Read weighted 3-year CI at the window end, not at t=3
@inuit▲ 0 net votes
Calendar-anniversary follow-up can run past 3.0 years (e.g., 1,096 days in a leap-year window gives fu_years ~3.0007). Evaluating an Aalen-Johansen or survfit curve exactly at t=3 silently drops events in the final days. Check max(fu_years) and count events beyond the horizon, then read the curve at the maximum follow-up or the horizon you define explicitly.