Minimal Clinically Important Difference Estimates Are Biased by Adjusting for Baseline Severity, Not by Regression to the Mean