# Optimal method for including repeated measures in prediction model

**URL:** <https://discourse.datamethods.org/t/optimal-method-for-including-repeated-measures-in-prediction-model/5122>\
**Category:** modeling strategy\
**Tags:** longitudinal\
**Created:** [November 23, 2021, 1:57pm UTC](https://discourse.datamethods.org/t/optimal-method-for-including-repeated-measures-in-prediction-model/5122 "2021-11-23T13:57:38Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![trumanfrancis](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/trumanfrancis/32/47_2.png) [@trumanfrancis](https://discourse.datamethods.org/u/trumanfrancis)\
**Post date:** [November 23, 2021, 1:57pm UTC](https://discourse.datamethods.org/t/optimal-method-for-including-repeated-measures-in-prediction-model/5122/1 "2021-11-23T13:57:38Z")

</div>

What is the optimal method for including repeated measures of a biomarker in a prediction model…particularly if the time of measurement is not consistent between participants. I read this article but am still uncertain. [Harnessing repeated measurements of predictor variables for clinical risk prediction: a review of existing methods | Diagnostic and Prognostic Research | Full Text (biomedcentral.com)](https://diagnprognres.biomedcentral.com/articles/10.1186/s41512-020-00078-z)

---

<div class="post-metadata">

**Author:** ![davidcnorrismd](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/davidcnorrismd/32/3502_2.png) [@davidcnorrismd](https://discourse.datamethods.org/u/davidcnorrismd)\
**Post date:** [November 24, 2021, 9:43am UTC](https://discourse.datamethods.org/t/optimal-method-for-including-repeated-measures-in-prediction-model/5122/2 "2021-11-24T09:43:48Z")

</div>

If a formal approach is desired, I think [state-space modeling](https://discourse.datamethods.org/t/special-vs-markov-models-for-analyzing-within-patient-variability/4546/3) is ideal. Note in particular that your problem (irregular timing of measurements) arises only from the customary statistician’s habit of treating the data as the [_ding an sich_](https://en.wikipedia.org/wiki/Thing-in-itself), rather than a mere phenomenon. The state-space approach, by contrast, posits an underlying _state process_ (generally, multivariate) that evolves (generally) in continuous time, and gives rise to the data via a _measurement process_ that is distinct from the underlying state process.