Band 8, Heft 1 -

Interrupted time series are increasingly being used to evaluate the population-wide implementation of public health interventions. However, the resulting estimates of intervention impact can be severely biased if underlying disease trends are not adequately accounted for. Control series offer a potential solution to this problem, but there is little guidance on how to use them to produce trend-adjusted estimates. To address this lack of guidance, we show how interrupted time series can be analysed when the control and intervention series share confounders, i. e. when they share a common trend. We show that the intervention effect can be estimated by subtracting the control series from the intervention series and analysing the difference using linear regression or, if a log-linear model is assumed, by including the control series as an offset in a Poisson regression with robust standard errors. The methods are illustrated with two examples.

Instrumental variables is a popular method in epidemiology and related fields, to estimate causal effects in the presence of unmeasured confounding. Traditionally, instrumental variable analyses have been confined to linear models, in which the causal parameter of interest is typically estimated with two-stage least squares. Recently, the methodology has been extended in several directions, including two-stage estimation and so-called G-estimation in nonlinear (e. g. logistic and Cox proportional hazards) models. This paper presents a new R package, ivtools , which implements many of these new instrumental variable methods. We briefly review the theory of two-stage estimation and G-estimation, and illustrate the functionality of the ivtools package by analyzing publicly available data from a cohort study on vitamin D and mortality.

Researchers interested in the effects of exposure spikes on an outcome need tools to identify unexpectedly high values in a time series. However, the best method to identify spikes in time series is not known. This paper aims to fill this gap by testing the performance of several spike detection methods in a simulation setting. We created simulations parameterized by monthly violence rates in nine California cities that represented different series features, and randomly inserted spikes into the series. We then compared the ability to detect spikes of the following methods: ARIMA modeling, Kalman filtering and smoothing, wavelet modeling with soft thresholding, and an iterative outlier detection method. We varied the magnitude of spikes from 10 to 50 % of the mean rate over the study period and varied the number of spikes inserted from 1 to 10. We assessed performance of each method using sensitivity and specificity. The Kalman filtering and smoothing procedure had the best overall performance. We applied each method to the monthly violence rates in nine California cities and identified spikes in the rate over the 2005–2012 period.

In studies of clinical phenotypes, such as dementia, disability, and frailty, participants are typically assessed at in-person clinic visits. Thus, the precise time of onset for the phenotype is unknown. The discreteness of the clinic visits yields grouped event time data. We investigate how to perform a risk factor analysis in the case of grouped data. Since visits can be months to years apart, numbers of ties can be large, causing the exact tie-handling method of the Cox model to be computationally infeasible. We propose two, new, computationally efficient approximations to the exact method: Laplace approximation and an analytic approximation. Through extensive simulation studies, we compare these new methods to the Prentice-Gloeckler model and the Cox model using Efron’s and Breslow’s tie-handling methods. In addition, we compare the methods in an application to a large cohort study (N = 3,605) on the development of clinical frailty in older adults. In our simulations, the Laplace approximation has low bias in all settings, and the analytic approximation has low bias in settings where the regression coefficient is not large in magnitude. Their corresponding confidence intervals also have approximately the nominal coverage probability. In the data application, the results from the approximations are nearly identical to that of the Prentice-Gloeckler model.

Suppose we are interested in the effect of variable X on variable Y . If X and Y both influence, or are associated with variables that influence, a common outcome, called a collider , then conditioning on the collider (or on a variable influenced by the collider – its “child”) induces a spurious association between X and Y , which is known as collider bias. Characterizing the magnitude and direction of collider bias is crucial for understanding the implications of selection bias and for adjudicating decisions about whether to control for variables that are known to be associated with both exposure and outcome but could be either confounders or colliders. Considering a class of situations where all variables are binary, and where X and Y either are, or are respectively influenced by, two marginally independent causes of a collider, we derive collider bias that results from (i) conditioning on specific levels of the collider or its child (on the covariance, risk difference, and in two cases odds ratio, scales), or (ii) linear regression adjustment for, the collider or its child. We also derive simple conditions that determine the sign of such bias.

Mediation analysis is popular in examining the extent to which the effect of an exposure on an outcome is through an intermediate variable. When the exposure is subject to misclassification, the effects estimated can be severely biased. In this paper, when the mediator is binary, we first study the bias on traditional direct and indirect effect estimates in the presence of conditional non-differential misclassification of a binary exposure. We show that in the absence of interaction, the misclassification of the exposure will bias the direct effect towards the null but can bias the indirect effect in either direction. We then develop an EM algorithm approach to correcting for the misclassification, and conduct simulation studies to assess the performance of the correction approach. Finally, we apply the approach to National Center for Health Statistics birth certificate data to study the effect of smoking status on the preterm birth mediated through pre-eclampsia.