How would you interpret a large fold change with a wide confidence interval?

Why

Suppose a gene has an estimated log2 fold change of +3, corresponding to eightfold higher expression, but its interval stretches from approximately zero to a much larger increase. That result can be compatible with minimal change and with a substantial response, so the large point estimate alone is a poor ranking criterion. Sparse genes, one extreme sample, a small number of independent donors or unstable dispersion estimates often contribute uncertainty.

Inspect per-sample normalized counts and QC, consider prespecified low-count filtering, and review whether a covariate or a batch drives the signal. An empirical Bayes shrinkage estimator may stabilize estimates for ranking while retaining an explicit account of uncertainty, but it is not evidence that the gene effect became biologically real. Explain the interval on the log scale or transform to fold changes when appropriate.

Interpret effect size jointly with interval, multiplicity adjustment and biological plausibility.