Cross-Tabulations and Proportions

Cross-tabulation summarizes the relationship between two categorical variables. In clinical research data management, cross-tabulations are useful for comparing outcome status by site, visit completion by visit, adverse event severity by relatedness, query status by site, or treatment arm by sex. A cross-tabulation is not automatically a statistical test; it is first a structured descriptive table. The simplest cross-tabulation uses `tabyl()`: ```r prepared_data |> tabyl(site, day28_status) ``` This produces counts of day 28 status within each site. To add row percentages: ```r prepared_data |> tabyl(site, day28_status) |> adorn_percentages("row") |> adorn_pct_formatting(digits = 1) ``` Row percentages answer the question: within each site, what percentage of participants fall into each day 28 status category? Column percentages answer a different question: within each day 28 status category, what percentage come from each site? The correct choice depends on the reporting question. The following example creates a cross-tabulation of treatment arm by sex: ```r prepared_data |> tabyl(treatment_arm, sex_label) |> adorn_totals("row") |> adorn_percentages("row") |> adorn_pct_formatting(digits = 1) ``` This may be useful for a baseline summary. However, the data manager should be careful not to overinterpret small differences. In randomized trials, baseline tables describe the sample; they are not usually intended to drive post-randomization decisions unless predefined procedures require it. Cross-tabulations can also reveal data quality problems: ```r prepared_data |> tabyl(site, visit_status) |> adorn_totals(c("row", "col")) ``` If one site has many `Pending` visits while other sites have few, the issue may be delayed data entry, true workflow differences, or misunderstanding of visit status coding. The table points to a question; it does not answer the question by itself. For more controlled reporting, cross-tabulations can be built using `count()` and `group_by()`: ```r visit_status_by_site <- prepared_data |> count(site, visit_status, name = "n") |> group_by(site) |> mutate( site_total = sum(n), percent = round(100 * n / site_total, 1) ) |> ungroup() visit_status_by_site ``` This long-format table is often easier to export, join, plot, or use in dashboards than a wide cross-tabulation. The appropriate format depends on the next step.

The denominator should be shown or explained in any table intended for decision-making. Many misunderstandings in clinical reporting arise not from complex statistics, but from unclear denominators.