Fix T1D DKA timing

docs/api/tasks/pyhealth.tasks.dka.rst

@@ -6,3 +6,7 @@ pyhealth.tasks.dka
     :undoc-members:
     :show-inheritance:
 
+.. autoclass:: pyhealth.tasks.dka.T1DDKAPredictionMIMIC4
+    :members:
+    :undoc-members:
+    :show-inheritance:

examples/clinical_tasks/t1dka_mimic4.py

@@ -9,7 +9,7 @@
 
 Target Population:
     - Patients with Type 1 Diabetes Mellitus (T1DM) ONLY
-    - Predicts DKA occurrence within 90 days of T1DM diagnosis
+    - Predicts DKA from the T1DM diagnosis through the next 90 days
     - Smaller, focused patient cohort
 """
 
@@ -179,4 +179,3 @@ def main():
 if __name__ == "__main__":
     main()
 
-

pyhealth/tasks/dka.py

@@ -282,6 +282,7 @@ class T1DDKAPredictionMIMIC4(BaseTask):
     Target Population:
         - Patients with Type 1 Diabetes (ICD-9 or ICD-10 codes)
         - Excludes patients without any T1DM diagnosis codes
+        - Excludes DKA events recorded before a separate T1DM diagnosis
 
     Label Definition:
         - Positive (1): Patient has DKA code within 90 days of T1DM diagnosis
@@ -459,20 +460,34 @@ def __call__(self, patient: Any) -> List[Dict[str, Any]]:
 
         has_t1dm = False
         t1dm_times: List[datetime] = []
+        t1dm_diagnosis_times: list[datetime] = []
+        dka_times: list[datetime] = []
 
         for diag in all_diagnoses:
             code = getattr(diag, "icd_code", None)
             version = getattr(diag, "icd_version", None)
+            diag_time = getattr(diag, "timestamp", None)
+            is_dka = self._is_dka_code(code, version)
             if self._is_t1dm_code(code, version):
                 has_t1dm = True
-                diag_time = getattr(diag, "timestamp", None)
                 if diag_time:
                     t1dm_times.append(diag_time)
+                    if not is_dka:
+                        t1dm_diagnosis_times.append(diag_time)
+            if is_dka and diag_time:
+                dka_times.append(diag_time)
 
         # Skip patients without T1DM diagnosis (early exit before sorting)
         if not has_t1dm:
             return []
 
+        if (
+            t1dm_diagnosis_times
+            and dka_times
+            and min(dka_times) < min(t1dm_diagnosis_times)
+        ):
+            return []
+
         # Get admissions and sort by timestamp
         admissions = patient.get_events(event_type="admissions")
         if not admissions:
@@ -609,9 +624,11 @@ def __call__(self, patient: Any) -> List[Dict[str, Any]]:
         # Determine label based on temporal relationship
         has_dka_within_window = False
         if has_dka and t1dm_times and dka_time:
-            for t1dm_time in t1dm_times:
-                delta = abs((dka_time - t1dm_time).days)
-                if delta <= self.dka_window_days:
+            diagnosis_times = t1dm_diagnosis_times or t1dm_times
+            for t1dm_time in diagnosis_times:
+                if t1dm_time <= dka_time <= t1dm_time + timedelta(
+                    days=self.dka_window_days
+                ):
                     has_dka_within_window = True
                     break
         elif has_dka and not t1dm_times:

tests/core/test_t1d_dka_window.py

@@ -49,7 +49,7 @@ def get_events(self, event_type: str, filters=None, return_df: bool = False):
 
 def _build_patient(t0: datetime, admission_days: list[int], diag_specs: list[tuple[str, int | str, int, str]]) -> DummyPatient:
     admissions = [DummyAdmission(f"a{i+1}", t0 + timedelta(days=day)) for i, day in enumerate(admission_days)]
-    base_diag = DummyDiagnosis("E10.10", 10, t0, "t1dm")
+    base_diag = DummyDiagnosis("E10.9", 10, t0, "t1dm")
     diagnoses = [base_diag] + [
         DummyDiagnosis(code, version, t0 + timedelta(days=day), hadm_id)
         for code, version, day, hadm_id in diag_specs
@@ -101,6 +101,60 @@ def test_sequences_do_not_exceed_window(self):
         self._assert_within_window(sample_long, expected_visits=2)
         self.assertEqual(sample_long["label"], 0)
 
+    def test_dka_before_t1dm_diagnosis_is_excluded(self):
+        patient = _build_patient(
+            self.t0,
+            admission_days=[-60, -30],
+            diag_specs=[
+                ("I10", 10, -60, "a1"),
+                ("E1011", 10, -30, "a2"),
+            ],
+        )
+
+        self.assertEqual(self.task(patient), [])
+
+    def test_dka_at_t1dm_diagnosis_is_positive(self):
+        patient = _build_patient(
+            self.t0,
+            admission_days=[-10, 0],
+            diag_specs=[
+                ("I10", 10, -10, "a1"),
+                ("E1011", 10, 0, "a2"),
+            ],
+        )
+
+        samples = self.task(patient)
+        self.assertEqual(len(samples), 1)
+        self.assertEqual(samples[0]["label"], 1)
+
+    def test_dka_inside_window_is_positive(self):
+        patient = _build_patient(
+            self.t0,
+            admission_days=[10, 30],
+            diag_specs=[
+                ("I10", 10, 10, "a1"),
+                ("E1011", 10, 30, "a2"),
+            ],
+        )
+
+        samples = self.task(patient)
+        self.assertEqual(len(samples), 1)
+        self.assertEqual(samples[0]["label"], 1)
+
+    def test_dka_at_end_of_window_is_positive(self):
+        patient = _build_patient(
+            self.t0,
+            admission_days=[30, 90],
+            diag_specs=[
+                ("I10", 10, 30, "a1"),
+                ("E1011", 10, 90, "a2"),
+            ],
+        )
+
+        samples = self.task(patient)
+        self.assertEqual(len(samples), 1)
+        self.assertEqual(samples[0]["label"], 1)
+
 
 if __name__ == "__main__":
     unittest.main()