SHAARPEC is a predictive analytics platform that structures your existing healthcare data into longitudinal patient trajectories, trains deep learning models on those trajectories, and delivers forward-looking commercial intelligence — deployed inside your environment, working with data you already own.
Most analytics platforms treat claims as rows in a table. SHAARPEC treats them as what they actually are: a record of a patient moving through the healthcare system over time. We structure raw claims data into clinical encounters, assemble those encounters into longitudinal patient trajectories, and store those trajectories in a graph-native data model that preserves every temporal relationship — not just what happened, but when, and in what sequence relative to everything else.
The platform starts by training a foundation model on the broadest possible population for a given condition — potentially hundreds of thousands of patients nationally. That foundation model learns the full range of how patients with that condition move through care. We then fine-tune it for each specific therapy, indication, and commercial question. This means every downstream model inherits population-scale learning and specializes from a position of depth, not from scratch.
Each step runs on a weekly processing cycle as new data arrives.
Raw claims — open and closed — enter the platform and pass through an encounter grouper that converts billing lines into discrete clinical encounters: infusion visits, surgical events, imaging studies, specialist consultations. Those encounters become the building blocks of patient trajectories, assembled into a graph database that captures the full temporal sequence of each patient's care journey. The closed claims corpus trains the foundation model, establishing a population-scale understanding of what complete, well-characterized patient trajectories look like for a given condition.
Each week, a new batch of open claims arrives. The same encounter grouper runs, converting incoming claims into encounters and appending them to each patient's growing longitudinal record — inserted at the correct point in time, preserving the trajectory's temporal integrity. A cohort surveillance module continuously evaluates whether each trajectory contains enough signal to identify the patient as a candidate for a specific indication.
Open claims are structurally incomplete — submission lag, partial adjudication, and reporting gaps mean that a patient's open claims record systematically undercounts clinical activity. The platform's trajectory completion model applies what the foundation learned from closed claims to reconstruct what open claims cannot see. The result is an enhanced trajectory that more accurately represents the patient's true care journey, with all reconstructed encounters flagged transparently.
For each outcome of interest — treatment initiation, therapy switching, new diagnosis — the platform has learned a patient archetype: the characteristic trajectory signature that precedes that outcome. An archetype is not a demographic profile; it is a temporal pattern, the specific sequence of clinical events and their timing that the model learned to associate with a given outcome across a population. As enhanced trajectories develop, the platform continuously evaluates how closely each patient's trajectory matches the relevant archetypes. This is the mechanism that bridges model output and commercial action — and the reason predictions are interpretable by clinical and commercial teams, not just data scientists.
A prediction readiness module evaluates each trajectory continuously. When archetype match and signal strength cross a defined confidence threshold, the prediction module fires — generating a timestamped prediction for that patient and that outcome. Predictions only register when confidence is sufficient. This threshold-gated design minimizes noise and ensures the signal reaching the commercial team reflects genuine emerging activity, not statistical artefact.
Individual patient-level predictions aggregate to the provider, medical center, and geography level — the unit of action for pharma commercial teams. Dashboards show where demand for a specific therapy is forming, which HCPs are likely to have eligible patients in the next three to six months, and how signal strength is trending week over week. Every prediction is tracked against what subsequently happens, feeding a continuous validation loop that keeps models calibrated as treatment landscapes evolve.
SHAARPEC does not require access to a proprietary data asset and does not create a third-party data dependency. The platform is deployed inside the client's own firewall or protected cloud environment, working with data the client already has — or has the right to access.
The platform runs inside your secure environment. Patient-level data never leaves your control. SHAARPEC brings the methodology; you bring the data.
Optimized for open and closed claims — the data most pharma commercial teams already have access to. Compatible with EHR, pharmacy, and lab data where available.
A single platform installation supports multiple therapy pipelines. Each pipeline runs independently, with its own foundation model, archetypes, and prediction models — sharing infrastructure, not data.
Every prediction the platform makes is tracked against what subsequently happens. Continuous validation loops recalibrate models automatically as treatment landscapes, approvals, and patient behavior evolve.
Signal designed to inform specific decisions — not generic utilization reports.
By HCP, medical center, and geography — 3 to 6 months before it appears in prescription data. Enables earlier field force positioning and earlier engagement with the right providers.
Trajectory-based prediction of time-to-treatment for patients currently in the diagnostic or pre-treatment phase. The output most natively aligned with launch strategy and resource allocation.
Forward-looking switching propensity at the patient and prescriber level — not who switched last quarter, but who is on a trajectory that predicts switching in the next 90 days.
Signal aggregated to the HCP and site-of-care level, ranked by predicted future activity. Replaces territory planning based on historical prescribing with planning based on forward-looking demand.
Encounter density anomalies at the prescriber level that indicate patients hitting step therapy, prior authorization friction, or other access barriers — surfaced for patient services and market access teams.
The platform architecture is outcome-agnostic. If the outcome can be defined and observed in claims data, a prediction model can be trained for it. New outcome models are added within the existing pipeline without rebuilding the foundation.
We start with your data question and show you what the signal looks like for your therapy.
Request a Demo