Back to AI Research

AI Research

LLM4CKD: Large Language Models for Early Stage Chro... | AI Research

Key Takeaways

  • LLM4CKD: Large Language Models for Early Stage Chronic Kidney Disease Screening This research explores whether Large Language Models (LLMs) can effectively s...
  • This study evaluates the effectiveness of large language models (LLMs) for CKD screening under zero-shot and few-shot in-context learning settings and compares them with traditional ML and DL methods.
  • We propose a framework that uses clinically selected tabular features and structured prompt templates to enable LLM-based inference without task-specific training.
  • LLM performance is evaluated across multiple prompt styles, feature configurations, and data settings, and compared with standard ML, DL, and tabular foundation model (TFM) baselines, and existing CKD screening tools.
  • The results show that LLMs can achieve competitive performance using only a small number of examples, often matching or outperforming traditional approaches in low-data settings.
Paper AbstractExpand

Early screening of chronic kidney disease (CKD) is critical for timely intervention, yet most machine learning (ML) and deep learning (DL) approaches require labeled data and model training, limiting their use in real-world screening settings. This study evaluates the effectiveness of large language models (LLMs) for CKD screening under zero-shot and few-shot in-context learning settings and compares them with traditional ML and DL methods. We propose a framework that uses clinically selected tabular features and structured prompt templates to enable LLM-based inference without task-specific training. LLM performance is evaluated across multiple prompt styles, feature configurations, and data settings, and compared with standard ML, DL, and tabular foundation model (TFM) baselines, and existing CKD screening tools. The results show that LLMs can achieve competitive performance using only a small number of examples, often matching or outperforming traditional approaches in low-data settings. However, their performance remains model-dependent and less stable as input complexity increases. In contrast, ML, DL, and TFM models show more consistent improvement with larger training data. Overall, the findings highlight a trade-off between data efficiency and stability, suggesting that LLMs may serve as a flexible complementary approach for CKD screening when labeled data are limited.

LLM4CKD: Large Language Models for Early Stage Chronic Kidney Disease Screening
This research explores whether Large Language Models (LLMs) can effectively screen for early-stage chronic kidney disease (CKD) without the need for extensive, task-specific training. Because traditional machine learning models often require large amounts of labeled data and stable clinical environments, they can be difficult to deploy in resource-constrained settings. This study investigates if LLMs can act as a flexible, data-efficient alternative by using in-context learning to identify CKD risk factors from patient data.

A Flexible Framework for Clinical Screening

The researchers developed a framework that converts structured patient information—such as medical history, demographic data, and clinical examination findings—into a text-based format that LLMs can process. By using "prompt templates," the team tested how different ways of presenting this data (such as list-based versus narrative-style descriptions) influenced the models' ability to classify patients as having CKD or not. The study evaluated five different LLMs, including both open-weight models like Llama-3 and Gemma-2, and proprietary models like GPT-4o-mini, across zero-shot (no examples provided) and few-shot (a small number of examples provided) settings. The same large language models question is explored in Harness-of-Harness, which adds a research perspective.

Comparing LLMs to Traditional Methods

The study compared the performance of these LLMs against standard machine learning, deep learning, and tabular foundation models. A key part of the evaluation involved "feature selection," where the researchers used a subset of clinically meaningful variables rather than the full range of available data. The results indicate that LLMs can achieve competitive performance in low-data environments, often matching or outperforming traditional models when only a small number of examples are available.

Key Findings and Performance Trends

The research highlights a notable trade-off between data efficiency and model stability. While LLMs show great promise in settings where labeled data is scarce, their performance is highly dependent on the specific model used and the structure of the prompt. The study found that simplifying the input data by selecting only the most relevant clinical features generally improved the accuracy of the LLMs. However, unlike traditional machine learning and deep learning models, which tend to show more consistent performance gains as they are provided with more training data, LLMs can become less stable as the complexity of the input increases. The same large language models question is explored in When Does Bigger Help? A Controlled..., which adds a research perspective.

Considerations for Clinical Use

The findings suggest that LLMs could serve as a valuable, complementary tool for CKD screening, particularly in regions where diagnostic resources and labeled datasets are limited. However, the researchers emphasize that LLM performance is not uniform; it varies significantly based on how the prompt is designed and which model is employed. While the models show an ability to prioritize clinically meaningful risk factors, the study underscores the need for careful prompt engineering and validation to ensure these tools remain reliable and consistent when applied to real-world clinical scenarios. as detailed in the full paper on Arxiv

More from Franklin

GRAINresearchLarge Language Models (LLMs) for Telecom...researchHarness-of-HarnessresearchWhen Does Bigger Help? A Controlled...researchpro-team at LLMs4OL 2026 Tasks Flagship...researchConstrained Entity Selection under Partial Knowledge...research

Comments (0)

No comments yet

Be the first to share your thoughts!