Extended Data Table 2.

Logistic regression analysis of primary endpoint imputing missing data with delta-adjusted pattern mixture model

The imputation was based on delta-adjusted pattern mixture model adjusting for stratification factors with a range of delta values. The pattern mixture model of a binary endpoint can be expressed in the following form:

$\mathrm{logit}\left[\mathrm{Pr}\left(Y=1\mid R_y,X\right)\right]={\gamma }_0+{\gamma }_1\mathrm{X}+\delta \left(1-R_y\right)$

Where $\mathrm{\delta }$ represents the difference in the log-odds of $Y=1$ between non-missing and missing, $X$ are observed covariates, which is stratification factors here, and $R_y$ is a missing indicator of $Y$. While $\delta =0$ implies data missing at random, $\delta >0$ implies better outcome in missing patients and $\delta <0$ otherwise. All the missing pCR was imputed with the same $\mathrm{\delta }$. For example, $\log \left(0.5\right)$, $\log \left(1\right)$ and $\log \left(1.5\right)$ indicates patients with missing pCR would have an odds of 0.5, 1 and 1.5 times achieving a pCR compared to non-missing patients.