It's reasonably consistently measurable and has demonstrated correlations with both causative factors and outcomes. You can legitimately question the nature of what is being measured, but it's obviously something real (even if it's a composite).