Skip to main content

task

Prefect task for the inclusion-criteria matching step.

Calls the pure functions in steps.criteria_matching to determine which scans meet the clinical trial eligibility criteria.

Module​

Functions​

criteria_matching_task​

def criteria_matching_task(    datasource: BaseSource,    config: CriteriaMatchConfig,    *,    scan_metrics: list[dict[str, typing.Any]] | None = None,    patient_data: list[CacheAccessor] | None = None,    filenames: Optional[list[str]] = None,    ga_metrics: dict[str, GAMetricsWithFovea | None] | CacheAccessor | None = None,    fluid_metrics: dict[str, FluidVolumeMetrics | None] | CacheAccessor | None = None,    cst_metrics: dict[str, CSTMetrics | None] | CacheAccessor | None = None,    gcc_metrics: dict[str, GCCMetrics | None] | CacheAccessor | None = None,    lesion_metrics: dict[str, LesionMetrics | None] | CacheAccessor | None = None,    ehr_patient_data: CacheAccessor | None = None,    enrichment_data: CacheAccessor | None = None,) ‑> CriteriaMatchResult:

Apply inclusion criteria and return eligibility results.

Inputs are two expandable channels: scan_metrics (a list of {"metrics": ..., "metrics_type": ...} sources, one per scan-derived metrics producer) and patient_data (a list of CacheAccessors, one per patient-data producer). The legacy scalar ga_metrics / fluid_metrics / cst_metrics / gcc_metrics / ehr_patient_data are still accepted and coerced into single-element channel lists by _coerce_channels (an explicit scan_metrics/patient_data channel list wins over any of the scalars). Each scalar metrics kwarg accepts either an in-memory {filename: metrics | None} dict or a CacheAccessor (e.g. a cache-backed step's wired .cache); a non-dict value is rebuilt into the dict via its typed *_metrics_from_cache wrapper.

The task adapts to the datasource:

  • For a FileSystemIterableSource the scan data is loaded with get_data_for_files (using filenames); patient_data and scan_metrics are joined onto it by bitfount patient id / ORIGINAL_FILENAME_METADATA_COLUMN respectively, when supplied.
  • For any other datasource the data is not loaded from the source. The single patient_data source's frame is the table to match against and filenames is unused; more than one patient_data source raises a ValueError, since merging multiple patient-grain sources by bitfount patient id is only supported on a file-system datasource. scan_metrics are only meaningful for file-system sources, so any supplied here are ignored (and a warning is logged).

scan_metrics is optional. If it is omitted (or ignored) while the config still specifies GA-derived filters, a warning is logged because those filters cannot be evaluated (their columns will be absent).

A field catalog is built from every wired scan_metrics source's field_specs() and passed to build_eligibility_filters so generic column_criteria grains resolve correctly. Any column_criteria column that neither the catalog nor the assembled frame has is logged as a warning — never raised, since column_criteria is a best-effort escape hatch. (Pathology-biomarker max_<label>_probability columns are only added to the catalog when a GA-derived scan_metrics source is wired — see _build_catalog — so a column_criteria targeting one of those triggers this warning on a GA-less run, since the column is genuinely never produced.)

Arguments

  • datasource: The data source to match criteria against.
  • config: The criteria-matching configuration.
  • scan_metrics: The expandable scan_metrics channel; see above.
  • patient_data: The expandable patient_data channel; see above.
  • filenames: Filenames to load scan data for, on a FileSystemIterableSource; unused otherwise.
  • ga_metrics: Legacy scalar GA metrics, coerced into scan_metrics; see above.
  • fluid_metrics: Legacy scalar fluid-volume metrics, coerced into scan_metrics; see above.
  • cst_metrics: Legacy scalar CST metrics, coerced into scan_metrics; see above.
  • gcc_metrics: Legacy scalar GCC metrics, coerced into scan_metrics; see above.
  • lesion_metrics: Metrics for lesion size, coerced into scan_metrics; see above.
  • ehr_patient_data: Legacy scalar EHR accessor, coerced into patient_data; see above.
  • enrichment_data: The patient_enrichment step's cache, rebuilt into a patient-grain frame and merged by bitfount patient id to contribute the supplied observations column, which observation criteria read alongside the EHR one. Optional; None leaves every frame untouched.

Returns The CriteriaMatchResult with one CriteriaEvaluation per row.