Real‐World Associations of KidneyIntelX Risk Stratification With Guideline‐Directed Therapy, Kidney Outcomes, and Metabolic Trajectories in Early Diabetic Kidney Disease

ABSTRACT

Background

KidneyIntelX is a prognostic blood test for the progression of diabetic kidney disease (DKD) within 5 years. Long‐term utility is important.

Materials and Methods

Adults with type 2 diabetes (T2D) and CKD (G1‐G3b) from two medical centres were assessed for the following: (1) baseline KidneyIntelX risk levels and their association with GDMT usage; (2) changes in temporal kidney and metabolic trajectories after testing; (3) probability for a kidney composite outcome of sustained eGFR decline of 40% or kidney failure by baseline KidneyIntelX risk strata; and (4) risk at 12‐month for repeat‐tested participants.

Results

2470 patients with T2D and CKD (median baseline eGFR 63 mL/min/1.73 m², UACR 51 mg/g, and HbA1c 7.2%), KidneyIntelX risk was 49.9% low, 40.6% intermediate, and 9.5% high. SGLT2i usage post‐test increased to 56%, 45%, 34% in high, intermediate, low‐risk groups. SGLT2i combined with GLP‐1 increased to 32%, 25%, 17% per risk group. Improvements in eGFR slope (−4.6 to −2.6 mL/min/1.73 m²/year), UACR (relative decrease of 23%), and HbA1c (relative decrease of 7.6%) were greatest in the high‐risk group. Hazard ratio (HR) for DKD progression for high vs. low was 10.4 [4.4–24.8] and intermediate vs. low HR was 4.0 [2.0–8.7]. With repeat testing (n = 1143), 29% transitioned to lower risk accompanied by biomarker reductions, specifically KIM‐1, and initiation of SGLT2i or GLP‐1 doubled odds of risk reduction.

Conclusions

KidneyIntelX was associated with targeted GDMT usage with favourable changes in kidney and metabolic clinical measures and risk trajectories, with lower risk linked to therapy initiation and reduction in biomarkers.

1. Introduction

Many individuals with diabetic kidney disease (DKD) progress to kidney failure without timely initiation of guideline‐directed medical therapy (GDMT) and lifestyle changes [1]. While there has been some recent progress in the increasing use of SGLT2i and GLP1a, focusing on primary care is essential for identifying patient risk, slowing disease progression and reducing morbidity [2].

KidneyIntelX test combines three biomarkers, soluble tumour necrosis factor (sTNFR1/2), and kidney injury molecule‐1 (KIM‐1) with routine clinical variables in an algorithm to stratify patients with type 2 diabetes and chronic kidney disease (CKD) stages G1‐G3b into low, intermediate, or high risk for progressive decline in kidney function [3, 4, 5, 6]. Prior studies demonstrated that the test was effective in predicting which patients with early‐stage DKD progressed over a 5‐year period [3, 5]. Subsequent post hoc analyses on patient samples from randomised controlled trials [4, 6] revealed that initiation of SGLT2i therapy was correlated with significant reductions in KidneyIntelX risk over a 1‐year period. These risk changes were strongly associated with improved kidney outcomes [6], underscoring the potential for early risk‐based intervention. Additionally, in previously published real‐world evidence (RWE) data, KidneyIntelX testing promoted changes in clinical decisions and outcome measures of kidney and metabolic health within a 12‐month period post‐testing at a single health care system [7].

We sought to extend these previous findings via the following goals: (1) to evaluate longer term baseline KidneyIntelX risk and association with GDMT in routine care; (2) to interrogate longitudinal kidney and metabolic trajectories; (3) to assess risk with a kidney composite outcome in a contemporary setting; and (4) to evaluate 12‐month risk upon repeat‐testing, all in participants enrolled in RWE studies across two health systems.

2. Methods

2.1. Study Design and Data Source

Participants were recruited into RWE studies at the Mount Sinai Health System (MSHS), New York, NY (NCT05198284), and Wake Forest/Atrium Health (WFA), North Carolina (NCT04960514). The present analysis evaluated the real‐world clinical impact, prognostic performance, and longitudinal risk dynamics associated with KidneyIntelX testing over a follow‐up period of up to 36 months. All analyses were conducted according to a statistician verified analytic plan. A patient consented Renalytix Accelerating Precision Diagnostics (RAPiD) Biobank was also instituted to collect clinical data and biospecimens (urine/blood) for future research and currently includes 2470 baseline patients with a subset of 1143 subjects having both baseline and follow‐up (median 12 months) samples.

2.2. Study Population

Adults ≥ 18 years with T2D and G1‐G3b CKD (estimated glomerular filtration rate [eGFR] ≥ 30 mL/min/1.73 m² and/or uACR ≥ 30 mg/g) were prospectively enrolled between October 2021 and October 2023 across both systems, representing ~200 unique providers. All participants had baseline KidneyIntelX testing with median follow‐up from baseline was 21.3 months. A subset of participants (n = 1143) had a repeat KidneyIntelX test approximately 12 months after baseline as per protocol‐defined longitudinal assessment (median 12.0 months; IQR 10.0–13.0).

2.3. Ethical Approval and Consent

The RWE and associated RAPiD Biobank protocols were individually approved by the MSH (IRB#s 21‐00165 and 21‐01927) and WFA (#IRB00070497). All participants provided written informed consent for biospecimen collection, longitudinal data linkage and secondary analyses.

2.4. Data Collection and Variable Definitions

Data were extracted from the Electronic Health Record (EHR) at both institutions using harmonised data pipelines. Collected variables included demographics, clinical diagnoses, outpatient and inpatient encounters, laboratory measurements, medication prescriptions, and KidneyIntelX biomarker and risk levels. Baseline clinical variables included age, sex, race, ethnicity, estimated glomerular filtration rate (eGFR), urine albumin‐to‐creatinine ratio (UACR), haemoglobin A1c (HbA1c), systolic blood pressure (SBP), and comorbid conditions. Medication exposure included prescriptions for sodium‐glucose cotransporter‐2 inhibitors (SGLT2i), glucagon‐like peptide‐1 receptor agonists (GLP‐1 RA), renin–angiotensin–aldosterone system inhibitors, and other glucose‐lowering agents. Medication use was based on active prescriptions in the EHR during predefined analysis windows of 12 months before baseline (prior therapy exposure), 0–12 months and 12–24 months post‐baseline. For combination therapy analyses, exposure required concurrent active prescriptions for both an SGLT2 inhibitor and a GLP‐1 receptor agonist; no minimum overlap duration was imposed to reflect real‐world prescribing patterns.

2.5. KidneyIntelX Test

KidneyIntelX is a prognostic test that combines three plasma biomarkers, sTNFR1, sTNFR2, and KIM‐1, with three clinical lab variables (UACR, HbA1c, and BUN) to estimate risk of a progressive decline in kidney function defined as sustained 40% decline in eGFR or kidney failure (eGFR < 15 mL/min/1.73 m²) [3, 4, 5, 6, 7]. Based on validated thresholds, patients are categorised into low, intermediate, or high‐risk for a progressive decline in kidney function over a 5 year period [3, 4, 5, 6, 7]. Baseline risk categorisation was defined by the first KidneyIntelX test obtained during the study period. For participants with repeat testing, risk transitions were characterised by changes in categorical risk assignment between baseline and follow‐up testing.

2.6. Observed Outcomes

2.6.1. Guideline‐Directed Medical Therapy

Initiation and sustained use of SGLT2 inhibitors alone or in combination with GLP‐1RA, as representative provider therapy choice, were evaluated longitudinally across baseline KidneyIntelX risk strata and adjusted for age, sex, baseline eGFR, log‐UACR, HbA1c, BMI, cardiovascular disease and health system (site).

2.6.2. Clinical and Laboratory Measurements

Clinical outcomes included temporal changes in cardiovascular, kidney, and metabolic measures, specifically eGFR and eGFR slope, UACR, HbA1c, and SBP in participants with at least one measurement available per time window. Slope of eGFR was calculated using linear regression of a minimum of three serial eGFR measurements taken in time windows before and after baseline KidneyIntelX testing, with slopes expressed as mL/min/1.73 m² per year.

2.6.3. KidneyIntelX Prognostic Performance

The risk of a progressive decline in kidney function was assessed over the follow‐up of up to 36 months. Time‐to‐event was from baseline KidneyIntelX to first occurrence of the composite outcome, censoring at last available follow‐up. A sustained eGFR decline of 40% was confirmed at least 90 days after the initial value and a kidney failure event with a confirmatory eGFR at least 30 days after the initial value with < 15 mL/min/1.73 m² threshold.

2.6.4. Longitudinal Risk Reassessment Analysis

Changes in KidneyIntelX risk levels were evaluated for associations with changes in biomarker concentrations (e.g., sTNFR1, sTNFR2, KIM‐1), clinical lab values (eGFR, UACR, HbA1c), and medication initiation. Risk transitions were categorised as improvement (movement to a lower risk category), stability, or worsening (movement to a higher risk category). Univariable and multivariable logistic regression models were used to evaluate factors associated with risk transitions.

2.6.5. Statistical Analysis

Baseline characteristics were summarised using medians with interquartile ranges for continuous variables and frequencies with percentages for categorical variables. Time‐to‐event analyses were conducted using Kaplan–Meier (KM) methods, with differences across risk groups assessed with Cox proportional hazards models employed to estimate unadjusted and adjusted hazard ratios (HRs) for the composite kidney outcome. Adjusted multivariable Cox models included sex, race, baseline eGFR, log uACR, and HbA1c and site. Proportional hazards were evaluated using Schoenfeld residuals, which demonstrated that proportionality was not violated. Continuous outcomes (UACR, HbA1c, eGFR, SBP) were analysed using patient‐level linear mixed‐effects models; skewed outcomes were log‐transformed; least‐squares means and contrasts versus baseline were exponentiated to yield percentage changes with 95% confidence intervals. Medication‐use percentages are descriptive observed proportions, and changes in between‐group contrasts over time are described using binomial models with group, time, and group‐by‐time interaction terms. Changes in medication use were evaluated longitudinally across baseline KidneyIntelX risk strata and logistic regression models adjusted for age, sex, baseline eGFR, log‐UACR, HbA1c, BMI, cardiovascular disease, and site. All statistical tests were two‐sided, with a significance threshold of p < 0.05. Analyses were performed using R (version 4.3.2). Figures were generated using ggplot2 and survival packages.

3. Results

2470 patients across two healthcare systems underwent KidneyIntelX testing at baseline (Table 1) with median follow‐up of 21.3 months (IQR 14.4–27.8). Median eGFR was 63 mL/min/1.73 m², UACR 51 mg/g and HbA1c 7.2%. KidneyIntelX risk distribution at baseline was low 49.9% (n = 1232), intermediate 40.6% (n = 1004) and high 9.5% (n = 234). Higher baseline HbA1c, SBP, UACR, biomarker concentrations and SGLT2i/GLP1‐RA usage were observed with increasing KidneyIntelX risk, whereas baseline eGFR did not show a monotonic trend. While baseline characteristics were similar between the two health systems, patients at Mount Sinai were much more likely to have been prescribed RAAS inhibitors (70% vs. 23%), SGLT2i (38% vs. 7%), and GLP1‐RAs (47 vs. 10%) at baseline (Table S1).

TABLE 1.

Baseline characteristics of the total population stratified by KidneyIntelX Risk Level.

Characteristic	Total	KidneyIntelX Risk Level			p trend
		Low	Intermediate	High
Participants, n	2470	1232	1004	234
Age in years, Median [IQR]	67 [59, 74]	68 [60, 75]	65 [57, 73]	64 [56, 70]	< 0.001
Female % (n)	54.3% (1340)	55.5% (684)	53.8% (540)	49.6% (116)	0.102
Self‐Reported Race, % (n)
Race: White	47.3% (1168)	48% (591)	50.1% (503)	31.6% (74)	0.009
Race: Black	34.7% (858)	34.3% (423)	32.7% (328)	45.7% (107)
Race: Other	18% (444)	17.7% (218)	17.2% (173)	22.7% (53)
Baseline Systolic blood pressure in mmHg, Median [IQR]	132 [121, 146]	129 [119, 139]	134 [122.8, 150]	140 [124, 156]	< 0.001
Baseline Body Mass Index in kg/m2, Median [IQR]	32.2 [28, 37.3]	31.8 [28, 36.6]	33 [28.3, 38.8]	30.8 [26.6, 35.3]	0.335
Baseline Body Mass Index in kg/m2, Missing % (n)	3.9% (97)	3% (37)	4.1% (41)	8.1% (19)
History of cardiovascular disease, % (n)	27.2% (673)	26.7% (329)	26.4% (265)	33.8% (79)	0.131
Lab values at baseline
Baseline A1C, Median [IQR]	7.2 [6.5, 8.5]	6.9 [6.4, 7.8]	7.6 [6.6, 9.5]	8.2 [7, 10.3]	< 0.001
Baseline A1C Missing % (n)	1.1% (28)	0.8% (10)	1.3% (13)	2.1% (5)
Urine albumin creatinine ratio in mg/g, Median [IQR]	51 [21, 153]	34 [12, 67]	84 [36.1, 257.5]	474 [107.5, 1378.3]	< 0.001
UACR < 30 mg/g, % (n)	29.1% (718)	41.9% (516)	18.1% (182)	8.5% (20)	< 0.001
UACR 30–300 mg/g, % (n)	53.9% (1331)	54.4% (670)	58.3% (585)	32.5% (76)
UACR > 300 mg/g, % (n)	14.2% (351)	1.4% (17)	21% (211)	52.6% (123)
UACR mg/g, Missing % (n)	2.8% (70)	2.3% (29)	2.6% (26)	6.4% (15)
Baseline estimated glomerular filtration rate in mL/min/1.73 m2, Median [IQR]	63 [49.7, 87.2]	60.3 [50.5, 84.2]	69.1 [50.7, 91.4]	53.9 [40.8, 74.6]	0.649
Baseline eGFR < 60 mL/min/1.73 m2, % (n)	44.7% (1105)	48.6% (599)	37.4% (375)	56% (131)	0.359
Baseline eGFR ≥ 60 mL/min/1.73 m2, % (n)	53.2% (1313)	49.9% (615)	60.4% (607)	38.9% (91)
Baseline eGFR mL/min/1.73 m2, Missing % (n)	2.1% (52)	1.5% (18)	2.2% (22)	5.1% (12)
Medication use
RAAS inhibition, %(n)	21.3% (525)	21.1% (260)	20.6% (207)	24.8% (58)	0.535
Baseline SGLT2 inhibitors, % (n)	19.2% (474)	17.8% (219)	19.4% (195)	25.6% (60)	0.011
Baseline GLP1 agonists, % (n)	23.8% (588)	21.9% (270)	24.6% (247)	30.3% (71)	0.005
Baseline SGLT2 inhibitors and GLP1 receptor agonists, % (n)	9.6% (237)	9.2% (113)	9.5% (95)	12.4% (29)	0.221
KidneyIntelX risk categories, % (n)
KidneyIntelX High Risk	9.5% (234)	0% (0)	0% (0)	100% (234)
KidneyIntelX Intermediate Risk	40.6% (1004)	0% (0)	100% (1004)	0% (0)
KidneyIntelX Low Risk	49.9% (1232)	100% (1232)	0% (0)	0% (0)
KIM1, Median [IQR]	104 [66, 175]	79 [54, 111]	137 [82, 212]	465 [361, 632.8]	< 0.001
TNFR1, Median [IQR]	3166.5 [2357.5, 4253]	2908 [2221.5, 3675.5]	3457.5 [2467.5, 4720.2]	4493 [3160.2, 5876]	< 0.001
TNFR2, Median [IQR]	11 370 [8481, 15102.8]	10242.5 [7932.2, 13156.2]	12557.5 [8918, 16971.8]	15 320 [11 527, 21 304]	< 0.001

Note: p‐trend values were obtained by modelling KidneyIntelX risk category as an ordinal numeric variable (Low = 1, Intermediate = 2, High = 3) and testing the significance of the ordinal term, representing a linear trend across categories. Linear regression was used for continuous variables, logistic regression for eGFR bins, and ordinal logistic regression for ordered UACR categories.

3.1. Association of KidneyIntelX Risk Levels With Changes in Care and Outcomes

3.1.1. Initiation of GDMT

An increase in SGLT2i use was observed over the 2 year follow up period across all risk levels with a relative percentage increase in the high‐risk group of 115% (absolute usage of 26% pretest to 56% post‐test), a 136% relative increase in the intermediate risk group (absolute usage of 19% pretest to 45% post‐test), and an 88% relative increase in low risk (18% pretest to 34% post‐test). Trends were similar across both sites, with absolute SGLT2i use among high‐risk patients reaching 70% at Mount Sinai. At Wake Forest/Atrium Health, although absolute utilisation of SGLT2i of 42% was lower than MSHS, the percentage increase from baseline in the high and intermediate risk groups was nearly fivefold (Table S2). When comparing the overall difference in utilisation between risk groups at baseline and over a two‐year period, there was a marked increase in SGLT2i usage in high and intermediate risk groups when compared to the low risk group; the difference in use was greater than fourfold, increasing from a 2.8% difference at baseline to a 13.1% difference 2 years post baseline (Figure 1A). After adjustment for age, sex, eGFR, log‐UACR, HbA1c, BMI, cardiovascular disease and site, intermediate/high vs. low baseline risk remained associated with greater odds of SGLT2i initiation within 24 months (adjusted OR 1.54, 95% CI 1.22–1.93; p = 0.0002).

FIGURE 1.

Changes in use of kidney protective therapies across risk levels from 12‐month period prior to baseline to 24‐month period post‐baseline. Panel A: Difference in the percentage of patients using SGLT2i in individual risk groups versus comparator risk group in the 12‐month period prior to baseline kidneyintelX.dkd test compared to the difference in percentage in the 24‐month period after the kidneyintelX.dkd test. Odds Ratio for SGLT2i usage in intermediate/high versus low group: 1.57 (95% CI 1.12–2.21). Panel B: Shows the same information for utilisation of SGLT2i in combination with GLP1 RA. Odds Ratio for SGLT2i/GLP1 RA usage in intermediate/high vs. low group: 1.94 (95% CI 1.32–2.85). To assess whether differences in SGLT2 and SGLT2 plus GLP‐1 use across risk groups changed from baseline to Year 2, binomial generalised linear models were built with a group‐by‐time interaction term to assess whether differences between risk groups changed over time. p‐values for comparisons of proportions for High vs. Low, High and Intermediate versus Low, and Intermediate versus Low were all < 0.05; expect for the following: p‐value for SGLT2i and GLP1 RA utilisation in high versus low was 0.09; p‐value for high vs. intermediate for SGLT2i and SGLT2i plus GLP1 RA utilisation were 0.68 and 0.87, respectively.

Combination SGLT2i and GLP1‐RA therapy in high‐risk patients increased from 12% at baseline to 32% over 2 years, with the difference in utilisation compared to the low‐risk group increasing from 3.2% at baseline to 14.4% over 2 years. Similar trends were observed when comparing high and/or intermediate risk groups to low risk group (Figure 1B). After adjustment for age, sex, eGFR, log‐UACR, HbA1c, BMI, and cardiovascular disease and site, intermediate/high versus low baseline risk remained associated with greater odds of combination SGLT2i plus GLP‐1 RA initiation (adjusted OR 1.54, 95% CI 1.17–2.05; p = 0.0025).

3.1.2. Trajectories of Cardiovascular, Kidney and Metabolic Measures

The least squares mean percentage change for UACR and HbA1C showed consistent and sustained reduction from the period 12 months prior to baseline to the periods 1 year and 2 years post baseline with the greatest improvements observed in the high (23% reduction for UACR; p = 0.003 vs. baseline, 7.6% reduction for HbA1c; p < 0.001 vs. baseline) and intermediate risk groups (24% reduction for UACR; p < 0.001 vs. baseline, 5.7% reduction for HbA1c; p < 0.001 vs. baseline). The low‐risk group also experienced a reduction of 11% for UACR and 2.2% for HbA1C (Figure 2A,B). Systolic blood pressure was largely stable across all three risk groups over time (Figure 2D).

FIGURE 2.

Changes in urinary albumin‐to‐creatinine ratio (UACR), estimated glomerular filtration rate, HbA1c%, systolic blood pressure, and eGFR slopes. Panel A–D shows the least‐squares mean percentage change from 12 months prior to baseline to the 12 months and 24 months period post baseline in the High, Intermediate, and Low risk groups for Panel A. Urinary albumin‐to‐creatinine ratio, Panel B. Haemoglobin A1c. Panel C. Estimated glomerular filtration rate (eGFR), Panel D. Systolic blood pressure. Panel E shows the median intra‐individual slope of eGFR across each time‐period. (A) UACR: % Reduction from Prior to Baseline over 1 year and 2‐year post baseline. (B) HbA1c: % Reduction from Prior to Baseline over 1 year and 2‐year post baseline. (C) eGFR: % Reduction from Prior to Baseline over 1 year and 2‐year post baseline. (D) Systolic Blood Pressure: % Reduction from Prior to Baseline over 1 year and 2‐year post baseline. (E) eGFR Slope by risk group over 1 year pre‐baseline, 1 year post‐baseline and 2‐year post baseline (ml/min/1.73 m² per year).

The median eGFR decreased over time and aligned with KidneyIntelX risk (Figure 2C). The rate of eGFR decline from pre‐baseline to post‐baseline ameliorated in all 3 risk groups, with the greatest relative attenuation seen in the patients that were high risk at baseline (pre‐test eGFR slope −4.6 mL/min/1.73 m² vs. −2.6 mL/min/1.73m² post‐test; p < 0.001; Figure 2E).

3.1.3. KidneyIntelX Prognostic Performance in Contemporary Setting

During a median follow‐up period of 21.3 months (IQR 14.4 to 27.8, maximum 36), 92 (3.9%) participants developed the composite kidney outcome (sustained 40% decline in eGFR or kidney failure (eGFR < 15 mL/min/1.73 m²)). The cumulative incidence probability of the outcome followed the baseline KidneyIntelX risk levels; 25.4% of the high‐risk group, 8.1% of the intermediate risk group, and 1.5% of the low‐risk group. The hazard ratio (HR) for high vs. low was 26.6 [95% CI 12.8–55.1] and for intermediate vs. low 6.5 [95% CI 3.2–13.2]. After adjustment for clinical covariates (race, sex, eGFR, log uACR, and HbA1c) and site, there was still significant independent associations between KidneyIntelX risk levels and the outcome (adjusted HR for high vs. low: 10.4 [95% CI 4.4–24.6], and for intermediate vs. low: 4.3 [95% CI 2.1–9.1]; Figure 3).

FIGURE 3.

Cumulative incidence curves for 40% decline in eGFR or kidney failure up to a maximum of 36 months follow up. *Hazard ratio.

3.1.4. Longitudinal Reassessment of KidneyintelX Risk Levels

A subset of 1143 (46.3%) patients had a repeat test at a median 12.0 months (IQR 10.0–13.0) with similar baseline HbA1c, eGFR and uACR levels but lower baseline SGLT2i (13.8% vs. 23.8%) and GLP‐1‐RA (17.1% vs. 29.5%) use than participants (n = 1327) without repeat testing (Table S3). On repeat testing, 31% of high‐risk patients at baseline moved to a lower risk level with 69% remaining in the high‐risk category. Conversely, for those at intermediate risk at baseline, just 7% of individuals increased to high‐risk with 29% moving to low risk. Among those initially categorised as low risk, 25% moved to a higher risk level and 75% remained at low risk (Figure 4).

FIGURE 4.

Transition between KidneyIntelX risk groups on re‐assessment at risk at Median of 12 months post baseline test. Sankey plot demonstrating the shifts in risk categories from baseline to 1 year.

3.1.5. Factors Associated With Changes in Risk Level

In participants that moved to a lower risk level or to a higher risk level when re‐tested by KidneyIntelX, the corresponding changes in all three individual biomarkers (KIM‐1, sTNFR1, and sTNFR2) were significantly associated with these risk level transitions (p < 0.05). In multivariable models, changes in all biomarkers and clinical inputs of UACR, BUN and HbA1c; only changes in KIM‐1 remained a significant driver across all risk transitions (Table S4).

In the combined high and intermediate risk groups, 401 subjects (35% of total retested population) were not on an SGLT2i or a GLP1 prior to baseline. Of these, 108 (27%) were started on one or both therapies between baseline and repeat test, with 45 (41.6%) moving to a lower risk group. The remaining 293 subjects (63%) not on SGLT2i or a GLP1 therapy at baseline remained naïve to these therapies between baseline and repeat test, with 79 (27%) moving to a lower risk group. This translated into an odds ratio of 1.93 (CI 1.22–3.07) for a reduction in risk level from high to intermediate/low or intermediate to low risk when treatment with SGLT2i or GLP1 is initiated in patients scored as high and intermediate risk at baseline.

4. Discussion

In two geographically diverse US health systems, real‐world implementation of KidneyIntelX was associated with a sustained increase in use of GDMT, attenuation in eGFR decline rate, improvements in measures of kidney and metabolic health over an extended 2‐year period post‐test and importantly, reduction of risk levels. Strong prognostic stratification for kidney outcomes was achieved by KidneyIntelX risk levels at baseline. Collectively, these findings support the clinical utility of KidneyIntelX for guiding risk‐based treatment decisions and therapeutic escalation where it is most likely to yield benefit in DKD [8].

Use of SGLT2i during the intervention period was observed to increase across all risk strata, with the steepest gains in the intermediate and high‐risk patients. In these combined risk groups, notable separation was observed from low‐risk with a positive differential gain in utilisation of 10.3 percentage points overall. Indeed, rates of SGLT2i inhibitors are increasing annually in indicated populations [9, 10, 11]. Although there was no matched control group in our analyses to strengthen the inference that the elevated KidneyIntelX risk was driving the increase observed in the absolute and relative usage rates post testing implies that the risk stratification was responsible, at least in part, for the drastic uptick in additional usage that could be explained by secular trends alone which have been demonstrated to be slow and gradually increasing over time [9, 10, 11].

Importantly, the combination of SGLT2i and GLP1‐RA therapy usage was observed to increase from 12% to 32% in pre vs. post‐tested high‐risk patients, consistent with clinicians using the risk classification to intensify care in accordance with clinical guidelines which suggest combination therapy may be most suitable for patients at the highest level of risk [12]. Furthermore, the more modest uptake in stable low‐risk patients suggests that KidneyIntelX may help avoid unnecessary polypharmacy in those unlikely to progress rapidly. These patterns are aligned with emerging care models which emphasise precision population health through individualised risk stratification targeting those patients most likely to benefit [8]. Leveraging prognosis can support prioritisation of GDMT intensification, more targeted patient counselling, closer follow‐up for high‐risk individuals, and more efficient allocation of specialty resources while avoiding overtreatment in stable low‐risk patients.

This study also demonstrates the durability of the improvements in rate of eGFR decline after the baseline over an extended 2‐year period. While the absolute number of composite kidney events was still low (3.9%) in the relatively short follow up period, the improved trajectories of eGFR can be expected to translate into favourable reductions events with extended follow‐up. As shown in prior studies [13], utilisation of SGLT2i could potentially delay Kidney Failure by 15 years for patients with DKD [14]. This attenuation in the rate of eGFR decline in our study was accompanied by significant reductions in UACR and HbA1C, while SBP was stable across all risk groups. In light of the increased use of SGLT2i post‐baseline we observed, our findings appear consistent with findings of RCTs of SGLT2i that demonstrate an initial reduction in eGFR due to the reduction in intraglomerular pressures, but an improvement in eGFR slope and reductions in UACR and HbA1c [13, 15, 16, 17, 18, 19].

The observed prognostic performance in the context of increased risk awareness and significant increases in use of GDMT over the relatively short follow up period is consistent with prior analyses [3, 4, 5]. The adjusted HR for the composite kidney outcome, even after adjustment remained robust.

Overall, most patients remained stable in their risk category from baseline to year 1; however, a meaningful proportion (30% overall) experienced a change in their future risk profile on repeat testing, potentially adding important information for care decisions to slow or prevent disease progression. As has been shown in published analyses from treatment and placebo arms from completed RCTs of SGLT2i, changes in KidneyIntelX risk levels were ‘prognostic at baseline, modifiable with treatment and dynamically and independently associated with clinical outcomes’ [6]. These observations also suggest KidneyIntelX may have utility as a longitudinal risk assessment tool, particularly in settings where periodic reassessment can establish an updated baseline following initiation of therapy. The changes in risk were orthogonal to absolute eGFR values, which primarily were lower at each time point, but were in alignment with time‐updated UACR at 1 and 2 years. The integration of 6 inputs in KidneyIntelX, including and yet additive to UACR, with the 3 biomarkers and 2 other clinical features, without eGFR as an input, can solidify biological responses in kidney health over time with greater precision than changes in UACR alone, which vary significantly at an individual level. We demonstrated that changes in all 3 biomarkers (KIM‐1, TNFR1, TNFR2) underpinned all risk transitions. In contrast, there was a less consistent or significant association of routinely monitored clinical features of eGFR, UACR and HbA1c with risk movements. In a multivariable model and in this setting of re‐assessment at 12 months in the presence of significantly increased utilisation of SGLT2i, KIM1 remained the most significantly associated feature with all risk level transitions, which is consistent with prior findings [6].

The results show the value and advantages of biomarker‐based risk assessment with KidneyIntelX. It indicates a role for incentivising both physician and patient behaviours towards effective use of GDMT, especially for those patients that stand to derive the most benefit, facilitating optimal and efficient care, as well as for gauging response to therapy. In theory, more aggressive actions and/or additional medications can and should be applied to those patients without positive changes in risk after 1 year after application of 1 or 2 pillars of GDMT [8].

There are limitations to this study including the observational design without randomised treatment assignment and lack of a formal control group which limit causal inference. While there are secular trends in adherence to GDMT, the observed inflection of increased usage in the current study is far greater than that explained by such trends [9, 10, 11]. These findings also extend prior analyses that demonstrated KidneyIntelX risk stratification influenced a marked increased usage of GDMT compared to an untested control group over the same time period at the MSHS [20]. Although both health systems represent diverse population with baseline differences in patient characteristics and GDMT the consistency of the association between KidneyIntelX risk and the uplift in usage of GDMT with associated improvements in cardiometabolic parameters, and prognostic performance support the generalisability of the findings. Another notable limitation was how medication exposure was captured, that is, as EHR orders rather than filled prescriptions, which may overestimate true adherence. Unfortunately, documented adherence and filled prescriptions were not consistently available across both systems. Future prospective trials are needed to test provider behaviour with respect to biomarker‐guided treatment intensification as well as longer‐term kidney and cardiovascular outcomes beyond the associations observed in this real‐world implementation cohort.

In conclusion, analyses from the KidneyIntelX Real‐World Evidence Clinical Utility studies and the consented linked RAPiD Biobank demonstrated that risk‐guided care using KidneyIntelX in real‐world settings was associated with clinically meaningful changes in both treatment patterns and risk trajectories, as well as favourable changes in kidney and metabolic clinical measures. These results support broader implementation of biomarker‐driven risk‐based strategies in the management of diabetic kidney disease, emphasising the importance of precision medicine as a driver behind learning health systems that promote preventative proactive care. Overview of population included in the study analysis stratified by health system and presence of baseline and repeat test (Figure S1).

Funding

The study was sponsored by Renalytix, PLC. The sponsor funded study operations and analysis support, there was a prespecified analytic plan and statistician verification, and all authors contributed to interpretation, drafting, and the decision to submit.

Conflicts of Interest

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: B.I.F. was a consultant for and received research support from Renalytix at the time of this research. M.J.D. and F.F. are employees of Renalytix; S.G.C. and G.N. are consultants for Renalytix. D.L. and J.T. received research support from Renalytix. The other authors declare no conflicts of interest. KidneyIntelX is based on technology developed by Mount Sinai faculty and licensed to Renalytix AI Inc. Mount Sinai faculty members are co‐founders and equity owners in the company. In addition, the Icahn School of Medicine at Mount Sinai has equity ownership in Renalytix AI.

Supporting information

Figure S1: Overview of population included in the study analysis stratified by health system and presence of baseline and repeat test.

Table S1: Baseline characteristics of the total population and subgroups defined by Health System enrollment.

Table S2: Longitudinal changes in use of guideline‐directed medical therapies and clinical parameters.

Table S3: Characteristics of participants that had repeat test at 1 year versus those without a repeat test.

Table S4: Association of changes in biomarkers and clinical features with risk level movements from baseline to year 1.

Table S5: Acknowledgements.

Lam D., Freedman B. I., Tokita J., et al., “Real‐World Associations of KidneyIntelX Risk Stratification With Guideline‐Directed Therapy, Kidney Outcomes, and Metabolic Trajectories in Early Diabetic Kidney Disease,” Diabetes, Obesity and Metabolism 28, no. 9 (2026): 8428–8437, 10.1111/dom.70993.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.