Do not read an allocation ratio as a sample size

data/compounds/retatrutide.json

@@ -1,3 +1,3 @@
 {
   "type": "compound",
   "preferred_name": "Retatrutide",
@@ -489,7 +489,7 @@
         }
       },
       {
-        "value": "Randomized controlled trial, humans, n=534, 2025: Percent reduction from baseline in total fat mass was 4·9% (SE 1·4%) with retatrutide 0·5 mg, 15·2% (3·2%) with retatrutide 4 mg (pooled), 26·1% (2·5%) with retatrutide 8 mg (pooled), 23·2% (3·0%) with retatrutide 12 mg, 2·6% (1·6%) with dulaglutide, and 4·5% (1·2%) with placebo.",
+        "value": "Randomized controlled trial, humans, n=281, 2025: Percent reduction from baseline in total fat mass was 4·9% (SE 1·4%) with retatrutide 0·5 mg, 15·2% (3·2%) with retatrutide 4 mg (pooled), 26·1% (2·5%) with retatrutide 8 mg (pooled), 23·2% (3·0%) with retatrutide 12 mg, 2·6% (1·6%) with dulaglutide, and 4·5% (1·2%) with placebo.",
         "evidence_label": "human-clinical-trial",
         "source_ids": [
           "pmid-40609566"
@@ -500,7 +500,7 @@
           "study": "Coskun 2025",
           "year": "2025",
           "design": "Randomized controlled trial",
-          "n": 534,
+          "n": 281,
           "species": "Humans",
           "dose": "0·5 mg; 4 mg",
           "route": "subcutaneous",
@@ -649,7 +649,7 @@
         }
       },
       {
-        "value": "Human study, humans, n=367, 2026: Background and hypothesis Retatrutide is an agonist of the glucose-dependent insulinotropic polypeptide, glucagon-like peptide-1 and glucagon receptors that reduced weight and hemoglobin A1c (HbA1c) in individuals with obesity and type 2 diabetes (T2D).",
+        "value": "Human study, humans, n=146, 2026: Background and hypothesis Retatrutide is an agonist of the glucose-dependent insulinotropic polypeptide, glucagon-like peptide-1 and glucagon receptors that reduced weight and hemoglobin A1c (HbA1c) in individuals with obesity and type 2 diabetes (T2D).",
         "evidence_label": "observational-human",
         "source_ids": [
           "pmid-41160422"
@@ -660,7 +660,7 @@
           "study": "Heerspink 2026",
           "year": "2026",
           "design": "Human study",
-          "n": 367,
+          "n": 146,
           "species": "Humans",
           "dose": "12 mg",
           "route": "not stated in abstract",
@@ -1204,6 +1204,10 @@
     {
       "date": "2026-09-29",
       "change": "Reviewed by Adam Mirando, PharmD, on 2026-09-29."
+    },
+    {
+      "date": "2026-09-29",
+      "change": "Corrected two more sample sizes that had stored the screening count: the body-composition substudy from 534 to 281 enrolled, and the cardiovascular trial from 367 screened to 146 randomised."
     }
   ],
   "guide": [

scripts/draft_claims.py

@@ -90,10 +90,14 @@
     r"\b" + NUM + r"\s+(?:\([^)]{0,40}\)\s*)?(?:\w+\s+){0,3}?"
     r"(?:were|was)\s+(?:randomly\s+assigned|randomi[sz]ed|allocated|enrolled)\b", re.IGNORECASE)
 SCREENED_AFTER = re.compile(r"^[^.]{0,40}?\b(?:were|was)\s+screened\b", re.IGNORECASE)
+# "randomized 1:1", "randomised 1:1:1:1:1" is an allocation ratio, not a count.
+# The bare verb pattern reads it as a sample of 1, which is why it sits below
+# the noun pattern and refuses a number followed by a colon.
+RATIO_AFTER = re.compile(r"^\s*:")
 N_RE = [
     ASSIGNED,
-    re.compile(r"\b(?:enrolled|randomi[sz]ed|included|analy[sz]ed|recruited)\s+" + NUM + r"\b", re.IGNORECASE),
     re.compile(r"\b" + NUM + r"\s+(?:patients|participants|subjects|volunteers|adults|individuals|people|men|women|children|infants|rats|mice|rabbits|dogs|pigs|animals)\b", re.IGNORECASE),
+    re.compile(r"\b(?:enrolled|randomi[sz]ed|included|analy[sz]ed|recruited)\s+" + NUM + r"\b", re.IGNORECASE),
     re.compile(r"\b[nN]\s?=\s?" + NUM + r"\b"),
 ]
 DESIGN_FROM_TITLE = [
@@ -227,9 +231,15 @@ def sample_size(abstract: str) -> int | None:
             n = int(re.sub(r"[ ,\u00a0\u202f]", "", m.group(1)))
             if not 1 <= n <= 2_000_000:
                 continue
+            # Both look at what follows the number itself, not the whole match:
+            # ASSIGNED ends after "randomly assigned", and the colon in
+            # "assigned: 134 to retatrutide 4 mg" is not an allocation ratio.
+            after = abstract[m.end(1):]
             # "930 participants were screened and 537 were randomly assigned":
             # the screening count is not the trial's size.
-            if rx is not ASSIGNED and SCREENED_AFTER.match(abstract[m.end():]):
+            if rx is not ASSIGNED and SCREENED_AFTER.match(after):
+                continue
+            if RATIO_AFTER.match(after):
                 continue
             return n
     return None

scripts/test_drafter.py

@@ -54,6 +54,29 @@ def test_screened_then_randomised_short_form(self):
     def test_plain_enrolment_still_wins(self):
         self.assertEqual(dc.sample_size("We enrolled 338 adults, 51.8% of whom were men."), 338)
 
+    def test_allocation_ratio_is_not_a_count(self):
+        # "randomized 1:1" put n=1 on kisspeptin, survodutide, semax, tesamorelin
+        # and thymosin-alpha-1 rows when the noun pattern lost precedence.
+        self.assertEqual(dc.sample_size(
+            "Of these, 194 patients have been randomized 1:1 to either Group A or Group B."), 194)
+        self.assertEqual(dc.sample_size(
+            "Totally 387 people (aged 18-75 years) were randomized 1:1:1:1:1 to once-weekly "
+            "subcutaneous survodutide or placebo."), 387)
+
+    def test_colon_after_assignment_is_not_a_ratio(self):
+        # The colon in "assigned: 134 to retatrutide 4 mg" introduces the arms,
+        # so it must not disqualify the number that precedes the verb.
+        self.assertEqual(dc.sample_size(
+            "930 participants were screened and 537 (296 [55%] female) were randomly assigned: "
+            "134 to retatrutide 4 mg, 133 to retatrutide 9 mg"), 537)
+
+    def test_screened_participants_phrasing(self):
+        self.assertEqual(dc.sample_size(
+            "Of 367 participants screened, 146 were randomized to study interventions."), 146)
+        self.assertEqual(dc.sample_size(
+            "534 participants were screened for inclusion into the main study; 253 were excluded "
+            "and 281 participants were enrolled and randomly assigned to the main study."), 281)
+
     def test_enrolment_beats_subgroup_n(self):
         text = "The renal impairment study included 33 participants (normal function, n = 14; mild impairment, n = 7)."
         self.assertEqual(dc.sample_size(text), 33)