Deep Learning on Chest X-Rays Predicts 5-Year CKD Risk in US Multicenter Study

A deep learning model using chest radiographs (CXR-CKD5) predicted 5-year incident chronic kidney disease (CKD) with an external validation C-index of 0.713 (95% CI 0.697-0.729). Combined with clinical data, C-index reached 0.734. Performance was strongest in non-diabetic patien…

• CXR-CKD5, a deep learning model using only chest radiographs, achieved external validation C-index 0.713 (95% CI 0.697-0.729) for predicting 5-year incident CKD.
• A combined imaging-clinical model reached C-index 0.734; performance was highest in non-diabetic patients (C-index 0.714) and substantially lower in diabetic patients (0.607).
• Absolute risk was overestimated at the external validation site, highlighting the need for recalibration before clinical deployment.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.