Patient vs Summary Mode

Introduction

CohortContrast Viewer supports two data modes:

• Patient: uses patient-level parquet files (for example data_patients.parquet).
• Summary: uses precomputed aggregate parquet files (for example concept_summaries.parquet).

In the Studies table, the Mode column shows this in one word: Patient or Summary.

Study selection with mode column

How each mode is produced

• Patient mode: produced directly by CohortContrast(..., createOutputFiles = TRUE).
• Summary mode: produced by precomputeSummary(studyPath = ..., outputPath = ...).

The bundled example studies include one patient-mode study and one summary-mode study:

patientStudyPath <- system.file("example", "st", "lc500", package = "CohortContrast")
summaryStudyPath <- system.file("example", "st", "lc500s", package = "CohortContrast")

data.frame(
  study = c("lc500", "lc500s"),
  mode = c(
    CohortContrast::checkDataMode(patientStudyPath)$mode,
    CohortContrast::checkDataMode(summaryStudyPath)$mode
  )
)
#>    study    mode
#> 1  lc500 patient
#> 2 lc500s summary

This mirrors the distinction shown in the Viewer study-selection table.

summary_result <- CohortContrast::precomputeSummary(
  studyPath = file.path(getwd(), "studies", "LungCancer_1Y"),
  outputPath = file.path(getwd(), "studies", "LungCancer_1Y_summary"),
  clusterKValues = c(2, 3, 4, 5),
  minCellCount = 5
)

# Open viewer and load the summary study
CohortContrast::runCohortContrastViewer(
  dataDir = file.path(getwd(), "studies")
)

What is different in the UI

• Mappings merge actions:
  • Patient: available (Manual Merge, Hierarchy Suggestions, Correlation Suggestions).
  • Summary: disabled (history is still visible).
• Clustering updates:
  • Patient: Recluster runs live clustering.
  • Summary: clustering uses precomputed artefacts for selected k when filters are applied.
• Data granularity:
  • Patient: row-level patient events are available.
  • Summary: only aggregate summaries are available.

Recommended usage

• Use Patient mode for interactive concept curation and merge decisions.
• Use Summary mode for faster sharing and reproducible visualization of precomputed results.