Introduction
Chronic kidney disease (CKD) is defined as a condition with kidney damage or reduced kidney function persisting for more than 3 months [1]. Risk factors for CKD include high blood pressure, dyslipidemia, diabetes mellitus (DM), cardiovascular disease, age greater than 65 years, and various other factors [2]. Early detection is crucial because medications such as angiotensin-converting-enzyme inhibitor drugs or angiotensin II receptor blockers can slow disease progression and reduce mortality rates. In addition to glomerular filtration rate (GFR), the presence or absence of albuminuria is significant for staging CKD. Even with normal GFR, the risk of CKD is increased when albuminuria is present [3]. Therefore, screening of high-risk patients for CKD involves assessment of GFR through blood tests and of albuminuria using urine analysis, particularly via urine dipstick tests in primary care settings. The urine dipstick is cost-effective and provides information on proteins, ketones, glucose, leukocytes, nitrite, and urinary pH [4]. However, the accuracy of urine dipstick tests for evaluating albuminuria remains uncertain [5–7]. Only patients with positive urine dipstick results are referred for further renal function evaluation, which includes tests such as urinary albumin-to-creatinine ratio (ACR), serum urea nitrogen, and serum creatinine. Our study aimed to investigate risk factors associated with false-negative urine dipstick results for renal dysfunction in a Korean adult population using the Korea National Health and Nutrition Examination Survey (KNHANES) data. In addition, we aimed to develop a predictive model for detecting albuminuria when urine dipstick results are negative.
Methods
Subject
We utilized data from the KNHANES from 2019 to 2022. The KNHANES is a comprehensive nationwide cross-sectional survey conducted by the Korea Ministry of Health and Welfare to represent the Korean population. Employing a clustered, multistage, stratified, and rolling sampling approach, each KNHANES cycle is composed of independent datasets of healthy individuals from South Korea for the respective survey years. Participants were randomly selected from 600 districts throughout the country. The KNHANES consists of three main sections: a health interview, health examination, and dietary survey. The KNHANES has been approved by the Institutional Review Board of the Korea Centers for Disease Control and Prevention (No. 2018-01-03-C-A, 2018-01-03-2C-A, 2018-01-03-5C-A, and 2018-01-03-4C-A). Eligibility criteria for our study were participants aged 18 years and older with negative urine dipstick results for protein. Of the initial 28,824 participants, 19,034 were eligible and selected based on predefined eligibility criteria. Among these, data from 14,541 participants from 2019 to 2021 was used to build the model, and data from 4,493 participants from 2022 was used to validate the model (Fig. 1).
Health examination
Survey demographic information, including age, sex, height, weight, and body mass index (BMI), was collected using a health questionnaire. Trained staff conducted measurements of body height and weight during the health examination. BMI was calculated as weight (kg)/square of height (m2). Blood pressure results were obtained using standard sphygmomanometer methods with the participant seated. Blood pressure was measured three times at 30-second intervals after a minimum of 5 minutes of rest in a seated position, and the average value of the second and third measurements was recorded.
Biochemical analysis
Blood and spot urine samples were collected after an 8-hour fast and transported to the Neodin Medical Institute in Seoul, Korea, for biochemical analysis. Serum creatinine was measured using the Jaffe rate-blanked and compensated method (Hitachi Automatic Analyzer 7600; Hitachi), and the estimated GFR (eGFR, mL/min/1.73 m2) was calculated using the Modification of Diet in Renal Disease. Random urine samples, collected during the first morning void, were used to measure urinary albumin concentration (μg) using a turbidimetric immunoassay (Hitachi Automatic Analyzer 7600). Dipstick urinalysis was performed using the Urisys 2400 cassette strip, and results were presented on a Urisys 2400 automated analyzer (Roche). Urinary creatinine concentration (mg) was measured using a colorimetric method (Hitachi Automatic Analyzer 7600). The ACR (mg/g) was computed by dividing the urinary albumin concentration by the urinary creatinine concentration. Albuminuria was defined according to the KDIGO (Kidney Disease: Improving Global Outcomes) definition as a urinary ACR of >30 to 300 mg/g, encompassing both microalbuminuria (30–300 mg/g) and macroalbuminuria (>300 mg/g) [3].
Socioeconomic data
For socioeconomic data, household income was categorized into five grades based on sex and age for each year of the survey: poor, poor-moderate, moderate, moderate-rich, and rich. Dietary intake was assessed using data from the 24-hour recall dietary intake survey conducted by the KNHANES. The survey includes measurements of total calorie, calcium, and sodium intakes as calculated according to the Korean Food Composition Table [8]. Participants were asked to recall and report both the types and quantities of food they consumed over the previous 24 hours.
Statistical analysis
All statistical analyses were performed using R software (version 4.2.1; R Foundation for Statistical Computing) to identify risk factors associated with false-negative results on urine dipstick tests for proteinuria and to develop a predictive score model. Univariate logistic regression analyses were initially conducted to identify potential risk factors associated with false-negative dipstick results. Variables included in the univariate analysis were age, sex, BMI, blood pressure, urine specific gravity, comorbidities (e.g., hypertension, DM), and laboratory measures (e.g., serum creatinine, eGFR, hemoglobin A1C [HbA1c]), considering their clinical utility and importance of potential risk. Variables with a p-value <0.05 were considered statistically significant and were included in the multivariate analysis. A multivariate logistic regression model was developed to assess the independent association between each risk factor and the likelihood of a false-negative dipstick result. Stepwise regression analysis using Akaike Information Criterion (AIC) was used to select the final model. This approach allowed the optimal combination of variables that provided the best fit for the data while minimizing the risk of overfitting. The final model included variables that significantly contributed to predicting the likelihood of a false-negative result. To assess multicollinearity between variables, the variance inflation factor (VIF) was checked to ensure that it did not exceed the threshold of 5. A risk score model was developed using the beta coefficients from the final multivariate logistic regression model. Each significant variable in the final model was assigned a score based on its beta coefficient, rounded to the nearest integer for simplicity in clinical application. Variables with p-value <0.05 were considered significant, and for ordinal categorical variables, significance was determined based on the global p-value. The total risk score for each participant was calculated by summing the individual scores. The predictive performance of the score model was assessed using the area under the receiver operating characteristic curve (ROC-AUC), as was the ability of the model to discriminate between false-negative and true negative cases. To determine the optimal threshold for classifying participants into high-risk and low-risk groups, we used Youden’s index, which maximizes the sum of sensitivity and specificity, providing a balanced approach to minimize both false positives and false negatives. External validation was conducted using the 2019 dataset, which was not used in the model development phase. The external validation cohort consisted of 4,493 participants. The performance of the score model in the validation cohort was also assessed using ROC-AUC. Missing data were addressed using complete case analysis (only participants with complete data for all variables included in the final model were analyzed). This approach was chosen to ensure robustness in the results and to avoid the potential bias introduced by imputation methods.
Results
Demographics of clinical characteristics
Our study analyzed data from 19,034 participants, comprising 8,093 males and 10,941 females. The prevalence of albuminuria among subjects with negative urine dipstick results was 4.0%. Demographic characteristics are summarized in Table 1. Individuals with albuminuria (false-negative cohort) were older (48.6 years vs. 62.0 years, p < 0.01) and had higher BMI (24.1 kg/m2 vs. 24.7 kg/m2, p < 0.01), abdominal circumference (83.9 cm vs. 87.5 cm, p < 0.01), systolic blood pressure (118.0 mmHg vs. 130.9 mmHg, p < 0.01), and diastolic blood pressure (74.9 mmHg vs. 78.0 mmHg, p < 0.01) but lower weight (66.0 kg vs. 63.8 kg, p < 0.01), total calorie intake (1,876.4 kcal vs. 1,657.7 kcal, p < 0.01), and total water intake (1,063.8 mL vs. 925.4 mL, p < 0.01) than subjects without albuminuria.
Comorbidities of subjects
Table 2 shows the prevalent comorbidities among participants. The albuminuria group exhibited significantly higher rates of diagnosed hypertension, DM, dyslipidemia, and CKD (p < 0.01). Hypertension was the most common comorbidity in the albuminuria group (52.0%), and only 5% of participants with albuminuria were previously diagnosed with CKD.
Laboratory results of subjects
Laboratory findings (Table 3) revealed higher levels of urea nitrogen, creatinine, fasting glucose, HbA1C, and triglycerides in the albuminuria group than in the non-albuminuria group. Urine analysis showed higher pH, glucose, and blood levels, and lower specific gravity in the albuminuria group.
Development of a prediction model for albuminuria based on a negative urine dipstick test
The eight variables of sex, age, weight, systolic blood pressure, comorbidities (hypertension, dyslipidemia, DM), and urine specific gravity were considered relevant to the false-negative in the urine dipstick test with clinical utility and were tested for significant association with the likelihood of false negatives using univariate logistic regression. From this analysis, all selected risk factors were identified as candidate predictors for model development with p-values less than 0.5 (Supplementary Table 1, available online). In the evaluation of multicollinearity among the variables considered for multivariate logistic regression analysis, the VIF did not exceed the threshold value of 5 for any predictor, indicating no significant multicollinearity concerns in the model (Supplementary Table 2, available online). Multivariate logistic regression was then performed using those predictors, and six predictors (sex, age, systolic blood pressure, comorbidity of hypertension or DM, and urine specific gravity) were seven selected for the final model after stepwise multivariate analysis based on AIC values. The result of multivariate analysis is shown in Table 4 including the odds ratio (OR) and beta coefficient (95% confidence interval [CI]) of each covariate. In the final model, urine specific gravity <1.005, systolic blood pressure >160 mmHg, and comorbidity of DM had the highest ORs of 5.338, 4.673, and 4.253, respectively. Score points were assigned by multiplying the beta coefficient from multivariate logistic regression by 10 and rounding to the closest integer (Table 4). The allocated prediction score for each variable ranged from 2 to 17, with a high score indicating a high risk of a false negative in the urine dipstick test, and the total score was calculated by summing the respective scores. The total scores of individuals in the case cohort were significantly greater than the scores in the control cohort (p < 0.05). Total scores were stratified into two groups—low risk (<19) and high risk (≥19)—using Youden’s index, which maximizes the combined predictive performance of sensitivity and specificity. The OR for the high-risk group compared with the low-risk group was 9.832 (95% CI, 8.05–12.00).
Validation of developed score
The score model to predict false-negative results in urine dipstick analysis, developed using data from 2019 to 2021, was validated with a 2022 cohort (Fig. 1). The score distribution of the validation cohort was similar to the corresponding validation cohort. (Supplementary Fig. 1, available online). The predictive model demonstrated strong performance, with an AUC-ROC of 0.82 which shows that the total score can predict the risk of false-negative urine dipstick in good predictability (Fig. 2). Also, the stratification by the model (high risk, ≥19 and low risk, <19) which could maximize the sum of sensitivity and specificity showed sensitivity of 0.769, specificity of 0.694 and accuracy of 0.735. These metrics indicate the model’s robust ability to distinguish true negatives albuminuria (3,011/4,337, 69.4%) and false-negatives albuminuria (120/156, 76.9%) when urine dipstick is negative, also showing superior performance compared to conventional methods. Specifically, when using a medical history of DM or hypertension alone, the analysis achieved sensitivity, specificity, and accuracy values of 0.705, 0.699, and 0.698 respectively. Additionally, assessments based on uncontrolled DM (HbA1c ≥7) or uncontrolled hypertension (systolic blood pressure ≥130 mmHg or diastolic blood pressure ≥80 mmHg) showed high sensitivity (0.936). However, these methods were ineffective at identifying true negative cases of albuminuria when urine dipstick results were negative, achieving only a specificity of 0.033 and an accuracy of 0.064. This limitation could lead to unnecessary urine albumin testing.
Web-based application for score model
The developed score model which predicts the risk of false-negative was implemented in R script, and a web-based R shiny application was developed for ease of use in clinics. A draft version of the software is now available at https://pipetapp.com/project/albuminuria-risk/ with free unlimited access. Supplementary Fig. 2 (available online) shows an example of the R Shiny application interface used to predict the risk of false negatives in the urine dipstick test. This application displays the calculated risk score and estimated relative risk, along with the false-negative probability derived from the developed score model. It also presents the median score and false-negative probability for the overall population. Additionally, it stratifies patients into low-risk and high-risk groups based on each patient’s score, using 19 as a cutoff value which demonstrated the highest combined specificity and sensitivity.
Discussion
In this study, we focused on false-negative albuminuria in healthy adults (i.e., subjects who had a negative urine dipstick for protein but an albumin level >30 mg/g), which was shown in 4% of the overall population with a negative urine dipstick test. We investigated the associated risk factors and comorbidities of false-negative albuminuria and developed a prediction model with an accompanying web-based application to identify albuminuria risk in cases of negative dipstick results. Our model demonstrated good predictive capability (AUC-ROC = 0.820), and its risk stratification approach proved superior to conventional methods that rely solely on a medical history of DM or hypertension.
These findings highlight the importance of CKD risk awareness among primary care clinicians, even when the urinary protein dipstick is negative. Early detection and intervention play pivotal roles in mitigating the progression of CKD and the associated complications. Unfortunately, CKD often remains asymptomatic, leading to low awareness and late diagnosed cases [9,10]. Intervening during the early stages of CKD is important because evidence indicates that early and consistent nephrology care can significantly reduce CKD-related morbidity and mortality [11]. Albuminuria is important because it detects CKD risk. The presence of albuminuria is associated with even when GFR is normal [3]. Previous studies have focused on the accuracy of urine dipstick in detecting albuminuria. Nielsen et al. [5] concluded that a urine dipstick is not a reliable screening tool for albuminuria. Panta and Techakehakij [6] concluded that the dipstick should not be recommended for mass screening of albuminuria among hypertensive patients due to its low sensitivity. Our study showed a 4% prevalence of albuminuria with a negative urine dipstick, illustrating its unreliability. Although major DM guidelines recommend annual albuminuria testing [12,13], current hypertension guidelines vary in their recommendations for screening albuminuria [14]. The 2018 European Society of Cardiology/European Society of Hypertension guidelines recommend albuminuria screening for all hypertensive patients, with annual albuminuria testing specifically for subjects with CKD [15]. In contrast, the 2017 American College of Cardiology/American Heart Association guidelines and 2020 International Society of Hypertension guidelines suggest routine urine dipstick testing, emphasizing that serial albuminuria testing can provide value as part of optimal care [16,17]. Our study identified hypertension as the most common comorbidity associated with false-negative albuminuria. Based on the findings, we recommend screening for albuminuria in addition to urine dipstick testing for patients with hypertension. The results showed that risk factors such as hypertension, DM, and dyslipidemia for patients with albuminuria align with risk factors for CKD. Our predictive model showed that a calculated score >19 reflecting the weighted risk factors for urine stick-negative patients indicates high risk. Therefore, in the medical history assessment, it is crucial to consider the presence of DM, dyslipidemia, and hypertension, especially based on the degree of elevated blood pressure measured during routine screenings. When the score exceeds the 19-point cutoff, selective ACR testing can aid in early CKD diagnosis. This research emphasizes the importance of albuminuria screening among high-risk patients in primary care settings to identify CKD risk factors. However, there are several limitations in our study. First, the study design was cross-sectional, prohibiting assessment of the cause-and-effect relationship. Longitudinal studies are needed to evaluate the effectiveness of the predictive model and its effect on early CKD detection and management. Second, the focus was on false-negative albuminuria in healthy adults, and the predictive model may not be generalizable to other populations. Last, although the predictive model provides a valuable tool for identifying high-risk individuals, its accuracy and generalizability should be further validated in diverse settings and larger cohorts. In conclusion, this study underscores the importance of albuminuria screening in high-risk patients, even when urine dipstick results are negative. The development of a predictive model for albuminuria in such a moderately increased risk of CKD cases highlights the importance of CKD screening in earlier detection and better management of the disease. In the long term, this model is anticipated to contribute to more cost-effective CKD screening by enabling selective ACR testing.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement table 1
Print
