Treatment group outcome variance difference after dropout as an indicator of missing-not-at-random bias in randomized clinical trials