No synthetic marker found in these counts. That is not proof of where the data came from.

The sorting behind these counts was checked against labels made by a council of models; for some classes too few entries could be checked. No person has read the individual rows, and no clinician has read the class definitions.

Trial endpoint atlas: data dictionary

Registry data from ClinicalTrials.gov, an online database; the U.S. National Library of Medicine maintains its website. This lab is independent of the registry; the grouping and counts are the lab's work, not the registry's.
Registry records change; the live record on ClinicalTrials.gov is current, this file is not.

ABBREVIATION
  The page's controls (links and switches) shorten cystic fibrosis (CF). Elsewhere the abbreviation appears only inside quoted registry wording, such as "non-CF".

UNITS
  A study is counted once in each class its main outcomes name, so class totals can add up to more than the number of studies.
  An entry is one main (primary) outcome as the trial team wrote it: a measure, a description and a time frame.
  The research tools state: studies are counted once per class; entries are primary-outcome entries; a study can name several classes

FIGURES
  Figure 1. Studies by class of main outcome and start year. Figure 2. Studies by class of main outcome and trial phase. Both count studies.

START KINDS
  actual: the registry marks the start date actual.
  planned: the registry marks the start date estimated.
  untyped: a start date is given but the registry does not say whether it is actual or planned; we do not guess.
  no date: no start date is given.
  A study with a start date is in exactly one of actual, planned and untyped; Figure 1 counts it once in its start year.

CSV FILE (trial-atlas-counts.csv)
  Columns: table (which table the row belongs to), class_id (the class id, empty for whole-atlas rows), class (the class name), key (a year, a phase code, a number's name or 'line N'), value (a whole number of studies or entries, or a line of text for the about rows).
  Values of the table column:
    about: The source, dates, live-record line, citation, every change the lab made, license and retention, one line per row.
    totals: studies and entries for the whole atlas.
    studies_by_start_year: All studies by start year (key: the year, or unknown).
    class: One number per class (key: studies, entries, studies_cf_only, lead_sponsors, studies_without_lead_sponsor).
    start_year: Studies of a class by start year.
    actual_start_year: Studies of a class by actual start year.
    planned_start_year: Studies of a class by planned start year.
    untyped_start_year: Studies of a class by start year, where a start date is given but the registry does not say whether it is actual or planned; we do not guess.
    first_posted_year: Studies of a class by first-posted year.
    phase: Studies of a class by phase code (key: the registry's code, or none when no phase is given).

JSON FILE (trial-atlas-counts.json)
  Top-level keys:
    synthetic_data: Present only in a preview: says the data are made up for testing and are not from ClinicalTrials.gov.
    check_note: Present only in the build published on the site: what the sorting behind the counts was checked against, and what no person has read.
    kind: What this file is.
    source_note: The source line: what ClinicalTrials.gov is and that the lab is independent of it (a placeholder line in a preview).
    live_record: Registry records change; the live record is current and this file is not.
    how_to_cite: A citation for this dated copy; fill in the date you read it.
    registry_terms: The registry's terms as the research tools recorded them, every item kept, with one ordered list of every change the lab made (the tools' changes first, then this file's).
    units: The counting units, as the research tools state them.
    data_processed_by_registry: The date and time the registry processed the data, as the registry gave it.
    snapshot_sha256: The hash of the dated copy these counts were made from.
    lexicon_version: The version of the rule lexicon that sorted the main outcomes.
    lexicon_sha256: The hash of that lexicon.
    studies: Studies in scope (the main unit of count).
    entries: Main-outcome entries of those studies (the unit that is sorted).
    entries_by_status: Entries sorted by a rule, by a model checked against an exact quote from the record, naming no measure (not stated), and not yet in a class (see the keys of entries_by_status).
    unsorted_entries: Entries not yet in a class: no rule sorted them and no accepted model tag placed them. Not the class Other, which a model or a person gives to a measure that fits none of the classes.
    flags: Entry counts for each flag (see the flags section).
    scope: How many studies the dated copy held, how many were kept, how many each scope rule left out, the split between studies whose every listed condition names cystic fibrosis and studies that list it among other conditions, and the start types (see the keys of scope).
    routes: The two searches: in-scope studies of each, and how many studies only one of them found (numbers only, no ids).
    studies_by_start_year: All studies in scope by start year, each counted once ('unknown': no start date).
    classes: One object per class, in the lexicon's order (see the class fields).
  Fields of each object in classes:
    id: The class id in the lexicon.
    class: The class name.
    domain: The domain the class belongs to.
    studies: Studies naming this class.
    entries: Main-outcome entries sorted into this class.
    studies_cf_only: Of those studies, the ones whose every listed condition names cystic fibrosis.
    lead_sponsors: How many different lead sponsors those studies have.
    studies_without_lead_sponsor: Studies with no lead sponsor recorded.
    by_start_year: Studies by start year.
    by_actual_start_year: Studies by start year, where the registry marks the start actual.
    by_planned_start_year: Studies by start year, where the registry marks the start planned (estimated).
    by_untyped_start_year: Studies by start year, where a start date is given but the registry does not say whether it is actual or planned; we do not guess.
    by_first_posted_year: Studies by the year the record was first posted.
    by_phase: Studies by phase, as the registry codes it.
  Keys of scope:
    studies_in_route: Studies in the dated copy's answer to the condition search, before the scope rules.
    included: Studies kept by the scope rules (the studies in scope).
    excluded_by_rule: Studies each scope rule left out, by rule id (see the scope rules).
    cf_only: Studies in scope whose every listed condition names cystic fibrosis.
    cf_among_others: Studies in scope that list cystic fibrosis among other conditions.
    start_actual: Studies in scope whose start date the registry marks actual.
    start_planned: Studies in scope whose start date the registry marks estimated (planned).
    start_untyped: Studies in scope where a start date is given but the registry does not say whether it is actual or planned; we do not guess (a type other than actual or estimated is counted here too).
    start_no_date: Studies in scope where no start date is given.
  Keys of entries_by_status:
    rule: Entries sorted by a rule of the lexicon.
    model: Entries sorted by a model, checked against an exact quote from the record.
    not_stated: Entries that name no measure.
    unclassified: Entries not yet in a class (the same number as unsorted_entries).
  Keys of routes:
    main_in_scope: Studies in scope found by the condition search; the counts come from this search.
    recall_in_scope: Studies in scope found by the term search.
    only_main: How many studies only the condition search found (counted).
    only_recall: How many studies only the term search found (not counted).
  Keys of registry_terms:
    source: The registry the data come from.
    data_processed_by_registry: The date and time the registry processed the data, as the registry gave it.
    snapshot_fetched_at: When the dated copy was taken.
    terms_url: The registry's Terms and Conditions page.
    terms_last_updated: The last-updated date that page gave.
    modifications: Every change the lab made, in one ordered list: the research tools' changes, then this file's.
    licence: What the lab's license covers.
    retention: How long the registry's terms apply.
    no_warranty: The registry's no-warranty statement, as the research tools recorded it.
    sponsor_responsibility: Who writes and answers for each record, as the research tools recorded it.
    disclaimer_url: The registry's Disclaimer page.
    disclaimer_last_updated: The last-updated date that page gave.
    third_party_copyright: The registry's note on third-party and international copyright, as the research tools recorded it.
    keep_current: The registry's request that published data be kept current, and how this dated copy stands.

FLAGS (counts of entries)
  composite: The entry's wording names a composite outcome (for example 'composite of', 'time to first of').
  multi_class: The entry was sorted into more than one class and is not marked composite.
  vague: The entry uses vague wording, such as 'lung function' or 'efficacy', that names no specific measure.
  no_primary_outcome: The study entered no main outcome at all.

SCOPE RULES (each left studies out; the counts are in scope.excluded_by_rule)
  A condition names cystic fibrosis when it holds the words "cystic fibrosis" or "mucoviscidosis", or is written just as "CF", and not only inside a negation such as "non-CF".
  X1: Observational study.
  X2: Expanded access.
  X3: The study type is missing, or is not interventional, observational or expanded access.
  X4: The condition list does not name cystic fibrosis (the words "cystic fibrosis" or "mucoviscidosis", or a condition written just as "CF").
  X5: The condition list names cystic fibrosis only in a negation, for example "non-CF bronchiectasis".
  X6: Excluded by hand, with the reason written in the exclusions file. X6 is applied first.
  Rules are applied in the order X6, X1, X2, X3, then X5/X4.

CO-OCCURRENCE FILE (trial-atlas-co-occurrence.csv)
  The pairs on the page Which measures are registered together. The unit is the study, not the entry. It shows what trials registered together, not which measures agree, replace or predict each other.
  Not paired, by the research tools' rule: other, not_stated.
  Columns:
    class_a: The class id of the first class of the pair, in the lexicon's order (or 'about' for a row of the terms).
    class_b: The class id of the second class of the pair (or, in an about row, one line of the source, dates, changes, license and retention).
    studies: Studies that registered a main outcome in both classes (empty in an about row). Each pair appears once; pairs with no study are not listed.

CLASS, YEAR AND PHASE FILE (trial-atlas-class-year-phase.csv)
  Studies by class, start type, start year and phase. Summed over phase, a row group gives the class's studies for that start type and year in the counts file.
  Columns:
    class_id: The class id in the lexicon (or 'about' for a row of the terms).
    class_label: The class name (or, in an about row, one line of the source, dates, changes, license and retention).
    start_type: actual, planned or untyped: whether the registry marks the start date actual or estimated; untyped: a start date is given but the registry does not say whether it is actual or planned; we do not guess. A study with no start date is in none of them, so it is not in this file.
    start_year: The study's start year (unknown: no start year).
    phase: The phase as the registry codes it, joined with / for a combined phase, or none when no phase is given.
    studies: Studies of this class with that start type, start year and phase.

CLASSES NOT SHOWN
  Some classes were counted but are not shown: on a check against model-made labels, the lab's tagging fell below its accuracy threshold. Every per-class number of those classes is left out of these files: they have no row in the CSV files, no object under classes and no pair in the co-occurrence file. The JSON file is therefore not a copy of the research tools' counts file. Their names, and one aggregate figure, are on the front page under Not shown.
  Totals that are not per class (studies, entries, entries_by_status, unsorted_entries, flags, scope, routes and studies_by_start_year) still include their studies and entries.

LICENSE
  The lab's licence covers its own tags, code and counts only; registry text and fields remain ClinicalTrials.gov data under its terms.
  The registry's terms apply for as long as the data are kept, in any copy, file or page made from them.