Compare it against the uptime_in_seconds metric in the "server" section.
Both of these are cumulative numbers. They accumulate seconds for the entire time the Redis server process is running. This makes the used_cpu_sys and used_cpu_user figures more useless the longer the process runs.
If your Redis server was idle for long periods of time then suddenly started to consume lots of cpu, the percentage would stay low, falsely indicating the process is not currently consuming much cpu. If your Redis server was very busy for several minutes after it first started up, then settled down to a low level, the percentage would stay high for several minutes, falsely indicating the process is consuming more cpu than it actually is.
My advice is to not use these numbers to monitor your Redis server process's cpu usage. Use independent monitoring that measures the usage on a minute-by-minute basis rather than the whole time the process is running.