How can I communicate hazard ratios from log-transformed biomarkers in a clinically interpretable way?

An alternative would be to recognize the log-transformed quantity by its own name. Consider the pH, which is familiar to everybody in medicine.