table,variable,type,description,allowed_values
claims,claim_id,integer,Unique identifier for the claim within the dataset.,
claims,doc_id,string (16-hex),"Anonymised, stable identifier of the source document. Join key to `documents`. Group by this to recover which claims came from the same document.",
claims,pub_date,date,Publication date of the source document (YYYY-MM-DD). Empty when date_known=0.,
claims,pub_year,integer,Publication year. 0 when date_known=0.,
claims,pub_month,integer 1-12,Publication month. 0 when date_known=0.,
claims,date_known,0/1,1 if a publication date was recovered. ~0.9% of claims have no recoverable date; do not treat them as any particular year.,
claims,disease,string,"The health condition named in the claim, lower-cased and trimmed. Free text as written in the source — NOT mapped to a controlled vocabulary. 'covid-19', 'covid' and 'coronavirus' are distinct values.",
claims,factor,string,"The exposure, treatment, behaviour or trait claimed to affect the disease. Free text, lower-cased, trimmed.",
claims,relation,categorical,How the source framed the disease–factor link.,association | risk_factor | treatment | cause | protective | mechanism | prevention
claims,direction,categorical,Whether the factor is said to increase or decrease the disease. 'na' when no direction is stated (~76% of claims).,increases | decreases | na
claims,epistemic_stance,categorical,The epistemic posture of the statement. Repaired from occasional truncated raw values.,asserted | correlational | speculative | attributed | other
claims,study_tier,categorical,"The kind of study the source document PRESENTS ITSELF as reporting. Self-described, not verified. See documents.pubmed_article_types for verified types where available.",clinical_rct | review_meta | observational | clinical | preclinical | unspecified
claims,evidence_strength,categorical,"Evidence strength as PRESENTED by the source. Near-deterministic from study_tier (see README); treat as one quality signal, not two.",minimal | low | moderate | high | very_high | unknown
claims,evidence_score,integer 0-5,"Ordinal encoding of evidence_strength: unknown=0, minimal=1, low=2, moderate=3, high=4, very_high=5.",
claims,tier_score,integer 0-5,"Ordinal encoding of study_tier: unspecified=0, preclinical=1, observational=2, clinical=3, review_meta=4, clinical_rct=5.",
claims,is_causal,0/1,1 if relation = cause.,
claims,is_directional,0/1,1 if direction is not 'na'.,
claims,contamination_flag,0/1,1 if the claim belongs to a detected promotional-content artifact (see README). Retained so researchers can decide; PDFCI-style analyses exclude these.,
claims,source_type,categorical,"Publisher type of the source document, from recovered provenance. literature_index = a PubMed record (PMID present, 99.4%) or a curated peer-reviewed-literature domain without a PubMed record (0.6%). Other types assigned by a curated map of the ~570 highest-volume publishers (manually reviewed); 'other' = resolved publisher outside that map (long tail of small sites); 'unresolved' = provenance not recoverable (15 claims).",literature_index | news_media | trade_press | press_release | promotional | other | unresolved
claims,pmid,integer,PubMed identifier when the source document is a PubMed record; 0 otherwise (63.7% of claims carry one). Join key to documents.pubmed_* and to external PubMed metadata.,
claims,pair_total_claims,integer,Total claims for this disease×factor pair across the whole dataset (all years).,
claims,pair_n_docs,integer,Distinct documents contributing claims to this pair.,
claims,pair_first_year,integer,Earliest year the pair appears.,
claims,pair_pct_increases,float 0-100,"Of the pair's directional claims, the % saying 'increases'. Empty if the pair has no directional claims.",
claims,pair_contested,0/1,1 if the pair has ≥25 claims in EACH direction.,
claims,disease_total_claims,integer,Total claims naming this disease.,
claims,disease_n_factors,integer,Distinct factors linked to this disease.,
claims,factor_total_claims,integer,Total claims naming this factor.,
claims,factor_n_diseases,integer,Distinct diseases linked to this factor.,
claims,doc_n_claims,integer,Number of claims extracted from the same source document.,
documents,doc_id,string (16-hex),Anonymised document identifier. Primary key.,
documents,pub_date,date,Publication date; empty if unknown.,
documents,date_known,0/1,,
documents,source_type,categorical,"Publisher type of the source document, from recovered provenance. literature_index = a PubMed record (PMID present, 99.4%) or a curated peer-reviewed-literature domain without a PubMed record (0.6%). Other types assigned by a curated map of the ~570 highest-volume publishers (manually reviewed); 'other' = resolved publisher outside that map (long tail of small sites); 'unresolved' = provenance not recoverable (15 claims).",
documents,pmid,integer,PubMed identifier when the source document is a PubMed record; 0 otherwise (63.7% of claims carry one). Join key to documents.pubmed_* and to external PubMed metadata.,
documents,n_claims,integer,Claims extracted from this document.,
documents,n_diseases,integer,Distinct diseases in this document's claims.,
documents,n_factors,integer,Distinct factors in this document's claims.,
documents,pubmed_pub_year,integer,From the PubMed record (0 if not a PubMed document). Independent of pub_date.,
documents,pubmed_journal,string,Journal title from the PubMed record.,
documents,pubmed_article_types,string,Pipe-separated PubMed publication types (e.g. 'Journal Article|Randomized Controlled Trial'). VERIFIED study type — compare against claims.study_tier.,
documents,pubmed_is_rct,0/1,1 if PubMed types include Randomized Controlled Trial.,
documents,pubmed_is_meta_analysis,0/1,1 if PubMed types include Meta-Analysis.,
documents,pubmed_is_systematic_review,0/1,1 if PubMed types include Systematic Review.,
documents,pubmed_is_review,0/1,1 if PubMed types include Review.,
documents,pubmed_is_clinical_trial,0/1,1 if PubMed types include any Clinical Trial type.,
documents,pubmed_country,string,Country of publication from the PubMed record.,
documents,pubmed_n_mesh_terms,integer,Number of MeSH descriptors on the record.,
documents,pubmed_has_abstract,0/1,1 if the PubMed record carries an abstract.,
pairs,disease,string,,
pairs,factor,string,,
pairs,total_claims,integer,,
pairs,n_docs,integer,,
pairs,first_year,integer,,
pairs,last_year,integer,,
pairs,n_increases,integer,,
pairs,n_decreases,integer,,
pairs,pct_increases,float,% of directional claims saying increases; empty if none.,
pairs,n_causal,integer,Claims phrased causally.,
pairs,n_top_tier,integer,Claims from documents presenting as RCT or review/meta-analysis.,
pairs,n_high_evidence,integer,Claims presented as high or very-high evidence.,
pairs,contested,0/1,≥25 claims in each direction.,
diseases,disease,string,,
diseases,total_claims,integer,,
diseases,n_factors,integer,,
diseases,n_docs,integer,,
diseases,first_year,integer,,
diseases,last_year,integer,,
factors,factor,string,,
factors,total_claims,integer,,
factors,n_diseases,integer,,
factors,n_docs,integer,,
factors,first_year,integer,,
factors,last_year,integer,,