I agree. However the hazard ratio is monotonously increasing across activity groups in the direction one would expect if there is an effect. So the data does i

I agree. However the hazard ratio is monotonously increasing across activity groups in the direction one would expect if there is an effect. So the data does indeed show a (meaningful) rise in risk going lower on the activity ladder, only it is not statistically significant. A continuous model might get more (statistical) traction here. They've got the data. Why use segments?