Why Not AI?

AI often looks like the natural answer to “how can we parse oncology documents and quickly suggest the next step to the patient”. For HealthOS Onco, that logic is unsafe if it turns consultation preparation into opaque medical advice.

HealthOS Onco is not built as an autonomous smart adviser that selects therapy, medicines, clinical trials, or clinics from patient documents. For oncology preparation, verifiability, reproducibility, explicit data gaps, and clear responsibility boundaries matter more.

Product Principle

The product should be structured so the patient and doctor can understand how the report was produced: which document was the source, which fact was confirmed by the patient, which information is stale, where there is a conflict, and what is missing.

The service may use OCR and NLP to extract draft facts from documents, but a draft does not become a confirmed oncology fact without review. Unknown is not treated as “no”, and conflicting data are explicitly marked.

Independent Review

For medical software, independent review of the basis for an output is important: the doctor should be able to see the basis without special trust in an internal model. In a patient product this principle is even stricter, because the patient does not have to understand the algorithm, but must see where there is a fact, a gap, a warning, and a question for the doctor.

HealthOS Onco therefore shows:

This approach fits oncology consultation preparation better than an opaque probabilistic recommendation.

Why It Matters

Automation Risk HealthOS Onco Approach
The system sounds certain with incomplete receptor, stage, or prior-therapy data It first shows gaps and readiness status
A document draft looks like a fact It separates draft, document-based fact, and patient confirmation
Cost starts pushing the patient toward a direction Risks and completeness come first, indicative cost second
The patient reads a comparison as a medical order The result is phrased as questions and topics for the doctor
An error is hidden inside the model Sources, versions, dates, and limits are shown
Model behavior changes after an update Reports are built on versioned rules and a price snapshot
The user over-trusts confident wording Uncertainty is shown explicitly instead of being masked as a finished answer

Automation Bias

The more confident an automatic answer looks, the greater the risk that a person will accept it as a ready decision. For an oncology patient this is especially sensitive: anxiety, urgency, cost, and the wish to “do everything right” can make a polished comparison look like medical direction.

Therefore the HealthOS Onco interface should not imitate a final doctor’s decision. It gathers facts, highlights gaps, explains risks in plain language, and prepares questions. Medical decisions stay with the doctor.

Determinism and Audit

Reproducibility matters for the doctor report. If the patient uploads a new pathology report, confirms a medicine list, or updates the region a week later, the system should show what changed and why. If a doctor sees the report, they should understand which rules, document versions, and price snapshot were used.

HealthOS Onco relies on deterministic completeness and safe-display rules:

Where AI Is Useful

AI can be useful as an auxiliary layer, but not as the decision layer. Appropriate tasks include:

In all these scenarios the result remains a draft or analytical signal. It does not become a medical order, does not replace patient confirmation, and does not remove the need for doctor consultation.

Role of Automation

Automation is needed for order: parse documents, collect fields, find gaps, highlight risks, and prepare a report. But medical decisions remain with the doctor, while the patient receives a tool for preparing the conversation.

HealthOS Onco does not diagnose, does not issue medical orders, does not select therapy for the patient, is not a drug marketplace, and does not guarantee the cost of medicines or medical care.