Specify baseline-adjusted ANCOVA clearly

State the comparison at follow-up

For a randomised trial with a continuous endpoint measured before and after treatment, an ANCOVA model can compare follow-up outcomes while adjusting for a prognostic baseline measurement. The purpose is usually improved precision, not correction of a failure of randomisation. The FDA covariate-adjustment guidance discusses prespecified baseline adjustment in randomised trials.

Write the estimand first: population, outcome, follow-up time and treatment comparison. Confirm that the baseline measurement precedes treatment assignment or cannot be affected by treatment. A post-treatment variable should not enter the model merely because it predicts the endpoint.

Specify the model before inspecting treatment results

Document the dependent variable, treatment coding, baseline covariate units and any additional prespecified covariates. Explain functional forms and whether treatment-by-baseline interaction is scientifically necessary. With an interaction, a single treatment coefficient may represent a contrast only at a reference baseline value; report the intended contrast rather than labelling that coefficient an overall effect automatically.

Define the missing-data strategy for baseline and follow-up measurements. Check residual patterns, unusual observations and the uncertainty estimator. Adjustment does not make a model reliable when the sample is too small for its complexity. Keep the planned analysis and justified sensitivity analyses distinguishable.

Hypothetical example

Consider a fictional model: follow-up score = 10 + 2 × treatment + 0.6 × baseline, where treatment is coded zero or one. At baseline 20, the predicted scores are 22 and 24, a difference of two points. At baseline 30, they are 28 and 30, again a difference of two. This constant contrast follows from the absence of an interaction in the example model.

The numbers illustrate coefficient interpretation only. They are not trial results, evidence of benefit or a reason to select this model for a particular instrument.

Report enough to reproduce the contrast

Give the adjusted difference with its confidence interval and units, the covariates, the analysis population and the handling of missing observations. Explain the direction of the scale so readers know what a positive difference means. Include descriptive baseline and follow-up summaries without using baseline significance tests to decide which covariates to retain.

Use the residual review worksheet to document model checks. If the treatment effect varies over baseline values, show clinically relevant contrasts or a prespecified standardised effect and explain the target population used for averaging.

Source verification: 6 November 2026. Examples are hypothetical.