Problem or Issue	Forecasting AKI and the need for CRRT in ICU patients.
What is Already Known	Existing AKI/CRRT prediction models are limited in: often tailored to specific patient groups like cardiac surgery patients, relying on static features or single-point measurements, ignoring non-zero gap times, and using unimodal frameworks.
What this Paper Adds	We developed a machine-learning, longitudinal, multi-modal model that leverages both clinical notes and time-series measurements. Our model enables early prediction of AKI and CRRT with at least 12 hours' lead time. Furthermore, our model identifies previously overlooked factors contributing to AKI and CRRT need.