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

Handling missing disease information due to death in diseases that need two visits to diagnose

In studies that assess disease status periodically, time of disease onset is interval censored between visits. Participants who die between two visits may have unknown disease status after their last visit. In this work, we consider an additional scenario where diagnosis requires two consecutive positive tests, such that disease status can also be unknown at the last visit preceding death. We show that this impacts the choice of censoring time for those who die without an observed disease diagnosis. We investigate two classes of models that quantify the effect of risk factors on disease outcome: a Cox proportional hazards model with death as a competing risk and an illness death model that treats disease as a possible intermediate state. We also consider four censoring strategies: participants without observed disease are censored at death (Cox model only), the last visit, the last visit with a negative test, or the second last visit. We evaluate the performance of model and censoring strategy combinations on simulated data with a binary risk factor and illustrate with a real data application. We find that the illness death model with censoring at the second last visit shows the best performance in all simulation settings. Other combinations show bias that varies in magnitude and direction depending on the differential mortality between diseased and disease-free subjects, the gap between visits, and the choice of the censoring time.

Estimation of mortality rate ratios for chronic conditions with misclassification of disease status at death

Estimation of mortality rates and mortality rate ratios (MRR) of diseased and non-diseased individuals is a core metric of disease impact used in chronic disease epidemiology. Estimation of mortality rates is often conducted through retrospective linkage of information from nationwide surveys such as the National Health Interview Survey (NHIS) and death registries. These surveys usually collect information on disease status during only one study visit. This infrequency leads to missing disease information (with right censored survival times) for deceased individuals who were disease-free at study participation, and a possibly biased estimation of the MRR because of possible undetected disease onset after study participation. This occurrence is called "misclassification of disease status at death (MicDaD)" and it is a potentially common source of bias in epidemiologic studies. In this study, we conducted a simulation analysis with a high and a low incidence setting to assess the extent of MicDaD-bias in the estimated mortality. For the simulated populations, MRR for diseased and non-diseased individuals with and without MicDaD were calculated and compared. Magnitude of MicDaD-bias depends on and is driven by the incidence of the chronic disease under consideration; our analysis revealed a noticeable shift towards underestimation for high incidences when MicDaD is present. Impact of MicDaD was smaller for lower incidence (but associated with greater uncertainty in the estimation of MRR in general). Further research can consider the amount of missing information and potential influencers such as duration and risk factors of the disease.

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

BACKGROUND

The increased stroke risk associated with atrial fibrillation (AF) burden exceeding 5 min is a matter of debate. In addition, the potential linear or nonlinear relationship between AF burden and stroke risk has been largely unexplored.

AIM

To determine the association between AF burden > 5 min and the increased risk of stroke and explore the potential dose-response relationship between these two factors.

METHODS

Sixteen studies from six databases with 53141 subjects (mean age 65 years) were included. Fifteen studies were observational studies, and one was a randomized controlled trial study. The potential nonlinear dose-response association was characterized using a restricted cubic splines regression model. AF burden for each 1 h and 2 h was associated with an increased risk of stroke. Trial sequential analysis with a random-effect model was used to evaluate the robustness of the evidence from the included 16 studies.

RESULTS

AF burden > 5 min was associated with an increased risk of clinical AF [adjusted risk ratio (RR) = 4.18, 95% confidence interval (CI): 2.26-7.74]. However, no association was found with an increased risk of all-cause mortality (adjusted RR = 1.55, 95%CI: 0.87-2.75). Patients with AF burden > 5 min had an increased risk of stroke (adjusted RR = 2.49, 95%CI: 1.79-3.47). Moreover, a dose-response analysis showed that the increased stroke risk was paralleled by an increase in AF burden at a rate of 2.0% per hour (P_nonlinear = 0.656, RR = 1.02, 95%CI: 1.01-1.03). Trial sequential analysis provided robust evidence of the association between AF burden > 5 min and an increased risk of stroke.

CONCLUSION

AF burden was a significant risk factor for clinical AF and future stroke. A significant linear association was documented between increased AF burden and risk of future stroke.

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

Aggregated data about the prevalence and incidence of chronic conditions is becoming more and more available. We recently proposed a method to estimate the age-specific excess mortality in chronic conditions from aggregated age-specific prevalence and incidence data. Previous works showed that in age groups below 50 years, estimates from this method were unstable or implausible. In this article, we examine how limited diagnostic accuracy in terms of sensitivity and specificity affects the estimates. We use a simulation study with two settings, a low and a high prevalence setting, and assess the relative importance of sensitivity and specificity. It turns out that in both settings, specificity, especially in the younger age groups, dominates the quality of the estimated excess mortality. The findings are applied to aggregated claims data comprising the diagnoses of diabetes from about 35 million men in the German Statutory Health Insurance. Key finding is that specificity in the lower age groups (<50 years) can be derived without knowing the sensitivity. The false-positive ratio in the claims data increases linearly from 0.5 per mil at age 25 to 2 per mil at age 50. As a conclusion, our findings stress the importance of considering diagnostic accuracy when estimating excess mortality from aggregated data using the method to estimate excess mortality. Especially the specificity in the younger age-groups should be carefully taken into account.

A multi-state model based reanalysis of the Framingham Heart Study: Is dementia incidence really declining?

Recent research by Satizabal and colleagues using data from the Framingham Heart Study demonstrated a linear decline in dementia incidence since the 1970s. The aim of this study is to re-examine these findings, given concerns that bias resulted from failure to account for the probability of acquiring dementia between the last dementia-free observation and death. This analysis included 5118 persons 60+ years of age, and determined the 5-year dementia incidence during four non-overlapping epochs. In addition to a replication using Cox proportional hazards, we applied separate Cox models (given unequal hazards across epochs) and a Spline-based penalized likelihood approach based on the illness-death multi-state model. In addition, we present a simulation study demonstrating the bias associated with the use of standard survival models. The simulation showed that estimates of disease incidence derived from the multi-state model-based approach were consistent with the true disease incidence, whereas Cox regression 'censoring' observations at death or at last observation consistently underestimated it. Using the Framingham data, the 5-year age- and sex-adjusted cumulative hazard rates for dementia as derived from the multi-state model-based approach were 3.84, 2.66, 3.29 and 3.13 per 100 persons in epochs 1, 2, 3 and 4 respectively. The findings do not support the conclusion that dementia incidence has declined in the Framingham Heart Study over the given time period. Previous findings of a decline may have been an artefact resulting from improper treatment of those cases in which death precluded the observation of dementia onset.