The Data Debt Under the Model

Most health systems that get serious about AI fail in the same place, and it is not where they expect. There are a lot of guys on the chopping block because they have made some poor decisions and they are not getting any ROI, because they cannot even get the implementation to the point where it is actually testing anything of value. They have got that black hole piece of it. They do not understand the data.

Pilots often stall where there are gaps in available data. The most noticeable gap in a deployed system is between a working demo and an actual scaled system. Demos may prove the model, but, they may also expose missing or incomplete data.

The debt that hides under the tech debt

Technical debt in the healthcare space is kind of understood, but kind of not. What is really not understood is the underlying data technical debt, because they have invested in the tech but they have not invested in curating the data. In healthcare especially, they have got so much data technical debt they cannot even begin to touch AI solutions. The technical debt is disguising and hiding the data technical debt, because they will spend tens of millions on the tech but will not deal with the data quality.

The record was never built to answer the question

The data is hard to use because of how the record was built, not because anyone was careless. You saw it up close. When I got hired in New York about ten years ago working for a health system, the team was about eighty people, primarily people who just dealt with the technology. Data architects, BI developers, some visualization specialists. The difficulties really came from the fact that the people dealing with the technology they were leaning on did not have any understanding of clinical workflow. They did not understand how things flowed through the record. They did not understand that a discharge diagnosis could actually be more than one field that says discharge diagnosis. It might be a timestamp issue. You do not even know.

Up until we fixed that, the stuff that was coming out of it was crap. No one trusted it. They should not have.

Growth by acquisition makes it worse

If you are growing by acquisition, you are inheriting a matrix that is as ugly as anything you could imagine. Every time you acquire another company, you have got another one of those matrices that you have got to figure out how to connect. And when it comes to clinical data, there are a few unicorns here and there, but there is not a large number of people who actually know how to look at the user interface, who understand clinical workflow, who understand data flow, who can start to figure out how to build a model that supports all the things they want to do.

Ask one question. The only thing they normalize is the stuff that runs the business. So why is it we cannot fix the pieces that actually touch people in healthcare?

The cloud did not buy you a solution

When the data is this hard, leaders try to spend around it. The big guys, you have got your AWS, Google, Microsoft, they double down, double-digit millions in commitments. Then the parade begins. All these people knocking on your door, trying to get you to give them a use case. My guys, I try to tell you this: you have now got very, very pricey cloud compute and cloud storage. You have no solutions. That is not what they do. Your team is still going to have to do all the same work tomorrow that they are doing today.

What the missing data does

People brace for the obvious failures, like seeing a wrong answer on the screen, and they should be noticed quickly. The bigger risk is losing the trust of the organization if these gaps are not identified along with a simple set of bread crumbs to be able to quickly diagnose and correct them. Think of an inherited dashboard a dozen clinical units relied on, printed on unit walls and used in meetings. When a single column stops updating, trust in the whole tool can collapse. It has never been more needful to ensure visibility into the core data set that supports and measures organizational performance. There may be an issue with the data, but, there may be a larger issue of missing translation or translator that makes sure operations are run efficiently. Yes, the data quality and compliance is essential. Equally essential are the minds of excellent teams that provide the data and insights that warrant trust.

Readiness starts with the risk, then finds the yes

Readiness is a tough spot for healthcare organizations for a variety of reasons, not the least of which is they were given some false impressions around some of the major tech spending they have done. I think about it in terms of AI readiness. You can have all your data in great shape, but if your cyber is not top-notch, you have got risks you do not even know about. So the very first thing is a risk assessment. See what exposure you already have.

The first thing I do is help them understand where they are, because usually the issues are not the ones they think they have. I find them the areas they can say yes to, because they do have data in solid enough condition to trust for certain things. They are not usually the ones they want, but those are the ones they can get while we figure out the steps to get the other data sources ready.

A data club is not data ownership

Most organizations have what I call a data club instead of a data governance. They waste time bringing high-level people together to make decisions. After two or three meetings those people go away, and the only thing they end up doing is explaining why they cannot do anything. It is gridlock.

The other reason no one clears the debt: the people who understand the data keep leaving. You are throwing contract dollars at something that is not building any core competencies into your DNA. Every time something gets finished, all the know-how just walks out the front door, and your team cannot support it. Then you end up contracting longer term for people who do not know your business and are not going to learn it.

Treat the data as the asset

This part is not theory for me. My team lived it. One of the things that saved us, believe it or not, was the pandemic. Data quality, we began to pivot and treat it like the asset that it was, and that really came through then.

I sat with the team. Guys, we have a catchment in here of about six million people. I do not want to see an index of six million people. The maximum number of patients who could be occupying a bed across all of our health system is about sixty-five hundred. It was closer to forty-five hundred before the pandemic, but when we realized we had to spin up beds, we converted conference rooms to ICUs. So when a person walks through the door, as soon as the order is done to admit them, we have an automated routine that goes out, looks for all of the history we have on them, and brings them into a finite index. It is not massive, but it is very, very efficient and easy to figure out where these people are, and it follows them all the way through the time they are in the hospital.

That is the argument in one story. We improved when we finally treated the data like it mattered, and we could trust everything we built on it. Identifying data debt before you buy a model, and you may get something that is worth the time and money.

Continue reading from Hutchins Data Strategy

AI Cannot Absorb Responsibility → https://hutchinsdatastrategy.com/the-ai-health-pulse/ai-cannot-absorb-responsibility

We Can Do That → https://hutchinsdatastrategy.com/the-ai-health-pulse/we-can-do-that

On the Signal Room podcast

The Hidden Reason Hospital AI Keeps Failing | Angel Mena, MD → https://signalroompodcast.com/episodes/hospital-ai-pilots-fail

Healthcare Data Readiness and AI Adoption: Why 85% of Organizations Aren't Prepared | Ratnadeep Bhattacharjee → https://signalroompodcast.com/episodes/data-readiness-ai-adoption

The Dark Side of the $50B AI Medical Boom | Lorraine Fernandes → https://signalroompodcast.com/episodes/dark-side-ai-medical-boom

More from Chris Hutchins

Subscribe — one email gets you The AI Health Pulse, the Signal Room podcast, and a free chapter of Beneath the Signal → https://hutchinsdatastrategy.com/free-chapter

Book Chris to Speak → https://www.chrisjhutchins.com/

Learn about our work to improve mental health around the world → https://continuamindhealth.com/

Do you have a podcast? We offer a Free growth assessment from PodcastPull.com → https://podcastpull.com/

Read on BeeHiiv → https://aihealthpulse.beehiiv.com/

Read on Substack → https://aihealthpulse.substack.com/

Tags: AI Health Pulse newsletter · healthcare data debt · data technical debt · AI readiness · healthcare AI pilots · growth by acquisition · data ownership Hashtags: #HealthcareAI #DataStrategy #AIinHealthcare #HealthIT #DigitalHealth