The relatively poor correlation between random and 24-hour urine protein excretion in patients with biopsy-proven glomerular diseases

Abstract

Random urine protein creatinine ratios are used to estimate 24-hour urine protein excretion which is considered a diagnostic gold standard. However, few studies are available of the sensitivity and specificity of this estimation in patients with glomerular proteinuria. To clarify this, we measured the urine protein and creatinine centrally in random and 24-hour urine collections at biopsy and longitudinally every 6 months in individuals participating in the Nephrotic Syndrome Study Network (NEPTUNE) cohort with glomerular disease. In the initial developmental cohort, 302 patients had same day random and 24-hour samples with a total of 827 paired measurements across all visits. The protein excretion (g/day) was higher in adult than pediatric patients. The correlation between the random urine protein creatinine ratio and 24-hour urine protein excretion was moderate in both groups (r of 0.60 and 0.67, respectively). However, the Log10 transformation of values strengthened correlations in both groups (r of 0.85 and 0.82, respectively). Associations were moderately stronger among obese patients. Prediction equations were developed and validated in 232 unique cases from NEPTUNE (R² of 0.65). Thus, in patients with glomerular disease and proteinuria, the urine protein creatinine ratio correlates only moderately with 24-hour urine protein excretion. However an estimating equation was developed to derive 24-hour urine protein excretion from random urine protein creatinine ratio values with improved precision. The long term prognostic value of Log10 transformed random protein creatinine ratios values requires future study.

Introduction

Proteinuria quantification is essential during the clinical evaluation of patients with glomerulonephritis since it is one of the strongest determinants of renal prognosis.^1-5 While therapeutic guidelines and targets are largely dependent upon the gold-standard 24-hour urine collection, this procedure is cumbersome for patients and potentially subject to variability in collection technique.^{6, 7} A random urine protein-to-creatinine ratio (UPCR) is therefore often used to estimate 24-hour urine protein excretion (24hUP). While random testing is recognized as an important screening tool for detection of very low-grade proteinuria in diabetic nephropathy, the correlation between random UPCR and 24hUP is not well-established in patients with glomerulonephritis who frequently excrete over 1g/day of protein in their urine. The time of random urine collection, degree of proteinuria and renal function, underlying histologic type of kidney disease, and handling of urine samples may all influence the accuracy of random UPCR,^8-10 and thus alter the correlation between protein measures in random and timed urine collections. Despite these limitations, data from normal and proteinuric adults and children suggest that random UPCR and protein: osmolality ratios provide reasonable approximations of 24-hour urinary excretion.^{11, 12} However, these relationships have not been validated in well-characterized cohorts of patients with defined glomerular diseases. To better understand the relationship between random urine UPCR and 24hUP, these laboratory values were evaluated in patients enrolled in the observational NEPTUNE cohort.

Results

Study population

Urine protein, albumin and creatinine in NEPTUNE were analyzed in batches in a centralized laboratory. Data for 302 patients across multiple study visits (827 total samples collected between 1/3/11 and 1/21/14) were available at the time of initial analysis (derivation sample; Table 1, Figure 1). Another set of 373 samples subsequently measured from 232 patients collected between 8/4/11 and 8/8/15 was used as the validation sample. Patients in the derivation sample were enrolled from 7/8/11 to 11/1/13; those in the validation sample enrolled from 7/8/11 to 9/92014; therefore derivation samples were more often from baseline, or earlier follow-up visits. Table 2 provides a description of random and 24-hour urine protein, creatinine, and albumin values by adults and pediatrics for both the derivation and validation cohorts. On average, proteinuria levels were higher in the derivation than the validation cohort. Proteinuria was also higher in adults than in pediatric patients (e.g., in the derivation cohort, adult and pediatric 24hUP = 1(0.3, 2.3) vs. 0.2(0, 0.7), respectively; Table 2).

Table 1.

Descriptive characteristics of patients in the derivation (n=302) and validation samples (n=232).

	Derivation sample n=827 observation n=302 patients	Validation sample n=373 observation n=232 patients
	N(%)/median(IQR)	N(%)/median(IQR)
Age
Pediatric (<18)	76 (25)	64 (28)
Adult (18+)	226 (75)	168 (72)
Sex
Male	192 (64)	145 (63)
Female	110 (36)	87 (37)
Race
Asian	42 (14)	30 (13)
Black/African American	56 (19)	42 (18)
Pacific Islander	1 (1)	2 (1)
White/Caucasian	175 (58)	142 (61)
Multiracial	14 (4)	9 (4)
Missing/did not report	14 (4)	7 (3)
Cohort
MCD	59 (20)	46 (20)
FSGS	92 (30)	70 (30)
MN	56 (19)	48 (21)
Other	95 (31)	68 (29)
Weight status (at first observation)
Underweight (BMI<18.5 or BMI %tile<5 for peds)	5 (2)	3 (1)
Normal weight (BMI: 18.5-25 or BMI %tile: 5-85 for peds)	78 (26)	62 (27)
Overweight (BMI: 25-30 or BMI %tile: 85-95 for peds)	89 (29)	72 (31)
Obese (BMI>30 or BMI %tile>95 for peds)	121 (40)	92 (40)
No baseline height/weight data	9 (3)	3 (1)
Proteinuria from all study visits
Random UP:C	1.2 (0.3, 3.1)	0.5 (0.1, 2.6)
24-hour urine protein (g)	0.7 (0.2, 2.1)	0.5 (0.1, 1.8)
Random UA:C	1.0 (0.2, 2.3)	0.4 (0.1, 1.8)
24-hour urine albumin (g)	0.6 (0.1, 1.7)	0.3 (0.1, 1.3)
Serum albumin from all study visits	3.8 (3.1, 4.2)	3.9 (3.4, 4.3)
Proteinuria at baseline
Random UP:C	2.5 (0.9, 4.9)	2.3 (0.4, 5.9)
24-hour urine protein (g)	1.6 (0.5, 3.2)	1.8 (0.5, 3.0)
Random UA:C	1.9 (0.7, 3.9)	1.7 (0.3, 4.3)
24-hour urine albumin (g)	1.3 (0.4, 2.5)	1.4 (0.3, 2.6)
Serum albumin at baseline	3.3 (2.6, 4.0)	3.0 (2.0, 3.7)
On steroids at baseline	90 (30)	74 (32)
On steroids at one or more study visits	106 (35)	84 (36)
Proteinuria >3.5g and serum albumin <3.5g/dL at baseline	73 (24)	61 (27)

Peds: pediatric patient cohort.

Figure 1. Study design.

The following adjustments were performed: we limited range of random UPCRs & 24-hour urine proteins to <10(g) (II) standardized 24-hour protein values to BSA (III) limited samples to 24-hour urine creatinine (12.1-28.9mg/kg for adult male and 10.7-26.0mg/kg for adult females) and (IV) log-transformed random UPCR & 24-hour urine protein. Subgroup variation in relationship between random and 24-hour urine using interaction terms between variables of interest and random UPCR. (Cohort, age sex, weight, BMI)

Table 2.

Description of central NEPTUNE urine data for the derivation sample cohort.

	Adults n= 666 observations n= 226 patients		Pediatrics n= 161 observations n= 76 patients
	Median (IQR)	Range	Median (IQR)	Range
24-Hour urine protein (g)	1 (0.3, 2.3)	0 - 31.3	0.2 (0, 0.7)	0 - 16.9
24-Hour urine creatinine (g)	0.8 (0.5, 1.1)	0 - 6.5	0.4 (0.2, 0.8)	0 - 2.7
24-Hour urine creatinine (mg/kg)	9.3 (6.2, 12.7)	0.1 - 82	10.1 (5.9, 15)	0.6 - 31.3
24-Hour urine albumin (g)	0.8 (0.2, 1.9)	0 - 23.8	0.1 (0, 0.6)	0 - 15.1
24-Hour UPCR	1.4 (0.4, 2.9)	0 - 33.1	0.5 (0.1, 1.6)	0 - 13.7
24-Hour UA:C	1.2 (0.3, 2.5)	0-34.6	0.4 (0.1, 1.3)	0 – 12.8
Random urine protein (mg/dL)	98.6 (26.2, 255.9)	0 - 6210	33.9 (3.7, 107.1)	0 - 3600
Random urine creatinine (mg/dL)	79.8 (50.1, 123.1)	7.9 - 519.3	85.8 (36.7, 139.2)	2.6 - 372.8
Random urine albumin (mg/L)	809.9 (187.7, 1774.9)	0 - 64200	254.9 (7.1, 858.7)	0 - 34700
Random UPCR	1.4 (0.4, 3.2)	0 - 39.7	0.4 (0.1, 1.7)	0 - 25.5
Random UA:C	1.2 (0.3, 2.5)	0 - 34.6	0.3 (0, 1.5)	0 - 24

Correlations between random and 24-hour collections

Pearson correlation coefficients estimating the relationship in the derivation sample between random UPCR and 24hUP are shown in Table 3. There was only a moderate correlation across all observations for adult and pediatric subjects (r=0.60 and 0.67, respectively). Correlations were stronger between random UPCR and 24-hour UPCR (aliquot from the 24h sample) than between random UPCR and 24hUP (r=0.79 and 0.84 for adults and pediatrics, respectively; See scatter plots in Figure 2). Adjusting for body surface area or lean body weight improved associations in pediatric but not adult subjects (Table 3). Log₁₀-transforming random UPCR and 24hUP strengthened the correlation to 0.85 for adults and 0.82 for pediatric subjects. Removing outliers (24-hour urine creatinine excretion outside of the reference range of 12.1-28.9mg/kg for males and 10.7-26.0mg/kg for females¹³) and log transforming values further improved the final correlation among adults (r=0.92), while in pediatric subjects removing outliers, log transforming and adjusting for body surface area resulted a correlation of r=0.80. The effect of urinary flow, as reflected by urinary creatnine concentration, did did/not have an additional effect on the correlation of random UPCR and 24hUP (Table 3).

Table 3.

Pearson correlation coefficients for correlations of random UPCR and 24hUP or UPCR taken from an aliquot of the 24-hour collection in NEPTUNE patients (pediatric and adult) with same-day matching samples from the derivation Cohort.

	Adults n= 226 patients			Pediatrics n= 76 patients
	n	24hUP (r)	24h aliquot UPCR (r)	n	24-Hour UP (r)	24h aliquot UPCR (r)
Overall	666	0.60	0.79	161	0.67	0.84
Single adjustments
Remove outliers* (I)	629	0.72	0.90	155	0.64	0.76
Adjust for BSA** (II)	650	0.62	---	159	0.74	---
Adjust for lean body weight¶	650	0.6	---	159	0.74	---
Drop inadequate samples*** (III)	204	0.65	---	---	---	---
Log-transform (IV)	666	0.85	0.94	161	0.82	0.88
Double adjustments
I + II	613	0.74	---	155	0.66	---
I + III	190	0.85	---	---	---	---
I + IV	629	0.84	---	155	0.80	---
II + III	204	0.69	---	---	---	---
II + IV	650	0.85	0.94	159	0.83	0.88
III + IV	204	0.92	---	---	---	---
Triple adjustments
I + II + III	190	0.85	---	---	---	---
I + II + IV	629	0.84	---	155	0.80	---
I + III + IV	190	0.93	---	---	---	---
II + III + IV	204	0.93	---	---	---	---
All adjustments
I + II + III + IV	190	0.94	---	---	---	---

^*Remove 24-Hour UP and random UPCR values>10,

^**Mostellar $\mathrm{BSA}=\sqrt{\mathit{height}\mathit{cm}\times \frac{\mathit{weight}\mathit{kg}}{3600}}$,

^***Include samples with 12.1-28.9mg/kg for males and 10.7-26.0mg/kg for females,

^¶LBW (male) = 1.10 × BWt − 0.0128 × BMI × BWt; LBW (female) = 1.07 × BWt − 0.0148 × BMI × BWt

Figure 2. Correlation of random UP:C and 24hUP from an aliquot of the 24h collection.

NEPTUNE patients with same-day matching samples from derivation (Panels A and C) and validation samples (Panels B and D) are shown separately. Correlations on a log-log scale are shown in Panels A and B, while receiver operating curves for various levels of 24hUP are shown in Panels C and D. *R² from a linear regression with 24-hour urine protein modeled as a function of predicted 24-hour urine protein. Separate prediction equations were used for children and adults. Adult=10^{(0.88×(log₁₀Spot UP:C))}; Pediatric=10^{(1.06×(log₁₀Spot UP:C))}. Pediatric prediction equation to give the units of 24-hour UP standardized to a body size of $1.73{\mathrm{m}}^2={10}^{\left(1.06\times \left({\log }_{10}\mathit{Random}\mathrm{UPCR}\right)\right)}\times \sqrt{height\left(\mathit{cm}\right)\times \mathit{weight}\left(\mathit{kg}\right)\times 1.73m^2}$

The regression equation best describing the relationship between random UPCR and 24h urine protein excretion (g/day) in adults was =10^{(0.88×(log₁₀Random UPCR))}. The corresponding pediatric equation was =10^{(1.06×(log₁₀Random UPCR))}. The pediatric prediction equation standardized to a body size of $1.73{\mathrm{m}}^2={10}^{\left(1.06\times \left({\log }_{10}\mathit{Random}\mathrm{UPCR}\right)\right)}\times \sqrt{\mathit{height}\left(\mathit{cm}\right)\times \mathit{weight}\left(\mathit{kg}\right)\times 1.73m^2}$. Case examples are shown in supplemental table 1.

Bland-Altman plots in Figure 3 demonstrate that the correlation between log-transformed 24-hour urine values and predicted values, based on the prediction equations above, were reasonably uniform across all ranges of the data in the derivation cohort, although the figures suggest somewhat greater variability at lower levels of protein (<0.1g). Overall the relationship between actual and predicted 24 hr UPCR showed less variability than between the predicted 24hUP and measured 24hUP.

Figure 3. Bland-Altman analysis comparing actual and predicted 24hUP (Panel A) and random and 24h UPCR (Panel B).

All values are log₁₀ transformed. The correlation is consistent across the range of values for both comparisons, although the relationship between actual and predicted 24 hr UPCR (Panel B) showed less variability than between the predicted 24hUP and measured 24hUP (Panel A).

Prediction of proteinuria thresholds

The ability of random UPCR to accurately predict 24hUP values above thresholds of 0.5g, 1.0g, 2.0g, 3.0g, 6.0g and 10.g are shown in Table 4. For example, when predicting values in adults, a threshold of random UPCR >0.60 accurately predicts a 24hUP >0.5g with positive and negative predictive values of 0.91 and 0.89 respectively. Accuracies were >0.80 for all thresholds among both adult and pediatric subjects. When using all 226 adult patients and all 666 adult observations, area under the curve statistics for each threshold were ≥0.90. Pediatric c-statistics were all ≥0.83 (Figure 4).

Table 4.

Characteristics of random UPCR ratio to detect different degrees of proteinuria from same day 24-hour urine from baseline samples from the derivation cohort.

	>0.5g	>1.0g	>2.0g	>3.0g	>6.0g	>10.0g
Adults (n=152)	123	105	78	49	23	10
Random UPCR threshold	0.60	1.54	2.03	5.45	11.1	16.45
Sensitivity	0.98	0.91	0.91	0.59	0.35	0.50
Specificity	0.59	0.72	0.70	0.92	0.97	0.99
Positive predictive value	0.91	0.88	0.76	0.78	0.67	0.83
Negative predictive value	0.89	0.79	0.88	0.83	0.89	0.97
Accuracy	0.91	0.86	0.81	0.82	0.88	0.96
Pediatrics (n=39)	19	15	10	6	3	0
Random UPCR threshold	0.44	1.32	4.23	18.1	18.1	---
Sensitivity	0.95	0.93	0.80	0.17	0.17	---
Specificity	0.75	0.88	0.86	1.00	1.00	---
Positive predictive value	0.78	0.82	0.67	1.00	1.00	---
Negative predictive value	0.94	0.95	0.93	0.87	0.87	---
Accuracy	0.85	0.90	0.85	0.87	0.87	---

Figure 4. ROC analysis of the ability of random UPCR to predict 24hUP above clinically relevant thresholds for adults (Panel A) and pediatrics (Panel B).

Repeated measures ROC estimated using generalized linear mixed models with variance components covariate structure

Table 5 contains tests for subgroup variation in the correlation of log-transformed UPCR and log-transformed 24-hour urine protein. The regression coefficient did not differ by sex in the adult analysis (β_Male=0.89 (0.83, 0.95) β_Female=0.87 (0.82, 0.92) p=0.40) or pediatric analysis (β_Male=0.88 (0.76, 1.0) β_Female=0.81 (0.71, 0.91) p=0.15). Regression coefficients also did not vary by disease cohort among adult or pediatric subjects. Regression coefficients significantly differed by weight status among adults, with the strongest correlation among obese patients (β_{Normal weight} =0.78 (0.69, 0.88); β_Overweight=0.85 (0.78, 0.93); β_Obese=0.93 (0.85, 1.0) p=0.03) Weight status did not affect correlations in the adult validation sample although it did in the pediatric validation cohort (p=.02).

Table 5. Subgroup variation in correlations between log(random UPCR) and log(24-hour urine protein) in the derivation and validation cohorts.

Interaction models were tested using linear mixed-effects models adjusting for repeated measures within patients using a compound symmetry covariance structure.

	Derivation		Validation
	Adults n=666 observations n=226 patients		Adults n=283 observations n=168 patients
	β(95%CI)	p-value	β(95%CI)	p-value
Sex		0.40		0.06
Males	0.89 (0.83, 0.95)		0.76 (0.61, 0.90)
Females	0.87 (0.82, 0.92)		0.88 (0.75, 1.01)
Race		0.41		0.04
Black/African- American	0.88 (0.83, 0.93)		0.90 (0.84, 0.97)
Non-Black/African American	0.84 (0.76, 0.93)		0.79 (0.68, 0.90)
Cohort		0.51		0.85
MCD	0.90 (0.77, 1.03)		0.85 (0.69, 1.00)
FSGS	0.81 (0.70, 0.90)		0.83 (0.72, 0.95)
MN	0.90 (0.81, 0.99)		0.90 (0.80, 0.99)
Other	0.87 (0.79, 0.95)		0.86 (0.73, 0.98)
Weight status		0.03		0.34
Normal weight	0.78 (0.69, 0.88)		0.92 (0.81, 1.03)
Overweight	0.85 (0.78, 0.93)		0.85 (0.74, 0.96)
Obese	0.93 (0.85, 1.00)		0.84 (0.74, 0.94)
	Pediatrics n=161 observations n=76 patients		Pediatrics n=90 observations n=64 patients
Sex		0.15		0.30
Males	0.88 (0.76, 1.00)		0.86 (0.78, 0.94)
Females	0.81 (0.71, 0.91)		0.89 (0.83, 0.95)
Race		0.06		0.26
Black/African- American	0.72 (0.61, 0.84)		0.77 (0.64, 0.91)
Non-Black/African American	0.82 (0.72, 0.93)		0.84 (0.72, 0.97)
Cohort		0.18		0.77
MCD	0.77 (0.64, 0.90)		0.79 (0.64, 0.95)
FSGS	0.86 (0.68, 1.03)		0.90 (0.53, 1.26)
MN	---		---
Other	1.09 (0.76, 1.43)		0.88 (0.63, 1.12)
Weight status		0.69		0.02
Normal weight	0.79 (0.64, 0.94)		0.76 (0.56, 0.95)
Overweight	0.89 (0.71, 1.01)		0.85 (0.65, 1.05)
Obese	0.85 (0.67, 1.03)		0.90 (0.68, 1.12)

Low 24-hour urine creatinine excretion in NEPTUNE cohort

Interestingly, total urinary creatinine excretion was systematically lower than that expected for adults when compared to published reference ranges for men and women (Figure 5).¹³ To determine if patient 24 hr urine collections were consistent over time, Intra patient urine creatinine variability was assessed by repeated measures ANOVA for patients with 3 or more available measures. A model converged on 48 subjects who have 24-hour urine creatinine data from V2, V4, V5, and V6. The results of this repeat measures ANOVA support internal consistency of collection by subject (P=0.4972; Supplemental Table 1). The observed bias compared to the published vales was independent of weight (Figure 5). Urinary creatinine excretion did not associate with serum albumin suggesting the lower urine creatinine excretion rates seen in this cohort was independent of nutritional status. Nevertheless steroid exposure at baseline modestly correlated with lower urine creatinine (27% of those with adequate UC were on steroids at that visit compared to 19% of those with inadequate 24-urine UC (p=0.02), suggesting reduced lean muscle mass might be a contributing factor. In addition within patient comparisons were stratified by steroid exposure: (1) paired visits where both observations where when the patient was on steroids (n=53 unique patients), (2) pairs where one observation was on and one was off steroids (n=79 unique patients), (3) paired observations where both were from when the patient was off steroids (n=135 unique patients). There was little difference across these three strata, and steroid exposure appeared to have no effect on the intrapatient correlation of timed urine creatinine values.

Figure 5. Correlation of individual patients’ 24-hour urine creatinine excretion with weight (n=226 patients/ 666 observations).

Reference ranges (2.5%-97.5%; mg/kg weight) for adult men (blue) and women are shown. The figure excludes pediatric patients. On average both men and women excreted less creatinine than expected.

Recently Ix and colleagues developed an equation to predict 24-hour creatinine excretion using data from several large cohorts with and without kidney disease.¹⁴ Among NEPTUNE adult patients, the measured creatinine excretion was systematically 50% lower than predicted by this equation (Figure 6). When this formula was applied to an adult cohort without nephrotic syndrome ¹⁵, the mean actual 24 hr creatinine excretion as measured in the same central laboratory was 84% of that predicted by the formula in 415 women and 87% of that predicted in 290 men.

Figure 6. Measured 24-hour urine creatinine excretion in adult NEPTUNE patients versus that predicted by the equation of Ix, et. al ¹⁴.

The distribution of measured urine creatinine (g/24hrs) had a median of 0.8 and an IQR of 0.5 to 1.1 compared to the predicted 24-hour creatinine excretion with a median of 1.6 and an IQR of 1.3 to 1.8.

Discussion

Quantification of proteinuria is vital for monitoring disease activity and response to therapies in patients with glomerular disease.^{1-5, 16}. The gold standard is considered 24hUP but random UPCR is often used because of the challenges inherent in 24 hour collections. Furthermore, clinicians often think in terms of urine protein excretion rates. However, this study using the NEPTUNE cohort of patients with MCD, FSGS, MN and other proteinuric glomerular diseases demonstrated that the random UPCR only modestly correlates with 24hUP. Fortunately a simple log₁₀ transformation of the random UPCR can be utilized to derive a more accurate estimate of the 24hUP among both adult and pediatric patients. Using this strategy a strong correlation (r=0.87) and modest predictability (R²=0.65) was observed between random UPCR and 24hUP in the validation sample. The strength of this correlation was consistent across clinical variables including gender, histologic diagnosis and severity of proteinuria, although weight influenced the association to some extent.

Given the importance of proteinuria for decisions such as initiation or withdrawal of immunosuppression, a statistically “moderate” correlation is of obvious concern. The prognosis of IgA nephropathy, for example, is closely correlated to proteinuria and small differences in 24hUP are associated with large differences in the rate of renal function decline.^{1, 17} A correlation coefficient of 0.60-0.67 would not offer sufficient precision to confidently estimate prognosis or make decisions regarding immunotherapy. Similarly in FSGS and MN, clinical guidelines regarding immunotherapy may be dependent upon small differences in proteinuria, therefore excellent correlation with random values is essential. This must be balanced, however, against the consideration that 24h urine collections are cumbersome for patients. Given the inherent limitations of each approach (difficulties in obtaining accurate timed collection versus missed biologic variability in a random sample), studies to directly compare the relative predictive value of a random UPCR and 24hUP are sorely needed. Of note the current study does provide good evidence that the random UPCR can reliably be used to predict clinically significant thresholds of 24-hour protein between 0.5 to 10 g/day (AUC 0.91-0.97; Figures 2, 4).

The first study to examine the relationship of random UPCR to 24hUP was performed in only 49 CKD patients among whom only three had a 24hUP > 3.5g per1.73m².¹⁸ The correlation coefficient was 0.97 in those who excreted >3.5g protein per 24-hours. Other studies suggest correlations between 0.56 and 0.98, although many of these were done on patients without glomerulonephritis (e.g.. preeclampsia) and with 24hUP in the sub-nephrotic range¹⁹. Two studies in renal transplant recipients found correlations of 0.77 and 0.76.^{20, 21} Other studies have evaluated cohorts of both normal individuals and patients with kidney disease.^11,12 In these cases, urinary protein to osmolality and protein to creatinine ratios were reasonable predictors of 24hUP. In particular, cut points for random urine samples could be optimized to differentiate pathological levels of proteinuria. Studies restricted to patients with chronic kidney disease and diabetic nephropathy have also shown weaker correlations in patients with nephrotic range proteinuria.^{22, 23}

Two meta-analyses have examined the diagnostic accuracy of random UPCR to predict 24hUP above clinical thresholds of 300mg/day in patients with preeclampsia.^{16, 24} Morris, et al., found sensitivities ranging 0.65 to 0.89 and specificities from 0.63 to 0.87, while Sanchez-Ramos et al., reported pooled sensitivities and specificities of 0.91 and 0.86.^{16, 24} This analysis showed a similar strength of diagnostic utility in patients with nephrotic syndrome and higher ranges of proteinuria. In a study restricted to adults, urinary protein/osmolality ratio was a better predictor than UPCR of pathological proteinuria.¹². In pediatrics, the opposite was found.¹¹ However, in both cohorts, protein/osmolality and creatinine ratios were very comparable for predicting 24hUP.

In this study the correlation of UPCR and 24hUP was further optimized when only samples with “adequately” collected creatinine were considered. This suggests that the relatively poor correlation between UPCR and 24hUP may in part be related to collection procedures. However we cannot eliminate the possibility that this also relates to differences in creatinine excretion among the patients with known glomerular disease that comprised our study population. Indeed in the current study, 24-hour creatinine excretion was less than that in published reference range studies (restricted to adults)^{13, 14}. Creatinine excretion was also, on average, about half that predicted by an equation recently developed in a population that included subjects with CKD, largely secondary to diabetic nephropathy (Figure 6).¹⁴ Often, the expected creatinine excretion, on a g/kg basis, is used to determine if a 24-hour collection is accurate.¹⁴ It is unclear if this cohort with known CKD and nephrotic syndrome has lower than expected creatinine generation or if they consistently under collected. In further analysis we found, in general, the 24-hour creatinine excretion was consistently biased low, suggesting that it was not just a matter of randomly inaccurate (over and under) collection. and more likely due to increased catabolic state of these individuals (many of whom were already receiving steroids and were hypoalbuminemic at their baseline visit²⁵). Patients in the study by Ix and colleagues though proteinuric (median 24 hr albumin of 4 g/L), were mostly diabetic and not likely to be on steroids or other immunosuppressants. Another potential factor that can influence correlations of random UPCR and 24hUP is the urinary flow rate. In a recent report from Yang and colleagues ²⁶, random UPCR from dilute urines (urine specific gravity ≤ 1.005) overestimated 24hUP, while for concentrated urines (urine specific gravity ≥ 1.015) the reverse was true. In our analysis urine concentration was taken into account as we looked at the correlation of random UPCR and 24hrUP. Indeed this interesting observation may in part explain why random and 24hr UPCR correlated better than random UPCR and 24hrUP. At a minimum these studies highlight the difficulties inherent in using 24 hr creatinine excretion to judge the completeness of a 24 hr urine collection.

Conversely, the current study revealed a relatively good correlation between the random and 24 hr protein creatinine ratios. This correlation was in fact much better than for random protein creatinine ratio and 24-hour protein excretion. One argument for a 24-hour collection is that protein excretion has biologic variability through the day, and a random collection may under or overestimate the average protein excretion rate.²⁷ The observation that random UPCR and 24 hr UPCR values correlate well argues to some extent that the protein creatinine ratio is relatively constant through the day, at least in a population with established proteinuric CKD. Thus, future studies should compare the utility of random UPCR (and perhaps 24 hr UPCR) versus 24hUP in predicting long term outcomes such as progression to end stage renal disease.

This study has certain limitations. We only report on total protein and not albumin excretion, although the literature in nephrotic syndrome is based upon total protein. Protein excretion rates were relatively lower in our validation cohort compared to the derivation samples. We also lack data on long term outcomes, accuracy of the 24-hour collections, or a gold standard way to assess true 24hUP (directly observed in a clinical research unit). Nevertheless, this is one of the largest comparisons of random UPCR and 24hUP published to date in a population of patients with biopsy-proven nephrotic diseases, across a wide range of protein excretion.

In conclusion, among patients with MCD, FSGS, and MN, UPCR correlates moderately well with 24hUP, and a non-linear estimating equation can be used to improve the estimated 24hUP from a random UPCR. Furthermore, random UPCR correlates well with 24-hour UPCR. Those patients with 24hUP above clinically relevant cut points can also be estimated with caution from the random UPCR. Individuals with these nephrotic diseases also appear to excrete less creatinine than predicted from published reference studies. Since 24hUP is susceptible to collection errors, and gold standards for predicted creatinine excretion are lacking, additional studies to evaluate random UPCR vs 24hUP and UPCR with clinical outcomes are warranted.

Methods

Patient protocol

Following Institutional Review Board (IRB) approval, patient/guardian consent and minor assent, eligible patients were enrolled in the NEPTUNE longitudinal cohort study at the time of clinically indicated initial kidney biopsy. Full details of the study design and baseline description of the cohort have been published elsewhere.^{28, 29} Eligibility criteria include: individuals of any age scheduled for a clinically-indicated renal biopsy with proteinuria ≥ 500 mg/day from a 24-hour or random (protein to creatinine ratio (UPCR) > 0.5 g/g) urine sample. Patients with sub-nephrotic proteinuria were included to capture the broad spectrum of clinical presentations. Exclusion criteria include patients with clinical or pathology evidence of other kidney diseases (e.g. systemic diseases such as diabetic nephropathy, systemic lupus erythematosus, vasculitis, Alport syndrome, amyloidosis, monoclonal gammopathy), prior solid organ transplant, life expectancy less than 6 months, unwillingness or inability to consent.

Patients were eligible for this analysis had a collected random and 24-hour urine sample collected on the same 24-hour day. Data for 302 patients across multiple study visits (827 total samples collected between 1/3/11 and 1/21/14) were available at the time of initial analysis and served as the development sample. Another 373 samples collected from 232 patients between 8/4/11 and 8/8/15) and subsequently measured were used as the validation sample.

Urine measurements

All urine samples were processed per the published NEPTUNE study protocol ³⁰. Whole urine aliquots frozen to −70°C were used for the current measurements. At the time of assay samples were thawed to 37°C and spun (1000g × 12 min) to remove cellular debris. All urine measures were performed in a centralized laboratory at the Mayo Clinic, Rochester, MN using a Cobas C311 autoanalyzer (Roche, Indianapolis, IN). Creatinine was measured using the standardized (isotope dilution mass spectrometry traceable) enzymatic creatinine assay (Roche). Total urine protein was measured using a pyrogallol red colorimetric assay (Wako Diagnostics, Richmond, VA).

Statistical analysis

Descriptive characteristics of the laboratory values and clinical characteristics of the patients are presented using medians, inter-quartile ranges, frequencies and percentages. The correlation between random UPCR and 24-hour urine protein in grams was measured using Pearson Correlation Coefficients, scatter plots, and Bland-Altman plots. All analyses were stratified by age: adults (18+) vs. pediatrics (<18). After estimating a raw correlation, the analysis was adjusted by (I) limiting the range of random UPCRs and 24-hour urine proteins to values <10(g) (II) standardizing the 24-hour urine protein values to body surface area (III) limiting the samples to those with adequate weights of 24-hour urine creatinine (12.1-28.9mg/kg for males and 10.7-26.0mg/kg for females)³¹, and (IV) log-transforming random UPCR and 24-hour urine protein. ROC logistic regression analyses were done to assess the diagnostic utility of random UPCR in predicting 24-hour urine protein values above specified clinically useful thresholds: >0.5g, >1.0g, >2.0g, >3.0g, >6.0g, and >10.0g. Accuracy was defined as the proportion of patients correctly identified in the entire ROC analysis $\left(\frac{\Sigma \mathit{True}\mathit{positive}+\Sigma \mathit{True}\mathit{negative}}{\Sigma \mathit{Total}\mathit{population}}\right)$. This combines information from positive and negative predictive values. These analyses were done in two iterations. First, only baseline samples were used. Next, all samples were used and a generalized linear mixed model was incorporated to account for the dependence of the repeated measures.^{32, 33} Subgroup variation in the relationship between random and 24-hour urine values was also tested using interaction terms between clinical variables of interest and random UPCR in a series of linear mixed effects models. Subgroup variation by disease cohort (FSGS, MCD, MN, other), sex, and weight (normal weight, overweight, obese) was assessed. All analyses were completed using SAS v9.4.

Participating Institutions: Universities with CTSAs are marked with *.

** Manuscript NEPTUNE Authors and Institutions

Katherine Dell, MD, Case Western Reserve University

John Sedor, MD, Case Western Reserve University

Kevin V Lemley, MD, PhD, Children’s Hospital Los Angeles

Tarak Srivastava, MD, Children’s Mercy Kansas City

Christine Sethna, MD, Cohen Children’s Hospital

Laurence Greenbaum, MD, PhD, Emory University* and Children’s Healthcare of Atlanta

Cynthia C Nast, MD, Cedars-Sinai Health System

Alicia M Neu, MD, Johns Hopkins University*

Fernando Fervenza, MD, PhD, Mayo Clinic

Fredrick Kaskel, MD, PhD, Montefiore Medical Center*

Michelle H Mokrzycki, MD, Montefiore Medical Center*

Jeffrey B Kopp, MD, NIH/NIDDK

Stephen M Hewitt, MD, PhD, NIH/NIDDK

Avi Z Rosenberg, MD, PhD, NIH/NIDDK

Olga Zhdanova, MD, New York University Langone Medical Center*

Howard Trachtman, MD, New York University Langone Medical Center*

Richard Lafayette, MD, Stanford University

Crystal Gadegbeku, MD, Temple University

Duncan Johnstone, MD, PhD, Temple University

Daniel C Cattran, MD, University of Toronto

Ambarash Athavale, MD, University of Illinois at Chicago

Alessia Fornoni, MD, PhD, University of Miami*

Laura Barisoni, MD, University of Miami*

Matthias Kretzler, MD, University of Michigan

Debbie Gipson, MD, MS, University of Michigan

Matthew Sampson, MD, MS, University of Michigan

Patrick Nachman, MD, University of North Carolina at Chapel Hill*

Keisha Gibson, MD, MS, University of North Carolina at Chapel Hill*

Lawrence Holzman, MD, University of Pennsylvania

Kevin Meyers, MD, University of Pennsylvania

Kamalanathan Sambandam, MD, University of Texas Southwestern

Elizabeth Brown, MD, University of Texas Southwestern

Peter Nelson, MD, PhD, University of Washington

Ashley Jefferson, MD, University of Washington

Sangeeta Hingorani, MD, University of Washington

Katherine Tuttle, MD, University of Washington

Barry Freedman, MD, Wake Forest University

Jen Jar Lin, MD, Wake Forest University