Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Multimodal Representation Learning in Paediatric Kidney Disease

Ana Durica, John Booth, Ivana Drobnjak

TL;DR

This study addresses the need for near-term monitoring of renal function in paediatric patients by predicting abnormal serum creatinine within $30$ days using longitudinal EHR data from GOSH. It employs a lightweight temporal representation with a GRU encoder that processes sequences of $100$ time points consisting of 15 laboratory markers (each with presence and abnormality flags) plus demographics, outputting embeddings for a simple $30$-day classifier. Evaluation on a held-out set with a stratified $70/10/20$ split and bootstrap-derived CIs demonstrates good discrimination for the near-term outcome, and embedding visualisation via t-SNE suggests the model captures meaningful predictive structure. The work modularly demonstrates feasibility of temporal modelling in paediatric nephrology and lays groundwork for future multimodal extensions incorporating richer signals and clinically meaningful renal endpoints to enable proactive care.

Abstract

Paediatric kidney disease varies widely in its presentation and progression, which calls for continuous monitoring of renal function. Using electronic health records collected between 2019 and 2025 at Great Ormond Street Hospital, a leading UK paediatric hospital, we explored a temporal modelling approach that integrates longitudinal laboratory sequences with demographic information. A recurrent neural model trained on these data was used to predict whether a child would record an abnormal serum creatinine value within the following thirty days. Framed as a pilot study, this work provides an initial demonstration that simple temporal representations can capture useful patterns in routine paediatric data and lays the groundwork for future multimodal extensions using additional clinical signals and more detailed renal outcomes.

Towards Multimodal Representation Learning in Paediatric Kidney Disease

TL;DR

This study addresses the need for near-term monitoring of renal function in paediatric patients by predicting abnormal serum creatinine within days using longitudinal EHR data from GOSH. It employs a lightweight temporal representation with a GRU encoder that processes sequences of time points consisting of 15 laboratory markers (each with presence and abnormality flags) plus demographics, outputting embeddings for a simple -day classifier. Evaluation on a held-out set with a stratified split and bootstrap-derived CIs demonstrates good discrimination for the near-term outcome, and embedding visualisation via t-SNE suggests the model captures meaningful predictive structure. The work modularly demonstrates feasibility of temporal modelling in paediatric nephrology and lays groundwork for future multimodal extensions incorporating richer signals and clinically meaningful renal endpoints to enable proactive care.

Abstract

Paediatric kidney disease varies widely in its presentation and progression, which calls for continuous monitoring of renal function. Using electronic health records collected between 2019 and 2025 at Great Ormond Street Hospital, a leading UK paediatric hospital, we explored a temporal modelling approach that integrates longitudinal laboratory sequences with demographic information. A recurrent neural model trained on these data was used to predict whether a child would record an abnormal serum creatinine value within the following thirty days. Framed as a pilot study, this work provides an initial demonstration that simple temporal representations can capture useful patterns in routine paediatric data and lays the groundwork for future multimodal extensions using additional clinical signals and more detailed renal outcomes.

Paper Structure

This paper contains 11 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Cohort selection and outcome definition
  4. Data extraction and preprocessing
  5. Sequence modelling
  6. Evaluation
  7. Results
  8. Cohort and temporal characteristics
  9. Predictive performance
  10. Embedding visualisation
  11. Discussion

Figures (3)

  • Figure 1: Overview of longitudinal data used for temporal modelling.
  • Figure 2: Predictive performance of the GRU model. Bootstrapped 95% confidence intervals are shown for both the ROC curve and the confusion-matrix cell counts.
  • Figure 3: t-SNE projection of GRU embeddings for the test set. Each point represents a patient embedding coloured by 30-day outcome.