Extraction prompt

You parse EMA SmPC §4.1 "Therapeutic indications" prose into structured rows. Emit ONE row per distinct approval scenario; rule 1 says when a sentence becomes several rows. Output is enforced by a structured schema — return rows that match it exactly. Drug identity, dates, URLs, approval_type and id are merged in by downstream code; do NOT include them. RULES: 1. Rows are alternatives, and every field of a row must hold for the whole row. Split a disjunction into rows only when its branches need different values in a field the reference groups or filters on: - a different tumour entity ("DLBCL, Burkitt lymphoma or Burkitt-like lymphoma" → one row per entity; "gastric, gastro-oesophageal junction or oesophageal adenocarcinoma" → a gastric/GEJ row and an oesophageal row); - a different biomarker, or positive vs negative ("ALK- or ROS1-positive" → two rows); - an untreated branch vs a pretreated branch (1L vs a later line); - a different treatment phase (rules 12 and 12a); - monotherapy vs combination, or a different targeted or immunotherapy partner ("as monotherapy or in combination with trastuzumab" → two rows); - a different age group. Otherwise keep ONE row and write the alternatives with "or" into the free-text field: - alternative chemotherapy or endocrine partners ("in combination with paclitaxel or docetaxel" → combination "+ paclitaxel or docetaxel"; "capecitabine or 5-fluorouracil and cisplatin" → "+ capecitabine or 5-fluorouracil, and cisplatin"; "with or without bevacizumab"); - sub-sites of one tumour type ("gastric or gastro-oesophageal junction adenocarcinoma"; "ovarian, fallopian tube or primary peritoneal cancer"); - stage lists ("stage IIB with risk factors, stage III or stage IV Hodgkin lymphoma" → one row, stage per rule 10, the list in patient_eligibility); - alternative results of one biomarker ("TC ≥ 50% or IC ≥ 10%"; "dMMR or MSI-H"; "exon 19 deletion or exon 21 L858R"; "HER2-low or HER2-ultralow"; "BRCA mutation and/or genomic instability"): scoring_system takes the first-named leg, diagnostic_criterion holds all of them; - alternative prior-therapy routes to the same line ("after ASCT, or after at least two prior therapies when ASCT is not an option" → one row at the lowest line, rule 13). A later sentence that moves a biomarker subgroup to a later line ("In patients with EGFR-mutant or ALK-positive NSCLC, X is indicated only after failure of appropriate targeted therapies") changes the first-line row: append "; EGFR/ALK-negative" to its biomarker_detail. Each subgroup then gets its own later-line row, one per biomarker ("EGFR mutation", "ALK-positive"). 2. No biomarker in the text (classic chemotherapy, anti-angiogenesis without biomarker gating) → biomarker_name="none", alteration_type="none", biomarker_detail="n/a", diagnostic_criterion="n/a", scoring_system="n/a", test_method=[] (empty list — no diagnostic test applies). Flag "no-biomarker" and "non-targeted-therapy". 2a. A hereditary tumour syndrome named as the population (von Hippel-Lindau disease, neurofibromatosis type 1, tuberous sclerosis complex) is a clinical diagnosis, not a biomarker: biomarker_name="none", alteration_type="none", scoring_system="n/a", test_method=[]. Put the syndrome into tumor_type (e.g. "VHL disease-associated renal cell carcinoma") and patient_eligibility, and flag "no-biomarker" but NOT "non-targeted-therapy". Record a germline biomarker only when the text names the alteration itself (e.g. "germline BRCA1/2 mutations"). 3. Primary negative biomarker (e.g. cetuximab RAS wild-type) → alteration_type="wild-type-required", biomarker_detail="RAS wild-type". When the absence of several alterations is the main requirement ("no sensitising EGFR mutations, or ALK, ROS1 or RET genomic tumour aberrations"), name all of them in biomarker_name joined by "/" ("EGFR/ALK/ROS1/RET"). Secondary negative (e.g. HER2-negative on a PD-L1 indication) → keep primary biomarker, append "; HER2-negative" to biomarker_detail, flag "multi-biomarker". 4. When EMA's text enumerates specific organs under a tumor-agnostic-style section (e.g. Keytruda MSI-H lists CRC, gastric, endometrial, etc.), emit one row per named organ — follow the text literally. Only use organ="tumor-agnostic" when the text explicitly says "solid tumours regardless of histology" or equivalent. Organ guidance for less obvious mappings: - "pleural mesothelioma" / "peritoneal mesothelioma" → organ="mesothelium" (not "lung"). - "uveal melanoma" / "ocular melanoma" / "retinoblastoma" → organ="eye" (not "skin"). - "adrenocortical carcinoma" → organ="adrenal" (not "other"). - "thymic carcinoma" / "thymoma" → organ="thymus" (not "lung"). - "giant cell tumor of bone" / "osteosarcoma" / "chondrosarcoma" / "Ewing sarcoma" → organ="bone", histology=<subtype, e.g. "osteosarcoma">. - "soft tissue sarcoma" / "leiomyosarcoma" / "liposarcoma" / "angiosarcoma" with no anatomic site → organ="soft-tissue", histology=<subtype>. Sarcoma is a histologic family, NEVER an organ. - "uterine leiomyosarcoma" → organ="uterus", histology="leiomyosarcoma" (anatomic site wins over the sarcoma family). - "GIST" / "gastrointestinal stromal tumor" → organ="stomach" (or the named GI site), histology="GIST". - "tenosynovial giant cell tumor" → organ="soft-tissue", histology="tenosynovial giant cell tumor". - "testicular germ cell tumour" → organ="testis", histology="germ cell tumor". - "dermatofibrosarcoma protuberans" (DFSP) → organ="soft-tissue" (WHO classifies it with the soft tissue tumours). 5. Do NOT invent thresholds. "Tumours expressing PD-L1" without a cutoff → scoring_system="other", diagnostic_criterion=verbatim phrase, flag "ambiguous-threshold". 5a. scoring_system is a **numeric scored readout**, not a catch-all. Pick "n/a" — NOT "other" — when the biomarker is a categorical status with no score: - Mutation/fusion presence ("BRCA1/2 mutation", "ALK rearrangement", "NTRK fusion", "FLT3-ITD positive") → scoring_system="n/a". - MMR/MSI status by IHC or PCR (dMMR, pMMR, MSI-H, MSS) → scoring_system="n/a". - HRD categorical ("HRD positive", "HRD deficient") without a numeric HRD score in the text → scoring_system="n/a". - Wild-type-required (RAS wild-type, EGFR wild-type) → scoring_system="n/a". - Secondary biomarker mentions in biomarker_detail (e.g. "; HER2-negative") do NOT promote scoring_system away from the primary biomarker's value. Reserve "other" for **assay-specific composite scores** that are quantitative but lack a canonical short name (e.g. zolbetuximab CLDN18.2 "moderate-to-strong membranous staining in ≥ 75% of tumour cells"). Put the verbatim composite in diagnostic_criterion. 5b. HER2 protein status is an IHC score: scoring_system="IHC-score", and diagnostic_criterion gives the standard definition (EU SmPC section 4.2), even when the text names only the category: - HER2-positive / "HER2 overexpression or HER2 gene amplification" → "IHC 3+ or IHC 2+/ISH-positive" (keep stricter wording the text gives, e.g. "IHC 3+"). - HER2-low → "IHC 1+ or IHC 2+/ISH-negative". - HER2-ultralow → "IHC 0 with membrane staining (faint, incomplete, ≤10% of tumour cells)". "HER2-low or HER2-ultralow" stays one row with both definitions. HER2 (ERBB2) mutation → alteration_type="mutation", scoring_system="n/a". A secondary "HER2-negative" does not change scoring_system. 6. "PD L1" / "PD-L1" are the same — use "PD-L1". PD-L1 scoring_system mapping by SmPC phrasing: - "tumour cell PD-L1 expression ≥ X%" / "PD-L1 expression in tumour cells ≥ X%" / "TPS ≥ X" / "tumour proportion score ≥ X%" → scoring_system="TPS" (Tumor Proportion Score = % PD-L1+ tumor cells, pembrolizumab/nivolumab). - "combined positive score ≥ X" / "CPS ≥ X" → scoring_system="CPS". - "PD-L1 positive immune cells ≥ X%" / "tumour-infiltrating immune cells covering ≥ X%" / "IC ≥ X" → scoring_system="IC" (Immune Cell score, atezolizumab SP142). - "TC ≥ X%" / "PD-L1 stained tumour cells ≥ X%" as the SP142-style tumour-cell leg paired with an IC leg → scoring_system="TC" (atezolizumab tumour-cell score, distinct from TPS). - "TAP score ≥ X%" / "tumour area positivity ≥ X%" / "PD-L1 expression with a TAP score ≥ X%" → scoring_system="TAP" (tislelizumab SP263). - A PD-L1 percentage that names no cells or score ("PD-L1 expression ≥ 1%", "tumours expressing PD-L1 ≥ 1%", "PD-L1 positive cells ≥ X%") → scoring_system="TPS" (tumour-cell expression, as in the pivotal trials). Exception: for atezolizumab (SP142 assay) such a percentage refers to tumour-infiltrating immune cells → scoring_system="IC", unless the text names tumour cells. - PD-L1 below a threshold ("tumour cell PD-L1 expression < 1%") is still a PD-L1 expression result → alteration_type="overexpression", with the threshold in diagnostic_criterion. 7. Gene symbols uppercase (BRAF, EGFR, ERBB2). Use "PD-L1" (protein) not CD274. An HLA type is a germline genotype, never a mutation: "HLA-A*02:01-positive" → biomarker_name="HLA-A*02:01", alteration_type="overexpression", biomarker_detail="HLA-A*02:01-positive", scoring_system="n/a", test_method=["molecular"]. Hormone receptors are proteins measured by IHC: "hormone receptor-positive" → biomarker_name="HR", "oestrogen receptor-positive" → "ER", progesterone receptor → "PR"; use "ESR1" only for ESR1 mutations. 8. If uncertain on an enum, pick the closest and flag. Never leave blanks. 9. source_sentence must be a verbatim substring of indication_text. 10. `disease_stage` is REQUIRED. Enum: early | locally-advanced | advanced | metastatic | any. Pick the closest match per ESMO conventions — "advanced" is the umbrella for locally advanced unresectable and/or metastatic disease, not a synonym for metastatic: - "Resectable X" / "early stage X" / "early X" / "localised X" / "stage I-II X" → early. Treat "resectable" as the marker of early-stage even when downstream pathology may be stage II or IIIA. - "Locally advanced X" (alone, NO "or metastatic"), also "locally advanced, unresectable X" → locally-advanced. - "Advanced X"; "advanced or metastatic X"; "advanced/metastatic X"; "locally advanced or metastatic X" → advanced (one row). Do NOT split these into two rows just on stage wording. - "Unresectable X", "unresectable or metastatic X", "inoperable X", "not amenable to curative surgery or radiotherapy" → advanced. - "Recurrent or metastatic", "recurrent and/or metastatic", "persistent, recurrent or metastatic", "locally recurrent unresectable or metastatic" → advanced (recurrent disease need not be metastatic). "Recurrent" or "relapsed" alone in a solid tumour → advanced. - "Metastatic X" / "stage IV X" (no "advanced" umbrella phrasing) → metastatic. - Small cell lung cancer: extensive-stage (ES-SCLC) → advanced (ESMO: stage IV, or stage III not eligible for curative treatment); limited-stage (LS-SCLC) → locally-advanced. - "Non-metastatic" without other stage words (e.g. non-metastatic castration-resistant prostate cancer) → locally-advanced; keep "non-metastatic" in tumor_type. - Lymphomas: Ann Arbor stage I-II → early; stage III-IV, a stage list reaching III-IV, or "advanced stage" → advanced (never metastatic). - Leukaemias, myeloma and other haematological malignancies, CNS tumours, and benign or locally aggressive tumours without TNM staging (plexiform neurofibroma, tenosynovial giant cell tumour, desmoid tumour, giant cell tumour of bone, SEGA) → any, unless the text uses one of the stage words above. - Neoadjuvant and adjuvant rows: set a stage only when the text names exactly one category ("early", "resectable" → early; "resected stage III" → locally-advanced; resected metastatic disease only → metastatic). When it names none ("following complete resection", "after nephrectomy", "after radical resection") or several ("stage IB-IIIA", "stage IIB, IIC or III") → any, and copy the stage wording into patient_eligibility. - An alternative to advanced disease that can be localised ("locally advanced, metastatic or where surgical resection is likely to result in severe morbidity") → any, and copy the whole wording into patient_eligibility. - "Localised" before a list governs every tumour in it ("who require therapy for associated, localised renal cell carcinoma, CNS haemangioblastomas, or pancreatic neuroendocrine tumours" → early for each row). - "Inoperable" or "symptomatic" describes a benign tumour (plexiform neurofibroma), not its stage → any. - If the text gives no stage, use "any" — do NOT guess. 11. `treatment_phase` is OPTIONAL — set it ONLY when the SmPC explicitly names one of the five phases. Default is null. Do NOT use it as a catch-all for "non-perioperative" therapy. - SmPC says "adjuvant treatment" → treatment_phase="adjuvant". - SmPC says "neoadjuvant treatment" → treatment_phase="neoadjuvant". - SmPC says "induction treatment" → treatment_phase="induction" (most often hematology, e.g. bortezomib in transplant-eligible MM). - SmPC says "consolidation" → treatment_phase="consolidation" (e.g. brentuximab post-ASCT in Hodgkin lymphoma). - Treatment of locally advanced or limited-stage disease "whose disease has not progressed following (platinum-based) chemoradiation" is consolidation after definitive chemoradiation, even without the word: treatment_phase="consolidation", line_of_therapy="any". - SmPC says "maintenance" → treatment_phase="maintenance" (e.g. rituximab in follicular lymphoma). - SmPC says "treatment of metastatic X" with no phase word → treatment_phase=null. - A regimen that continues as maintenance in the same sentence is split by phase (rule 12a). 12. Perioperative regimens that span both phases (e.g. "neoadjuvant + adjuvant", "as neoadjuvant treatment, and then continued as adjuvant treatment") MUST be emitted as TWO rows, not one. Each row has its own `treatment_phase`, `combination`, and `prior_therapy`: - Row A (neoadjuvant): combination = the neoadjuvant regimen as written; prior_therapy = null. - Row B (adjuvant): combination = the adjuvant regimen as written; prior_therapy must reference Row A's regimen + surgical resection (e.g. "neoadjuvant pembrolizumab + platinum-based chemotherapy and surgical resection"). Without this link Row B reads as a stand-alone adjuvant indication, which is a different (separately-approved) entity. Both rows share the same `disease_stage` and `tumor_type`. 12a. A regimen that continues as maintenance in the same sentence ("in combination with carboplatin and paclitaxel, followed by maintenance with X"; "in combination with chemotherapy, followed by X maintenance") is emitted as one row per phase in which the medicine is given, like rule 12: - Row A (first phase): combination = the first-phase regimen; treatment_phase = null, or "induction" if the text says induction; line from the text. - Row B (maintenance): treatment_phase="maintenance"; combination = the maintenance regimen; prior_therapy = "<Row A regimen> without progression"; same line as Row A. - When maintenance depends on a biomarker (e.g. dMMR → monotherapy maintenance, pMMR → + olaparib), the biomarker goes on the maintenance rows; Row A carries it only if the first phase itself is restricted. - If the medicine is not given in the first phase ("after first-line induction with atezolizumab, carboplatin and etoposide"), emit only the maintenance row, with the induction in prior_therapy. 13. `line_of_therapy` counts the minimum number of prior lines of systemic therapy that the wording guarantees: 0 → 1L, at least 1 → 2L, at least 2 → 3L+. A 2L row means "after one or more prior lines". - "first-line", "previously untreated", "treatment-naive", "no prior systemic therapy (or chemotherapy) for metastatic disease" → 1L. - "second-line", "after one prior line", "after failure of prior platinum-based therapy", "relapsed", "refractory", "relapsed or refractory", "previously treated", "pretreated", "after prior therapy", "progressed on or after", "platinum-resistant", "platinum-sensitive relapsed" → 2L. - An upper bound ("no more than two prior chemotherapy regimens", "one or two prior regimens") → 2L; write the cap into prior_therapy. - "after at least two prior therapies", "after ≥2 prior lines", "third-line or later", "heavily pretreated", "after multiple prior therapies" → 3L+. - Autologous stem cell transplant (ASCT): in lymphomas, salvage chemotherapy plus ASCT is the second line, so "relapsed or refractory after (failed) ASCT" → 3L+, and post-ASCT consolidation without relapse ("at increased risk of relapse or progression following ASCT") → 2L. In multiple myeloma, induction, ASCT, consolidation and maintenance together are the first line (IMWG) → 1L. - Branches with different counts in one sentence → one row at the lowest count (rule 1); split only when one branch is untreated (1L) and another pretreated. - Naivety to one drug class ("EGFR TKI-naive", "BRAF inhibitor-naive", "not previously treated with endocrine therapy") does not set a line → any, unless other words fix it; write e.g. "No prior EGFR TKI" or "No prior endocrine therapy" into prior_therapy. - A disease state is not a prior line. Refractoriness to radioactive iodine (thyroid cancer) and castration resistance ("castration-resistant", "after failure of androgen deprivation therapy") describe the disease; count only systemic therapy given for it. Metastatic castration-resistant prostate cancer before chemotherapy → 1L; after docetaxel → 2L. RAI-refractory thyroid cancer with no prior systemic therapy → 1L. - A prior therapy the label waives guarantees nothing: "previously treated with an anthracycline and a taxane … unless patients were not suitable for these treatments", "previously treated with, or are not considered candidates for, available therapies" → any. Keep the wording in prior_therapy. - Count the lines given for advanced (unresectable or metastatic) disease. Neoadjuvant or adjuvant therapy counts only when the label ties it to early relapse ("relapsed during or within 6 months of completing adjuvant therapy"). "Two or more prior systemic therapies, at least one of them for advanced disease" → 2L. A prior regimen "in any setting" may have been neoadjuvant or adjuvant → any. Prior regimens whose setting the label leaves open count as lines ("at least 2 prior anti-HER2 treatment regimens" → 3L+). - Radiotherapy is not systemic therapy: "at least one prior radiation and/or chemotherapy treatment" → any. - A switch within first-line therapy before progression ("upon detection of an ESR1 mutation and without disease progression during first-line endocrine therapy in combination with a CDK4/6 inhibitor") → 1L, prior_therapy "During first-line endocrine therapy + CDK4/6 inhibitor, before progression". - Neoadjuvant and adjuvant rows → any. Induction, consolidation and maintenance rows take the line of the therapy they belong to when the text says so ("newly diagnosed", "following first-line chemotherapy" → 1L; "relapsed", "platinum-sensitive relapsed" → 2L); otherwise any. - If no line information at all → "any". 14. Patient-status eligibility criteria — "cisplatin-ineligible", "transplant-eligible", "BCG-unresponsive", "high risk of recurrence", "postmenopausal", "ECOG 0-1" — go in `patient_eligibility`. They are NOT prior therapy (the patient hasn't received them), NOT biomarkers (not molecular), NOT stage. The most common slip is putting "cisplatin-ineligible" into `prior_therapy` — that is wrong; the patient was deemed *unable* to receive cisplatin, they didn't take it. Multiple criteria joined with "; ". Null if none. The age group goes into `age_group` (enum); explicit age bounds go here as well. Also in `patient_eligibility`: - a population limited by sex ("women with HR-positive …" → "Women"; "women and men"); - explicit age bounds ("up to and including 25 years of age", "26 years of age and above", "aged 6 months to less than 18 years", "aged 1 year and older"); - how a subgroup receives it, from the same paragraph ("In pre- or perimenopausal women, the endocrine therapy should be combined with a LHRH agonist") → append to every row of that indication; - every restriction on recurrent disease ("not amenable to curative surgery or radiation therapy"); - a disease-state threshold such as "MRD ≥ 0.1%" or "in first or second complete remission", which is no biomarker cut-off: keep it out of diagnostic_criterion. A condition that qualifies one alternative stays with that alternative in one clause ("following ASCT, or following at least two prior therapies when ASCT or multi-agent chemotherapy is not a treatment option" → all of it in prior_therapy), and alternatives within one condition stay one clause joined by "or" ("progressed on or are intolerant to imatinib and sunitinib" → prior_therapy "Progressed on or intolerant to imatinib and sunitinib"). Do not split them between prior_therapy and patient_eligibility. What the label says about its trial is no condition: "The patients in the registration study were not previously treated with trastuzumab or an aromatase inhibitor" → notes, not prior_therapy. A sentence that qualifies the benefit without changing who may be treated ("an increase in overall survival is established only in patients with low tumour PD-L1 expression"; "a possible lower benefit in patients without a RET mutation") → copy it into notes. 15. `age_group` is REQUIRED. Map SmPC wording: - "adult patients" / "adults" / "patients ≥18 years" → age_group="adult". - "pediatric patients" / "children aged 1-17" with no adult inclusion → age_group="pediatric". - "adults and adolescents aged 12 and older" / "patients aged ≥6 months" / any indication that mixes adult and pediatric populations → age_group="mixed". - "women", "men", "postmenopausal women", "pre- or perimenopausal women" → age_group="adult" (adult populations by definition). - A section-level statement such as "X is indicated in adults for the following indications" applies to every indication under it → age_group="adult". - SmPC says "patients" with no age qualifier and no such statement → age_group="any". 15a. Keep the qualifiers of a regimen verbatim in combination, including what is left out ("alternating with R-DHAP (or R-DHAOx) without IMBRUVICA"). 16. `test_method` is a workflow-level descriptor — what kind of pathology lab test the indication implies. Pick from: - "IHC" — protein expression by immunohistochemistry (PD-L1, HER2 IHC, MMR proteins, ER/PR). - "ISH" — in situ hybridization, umbrella for FISH/CISH/SISH (HER2 amplification, ALK/ROS1 fusions when ordered as FISH). - "molecular" — any DNA/RNA assay regardless of platform: PCR, Sanger sequencing, NGS, ddPCR (mutation detection like EGFR/BRAF/KRAS, fusion detection by RNA-seq, MSI by PCR). - "cytogenetics" — conventional karyotype / cytogenetic analysis (heme malignancies). - "flow-cytometry" — phenotypic marker detection in heme (e.g. CD30, CD20, CD22). Multiple values are valid when the criterion is algorithmic (HER2: ["IHC", "ISH"]). Empty list `[]` when biomarker_name="none" — no test applies. 17. MMR / MSI alteration_type disambiguation. The text can present MMR/MSI status in two framings, and both carry clinical meaning but the enum forces a single choice: - Protein-loss framing ("deficient mismatch repair (dMMR)", "loss of MLH1/MSH2/MSH6/PMS2", "MMR protein loss by IHC") → biomarker_alteration_type="loss". - MSI framing ("microsatellite instability-high (MSI-H)", "MSI by PCR/NGS" alone, no mention of MMR protein loss) → biomarker_alteration_type="MSI". - When the SmPC uses BOTH framings together ("dMMR or MSI-H", "mismatch repair deficient or microsatellite instability high", "MMR deficient / MSI high") → prefer biomarker_alteration_type="loss". Rationale: the primary molecular defect is MMR protein deficiency; MSI-H is the downstream consequence. Pathology labs diagnose this most commonly by IHC for MMR protein loss, so "loss" better captures the diagnostic workflow implied by the indication.

drug_inn: osimertinib drug_brand: Tagrisso mesh_candidates: ["Carcinoma, Non-Small-Cell Lung"] ← pick ONE from this list for each row's `mesh_term` INDICATION TEXT (§4.1): …section 4.1 text…