their doctor a lot less often than that. Many people only go once every 18-24 months - too infrequently to pass the standards the government is imposing.

Three, the system as it stands only funds brief visits, often inadequate to properly taking care of patients.

Four, all of these things have to be documented in a very rigid way, to ensure that the EMR can ultimately be used to upload the results to the government or insurer so that they can assess how you did.

Five, the EMR strong-arms you into doing these things, jumping through a bunch of bad UI along the way. If it's inappropriate to provide, you still have to do a bunch of mandatory paperwork, because ... well, the EMR is really trying to strong-arm you.

All of these issues compound and intersect. The EMR is there to force you to do the thing that isn't yours to do and that the patient doesn't want and you have to choose between that and what they want because the funding is built around visits too brief to do both.

But, isn't it good to try and get docs to actually give everyone flu shots? Yes it is. I used to administer one of these programs, and I assure you, doctors' perception of how thoroughly they provide preventive care is wildly divergent from how thoroughly they actually do it. But instead of offering a carrot, we've created a stick, and system to apply it to the bottom of doctors' feet.

> - A lot of doctors dislike the underlying mechanic - being forced to actually write down everything they're doing and why - on top of sub-par UIs. The goals of the system conflict with the goals of the practitioner.

That's not the problem. The problem is that the reimbursement agency - be it CMS or an insurer - wants things documented in a very particular, rigid way that requires a shit-ton of boxes to be clicked and fields to be filled out, so that every little task becomes magnified by an order of magnitude.

Teaching docs to link together assessment + plan wasn't very difficult. Convincing them to fill out a form every time they don't give a hypertensive an ACE/ARB is more of a pain in the ass ("Because this guy's going to be dead in three to six months, so the monetary cost and potential side-effects aren't offset by any clinical gains whatso-fucking-ever" is annoying to repeatedly fill out. Or more realistically, "because he's telling me his urine output in response to his diuretics is going down and I'm concerned we're boxing his kidneys, so maybe I'll give him an anti-hypertensive that doesn't directly beat on his kidneys... even though his creatinine hasn't jumped yet, so I have no actual evidence to present to the government/insurer to support this, so it looks like I'm just ignoring standards willy-nilly.")

Is your experience with Epic from over a decade ago? Transferring patient data, especially between two Epic systems, is much easier than that. As far as the UI goes, over the last four years Epic has had a big push to hire experts and make processes to help develop and test good user experience, as well as trying to make it look better. But that takes time in a big, old, complicated system, and you still have hospitals who haven't upgraded from 3+ year old versions. The latest version already looks immensely better in lots of areas, but you also can only do so much to make an easy to use program that also gets dense medical data in front of the right pairs of eyes so that patients don't die.

Yes, the UI is complete garbage.

Re: record transfer, I recently found that there is a way to create an encrypted file from the patient’s chart that can be emailed , put on a flash drive, whatever. I’m planning to give it a try soon, but in the meantime, my clinic still uses the prehistoric method you describe.