Discretizing Continuous Event Time Data

Area of Science:

• Biostatistics
• Survival Analysis
• Data Science

Background:

• Statistical analyses often require discretizing continuous time data into intervals.
• Accurate assignment of events, including outcomes and loss to follow-up (LTFU), is crucial for reliable results.
• Existing methods for handling LTFU in discretized time can introduce errors.

Purpose of the Study:

• To determine the optimal method for assigning outcomes and LTFU events when discretizing continuous time data.
• To evaluate the accuracy of different LTFU assignment strategies in simulated and real-world datasets.
• To compare discretized time analysis results with continuous time analyses.

Main Methods:

• Simulated data was used to demonstrate outcome assignment to the end of an interval.
• Four distinct methods for assigning LTFU events were compared using simulated and 20 real-world datasets.
• Cumulative risk curves generated from discretized time analyses were compared against continuous time analyses.

Main Results:

• Assigning outcomes to the end of the interval was shown to be appropriate.
• An approach assigning LTFU to the closest interval start or end demonstrated the least error across all tested scenarios.
• This optimal LTFU assignment method outperformed strategies that consistently censored at the interval's beginning or end.

Conclusions:

• The optimal method for discretizing time involves assigning outcomes to interval ends and LTFU to the nearest interval boundary (start or end).
• This refined approach minimizes cumulative risk errors compared to traditional censoring methods in survival analysis.
• Accurate event time discretization is vital for robust statistical inference in time-to-event data analysis.