Audit Laboratory Panel Completeness and Discordant Missingness — panel_missingness

Analytes ordered together (e.g. BMP or CBC) should exhibit consistent missingness rates. When a component is missing significantly more often than its panel median, or deviates strongly from peer hospital sites, it flags a likely unmapped analyte.

Usage

panel_missingness_audit(
  features,
  panels = NULL,
  by = "site",
  threshold = 0.15,
  peer_gap = 0.2
)

Arguments

features: Data frame containing patient features with analyte columns.
panels: Named list of character vectors mapping panel names to analyte column names (default BMP and CBC).
by: Column name to stratify by (default 'site'); if NULL or absent, audited as 'ALL'.
threshold: Maximum allowable missingness gap above panel median (default 0.15 = 15 percent).
peer_gap: Cross-site discrepancy gap flagging peer outliers across hospitals (default 0.20 = 20 percent).

Value

A data frame with columns: site, panel, analyte, missing_count, total_count, missing_rate, panel_median_missing, missingness_gap, flagged, reason.