Scholarlink

<p>Real-world data (RWD) from electronic health records and digital health databases present unique opportunities to study causal effects in healthcare. While Difference-in-Differences (DiD) analysis is widely used for such analyses, it can be biased when time-varying unmeasured confounding violates the parallel trends assumption. We propose a negative control-calibrated difference-in-difference (NC-DiD) approach that uses negative control outcomes (NCOs) both before and after the intervention to detect and adjust for such confounding. The method remains robust even with partially unreliable controls. In simulations, NC-DiD reduces bias, controls type-I error, and improves estimation accuracy. We applied NC-DiD to assess racial/ethnic disparities in post COVID-19 health outcomes using RWD emulated from pediatric 15,373 patients across eight children’s hospitals. Results revealed worse long-term outcomes for minority groups compared to Non-Hispanic White patients. NC-DiD offers a robust framework for deriving reliable causal insights from digital health data, supporting evidence-based clinical decision-making and potentially improving patient outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Negative control-calibrated difference-in-difference analyses: addressing unmeasured confounding in RWD with application to racial/ethnic differences

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月17日

查看全文
  • Dazheng Zhang,
  • Bingyu Zhang,
  • Huiyuan Wang,
  • Yiwen Lu,
  • Charles J. Wolock,
  • Wenjie Hu,
  • Linbo Wang,
  • George Hripcsak,
  • Yong Chen

摘要

Real-world data (RWD) from electronic health records and digital health databases present unique opportunities to study causal effects in healthcare. While Difference-in-Differences (DiD) analysis is widely used for such analyses, it can be biased when time-varying unmeasured confounding violates the parallel trends assumption. We propose a negative control-calibrated difference-in-difference (NC-DiD) approach that uses negative control outcomes (NCOs) both before and after the intervention to detect and adjust for such confounding. The method remains robust even with partially unreliable controls. In simulations, NC-DiD reduces bias, controls type-I error, and improves estimation accuracy. We applied NC-DiD to assess racial/ethnic disparities in post COVID-19 health outcomes using RWD emulated from pediatric 15,373 patients across eight children’s hospitals. Results revealed worse long-term outcomes for minority groups compared to Non-Hispanic White patients. NC-DiD offers a robust framework for deriving reliable causal insights from digital health data, supporting evidence-based clinical decision-making and potentially improving patient outcomes.