In healthcare analytics, what we know about patients depends on the data we can see. That sounds obvious, but the implications are anything but.
At BiotechX USA 2026, I presented on what happens when you rely on a single data source to identify patients, and why the field needs to move beyond that model entirely. The short version: A clinical signal is not the same as sufficient evidence to act. And in rare disease, the gap between those two things can mean everything.
The Problem With Silos
Take a patient presenting with low hemoglobin and elevated LDH. Those findings may raise concern for anemia and hemolysis, but they do not establish the underlying cause or confirm PNH. However, the labs don’t confirm PNH. Act on that signal alone and you risk sending a medical science liaison into a conversation that has nothing to do with the disease in question. Ignore it, and you may miss a patient who genuinely needs attention.
Layer in claims data and the picture sharpens: an ICD-10 code for PNH, a hematologist visit, a complement inhibitor on record. That’s better, but still not what we’re actually looking for. A diagnosed, treated patient isn’t necessarily a patient with an unmet need.
It’s only when you add the EHR — clinical notes capturing breakthrough episodes, weight changes, documented discussions about switching therapy — that the full picture comes into focus. Not just who this patient is, but where they are in their treatment journey and what the right next step looks like. The picture becomes more clinically meaningful: not just whether the patient has PNH, but whether there is evidence of ongoing disease burden or a potential treatment decision.
Scale Changes Everything
Doing this for one patient is a thought exercise. Doing it for millions of patients across billions of data points is a fundamentally different problem, and that’s where AI becomes essential.
No two patients look the same, and no two providers document symptoms the same way, so data is fragmented and gaps exist. AI helps extract relevant information from clinical notes, identify patterns across sources, and flag where evidence is incomplete. It makes that work possible at scale, while experts define the criteria, evaluate the evidence, and determine what warrants action.
The Expert Still Owns the Threshold
What often gets lost in conversations about AI and automation is that different decisions require different standards of evidence. A market sizing exercise can tolerate more uncertainty than a regulatory submission. A field team needs timely data, even if it means accepting some signal gaps.
The technology surfaces the candidates, but humans determine whether the evidence is credible, relevant, and strong enough to act on. That judgment isn’t something you can outsource.
A Question Worth Asking
Here’s a challenge I’d extend to anyone working in this space: The next time someone hands you a patient count, a provider list, or a model output, ask what decision you’re actually trying to make. Does the data give you enough evidence to make it? What are you still uncertain about, and would more data change what you’d recommend?
If you can answer those questions, you can turn a promising signal into action. If you can’t, you know where to look next.
That’s the evolution of real-world evidence — not just more data, but smarter, connected intelligence that tells you what to do with it. That’s the evolution of real-world evidence: connected intelligence that helps us understand what the data supports, where uncertainty remains, and when there is enough evidence to act.
NorstellaLinQ links claims, EHR, lab, and electronic prescription data into a single patient view — decision-grade and AI-ready. See what it can do for your team. Contact us.
