Regulatory compliance (SaMD) and clinical validation
Milestone document. SauBit is a decision-support tool (the prescriber decides). Formal regulatory framing is conducted with specialist assessment before commercial expansion.
Software as a Medical Device (SaMD)
Given its intended use (prescription validation, risk classification and recommendations), SauBit runs a formal SaMD framing analysis with ANVISA (RDC 657/2022 and related 751/2022, 848/2024). We do not assert the final class here — it depends on the intended use and specialist assessment.
Maintained documentation
- Intended use, intended users and allowed/forbidden scenarios.
- Risk management (ISO 14971) + clinical hazard matrix.
- Software lifecycle (IEC 62304) and product safety (IEC 82304-1).
- Cybersecurity in the lifecycle (IEC 81001-5-1) and information security (ISO 27799).
- Traceability requirement → rule → test → evidence (see rule catalog and
GET /api/v1/rules). - Change control, formal version release, post-market surveillance, CAPA and a rollback plan; record of who approved each clinical rule (rule governance).
Clinical validation dossier (structure)
Overall accuracy is not the key metric: in medication safety, a negative-heavy set can mask critical false negatives. We report:
| Metric | Why |
|---|---|
| Sensitivity for contraindications and severe interactions | Catch what must not pass. |
| False-negative on critical risk | Safety metric #1. |
| Specificity · PPV · NPV · false-positive | Alert fatigue. |
| Normalization coverage (PT/EN/brand) | Normalization miss = false negative. |
| Dose precision (UCUM) | Max dose/adjustments. |
| Partial response and unverified pairs rate | Coverage transparency. |
| Results by population (pediatrics, geriatrics, renal, pregnancy) | Generalization. |
| Confidence intervals · external review · version comparison | Reproducibility. |
Each release has pass/fail criteria and the list of known uncovered cases, with the evaluation cut-off date.
Status: structure defined; the quantitative dossier is published as the labeled validation set grows. See also Security and LGPD.