Per km driven? Not clear from the report. They give the rate per million km for all collisions (including "slight") which fell - but they don't give a rate per km for fatal and serious collisions. Eyeballing the data (page 51) the overall reduction in rate looks likely to be entirely accounted for by the reduction in "slight" collisions.
It's even quite hard to work out how traffic volumes changed.
"Traffic growth on the route had increased by an average of 10% from 2009 to 2018 (Comparing average weekday traffic (AWT) between 2009 and 2018)"
Growth has increased????
Interestingly they compute a reduction in "fatality equivalents" - by weighting "slight" collisions as 0.01 of a fatal collision. So they report an overall reduction in "fatality" equivalents when there has been an increase in fatal and serious collisions.
The data are sliced and presented in many different ways - but without any clear presentation of the fact that the number of fatal and serious collisions per year (and likely per km) on the modified stretch of road actually increased.