A Predictive Model Based on Cardio and Lung Ultrasound Features for Differential Diagnosis of Cardiogenic and Non‐cardiogenic Dyspnea in Patients With Chronic Kidney Disease: A Single‐Center Observational Study

Abstract

Objectives

Lung ultrasound (LUS) quantifying B‐lines is valuable to assess fluid status during hemodialysis. However, B‐lines may indicate not only cardiogenic pulmonary edema (CPE) but also non‐cardiogenic pulmonary diseases, such as pneumonia or interstitial lung disease, posing challenges to guiding fluid therapy decisions. We developed a combined cardiac‐LUS diagnostic model to differentiate these etiologies, aiming to prevent misdiagnosis and optimize treatment.

Methods

This retrospective observational study enrolled chronic kidney disease (CKD) stage 3–5 patients divided into training and validation cohorts. Participants were classified into CPE group or non‐cardiogenic dyspnea (NCD) group. Multivariable logistic regression identified discriminative echocardiographic and LUS parameters to build a diagnostic model. Predictive performance was assessed using receiver operator characteristic (ROC), calibration and decision curve analysis. A simplified β‐weighted scoring system was developed and evaluated.

Results

This retrospective study analyzed 215 training and 56 validation participants. Multivariable regression identified E/e′ (OR 1.180, 95% CI 1.067–1.304, p = .001), inferior vena cava diameter (OR 1.240, 95% CI 1.023–1.503, p = .029), bilateral symmetric B‐pattern (OR 9.720, 95% CI 2.707–34.899, p < .001), bilateral continuous B‐pattern (OR 13.777, 95% CI 3.666–51.770, p < .001), ≥1 irregular pleural line region (OR 0.078, 95% CI 0.018–0.326, p < .001), ≥1 consolidation region (OR 0.039, 95% CI 0.008–0.196, p < .001), ≥1 pleural effusion region (OR 4.732, 95% CI 1.148–19.510, p = .032) as independent CPE predictors. The nomogram showed good discrimination, with area under the curve (AUC) of 0.9703 (training) and 0.8910 (validation). The β‐weighted scoring system (cut‐off as 42.99 scores) showed an AUC of 0.968 (training) and 0.888 (validation).

Conclusion

The nomogram model demonstrated that the ultrasound parameters and β‐weighted scoring system possess reliable clinical application value and high predictive accuracy.

Abbreviations

ARDS

acute respiratory distress syndrome

AUC

area under the curve

CKD

chronic kidney disease

CPE

cardiogenic pulmonary edema

EVLW

extravascular lung water

IVC

inferior vena cava

KDIGO

Kidney Disease: Improving Global Outcomes

LAD

left atrial diameter

LUS

lung ultrasound

LVEDD

left ventricular end diastolic diameter

LVEF

left ventricular ejection fraction

NCD

non‐cardiogenic dyspnea

NT‐proBNP

N‐terminal pro‐B‐type natriuretic peptide

PASP

pulmonary artery systolic pressure

PCWP

pulmonary capillary wedge pressure

ROC

receiver operator characteristic

RVEDD

right ventricular end diastolic diameter

WBC

white blood cell

Chronic kidney disease (CKD) affects approximately 15% of the population with poor reversibility. ¹ For patients progressing to end‐stage CKD requiring hemodialysis, cardiovascular complications account for nearly half of all deaths. ² , ³ , ⁴ Chronic volume overload is highly prevalent in CKD patients, particularly those undergoing hemodialysis. ⁵ Due to impaired renal excretory function in CKD patients, excessive fluid intake surpassing excretion leads to increased intravascular volume, thereby elevating cardiac preload. Persistent volume overload can result in sustained cardiac enlargement and functional deterioration, further increasing the risk of pulmonary edema by raising pulmonary circulatory pressure—a common complication in CKD patients. ⁶ Notably, volume overload is strongly associated with increased mortality in CKD stages 3–5. ⁷ , ⁸ , ⁹

Assessment of volume status, cardiac function, and extravascular lung water (EVLW) is crucial in CKD patients. ¹⁰ Currently, lung ultrasound (LUS) and bioimpedance spectroscopy are considered practical and accurate tools for evaluating volume overload. ⁵ , ¹⁰ In CKD patients with cardiac dysfunction and volume overload leading to cardiogenic dyspnea, echocardiography and LUS enable rapid, bedside qualitative and quantitative assessments. ⁴ Cardiogenic dyspnea, often caused by heart failure‐induced cardiogenic pulmonary edema (CPE), typically requires fluid removal via diuretics or hemodialysis. Previous studies have demonstrated that excessive EVLW detected by LUS is associated with future cardiovascular events and mortality. ¹¹ LUS‐guided hemodialysis effectively reduces intradialytic blood pressure fluctuations, ¹² with a significant post‐dialysis reduction in B‐lines—correlating with decreased pulmonary congestion and improved dyspnea scores. ¹³ Moreover, echocardiography and LUS provide valuable assessments of cardiac function and EVLW in CKD patients, offering critical guidance for clinical management. ⁴

However, it is important to note that B‐lines on LUS are not specific to CPE. In CKD patients with non‐cardiogenic dyspnea (NCD) (eg, due to pneumonia, interstitial lung disease, or other conditions), exudative changes in the pulmonary interstitium can also manifest as B‐line artifacts on LUS. ¹⁴ CKD patients exhibit compromised immune function (particularly those receiving immunosuppressive therapy), rendering them significantly more susceptible to infections compared to the general population. ¹⁵ , ¹⁶ During the COVID‐19 pandemic, 15.74% of CKD patients infected with the SARS‐CoV‐2 Omicron variant progressed to critical COVID‐19 or death. ¹⁷ Notably, both COVID‐19 pneumonia and its secondary complication, acute respiratory distress syndrome (ARDS), predominantly manifest as bilateral multiple B‐lines on LUS. ¹⁸ , ¹⁹ Case reports further indicate that pulmonary interstitial fibrosis caused by anti‐GBM disease concurrent with ANCA‐associated vasculitis may also present with bilateral B‐lines on LUS ²⁰ These overlapping presentations may increase the risk of misdiagnosis when applying LUS for pulmonary evaluation in CKD patients.

Although CPE originates from cardiac dysfunction, pneumonia may also coexist in heart failure patients. When a CKD patient with pre‐existing cardiac insufficiency is hospitalized for dyspnea caused by pulmonary infection, the presence of cardiomegaly, reduced systolic function, and pulmonary B‐lines can create diagnostic ambiguity between CPE and pneumonia. Misguided dehydration therapy in such cases may precipitate hypovolemia. Accurate differential diagnosis is therefore critical for appropriate therapeutic decision‐making.

Although numerous studies have investigated the application of LUS B‐lines for assessing EVLW changes before and after hemodialysis in CKD patients, ²¹ the critical differentiation of B‐lines' etiology (Cardiogenic versus Non‐Cardiogenic CPE) remains underemphasized. Currently, no validated predictive model exists for distinguishing dyspnea types in CKD patients using ultrasonographic parameters. This study developed a comprehensive and effective scoring system based on the analysis of echocardiography and LUS parameters and characteristics to differentiate between cardiogenic and NCD in CKD patients and establish a predictive model.

Materials and Methods

Study Design and Population

This study was a single‐center, retrospective observational study involving patients diagnosed with CKD who presented with dyspnea. Selected patients received diagnosis and treatment in the Department of Nephrology or Hemodialysis ward of Hebei General Hospital between 2018 and 2025. Anonymous patient data were extracted from the electronic health record system, including the ultrasonography reporting system and ultrasound examination results documentation forms. The diagnosis of CKD was established by clinicians in accordance with the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines. ²² The study was approved by the Ethics Committee of Hebei General Hospital (Approval No. 2024‐150) and was performed following the last revised version of the Helsinki Declaration. As this study was a retrospective observational investigation, it was granted exemption from requiring informed consent.

The inclusion criteria were as follows: (1) Patients with a confirmed diagnosis of CKD (stages 3–5); (2) during hospitalization, acute dyspnea occurred and underwent cardio and LUS examination, with images or data retained; (3) ≥18 years old; (4) the final diagnosis of dyspnea was confirmed by clinical physicians. Exclusion criteria: (1) congenital organic heart disease, vascular malformations; (2) malignant tumors in the chest, mediastinum, or lungs; (3) chronic obstructive pulmonary disease, pneumothorax, subcutaneous emphysema; (4) combined acute cardiovascular events and acute kidney injury (AKI); (5) postrenal oliguria factors such as ureteral or bladder obstruction; (6) previous nephrectomy or kidney transplantation; (7) unclear diagnosis of dyspnea; (8) poor ultrasound image quality or incomplete data. AKI was diagnosed using the KDIGO criteria: a rise in serum creatinine by ≥0.3 mg/dL (≥26.5 μmol/L) within 48 hours, or by ≥50% from baseline within 7 days. ²³ CKD patients with combined AKI were excluded because changes in creatinine may affect the stage of CKD.

Data Collection

We divided the cases chronologically into a training cohort (November 2018 to April 2024) and a validation cohort (May 2024 to April 2025) (Figure 1). Eligibility for both cohorts was determined by applying the same inclusion and exclusion criteria. Baseline data were collected on demographic characteristics (age, gender), vital signs (respiratory rate, heart rate, systolic, and diastolic blood pressure), stage of CKD, comorbidities (Hypertension, diabetes, coronary heart disease, heart failure and its classification of New York Heart Association [NYHA]), laboratory variables (serum creatinine, white blood cell [WBC] count).

Figure 1.

Flow chart showing CKD patients included in the analysis.

All ultrasonographic examinations were performed using 2 identical bedside color Doppler ultrasound systems (Vivid IQ, GE Healthcare, USA). A phased‐array transducer (1.7–3.4 MHz) was employed for cardiac and inferior vena cava (IVC) imaging, and a convex‐array transducer (3.5–5 MHz) was used for LUS. The internal settings and operational functions of the 2 machines were identical, ensuring technical consistency. All cardiac and LUS examinations in this study were performed by a single operator: an operator who was an attending physician from the Department of Ultrasound. He is board‐certified with a Medical Practitioner License and specialized credentials in diagnostic ultrasound, with over 10 years of experience in critical care ultrasonography (over 5000 bedside examinations performed). As a physician, he was fully qualified to independently acquire, interpret, and report all ultrasound findings. Furthermore, this operator has obtained formal certification from the Chinese Critical Ultrasound Study Group (CCUSG) after passing its competency examination, and has also participated in international training conducted by the World Interactive Network Focused on Critical Ultrasound (WINFOCUS), ensuring compliance with global proficiency standards.

The predictive model was developed based on ultrasound parameters, which comprised the following cardiac data: left atrial diameter (LAD), left ventricular end diastolic diameter (LVEDD), right ventricular end diastolic diameter (RVEDD), left ventricular ejection fraction (LVEF), early diastolic and late diastolic peak velocity of mitral valve (E and A), early diastolic tissue velocity of interventricular septum (e′), pulmonary artery systolic pressure (PASP), and E/A. E/e′ were calculated. The IVC diameter was measured during end‐expiration at a location 2–3 cm distal to the right atrium along the IVC long axis, and the IVC variation was calculated as follows: IVC variation = (IVC diameter − IVC diameter at the end of inspiration)/IVC diameter. The detailed operational procedure is outlined in Video S1.

Based on our clinical experience with LUS, we defined several critical sonographic features (Table 1). All LUS examinations were performed and documented using the 12‐region scanning protocol (Figure 2). A‐pattern (A‐lines in all 12 lung regions or only 1 region has less than 3 B‐lines) was excluded from this study. B‐pattern was defined as ≥3 B‐lines in at least 2 regions, with or without consolidation and pleural effusion. Furthermore, we defined the distribution of B‐lines in both lungs (refer to Table 1, Figure 3).

Table 1.

Definition of Lung Ultrasound Features in This Study

Definition of Lung Ultrasound	Characterization
Lung ultrasound pattern (total 12 regions)
A‐pattern	Only A‐lines in both lungs or less than 3 B‐lines in only 1 region
B‐pattern	Within the 12 regions of both lungs, at least 2 regions presents ≥3 B‐lines, with or without consolidation and pleural effusion
Lung ultrasound features (single region)
Irregular pleural line region	In the B‐lines region, the pleural line is noticeably irregular and uneven
Consolidation region	In the scanning region, visible consolidation, with fragmented signs or bronchial signs as the main signs (exceeding 50% of the region), with or without pleural effusion
Pleural effusion region	In the scanning region, pleural effusion and compressed lung tissue can be seen, with pleural effusion accounting for at least 50% of the region
Lung ultrasound features (comparison of bilateral lungs)
Bilateral symmetric B‐pattern	B‐lines exists in at least 1 region of the bilateral lungs, respectively, which are symmetrically distributed
Asymmetric B‐pattern	One of the following conditions: (1) B‐lines appears in bilateral lungs, but asymmetrically; (2) B‐lines visible in 1 lung, and no B‐lines (A‐lines, consolidation or pleural effusion) in the other lung
Bilateral continuous B‐pattern	B‐lines can be observed continuous in the intercostal spaces of 1 lung, and when both lungs meet this characteristic, it is called “bilateral continuous B‐pattern”
Discontinuous B‐pattern	B‐lines can be observed discontinuous in the intercostal spaces of 1 lung, with intermittent A‐line or lung consolidation visible. At least 1 lung in both lungs meets this characteristic, which is called “discontinuous B‐pattern”

Figure 2.

The 12‐region lung ultrasound scanning protocol.

Figure 3.

Characteristics of different signs and patterns of lung ultrasound. A–D showed the LUS signs of cardiogenic dyspnea, including diffuse and continuous B‐lines (blue arrow) in both longitudinal planes (A) and transverse planes (B), smooth pleural line (C), and transudative pleural effusion (D). E–H showed the LUS signs of non‐cardiogenic dyspnea, such as pneumonia, including focal and discontinuous B‐lines (blue arrow) coexisting with A‐lines (red arrow) in both longitudinal planes (E) and transverse planes (F), irregular pleural line (G), and consolidation (H).

Diagnosis of CPE and Clinical Group

The diagnosis of dyspnea was made by 1 nephrologist and 2 intensive care physicians based on a posteriori review of medical records and medical history. The diagnosis of CPE included typical clinical manifestations (respiratory distress, coughing, orthopnea, etc.), extensive moist rales in both lungs throughing auscultation, evidence of left ventricular dysfunction and fluid overload, imaging supporting typical pulmonary edema. In addition, symptoms improved or imaging infiltrations significantly improved after receiving diuretic dehydration treatment.

Patients meeting diagnostic criteria for CPE were assigned to the CPE group, while those with NCD unrelated to CPE comprised the NCD group, which primarily included pneumonia, interstitial lung disease, alveolar hemorrhage, and non‐CPE (eg, ARDS). Cases with no evidence of pulmonary exudative lesions were excluded, as were those with mixed CPE and other pulmonary pathologies that could not be clearly differentiated.

Statistical Analysis

The SPSS version 21.0 (IBM; Armonk, NY, USA) and R4.2.2 software (R Core Team, Vienna, Austria) were used for data analysis. Continuous variables were expressed as a mean ± standard deviation (SD) or median (IQR) according to the data distribution and analyzed by t tests or Mann–Whitney U tests. Categorical variables were expressed as frequencies (percentages) and analyzed by chi‐square test or Fisher's exact test. Potential risk factors for CPE were initially screened using least absolute shrinkage and selection operator (LASSO) regression, with variables yielding p‐values <.1 subsequently incorporated into multivariable logistic regression analysis to identify independent predictors and develop a nomogram. Model performance was evaluated by generating receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). Calibration curves were plotted to assess prediction accuracy, while the Hosmer‐Lemeshow goodness‐of‐fit test was employed to evaluate the nomogram's calibration. A β‐weighted scoring system was developed by assigning “1 score” to the smallest positive partial regression β‐coefficient in the logistic regression model, with other predictors weighted proportionally based on their β‐coefficient ratios relative to this reference value. Each patient's ultrasound features were then scored accordingly to construct the predictive model scoring table. Inter‐observer agreement was quantified by Cohen's kappa statistic. The strength of agreement was: kappa <0.40, poor to fair; kappa = 0.41–0.75, moderate to good; kappa >0.75, excellent. Statistical significance was set at a p value <.05. All analyses of R software using the “readr,” “glmnet,” “Matrix,” “rms,” “foreign,” “psych,” “tidyverse,” “pROC,” “ResourceSelection,” and “rmda” package. p‐value <.05 was considered to concede statistical significance.

Results

Baseline Characteristic

During the study period, 215 patients (118 males, 97 females) were enrolled in the training cohort (86 with CPE and 129 with NCD). Fifty‐six patients (29 males, 27 females) were enrolled in the validation cohort (27 with CPE and 29 with NCD). Baseline comparisons revealed that the training cohort had lower median age (60.0 [46.0–71.0] versus 65.0 [53.0–77.25] years, p < .05), respiratory rate (21 [20–23] versus 22 [20–25] breaths/min, p < .05), and proportion of diabetes (47.4% versus 69.6%, p < .05) compared to the validation cohort (Table 2). No other statistically significant differences were observed between cohorts (p > .05).

Table 2.

Baseline Clinical Characteristics of Patients with CKD

Variable	Total (n = 271)	Training Cohort (n = 215)	Validation Cohort (n = 56)	p Value
Age (years)	61.00 (48.00, 72.00)	60.00 (46.00, 71.00)	65.00 (53.00, 77.25)	.011
Gender, n (%)				.679
Male	147 (54.2%)	118 (54.9%)	29 (51.8%)
Female	124 (45.8%)	97 (45.1%)	27 (48.2%)
Respiratory rate (breaths/min)	21.00 (20.00, 23.00)	21.00 (20.00, 23.00)	22.00 (20.00, 25.00)	.031
Heart rate (beats/min)	83.00 (74.00, 93.00)	83.00 (74.00, 93.00)	83.00 (72.50, 92.75)	.631
Systolic pressure (mm Hg)	146.49 ± 23.04	145.72 ± 22.70	149.45 ± 24.30	.282
Diastolic pressure (mm Hg)	81.00 (71.00, 93.00)	80.00 (71.00, 92.00)	81.00 (72.50, 93.75)	.735
Comorbidities, n (%)
Hypertension	246 (90.8%)	192 (89.3%)	54 (96.4%)	.101
Diabetes	141 (52.0%)	102 (47.4%)	39 (69.6%)	.003
Coronary heart disease	88 (32.5%)	68 (31.6%)	20 (35.7%)	.561
Heart failure	126 (46.5%)	97 (45.1%)	29 (51.8%)	.373
NYHA classification, n (%)				.727
Level II	12 (4.4%)	10 (4.7%)	2 (3.6%)
Level III	39 (14.4%)	31 (14.4%)	8 (14.3%)
Level IV	75 (27.7%)	56 (26.1%)	19 (33.9%)
CKD stages, n (%)
Stage‐3	4 (1.5%)	3 (1.4%)	1 (1.8%)	.759
Stage‐4	11 (4.0%)	8 (3.7%)	3 (5.3%)
Stage‐5	256 (94.5%)	204 (94.9%)	52 (92.9%)
Creatinine [μmol/L]	610.20 (437.20, 802.50)	612.30 (439.80, 818.00)	572.30 (397.88, 741.40)	.200
WBC (×109)	7.58 (5.91, 9.89)	7.77 (6.06, 10.21)	6.73 (5.33, 9.54)	.126

CKD, chronic kidney disease; NYHA, New York Heart Association; WBC, white blood cell.

Cardiac Ultrasound and LUS Characteristic of Training Cohort

The training cohort served as the foundation for model development. Within this cohort, the CPE group demonstrated significantly higher values for LAD, LVEDD, E wave, E/A, E/e′, PASP, IVC diameter, and a lower value of e′, IVC variation (p < .05). Regarding LUS characteristics, the CPE group exhibited significantly higher proportions of bilateral symmetric B‐pattern, bilateral continuous B‐pattern, and ≥1 pleural effusion region, but lower proportions of ≥1 irregular pleural line region and ≥1 consolidation region (p < .05) relative to the NCD group (Table 3).

Table 3.

Comparison of Ecocardiography and LUS Parameters in Training Cohort

Parameters	Total (n = 215)	CKD‐CPE (n = 86)	CKD‐NCD (n = 129)	p Value
Ecocardiography parameters
LAD (mm)	42.00 (38.00, 46.00)	44.00 (40.00, 48.00)	41.00 (38.00, 45.00)	<.001
LVEDD (mm)	50.00 (45.00, 54.00)	52.00 (46.75, 56.00)	48.00 (45.00, 52.00)	<.001
RVEDD (mm)	32.00 (30.00, 35.00)	32.00 (30.00, 36.25)	31.00 (29.00, 34.00)	.054
LVEF (%)	59.00 (51.00, 64.00)	55.00 (42.00, 61.00)	61.00 (57.00, 65.00)	<.001
E (cm/s)	96.00 (75.00, 120.00)	110.50 (94.00, 135.00)	83.00 (69.00, 108.00)	<.001
A (cm/s)	91.00 (65.00, 106.00)	92.50 (62.50, 107.75)	91.00 (66.50, 106.00)	.930
E/A	1.09 (0.78, 1.50)	1.29 (0.95, 1.85)	0.93 (0.71, 1.38)	<.001
Septal e′ (cm/s)	5.50 (4.60, 6.50)	5.00 (4.00, 5.90)	5.90 (5.00, 6.80)	<.001
E/e′	17.02 (12.88, 22.77)	22.05 (17.41, 27.45)	14.22 (11.75, 17.47)	<.001
PASP (mm Hg)	41.00 (32.00, 50.00)	45.00 (40.00, 58.00)	37.00 (30.50, 47.00)	<.001
IVC parameters
IVC diameter (mm)	16.88 ± 4.37	19.25 ± 3.17	15.30 ± 4.35	<.001
IVC variation (%)	37.0 (25.27, 48.15)	28.21 (18.91, 35.34)	43.75 (32.04, 51.94)	<.001
LUS pattern
Bilateral symmetric B‐pattern, n (%)	87 (%)	62 (72.1%)	25 (19.4%)	<.001
Bilateral continuous B‐pattern, n (%)	102 (%)	63 (73.3%)	29 (22.5%)	<.001
≥1 irregular pleural line region, n (%)	147 (%)	41 (47.7%)	106 (82.2%)	<.001
≥1 consolidation region, n (%)	79 (%)	17 (19.8%)	68 (52.7%)	<.001
≥1 pleural effusion region, n (%)	69 (%)	38 (44.2%)	29 (22.5%)	.001

LUS, lung ultrasound; LAD, left atrial diameter; LVEDD, left ventricular end‐diastolic diameter; RVEDD, right ventricular end‐diastolic diameter; LVEF, left ventricular ejection fraction; E, early diastolic peak velocity of mitral valve; A, late diastolic peak velocity of mitral valve; e′, early diastolic tissue velocity of interventricular septum; PASP, pulmonary artery systolic pressure; IVC, inferior vena cava.

Model Development by Univariate and Multivariate Logistic Regression Analysis

The result of LASSO cross‐validation will be obtained, and the lambda when it is 1 standard error away from the mean square error will be considered optimal (at this time, the mean square error is small, and the number of independent variables is also small) (Figure 4A,B). Using CPE as the dependent variable, It showed the results of univariate and multivariate logistic regression analyses, identifying 2 cardiac ultrasound and 5 LUS prediction factors as follows: E/e′ (odds ratio [OR] 1.180, 95% confidence interval [CI] 1.067–1.304, p = .001), IVC diameter (OR 1.240, 95% CI 1.023–1.503, p = .029), bilateral symmetric B‐pattern (OR 9.720, 95% CI 2.707–34.899, p < .001), bilateral continuous B‐pattern (OR 13.777, 95% CI 3.666–51.770, p < .001), ≥1 irregular pleural line region (OR 0.078, 95% CI 0.018–0.326, p < .001), ≥1 consolidation region (OR 0.039, 95% CI 0.008–0.196, p < .001), ≥1 pleural effusion region (OR 4.732, 95% CI 1.148–19.510, p = .032) (refer to Table 4, Table S1).

Figure 4.

LASSO regression screening of ultrasound influencing factors in CKD patients with concomitant CPE. A, The cross‐validation curve for the penalty term. B, LASSO regression coefficients plotted against different penalty parameter values.

Table 4.

Univariate and Multivariate Logistic Regression Analysis of Ultrasound Predictors

Parameters	Univariate Analysis		Multivariate Analysis
	OR (95% CI)	p	OR (95% CI)	p
LAD	1.114 (1.057–1.174)	<.001
LVEDD	1.070 (1.025–1.117)	.002
RVEDD	1.088 (1.023–1.158)	.008
LVEF	0.950 (0.926–0.976)	<.001
E/A	2.303 (1.474–3.598)	<.001
E/e′	1.209 (1.141–1.281)	<.001	1.180 (1.067–1.304)	.001
PASP	1.041 (1.019–1.063)	<.001
IVC diameter	1.305 (1.195–1.426)	<.001	1.240 (1.023–1.503)	.029
IVC variation	0.928 (0.906–0.951)	<.001
Bilateral symmetric B‐pattern	10.747 (5.653–20.429)	<.001	9.720 (2.707–34.899)	<.001
Bilateral continuous B‐pattern	9.445 (5.023–17.762)	<.001	13.777 (3.666–51.770)	<.001
≥1 irregular pleural line region	0.198 (0.107–0.367)	<.001	0.078 (0.018–0.326)	<.001
≥1 consolidation region	0.221 (0.117–0.416)	<.001	0.039 (0.008–0.196)	<.001
≥1 pleural effusion region	2.730 (1.508–4.941)	.001	4.732 (1.148–19.510)	.032

OR, odds ratio; CI, confidence interval; LAD, left atrial diameter; LVEDD, left ventricular end‐diastolic diameter; RVEDD, right ventricular end‐diastolic diameter; LVEF, left ventricular ejection fraction; E, early diastolic peak velocity of mitral valve; A, late diastolic peak velocity of mitral valve; e′, early diastolic tissue velocity of interventricular septum; PASP, pulmonary artery systolic pressure; IVC, inferior vena cava.

Model Development and External Validation

Combining the above 7 predictive factors, a nomogram was developed (Figure 5). The ROC curve showed that the AUC of the training cohort is 0.9703 (95% CI 0.9502–0.9903), with a sensitivity of 95.3% and a specificity of 92.2%, and the AUC of the validation cohort was 0.8910 (95% CI 0.8000–0.9830), with a sensitivity of 81.5% and a specificity of 93.1%, confirming the model's strong predictive value (Figure 6A,B, Table S2). Furthermore, the calibration curve suggested appropriate agreement between the model prediction and actual observations in both training and validation cohort in the Hosmer‐Lemeshow test (Figure 7A,B). Decision curve analysis (DCA) showed that the nomogram provides greater net benefits in predicting value, suggesting that the nomogram prediction model could effectively guide clinical practice (Figure 7C,D).

Figure 5.

Nomogram prediction model for differential diagnosis of dyspnea in CKD patients.

Figure 6.

The ROC curve of the nomogram. A, Training cohort. B, Validation cohort.

Figure 7.

The calibration curve and decision curve analysis (DCA) of the nomogram prediction model. A, Calibration curve of training cohort. B, Calibration curve of validation cohort. C, DCA of training cohort. D, DCA of validation cohort.

Construction of Scoring Method for Prediction Model

We assigned 1 score to the smallest positive β‐coefficient (E/e′, β: 0.165 = 1 score) from multivariate logistic regression, with the remaining 6 variables weighted proportionally to their β‐coefficient relative to E/e′ (refer to Table S3). It showed good discrimination of the β‐weighted scoring system demonstrated in the training cohort (cut‐off 42.99 scores, AUC 0.968, 95% CI 0.946–0.990; sensitivity 93.02%, specificity 91.47%) (Figure 8A), with satisfactory performance in the validation cohort using the same cut‐off (AUC 0.888, 95% CI 0.795–0.980; sensitivity 85.19%, specificity 86.21%) (Figure 8B, Table S4).

Figure 8.

The ROC curve of β‐weighted scoring system. A, Training cohort. B, Validation cohort.

Figure 9 showed the flowchart based on β‐weighted scoring system. The A‐pattern was described as “non‐cardiogenic dysnea” or “normal lung tissue” because cases presenting with this pattern were excluded from the inclusion criteria of the training cohort. When B‐pattern is present as the LUS finding, the LUS features (binary variables) and cardiac ultrasound measurements (continuous variables) are each multiplied by their respective β‐weighted coefficients, and the total score is calculated. The optimal cutoff value for this score is 42.99 scores.

Figure 9.

Differential diagnosis flowchart based on β‐weighted scoring system. The β‐weighted scoring system consists of the sum of 2 parts. For lung ultrasound, features are binary (present = 1, absent = 0) and multiplied by their β‐weighted coefficients. For cardiac ultrasound, the measured values of IVCD and E/e′ are continuous variables and are each multiplied by their respective β‐weighted coefficients. IVCD, inferior vena cava diameter; E, early diastolic peak velocity of mitral valve; e′, early diastolic tissue velocity of interventricular septum.

Inter‐Observer Agreement

Inter‐observer agreement was assessed using Cohen's kappa coefficient. For the clinical diagnosis, a random sample of 30 patients was independently evaluated by the study nephrologist and the study intensivist involved in the study. The agreement was excellent (kappa = 0.933, p < .001) (Table S5). For the interpretation of LUS signs, 80 LUS images (featuring A‐lines, B‐lines, consolidations, and pleural effusions) were randomly selected and independently interpreted by 2 sonographers: the primary study sonographer (who performed LUS examinations in this study, with 10 years of experience) and another sonographer (with 7 years of experience). The inter‐observer agreement was excellent (kappa = 0.817, p < .001) (Table S6).

Discussion

In our study, Cardiac and IVC ultrasonography revealed significantly elevated E/e′ ratios and IVC diameters in the CPE group, while LUS analysis identified several diagnostically significant features based on established CPE characteristics and our clinical ultrasonographic experience. This study revealed that bilateral symmetric B‐pattern, bilateral continuous B‐pattern, and ≥1 pleural effusion region strongly supported CPE diagnosis whereas ≥1 irregular pleural line region and ≥1 consolidation region favored non‐cardiogenic pulmonary disease, with the training cohort showing excellent discriminative performance (AUC 0.9703, 95% CI 0.9502–0.9903) and external validation confirming good diagnostic accuracy (AUC 0.8910, 95% CI 0.8000–0.9830) for distinguishing CPE‐related dyspnea in CKD patients.

In this study, we developed a scoring system weighted by the β‐coefficients of variables selected through multivariate regression analysis. The optimal cut‐off value of 42.99 scores in the training cohort demonstrated excellent diagnostic performance (AUC 0.968, 95% CI 0.946–0.990), with 93.02% sensitivity and 91.47% specificity. External validation using this cut‐off maintained satisfactory discrimination (AUC 0.888, 95% CI 0.795–0.980), showing 85.19% sensitivity and 86.21% specificity, thereby confirming the model's accuracy. The β‐weighted scoring system significantly enhanced clinical utility by providing both intuitive interpretation and practical application. This enables clinicians to promptly determine dyspnea etiology based on initial ultrasonographic findings, facilitating timely therapeutic decision‐making at the point of care.

CKD patients with fluid overload may contribute to secondary cardiovascular complications, which also elevate mortality risk. ³ Various methods are available for assessing volume status in dialysis‐dependent CKD patients, including bioimpedance spectroscopy, N‐terminal pro‐B‐type natriuretic peptide (NT‐proBNP) measurement, physical examination for peripheral edema, chest radiography, IVC parameters, and quantification of B‐lines on LUS. ¹⁰ Parameters of the IVC include static indices (diameter) and dynamic indices (variation), which closely correlate with central venous pressure proximal segment. ²⁴ Patients with cardiogenic dyspnea typically demonstrate wider IVC diameters and reduced collapsibility, reflecting their characteristic hemodynamic profile. ²⁵ During hemodialysis sessions, serial IVC diameter measurements showed significant correlation with reductions in total body water volume. ²⁶ Notably, elevated IVC diameters are associated with both major adverse cardiovascular events and mortality. ²⁷ Study by Sekiguchi et al suggested that an IVC minimal diameter >23 mm demonstrates diagnostic value for distinguishing CPE from non‐CPE. ²⁸ The pathophysiological basis of CPE involves elevated pulmonary capillary wedge pressure (PCWP), which directly correlates with left atrial pressure and left ventricular function. ²⁹ The E/e′ ratio, a key indicator of left ventricular filling pressure and diastolic function, plays a more significant role than systolic function (as measured by LVEF) in the pathophysiology of PCWP elevation. This is corroborated by 2 studies on ventilator‐weaning induced pulmonary edema, where elevated E/e′—but not LVEF—emerged as a significant predictor. ³⁰ , ³¹ Our findings align with this mechanistic understanding, as E/e′ was independently associated with CPE in our cohort.

LUS has demonstrated significant clinical utility across diverse settings, including intensive care units, internal medicine, and surgical departments. ³² , ³³ , ³⁴ In hemodialysis‐dependent CKD patients, LUS exhibits superior sensitivity to CT for evaluating fluid removal before and after dialysis sessions. ²¹ Ross et al reported the critical need to differentiate CPE‐related B‐lines from those caused by interstitial lung diseases. ³⁵ Clinically, distinguishing between CPE and pneumonia is essential, with LUS effectively capturing discriminative features between these conditions. ³⁶ Two meta‐analyses had quantified LUS diagnostic performance for CPE (88% sensitivity, 90% specificity) and for pneumonia (92% sensitivity, 94% specificity). ³⁷ , ³⁸ However, these studies primarily compared pathological findings against normal lungs, whereas real‐world clinical practice demands discrimination between CPE and non‐cardiogenic pulmonary disease.

In LUS, B‐lines associated pathologies were collectively termed alveolar‐interstitial syndrome, which encompasses various pathological conditions affecting the alveoli and/or interstitium. ³⁹ Clinically, this syndrome is classified into hemodynamic (cardiogenic) and non‐hemodynamic (non‐cardiogenic) subtypes to guide therapeutic decision‐making. Previous studies have focused on developing LUS‐based differentiation methods by analyzing specific characteristics of B‐lines distribution patterns in bilateral lung and pleural line abnormalities to distinguish between these etiological subtypes of dyspnea.

In this study, pneumonia emerged as the primary etiology of NCD in CKD patients. As Sultan et al summarized, ultrasonographic characteristics of pneumonia include consolidations, pleural line irregularities, A‐lines, and B‐lines. ⁴⁰ These sonographic features may manifest either collectively or selectively in patients with pulmonary infections. ¹⁴ Such as COVID‐19 pneumonia typically demonstrates multifocal B‐lines with rare consolidations/effusions, reflecting its interstitial predominance. ⁴¹ In contrast, lobar consolidations predominate in bacterial pneumonias. ⁴² , ⁴³ Therefore, the ultrasonographic features of pneumonia exhibit diversity, ³⁹ which is associated with the biological type of inflammatory lesions, their distribution in the lungs, severity, and the patient's immune status. Within the examined region, consolidation signs with multiple air bronchograms and shred signs predominantly suggest lobar pneumonia. In contrast, consolidation caused by massive pleural effusion compressing the lung lobe may be related to transudate due to CPE. ²⁸ Therefore, in this study, we classified consolidation into 2 types: consolidation‐dominant region and pleural effusion‐dominant region. The results demonstrated that, compared with the NCD group, the CPE group showed a significantly higher proportion of ≥1 pleural effusion region (OR 4.732, 95% CI 1.148–19.510, p = .032) and a lower proportion of ≥1 consolidation region (OR 0.039, 95% CI 0.008–0.196, p < .001), both of which remained independent predictors in multivariate regression analysis.

In cases with typical consolidation signs, differentiation from CPE is straightforward. However, distinguishing non‐cardiac pulmonary diseases (especially characterized by interstitial multiple focal B‐lines) from CPE poses diagnostic challenges. Under these circumstances, additional B‐lines characteristics become crucial, including pleural smoothness or irregularity. ⁴⁴ , ⁴⁵ In 2008, the BLUE‐protocol pioneered the classification of CPE as B‐profile (B‐lines with lung sliding) and pneumonia as B′‐profile (B‐lines without lung sliding), highlighting the diagnostic value of pleural characteristics. ³² Current LUS guidelines and consensus statements recognized pleural line morphology as a critical feature for both qualitative and quantitative assessments. ⁴⁶ , ⁴⁷ This has been robustly validated in animal studies. ⁴⁸ Among ICU patients, fragmented, thickened, or irregular pleural line demonstrated 89.3% sensitivity and 100% specificity for differentiating CPE from other pulmonary disorders. ⁴⁹ In this study, the presence of ≥1 irregular pleural line region showed a stronger association with NCD and remained an independent predictor in multivariate regression analysis (OR 0.078, 95% CI 0.018–0.326, p < .001).

The study of Wang et al focuses on lung phenotypes on LUS revealed significant differences in B‐lines distribution patterns across various diseases. CPE predominantly exhibited a nongravity‐dependent phenotype, whereas non‐cardiogenic pulmonary diseases (such as severe pneumonia and ARDS) showed a higher prevalence of gravity‐dependent phenotype. ⁵⁰ Dojcinovic et al demonstrated that the characteristic distribution of LUS findings could effectively differentiate cardiogenic from non‐cardiogenic causes of dyspnea. ²⁵ In our study, we further classified B‐lines patterns based on their distribution characteristics of continuity within ipsilateral lung and symmetry between bilateral lungs. Multivariate analysis confirmed both bilateral symmetric B‐pattern (OR 9.720, 95% CI 2.707–34.899, p < .001) and bilateral continuous B‐pattern (OR 13.777, 95% CI 3.666–51.770, p < .001) as independent predictors of CPE.

This study has several limitations. First, although pleural line characteristics are crucial for differentiating cardiogenic from NCD, current assessment techniques remain limited. For static pleural line evaluation, visual assessment (smoothness or irregularity) is currently used, which may introduce observational bias. Moreover, dynamic pleural line assessment relies on subjective visual evaluation of sliding amplitude (strong, weak, or absent), lacking validated quantitative parameters. Two‐dimensional speckle tracking showed promise for quantifying pleural sliding, though larger validation studies are required. ⁵¹ Second, Recent studies suggest that 2‐dimensional speckle tracking and superb microvascular imaging may serve as potential quantitative indicators for assessing pleural sliding, though large‐scale studies are still required for further validation. Third, this retrospective analysis may be subject to selection biases and unmeasured confounders. Although some patients received multiple ultrasound scans, only the findings from the first examination (ie, performed during the early phase of dyspnea) were included in the analysis. This was done to ensure methodological rigor. Fourth, the inherent limitations associated with chronic structural and functional cardiac alterations in the CKD population must be acknowledged. Chronic changes commonly found in advanced CKD, such as left ventricular hypertrophy and myocardial fibrosis, can significantly confound the interpretation of diastolic parameters. For instance, E/A and e′ may predominantly reflect chronic myocardial remodeling (eg, due to uremic cardiomyopathy and long‐standing hypertension) rather than purely acute volume status. Consequently, although our study observed that cardiac ultrasound parameters have good application value in differentiating cardiogenic and NCD, these findings must be interpreted with caution within the context of chronic cardiac remodeling. Future studies incorporating advanced imaging techniques (eg, cardiac magnetic resonance) or longitudinal follow‐up may be needed to better disentangle the contributions of acute hemodynamic shifts from chronic myocardial disease. Finally, this is a single‐center study, which may limit the generalizability of findings. More research from multiple centers should be analyzed to validate the nomogram.

Conclusions

We developed a diagnostic nomogram specifically designed to differentiate cardiogenic from NCD in CKD patients presenting with unexplained respiratory distress. This tool integrates multiple cardiac, IVC, and LUS parameters to facilitate early classification of dyspnea etiology. The nomogram effectively identifies discriminative indicators with high accuracy. In addition, a weighted scoring system was derived from β‐coefficients, demonstrating robust diagnostic performance while optimizing clinical usability. By focusing on the clinical utility of ultrasonographic differentiation in CKD populations, our findings provide actionable guidance for therapeutic decision‐making.

Supporting information

Table S1. Variable value assignment table.

Table S2. ROC curve of nomogram.

Table S3. β‐weighted scoring system.

Table S4. The ROC curve of β‐weighted scoring system.

Table S5. Kappa coefficients of inter‐observer agreement for CPE and NCD diagnoses.

Table S6. Kappa for agreement between 2 observers on the identification of lung ultrasound features.

Video S1. Measurement methods of echocardiography and inferior vena cava ultrasound.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.