📄 transplantation

Improving diagnostic performance of kidney allograft rejection with a model combining relative fraction and absolute copies of donor-derived cell-free DNA - results from five independent cohorts

By integrating relative fraction and absolute copies of donor-derived cell-free DNA into a novel composite score (CM-Score), this study demonstrates significantly improved diagnostic accuracy for kidney allograft rejection across five independent cohorts, offering a robust tool to better rule in and rule out rejection compared to existing single-metric approaches.

Original authors: Benning, L., Akifova, A., Oellerich, M., Osmanodja, B., Morath, C., Beck, J., Schuetz, A., Bornemann-Kolatzki, K., Schrezenmeier, E. V., Tran, T. H., Schwenger, V., Shipkova, M., Wieland, E., Schuetz

Published 2026-01-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Benning, L., Akifova, A., Oellerich, M., Osmanodja, B., Morath, C., Beck, J., Schuetz, A., Bornemann-Kolatzki, K., Schrezenmeier, E. V., Tran, T. H., Schwenger, V., Shipkova, M., Wieland, E., Schuetz, E., Budde, K.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Picture: A Better "Smoke Detector" for Kidney Transplants

Imagine you have a kidney transplant. To make sure the new kidney is healthy, doctors need to know if the body is attacking it (rejection). Currently, they use a blood test called dd-cfDNA. Think of this test as a "smoke detector" for the kidney.

When the kidney gets hurt, tiny pieces of its DNA leak into the blood. The test measures this "smoke" in two different ways:

The Percentage: What portion of the total "smoke" in the air comes from the kidney? (Like asking: "Is 5% of the smoke in this room from the kitchen?")
The Absolute Count: How many actual pieces of kidney DNA are there? (Like asking: "Are there 50 pieces of smoke in this room?")

The Problem: The Old Detectors Get Confused

The authors explain that using just one of these measurements has flaws:

The Percentage Problem: If the patient gets sick with a flu or has a urinary tract infection, their body produces a lot of its own DNA. This dilutes the kidney DNA, making the percentage look low even if the kidney is actually in trouble. It's like adding a huge fan to a smoky room; the smoke is still there, but it looks less concentrated because the air volume increased.
The Absolute Count Problem: If the patient has very little blood flow or the test isn't perfect, the count might look low even if there is damage.
The "Two-Alarm" Confusion: Some doctors try to use both numbers. If either number is high, they sound the alarm. But this leads to false alarms. It's like a security system that beeps if the door opens OR if the window cracks. It works, but it beeps too often for things that aren't actually break-ins.

The Solution: The "CM-Score" (The Smart Composite Score)

The researchers created a new tool called the CM-Score. Instead of looking at the percentage or the count separately, they combined them into a single, continuous number using a specific mathematical formula.

The Analogy:
Imagine you are trying to guess if a car engine is overheating.

Old Way: You look at the temperature gauge (Percentage) OR you count how many steam puffs come out of the hood (Absolute Count). Sometimes the gauge is stuck, or the wind blows the steam away, and you get the wrong answer.
New Way (CM-Score): You have a smart computer that looks at both the gauge and the steam puffs simultaneously. It weighs them together to give you one final "Overheat Score."

What Did They Find?

The team tested this new score on data from 383 patients across five different groups (some were routine checks, some were sick patients). They compared their new score against the old methods and against data from 11 other major studies.

Here are the results in plain English:

Fewer False Alarms (Better "Rule-In"):
The old methods were great at saying "Everything is fine" (high Negative Predictive Value), but they were bad at saying "Something is wrong." They often sounded the alarm when nothing was actually wrong (low Positive Predictive Value).
- The Result: The CM-Score fixed this. It kept the ability to say "Everything is fine" with high confidence, but it became much better at saying "Something is wrong."
- The Numbers: In the study, when the CM-Score said there was a problem, it was right 81% of the time. The old methods were only right about 54% of the time.
Better at Distinguishing Real Trouble:
The new score was better at telling the difference between a kidney that is being rejected (the body attacking it) and a kidney that is just tired or slightly damaged by other things (like medication toxicity).
It Works Like a "Sliding Scale":
Instead of a simple "Yes/No" line (like a cutoff of 0.5%), the CM-Score gives a number that can be anywhere from negative to positive.
- Negative Score: Very likely no rejection. You can relax.
- High Positive Score: Very likely rejection. You need to act.
- Middle Score: It gives doctors a "probability" (e.g., "There is a 64% chance of rejection") rather than a binary guess. This helps doctors decide if a biopsy (a needle test) is actually necessary.

Why Does This Matter?

Currently, because the old tests have so many false alarms, doctors often have to perform unnecessary biopsies (taking a tiny piece of the kidney with a needle) to check if the patient is really having a rejection. This is invasive and stressful for the patient.

The paper claims that by using the CM-Score, doctors can be much more confident when they see a high result. This means:

They can stop doing unnecessary biopsies for false alarms.
They can catch real rejections earlier and more accurately.
They can make better decisions about whether to change a patient's medication.

Summary

The paper introduces a new math formula (the CM-Score) that combines two different ways of measuring kidney DNA. By doing this, it turns a "noisy" smoke detector into a "smart" one that rarely misses a real fire and rarely screams when there is just a burnt piece of toast. This helps doctors make clearer decisions about kidney transplant patients.

Technical Summary: Improving Diagnostic Performance of Kidney Allograft Rejection with a CM-Score Model

Problem Statement
Donor-derived cell-free DNA (dd-cfDNA) has become a standard-of-care biomarker for monitoring kidney transplant recipients. However, current clinical practice relies on reporting dd-cfDNA either as a relative percentage of total cell-free DNA (dd-cfDNA%) or as an absolute quantity in genomic copies per milliliter (cp/mL). Each metric has distinct limitations: dd-cfDNA% is susceptible to fluctuations in the recipient's total cfDNA levels, potentially leading to false classifications independent of graft health. Conversely, while absolute quantification (cp/mL) has demonstrated superior performance in some studies, relying on two separate metrics can lead to divergent results when using fixed thresholds. Existing parallel testing approaches (where an elevated result in either metric triggers a positive call) improve sensitivity but often reduce specificity, leading to high rates of false positives and unnecessary biopsies. The authors hypothesize that integrating both metrics into a single continuous composite score could overcome these limitations, reducing false classifications and improving the accuracy of rejection diagnosis.

Methodology
The study analyzed 443 dd-cfDNA measurements from 383 patients across five independent kidney transplant cohorts. The dataset included samples from patients undergoing indication biopsies (including T-cell-mediated rejection [TCMR], antibody-mediated rejection [ABMR], BK-virus associated nephropathy, and drug toxicity) and clinically stable surveillance patients. dd-cfDNA was quantified using SNP-based droplet-digital PCR (ddPCR).

The methodology involved a two-stage approach:

Model Derivation: A random discovery set comprising 25 biopsy-proven rejections and 26 non-rejections was used to derive a continuous composite score (CM-Score). The authors evaluated linear, logarithmic, and quadratic models to integrate dd-cfDNA% and cp/mL. The optimal model (Eq. 1) was selected based on maximizing the Area Under the Receiver Operating Characteristic Curve (AUCROC).
Validation: The derived CM-Score was validated in a larger cohort of 75 rejections and 279 non-rejections. The study also performed a meta-analysis comparing the CM-Score against 11 published clinical cohorts (totaling 6,536 samples) to contextualize performance.
Statistical Analysis: Performance was evaluated using sensitivity, specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and Likelihood Ratios (LR). The analysis accounted for varying rejection prevalences (e.g., 25% for general validation, 6–11% for surveillance settings) using Bayes' theorem to calculate post-test probabilities. Decision curve analysis was employed to assess net clinical benefit.

Key Contributions
The primary contribution is the development and validation of the CM-Score, a continuous mathematical model that combines relative fraction and absolute copies of dd-cfDNA.

Mathematical Integration: The model utilizes the formula: $MV = (Percent Value^{0.71})^{0.5} + (cp/mL Value / 32)^{0.6}$ . This continuous score is transformed to a scale where a cut-off of zero distinguishes rejection from non-rejection.
Superior Diagnostic Metrics: The study demonstrates that the CM-Score outperforms both dd-cfDNA% and cp/mL individually. While the AUCROC was not significantly different from single metrics, the CM-Score significantly improved the Positive Predictive Value (PPV) while maintaining a high Negative Predictive Value (NPV).
Clinical Utility Framework: The authors provide a framework for interpreting the CM-Score using Likelihood Ratios across a continuous range rather than a single dichotomous threshold. This allows clinicians to adjust post-test probabilities based on specific pre-test clinical contexts (e.g., for-cause vs. surveillance, high vs. low immunological risk).

Results

Performance Comparison: In the validation cohort (at 25% prevalence), the CM-Score achieved a PPV of 81% (95% CI: 72–89%), a significant improvement over the weighted average PPV of 54% reported in published literature for standard dd-cfDNA metrics. The NPV remained high at 91% (89–93%).
Metric Hierarchy: The results consistently showed that the CM-Score provided better diagnostic performance than cp/mL alone, which in turn was superior to dd-cfDNA% alone.
Specificity and Rule-In Capability: The CM-Score improved the specificity for rejection to nearly 95%, addressing the issue of low specificity in current assays where values are often elevated in non-rejection pathologies (e.g., BKVAN, CNI toxicity). The Likelihood Ratio for a positive result (LR+) increased by approximately 168% compared to literature values, enhancing the ability to "rule in" rejection.
Surveillance Impact: In simulated surveillance settings (10% prevalence), the CM-Score reduced the odds of overdiagnosis (unnecessary biopsies) from 2.6:1 (typical of current methods) to 0.68:1, and to ≤0.54:1 for high CM-Score values (≥0.8).
Robustness: Bootstrap validation and random split analyses confirmed the model's robustness against overfitting. The addition of clinical variables (eGFR, time post-transplant, HLA mismatch) to the CM-Score did not yield a meaningful improvement in discrimination.

Significance and Claims
The paper claims that the CM-Score represents a robust, clinically meaningful tool that improves the diagnostic accuracy of dd-cfDNA as a stand-alone metric. By moving from fixed thresholds to a continuous composite score, the model addresses the "rule-out" limitation of current dd-cfDNA testing (high NPV but low PPV) and establishes its viability as a "rule-in" biomarker.

The authors emphasize that this approach allows for personalized clinical decision-making. Instead of a "one-size-fits-all" interpretation, clinicians can use the continuous Likelihood Ratios to update the probability of rejection based on the specific pre-test probability of the patient (e.g., a patient with rising creatinine vs. a routine surveillance patient). The study concludes that while the CM-Score significantly improves diagnostic confidence and reduces unnecessary biopsies, prospective clinical validation is required to confirm its impact on long-term patient outcomes and immunosuppression management. The authors explicitly state that the findings are based on real-world data and that the model is currently in a verification stage.