Confounding control in high-throughput studies – principles and applications

Background: High throughput studies that can simultaneously examine a myriad of exposures or outcomes are becoming increasingly prominent in real-world data analyses. Inherent in their desired large-scale processing are also limitations that may impact control of confounding. In most cases, the aims of high throughput studies are to screen as a preliminary step to a more nuanced study. Finding the optimal balance of “good enough” confounding control to implement high throughput analyses without sacrificing too much speed or rigor is the goal. Semi-automated tools can be key catalysts to achieving that balance.

Objectives: This symposium will discuss several semi-automated methods to address confounding in high-throughput study designs. It will be particularly beneficial to researchers involved or interested in applying high throughput approaches for signal detection in large, electronic health databases.

Description: The symposium will feature 5 speakers from academia and industry in North America and Europe who will share their experience with a broad array of approaches to address confounding in high-throughput scenarios. The session will feature a discussion of salient challenges in high-throughput screening studies, such as the limited ability to maximize restriction (e.g., selection of active comparators or customized exclusions for specific clinical scenarios) (AG Winterstein). This will be followed by 4 presentations on distinct approaches to address these challenges, including automated selection of inclusion criteria (J Maro) or active comparators (J Bohn) to facilitate restriction, use of negative controls to quantify confounding (R Desai), and use of self-controlled designs to reduce confounding (T Miano). The symposium will close with a panel discussion on the advantages and limitations of these and other semi-automated approaches in contrast to expert informed designs to support causal inferences (moderator J Gagne).
1. The problem: when restriction by design or automated confounder selection are not sufficient (A Winterstein)
2. Solution 1: Restriction – automated selection of inclusion criteria in high throughput studies (J Maro)
3. Solution 2: Comparator selection – automated screening of active comparator candidates for high throughput studies (J Bohn)
4. Solution 3: Self-control designs in high throughput studies (T Miano)
5. Solution 4: Negative control variables identified in a semi-automated way to assess and address unmeasured confounding (R Desai)