By Kristy Drollinger, Chief Product Officer, Lucem Health
LinkedIn: Kristy Boucher Drollinger
LinkedIn: Lucem Health
Advanced AI predictive models are getting better at finding patients with undiagnosed disease before symptoms, or a crisis, make the diagnosis unmistakable. They can interpret subtle clues already sitting in EHR data to surface elevated-risk patients who warrant further evaluation. At the same time, advances in diagnostic testing and new therapies that can delay progression, or even cure disease, when caught early, are making that early window more consequential than it’s ever been.
Health IT leaders are feeling this from both directions at once. Clinical leaders want to act quickly on these breakthroughs. And the financial case makes it hard to say no: as disease progression slows, many of the most severe (and most expensive) complications are avoided.
As value-based care expands, models like CMS’s Long-Term Enhanced ACO Design (LEAD) are asking organizations to hold risk over longer periods. The longer the risk window, the earlier action pays off. But as the opportunities to intervene earlier multiply, so do the demands on health IT.
Leaders now must decide where to point already oversubscribed technology resources and how to prioritize those investments alongside clinical teams that are equally stretched. The challenge is no longer simply whether health systems can identify risk earlier. It’s deciding where earlier detection can make the greatest difference, and turning that insight into action.
A Predictive Model Doesn’t Produce Outcomes
Building a high-performance model that meets the bar for bias, generalizability, and clinical performance is real work. Running it across longitudinal data at population scale adds another layer of cost and complexity. And the model has to be precise enough that a care team can actually act on the results.
However, identifying high risk patients is just the first step. Getting a patient from high risk to diagnosis can require guiding the patient through a multi-step program including disease education, diagnostic tests or procedures, new treatment, insurance coverage verification, navigation around financial or access barriers and more. Delivering high quality proactive care requires a designed plan that fits into clear clinical workflows.
This is the distinction that determines whether an early-detection effort becomes a pet AI project or a program that delivers real clinical value: the model and the pathway around it must be designed as one system. Data requirements, identification, patient engagement, diagnostics, and intervention need to work together. A weak link anywhere breaks the chain between an accurate prediction and earlier care.
Healthcare doesn’t make this easy. Every disease has its own diagnostic process and its own treatment path, which means “build a predictive model” is never really the whole task.
Two Diseases, Two Very Different Programs
Chronic kidney disease (CKD) and adult-onset type 1 diabetes (T1D) both benefit enormously from earlier diagnosis. But they demand almost opposite program designs.
CKD is a scale problem. Roughly 35.5 million U.S. adults have it, and the CDC estimates about 90% don’t know it. Early-stage CKD is usually symptom-free, and because the U.S. Preventive Services Task Force doesn’t recommend population screening, most organizations don’t have a proactive kidney disease program. That means patients are often caught in late stage 3 or stage 4, after real damage is done.
New genetic tests can now pinpoint disease etiology, and new drug classes can slow progression, which makes early diagnosis more valuable than ever. But population-wide screening still isn’t cost-effective. The only way to match early disease patients with therapies that can help them is a high-quality predictive model doing the screening a blanket program can’t.
T1D is the opposite problem: not scale, but visibility. It’s far less prevalent, still incorrectly perceived as a childhood disease, and easily mistaken for type 2 when symptoms first appear. The result is that many adults aren’t diagnosed until they’re in a life-threatening diabetic ketoacidosis crisis.
New T1D therapies can meaningfully change the disease’s trajectory but only within a narrow treatment window, and an undiagnosed patient can silently progress past it. Here, a predictive model paired with disease-specific education, testing, and follow-up isn’t just useful, it can be the only thing standing between a patient and that crisis.
CKD and T1D are just two current examples of the same pattern: therapeutic breakthroughs changing the math on broader testing and early diagnosis. That pattern isn’t slowing down.
You Can’t Build All of This Yourself
Each new program pulls on the same finite team: data scientists to build and validate models, informaticists to design workflow and data flow, interoperability specialists to move data across organizational boundaries, analysts to track program performance, and program managers to hold it all together.
The pace of therapeutic innovation is outrunning most teams’ capacity to build everything in-house. Even the best-resourced organizations can’t develop every new capability themselves, especially as compelling new breakthroughs continue to emerge. Turning early detection into consistent clinical impact means deciding, deliberately, where to build and where to bring in a partner who’s already built it.
Build where you have a genuine advantage. Partner where someone else does.
This is where the build-versus-partner question gets misjudged.
It’s tempting to frame it as a technology evaluation: whose model performs best, whose data pipeline is cleanest, whose AI is most sophisticated. But a highly predictive model, on its own, won’t move outcomes. It only works if it’s embedded in a full program: provider support, patient education, patient engagement, a designed care pathway, and performance metrics that show it’s working.
That means the technology team can’t evaluate partners alone, and a partner can’t be chosen by clinical leadership in isolation from IT either. A dyad needs to assess the partner’s program as a whole, not just the tech underneath it. Ask what happens after the model flags a patient. Ask how the partner handles the diagnostic and treatment path for that specific disease. Ask whether their engagement approach actually reaches patients, and whether their metrics show real clinical movement, not just identification counts.
Your organization still owns the clinical relationship, the technology environment, and every decision about how care gets delivered. The build-versus-partner decision should come down to where internal teams have the expertise and capacity to create meaningful differentiation, and where outside capabilities can help accelerate the work without adding unnecessary complexity.
As earlier detection becomes critical across more diseases, health IT leaders need a repeatable way to engage clinical stakeholders, assess which opportunities are worth pursuing, and determine how to operationalize them. The organizations that do this well will be the ones that focus their internal resources where they can have the greatest impact, and look outside their walls when the expertise, infrastructure or experience already exists elsewhere.