Abstract: SA-OR86
Kidney Rejection Prediction Model Combining Blood Gene Expression, Donor-Derived Cell-Free DNA, Urine Chemokines, and Torque Teno Virus
Session Information
- Transplantation: Clinical Management and Monitoring
October 26, 2024 | Location: Room 25, Convention Center
Abstract Time: 05:10 PM - 05:20 PM
Category: Transplantation
- 2102 Transplantation: Clinical
Authors
- Chen, Kenny W., Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
- Zhao, Lihui, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
- Paudel, Sujay Dutta, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
- Park, Sookhyeon, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
- Desai, Amishi, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
- Lacombe, Ronnie, Eurofins Viracor BioPharma Services, Lenexa, Kansas, United States
- Sinha, Rohita, Eurofins Viracor BioPharma Services, Lenexa, Kansas, United States
- Kleiboeker, Steven, Eurofins Viracor BioPharma Services, Lenexa, Kansas, United States
- Friedewald, John J., Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States
Background
Blood gene expression (GEP), dd-cfDNA, and urine chemokines have all been shown to be useful non-invasive biomarkers in predicting kidney rejection. Torque teno virus (TTV) is an emerging biomarker of net IS. We aim to integrate these 5 biomarkers into a prediction model for kidney rejection.
Methods
We analyzed 638 biopsy-paired blood and urine samples from the CTOT-08 trial (NCT01289717).
We fit a logistic regression model using log dd-cfDNA, GEP, urine CXCL9 and CXCL10 normalized to Ucreat, and log TTV viral load to predict kidney rejection. Sensitivity, specificity, PPV, NPV and AUC was used to evaluate diagnostic performance.
Results
In the 5-biomarker model, dd-cfDNA, GEP, and TTV VL were statistically significant (p-value < 0.05) adjusting for other covariates (TABLE). Model AUC was 0.816 (Figure). When maximizing Youden’s index, we obtained a sensitivity of 0.71, specificity of 0.78, accuracy of 0.76, NPV of 0.87, and a PPV of 0.56. Out of all rejection types, the model correctly predicted 51/66 AMR cases, 33/41 mixed AMR/ACR, 42/71 ACR, and 360/460 no rejection cases.
Conclusion
Combining non-invasive biomarkers into prediction models to monitor kidney allograft rejection could assist in reducing the need for biopsies and better inform safe immunosuppression titration. We found that dd-cfDNA, blood GEP and TTV VL all significantly improved the prediction model, whereas the addition of urine CXCL9/10 did not significantly impact model performance in predicting kidney rejection.
Model Covariates
| Covariate | log(OR) | 95% CI | p-value |
| Log(ddcf-DNA) | 0.77 | 0.57, 0.99 | <0.001 |
| GEP | 0.05 | 0.03, 0.06 | <0.001 |
| CXCL9 | 0.01 | -0.01, 0.04 | 0.4 |
| CXCL10 | 0.05 | -0.08, 0.19 | 0.4 |
| Log(TTV) | -0.15 | -0.28, -0.02 | 0.030 |
OR = Odds Ratio, CI = Confidence Interval
Funding
- Other NIH Support – Eurofins - Viracor, Eurofins - Transplant Genomics