# Clinically Relevant Risk Factors - Survival Analysis

**URL:** <https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860>\
**Category:** general\
**Created:** [August 9, 2022, 2:11pm UTC](https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860 "2022-08-09T14:11:17Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![Vitaly\_Druker](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/vitaly_druker/32/192_2.png) [@Vitaly\_Druker](https://discourse.datamethods.org/u/Vitaly_Druker)\
**Post date:** [August 9, 2022, 2:11pm UTC](https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860/1 "2022-08-09T14:11:17Z")

</div>

I’m currently working with some clinicians on exploring the role of a specific risk factor for survival in an oncology trial. I know there has been discussion about clinically meaningful effects for treatments, but I haven’t seen any discussion about a similar concept for risk factors.

Has anyone come across any guidance on how to choose/define a clinically meaningful impact of a risk factor in a model? I think one of the main additional considerations would be prevalence or variability of the risk factor.

I am using methods similar to those described [here](https://discourse.datamethods.org/t/statistically-efficient-ways-to-quantify-added-predictive-value-of-new-measurements/2013) to actually see if the predictor is ‘useful’, i.e, improves the model. I plan on using the R^2 to describe improvement in the model, but it’s not clear how to turn that into a clinically meaningful measure.

Specifically, there is a list of prognostic markers that are most likely relevant in non small cell lung cancer:

1. Age
2. PDL1
3. Smoking status etc.

We want to see how useful a new biomarker (for example TMB - a continuous variable) is in predicting survival. I’ll be able to do that using a likelihood ratio test. However, how could you communicate the additional usefulness of this variable in a risk model?

Thank you in advance

---

<div class="post-metadata">

**Author:** ![f2harrell](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/f2harrell/32/165_2.png) [@f2harrell](https://discourse.datamethods.org/u/f2harrell)\
**Post date:** [August 9, 2022, 3:47pm UTC](https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860/2 "2022-08-09T15:47:50Z")

</div>

Not sure if this is what you want but take a look at [Statistically Efficient Ways to Quantify Added Predictive Value of New Measurements | Statistical Thinking](http://fharrell.com/post/addvalue)

---

<div class="post-metadata">

**Author:** ![Vitaly\_Druker](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/vitaly_druker/32/192_2.png) [@Vitaly\_Druker](https://discourse.datamethods.org/u/Vitaly_Druker)\
**Post date:** [August 9, 2022, 4:48pm UTC](https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860/3 "2022-08-09T16:48:43Z")

</div>

Thank you - this is definitely down the right path. The publication you link to [Fronkzec 2021](https://ccforum.biomedcentral.com/articles/10.1186/s13054-021-03632-3) is helpful. Have you come across anything that does this for survival analyses? Perhaps I can use something like RMST at a fixed value or something like that.

---

<div class="post-metadata">

**Author:** ![f2harrell](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/f2harrell/32/165_2.png) [@f2harrell](https://discourse.datamethods.org/u/f2harrell)\
**Post date:** [August 10, 2022, 12:06am UTC](https://discourse.datamethods.org/t/clinically-relevant-risk-factors-survival-analysis/5860/4 "2022-08-10T00:06:11Z")

</div>

Most of those measures work quite generally including for censored Y.