Table 3.

Factors attributing to mean tumor ADC

Factors:	Proportion of variance in ADC predicted by these factors (LOOCV R2)
A. Hardware and acquisition parameters:	64.3%
Repetition time (TR)*
Echo time (TE)*
Flip angle
Pixel Bandwidth
In plane resolution*
Slice thickness
Number of signal averages (NSA)*
Maximum b-value*
Number of b-values*
Signal to noise ratio (SNR)*
Scanner model*
Magnetic field strength
B. Patient-intrinsic parameters	0.4%
Age
Sex
cT-stage (assessed on baseline MRI)
cN-stage (assessed on baseline MRI)
Response to chemoradiotherapy (complete versus incomplete response)
Tumor volume
C. Center	32.5%
Umbrella variable to account for any additional unknown variations between centers (e.g., patient preparation protocols, types of coils used, fat suppression techniques, etc.)
Significant parameters: center
All (A + B + C) combined	63.5%
Significant parameters: center, age, TR, TE, in-plane resolution, slice thickness, NSA, maximum b-values, number of b-values, scanner model, and SNR

Further details of the regression analysis can be found in Supplementary Materials 2. The LOOCV R² value is a leave-one-out cross-validated goodness-of-fit measure indicating the proportion of the variance in the dependent (i.e., ADC) variable that can be explained by the independent variables using a linear regression model

^*Indicate the variables that were significant with a p value < 0.05 based on a likelihood ratio test. For continuous variables, all polynomial terms were tested jointly