First of all, fit4NM 3.3.7 does not support NONMEM 72.
Features revised were as follows.
1. Prediction probability calculated using Somers' d.
The algorithm to caclculate Somers' d was revised as in SAS. We introduced the concept of rounding predictor or observation to the calculation of prediction probability for efficiency of computing. But we recommend the users to keep the number of digits in the original data for accuracy, because rounding categorizes data, which results in erroneously high prediction probability. This can be done by adjusting the number of rounding digits to include all the digits in your data. The reason we consider rounding is to quickly obtain a rough trend of prediction probability. R may take a long time to calculate prediction probability because of a big matrix generated to obtain Somers' d, when we do not using rounding.

2. The 95% confidence intervals of prediction intervals provided by the predictive checks.
We newly introduced the algorithm to calculate 95% CI of PIs. Although the plots of predictive checks including 95% CIs of PIs look very complex, you can draw more beautiful plots using other graphic softwares such Graphpad Prism or Sigmaplot.
A new version to support NONMEM 72 will be released in a while.