Figure 2
From: Prevalent cases in observational studies of cancer survival: do they bias hazard ratio estimates?

ER-specific annual mortality rates and hazard ratios by survival time. (A) Annual mortality rate for ER-negative and -positive tumours in incident cases by survival time in years. (B) Corresponding observed incident HRs for ER status (ER negative is referent) and the expected full cohort left-truncated HRs by survival time in years. The reference line shows the overall observed incident HR estimate calculated using a standard Cox proportional hazards model without adjustment for time-dependent effects. Observed incident HRs were calculated using a standard Cox proportional hazards model for incident cases only split at various time periods. Expected HRs were calculated using the extended Cox model formula at each year.