AI for Commercial

Patient Journey Mapping with AI and Real-World Data

ANG Associates
Life Sciences & AI Consulting

Oct 2025 7 min read

Why the Patient Journey Still Hides in Plain Sight

For most rare disease patients, the path from first symptom to correct diagnosis is not a straight line. It is a sequence of missed signals scattered across primary care visits, emergency departments, specialist referrals, and lab orders that no single clinician ever sees end to end. The scale of the problem is well documented. In a Rare Barometer survey of 6,507 people living with 1,675 different rare diseases across 41 European countries, published in the European Journal of Human Genetics, the average total time to diagnosis was 4.7 years, and 56 percent of respondents waited more than six months after their first medical contact before receiving a diagnosis. Twenty-two percent consulted at least eight healthcare professionals during their search, and 73 percent were misdiagnosed at least once along the way. These are not gaps in medical knowledge alone. They are gaps in how fragmented healthcare data is connected and interpreted, which is precisely the problem that AI-driven patient journey mapping is built to address.

Methodology: Turning Longitudinal Records into Sequences

Patient journey mapping with real-world data (RWD) treats each patient's history, claims codes, EMR encounters, lab results, and increasingly patient-reported outcomes, as an ordered sequence of events rather than a static snapshot. The analytical approach generally follows a few consistent steps.

Data harmonization: claims, EMR, and lab feeds are linked at the patient level and normalized into a common timeline, since each source uses different coding standards and capture windows.
Sequence representation: techniques such as document-embedding algorithms (for example Doc2Vec) convert each patient's chronological event history into a numerical vector that preserves order and context, so that patients with similar clinical trajectories end up numerically close to one another.
Clustering: unsupervised methods such as k-means or hierarchical clustering are then applied to these vectors to group patients into distinct pathway archetypes, for instance a "fast diagnosis" cluster versus a "long odyssey with repeated misdiagnosis" cluster.
Validation against outcomes: clusters are checked against known clinical endpoints (time to treatment, mortality, hospitalization length) to confirm the groupings are clinically meaningful rather than statistical artifacts.

A 2025 study in BMC Medical Informatics and Decision Making illustrates the approach directly: researchers applied Doc2Vec embeddings and k-means clustering to the pediatric intensive care journeys of 1,853 patients at Great Ormond Street Hospital, built from 647 unique lab and medication event types. The analysis surfaced five distinct patient clusters with statistically significant differences in mortality, age at admission, and length of stay, including one high-mortality cluster dominated by children with neoplasms and one short-stay cluster resolving in under three days for most patients. The same embedding-plus-clustering logic scales to outpatient and claims-based rare disease pathways, where the "events" are referrals, specialist visits, and diagnostic tests rather than ICU lab draws.

What the Evidence Shows

Applied to rare disease specifically, this kind of sequence analysis is already producing findings that go beyond what manual chart review could surface. Komodo Health, working with Regeneron, built a machine-learning model on claims data from its Healthcare Map platform to flag patients likely to have homozygous familial hypercholesterolemia (HoFH), a condition affecting roughly 1 in 250,000 people that is frequently missed because it lacks a distinct diagnostic code. The model, described in a peer-reviewed paper in Scientific Reports (2024), identified 331 confirmed HoFH patients within the claims database by recognizing patterns across comorbidities, lab values, and treatment histories that individually looked unremarkable but collectively pointed to the disease.

Even among rare disease patients diagnosed within one year, 28 percent had still consulted five or more healthcare professionals, according to the Rare Barometer analysis published in the European Journal of Human Genetics.

That statistic is a useful reminder that diagnostic delay and referral churn are related but distinct problems: some patients reach a diagnosis relatively quickly but only after an inefficient tour through multiple specialists, which is exactly the kind of referral bottleneck that sequence clustering is designed to detect and quantify. IQVIA's own discussion of patient journey analytics, drawing on a multi-country expert panel from Brazil, France, the UK, the US, and Japan, frames the core obstacle similarly: patient data is fragmented across public and private providers, and integrating EMRs, claims, labs, and other sources into a single longitudinal view is what allows analysts to identify substantially more of the right patients, faster, than manual approaches allow.

Building the Capability: Where a Firm Like ANG Associates Fits

None of this happens by installing a clustering library. It requires a data engineering foundation capable of ingesting and harmonizing claims, EMR, and patient-reported sources under real privacy and provenance constraints, a validated analytics layer that a GxP-regulated organization can defend to auditors and regulators, and a delivery structure, typically SAFe or Agile at scale, that gets pilot models into production without stalling in a proof-of-concept loop. ANG Associates works with life sciences organizations across exactly this stack: helping teams define a defensible AI and RWD strategy up front, building and validating the underlying data pipelines to GxP standards, and running the SAFe-based delivery cadence that turns a promising patient journey model into a capability the medical affairs, market access, or clinical development team can actually rely on. For organizations earlier in the rare disease space, where signal is scarce and every data source needs careful validation, that combination of strategic clarity, compliant engineering, and disciplined delivery is usually the difference between an interesting analysis and a capability that changes how patients are found and referred.

Sources

patient journey mappingreal-world datarare diseaseartificial intelligencemachine learning in healthcare

Interested in this topic?

Let's discuss how we can apply these approaches to your organization.