ictrp-mcp-server 0.1.1 → 0.1.2

package/CHANGELOG.md

@@ -3,6 +3,31 @@
 Notable changes to the npm package `ictrp-mcp-server`. The Python package
 `ictrp-mcp-service` is versioned separately in `pyproject.toml`.
 
+## 0.1.2
+
+**Default search view is now the clinician-triage view.** `ictrp_search` used to
+return the 10 most common columns; it now returns the columns a clinician needs
+to triage a target: title and scientific title, sponsor and secondary sponsor,
+contact (PI) name/affiliation/email, recruitment status, phase, registration
+date, condition, countries, age bounds and gender, target size, and a
+`line_of_therapy_hint`. A new derived field, `line_of_therapy_hint`, is a
+non-authoritative extraction of treatment line from the free-text
+inclusion/exclusion criteria and the scientific title. It returns
+`first-line`/`second-line`/`third-line`/`fourth-line`/`later-line`, or `null`
+when nothing is stated -- `null` is documented as "not stated", never as
+"first-line", because the export is incomplete and the column does not exist.
+
+**Line-hint fix** — the hint missed trials whose line is stated only as a number
+or only in the scientific title (e.g. "Received >=2 Prior Lines of Therapy"). It
+now scans the scientific title and recognizes numeric line statements; a minimum
+of N lines resolves to `first-line` only when N<=1, otherwise `later-line`.
+
+Documented structural limits surfaced by the wider view: ICTRP has no "PI" and
+no "participating-hospital list" column -- sponsor/PI come from
+`primary_sponsor` + `contact_*`, and "participating hospitals" degrades to
+`countries` + `contact_affiliation`. CT.gov `exclusion_criteria` is frequently
+blank upstream, which is a source gap, not a parse failure.
+
 ## 0.1.1
 
 **Retry on transient stalls** — the search POST intermittently stalls past a

package/dist/index.js

@@ -21,7 +21,7 @@ import { getEnvReport } from "./runtime/env-probe.js";
 import { callSidecar, pingSidecar, SEARCH_TIMEOUT_MS, toErrorText, } from "./runtime/sidecar-client.js";
 import { ensureSidecar, installShutdownHooks, sidecarLogTail } from "./runtime/supervisor.js";
 const SERVER_NAME = "ictrp-mcp-server";
-const SERVER_VERSION = "0.1.1";
+const SERVER_VERSION = "0.1.2";
 /**
  * Attached to every payload that returns rows.
  *

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ictrp-mcp-server",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "MCP server for the WHO International Clinical Trials Registry Platform (ICTRP). Plain HTTP, no browser. Every response is explicit about the export's known incompleteness.",
   "type": "module",
   "main": "dist/index.js",

package/sidecar/vendor/ictrp_mcp/data/normalize.py

@@ -323,6 +323,18 @@ def to_trial(row: list[str], header: tuple[str, ...]) -> dict[str, Any]:
     trial["inclusion_age_min"] = _as_int(age_min)
     trial["inclusion_age_max"] = _as_int(age_max)
 
+    # ICTRP has no structured "line of therapy" column. Treatment line is stated
+    # only inside the free-text inclusion/exclusion criteria (e.g. "first-line",
+    # "previously treated", "一线") and, frequently, the scientific title
+    # ("...Received >=2 Prior Lines of Therapy"). Derive a non-authoritative hint
+    # from that text so a caller can surface it without re-parsing. It is a hint,
+    # never a fact: absence here does NOT mean the trial is treatment-naive.
+    trial["line_of_therapy_hint"] = derive_line_of_therapy_hint(
+        raw.get("inclusion_criteria"),
+        raw.get("exclusion_criteria"),
+        raw.get("scientific_title"),
+    )
+
     # `results yes no` is a flag, not content, so it is excluded from the test.
     # Measured: the substantive `results *` columns are ~0.0% populated for
     # ChiCTR records, so this is a property of the registry rather than of an
@@ -346,3 +358,66 @@ def _as_int(value: str | None) -> int | None:
         return None
     m = re.search(r"-?\d+", text)
     return int(m.group(0)) if m else None
+
+
+#: Treatment-line vocabulary, as it actually appears in live inclusion/exclusion
+#: text. Both English and Chinese markers are matched; "refractory"/"relapsed"/
+#: "previously treated" indicate a later line without naming a number.
+_LINE_PATTERNS: tuple[tuple[str, str], ...] = (
+    (r"first[-\s]?line|1st[-\s]?line|一线|treatment[-\s]?naive|previously untreated", "first-line"),
+    (r"second[-\s]?line|2nd[-\s]?line|二线", "second-line"),
+    (r"third[-\s]?line|3rd[-\s]?line|三线", "third-line"),
+    (r"fourth[-\s]?line|4th[-\s]?line|四线", "fourth-line"),
+    (r"relapsed|refractory|previously treated|prior (therapy|treatment)|已接受.*治疗|经治", "later-line"),
+)
+
+#: Numeric "line" statements, e.g. ">=2 Prior Lines of Therapy" or "2-line",
+#: where a word may sit between the number and "line" ("2 prior lines"). The
+#: captured number is the *minimum* line; ">=2" means second-line or later, never
+#: first-line, so it resolves to `later-line` rather than a specific line.
+_LINE_NUMBER = re.compile(r"(\d+)\s*(?:[-\w.,]+\s)*?(?:line|lines|线)")
+
+_WORD_NUMBERS = {
+    "one": 1, "two": 2, "three": 3, "four": 4, "five": 5,
+    "一": 1, "二": 2, "三": 3, "四": 4, "五": 5,
+}
+_WORD_NUMBER = re.compile(r"(one|two|three|four|five|一|二|三|四|五)\s*(?:line|lines|线)")
+
+
+def derive_line_of_therapy_hint(
+    inclusion: str | None, exclusion: str | None, scientific_title: str | None = None
+) -> str | None:
+    """Best-effort treatment-line label from free-text criteria.
+
+    Returns the earliest explicit line if one is named (first > second > third >
+    fourth), else "later-line" when a numeric "N lines" (N>=2) or relapse/
+    refractory language is present, else `None`. `None` is deliberate: it means
+    "not stated in the criteria we hold", which must not be read as "first-line"
+    -- the export is incomplete and the column does not exist.
+
+    The scientific title is included because the line is often stated only there
+    (e.g. "...Received >=2 Prior Lines of Therapy"), not in the criteria body.
+    """
+    text = f"{inclusion or ''} {exclusion or ''} {scientific_title or ''}"
+    if not text.strip():
+        return None
+    from .query import _coerce  # local import avoids a cycle at module load
+
+    lowered = _coerce(text)
+
+    # Explicit worded lines take precedence -- they name a specific line.
+    for pattern, label in _LINE_PATTERNS:
+        if re.search(pattern, lowered):
+            return label
+
+    # Numeric "N line(s)": a minimum of N, so only the first-line case is exact;
+    # anything >=2 is "later-line", never a specific later number.
+    for match in _LINE_NUMBER.finditer(lowered):
+        if int(match.group(1)) <= 1:
+            return "first-line"
+        return "later-line"
+    for match in _WORD_NUMBER.finditer(lowered):
+        if _WORD_NUMBERS[match.group(1)] <= 1:
+            return "first-line"
+        return "later-line"
+    return None

package/sidecar/vendor/ictrp_mcp/tools.py

@@ -35,16 +35,33 @@ from .provenance import INCOMPLETENESS_NOTICE, Provenance, derive_set_provenance
 
 #: Fields shown by default when a caller does not ask for specific ones. Chosen to
 #: be the high-coverage, decision-relevant columns rather than all 58.
+#: The default view. Deliberately NOT just the 10 most common columns: it is the
+#: view a clinician actually needs to triage a target -- name, sponsor/PI, where
+#: it runs, eligibility shape, and a treatment-line hint. Sponsor/PI come from the
+#: `primary_sponsor` + `contact_*` columns (ICTRP has no dedicated "PI" or
+#: "participating-hospital" field; hospitals exist only as `countries` and the
+#: contact's affiliation), so those are surfaced here. `line_of_therapy_hint` is a
+#: non-authoritative extraction from the free-text criteria -- see normalize.py.
 DEFAULT_FIELDS: tuple[str, ...] = (
     "trial_id",
     "source_register",
     "public_title",
+    "scientific_title",
+    "primary_sponsor",
+    "secondary_sponsor",
+    "contact_firstname",
+    "contact_lastname",
+    "contact_affiliation",
     "recruitment_status",
     "phase_code",
     "registration_date",
     "condition",
     "countries",
+    "inclusion_age_min",
+    "inclusion_age_max",
+    "inclusion_gender",
     "target_size_total",
+    "line_of_therapy_hint",
     "last_refreshed_display",
 )