Local causal discovery in epidemiology: an application to quantifying the effect of diabetes on severe liver fibrosis in patients with viral hepatitis
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Assessing the controlled direct effect (CDE) of an exposure on a disease from observational data is challenging when the DAG is unknown. In such cases, causal discovery methods can be used to infer a partiallly oriented DAG from which potential adjustment sets for effect estimation can be deduced. How-ever, classical causal discovery approaches that attempt to reconstruct the entire graph often face important limitations, including strong assumptions, numerous statistical tests, and large sample size requirements, which are further exacerbated by missing data. To address these issues, we rely on a local causal discovery algorithm, which focuses on identifying only the relevant part of the graph. We apply this approach to cohort data from patients with viral hepatitis to estimate the CDE of diabetes on severe liver fibrosis.
Methods
We investigated the CDE of diabetes on liver fibrosis among patients infected with HBV or HCV using baseline data from the French ANRS CO22 HEPATHER cohort. We used a local causal discovery algorithm, called, LocalPC-CDE. This method focuses on learning only the part of the graph necessary to answer the target causal question, thereby avoiding some unrealistic assumptions, reducing the number of required tests, and automatically selecting the adjustment set for valid CDE estimation. To further enhance robustness, we incorporated a bootstrap augmentation of the algorithm to quantify the stability of each identified direct cause (variables of the adjustment set). Only the variables least affected by sampling fluctuations were retained in the final adjustment set. When the CDE was identifiable, it was estimated using logistic regression, yielding odds ratios (OR) with 95% confidence intervals (CI).
Results
A total of 20858 patients were included for causal discovery, and the estimation was performed on 8,130 complete-case observations. The set of direct causes of severe fibrosis (excluding diabetes) identified by the causal discovery algorithm comprised 9 variables: diabetes, geographical origin, age, type of hepatitis, total cholesterol, HDL cholesterol, past alcohol consumption, blood glucose, and sex, which constituted the adjustment set. The CDE of diabetes on severe fibrosis among patients with viral hepatitis was significantly positive, with an estimated odds ratio of 1.63 (95% CI [1.40, 1.90]).
Conclusions
Even after adjusting for confounding using a targeted, data-driven approach, diabetes still appears to have a direct and statistically significant effect on liver fibrosis in patients with chronic viral hepatitis. This suggests that managing diabetes may be a direct lever for mitigating fibrosis progression in patients with chronic viral hepatitis, not just an indirect one through other metabolic and lifestyle factors.
