Diagnostic Performance and Resource Utilization of Combining Blood Gene Expression, Cell-Free DNA, and Urine Chemokines for Monitoring Kidney Rejection

Visual Abstract

Abstract

Key Points

• Two-stage urine chemokine screening improved discrimination and positive predictive value versus blood-only testing while preserving high negative predictive value.

• In validation, urine chemokine C-X-C motif ligand 9/creatinine screening referred only 79% of samples for donor-derived cell-free DNA testing, reducing blood test use.

• Standalone blood biomarkers were more consistent for antibody-mediated rejection than cellular rejection; urine chemokine screening improved diagnostic yield.

Background

After kidney transplantation, subclinical acute rejection occurs in 20%–25% of clinically stable recipients; surveillance biopsies are invasive and often indicate no rejection. Moreover, the widespread use of noninvasive biomarkers is limited by cost.

Methods

We developed and sought to validate two-stage diagnostic models integrating urine chemokine assays (chemokine C-X-C motif ligand 9 [CXCL9] and C-X-C motif ligand 10) with confirmatory blood-based gene expression profiling (GEP) and donor-derived cell-free DNA (dd-cfDNA). Data from two studies, 476 biopsy-paired samples from 226 recipients (discovery) and 144 samples from 134 recipients (validation), were analyzed retrospectively. Rapid, low-cost urine chemokine assays were used for stage 1 screening; recipients with elevated urine chemokines underwent additional stage 2 testing with GEP and/or dd-cfDNA.

Results

In the validation cohort, dd-cfDNA outperformed GEP for overall rejection and antibody-mediated rejection, whereas neither test validated significant discrimination for cellular rejection. CXCL9-based two-stage testing improved the area under the receiver operating characteristic curve and positive predictive value versus standalone stage-2 testing, while maintaining similar negative predictive value and reducing stage 2 test use. At a probability screening threshold of 0.10, the CXCL9/creatinine-based logistic regression, adjusted for BK virus, urinary tract infection, sex, age, time post-transplant, and donor type, followed by dd-cfDNA validated a higher positive predictive value (0.39 versus 0.34), lower sensitivity (0.56 versus 0.62), and similar negative predictive value around 0.94, compared with dd-cfDNA alone, while recommending dd-cfDNA testing in only 0.79 (95% confidence interval, 0.72 to 0.85) of validation cases. Combined GEP and dd-cfDNA testing validated higher area under the receiver operating characteristic curves than dd-cfDNA as a second-stage test for overall (0.85 [0.79 to 0.94] versus 0.82 [0.74 to 0.92]), antibody-mediated (0.96 [0.93 to 1.00] versus 0.90 [0.79 to 1.00]), and cellular rejections (0.69 [0.61 to 0.89] versus 0.63 [0.53 to 0.81]).

Conclusions

This staged approach reduces unnecessary biopsies and guides the efficient use of costly blood-based biomarkers while maintaining good clinical performance, particularly for subclinical antibody-mediated rejections.

Clinical Trial registry name and registration number:

ClinicalTrials.gov, NCT01289717.

Introduction

Approximately 20%–25% of clinically stable kidney transplant (KT) patients experience subclinical acute rejection (AR).^1–3 Surveillance biopsies are the gold standard for diagnosing subclinical AR, but they are invasive, and interpretation varies among pathologists.⁴ Moreover, 75%–80% of surveillance biopsies show no rejection (TX), limiting their specificity for clinical stratification of KT recipients.⁵ Less-invasive tools, such as serum creatinine (Cr) and immunosuppression levels, are commonly used but have suboptimal sensitivity and specificity and may lag behind actual histologic rejection.^6,7

Urine chemokines, blood-based gene expression profiling (GEP), and donor-derived cell-free DNA (dd-cfDNA) have all been shown to correlate with the presence or absence of rejection.^3,7–11 Recent studies have underscored the promise of integrating multiple biomarkers to enhance the detection of subclinical AR in KT recipients.¹² Combining blood-based GEP with dd-cfDNA significantly improved diagnostic accuracy over either marker alone, offering enhanced negative predictive value (NPV) and positive predictive value (PPV) for detecting subclinical AR episodes.³ Despite these advantages, the high cost of GEP and dd-cfDNA assays relative to urine chemokines limits their widespread use.^9,13,14 This challenge has motivated our investigation into the role of urine chemokines as an initial screening tool to guide a resource-efficient and noninvasive diagnostic approach.

This study aims to (1) develop and assess the reliability of a model for detecting subclinical AR using a combination of previously studied blood biomarkers: GEP, dd-cfDNA, and urine chemokines and (2) investigate stepwise models for efficient resource utilization while maximizing the diagnostic performance.

Methods

Study Population

We retrospectively investigated cohorts from the Clinical Trials in Organ Transplantation-08 (CTOT-08) study (NCT01289717) and the Northwestern University Mini-Kidney Biorepository Study (MKBS; NCT01531257) as discovery and validation sets, respectively. CTOT-08 was a multicenter observational study of 307 individuals who underwent KT between March 2011 and March 2014.^3,5 MKBS is an ongoing observational study of 777 subjects that mirrors CTOT-08, employing the same inclusion and exclusion criteria. The MKBS subjects included in this analysis underwent KT between April 2005 and October 2021. We reported the demographic characteristics collected in both studies in Table 1.⁵

Table 1.

Patient-level demographics and clinical information in Clinical Trials in Organ Transplantation-08 (discovery) and Mini-Kidney Biorepository Study (validation) data set

Characteristic	CTOT-08 (Discovery Set)	MKBS (Validation Set)
	N=226a	N=134a
Recipient age	52 (42–63)	50 (36–58)
Recipient sex
Female	80 (35%)	52 (39%)
Male	146 (65%)	82 (61%)
Recipient ethnicity
Hispanic or Latino	32 (14%)	31 (23%)
Not Hispanic or Latino	184 (81%)	101 (75%)
Unknown or not reported	10 (4.4%)	2 (1.5%)
Recipient race
American Indian or Alaska Native	4 (1.8%)	2 (1.5%)
Asian	11 (4.9%)	8 (6.0%)
Black or African American	45 (20%)	26 (19%)
More than one race	2 (0.9%)	0 (0%)
Native Hawaiian or other Pacific Islander	2 (0.9%)	1 (0.7%)
Unknown or not reported	18 (8.0%)	22 (16%)
White	144 (64%)	75 (56%)
Donor age	41 (27–50)	43 (32–54)
Donor sex
Female	110 (49%)	61 (46%)
Male	116 (51%)	71 (53%)
Unknown/not reported	0 (0%)	2 (1.5%)
Donor ethnicity
Hispanic or Latino	29 (13%)	21 (16%)
Not Hispanic or Latino	171 (76%)	109 (81%)
Unknown or not reported	26 (12%)	4 (3.0%)
Donor race
American Indian or Alaska Native	2 (0.9%)	0 (0%)
Asian	8 (3.5%)	7 (5.2%)
Black or African American	26 (12%)	12 (9.0%)
Native Hawaiian or other Pacific Islander	1 (0.4%)	1 (0.7%)
Unknown or not reported	26 (12%)	36 (27%)
White	163 (72%)	78 (58%)
Reason for ESKD
Cystic (includes PKD)	24 (11%)	10 (7.5%)
Diabetes mellitus	44 (19%)	22 (16%)
GN	63 (28%)	26 (19%)
Hypertension	41 (18%)	26 (19%)
Other	54 (24%)	48 (36%)
Unknown or not reported	0 (0%)	2 (1.5%)
PRA class 1 %	0 (0–2)	0 (0–7)
PRA class 2 %	0 (0–0)	0 (0–4)
cPRA %	9 (0–64)	0 (0–7)
Deceased donor
Deceased donor	96 (42%)	54 (40%)
Living donor	130 (58%)	78 (58%)
Unknown or not reported	0 (0%)	2 (1.5%)
Induction therapy
Alemtuzumab	113 (50%)	96 (72%)
Anti-thymocyte globulin	65 (29%)	0 (0%)
Steroid	183 (81%)	131 (98%)
Basiliximab	42 (19%)	30 (22%)
Immunosuppression
Steroid	130 (58%)	108 (81%)
Tacrolimus	225 (100%)	133 (99%)
Cyclosporine	12 (5.3%)	8 (6.0%)
MMF	223 (99%)	134 (100%)
mTORi	17 (7.5%)	26 (19%)
Leflunomide	3 (1.3%)	4 (3.0%)
Belatacept	1 (0.4%)	19 (14%)
Days post-transplant	363 (120–721.25)	380 (242.5–732.0)

cPRA, calculated Panel Reactive Antibody; CTOT-08, Clinical Trials in Organ Transplantation-08; MMF, mycophenolate mofetil; mTORi, mammalian target of rapamycin inhibitor; MKBS, Mini-Kidney Biorepository Study; PKD, polycystic kidney disease; PRA, panel reactive antibody.

^aMedian (Q1–Q3); n (%).

Study Outcome

Participants underwent surveillance kidney biopsies at 2–6, 12, and 24 months post-transplant with stable kidney function.⁵ Biopsies were classified as subclinical AR or TX. Subclinical AR was defined based on Banff 2019 criteria AND stable kidney function with serum Cr <2.3 mg/dl and <20% increase in Cr from a minimum of two to three prior values over a mean of 132 days (range, 75–187 days). Because detailed donor-specific antibody (DSA) results were unavailable for some patients, antibody-mediated rejection (AMR) was diagnosed using histologic findings and DSA when available; otherwise, biopsies were classified as AMR if they met two of the Banff 2019 histologic criteria for AMR and one for suspicious.

Study Exposures

This study included four biomarkers: two urine chemokines, a GEP, and a dd-cfDNA assay.

Urine Chemokines

Urine specimens were collected immediately before the allograft biopsies. Samples were analyzed using the Ella method, an automated, benchtop ELISA platform (Ella Automated Immunoassay System; Bio-Techne, Minneapolis, MN). Levels of chemokine C-X-C motif ligand 9 (CXCL9) and chemokine C-X-C motif ligand 10 (CXCL10) were measured.^15,16 The results were normalized to urinary Cr by calculating the CXCL9/Cr and the CXCL10/Cr ratios (micrograms of protein per gram of urinary Cr; μg/g). Urine Cr measurements were performed on the same samples using the Cr Parameter Assay Kit (Bio-Techne).

Gene Expression Profile

Blood samples for the GEP assay were collected directly into PAXgene tubes (BD BioSciences, San Jose, CA) at the time of the surveillance biopsy. The GEP in the discovery cohort was analyzed using the TruGraf algorithm, a DNA microarray-based method that examines differential expression across 120 genes (Eurofins-Transplant Genomics, Lenexa, KS).^17,18 For the validation cohort, the GEP assays were processed using reverse transcriptase PCR and microfluidics on the Fluidigm Biomark high definition system (Fluidigm, South San Francisco, CA). The results were presented as a probability score normalized to 0–100 and dichotomized at 50, with scores above 50 classified as “positive” and scores ≤50 classified as “negative.”⁵

dd-cfDNA

Blood samples for dd-cfDNA analysis were collected during the surveillance biopsy in plasma separation tubes (BD Vacutainer plasma preparation tube; BD BioSciences, San Jose, CA). Next-generation sequencing data were aligned to a reference genome and analyzed alongside recipient genotype data using a bioinformatics pipeline licensed from Stanford University.¹⁹ The results were reported as the fraction of dd-cfDNA (TRAC assay, Eurofins-Viracor, Lenexa, KS), with a threshold of 0.7% based on the manufacturer's recommendations. Values >0.7% were classified as “positive.” In comparison, those ≤0.7% were classified as “negative.”

Clinical Factors

BK viremia (BKV) and urinary tract infection (UTI) events were captured and verified by clinical sites. For our analyses, a biopsy was mapped to any infection event reported within the preceding 30 days. Relative eGFR was defined as the relative deviation from each patient's baseline at post-transplant day 30, calculated using the CKD Epidemiology Collaboration equation.²⁰ Because some subjects lacked concurrent DSA results, all biopsies performed after a DSA-positive test were considered DSA-positive, whereas biopsies performed before a DSA-negative test were considered DSA-negative.

Two-Stage Model Development

We evaluated two-stage diagnostic models for subclinical AR using a rapid, low-cost urine chemokine assay as the stage 1 rule-out test. A costlier but more accurate stage 2 test, GEP, dd-cfDNA, or combined GEP and dd-cfDNA, was performed only when stage 1 indicated suspicion of subclinical AR. The combined GEP+dd-cfDNA test was considered negative only when both component assays were negative.

For stage 1, we considered baseline logistic regression (LR) models with CXCL9/Cr, CXCL10/Cr, or both, adjusted for recent BKV and UTI, relative eGFR, sex, age, living or decreased donor type, and time post-transplant. For each chosen probability screening threshold (PST) for the stage 1 LR, the predicted probability determined how often stage 2 testing was recommended.^21,22

For each two-stage model, we reported the area under the receiver operating characteristic curve (AUC), and at a given stage 1 PST, the diagnostic performance: sensitivity, specificity, PPV, and NPV, evaluated relative to each stage 2 comparator (GEP, dd-cfDNA, or combined GEP+dd-cfDNA). We also reported the proportion of cases referred for stage 2 testing. Sensitivity analyses examined model performance by: (1) reclassifying borderline (BL) biopsies as rejection; (2) including chronic rejections alongside subclinical AR; (3) subgroups defined by age, time post-transplant, donor type, and sex; and (4) incorporating DSA data into the stage 1 LR model.

Statistical Analysis

Descriptive statistics were reported as mean with SD and median with interquartile range for continuous variables and counts and percentages for categorical variables. We also used the t test or Wilcoxon rank-sum test, and the chi-squared test or Fisher exact test to analyze continuous or categorical characteristics, respectively. Bootstrapping was used to report the 95% confidence intervals (CIs) for performance metrics. Because phenotypes were unknown when applying biomarkers, models were trained for any subclinical AR versus TX, with phenotype-specific performance reported secondarily. All statistical analyses were conducted using R version 4.4.0.

Results

We selected 226 subjects from the CTOT-08 (discovery) dataset with 476 biopsies showing stable kidney function and available urine CXCL9/CXCL10, GEP, and dd-cfDNA measurements (Figure 1). Of these, 78 biopsies were subclinical AR (64 AMR, six cellular, eight mixed), and 382 were TX, including 19 BLs. The MKBS (validation) cohort comprised 134 subjects with 144 allograft biopsies with stable kidney function and complete measurements of all four biomarkers (Supplemental Figure 1). These included 16 subclinical AR (ten AMR, five cellular, one mixed) and 128 TX, with 30 BL cases. Across cohorts, AMR accounted for the majority of rejection events. Both cohorts had broadly similar demographics, except for differences in age, days post-transplant, and immunosuppression (Table 1). CTOT-08 recipients were slightly older (median 52 versus 50 years, P = 0.04) and earlier in follow-up (median 363 versus 380 days, P < 0.001) than MKBS recipients. Immunosuppression differed most substantially: MKBS subjects more often received alemtuzumab and maintenance mammalian target of rapamycin inhibitors or Belatacept, whereas antithymocyte globulin induction was used only in CTOT-08 (all P < 0.001).

Figure 1.

CONSORT diagram illustrating the number of patients and the samples available for analysis based on inclusion and exclusion criteria and sample availability for the discovery CTOT-08 cohort. CONSORT, Consolidated Standards of Reporting Trials; CTOT-08, Clinical Trials in Organ Transplantation-08; CXCL9, chemokine C-X-C motif ligand 9; CXCL10, chemokine C-X-C motif ligand 10; dd-cfDNA, donor-derived cell-free DNA.

Single-Stage Models

In the baseline CXCL9/Cr LR, a 0.1 μg/g increase in CXCL9/Cr was independently associated with subclinical AR (odds ratio [OR], 1.08; 95% CI, 1.04 to 1.12; P < 0.001), living donor transplant with lower odds (OR, 0.41; 0.24 to 0.68; P < 0.001), and months post-transplant with higher odds (OR, 1.05; 1.02 to 1.08; P = 0.002). The baseline CXCL10/Cr LR showed similar associations: a 0.1 μg/g increase in CXCL10/Cr (OR, 1.21; 1.06 to 1.38; P = 0.01), living donor transplant (P < 0.001), and months from transplant (P = 0.003). In the combined CXCL9/Cr and CXCL10/Cr LR, CXCL9/Cr remained independently associated (OR, 1.07; 1.02 to 1.12; P = 0.01), whereas CXCL10/Cr did not (P = 0.55); living-donor transplant (P < 0.001) and months from transplant (P = 0.002) remained significant covariates with similar odds (Supplemental Figures 2 and 3 and Supplemental Table 1).

The three baseline LR models, tuned at Youden-optimal probability thresholds of 0.17 (CXCL9/Cr), 0.15 (CXCL10/Cr), and 0.16 (combined CXCL9/Cr and CXCL10/Cr), showed moderate discrimination for subclinical AR versus TX with AUCs of 0.70 to 0.72 in both CTOT-08 and MKBS cohorts (Table 2). In MKBS, PPVs were low (0.16 to 0.17), whereas NPVs were consistently high (0.94 to 0.96) with slightly higher sensitivity (0.75 to 0.81) and lower specificity (0.47 to 0.52) than that of the CTOT-08 cohort. The combined CXCL9/Cr and CXCL10/Cr LR demonstrated stronger rule-out performance than individual LR models in both cohorts. By phenotype, all three LR models validated significant discrimination for AMR (e.g., CXCL9/Cr LR AUC=0.86 [0.75 to 0.96]) but did not validate significant discrimination for cellular rejection (e.g., CXCL9/Cr LR AUC=0.51 [0.41 to 0.81]), owing to the small sample size. Overall, urine chemokine LR provided efficient rule-out performance with high NPV in both cohorts.

Table 2.

Performance of single-stage: urine chemokines, gene expression profile, and donor-derived cell-free DNA to detect subclinical acute rejection in discovery (Clinical Trials in Organ Transplantation-08) and validation (Mini-Kidney Biorepository Study) cohorts

Model	Prediction	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	Dataset
Baseline CXCL9/Cr LR >0.17	Subclinical AR (n=78) versus TX (n=382)	0.72 (0.65 to 0.77)	0.71 (0.60 to 0.81)	0.68 (0.63 to 0.72)	0.31 (0.26 to 0.35)	0.92 (0.89 to 0.94)	CTOT-08
	Cellular (n=6) versus TX (n=382)	0.70 (0.52 to 0.89)	0.83 (0.50 to 1.00)	0.68 (0.63 to 0.72)	0.04 (0.02 to 0.05)	1.00 (0.99 to 1.00)
	AMR (n=64) versus TX (n=382)	0.73 (0.67 to 0.79)	0.70 (0.58 to 0.81)	0.68 (0.63 to 0.72)	0.27 (0.23 to 0.31)	0.93 (0.91 to 0.96)
	Mixed (n=8) versus TX (n=382)	0.61 (0.48 to 0.80)	0.62 (0.25 to 1.00)	0.68 (0.63 to 0.72)	0.04 (0.02 to 0.06)	0.99 (0.98 to 1.00)
Baseline CXCL10/Cr LR >0.15	Subclinical AR (n=78) versus TX (n=382)	0.70 (0.63 to 0.75)	0.77 (0.68 to 0.86)	0.59 (0.54 to 0.63)	0.28 (0.24 to 0.31)	0.93 (0.90 to 0.95)
	Cellular (n=6) versus TX (n=382)	0.63 (0.49 to 0.79)	0.83 (0.50 to 1.00)	0.59 (0.54 to 0.64)	0.03 (0.02 to 0.04)	1.00 (0.99 to 1.00)
	AMR (n=64) versus TX (n=382)	0.73 (0.66 to 0.79)	0.80 (0.69 to 0.89)	0.59 (0.54 to 0.63)	0.25 (0.21 to 0.28)	0.95 (0.92 to 0.97)
	Mixed (n=8) versus TX (n=382)	0.51 (0.47 to 0.72)	0.50 (0.12 to 0.88)	0.59 (0.54 to 0.64)	0.02 (0.01 to 0.04)	0.98 (0.97 to 1.00)
Baseline CXCL9/Cr+CXCL10/Cr LR >0.16	Subclinical AR (n=78) versus TX (n=382)	0.72 (0.66 to 0.77)	0.74 (0.65 to 0.83)	0.64 (0.58 to 0.68)	0.29 (0.26 to 0.33)	0.92 (0.90 to 0.95)
	Cellular (n=6) versus TX (n=382)	0.69 (0.52 to 0.88)	0.83 (0.50 to 1.00)	0.64 (0.59 to 0.68)	0.03 (0.02 to 0.05)	1.00 (0.99 to 1.00)
	AMR (n=64) versus TX (n=382)	0.74 (0.67 to 0.79)	0.75 (0.64 to 0.86)	0.64 (0.59 to 0.68)	0.26 (0.22 to 0.29)	0.94 (0.91 to 0.96)
	Mixed (n=8) versus TX (n=382)	0.61 (0.48 to 0.80)	0.62 (0.25 to 1.00)	0.64 (0.59 to 0.68)	0.03 (0.01 to 0.05)	0.99 (0.98 to 1.00)
GEP >50%	Subclinical AR (n=78) versus TX (n=382)	0.68 (0.62 to 0.74)	0.40 (0.29 to 0.51)	0.79 (0.75 to 0.83)	0.28 (0.22 to 0.35)	0.87 (0.84 to 0.89)
	Cellular (n=6) versus TX (n=382)	0.71 (0.55 to 0.85)	0.33 (0.00 to 0.67)	0.79 (0.75 to 0.83)	0.02 (0.00 to 0.05)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX (n=382)	0.68 (0.61 to 0.75)	0.41 (0.28 to 0.53)	0.79 (0.75 to 0.83)	0.25 (0.18 to 0.32)	0.89 (0.87 to 0.91)
	Mixed (n=8) versus TX (n=382)	0.69 (0.44 to 0.85)	0.38 (0.00 to 0.75)	0.79 (0.75 to 0.84)	0.04 (0.00 to 0.07)	0.98 (0.98 to 0.99)
dd-cfDNA >0.7%	Subclinical AR (n=78) versus TX (n=382)	0.81 (0.76 to 0.86)	0.64 (0.54 to 0.74)	0.86 (0.83 to 0.90)	0.49 (0.42 to 0.57)	0.92 (0.90 to 0.94)
	Cellular (n=6) versus TX (n=382)	0.59 (0.46 to 0.79)	0.17 (0.00 to 0.50)	0.86 (0.83 to 0.90)	0.02 (0.00 to 0.06)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX (n=382)	0.84 (0.79 to 0.89)	0.69 (0.58 to 0.80)	0.86 (0.83 to 0.90)	0.46 (0.39 to 0.53)	0.94 (0.92 to 0.96)
	Mixed (n=8) versus TX (n=382)	0.77 (0.48 to 0.95)	0.62 (0.25 to 1.00)	0.86 (0.83 to 0.90)	0.09 (0.04 to 0.14)	0.99 (0.98 to 1.00)
Combined GEP+dd-cfDNA	Subclinical AR (n=78) versus TX (n=382)	0.74 (0.69 to 0.79)	0.78 (0.69 to 0.87)	0.69 (0.65 to 0.73)	0.34 (0.30 to 0.38)	0.94 (0.91 to 0.96)
	Cellular (n=6) versus TX (n=382)	0.60 (0.52 to 0.78)	0.50 (0.17 to 0.83)	0.69 (0.64 to 0.73)	0.02 (0.01 to 0.04)	0.99 (0.98 to 1.00)
	AMR (n=64) versus TX (n=382)	0.75 (0.70 to 0.80)	0.81 (0.72 to 0.91)	0.69 (0.64 to 0.74)	0.31 (0.27 to 0.35)	0.96 (0.94 to 0.98)
	Mixed (n=8) versus TX (n=382)	0.72 (0.54 to 0.85)	0.75 (0.38 to 1.00)	0.69 (0.64 to 0.74)	0.05 (0.03 to 0.07)	0.99 (0.98 to 1.00)
Baseline CXCL9/Cr LR >0.17	Subclinical AR (n=16) versus TX (n=128)	0.70 (0.53 to 0.85)	0.75 (0.50 to 0.94)	0.52 (0.44 to 0.60)	0.16 (0.12 to 0.21)	0.94 (0.89 to 0.99)	MKBS
	Cellular (n=5) versus TX (n=128)	0.51 (0.41 to 0.81)	0.60 (0.20 to 1.00)	0.52 (0.44 to 0.60)	0.05 (0.02 to 0.08)	0.97 (0.94 to 1.00)
	AMR (n=10) versus TX (n=128)	0.86 (0.75 to 0.96)	0.90 (0.70 to 1.00)	0.52 (0.44 to 0.61)	0.13 (0.10 to 0.16)	0.99 (0.96 to 1.00)
	Mixed (n=1) versus TX (n=128)	0.95 (0.95 to 0.98)	0.00 (0.00 to 0.00)	0.52 (0.43 to 0.60)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)
Baseline CXCL10/Cr LR >0.15	Subclinical AR (n=16) versus TX (n=128)	0.72 (0.56 to 0.87)	0.81 (0.62 to 1.00)	0.47 (0.38 to 0.55)	0.16 (0.12 to 0.20)	0.95 (0.90 to 1.00)
	Cellular (n=5) versus TX (n=128)	0.51 (0.41 to 0.82)	0.60 (0.20 to 1.00)	0.47 (0.38 to 0.55)	0.04 (0.01 to 0.07)	0.97 (0.93 to 1.00)
	AMR (n=10) versus TX (n=128)	0.89 (0.81 to 0.97)	1.00 (1.00 to 1.00)	0.47 (0.39 to 0.55)	0.13 (0.11 to 0.15)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX (n=128)	0.92 (0.91 to 0.94)	0.00 (0.00 to 0.00)	0.47 (0.38 to 0.56)	0.00 (0.00 to 0.00)	0.98 (1.00 to 1.00)
Baseline CXCL9/Cr+CXCL10/Cr LR >0.16	Subclinical AR (n=16) versus TX (n=128)	0.71 (0.55 to 0.86)	0.81 (0.62 to 1.00)	0.51 (0.43 to 0.59)	0.17 (0.13 to 0.21)	0.96 (0.91 to 1.00)
	Cellular (n=5) versus TX (n=128)	0.51 (0.42 to 0.82)	0.60 (0.20 to 1.00)	0.51 (0.42 to 0.59)	0.05 (0.01 to 0.08)	0.97 (0.94 to 1.00)
	AMR (n=10) versus TX (n=128)	0.87 (0.78 to 0.96)	1.00 (1.00 to 1.00)	0.51 (0.43 to 0.59)	0.14 (0.12 to 0.16)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX (n=128)	0.95 (0.95 to 0.98)	0.00 (0.00 to 0.00)	0.51 (0.43 to 0.60)	0.00 (0.00 to 0.00)	0.98 (1.00 to 1.00)
GEP >50%	Subclinical AR (n=16) versus TX (n=128)	0.60 (0.43 to 0.74)	0.50 (0.25 to 0.75)	0.72 (0.64 to 0.79)	0.18 (0.10 to 0.27)	0.92 (0.88 to 0.96)
	Cellular (n=5) versus TX (n=128)	0.46 (0.39 to 0.76)	0.40 (0.00 to 0.80)	0.72 (0.64 to 0.80)	0.05 (0.00 to 0.11)	0.97 (0.95 to 0.99)
	AMR (n=10) versus TX (n=128)	0.67 (0.46 to 0.84)	0.60 (0.30 to 0.90)	0.72 (0.64 to 0.80)	0.14 (0.07 to 0.22)	0.96 (0.93 to 0.99)
	Mixed (n=1) versus TX (n=128)	0.84 (0.80 to 0.90)	0.00 (0.00 to 0.00)	0.72 (0.64 to 0.79)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)
dd-cfDNA >0.7%	Subclinical AR (n=16) versus TX (n=128)	0.81 (0.70 to 0.91)	0.62 (0.38 to 0.81)	0.85 (0.79 to 0.91)	0.34 (0.23 to 0.48)	0.95 (0.92 to 0.97)
	Cellular (n=5) versus TX (n=128)	0.63 (0.41 to 0.81)	0.20 (0.00 to 0.60)	0.85 (0.79 to 0.91)	0.05 (0.00 to 0.15)	0.96 (0.95 to 0.98)
	AMR (n=10) versus TX (n=128)	0.89 (0.77 to 0.98)	0.80 (0.50 to 1.00)	0.85 (0.79 to 0.91)	0.30 (0.19 to 0.42)	0.98 (0.96 to 1.00)
	Mixed (n=1) versus TX (n=128)	1.00 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.85 (0.79 to 0.91)	0.00 (0.00 to 0.00)	1.00 (1.00 to 1.00)
Combined GEP+dd-cfDNA	Subclinical AR (n=16) versus TX (n=128)	0.71 (0.60 to 0.82)	0.81 (0.62 to 1.00)	0.62 (0.54 to 0.70)	0.21 (0.16 to 0.27)	0.96 (0.92 to 1.00)
	Cellular (n=5) versus TX (n=128)	0.51 (0.30 to 0.73)	0.40 (0.00 to 0.80)	0.62 (0.54 to 0.70)	0.04 (0.00 to 0.08)	0.96 (0.94 to 0.99)
	AMR (n=10) versus TX (n=128)	0.81 (0.77 to 0.85)	1.00 (1.00 to 1.00)	0.62 (0.53 to 0.70)	0.17 (0.14 to 0.21)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX (n=128)	0.81 (0.78 to 0.85)	1.00 (1.00 to 1.00)	0.62 (0.53 to 0.70)	0.00 (0.00 to 0.00)	1.00 (1.00 to 1.00)

AMR, antibody-mediated rejection; AR, acute rejection; AUC, area under the receiver operating characteristic curve; CI, confidence interval; Cr, creatinine; CTOT-08, Clinical Trials in Organ Transplantation-08; CXCL9, chemokine C-X-C motif ligand 9; CXCL10, chemokine C-X-C motif ligand 10; dd-cfDNA, donor-derived cell-free DNA; GEP, gene expression profile; LR, logistic regression; MKBS, Mini-Kidney Biorepository Study; NPV, negative predictive value; PPV, positive predictive value; TX, no rejection.

Similar to the performance in the CTOT-08 cohort, dd-cfDNA >0.7% outperformed GEP >50% on overall discrimination and rule-in metrics in MKBS (AUC=0.81 [0.70 to 0.91], specificity=0.85 [0.79 to 0.91], PPV=0.34 [0.23 to 0.48] versus AUC=0.60 [0.43 to 0.74], specificity=0.72 [0.64 to 0.79], PPV=0.18 [0.10 to 0.27]; Table 2). The combined GEP and dd-cfDNA test maximized sensitivity (0.81) and NPV (0.96) but at the cost of lower specificity (0.62) and PPV (0.21). Phenotype-level results were concordant with biologic expectations: dd-cfDNA performed best in AMR (AUC=0.89 [0.77 to 0.98]) and the combined blood test increased sensitivity across phenotypes. However, the three blood-based tests did not validate significant discrimination for cellular rejections (Table 2). Across cohorts, dd-cfDNA and the combined GEP and dd-cfDNA tests provided better rule-in performance with higher PPV than urine-chemokine-based LR models for AMR.

Two-Stage Model

In both cohorts, the baseline CXCL9/Cr LR then dd-cfDNA and CXCL9/Cr LR then the combined GEP and dd-cfDNA two-stage models showed improved AUCs and PPVs over the individual dd-cfDNA and the combined GEP and dd-cfDNA testing, respectively (Figure 2 and Tables 2 and 3). At the same time, both models reduced the use of second-stage testing across all PSTs of 0.05, 0.10, 0.15, 0.20, and 0.25 while showing similar sensitivity and NPV to the individual second-stage tests. At a PST of 0.10, the CXCL9/Cr LR then dd-cfDNA test increased PPV from 0.34 to 0.39, with slightly lower sensitivity, 0.56 versus 0.60, and similar NPV, 0.94 versus 0.95, compared with dd-cfDNA alone in the validation cohort. This two-stage model recommended dd-cfDNA testing for only 0.79 (0.72 to 0.85) of validation cases. Similarly, the baseline CXCL9/Cr LR then the combined GEP and dd-cfDNA model with a PST of 0.10 increased PPV from 0.21 to 0.24, with lower sensitivity, 0.75 versus 0.81, and similar NPV, 0.96 versus 0.96, compared with the combined GEP and dd-cfDNA testing in the validation cohort. Across phenotypes, the above two tests validated better discrimination than their respective standalone stage 2 testing, although performance was substantially stronger for AMR than for cellular rejection. Full results of all the two-stage models appear in the Supplemental Material (Supplemental Tables 2–19).

Figure 2.

A two-stage diagnostic testing model with CXCL9/Cr and CXCL10/Cr-based model as stage 1 test and stage 2 tests is recommended only when the chances of positive prediction by the first stage test are above the PST for the stage 1 urine chemokines test. After the stage 2 test, subclinical AR or TX is predicted based on the second stage test result. We considered baseline CXCL9/Cr, CXCL10/Cr, and the combined CXCL9/Cr and CXCL10/Cr LR models as stage 1 tests; whereas, GEP, dd-cfDNA, and the combined GEP and dd-cfDNA as stage 2 tests. AR, acute rejection; Cr, creatinine; GEP, gene expression profile; LR, logistic regression; PST, probability screening threshold; TX, no rejection.

Table 3.

Performance of baseline chemokine C-X-C motif ligand 9/creatinine logistic regression then donor-derived cell-free DNA, and combined gene expression profile and donor-derived cell-free DNA two-stage models for discovery (Clinical Trials in Organ Transplantation-08) cohort with 78 subclinical acute rejection and 382 no rejection cases, and validation (Mini-Kidney Biorepository Study) cohort with 16 subclinical acute rejection and 128 no rejection cases

Dataset	Prediction	Stage-1 PST	Stage-2 Test Use	Baseline CXCL9/Cr LR Then dd-cfDNA					Baseline CXCL9/Cr LR Then Combined GEP and dd-cfDNA
				AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)
CTOT-08	Subclinical AR versus TX	0.05	0.95 (0.93 to 0.97)	0.82 (0.78 to 0.87)	0.64 (0.54 to 0.74)	0.88 (0.85 to 0.91)	0.52 (0.45 to 0.60)	0.92 (0.90 to 0.94)	0.84 (0.81 to 0.89)	0.78 (0.69 to 0.87)	0.72 (0.68 to 0.76)	0.36 (0.32 to 0.41)	0.94 (0.92 to 0.96)
	Cellular (n=6) versus TX		0.94 (0.91 to 0.96)	0.60 (0.52 to 0.86)	0.17 (0.00 to 0.50)	0.88 (0.85 to 0.91)	0.02 (0.00 to 0.07)	0.99 (0.98 to 0.99)	0.75 (0.63 to 0.90)	0.50 (0.17 to 0.83)	0.72 (0.68 to 0.76)	0.03 (0.01 to 0.05)	0.99 (0.98 to 1.00)
	AMR (n=64) versus TX		0.95 (0.92 to 0.96)	0.85 (0.80 to 0.90)	0.69 (0.56 to 0.80)	0.88 (0.85 to 0.91)	0.49 (0.41 to 0.57)	0.94 (0.92 to 0.96)	0.85 (0.82 to 0.91)	0.81 (0.72 to 0.91)	0.72 (0.68 to 0.76)	0.33 (0.29 to 0.37)	0.96 (0.94 to 0.98)
	Mixed (n=8) versus TX		0.94 (0.91 to 0.96)	0.77 (0.61 to 0.97)	0.62 (0.25 to 0.88)	0.88 (0.85 to 0.91)	0.10 (0.04 to 0.16)	0.99 (0.98 to 1.00)	0.84 (0.69 to 0.98)	0.75 (0.38 to 1.00)	0.72 (0.67 to 0.76)	0.05 (0.03 to 0.07)	0.99 (0.98 to 1.00)
	Subclinical AR versus TX	0.10	0.73 (0.69 to 0.77)	0.82 (0.78 to 0.87)	0.60 (0.50 to 0.71)	0.90 (0.87 to 0.93)	0.55 (0.47 to 0.65)	0.92 (0.90 to 0.94)	0.84 (0.81 to 0.89)	0.71 (0.60 to 0.79)	0.78 (0.74 to 0.82)	0.40 (0.34 to 0.45)	0.93 (0.91 to 0.95)
	Cellular (n=6) versus TX		0.69 (0.65 to 0.74)	0.60 (0.52 to 0.86)	0.17 (0.00 to 0.50)	0.90 (0.87 to 0.93)	0.03 (0.00 to 0.08)	0.99 (0.98 to 0.99)	0.75 (0.63 to 0.90)	0.33 (0.00 to 0.67)	0.78 (0.74 to 0.82)	0.02 (0.00 to 0.05)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX		0.73 (0.69 to 0.77)	0.85 (0.80 to 0.90)	0.66 (0.53 to 0.77)	0.90 (0.87 to 0.93)	0.53 (0.44 to 0.62)	0.94 (0.92 to 0.96)	0.85 (0.82 to 0.91)	0.77 (0.66 to 0.86)	0.78 (0.74 to 0.82)	0.37 (0.32 to 0.43)	0.95 (0.93 to 0.97)
	Mixed (n=8) versus TX		0.69 (0.65 to 0.74)	0.77 (0.61 to 0.97)	0.50 (0.12 to 0.88)	0.90 (0.87 to 0.93)	0.10 (0.03 to 0.16)	0.99 (0.98 to 1.00)	0.84 (0.69 to 0.98)	0.50 (0.12 to 0.88)	0.78 (0.74 to 0.82)	0.05 (0.01 to 0.07)	0.99 (0.98 to 1.00)
	Subclinical AR versus TX	0.15	0.47 (0.43 to 0.52)	0.82 (0.78 to 0.87)	0.49 (0.37 to 0.59)	0.92 (0.90 to 0.95)	0.57 (0.47 to 0.68)	0.90 (0.88 to 0.92)	0.84 (0.81 to 0.89)	0.56 (0.45 to 0.67)	0.86 (0.82 to 0.90)	0.45 (0.38 to 0.53)	0.91 (0.88 to 0.93)
	Cellular (n=6) versus TX		0.42 (0.38 to 0.47)	0.60 (0.52 to 0.86)	0.17 (0.00 to 0.50)	0.92 (0.90 to 0.95)	0.03 (0.00 to 0.10)	0.99 (0.98 to 0.99)	0.75 (0.63 to 0.90)	0.33 (0.00 to 0.67)	0.86 (0.82 to 0.89)	0.04 (0.00 to 0.08)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX		0.47 (0.43 to 0.51)	0.85 (0.80 to 0.90)	0.53 (0.42 to 0.66)	0.92 (0.90 to 0.95)	0.54 (0.44 to 0.66)	0.92 (0.90 to 0.94)	0.85 (0.82 to 0.91)	0.61 (0.50 to 0.73)	0.86 (0.83 to 0.90)	0.42 (0.35 to 0.51)	0.93 (0.91 to 0.95)
	Mixed (n=8) versus TX		0.42 (0.37 to 0.47)	0.77 (0.61 to 0.97)	0.38 (0.00 to 0.75)	0.92 (0.90 to 0.95)	0.09 (0.00 to 0.19)	0.99 (0.98 to 0.99)	0.84 (0.69 to 0.98)	0.38 (0.00 to 0.75)	0.86 (0.82 to 0.89)	0.05 (0.00 to 0.11)	0.99 (0.98 to 0.99)
	Subclinical AR versus TX	0.20	0.29 (0.25 to 0.33)	0.82 (0.78 to 0.87)	0.35 (0.24 to 0.46)	0.95 (0.93 to 0.97)	0.59 (0.46 to 0.72)	0.88 (0.86 to 0.90)	0.84 (0.81 to 0.89)	0.38 (0.28 to 0.50)	0.91 (0.88 to 0.94)	0.47 (0.37 to 0.58)	0.88 (0.86 to 0.90)
	Cellular (n=6) versus TX		0.26 (0.22 to 0.30)	0.60 (0.52 to 0.86)	0.17 (0.00 to 0.50)	0.95 (0.93 to 0.97)	0.05 (0.00 to 0.14)	0.99 (0.98 to 0.99)	0.75 (0.63 to 0.90)	0.17 (0.00 to 0.50)	0.91 (0.88 to 0.94)	0.03 (0.00 to 0.08)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX		0.29 (0.25 to 0.33)	0.85 (0.80 to 0.90)	0.38 (0.27 to 0.48)	0.95 (0.93 to 0.97)	0.56 (0.43 to 0.70)	0.90 (0.88 to 0.92)	0.85 (0.82 to 0.91)	0.42 (0.30 to 0.55)	0.91 (0.88 to 0.94)	0.44 (0.34 to 0.55)	0.90 (0.89 to 0.92)
	Mixed (n=8) versus TX		0.25 (0.21 to 0.30)	0.77 (0.61 to 0.97)	0.25 (0.00 to 0.62)	0.95 (0.93 to 0.97)	0.10 (0.00 to 0.23)	0.98 (0.98 to 0.99)	0.84 (0.69 to 0.98)	0.25 (0.00 to 0.62)	0.91 (0.88 to 0.94)	0.06 (0.00 to 0.14)	0.98 (0.98 to 0.99)
	Subclinical AR versus TX	0.25	0.18 (0.14 to 0.21)	0.82 (0.78 to 0.87)	0.23 (0.14 to 0.32)	0.97 (0.95 to 0.98)	0.58 (0.42 to 0.75)	0.86 (0.85 to 0.87)	0.84 (0.81 to 0.89)	0.26 (0.17 to 0.35)	0.95 (0.92 to 0.97)	0.50 (0.36 to 0.64)	0.86 (0.85 to 0.88)
	Cellular (n=6) versus TX		0.15 (0.11 to 0.18)	0.60 (0.52 to 0.86)	0.17 (0.00 to 0.50)	0.97 (0.95 to 0.98)	0.07 (0.00 to 0.21)	0.99 (0.98 to 0.99)	0.75 (0.63 to 0.90)	0.17 (0.00 to 0.50)	0.95 (0.93 to 0.97)	0.05 (0.00 to 0.15)	0.99 (0.98 to 0.99)
	AMR (n=64) versus TX		0.17 (0.14 to 0.21)	0.85 (0.80 to 0.90)	0.23 (0.14 to 0.34)	0.97 (0.95 to 0.98)	0.54 (0.38 to 0.73)	0.88 (0.87 to 0.90)	0.85 (0.82 to 0.91)	0.27 (0.16 to 0.38)	0.95 (0.93 to 0.97)	0.46 (0.32 to 0.61)	0.89 (0.87 to 0.90)
	Mixed (n=8) versus TX		0.15 (0.11 to 0.18)	0.77 (0.61 to 0.97)	0.25 (0.00 to 0.50)	0.97 (0.95 to 0.98)	0.13 (0.00 to 0.31)	0.98 (0.98 to 0.99)	0.84 (0.69 to 0.98)	0.25 (0.00 to 0.50)	0.95 (0.92 to 0.97)	0.09 (0.00 to 0.21)	0.98 (0.98 to 0.99)
MKBS	Subclinical AR versus TX	0.05	0.96 (0.92 to 0.99)	0.82 (0.74 to 0.92)	0.62 (0.38 to 0.81)	0.86 (0.80 to 0.91)	0.36 (0.23 to 0.50)	0.95 (0.92 to 0.97)	0.85 (0.79 to 0.94)	0.81 (0.62 to 1.00)	0.64 (0.56 to 0.73)	0.22 (0.16 to 0.28)	0.96 (0.93 to 1.00)
	Cellular (n=5) versus TX		0.95 (0.92 to 0.98)	0.63 (0.53 to 0.81)	0.20 (0.00 to 0.60)	0.86 (0.80 to 0.91)	0.05 (0.00 to 0.16)	0.96 (0.95 to 0.98)	0.69 (0.61 to 0.89)	0.40 (0.00 to 0.80)	0.64 (0.55 to 0.72)	0.04 (0.00 to 0.09)	0.96 (0.94 to 0.99)
	AMR (n=10) versus TX		0.96 (0.93 to 0.99)	0.90 (0.79 to 1.00)	0.80 (0.50 to 1.00)	0.86 (0.80 to 0.91)	0.31 (0.21 to 0.45)	0.98 (0.96 to 1.00)	0.96 (0.93 to 1.00)	1.00 (1.00 to 1.00)	0.64 (0.56 to 0.72)	0.18 (0.15 to 0.22)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX		0.96 (0.92 to 0.99)	1.00 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.86 (0.80 to 0.92)	0.05 (0.00 to 0.00)	1.00 (1.00 to 1.00)	1.00 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.64 (0.56 to 0.72)	0.02 (0.00 to 0.00)	1.00 (1.00 to 1.00)
	Subclinical AR versus TX	0.10	0.79 (0.72 to 0.85)	0.82 (0.74 to 0.92)	0.56 (0.31 to 0.81)	0.89 (0.84 to 0.94)	0.39 (0.25 to 0.56)	0.94 (0.91 to 0.97)	0.85 (0.79 to 0.94)	0.75 (0.56 to 0.94)	0.70 (0.62 to 0.77)	0.24 (0.17 to 0.31)	0.96 (0.92 to 0.99)
	Cellular (n=5) versus TX		0.78 (0.71 to 0.85)	0.63 (0.53 to 0.81)	0.20 (0.00 to 0.60)	0.89 (0.84 to 0.94)	0.07 (0.00 to 0.19)	0.97 (0.96 to 0.98)	0.69 (0.61 to 0.89)	0.40 (0.00 to 0.80)	0.70 (0.61 to 0.77)	0.05 (0.00 to 0.10)	0.97 (0.95 to 0.99)
	AMR (n=10) versus TX		0.80 (0.73 to 0.86)	0.90 (0.79 to 1.00)	0.80 (0.50 to 1.00)	0.89 (0.83 to 0.94)	0.36 (0.24 to 0.53)	0.98 (0.96 to 1.00)	0.96 (0.93 to 1.00)	1.00 (1.00 to 1.00)	0.70 (0.62 to 0.77)	0.20 (0.17 to 0.26)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX		0.78 (0.71 to 0.84)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.89 (0.84 to 0.94)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.70 (0.61 to 0.78)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)
	Subclinical AR versus TX	0.15	0.59 (0.51 to 0.67)	0.82 (0.74 to 0.92)	0.56 (0.31 to 0.81)	0.92 (0.87 to 0.96)	0.47 (0.31 to 0.67)	0.94 (0.92 to 0.98)	0.85 (0.79 to 0.94)	0.75 (0.56 to 0.94)	0.78 (0.71 to 0.84)	0.30 (0.22 to 0.40)	0.96 (0.93 to 0.99)
	Cellular (n=5) versus TX		0.56 (0.49 to 0.65)	0.63 (0.53 to 0.81)	0.20 (0.00 to 0.60)	0.92 (0.87 to 0.96)	0.09 (0.00 to 0.29)	0.97 (0.96 to 0.98)	0.69 (0.61 to 0.89)	0.40 (0.00 to 0.80)	0.78 (0.70 to 0.85)	0.07 (0.00 to 0.15)	0.97 (0.95 to 0.99)
	AMR (n=10) versus TX		0.59 (0.51 to 0.68)	0.90 (0.79 to 1.00)	0.80 (0.50 to 1.00)	0.92 (0.88 to 0.96)	0.44 (0.30 to 0.63)	0.98 (0.96 to 1.00)	0.96 (0.93 to 1.00)	1.00 (1.00 to 1.00)	0.78 (0.70 to 0.84)	0.26 (0.21 to 0.33)	1.00 (1.00 to 1.00)
	Mixed (n=1) versus TX		0.56 (0.47 to 0.65)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.92 (0.87 to 0.96)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.78 (0.71 to 0.84)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)
	Subclinical AR versus TX	0.20	0.44 (0.37 to 0.53)	0.82 (0.74 to 0.92)	0.50 (0.25 to 0.75)	0.92 (0.87 to 0.96)	0.44 (0.26 to 0.68)	0.94 (0.91 to 0.97)	0.85 (0.79 to 0.94)	0.69 (0.44 to 0.88)	0.85 (0.79 to 0.91)	0.37 (0.24 to 0.50)	0.96 (0.92 to 0.98)
	Cellular (n=5) versus TX		0.41 (0.33 to 0.49)	0.63 (0.53 to 0.81)	0.20 (0.00 to 0.60)	0.92 (0.88 to 0.97)	0.09 (0.00 to 0.30)	0.97 (0.96 to 0.98)	0.69 (0.61 to 0.89)	0.40 (0.00 to 0.80)	0.85 (0.79 to 0.91)	0.10 (0.00 to 0.21)	0.97 (0.96 to 0.99)
	AMR (n=10) versus TX		0.45 (0.38 to 0.54)	0.90 (0.79 to 1.00)	0.70 (0.40 to 0.90)	0.92 (0.88 to 0.96)	0.41 (0.25 to 0.64)	0.98 (0.95 to 0.99)	0.96 (0.93 to 1.00)	0.90 (0.70 to 1.00)	0.85 (0.79 to 0.91)	0.32 (0.24 to 0.45)	0.99 (0.97 to 1.00)
	Mixed (n=1) versus TX		0.41 (0.34 to 0.50)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.92 (0.88 to 0.97)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.85 (0.79 to 0.91)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)
	Subclinical AR versus TX	0.25	0.29 (0.22 to 0.36)	0.82 (0.74 to 0.92)	0.44 (0.19 to 0.69)	0.95 (0.91 to 0.98)	0.54 (0.31 to 0.80)	0.93 (0.90 to 0.96)	0.85 (0.79 to 0.94)	0.56 (0.31 to 0.81)	0.92 (0.88 to 0.96)	0.47 (0.29 to 0.68)	0.94 (0.92 to 0.98)
	Cellular (n=5) versus TX		0.26 (0.19 to 0.34)	0.63 (0.53 to 0.81)	0.20 (0.00 to 0.60)	0.95 (0.91 to 0.98)	0.14 (0.00 to 0.50)	0.97 (0.96 to 0.98)	0.69 (0.61 to 0.89)	0.40 (0.00 to 0.80)	0.92 (0.88 to 0.96)	0.17 (0.00 to 0.38)	0.98 (0.96 to 0.99)
	AMR (n=10) versus TX		0.29 (0.22 to 0.36)	0.90 (0.79 to 1.00)	0.60 (0.30 to 0.90)	0.95 (0.91 to 0.98)	0.50 (0.27 to 0.78)	0.97 (0.94 to 0.99)	0.96 (0.93 to 1.00)	0.70 (0.40 to 1.00)	0.92 (0.87 to 0.97)	0.41 (0.25 to 0.64)	0.98 (0.95 to 1.00)
	Mixed (n=1) versus TX		0.26 (0.19 to 0.34)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.95 (0.91 to 0.98)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)	1.00 (1.00 to 1.00)	0.00 (0.00 to 0.00)	0.92 (0.87 to 0.97)	0.00 (0.00 to 0.00)	0.99 (1.00 to 1.00)

AMR, antibody-mediated rejection; AR, acute rejection; AUC, area under the receiver operating characteristic curve; CI, confidence interval; Cr, creatinine; CTOT-08, Clinical Trials in Organ Transplantation-08; CXCL9, chemokine C-X-C motif ligand 9; dd-cfDNA, donor-derived cell-free DNA; GEP, gene expression profiling; LR, logistic regression; MKBS, Mini-Kidney Biorepository Study; NPV, negative predictive value; PPV, positive predictive value; PST, probability screening threshold; TX, no rejection.

For the discovery cohort at a PST of 0.10, the baseline CXCL9/Cr LR in both models correctly identified 30.9% of TX cases at stage 1, which included four BL cases. These stage 1 true negatives were early post-transplant TX cases (mean=183.4 [156.1 to 210.7] days); 11.0% of TX cases were BKV-positive, and 10.2% had UTI, with a mean relative eGFR change of 2.12 (1.66 to 2.59; Table 4). Stage 1 false negatives represented 7.7% of subclinical AR cases and consisted of three AMR, one cellular, and two mixed rejections, with a lower mean relative eGFR change of 1.54 (−2.30 to 5.37). These cases also occurred early post-transplant (mean=220.3 [71.5 to 329.2] days) and could have been detected by the combined GEP and dd-cfDNA test (GEP positive for one AMR, one cellular, and one mixed; dd-cfDNA positive for the other two AMR and one mixed). The two-stage model with dd-cfDNA as the second-stage test missed 24.4% of rejections versus 15.4% for the model with the combined GEP and dd-cfDNA. Although the two-stage model using the combined GEP and dd-cfDNA test increased false positives (22.0% of TX cases versus 9.9% for dd-cfDNA), these included ten BL cases, some of which were classified as negative by dd-cfDNA. Moreover, the two-stage model with combined GEP and dd-cfDNA second-stage testing improved AMR (identifying 76.6% versus 64.6% of all AMR) and cellular rejection (identifying 33.3% versus 16.7% of all cellular) detection in the second stage, with similar detection of mixed rejections compared with the dd-cfDNA. The composition of cases at each stage was similar in the validation cohort (Table 4).

Table 4.

Details of baseline chemokine C-X-C motif ligand 9/creatinine logistic regression then donor-derived cell-free DNA, and combined gene expression profiling and donor-derived cell-free DNA two-stage model prediction at probability screening threshold of 0.10 for discovery (Clinical Trials in Organ Transplantation-08) cohort with 78 subclinical acute rejection and 382 no rejection cases, and validation (Mini-Kidney Biorepository Study) cohort with 16 subclinical acute rejection and 128 no rejection cases

Dataset	Metric	Baseline CXCL9/Cr LR Then dd-cfDNA						Baseline CXCL9/Cr LR Then Combined GEP and dd-cfDNA
		Stage-1 FN	Stage 1 TN	Stage 2 FN	Stage 2 TN	Stage 2 FP	Stage 2 TP	Stage 1 FN	Stage 1 TN	Stage 2 FN	Stage 2 TN	Stage 2 FP	Stage 2 TP
CTOT-08	N	6	118	25	226	38	47	6	118	17	180	84	55
	T-days (95% CI)	200.3 (71.5 to 329.2)	183.4 (156.1 to 210.7)	402.5 (309.9 to 495.1)	422.7 (388.4 to 457.0)	468.7 (382.7 to 554.7)	481.5 (413.9 to 549.1)	200.3 (71.5 to 329.2)	183.4 (156.1 to 210.7)	417.1 (294.9 to 539.3)	419.0 (380.0 to 458.0)	451.5 (396.2 to 506.7)	465.5 (403.8 to 527.2)
	AMR	3	0	19	0	0	42	3	0	12	0	0	49
	TX	0	118	0	226	38	0	0	118	0	180	84	0
	BL	0	4	0	10	5	0	0	4	0	5	10	0
	TCMR	1	0	4	0	0	1	1	0	3	0	0	2
	Mixed	2	0	2	0	0	4	2	0	2	0	0	4
	BKV positive	17%	11%	0%	3%	5%	4%	17%	11%	0%	2%	6%	4%
	UTI positive	0%	10%	0%	0%	0%	0%	0%	10%	0%	0%	0%	0%
	Rel-eGFR (95% CI)	1.54 (−2.30 to 5.37)	2.12 (1.66 to 2.59)	2.42 (1.33 to 3.50)	3.48 (2.97 to 3.99)	4.43 (2.73 to 6.14)	5.12 (2.79 to 7.45)	1.54 (−2.30 to 5.37)	2.12 (1.66 to 2.59)	2.88 (1.42 to 4.33)	3.86 (3.26 to 4.46)	3.09 (2.19 to 3.99)	4.59 (2.57 to 6.60)
MKBS	N	2	28	5	86	14	9	2	28	2	61	39	12
	T-days (95% CI)	235.0 (−1594.7 to 2064.7)	240.4 (175.2 to 305.5)	495.6 (280.7 to 710.5)	683.2 (530.9 to 835.5)	871.1 (215.3 to 1526.9)	2365.6 (1162.6 to 3568.6)	235.0 (−1594.7 to 2064.7)	240.4 (175.2 to 305.5)	370.5 (237.1 to 503.9)	715.9 (519.9 to 911.9)	699.5 (436.9 to 962.1)	1918.9 (926.4 to 2911.5)
	AMR	0	0	2	0	0	8	0	0	0	0	0	10
	TX	0	28	0	86	14	0	0	28	0	61	39	0
	BL	0	4	0	24	2	0	0	4	0	15	11	0
	TCMR	1	0	3	0	0	1	1	0	2	0	0	2
	Mixed	1	0	0	0	0	0	1	0	0	0	0	0
	BKV positive	50%	14%	0%	4%	0%	0%	50%	14%	0%	5%	0%	0%
	UTI positive	0%	4%	0%	0%	0%	0%	0%	4%	0%	0%	0%	0%
	Rel-eGFR (95% CI)	0.03 (−2.91 to 2.96)	0.06 (0.02 to 0.11)	0.51 (−0.54 to 1.56)	0.20 (0.12 to 0.29)	0.20 (−0.05 to 0.45)	0.01 (−0.33 to 0.35)	0.03 (−2.91 to 2.96)	0.06 (0.02 to 0.11)	0.31 (−2.58 to 3.20)	0.25 (0.13 to 0.36)	0.14 (0.03 to 0.25)	0.17 (−0.26 to 0.60)

AMR, antibody-mediated rejection; BKV, BK viremia; BL, borderline; CI, confidence interval; Cr, creatinine; CTOT-08, Clinical Trials in Organ Transplantation-08; CXCL9, chemokine C-X-C motif ligand 9; dd-cfDNA, donor-derived cell-free DNA; FN, false negative; FP, false positive; GEP, gene expression profiling; LR, logistic regression; MKBS, Mini-Kidney Biorepository Study; Rel-eGFR, mean relative eGFR change; TCMR, T-cell mediated rejection; T-days, days from transplant; TN, true negative; TP, true positive; TX, no rejection; UTI, urinary tract infection.

As part of sensitivity analysis, we included the BL cases as subclinical AR instead of TX in the two-stage models and, in MKBS, we observed lower AUC and NPV, while increased PPV with similar specificity compared with models treating BL cases as TX (Supplemental Tables 20 and 21). We also evaluated chronic rejections (n=16 only in the discovery cohort) together with subclinical AR versus TX (which included BL cases) (Supplemental Tables 22 and 23). In CTOT-08, the two-stage models starting with baseline CXCL9/Cr LR, then dd-cfDNA or the combined GEP and dd-cfDNA tests, showed reduced AUCs of 0.80 and 0.81, respectively, for both rejection types versus TX prediction, and AUCs of 0.70 and 0.77, respectively, for chronic rejection versus TX. At a PST of 0.10, the second stage testing utilization increased to 0.76 (0.72 to 0.80) of cases, higher than in models limited to subclinical AR versus TX.

Across cohorts, at a PST of 0.10, these two-stage models showed similar diagnostic performance in patients younger than the median age of 52 years and those ≥52 years (P = 0.08; Supplemental Table 24); however, utilization of the second-stage test was significantly higher for younger patient samples (P < 0.001). Patterns were similar when comparing biopsies from the first-year versus the second-year post-transplant (Supplemental Table 25). In MKBS, testing for patients with living versus deceased donors showed similar age- and time-related differences (Supplemental Table 26). We observed similar performance between male and female patients in both cohorts (Supplemental Table 27).

We also performed a sensitivity analysis for biopsy samples with available DSA information. DSA data were available for 211 CTOT-08 and 45 MKBS biopsies, with a subclinical AR prevalence of 21.3% and 13.3%, respectively (Supplemental Table 28). We developed a new stage 1 baseline CXCL9/Cr LR model, additionally adjusted for DSA status, and found a similar OR for CXCL9/Cr as in the baseline CXCL9/Cr LR without DSA; however, DSA positivity was associated with higher odds of rejection (OR, 7.05; 2.80 to 17.74; P < 0.001; Supplemental Tables 29 and 30). Across cohorts, using the baseline CXCL9/Cr LR with DSA as the stage 1 test at a PST of 0.15, then dd-cfDNA or the combined GEP and dd-cfDNA, also improved performance over the standalone second-stage tests (Table 2 and Supplemental Table 31). At the same time, these two-stage models demonstrated higher PPV and specificity, with slightly lower sensitivity and NPV compared with the models without DSA consideration, while significantly reducing the utilization of second-stage testing (0.31 [0.18 to 0.44] versus 0.79 [0.72 to 0.85]; Table 3 and Supplemental Table 31).

Discussion

We developed and validated two-stage subclinical AR diagnostic models using urine chemokines as an initial screening test, followed by GEP and/or dd-cfDNA tests for confirmation. As single-stage tests, the baseline urine chemokine-based LR models, adjusted for BKV, UTI, relative eGFR, sex, age, time post-transplant, and donor type, validated a good rule-out performance with high NPV (0.94 [0.89 to 0.99]). The blood-based tests dd-cfDNA and the combined GEP and dd-cfDNA validated a good rule-in performance, with higher AUCs of 0.81 (0.70 to 0.91) and 0.71 (0.60 to 0.82), and PPVs of 0.34 (0.23 to 0.48) and 0.21 (0.16 to 0.27), respectively, than urine chemokines. Hence, the urine chemokines were used as screening tests, and the two blood-based tests were used as confirmatory tests in the two-stage models. The resulting two-stage model with CXCL9/Cr LR at a PST of 0.10, then dd-cfDNA testing, demonstrated an AUC of 0.82 (0.74 to 0.92), a sensitivity of 0.56 (0.31 to 0.81), a specificity of 0.89 (0.84 to 0.94), a PPV of 0.39 (0.25 to 0.56), and an NPV of 0.94 (0.91 to 0.97) in the validation cohort. Although the combined CXCL9/Cr and CXCL10/Cr LR delivered high rule-out performance, we evaluated the baseline CXCL9/Cr LR model as the screening test for the two-stage model development since CXCL9/Cr showed significance (OR, 1.07; 1.02 to 1.12; P = 0.01) and used fewer resources than the combined LR model. The two-stage models validated improved AUC and PPV with similar sensitivity and NPV compared with the individual second-stage test, while reducing the utilization of the second-stage test (Tables 2 and 3).

Two-stage models using dd-cfDNA or the combined GEP and dd-cfDNA as the stage-2 test outperformed models using GEP alone as the stage-2 test in both cohorts. Moreover, the combined GEP and dd-cfDNA as the second-stage test validated higher AUCs than dd-cfDNA alone for overall rejection (0.85 [0.79 to 0.94] versus 0.82 [0.74 to 0.92]), for AMR (0.96 [0.93 to 1.00] versus 0.90 [0.79 to 1.00]), and for cellular rejections (0.69 [0.61 to 0.89] versus 0.63 [0.53 to 0.81]), although with wide CI for cellular rejection owing to the small sample size. Specifically, at the PST of 0.10 for the baseline CXCL9/Cr LR, the combined GEP and dd-cfDNA-based two-stage test detected more AMR and cellular rejections than dd-cfDNA, with a similar detection rate for mixed rejections in both cohorts. Although the former two-stage model demonstrated a higher false-positive rate at the second stage than the dd-cfDNA-based two-stage model, the increase was due to the sensitivity of GEP toward BL rejections.^3,5

Several studies have explored the diagnostic utility of urine chemokine assays as standalone tests for detecting AR.^{8,13,16,23–29} Van Loon et al., for example, integrated urine CXCL9 and CXCL10 with clinical risk factors, reporting a sensitivity of 0.42, specificity of 0.95, PPV of 0.48, and NPV of 0.94 in a cohort with an AR prevalence of 25%. Similarly, Goutaudier et al., Tinel et al., and Yang et al. demonstrated that urine chemokine assays possess high negative predictive utility in various KT populations.^16,28,29 Although these studies emphasize the strong NPV of urine chemokines, they generally assess the assays in isolation. By contrast, our two-stage model leverages the affordability and screening value of urine chemokines to guide costlier blood-based tests.

Optimal threshold selection for urine chemokine assays has been addressed in several investigations.^23,27,30 Mačionienė et al. reported that CXCL9/Cr and CXCL10/Cr achieved moderate diagnostic performance (AUC=0.73) with optimal cutoffs of approximately 0.11 and 0.42 (ng/mmol), respectively.²⁷ In pediatric KT recipients, Blydt-Hansen et al. proposed a two-level threshold strategy for CXCL10—using a low threshold to safely defer biopsies and a high threshold to flag potential rejection.²³ Millán et al. also identified a 100 pg/ml threshold for plasma CXCL10.³⁰ These studies focused on maximizing sensitivity and specificity for optimal threshold selection. In our setting, although a classical Youden-index approach selected a baseline CXCL9/Cr LR probability threshold of 0.17, our integrated diagnostic and resource utilization analysis of the two-stage approach indicates that a lower PST of 0.10 can yield improved diagnostic performance with reduced resource utilization compared with a standalone stage 2 test, which was used for 0.79 (0.72 to 0.85) of validation cases.

Although most biomarker discovery has focused on ARs rather than chronic or other lesions, as indicated by allograft biopsy, we considered the inclusion of chronic rejection along with subclinical AR versus TX and showed that the two-stage models performed similarly for subclinical AR (Supplemental Tables 22 and 23).³¹ Sensitivity analysis of considering BL rejections as cellular rejection validated an increase in PPV from 0.24 (0.17 to 0.31) to 0.41 (0.31 to 0.51) for the two-stage model with combined GEP+dd-cfDNA testing due to greater detection of cellular rejections; however, we observed a lower AUC and NPV, underscoring the limited statistical power for the cellular phenotype. Further sensitivity analyses for two patient subgroups, based on age, time from transplant, donor type, and sex, validated similar diagnostic performance. However, the utilization of the second-stage test was significantly higher for testing in younger patients, testing in the second year, and testing deceased-donor recipients, owing to the lower diagnostic performance of urine chemokines in these patient subgroups.^23,32,33

Sensitivity analyses restricted to biopsies with available DSA results validated improvement in performance of all models while reducing stage-2 utilization. PPV and specificity were higher, particularly for AMR and mixed rejections, highlighting the added value of using DSA information for AMR diagnosis in the staged approach. At the same time, the two-staged approach without DSA information, where dd-cfDNA was used as a second-stage test, validated better overall diagnostic performance than standalone DSA testing, highlighting the importance of dd-cfDNA as an unbiased injury marker that may capture injury not reflected by DSA alone (Table 3 and Supplemental Table 28).^34,35 Our sensitivity analyses demonstrate that the two-stage models validated good performance both with and without DSA information, providing flexibility for implementation across different clinical settings. For centers that already perform prospective DSA screening, the staged urine chemokine approach may serve as a complementary, and more frequent screening tool between protocol DSA assessments, since urine chemokine assays can be performed more frequently and at lower cost than blood draws for DSA. Moreover, combining DSA with the staged approach achieved the highest AUCs for AMR, suggesting an additive benefit rather than redundancy.

In the validation cohort, the baseline CXCL9/Cr LR with a PST of 0.10 as a stage 1 test correctly identified 21.88% of the TX cases, and these cases were also early in the post-transplant follow-up period (mean=240.4 days [175.2 to 305.5]) and had a positive relative eGFR (mean=0.06 [0.02 to 0.11]). However, the very few false negatives at stage 1 (12.5% of subclinical AR cases) had low relative eGFR (mean=0.03 [−2.91 to 2.96]), possibly contributing to positive results by the CXCL9/Cr testing.²³ Moreover, half of these stage 1 false positives could have been diagnosed by the dd-cfDNA or the combined GEP and dd-cfDNA test, representing the trade-off between sensitivity and second-stage test utilization. However, by using the second-stage test only for 0.79 (0.72 to 0.85) of validation cases, the combined GEP and dd-cfDNA as the second-stage test correctly identified 75% of the rejections and an additional 47.7% of TXs. Thus, the two-stage strategy improved overall diagnostic yield relative to stage 2 testing alone, and reduced reliance on costly stage 2 testing and unnecessary invasive biopsies, improving patient experience while reducing resource utilization and, hence, the cost of care for KT recipients.

Despite its promising findings, our study has several limitations. First, we did not incorporate factors, such as proteinuria levels, known to affect urine chemokine levels.^8,16 Second, GEP in the validation dataset was measured using a PCR-based method, whereas the discovery cohort used a microarray-based assay with binary interpretation, reflecting evolution in testing methodology over time. Notably, the newer method has been validated and approved by Medicare for clinical use (Supplemental Tables 32 and 33). Although the choice of GEP platform likely has minimal effect on the binary GEP outcome, it could influence the precise validation threshold. Third, because cellular and mixed rejections were rare in both the discovery and validation cohorts, phenotype-specific estimates were underpowered and should be interpreted as exploratory; consequently, overall performance was primarily driven by the majority AMR phenotype and varied across phenotypes. For example, we were not able to validate a significant discrimination of cellular rejections by the single-stage models. Finally, due to limited DSA information, AMR phenotype included 33 suspicious AMR cases out of 64 in the CTOT-08 data.

In summary, our findings indicate that a two-stage diagnostic approach was more resource-efficient and diagnostically robust for surveillance of AMR in KT recipients than standalone testing. While the two-stage models also validated significant discrimination for cellular rejection, performance was substantially stronger and more clinically reliable for AMR. By leveraging the high NPV and affordability of urine chemokine assays to triage patients, this strategy minimizes the use of expensive blood-based tests, thereby reducing resource utilization while preserving clinical accuracy. Future prospective trials should validate these findings for cellular and mixed rejection and evaluate integrated surveillance strategies incorporating DSA as a key AMR indicator.

Supplementary Material

cjasn-21-1414-s001.pdf ^{(1.7MB, pdf)}

cjasn-21-1414-s002.pdf ^{(1.4MB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/CJN/C745.

Author Contributions

Conceptualization: Kenny Chen, John Friedewald, Steve Kleiboeker, Sanjay Mehrotra, Sook Park, Akhil Singla, Rohita Sinha, Lihui Zhao.

Data curation: Kenny Chen, Genevieve Hanna, Ronnie LaCombe, Sook Park, Christabel Rebello, Blade Robelly, Akhil Singla.

Formal analysis: Kenny Chen, Sanjay Mehrotra, Sook Park, Akhil Singla.

Funding acquisition: John Friedewald, Sook Park.

Investigation: Steve Kleiboeker, Sook Park, Akhil Singla.

Methodology: John Friedewald, Sanjay Mehrotra, Sook Park, Akhil Singla, Rohita Sinha, Lihui Zhao.

Project administration: Sanjay Mehrotra, Sook Park, Christabel Rebello.

Resources: Sook Park, Christabel Rebello.

Software: Sook Park, Christabel Rebello, Blade Robelly.

Supervision: John Friedewald, Steve Kleiboeker, Sook Park, Akhil Singla.

Validation: Kenny Chen, Sook Park, Akhil Singla.

Visualization: Sook Park, Akhil Singla.

Writing – original draft: Sook Park, Havisha Pedamallu, Christabel Rebello, Akhil Singla.

Writing – review & editing: John Friedewald, Steve Kleiboeker, Ronnie LaCombe, Connor Lantz, Sanjay Mehrotra, Sook Park, Akhil Singla, Lihui Zhao.

Funding

J. Friedewald: Division of Intramural Research, National Institute of Allergy and Infectious Diseases (U01 AI084146) and Eurofins Viracor BioPharma. This work was supported by Division of Intramural Research, National Institute of Allergy and Infectious Diseases (3U01 AI063594-07S1, 1U01AI088635,2U19 AI063603, and R34 AI118493).

Declarative Statements

This study includes clinical experimentation and received Institutional Review Board or Ethics Committee approval. All patients provided written informed consent. This study includes clinical experimentation and complies with the Declaration of Helsinki. The clinical and research activities being reported are consistent with the Principles of the Declaration of Istanbul as outlined in the Declaration of Istanbul on Organ Trafficking and Transplant Tourism.

Data Availability Statements

Original data generated for the study will be made available upon reasonable request to the corresponding author. Original data generated for the study are or will be made available in a public access repository upon publication. Data Type: Raw Data/Source Data and Clinical Trial Data. Reason for Restricted Access: The data is restricted because it is single center data that resides on university servers. Original data generated for the study will be made available upon reasonable request to the corresponding author. Repository Name: Gene Expression Omnibus. Linkable Citation: Gene Expression Omnibus Accession No. GSE107509.

Supplemental Material

This article contains supplemental material online, published as provided by the authors, at http://links.lww.com/CJN/C746.

Supplemental Methods and Materials

Supplemental Figure 1. Consolidated Standards of Reporting Trials diagram illustrating the number of patients and the samples available for analysis based on inclusion and exclusion criteria and sample availability for the validation MKBS cohort.

Supplemental Figure 2. Correlation matrix of the covariates for baseline LR models using urine chemokines CXCL9/Cr and CXCL10/Cr, along with time from transplant (TTR_DAYS), age, recent BKV positive (bkv_indicator), recent UTI (uti_indicator), and relative eGFR.

Supplemental Figure 3. Calibration of the baseline LR models for (A) CXCL9/Cr only, (B) CXCL10/Cr only, (C) for both CXCL9/Cr and CXCL10/Cr. Across all three models, the calibration intercept was near zero and the calibration slope was approximately one, indicating that the predicted probabilities were neither systematically too high nor too low.

Supplemental Table 1. Overfitting analysis of the baseline LR model showed modest or no performance.

Supplemental Table 2. Diagnostic performance of the CXCL9/Cr then GEP two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 3. Validation performance of the CXCL9/Cr then GEP two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 4. Diagnostic performance of the CXCL9/Cr then dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 5. Validation performance of the CXCL9/Cr then dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 6. Diagnostic performance of the CXCL9/Cr then the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 7. Validation performance of the CXCL9/Cr then the combined GEP and dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 8. Diagnostic performance of the CXCL10/Cr then GEP two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 9. Validation performance of the CXCL10/Cr then GEP two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 10. Diagnostic performance of the CXCL10/Cr then dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 11. Validation performance of the CXCL10/Cr then dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 12. Diagnostic performance of the CXCL10/Cr then the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 13. Validation performance of the CXCL10/Cr then the combined GEP and dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 14. Diagnostic performance of the combined CXCL9/Cr and CXCL10/Cr then GEP two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 15. Validation performance of the combined CXCL9/Cr and CXCL10/Cr then GEP two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 16. Diagnostic performance of the combined CXCL9/Cr and CXCL10/Cr then dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 17. Validation performance of the combined CXCL9/Cr and CXCL10/Cr then dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 18. Diagnostic performance of the combined CXCL9/Cr and CXCL10/Cr then the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 19. Validation performance of the combined CXCL9/Cr and CXCL10/Cr then the combined GEP and dd-cfDNA two-stage models for validation (MKBS) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25.

Supplemental Table 20. Diagnostic performance of the CXCL9/Cr then dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25, and validation at PST=0.10 in MKBS cohort. Rejections include the BL phenotypes instead of TX.

Supplemental Table 21. Diagnostic performance of the CXCL9/Cr then the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort with different PST of 0.05, 0.1, 0.15, 0.20, and 0.25, and validation at PST=0.10 in MKBS cohort.

Supplemental Table 22. Diagnostic performance of the CXCL9/Cr then dd-cfDNA two-stage models for discovery (CTOT-08) cohort with PST of 0.10.

Supplemental Table 23. Diagnostic performance of the CXCL9/Cr then the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort with PST of 0.10.

Supplemental Table 24. Diagnostic and resource utilization results for the two-stage model for two subgroups of patient samples, divided based on age, being less than the median age (n=247), and more than the median age (n=213) in the CTOT-08 cohort and validation in the MKBS cohort.

Supplemental Table 25. Diagnostic and resource utilization results for the two-stage model for two subgroups of patients, divided based on time from transplant, being <365 days (n=272) and more than 365 days (n=188) in the CTOT-08 cohort and validation in the MKBS cohort.

Supplemental Table 26. Diagnostic and resource utilization results for the two-stage model for two subgroups of patients, divided based on the type of kidney, whether living donor (n=295) or deceased donor (n=165) in the CTOT-08 cohort and validation in the MKBS cohort.

Supplemental Table 27. Diagnostic and resource utilization results for the two-stage model for two subgroups of patients, divided based on sex, being male (n=296) or female (n=164) in the CTOT-08 cohort and validation in the MKBS cohort.

Supplemental Table 28. Biopsy phenotype for 211 biopsies for CTOT-08 and 45 for MKBS with mapped DSA information and the DSA status matrix.

Supplemental Table 29. Baseline LR model with CXCL9/Cr adjusted for DSA information.

Supplemental Table 30. Performance of single-stage: urine chemokines to detect subclinical AR in discovery (CTOT-08) cohort for 211 biopsies with available DSA information.

Supplemental Table 31. Diagnostic performance of the CXCL9/Cr then dd-cfDNA and the combined GEP and dd-cfDNA two-stage models for discovery (CTOT-08) cohort for 211 biopsies and validation (MKBS) cohort for 45 biopsies with available DSA information with a PST of 0.15.

Supplemental Table 32. Validation of microarray- and quantitative polymerase chain reaction-based GEP assay on the same 445 samples run by both assays, reporting the diagnostic performance compared with biopsy-confirmed rejection and TX.

Supplemental Table 33. Comparison of genes comprising microarray and PCR TruGraf (GEP) assays. Table lists each gene in both assays and provides a direct comparison of gene overlap between the microarray and 139 gene PCR assay.