ViSTA improves clinical prediction from patient time series data
ViSTA: A Simple Bridge Extends Visual Alignment to Clinical Time-Series Understanding in Multimodal LLMs
Computation and LanguageArtificial Intelligence
Summary
Doctors often need to predict patient outcomes based on measurements taken over time, but existing language models aren’t great at this kind of numerical data. The authors created ViSTA, a small add-on for visual language models that helps them understand these changing numbers better without retraining the entire system. ViSTA boosts prediction accuracy for kidney injury and death risk, and also helps answer questions about a patient’s ongoing condition more efficiently than previous methods.
What this means in practice
- •For hospital data teams: Improve risk prediction models for patient outcomes using irregular clinical measurements integrated into visual language models.
- •For healthcare software developers: Build medical question-answering tools that understand patient time-series data with fewer trainable parameters and higher efficiency.
Authors
Junyi Gao, Yu Shi, Pingzhao Hu, Ewen M Harrison
Abstract
Clinical prediction models estimate risk from patient measurements, while large language models support medical text understanding and question answering. Yet their language capabilities do not ensure accurate prediction from structured, high-dimensional clinical time series. Improving this ability would connect risk estimation with flexible questions about a patient's evolving condition. We introduce ViSTA, a compact adapter that incorporates irregular numerical measurements into a pretrained vision-language model's chart representations. It learns corrections to visual tokens while leaving all pretrained parameters unchanged. On MIMIC-IV, ViSTA has the highest mean scores among the compared adaptations on all four metrics for acute kidney injury and mortality prediction across models with 2-9 billion parameters. With 0.516 million trainable parameters, the 2-billion-parameter model reaches an area under the ROC curve of 0.7376 for acute kidney injury, compared with GPT-5.6 Sol's 0.7380 with text input and high reasoning effort. Training for temporal question answering yields 69.27% accuracy at 4 billion parameters with over 90% fewer trainable parameters than low-rank adaptation using charts or numerical text, at a 2.82-4.88 percentage-point accuracy gap. ViSTA extends pretrained language models to numerical prediction and temporal questions.