for sure direct global access, especially looping on indexes is faster than SQL. It seems that the index global you used: ^PatientD("Date",FromDate)) should be: ^PatientI("Date",FromDate)) 

Using queue manager (for each "FromDate") will do this in parallel which will speed things much more!

The is a core process called GARCOL which is used to handle large global kills.
Securing Caché, Ensemble, And Operating System Resources | Caché & Ensemble 2018.1.0-2018.1.3
 

I would consider, replacing the lock + and lock - with setting/killing a (temporary) global, which can be mapped to IRISTEMP for better performance, and avoid journaling. It might be easy to implement such in parallel to the current lock mechanism (add those while keep the old in place) then migrate the management page in the application to loop on this global instead of the entries in lock table. The new approach can also have the timestamp for the lock and then have a function that will auto-retain locks (kill) based on a time interval.