After upgrading Cassandra from 3.11.3 to 3.11.9, Cassandra read latency 99% increased significantly. Getting back to 3.11.3 immediately fixed the issue.
I have observed "SStable reads" increases after upgrading to 3.11.9.
The same behavior was observed by some other users: https://email@example.com/msg61247.html
According to Paulo Motta's comment, this behavior may be caused by https://issues.apache.org/jira/browse/CASSANDRA-15690 which was introduced on 3.11.7 and removed an optimization that may cause a correctness issue when there are partition deletions.