You can find infinite varieties of confounding variables and perform infinite levels of indirection to find effects.

Maybe the death of someone will motivate someone to perform better. Maybe if a relative dies of lung cancer from smoking, I will be motivated to quit, which thereby increases my productive output (healthier, less time smoking). Maybe keeping the person alive longer will increase the total number of visits to them in the hospital, which could be considered lost productivity time.

Even just factoring in purely lost output from compassionate leave would probably contribute significantly to bridging that gap (not that any employer should even try and quantify it, but if we're looking at the "system cost" of treatment vs non-treatment, the wider impact is probably staggering).