feat: résumé clinique enrichi + preuves cliniques + validation QC batch

Améliore la qualité du codage CIM-10 sur 3 axes : - Contexte clinique enrichi (interprétations bio, traitements indicatifs, marqueurs sévérité) - Preuves cliniques structurées par diagnostic (evidence linking dans le prompt LLM) - Validation batch post-codage (1 appel LLM/dossier, ajustement confiance, alertes QC) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-17 21:47:27 +01:00
parent dbc5bdbaf4
commit 94fa4e5f3b
7 changed files with 988 additions and 16 deletions
--- a/tests/test_clinical_context.py
+++ b/tests/test_clinical_context.py
@@ -0,0 +1,264 @@
+"""Tests pour le module d'enrichissement du contexte clinique."""
+
+import pytest
+
+from src.medical.clinical_context import (
+    BIO_INTERPRETATIONS,
+    TREATMENT_INDICATORS,
+    interpret_bio_value,
+    detect_treatment_indicators,
+    detect_severity_markers,
+    build_enriched_context,
+    format_enriched_context,
+)
+from src.config import (
+    BiologieCle,
+    Diagnostic,
+    DossierMedical,
+    Imagerie,
+    Sejour,
+    Traitement,
+)
+
+
+# --- interpret_bio_value ---
+
+class TestInterpretBioValue:
+    def test_crp_major(self):
+        assert interpret_bio_value("CRP", "180", True) == "syndrome inflammatoire majeur"
+
+    def test_crp_moderate(self):
+        assert interpret_bio_value("CRP", "45", True) == "syndrome inflammatoire modéré"
+
+    def test_crp_minor(self):
+        assert interpret_bio_value("CRP", "8", True) == "syndrome inflammatoire mineur"
+
+    def test_crp_normal(self):
+        assert interpret_bio_value("CRP", "3", False) is None
+
+    def test_lipase_pancreatite(self):
+        assert interpret_bio_value("Lipasémie", "450", True) == "pancréatite biologique (>3N)"
+
+    def test_lipase_moderee(self):
+        assert interpret_bio_value("Lipasémie", "90", True) == "élévation modérée de la lipase"
+
+    def test_hemoglobine_severe(self):
+        assert interpret_bio_value("Hémoglobine", "6.5", True) == "anémie sévère (transfusion probable)"
+
+    def test_hemoglobine_moderee(self):
+        assert interpret_bio_value("Hémoglobine", "9", True) == "anémie modérée"
+
+    def test_plaquettes_severe(self):
+        assert interpret_bio_value("Plaquettes", "30", True) == "thrombopénie sévère"
+
+    def test_leucocytes_high(self):
+        assert interpret_bio_value("Leucocytes", "25", True) == "hyperleucocytose majeure (infection, inflammation)"
+
+    def test_leucocytes_low(self):
+        assert interpret_bio_value("Leucocytes", "1.5", True) == "leucopénie sévère (aplasie, immunodépression)"
+
+    def test_creatinine_severe(self):
+        assert interpret_bio_value("Créatinine", "350", True) == "insuffisance rénale sévère"
+
+    def test_unknown_test(self):
+        assert interpret_bio_value("Glycémie", "2.5", True) is None
+
+    def test_invalid_value(self):
+        assert interpret_bio_value("CRP", "positive", True) is None
+
+    def test_comma_separator(self):
+        assert interpret_bio_value("Hémoglobine", "6,5", True) == "anémie sévère (transfusion probable)"
+
+    def test_bilirubine_ictere(self):
+        assert interpret_bio_value("Bilirubine totale", "55", True) == "ictère franc"
+
+    def test_asat_cytolyse_majeure(self):
+        assert interpret_bio_value("ASAT", "250", True) == "cytolyse hépatique majeure (>5N)"
+
+    def test_asat_cytolyse_moderee(self):
+        assert interpret_bio_value("ASAT", "100", True) == "cytolyse hépatique modérée (>2N)"
+
+
+# --- detect_treatment_indicators ---
+
+class TestDetectTreatmentIndicators:
+    def test_insuline(self):
+        traitements = [Traitement(medicament="INSULINE LANTUS 20UI")]
+        result = detect_treatment_indicators(traitements)
+        assert len(result) == 1
+        assert result[0]["condition"] == "diabète insulino-traité"
+
+    def test_antibiotique_iv(self):
+        traitements = [Traitement(medicament="CEFTRIAXONE 1g IV")]
+        result = detect_treatment_indicators(traitements)
+        assert len(result) == 1
+        assert result[0]["condition"] == "antibiothérapie IV"
+
+    def test_multiple(self):
+        traitements = [
+            Traitement(medicament="Metformine 1000mg"),
+            Traitement(medicament="Enoxaparine 4000UI"),
+            Traitement(medicament="Paracétamol 1g"),
+        ]
+        result = detect_treatment_indicators(traitements)
+        assert len(result) == 2
+        conditions = {r["condition"] for r in result}
+        assert "diabète type 2" in conditions
+        assert "anticoagulation (HBPM)" in conditions
+
+    def test_no_match(self):
+        traitements = [Traitement(medicament="Paracétamol 1g")]
+        result = detect_treatment_indicators(traitements)
+        assert result == []
+
+    def test_dedup_conditions(self):
+        traitements = [
+            Traitement(medicament="Enoxaparine 4000UI"),
+            Traitement(medicament="Lovenox 4000UI"),
+        ]
+        result = detect_treatment_indicators(traitements)
+        # Les deux sont HBPM, mais une seule condition
+        assert len(result) == 1
+
+    def test_dict_input(self):
+        traitements = [{"medicament": "morphine 10mg"}]
+        result = detect_treatment_indicators(traitements)
+        assert len(result) == 1
+        assert result[0]["condition"] == "analgésie palier 3 (douleur sévère)"
+
+
+# --- detect_severity_markers ---
+
+class TestDetectSeverityMarkers:
+    def test_sejour_prolonge(self):
+        dossier = DossierMedical(sejour=Sejour(duree_sejour=20))
+        markers = detect_severity_markers(dossier)
+        assert any("séjour prolongé" in m for m in markers)
+
+    def test_sejour_gt7(self):
+        dossier = DossierMedical(sejour=Sejour(duree_sejour=10))
+        markers = detect_severity_markers(dossier)
+        assert any("séjour >7 jours" in m for m in markers)
+
+    def test_patient_tres_age(self):
+        dossier = DossierMedical(sejour=Sejour(age=85))
+        markers = detect_severity_markers(dossier)
+        assert any("très âgé" in m for m in markers)
+
+    def test_patient_age(self):
+        dossier = DossierMedical(sejour=Sejour(age=72))
+        markers = detect_severity_markers(dossier)
+        assert any("patient âgé" in m for m in markers)
+
+    def test_obesite_morbide(self):
+        dossier = DossierMedical(sejour=Sejour(imc=42.0))
+        markers = detect_severity_markers(dossier)
+        assert any("obésité morbide" in m for m in markers)
+
+    def test_complications(self):
+        dossier = DossierMedical(complications=["Fièvre", "Hématome"])
+        markers = detect_severity_markers(dossier)
+        assert any("2 complication(s)" in m for m in markers)
+
+    def test_no_markers(self):
+        dossier = DossierMedical(sejour=Sejour(age=45, duree_sejour=3))
+        markers = detect_severity_markers(dossier)
+        assert markers == []
+
+
+# --- build_enriched_context ---
+
+class TestBuildEnrichedContext:
+    def test_basic_context(self):
+        dossier = DossierMedical(
+            sejour=Sejour(sexe="M", age=65, duree_sejour=5),
+            biologie_cle=[
+                BiologieCle(test="CRP", valeur="150", anomalie=True),
+            ],
+            traitements_sortie=[
+                Traitement(medicament="Ceftriaxone 1g"),
+            ],
+        )
+        ctx = build_enriched_context(dossier)
+
+        assert ctx["sexe"] == "M"
+        assert ctx["age"] == 65
+        assert len(ctx["interpretations_bio"]) == 1
+        assert ctx["interpretations_bio"][0]["interpretation"] == "syndrome inflammatoire majeur"
+        assert len(ctx["conditions_traitements"]) == 1
+        assert ctx["conditions_traitements"][0]["condition"] == "antibiothérapie IV"
+
+    def test_no_abnormal_bio(self):
+        dossier = DossierMedical(
+            biologie_cle=[
+                BiologieCle(test="CRP", valeur="3", anomalie=False),
+            ],
+        )
+        ctx = build_enriched_context(dossier)
+        assert ctx["interpretations_bio"] == []
+
+
+# --- format_enriched_context ---
+
+class TestFormatEnrichedContext:
+    def test_with_interpretations(self):
+        ctx = {
+            "sexe": "F",
+            "age": 70,
+            "imc": None,
+            "duree_sejour": 10,
+            "antecedents": ["HTA", "Diabète"],
+            "biologie_cle": [("CRP", "180", True)],
+            "imagerie": [],
+            "complications": [],
+            "dp_texte": None,
+            "das_codes_existants": None,
+            "interpretations_bio": [
+                {"test": "CRP", "valeur": "180 mg/L", "interpretation": "syndrome inflammatoire majeur"},
+            ],
+            "conditions_traitements": [
+                {"medicament": "Insuline", "condition": "diabète insulino-traité"},
+            ],
+            "marqueurs_severite": ["séjour >7 jours (10 jours)"],
+        }
+        result = format_enriched_context(ctx)
+        assert "Patient : F, 70 ans" in result
+        assert "Durée séjour : 10 jours" in result
+        assert "HTA" in result
+        assert "CRP 180" in result
+        assert "INTERPRÉTATION CLINIQUE" in result
+        assert "syndrome inflammatoire majeur" in result
+        assert "diabète insulino-traité" in result
+        assert "séjour >7 jours" in result
+
+    def test_empty_context(self):
+        ctx = {
+            "sexe": None, "age": None, "imc": None,
+            "duree_sejour": None, "antecedents": [],
+            "biologie_cle": [], "imagerie": [],
+            "complications": [], "dp_texte": None,
+            "das_codes_existants": None,
+            "interpretations_bio": [],
+            "conditions_traitements": [],
+            "marqueurs_severite": [],
+        }
+        result = format_enriched_context(ctx)
+        assert result == "Non précisé"
+
+    def test_backward_compat_bio_format(self):
+        """Le format bio tuple (test, valeur, anomalie) doit rester compatible."""
+        ctx = {
+            "sexe": None, "age": None, "imc": None,
+            "duree_sejour": None, "antecedents": [],
+            "biologie_cle": [("CRP", "180", True)],
+            "imagerie": [],
+            "complications": [],
+            "dp_texte": None,
+            "das_codes_existants": None,
+            "interpretations_bio": [],
+            "conditions_traitements": [],
+            "marqueurs_severite": [],
+        }
+        result = format_enriched_context(ctx)
+        assert "CRP 180" in result
+        assert "(\u2191)" in result
--- a/tests/test_justification.py
+++ b/tests/test_justification.py
@@ -0,0 +1,245 @@
+"""Tests pour la validation batch des justifications (QC post-codage)."""
+
+from unittest.mock import patch, MagicMock
+
+import pytest
+
+from src.config import (
+    Diagnostic,
+    DossierMedical,
+    PreuveClinique,
+    Sejour,
+    BiologieCle,
+)
+
+
+class TestPreuveClinique:
+    def test_create(self):
+        p = PreuveClinique(type="biologie", element="CRP 180 mg/L", interpretation="syndrome inflammatoire majeur")
+        assert p.type == "biologie"
+        assert p.element == "CRP 180 mg/L"
+        assert p.interpretation == "syndrome inflammatoire majeur"
+
+    def test_diagnostic_with_preuves(self):
+        d = Diagnostic(
+            texte="Pancréatite aiguë",
+            cim10_suggestion="K85.9",
+            preuves_cliniques=[
+                PreuveClinique(type="biologie", element="Lipasémie 450 UI/L", interpretation="pancréatite biologique"),
+                PreuveClinique(type="imagerie", element="TDM: pancréatite stade D", interpretation="confirmation"),
+            ],
+        )
+        assert len(d.preuves_cliniques) == 2
+        assert d.preuves_cliniques[0].type == "biologie"
+
+    def test_diagnostic_default_empty_preuves(self):
+        d = Diagnostic(texte="Test")
+        assert d.preuves_cliniques == []
+
+    def test_serialization_round_trip(self):
+        d = Diagnostic(
+            texte="Test",
+            preuves_cliniques=[
+                PreuveClinique(type="clinique", element="fièvre 39°C", interpretation="syndrome infectieux"),
+            ],
+        )
+        data = d.model_dump()
+        assert data["preuves_cliniques"][0]["type"] == "clinique"
+        d2 = Diagnostic(**data)
+        assert d2.preuves_cliniques[0].element == "fièvre 39°C"
+
+
+class TestApplyLlmResultPreuves:
+    """Teste le stockage des preuves cliniques dans _apply_llm_result_diagnostic."""
+
+    def test_preuves_stored(self):
+        from src.medical.rag_search import _apply_llm_result_diagnostic
+
+        diag = Diagnostic(texte="Pneumopathie")
+        llm_result = {
+            "code": "J18.9",
+            "confidence": "high",
+            "justification": "Pneumopathie confirmée",
+            "preuves_cliniques": [
+                {"type": "biologie", "element": "CRP 120 mg/L", "interpretation": "syndrome inflammatoire"},
+                {"type": "imagerie", "element": "Radio thorax: opacité", "interpretation": "foyer pulmonaire"},
+            ],
+        }
+        _apply_llm_result_diagnostic(diag, llm_result)
+        assert len(diag.preuves_cliniques) == 2
+        assert diag.preuves_cliniques[0].type == "biologie"
+        assert diag.preuves_cliniques[1].element == "Radio thorax: opacité"
+
+    def test_preuves_empty_list(self):
+        from src.medical.rag_search import _apply_llm_result_diagnostic
+
+        diag = Diagnostic(texte="Test")
+        llm_result = {"code": "K85.9", "confidence": "medium", "preuves_cliniques": []}
+        _apply_llm_result_diagnostic(diag, llm_result)
+        assert diag.preuves_cliniques == []
+
+    def test_preuves_missing(self):
+        from src.medical.rag_search import _apply_llm_result_diagnostic
+
+        diag = Diagnostic(texte="Test")
+        llm_result = {"code": "K85.9", "confidence": "medium"}
+        _apply_llm_result_diagnostic(diag, llm_result)
+        assert diag.preuves_cliniques == []
+
+    def test_preuves_malformed_skipped(self):
+        from src.medical.rag_search import _apply_llm_result_diagnostic
+
+        diag = Diagnostic(texte="Test")
+        llm_result = {
+            "code": "K85.9",
+            "confidence": "high",
+            "preuves_cliniques": [
+                {"type": "bio"},  # manque 'element' → ignoré
+                {"type": "imagerie", "element": "TDM ok", "interpretation": "normal"},
+                "not a dict",  # ignoré
+            ],
+        }
+        _apply_llm_result_diagnostic(diag, llm_result)
+        assert len(diag.preuves_cliniques) == 1
+        assert diag.preuves_cliniques[0].element == "TDM ok"
+
+
+class TestValidateJustifications:
+    """Teste la fonction _validate_justifications."""
+
+    @patch("src.medical.ollama_client.call_ollama")
+    def test_confidence_adjusted(self, mock_ollama):
+        from src.medical.cim10_extractor import _validate_justifications
+
+        mock_ollama.return_value = {
+            "validations": [
+                {
+                    "numero": 1,
+                    "code": "K85.9",
+                    "verdict": "maintenir",
+                    "confidence_recommandee": "high",
+                    "commentaire": "bien justifié",
+                },
+                {
+                    "numero": 2,
+                    "code": "I10",
+                    "verdict": "maintenir",
+                    "confidence_recommandee": "low",
+                    "commentaire": "pas de preuve tensionnelle",
+                },
+            ],
+            "alertes_globales": [],
+        }
+
+        dossier = DossierMedical(
+            sejour=Sejour(sexe="M", age=60),
+            diagnostic_principal=Diagnostic(
+                texte="Pancréatite aiguë",
+                cim10_suggestion="K85.9",
+                cim10_confidence="medium",
+            ),
+            diagnostics_associes=[
+                Diagnostic(
+                    texte="HTA",
+                    cim10_suggestion="I10",
+                    cim10_confidence="high",
+                ),
+            ],
+        )
+
+        _validate_justifications(dossier)
+
+        # DP: medium → high
+        assert dossier.diagnostic_principal.cim10_confidence == "high"
+        # DAS: high → low
+        assert dossier.diagnostics_associes[0].cim10_confidence == "low"
+        # Alertes de confiance
+        assert any("QC:" in a and "I10" in a for a in dossier.alertes_codage)
+
+    @patch("src.medical.ollama_client.call_ollama")
+    def test_das_supprimer_alerte(self, mock_ollama):
+        from src.medical.cim10_extractor import _validate_justifications
+
+        mock_ollama.return_value = {
+            "validations": [
+                {
+                    "numero": 1,
+                    "code": "K85.9",
+                    "verdict": "maintenir",
+                    "confidence_recommandee": "high",
+                    "commentaire": "ok",
+                },
+                {
+                    "numero": 2,
+                    "code": "R10.4",
+                    "verdict": "supprimer",
+                    "confidence_recommandee": "low",
+                    "commentaire": "symptôme couvert par le DP",
+                },
+            ],
+            "alertes_globales": ["Vérifier la spécificité du DP"],
+        }
+
+        dossier = DossierMedical(
+            diagnostic_principal=Diagnostic(
+                texte="Pancréatite aiguë",
+                cim10_suggestion="K85.9",
+                cim10_confidence="high",
+            ),
+            diagnostics_associes=[
+                Diagnostic(
+                    texte="Douleur abdominale",
+                    cim10_suggestion="R10.4",
+                    cim10_confidence="medium",
+                ),
+            ],
+        )
+
+        _validate_justifications(dossier)
+
+        # Le DAS n'est pas supprimé automatiquement, mais une alerte est ajoutée
+        assert any("à reconsidérer" in a for a in dossier.alertes_codage)
+        assert any("Vérifier la spécificité" in a for a in dossier.alertes_codage)
+
+    @patch("src.medical.ollama_client.call_ollama")
+    def test_ollama_returns_none(self, mock_ollama):
+        from src.medical.cim10_extractor import _validate_justifications
+
+        mock_ollama.return_value = None
+        dossier = DossierMedical(
+            diagnostic_principal=Diagnostic(
+                texte="Test",
+                cim10_suggestion="K85.9",
+                cim10_confidence="high",
+            ),
+        )
+        _validate_justifications(dossier)
+        assert dossier.alertes_codage == []
+
+    def test_no_diags(self):
+        from src.medical.cim10_extractor import _validate_justifications
+
+        dossier = DossierMedical()
+        _validate_justifications(dossier)
+        assert dossier.alertes_codage == []
+
+    @patch("src.medical.ollama_client.call_ollama")
+    def test_invalid_validation_nums_skipped(self, mock_ollama):
+        from src.medical.cim10_extractor import _validate_justifications
+
+        mock_ollama.return_value = {
+            "validations": [
+                {"numero": 0, "code": "X", "verdict": "supprimer", "confidence_recommandee": "low", "commentaire": "oob"},
+                {"numero": 99, "code": "Y", "verdict": "supprimer", "confidence_recommandee": "low", "commentaire": "oob"},
+                {"numero": "abc", "code": "Z", "verdict": "supprimer", "confidence_recommandee": "low", "commentaire": "type"},
+            ],
+            "alertes_globales": [],
+        }
+
+        dossier = DossierMedical(
+            diagnostic_principal=Diagnostic(texte="T", cim10_suggestion="A00", cim10_confidence="high"),
+        )
+        _validate_justifications(dossier)
+        # Aucune modification, tous les numéros sont invalides
+        assert dossier.diagnostic_principal.cim10_confidence == "high"
+        assert dossier.alertes_codage == []