Thank you for the feedback. Let me clarify my reasoning.
For instance, the difference between sample biomarker values measured at 30 and 40 years has a p-value (30 vs 40) = 0.01. The same biomarker value difference measured at 40 and 50 has a p-value (40 vs 50) = 0.0001. IPR = 0.01/0.0001 = 100. So, greater than 1 and the biomarker is good.
Let’s say in your second example p-value (30 vs 40) = 0.5, and p-value (40 vs 50) = 0.05. But I wouldn’t calculate IPR, because the former p-value is greater than the cut-off 0.05. This makes the biomarker bad. If it were p-value (30 vs 40) = 0.05, the IPR would be 1. That’s how I see it.
The second biomarker (IPR = 1) is okay, but not as good as the first one (IPR = 100).
Effect size generally correlates with p-value. And the use of non-parametric tests would make what I propose more robust.