{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihej5be4y3dvfmqtl3bkrdefdwda6qwjvr6anqyn3cn45dg7hqomm",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3milhswh76vc2"
  },
  "path": "/t/how-can-i-communicate-hazard-ratios-from-log-transformed-biomarkers-in-a-clinically-interpretable-way/28692#post_1",
  "publishedAt": "2026-04-02T23:51:59.000Z",
  "site": "https://discourse.datamethods.org",
  "textContent": "Dear all experts,\n\nI am working with a Cox proportional hazards model where a biomarker has been log-transformed, and I am trying to interpret the hazard ratio in a way that is intuitive for non-statisticians.\n\nFor example, if the hazard ratio is 1.2, the standard interpretation is that a one-unit increase in the log-transformed biomarker is associated with a 20% increase in the hazard of the event (e.g. healing).\n\nHowever, this is difficult to communicate because a “one-unit increase” on the log scale does not translate easily into changes on the original biomarker scale.\n\nTo improve interpretability, I have plotted the estimated partial hazard (not account for baseline hazard) against the original biomarker values for different HR values. Based on the plot shown below, my interpretation is too technical (in my opinion): The plot reflects the effect of the log transformation on the hazard: the same absolute change in biomarker level results in a larger hazard difference when it occurs at lower values, where the curve is steep, than at higher values, where the curve is flat. The direction of this contrast, specifically whether lower biomarker values are associated with higher or lower hazard, is determined by whether the model HR is less than or greater than 1. As expected, the steepness of each curve, and therefore the magnitude of the hazard contrast for a given absolute difference, depends on how far the model HR deviates from the null value of 1.\n\n**My question is** : Are there better or more interpretable ways to express the effect of a log-transformed biomarker on the hazard, preferably in terms of changes in the original biomarker values, rather than a one-unit change on the log scale?\n\nThank you so much in advance.",
  "title": "How can I communicate hazard ratios from log-transformed biomarkers in a clinically interpretable way?"
}