Background: Identifying novel risk markers in cardiovascular patients remains a research priority. Longer follow-up generally is considered favorable in such studies, but associations of interest may become attenuated with increasing follow-up. This issue has not been adequately addressed in the context of patient cohorts. The current study analyzed the extent and mechanisms of attenuating associations in a cardiovascular patient cohort.
Methods: The associations of numerous biomarkers with all-cause mortality were estimated by multiple Cox regression in the Langzeiterfolge der KARdiOLogischen Anschlussheilbehandlung (KAROLA) prospective cohort study of 1204 patients who had participated in an inpatient rehabilitation program after an acute coronary syndrome (ACS) or coronary bypass operation. Hazard ratios were estimated based on the entire follow-up period (13 years), and after truncation at previous follow-up times (3, 4.5, 6, 8, 10 years).
Results: For the majority of markers, a clear and sometimes very pronounced attenuation of the hazard ratios could be observed with increasing follow-up duration. Differential attrition generally was not a sufficient explanation for this phenomenon, whereas further analyses suggested a role for reverse causality for some of the markers. Power analyses showed that the relationship of follow-up duration and statistical power can be counterintuitive in the presence of realistic amounts of attenuation.
Conclusions: The attenuation of estimates of association in patient cohorts is a much more substantial and complex issue than currently appreciated. This has important implications for the design and interpretation of prognostic, as well as etiologic, studies which may be particularly relevant in the case of patient cohorts defined by an initial acute event.
- Received for publication July 6, 2016.
- Accepted for publication August 18, 2016.
- © 2016 American Association for Clinical Chemistry