Standardizing to specific target populations in distributed networks and multisite pharmacoepidemiologic studies

Abstract

Distributed network studies and multisite studies assess drug safety and effectiveness in diverse populations by pooling information. Targeting groups of clinical or policy interest (including specific sites or site combinations) and applying weights based on effect measure modifiers (EMMs) prior to pooling estimates within multisite studies may increase interpretability and improve precision. We simulated a 4-site study, standardized each site using inverse odds weights (IOWs) to resemble the 3 smallest sites or the smallest site, estimated IOW-weighted risk differences (RDs), and combined estimates with inverse variance weights (IVWs). We also created an artificial distributed network in the Clinical Practice Research Datalink (CPRD) Aurum consisting of 1 site for each geographic region. We compared metformin and sulfonylurea initiators with respect to mortality, targeting the smallest region. In the simulation, IOWs reduced differences between estimates and increased precision when targeting the 3 smallest sites or the smallest site. In the CPRD Aurum study, the IOW + IVW estimate was also more precise (smallest region: RD = 5.41% [95% CI, 1.03-9.79]; IOW + IVW estimate: RD = 3.25% [95% CI, 3.07-3.43]). When performing pharmacoepidemiologic research in distributed networks or multisite studies in the presence of EMMs, designation of target populations has the potential to improve estimate precision and interpretability. This article is part of a Special Collection on Pharmacoepidemiology.

Keywords: distributed networks; external validity; standardization; target populations.

Figure 1

A graphical representation of the structure of multisite studies and distributed networks.

Figure 2

Directed acyclic graphs representing the causal structure in 4 simulated nodes included in a hypothetical distributed network. Panel A shows the first 3 sites, and panel B shows the fourth site. Solid arrows are causal effects that are present in every simulation scenario, while dashed arrows are limited to scenarios where treatment is not randomized.

Figure 3

Treatment effect estimates (diamonds) across the 4 nodes in a simulation study with a confounded treatment effect. Panel A shows the estimates obtained in the absence of heterogeneity, while panel B shows estimates in the presence of heterogeneity. Bars show 95% CIs. IPTW, inverse probability of treatment weighting; RD, risk difference.

Figure 4

Risk differences (RDs) obtained when estimating various treatment effects using data from 4 simulated nodes and inverse probability of treatment weighting (IPTW). Blue diamonds represent the estimates within each node, green diamonds represent the estimates when standardizing each node to the covariates of the 3 smallest nodes using inverse odds weights (IOW), and yellow diamonds represent a gold standard analysis of those 3 nodes directly. Bars show 95% CIs.

Figure 5

Risk differences (RDs) in various target populations among 4 simulated nodes estimated using various different methodologies. The blue diamond is the result of combining estimates with inverse variance weighting (IVW), the green diamonds are the result of combining estimates with IVW after targeting a specific population with inverse odds weights (IOW), and the gold diamonds are the result of analyzing those targets directly. Bars show 95% CIs.

Figure 6

Risk difference (RD) estimates for the effect of initiating use of metformin versus sulfonylurea on 1-year mortality obtained within each region of CPRD Aurum using inverse probability of treatment weights (IPTW) (A) and when combining IPTW with inverse odds weights (IOW) to target the region 11 subpopulation (B). The dashed line represents the summary estimate obtained from applying inverse variance weights to the IPTW estimates from each region without the use of any IOWs. Bars show 95% CIs. CPRD, Clinical Practice Research Datalink.