Aggregate source data for Figures 1–3

Fig1_source.csv: sequential participant counts; not PSG-record counts.
Fig2_source.csv: Cox hazard ratios and 95% confidence limits for five specifications per outcome. CVD n=2485, events=104; all-cause n=2486, events=304.
Fig2_paired_contrast_source.csv: difference in log-HB-change coefficients, full AHI3 minus full AHI4; low/high are percentile 95% intervals from 1000 paired participant bootstrap replicates.
Fig3_source.csv: pointwise time-specific hazard ratios and 95% limits at 1, 3 and 5 years; at_risk is the risk-set count and cumulative_CVD_deaths is the cumulative number of cardiovascular deaths at or before the time point. Each extended model has 18 coefficients.

All HB-change hazard ratios use the same fixed SD=0.6952283 of log(1+HB at visit 2) minus log(1+HB at visit 1). Change is not annualized. AHI3 and AHI4 correspond to the desaturation-only rdi3p and rdi4p indices. Model details are in the manuscript and Additional file 1.
Values are aggregate saved analysis outputs, not participant-level data. They support figure reconstruction, not independent replication of participant-level model fitting. Estimates are exploratory conditional associations; bootstrap intervals do not account for prior research-direction selection. Time-specific intervals are pointwise, not simultaneous.
