matiasmolinas/dental-analysis/oral-systemic-analysis

Oral-Systemic Analysis Skill (HISTORA)

Analyze a combined periodontal + medical record and produce a structured, traceable, non-diagnostic research output linking oral and systemic health.

When to use

A user provides integrated dental and medical data (probing depths, bleeding on probing, bone loss, treatments, plus blood pressure, diabetes/HbA1c, lipids, smoking, medications, cardiovascular history) and wants relationships, risk profiles, or research hypotheses — never a diagnosis or treatment plan.

Input format (optimized)

Present the record with named sections and glossed clinical terms, keeping oral and systemic data co-present so cross-domain relations are representable:

## SHARED RISK FACTORS
smoking, diabetes/HbA1c, hypertension ...
## CARDIOVASCULAR PROFILE
lipids, hs-CRP (mark MISSING if absent), family history, statin ...
## PERIODONTAL PROFILE (longitudinal)
PPD, CAL, BOP (= gingival inflammation marker), bone loss, diagnosis, maintenance ...
## QUESTION
oral-systemic risk profile and relational axes?

Gloss terms that carry mediating meaning (e.g. "BOP = bleeding on probing, a marker of gingival inflammation"). Include a one- to two-sentence mechanistic bridge when the inflammatory link is relevant (periodontitis -> systemic inflammatory burden -> cardiovascular risk via markers such as CRP).

Mediating concepts to reason through

inflammation, C-reactive protein, cytokines, atherosclerosis, endothelial dysfunction, bacteremia, oxidative stress, cardiovascular risk. Shared factors (diabetes, smoking) belong to the shared_behavioral / metabolic axes.

Output

Valid JSON per schemas/output_schema.json: risk_profile, relational_axes[] (each with oral evidence, systemic evidence, hypothesized mechanism, confidence, traceability), required_missing_data[] (fields to COLLECT — never imputed patient values), research_hypotheses[], non_diagnostic_disclaimer: true.

Guardrails

• Non-diagnostic; no treatment advice.
• Never impute a patient-specific value. Missing mediating data becomes a

collection flag, not a guess.

• Every axis cites the exact input fields it used.
• Factor-grounded / counterfactually coherent. Each relational axis must cite

the specific present patient factors that drive it; if a driving factor is absent from the record, that axis must be correspondingly weaker or omitted (do not assert an axis a removed factor would not support). Confidence must track the factors actually present in this record, not general population priors.