Authors' Reply: Clonal Hematopoiesis of Indeterminate Potential and Cardiovascular Events: Issues to Be Further Explored

We thank Wang and Liu¹ for their comments on our recent JASN article.² Collider bias can either inflate or attenuate an effect estimate between a risk factor and an outcome (i.e., clonal hematopoiesis of indeterminate potential [CHIP] and cardiovascular disease) if both the risk factor and outcome affect a stratifying variable (i.e., CKD). The concern for collider bias is applicable to most observational studies of cardiovascular risk factors in patients with CKD. One way to overcome collider bias is to evaluate the risk factor in the whole population, rather than a stratified subgroup, but the opportunity to evaluate the consistency of the signal across subgroups is lost. The effect of CHIP on cardiovascular disease in the general population is well established. Our goal was to assess whether CHIP heightens cardiovascular risk among individuals with CKD, a clinically relevant high-risk subgroup. When the target population is the same as the analysis population, the observed association is informative for prediction and risk stratification even if a collision is present.

If both CHIP and cardiovascular disease increase the likelihood of CKD, the collider bias would be expected to attenuate the observed effect size between CHIP and cardiovascular disease among those with CKD. However, modeling and evaluating collider bias is difficult in the presence of multiple direct and indirect bidirectional causal pathways as in CKD (Figure 1). Hypotheses can be made of collisions leading to both upward and downward biases in effect estimates.

Figure 1.

Collider bias in CKD studies. Causal effects of clonal hematopoiesis of indeterminate potential (CHIP) and cardiovascular disease on CKD (red arrows) could result in collider bias on the effect estimate of CHIP on cardiovascular disease (blue arrow) if participants are stratified by the presence of CKD. However, multiple direct and indirect bidirectional causal relationships make modeling collider bias impractical in CKD, and the effect estimate is still useful for prediction and risk stratification in patients with CKD.

The use of individual-level data from four patient cohorts including prospective observational studies, a randomized control trial, and biobank data is a strength of our analysis. Meta-analysis of effect estimates from each cohort, including sensitivity analyses of eGFR strata, diabetes diagnosis, race, APOL1 and IL6R genotype, follow-up time, presence of baseline cardiovascular disease, and recurrent cardiovascular events was conducted as part of the rigorous JASN review process. We agree with Wang and Liu that further studies to evaluate heterogeneity in causal variants, variant allele fraction, and clonal expansion rate are needed, but they require even larger sample sizes to have adequate power.

Competing risk analysis is vital when evaluating CKD outcomes because patients may die before or simultaneously with kidney failure events. In our study, cardiovascular events were collected even after the onset of kidney failure. A Fine–Gray analysis of a composite of cardiovascular disease and death showed consistent results (results presented in Supplemental Table 2).

Supplementary Material

jasn-36-2314-s001.pdf ^{(1.3MB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/F270.

Author Contributions

Supervision: Tanika N. Kelly, Matthew B. Lanktree.

Writing – original draft: Matthew B. Lanktree.

Writing – review & editing: Tanika N. Kelly, Matthew B. Lanktree, Yang Pan, Cassianne Robinson-Cohen, Caitlyn Vlasschaert.

Funding

None.