CHEQBACK™ — AI Care-Plan Consistency Checker

CHEQBACK — Don't Just Tell. Check Back.

Healthcare instructions can be individually clear—and still become collectively inconsistent.

A medication may be continued in one encounter, held for a procedure in another, and described differently at discharge. The patient may understand every instruction correctly, yet the overall care plan can still drift as information moves across clinicians, care teams, documents, and time.

That observation inspired CHEQBACK.

The idea

CHEQBACK is an AI-powered care-consistency layer designed to continuously check whether a patient's evolving care plan remains consistent across the healthcare ecosystem.

It is not a diagnostic system and does not prescribe or make clinical decisions.

Instead, CHEQBACK asks a simpler but important question:

"Do these instructions still agree?"

It brings together relevant information from sources such as:

Clinical records and medication orders Care-team communications Discharge instructions Patient-reported text or voice Encounter, procedure, and timing context

The system extracts clinically relevant information, normalizes terminology, considers temporal context, and compares instructions across sources.

It then produces one of three safety signals:

ALIGNED — no meaningful discrepancy detected UNCLEAR — information is incomplete or ambiguous; clarification is needed CONFLICT — a potentially meaningful inconsistency requires human review

The goal is deliberately narrow: detect the gap, surface the evidence, and put it in front of the right human.

Why this matters

The idea is grounded in evidence that medication understanding and continuity can break down even when individual instructions appear understandable.

Research has reported substantial patient–physician disagreement about planned medication changes, medication discrepancies during transitions of care, and unintentional medication discrepancies after discharge.

This led us to a different framing of the problem:

The challenge is not simply patient understanding. It is maintaining consistency as the care plan changes.

What makes CHEQBACK different?

Most reconciliation approaches focus on a particular encounter, document, or medication list.

CHEQBACK is designed around the continuity of the care plan.

It considers:

Meaning + Source + Time + Clinical Context

rather than simply matching keywords or comparing two medication lists.

For example, "continue aspirin" and "hold aspirin before the procedure" may both be valid statements—depending on when they apply.

CHEQBACK is designed to recognize that temporal context rather than incorrectly treating every difference as an error.

Technology concept

The proposed architecture combines:

LLM / NLP for extracting and understanding clinical information Clinical terminology mapping for normalization Temporal reasoning to understand when instructions apply Clinical rules and safety constraints to constrain AI interpretation FHIR-ready integration to connect with existing healthcare systems Speech-to-text for patient-reported voice input

The architecture is intentionally designed as a layer over existing healthcare infrastructure, rather than requiring hospitals to replace their EHR, CRM, ERP, patient portal, or other systems.

A key design principle is:

AI detects. Evidence explains. A clinician verifies.

What we learned while developing the concept

The biggest insight was that the problem becomes more interesting when viewed as a consistency problem rather than a comprehension problem.

A patient can correctly understand an instruction and still receive another instruction later that conflicts with it.

Similarly, two clinical documents can each be internally correct while becoming inconsistent when considered together.

That means a useful healthcare AI system should not only ask:

"What does this document say?"

It should also ask:

"Does this still agree with everything else we know about this patient's care?"

Challenges

The hardest part is not simply applying an LLM to medical text.

Healthcare requires a much higher bar for safety.

CHEQBACK therefore needs to address challenges such as:

Ambiguous clinical language Conflicting but time-dependent instructions Incomplete patient information Clinical terminology normalization False-positive alerts and alert fatigue Confidence and evidence traceability Privacy and healthcare data protection Safe human-in-the-loop verification

For that reason, CHEQBACK is intentionally positioned as a decision-support safety signal—not an autonomous clinical decision-maker.

Path forward

The proposed development path is:

Prototype → Validation → Limited Clinical Pilot → Scale

The first stage would focus on representative, de-identified care scenarios and test whether CHEQBACK can correctly identify meaningful inconsistencies while minimizing false alerts.

Subsequent validation would compare its signals against expert clinical review before considering deployment in real clinical workflows.

Ultimately, CHEQBACK aims to add something simple to healthcare:

A continuous "check back" on the care plan—before the next person, system, or setting acts on it.

Built With

• ai
• fhir
• healthcare
• llm
• machine-learning
• natural-language-processing
• speech