FIG 4.

Predicted probabilities and 95% CIs for the four categories of self-reported frequency to changes to clinical practice, by caring for patients with COVID-19 (yes v no), on the basis of the generalized ordinal logistic regression. (A), (B) Self-reported frequency to changes to clinical practice when asked “Please choose the proportion of your practice that has been affected by the following changes since March 2020 compared to last year.” Among the proposed options, the most clinically relevant are shown: (A) change to a less effective regimen, but with fewer complications or side effects, (B) stop giving chemotherapy to the patients who were already on treatment, and (C) delay the start of chemotherapy. Reported frequencies are based on Likert scale options: in all cases, in most cases, in some cases, in the minority of cases, no changes, and not applicable. Because of the small sample size across categories, in all cases and most cases were collapsed into one category. Respondents who selected not applicable are excluded.