Table 3 Best practice for data extraction from reported Kaplan–Meier curves

• Agree KM curve method in advance of data extraction

 - Choose to use the KM curve and information on follow-up (Scenario 12)

 - Choose to use the KM curve and numbers at risk (Scenario 13)

 - Choose to use both KM methods (e.g., for comparison)

 - But allow flexibility to deviate if specific issues arise (e.g., zero events in an arm or too few intervals with numbers at risk)

• For Scenario 12, agree on the time intervals in advance of data extraction

 - Select intervals guided by the shape of the KM curve (i.e., the event rates), rather than at regular time points

 - Select more time intervals where more events have occurred (i.e., the curves are steeper). Aim to include not more than a 15% event rate in an interval

 - Select fewer intervals where fewer events have occurred (i.e., the curves are flatter). Aim to include at least a 5% event rate in an interval

 - Avoid time intervals with zero events in one or both arms—make the time interval larger to avoid this

• For Scenario 13, select only time intervals that have accompanying numbers at risk

• For both scenarios, avoid extracting data at time points where few or no events have occurred (i.e., the curves have leveled out)

• Duplicate or cross-check data extracted by another researcher