Renal Ultrasound Findings and Estimated Glomerular Filtration Rate (eGFR): A Cross-Sectional Observational Study

Abstract

Background: Ultrasound (US) imaging is widely used in Nephrology for the non-invasive assessment of renal morphology and perfusion. This study investigates correlations between sonographic parameters and renal function measured as estimated glomerular filtration rate (eGFR). Methods: This single-center prospective cross-sectional study enrolled 130 patients undergoing renal ultrasound. Parameters included renal length, parenchymal thickness, cortical–medullary differentiation, renal volume, and intrarenal resistive index (IR). eGFR was calculated using the CKD-EPI formula. Statistical analysis assessed correlations and developed a multivariable predictive model. Results: Renal length and parenchymal thickness correlated positively with eGFR (r = 0.381 and 0.364, p < 0.001), while IR correlated negatively (r = −0.549, p < 0.001). Multivariate regression identified sex, renal length, IR, cortical–medullary differentiation, and solitary/shrunken kidney as significant predictors of eGFR. The final model showed a predictive correlation coefficient of r = 0.6632. Specific ultrasound parameters, particularly renal length and IR, show significant correlation with eGFR. Conclusions: A predictive model incorporating these factors may assist in estimating renal function non-invasively.

1. Introduction

Chronic kidney disease (CKD) is a global health problem, affecting an estimated 10–15% of the population worldwide. CKD progression is associated with an increased risk of cardiovascular events, kidney failure, and mortality. Early detection and monitoring are critical to guide therapeutic interventions and slow disease progression.

Renal function is commonly assessed through serum creatinine and estimation of the glomerular filtration rate (eGFR). However, these methods may not fully capture structural renal changes. Imaging, particularly ultrasonography, plays a crucial role in nephrology by offering a non-invasive, real-time, and radiation-free evaluation of renal morphology and perfusion.

Determining estimated GFR by ultrasound evaluation is not possible. Solitary kidney, multiple renal cysts, a limited acoustic window, and the experience of the operator performing the measurements can all bias the results. However, apart from the above conditions, sonographic findings such as renal size, parenchymal thickness, cortical–medullary differentiation, and resistive index (IR) might provide valuable information regarding kidney health. Some studies have already shown that certain ultrasound parameters correlate with estimated glomerular filtration rate [1,2,3]. However, the correlation was only demonstrated in patients with advanced chronic kidney disease, presenting with very evident renal ultrasound changes. A study evaluating the association of ultrasound parameters with estimated glomerular filtration rate in the early stages of kidney disease is lacking. Furthermore, it would be of great interest to have a comprehensive model integrating multiple ultrasound variables to eGFR.

This study aims to investigate the relationship between renal ultrasound parameters and eGFR and to develop a predictive model based on these findings. If validated, such a model could help radiologists in identifying patients requiring a more comprehensive evaluation including biochemical markers such as serum creatinine and nephrology consultation.

2. Materials and Methods

2.1. Study Design and Population

This prospective, observational, cross-sectional study was conducted at the Nephrology and Dialysis Unit of ICS Maugeri in Pavia, Italy. A total of 130 patients were recruited consecutively between January and December 2021. All patients gave written informed consent. Given the observational nature of the study and that the patients underwent renal ultrasound requested by their general practitioner, the submission and approval of the ethics committee was waived. Patients referred for renal ultrasound as part of their routine clinical care were eligible for inclusion.

Inclusion criteria consisted of adults aged ≥ 18 years with available eGFR values within 30 days of the ultrasound exam. Exclusion criteria included: advanced CKD with marked reduction in kidney size, known polycystic kidney disease, severe obesity impairing imaging quality, recent diagnosis or suspicion of acute kidney injury, renal transplantation, liver cirrhosis, and lack of recent serum creatinine data.

2.2. Ultrasound Evaluation

All ultrasound examinations were performed by a single trained operator using the Esaote MyLabX7 system paired with a 3.5–5 MHz convex probe (Esaote SpA Genova 16152, Italy). The operator was unaware of the renal function of the participants to minimize potential bias. All participants were well hydrated and with a full urinary bladder at the time of the ultrasound procedure. Participants underwent renal ultrasound in the supine position and the following parameters were assessed:

• Renal length: Defined as the maximal pole to pole distance and measured in a longitudinal view from upper to lower pole in both kidneys (mm).
• Parenchymal thickness: Measured as distance between the sinus fat and the renal capsule, assessed at the mesorenal level, avoiding cysts or structural anomalies (mm).
• Renal volume: Estimated using the ellipsoid formula (length × width × depth × 0.523) when feasible.
• Cortical–medullary differentiation: Echogenicity generally refers to how bright or dark the kidney parenchyma appears in comparison to the liver. In a normal kidney, the cortex is typically hypoechoic (darker) relative to the echogenic (brighter) central renal sinus. Thus we classified the cortical–medullary differentiation as preserved when renal cortex was less echogenic than liver and spleen or as reduced or absent when renal cortex echogenicity was equal to liver and spleen. To reduce operator/observer bias, patients were classified by combining subjects with reduced or moderately non-preserved corticomedullary differentiation with those showing absent or non-preserved corticomedullary differentiation.
• Intrarenal resistive index (IR): Measured at interlobar arteries using pulsed-wave Doppler ultrasound. Three measurements per kidney were averaged.

Figure 1 shows how the renal ultrasound parameters used in the statistical analysis were measured.

2.3. Laboratory and Clinical Data

Serum creatinine values obtained within 30 days of the ultrasound were used to calculate eGFR using the CKD Epidemiology Collaboration equation (CKD-EPI equation).

Additional clinical data included age, sex, history of hypertension or diabetes mellitus, and renal anatomical anomalies such as a solitary kidney or shrunken kidney.

2.4. Statistical Analysis

Continuous variables are presented as mean ± standard deviation, reporting also minimum and maximum values and 1st–3rd quartile for transparency. Categorical variables are reported as frequencies and percentages. The Shapiro–Wilk test and QQ-plot representation were used to assess the normality of the distribution.

Pearson correlation coefficients were calculated to evaluate associations between eGFR and continuous ultrasound parameters.

A multivariable linear regression model was developed to identify independent predictors of eGFR. Candidate variables were selected using a bidirectional stepwise method based on the AIC value. We considered features with no missing values in the full model. We first specified a full multivariable linear regression model including all candidate covariates without missing data. To identify the most parsimonious set of independent predictors of eGFR, we applied a bidirectional stepwise selection procedure based on the Akaike Information Criterion (AIC). The final model obtained through this procedure was then evaluated using adjusted R². To assess predictive accuracy and generalizability, leave-one-out cross-validation (LOOCV) was performed and the cross validated R² was reported.

All analyses were conducted using R version 4.2.1. A p-value < 0.05 was considered statistically significant.

3. Results

3.1. Patients’ Characteristics

A total of 130 patients were included (66 males, 64 females), with a mean age of 65.9 ± 13.6 years. Mean eGFR was 63 ± 27 [Q1; Q3: 44; 87.8—Min; Max: 11; 124] mL/min/1.73 m² body surface area. As shown in

Table 1. Patient characteristics (N = 130). Data are presented as counts (percentages) or as mean (SD), median [Q1; Q3], and range (minimum–maximum). No missing values are present.

Females	64 (49.2%)
Age
Mean (SD)	65.9 (13.6)
Median [Q1; Q3]	67.5 [59; 74.8]
Min; Max	24.0; 89.0
Hypertension	77 (59.2%)
Diabetes	22 (16.9%)
eGFR (Renal stage)
Stage 1	25 (19.2%)
Stage 2	43 (33.1%)
Stage 3°	28 (21.5%)
Stage 3b	17 (13.1%)
Stage 4	16 (12.3%)
Stage 5	1 (0.8%)
Solitary/shrunken kidney	13 (10.0%)

the majority of the examined patients were hypertensive (59.2%), whereas only 16.9% had a history of diabetes mellitus. Of the patients, 10% presented with either a solitary kidney or a unilateral small-sized kidney.

3.2. Ultrasound Findings

Kidney volume measurement was time-consuming and was thus measured only in 10% of the participants. Mean value was 206.7 ± 72.4 cm³. (

Table 2. Ultrasound and Doppler findings. Patients with a shrunken kidney or a solitary kidney: the average value refers only to the functioning kidney. Data are presented as mean (SD), median [Q1; Q3], minimum–maximum, and number (%) of missing values.

	Mean (SD)	Median [Q1; Q3]	Min; Max	Missing
Renal volume (cm3)	206.7 (72.4)	200.5 [152.8; 230.7]	115.0; 371.0	118 (90.8%)
Right kidney length (cm)	9.8 (1.1)	10.0 [9.1; 10.4]	5.6; 13.0	4 (3.1%)
Left kidney length (cm)	10.2 (1.0)	10.3 [9.5; 10.9]	7.5; 12.8	7 (5.4%)
Mean kidney length (cm)	10.1 (1.0)	10.1 [9.5; 10.5]	7.0; 13.0
Right kidney parenchymal thickness (cm)	1.7 (0.4)	1.8 [1.5; 2.0]	0.3; 3.0	4 (3.1%)
Left kidney parenchymal thickness (cm)	1.8 (0.4)	1.8 [1.4; 2.1]	0.7; 2.7	7 (5.4%)
Mean kidney parenchymal thickness (cm)	1.8 (0.4)	1.8 [1.6; 2.0]	0.7; 3.0
Mean kidney cortical thickness (cm)	1.2 (1.8)	0.7 [0.6; 0.9]	0.5; 8.8	104 (80.0%)
Right kidney arteriolar resistance	0.68 (0.07)	0.70 [0.64; 0.73]	0.51; 0.85	5 (3.8%)
Left kidney arteriolar resistance	0.68 (0.06)	0.71 [0.64; 0.74]	0.55; 0.80	7 (5.4%)
Mean kidney arteriolar resistance	0.68 (0.06)	0.71 [0.64; 0.73]	0.55; 0.83

) Echogenicity was preserved in 21 participants (16.2%).

3.3. Correlation Analysis

Among all the ultrasound parameters measured, parenchymal thickness, renal length, and resistive index demonstrated a moderate positive correlation with eGFR values (Person’s r = 0.55; p < 0.001)). Correlations between eGFR and parenchymal thickness (Person’s r = 0.36; p < 0.001) or renal length (Person’s r = 0.38; p < 0.001) were modest in magnitude.

The predictive variables initially considered were sex, hypertension, diabetes, age, presence of a shrunken/solitary kidney, state of preservation of corticomedullary echogenicity, average renal length, average parenchymal thickness, and average arteriolar resistance of the two kidneys (

Table 3. Univariable linear regression models for covariates without missing values. For each parameter, the table reports the regression coefficient (predictor estimates, left columns) and the corresponding intercept estimates (right columns), along with standard error, 95% confidence interval, and p-value.

	Predictors				Intercept
	Estimates	Std. Error	95%CI	p	Estimates	Std. Error	95%CI	p
Age (years)	−0.94	0.16	−1.25; −0.63	<0.001	125.48	10.46	104.78; 146.18	<0.001
Biological sex (female)	−0.11	4.78	−9.57; 9.36	0.982	63.53	3.36	56.89; 70.17	<0.001
Hypertension (yes)	−3.30	4.86	−12.92; 6.31	0.498	65.43	3.74	58.04; 72.83	<0.001
Diabetes (yes)	−9.49	6.32	−22.00; 3.02	0.136	65.08	2.60	59.94; 70.23	<0.001
Solitary/shrunken kidney (yes)	−19.16	7.79	−34.58; −3.75	0.015	65.39	2.46	60.52; 70.27	<0.001
Echogenicity (preserved)	23.85	6.15	11.69; 36.02	<0.001	59.62	2.47	54.74; 64.51	<0.001
Mean kidney length (cm)	10.20	2.19	5.87; 14.54	<0.001	−39.22	22.16	−83.08; 4.63	0.079
Mean kidney parenchymal thickness (cm)	24.77	5.60	13.69; 35.85	<0.001	19.86	10.11	−0.14; 39.86	0.052
Mean kidney arteriolar resistance	−233.48	31.46	−295.72; −171.23	<0.001	224.20	21.75	181.17; 267.23	<0.001

).

3.4. Linear Regression Model

The final model included mean kidney length, mean resistive index, preserved echogenicity and presence of solitary/shrunken kidney and biological sex (

Table 4. Multivariable linear regression models for covariates selected by bidirectional stepwise AIC.

Predictors	Estimates	Std. Error	95%CI	p
(Intercept)	84.03	30.51	23.65; 144.42	0.007
Solitary/shrunken kidney (Yes)	−26.91	6.05	−38.88; −14.94	<0.001
Echogenicity (Preserved)	12.17	4.96	2.35; 21.98	0.016
Biological Sex (Female)	5.20	3.63	−1.99; 12.40	0.155
Mean kidney length (cm)	9.97	1.91	6.19; 13.74	<0.001
Mean resistive index (IR)	−178.23	28.96	−235.55; −120.90	<0.001

Observations (N): 130; R²: 0.488; R² adjusted: 0.467; AIC: 1153.375; RMSE: 19.36.

). The model adjusted R² estimated on a LOOCV was 0.44, and the Pearson correlation between predicted and observed eGFR value was r = 0.66 (p < 0.001).

4. Discussion

Our findings confirm that renal length, parenchymal thickness, and IR are reliable indicators of renal function. Cortical–medullary differentiation, though age-dependent, remains a useful marker. The final model, though preliminary, demonstrates that ultrasound can provide valuable functional insight in CKD. Our data demonstrate that renal volume measured with the ellipsoid formula could surely be another reliable indicator of renal function. However, the evaluation of renal volume requires the measurements of three diameters, craniocaudal, anteroposterior and transverse, and the adjustment to the patient’s body mass index [4], a time-consuming procedure not applicable to the routine of an outpatient clinic. In our clinical practice, measurement of kidney volume is replaced by evaluation of kidney length. This is due to several reasons: ultrasound measurement of renal length is often simpler, more reproducible, and less operator-dependent. Furthermore, some studies have shown that renal volume evaluation by ultrasound is often underestimated and inaccurate compared to CT measurement [5]. To obtain an adequate and reliable measurement, good ultrasound vision is often required in all three measurement axes, which is not easy and more technically challenging. For this reason, we decided to measure volume only in those patients whose measurement was very simple and for whom ultrasound visualization was optimal.

These technical issues limit the use of ultrasound evaluated kidney volume as an indicator of renal function; however, it would be useful to also consider this aspect in a future study.

The results of our study are consistent with the previous literature. For example, Lucisano et al. [6] demonstrated that renal length and parenchymal thickness measured by ultrasound correlate with renal function and may predict CKD progression, although they did not perform either Doppler ultrasound of the renal vasculature or evaluation of resistive indexes, which are, according to our results, very important indicators of renal function. In addition, their sample size was smaller and included patients with CKD stage 4, while our study evaluated patients in the early stages of kidney disease as well. Sanusi et al. [7] also found a significant relationship between renal volume and measured GFR, although ultrasound appeared to underestimate true renal volume compared to MRI. The resistive index (IR), as shown in this study, aligns with results by Radermacher et al. [8], showing that elevated IR values were associated with worse renal function and poorer graft survival in transplant recipients. The results of our study confirm those of Avramovski et al., who, in their study, found that renal RI inversely strongly correlates with the parenchymal thickness and GFR [9]. Ikee et al. [10] also noted IR as a marker of microvascular damage, especially in diabetic and hypertensive nephropathy. In addition, based on their results, they suggested the resistive index at renal biopsy as a useful indicator of renal outcomes. Our study extends this evidence by integrating multiple ultrasound and clinical parameters into a single predictive model. While prior research has assessed individual sonographic features, few studies have proposed a multivariate equation to estimate eGFR, which could be useful in settings lacking access to laboratory testing or in routine nephrology follow-up. Moghazi S. and collaborators [11] demonstrated that cortical echogenicity is the sonographic parameter that correlates best with renal histopathology. In addition, they demonstrated that tubular atrophy and interstitial inflammation, but not interstitial fibrosis, were significant determinants of cortical echogenicity. Unfortunately only a few of our patients had undergone kidney biopsy therefore we could not include kidney biopsy parameters in our statistical analysis. Additionally we could not even consider proteinuria in the analysis since it was performed only by a few patients in the 30 days preceding the ultrasound procedure. Our analysis included patients with and without diabetes; however, without a kidney biopsy and proteinuria, we were not able to assess with certainty whether or not they had diabetic nephropathy.

Several studies in the literature demonstrate that renal cortical thickness (RCT) is one of the ultrasonographic parameters that correlates most significantly with renal function. Unfortunately, due to the observational nature of our study, this parameter was evaluated in only a small subset of the enrolled patients, which represents a significant limitation. RCT is generally measured in the sagittal plane by drawing a perpendicular line from the renal capsule to the medullary pyramid. Typically, measurements are taken at the upper, mid, and lower part of the kidney, and these values are then averaged. Consequently, this measurement is not straightforward and is time-consuming particularly in advanced stages of chronic kidney disease (CKD), where corticomedullary differentiation is poor. So in clinical routine, it is much simpler to measure renal length. In addition, according to several studies, renal length maintains a significant statistical relationship with the estimated glomerular filtration rate (eGFR) [12,13]. Furthermore, in some ethnic groups, bipolar renal length correlates better with eGFR than renal cortical thickness [14]. Lastly Wei S. et al. [15] observed in patients with malignant hypertension a positive correlation between kidney length and renal cortical thickness. This suggests that bipolar kidney length can be used as a proxy of renal cortical thickness.

The significant correlation between ultrasound-derived renal measurements and eGFR may be the results of histological changes characteristic of chronic kidney disease. Tubular atrophy, interstitial fibrosis, sclerosed glomeruli and remodeling of renal vessels may result in decreased renal parenchymal thickness and size [2,11].

Recently, a very accurate prediction of estimated GFR has been obtained with machine learning methodologies. A strategy using a machine learning model improves the accuracy of GFR estimation at the population level and provides assessments at the individual level [16,17]. Although this system appears very fascinating, it may not be used worldwide due to the limited distribution of this technology, whereas our method could be easily reproduced and rapidly integrated in clinical practice in order to help radiologists identifying patients requiring a more comprehensive evaluation inclusive of biochemical markers such as serum creatinine and nephrology consultation. It is important to emphasize that the primary aim of the study was neither to develop a predictive score (e.g., a nomogram) nor to perform extensive subgroup analyses. Two main reasons justify this: the overall cohort size was limited, and the study was conducted in a single center. Our objective was instead to highlight the potential usefulness of an ultrasound parameter as an early warning sign for clinicians evaluating patients whose renal function is unknown. Even in the context of an incidental finding, such a parameter could prompt the clinician to initiate appropriate diagnostic investigations when suspicion arises.

Limitations include: a single-center design; reliance on creatinine-based eGFR due to absence of cystatin C measurements performed within 30 days of the ultrasound procedure in our patients; missed registration of data regarding patient height, which is a known predictor of kidney length; missed measurements of renal cortical thickness that is considered a better predictor of renal function among the ultrasonography parameters; and absence of longitudinal validation.

5. Conclusions

Ultrasound-derived renal measurements, especially renal length and IR, correlate significantly with eGFR. With a limited predictive accuracy the proposed model suggests that ultrasound could support estimation of renal function although the current evidence does not justify ultrasound as a substitute for biochemical eGFR estimation. Nevertheless, this study provides a foundation for future research into US-based renal scoring systems.