Estimating hazard ratios in cohort data with missing disease information due to death

Abstract

In clinical and epidemiological studies information on the primary outcome of interest, that is, the disease status, is usually collected at a limited number of follow-up visits. The disease status can often only be retrieved retrospectively in individuals who are alive at follow-up, but will be missing for those who died before. Right-censoring the death cases at the last visit (ad-hoc analysis) yields biased hazard ratio estimates of a potential risk factor, and the bias can be substantial and occur in either direction. In this work, we investigate three different approaches that use the same likelihood contributions derived from an illness-death multistate model in order to more adequately estimate the hazard ratio by including the death cases into the analysis: a parametric approach, a penalized likelihood approach, and an imputation-based approach. We investigate to which extent these approaches allow for an unbiased regression analysis by evaluating their performance in simulation studies and on a real data example. In doing so, we use the full cohort with complete illness-death data as reference and artificially induce missing information due to death by setting discrete follow-up visits. Compared to an ad-hoc analysis, all considered approaches provide less biased or even unbiased results, depending on the situation studied. In the real data example, the parametric approach is seen to be too restrictive, whereas the imputation-based approach could almost reconstruct the original event history information.

Atrial fibrillation burden and the risk of stroke: A systematic review and dose-response meta-analysis

“…The heterogeneity might be associated with the variations in patient populations, hypertension, prior AF and antithrombotic therapy, etc . [ 33 ]. The population included in our study had different comorbidities, including patients with symptomatic atrial tachyarrhythmias[ 7 , 20 ], sinus node disease[ 19 ] and heart failure[ 22 ].…”

Impact of diagnostic accuracy on the estimation of excess mortality from incidence and prevalence: simulation study and application to diabetes in German men

“…Strictly speaking, the estimates of R from [ Scheidt-Nave 2019 ] might have undergone diagnostic inaccuracies as well. However, the estimates are based on individual data (ID) and potential biases in ID analyses (e.g., by missing disease status at death [ Binder et al , 2017 ]), are beyond the scope of this article. Thus, for simplicity we assume R = R (obs) .…”

Failure time studies with intermittent observation and losses to follow‐up

“…We stress that such registries usually involve random reporting delays and that information about the distribution of delays is needed to avoid potential bias. A number of authors, including Binder and Schumacher (2014) and Binder et al (2017) , (2019) have considered bias in the estimation of failure time models caused by death being treated as independent LTF. There have not been thorough investigations of the effects of reporting delays, however, and we are currently undertaking this.…”

“…However, aside from aforementioned studies on how to define LTF times, there has been very little study of settings with intermittent observation. Exceptions include a brief investigation of bias in standard estimates (Lawless, 2013) and studies of the special situation where LTF corresponds to death, for which exact times are ascertainable (e.g., Binder et al, 2019; Binder, Hernböck, & Schumacher, 2017; Binder & Schumacher, 2014; Joly, Commenges, Helmer, & Letenneur, 2002). We focus here on situations where exact LTF times are not generally ascertainable.…”