Approaches to handle missing follow-up time: A comparative analysis of contralateral breast cancer incidence

Abstract

Background

We aimed to compare various common approaches for handling missing vital status or follow-up time. As a case study for application of these methods, we estimated incidence of metachronous contralateral breast cancer (CBC).

Methods

For 1980–2016, incidence of metachronous CBC with follow-up through 2024 was estimated using Poisson regression with overdispersion, by age at incidence, year of incidence, histology and follow-up period. Missing follow-up time was ignored in the naive approach, simulated once using the average hazard derived from published Swiss cancer registry data, or multiply imputed using 3 different imputation models.

Results

24,612 women aged 20–84 had unilateral breast cancer between 1980 and 2016 in the Swiss cantons of Zurich and Zug. Of those, 5 % (n = 1264) were lost to follow-up. Over 291,463 person-years, 1145 contralateral breast malignancies were diagnosed, corresponding to 393 per 100,000 person-years (95 % CI 353–438). Incidence rates have been decreasing over time to 238 (171−333) for the incidence period 2010–2016. The same overall pattern was observed regardless of how we handled missing follow-up times. However, using a single imputation generally produced lower incidence rates compared to the naive approach, with multiple imputation giving higher estimates. The most complex multiple imputation model gave incidence estimates that were very similar to those from the naive approach.

Conclusion

Different methods to handle missing follow-up times yielded similar results: that CBC incidence has declined in recent decades. Multiple imputation is likely an appropriate method to handle missing follow-up data, enabling researchers to include all eligible individuals in the analysis.