Machine learning prediction models for diabetic kidney disease: A systematic review and meta-analysis.
Chenglong Zhou, Xiaochu Wu, Zhili Tian et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated machine learning models for predicting diabetic kidney disease (DKD) across various studies. The analysis included 66 models, revealing a pooled area under the curve (AUC) of 0.926, with sensitivity at 0.83 and specificity at 0.90. Neural networks demonstrated the highest AUC of 0.973.
Key findings
- The pooled AUC for ML models predicting DKD was 0.926 (95% CI: 0.898-0.947).
- Sensitivity was 0.83 (95% CI: 0.79-0.86) and specificity was 0.90 (95% CI: 0.87-0.93).
- Neural networks achieved the highest AUC of 0.973 among the models evaluated.
- Model type and study design were significant effect modifiers (P = 0.019 and P = 0.037, respectively).
Why it matters
Understanding the performance of machine learning models in predicting diabetic kidney disease can enhance early diagnosis and management strategies, potentially improving patient outcomes. However, the high heterogeneity and methodological limitations highlight the need for cautious interpretation.
What to keep in mind
The evidence is limited by high heterogeneity, lack of external validation, and poor methodological transparency.