Fehlende Information aufgrund von Tod bei Ereigniszeitanalysen von klinischen und epidemiologischen Studien
Zusammenfassung der Projektergebnisse
In most clinical and epidemiological studies information on disease status is collected at regular follow-up times. Often, this information can only be retrieved in those individuals who are alive at follow-up resulting in data with missing disease information due to death (MDID) for individuals who died before follow-up. The resulting complications are of particular relevance in long-term studies and when studying elderly populations. There are many examples in the biomedical literature that MDID is not adequately dealt with: individuals are censored at death or last follow-up or are even excluded from the analysis. In this project, we showed that such naive analyses can lead to serious bias in incidence estimates that is then transferred to bias in estimates of hazard ratios corresponding to risk or prognostic factors. For roughly judging magnitude and direction of MDID bias we developed a formula based on particular study characteristics. We conducted a systematic literature review of major epidemiological and geriatric journals leading to an estimate of 58 out of 125 identified studies (46%) that were considered susceptible to MDID bias. For an adequate analysis, we developed a comprehensive statistical analysis strategy based on an illness-death multistate model allowing for an explicit consideration of MDID and interval censoring in such studies and investigated various variants of statistical techniques extensively in simulated data as well as in real data from published studies. We found that MDID constitutes a particular problem in the current debate on the declining incidence of dementia over the last 30 to 40 years. Based on a recently conducted re-analysis of the Framingham Heart study we suspect that the originally observed decline can at least partially be attributed to MDID bias.
Projektbezogene Publikationen (Auswahl)
- Missing information caused by death leads to bias in relative risk estimates. J Clin Epidemiol. 2014;67(10):1111-20
Binder N, Schumacher M.
(Siehe online unter https://doi.org/10.1016/j.jclinepi.2014.05.010) - Re: "Incidence of dementia among participants and nonparticipants in a longitudinal study of cognitive aging". Am J Epidemiol. 2015;181(4):291-2
Binder N, Schumacher M, Joly P.
(Siehe online unter https://doi.org/10.1093/aje/kwu475) - Incidence of Dementia over Three Decades in the Framingham Heart Study. N Engl J Med. 2016 Jul 7;375(1):92-3
Binder N, Schumacher M
(Siehe online unter https://doi.org/10.1056/NEJMoa1504327) - Perils and potentials of self-selected entry to epidemiological studies and surveys. Journal of the Royal Statistical Society: Series A (Statistics in Society) 2016; 179:319–376
Schumacher M, Beyersmann J, Binder N. Discussion on Keiding, N. and Louis, T.A.
(Siehe online unter https://doi.org/10.1111/rssa.12136) - Estimating hazard ratios in cohort data with missing disease information due to death. Biom J. 2017;59(2):251-269
Binder N, Herrnböck AS, Schumacher M
(Siehe online unter https://doi.org/10.1002/bimj.201500167) - The combined association of alcohol consumption with dementia risk is likely biased due to lacking account of death cases. Eur J Epidemiol. 2017;32(7):627-629
Binder N, Manderscheid L, Schumacher M.
(Siehe online unter https://doi.org/10.1007/s10654-017-0252-0) - Cohort studies were found to be frequently biased by missing disease information due to death. J Clin Epidemiol. 2019;105:68-79
Binder N, Blümle A, Balmford J, Motschall E, Oeller P, Schumacher M.
(Siehe online unter https://doi.org/10.1016/j.jclinepi.2018.09.010)
