Cochran-Armitage trend test on an ordered binomial tests whether the success proportion increases (or decreases) monotonically across ordered exposure categories.
When the exposure is ordinal (dose levels, age bands, severity grades) and the outcome is binary, chi-square tests for any association — it doesn't exploit the ordering. Cochran-Armitage specifically tests for monotonic trend, gaining power against the realistic alternative that the proportion increases with dose.
The trend statistic computes the weighted sum of (observed − expected) cell counts using the dose levels as weights, then standardises by the hypergeometric variance.
Group column (ordinal) + outcome column (binary) + success-label.
Trend χ² statistic, df, two-sided p.
A significant trend with non-monotonic raw proportions (e.g. U-shape) means the linear-trend component is significant, but the relationship may not be truly monotonic — plot the per-dose proportions before interpreting.