Treatment selection bias for chemotherapy persists in colorectal cancer patient cohort studies even in comprehensive propensity score analyses