A, Random UAC (9 studies). B, Random ACR (13 studies). ACR indicates urine sample measuring the ratio of albumin to creatinine; AUC, area under the curve; SENS, sensitivity; SPEC, specificity; SROC, summary receiver operating characteristic; and UAC, urine sample measuring the albumin concentration.
eAppendix. Study Protocol and Search Strategies
eReferences. References for Study Protocol
eFigure 1. Summary of Study Identification and Selection
eFigure 2. Summary for Risk of Bias of Included Studies
eFigure 3. Risk of Bias Graph of Included Studies
eFigure 4. Funnel Plot for Publication Bias Assessment of Studies
eFigure 5. Paired Forest Plots of the Sensitivity and Specificity of Random Urine Albumin Concentration for the Detection of Microalbuminuria in Patients With Diabetes
eFigure 6. Paired Forest Plots of the Sensitivity and Specificity of Random Urine Albumin to Creatinine Ratio for the Detection of Microalbuminuria in Patients With Diabetes
Wu H, Peng Y, Chiang C, Huang J, Hung K, Wu K, Tu Y, Chien K. Diagnostic Performance of Random Urine Samples Using Albumin Concentration vs Ratio of Albumin to Creatinine for Microalbuminuria Screening in Patients With Diabetes MellitusA Systematic Review and Meta-analysis. JAMA Intern Med. 2014;174(7):1108-1115. doi:10.1001/jamainternmed.2014.1363
A random urine sample measuring the albumin concentration (UAC) without simultaneously measuring the urinary creatinine is less expensive than measuring the ratio of albumin to creatinine (ACR), but comparisons of their diagnostic performance for microalbuminuria screening among patients with diabetes mellitus (DM) have not been undertaken in previous meta-analyses.
To compare the diagnostic performance of the UAC vs the ACR in random urine samples for microalbuminuria screening among patients with DM.
Electronic literature searches of PubMed, MEDLINE, and Scopus for English-language publications from the earliest available date of indexing through July 31, 2012.
Clinical studies assessing the UAC or the ACR of random urine samples in detecting the presence of microalbuminuria among patients with DM using a urinary albumin excretion rate of 30 to 300 mg/d in 24-hour timed urine collections as the criterion standard.
Data Extraction and Synthesis
Bivariate random-effects models for analysis and pooling of the diagnostic performance measures across studies, as well as comparisons between different screening tests.
Main Outcomes and Measures
The primary end point was the diagnostic performance measures of the UAC or the ACR in random urine samples, as well as comparisons between them.
We identified 14 studies, with a total of 2078 patients; 9 studies reported on the UAC, and 12 studies reported on the ACR. Meta-analysis showed pooled sensitivities of 0.85 and 0.87 for the UAC and the ACR, respectively, and pooled specificities of 0.88 and 0.88, respectively. No differences in sensitivity (P = .70), specificity (P = .63), or diagnostic odds ratios (P = .59) between the UAC and the ACR were found. The time point of urine collection did not affect the diagnostic performance of either test.
Conclusions and Relevance
The UAC and the ACR yielded high sensitivity and specificity for the detection of microalbuminuria. Because the diagnostic performance of the UAC is comparable to that of the ACR, our findings indicate that the UAC of random urine samples may become the screening tool of choice for the population with DM, considering the rising incidence of DM and the constrained health care resources in many countries.
Diabetes mellitus (DM) is a global epidemic and the leading cause of chronic kidney disease because of a worldwide increase in the prevalence of type 2 DM and obesity.1,2 Microalbuminuria is an independent risk factor for the decline in renal function as well as cardiovascular morbidity and mortality,1,3,4 and published guidelines advocate screening for microalbuminuria in patients with DM.1,5- 7Microalbuminuria is defined as a urinary albumin excretion rate (UAER) in the range of 30 to 300 mg/d, and the definitive measurement is based on a timed urine collection during 24 hours.6- 8
The 24-hour or timed collections are arduous, and several studies8- 11 have shown poor compliance, which gives rise to inaccuracy. Therefore, most professional organizations prefer the use of random urine samples for microalbuminuria screening.1,5- 7 A routine urine dipstick is often used to detect macroalbuminuria (UAER >300 mg/d) but is insensitive to microalbuminuria.7 The measurement of a random urine sample measuring the albumin concentration (UAC) is much more sensitive but may be influenced by the patient’s hydration status.5,12,13 Nevertheless, screening by the ratio of albumin to creatinine (ACR) in a random urine sample also requires the measurement of urinary creatinine, which is subject to additional variations across individual patients as well as different methods and laboratory conditions of the measurement.12,14- 16 In addition, women, the elderly, and patients with decreased muscle mass tend to excrete less urinary creatinine, and using the ACR is more likely to diagnose those subpopulations as having microalbuminuria.12,15 Furthermore, previous studies17,18 reported that the ACR costs 15% more than the UAC in the United States and 43% more in Brazil, resulting in substantial costs for large numbers of urine samples obtained in mass screenings.
Most studies evaluating the UAC or the ACR in patients with DM have included few participants. Therefore, considerable overlap exists between the 95% CIs for the diagnostic performance of the UAC and the ACR vs timed urine collection procedures. The few studies directly comparing the diagnostic performance of the UAC and the ACR have had conflicting results. Some studies14,19 have demonstrated that the ACR is better than the UAC, while other studies20,21 have shown an equivalent diagnostic performance for the UAC and the ACR. To date, no published systematic review has compared the UAC with the ACR for their diagnostic performance. In this study, we compared the UAC vs the ACR for detecting microalbuminuria in random urine samples among patients with DM using 24-hour timed urine collection as the criterion standard.
We conducted electronic English-language literature searches of PubMed, MEDLINE, and Scopus from the earliest available date of indexing through July 31, 2012. We also hand-searched the reference lists of identified publications for additional studies. The detailed study protocol and search strategies are provided in the eAppendix in the Supplement.
We included a study if it met the following conditions: (1) It assessed the UAC or the ACR of a random urine sample as an index test (ie, the test under investigation) to evaluate the presence of microalbuminuria among patients with DM and age older than 18 years. (2) The reference standard (ie, the criterion standard) of microalbuminuria was defined as a UAER of 30 to 300 mg/d by 24-hour timed urine collection. (3) It reported cases in absolute numbers of true-positive, false-positive, false-negative, and true-negative results, or these data were derivable from the published results.
If a study presented multiple cutoff values for an index test and, as a consequence, reported multiple pairs of sensitivity and specificity estimates, the data with the best estimates were extracted. If a study presented different cutoff values of an index test for male and female patients, the data of the different sexes were analyzed as separate studies. Eligible studies included patients with a UAER not exceeding 300 mg/d by 24-hour timed urine collection. Studies of patients with any type of DM, as well as studies reporting outcomes from diabetic subgroups, were included. Eligible studies had to be published as full-length articles or letters in peer-reviewed English-language journals.
Two of us (H.-Y.W. and Y.-S.P.) independently performed data extraction. Extracted information included details of the study design and outcomes as well as patient demographics. When relevant information regarding design or outcomes was unclear or when doubt existed about duplicate publications, the study authors were contacted for clarification. Two of us (H.-Y.W. and Y.-S.P.) independently assessed methodological quality using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool.22 Disagreement between the 2 authors was resolved by discussion.
All data from each eligible study were extracted and entered into a spreadsheet (Excel 2007; Microsoft Corporation). Categorical variables are presented as frequencies or percentages, and continuous variables are presented as mean values unless stated otherwise. Measures of the diagnostic performance, including sensitivity, specificity, and diagnostic odds ratios (DORs), are reported as point estimates with 95% CIs. A DOR can be calculated as the ratio of the odds of positivity in a disease state relative to the odds of positivity in the nondisease state, with higher values indicating better discriminatory test performance.23 Between-study statistical heterogeneity was assessed using I2 and the Cochrane Q test on the basis of the random-effects analysis.24 Publication bias was examined using the effective sample size funnel plot and associated regression test of asymmetry described by Deeks and colleagues.25
We used the bivariate random-effects model for analysis and pooling of the diagnostic performance measures across studies, as well as comparisons between different index tests.26,27 The bivariate model estimates pairs of logit-transformed sensitivity and specificity from studies, incorporating the correlation that might exist between sensitivity and specificity. We also used the model to create hierarchical summary receiver operating characteristic curves and to estimate the area under the curve.28 When statistical heterogeneity was substantial, we performed meta-regression to identify potential sources of bias.29 Pooled estimates were also calculated for subgroups of studies that were defined according to specific study designs. Two-sided P ≤ .05 was considered statistically significant. Statistical analyses were performed with commercial software programs (SAS, version 9.2; SAS Institute and STATA, version 11.1; StataCorp LP).
The flowchart in eFigure 1 in the Supplement shows the literature search process. We identified 281 articles from PubMed, 160 articles from MEDLINE, 87 articles from Scopus, and 2 additional articles from hand-searching. After these searches were combined and reviewed, 14 studies met our inclusion criteria.
The clinical and methodological characteristics, as well as the main results of each trial, are summarized in Table 1 and Table 2. We retrieved 14 eligible studies, which enrolled a total of 2078 patients. Nine studies reported on the UAC, with a mean microalbuminuria prevalence of 33.4%, and 12 studies reported on the ACR, with a mean microalbuminuria prevalence of 33.6%. One study35 reported on the ACR using different cutoff values for male and female patients. eFigure 2 and eFigure 3 in the Supplement summarize the QUADAS assessment. Overall, the quality of the studies was deemed satisfactory. However, the QUADAS tool showed that unclear blinding during interpretation of results and the lack of reporting for uninterpretable results might be potential sources of bias. Withdrawals from some studies were not clearly explained, which may also result in bias. The funnel plots (eFigure 4 in the Supplement) and regression tests indicated no significant publication bias (P = .92 and P = .48 for the UAC and the ACR, respectively).
Pooled estimates of sensitivity, specificity, and DORs of the 2 index tests are summarized in Table 2 and in eFigure 5 and eFigure 6 in the Supplement. The pooled sensitivity across studies on the UAC was 0.85 (95% CI, 0.80-0.89), the pooled specificity was 0.88 (95% CI, 0.80-0.93), and the pooled DOR was 39.7 (95% CI, 20.9-75.3). The pooled sensitivity across studies on the ACR was 0.87 (95% CI, 0.81-0.91), the pooled specificity was 0.88 (95% CI, 0.84-0.91), and the pooled DOR was 46.4 (95% CI, 26.0-82.7). The Figure shows hierarchical summary receiver operating characteristic curves for both index tests and indicates that the areas under the curve were 0.91 (95% CI, 0.88-0.93) and 0.94 (95% CI, 0.91-0.95) for the UAC and the ACR, respectively. In a comparison of the diagnostic performance of the UAC vs the ACR, no significant differences were observed in sensitivity (P = .70), specificity (P = .63), or DORs (P = .59).
Between-study heterogeneity was present for sensitivity and specificity among studies of both index tests. Table 3 lists the results of univariate meta-regression analyses for identifying potential sources of heterogeneity. For studies evaluating the UAC, the age of the study participants could be the most probable source of heterogeneity. For studies evaluating the ACR, the number of participants included in each study, as well as the quality and year of publication of the study, explained some of the heterogeneity. For both index tests, whether urine samples were collected in the morning or not did not result in heterogeneity.
Table 4 summarizes the results of subgroup analyses for studies using urine collected in the morning12,18- 20,32- 37 or using urine collected at any time during the day.13,21,30,31 Comparing the diagnostic performance of the UAC vs the ACR, no significant differences were observed for sensitivity or specificity in either subgroup. We also analyzed 7 studies that evaluated both the UAC and the ACR for the same patients, and no difference in the diagnostic performance was found between the index tests.12,13,18- 21,32
To the best of our knowledge, this study is the first systematic review and meta-analysis comparing the diagnostic performance of different tests of random urine samples for microalbuminuria screening in patients with DM. Our findings show that the UAC, which has high sensitivity and specificity, is comparable to the ACR for accurate detection of microalbuminuria among patients with DM using a UAER of 30 to 300 mg/d in 24-hour timed urine collections as the criterion standard. The strengths of our study are that we followed a standard protocol and used a comprehensive search strategy. We applied rigorous methods for data analysis, including the bivariate random-effects model and hierarchical summary receiver operating characteristic curve analyses. To explain the heterogeneity, we added covariates to the bivariate model for meta-regression.
Few systematic reviews have evaluated the diagnostic performance of screening tests among random urine samples for microalbuminuria in patients with DM. Because of the narrative nature of reviews,14,38 no conclusion could be made regarding the comparison of the UAC and the ACR. Moreover, the previous reviews included studies with criterion standards of different cutoff values or urine collection times, which made pooling of the diagnostic performance difficult. Only one meta-analysis11 has evaluated the diagnostic performance of the ACR and showed a pooled DOR of 45.8 (95% CI, 28.5-73.4) using the criterion standard of 30 to 300 mg/d in 24-hour timed urine collections, which was similar to our pooled results. However, no previous meta-analysis has evaluated the UAC or compared the UAC with the ACR for their diagnostic performance. In this study, no difference was found in the ability of the UAC vs the ACR to detect microalbuminuria.
Previous studies1,6,7 have shown that the ACR in mid-morning urine samples correlates most closely with 24-hour specimens. Considering the possible influence of the time point of urine collection, we further assessed this issue by meta-regression and subgroup analyses. For both the UAC and the ACR, whether urine samples were collected in the morning or not did not cause significant heterogeneity. In addition, no significant difference in the diagnostic performance between the UAC and the ACR was noted in either subgroup with the use of morning specimens or with urine collected at any time during the day. Therefore, whether or not urine samples were collected in the morning did not seem to significantly influence our study results. However, the results of meta-regression or subgroup analyses should be interpreted with caution because of limited statistical power.
Only 2 studies13,18 have assessed the cost of the UAC vs the ACR in the same participants; both concluded that the UAC had a similar diagnostic performance but with lower cost compared with the ACR and recommended the UAC as the screening test of choice. Compared with the UAC, the additional expense for the ACR is dependent on the costs of laboratory materials, as well as the hourly rates of technicians and physicians, and can cost up to almost 50% more.12,18 To attain better cost-effectiveness, microalbuminuria screening requires tests that are similarly accurate but less expensive and less burdensome, such as the UAC of random urine samples.14,39
Our study nonetheless has several limitations. First, we neither identified unpublished studies nor included non–English-language articles. However, we performed a comprehensive systematic review of 3 electronic databases with a predefined protocol, and no publication bias was identified among the retrieved studies. Therefore, the results of our meta-analysis should be robust. Second, whereas 7 studies evaluated both the UAC and the ACR in the same participants, a precise estimate for the comparison of different index tests requires the original individual patient-level data for those studies. However, because of the difficulty in retrieving these complete data, we applied the bivariate model to allow for random effects to be correlated, which can yield a similar result. In addition, subgroup analysis for the 7 studies also showed no difference in the diagnostic performance between the UAC and the ACR. Third, considerable heterogeneity was found in the pooled estimates, and despite our attempts to explain it through meta-regression, substantial heterogeneity might remain unexplained. Many factors possibly contributing to the residual heterogeneity could not be assessed because they were not reported. For example, uninterpretable test results are likely to have an important effect on test performance,22 but none of the included studies reported them. Furthermore, the few included studies limited the power for further exploration of heterogeneity with multivariate meta-regression. Fourth, the microalbuminuria prevalence in the included studies is similar to that of recent large trials of patients with DM.40,41 Because all included studies required 24-hour timed urine collections as the criterion standard, patients who participated in those studies were more likely visiting subspecialty clinics rather than generalist clinics. Therefore, the diagnostic performance may not be the same when the test is applied in generalist clinics, where screening is commonly performed. Fifth, we emphasize that the advantage and usefulness of the UAC are purely under consideration as a screening test for microalbuminuria in patients with DM. A positive result on a screening test should be confirmed by timed urine collection measurement or by several additional first-void specimens collected sequentially.5,12 Other issues that deserve further study include monitoring patients over time; providing therapy in those with confirmed diagnoses; evaluating clinical outcomes, such as macroalbuminuria or dialysis; and screening the population without DM.
Our meta-analysis showed that the UAC and the ACR both yield high sensitivity and specificity for the detection of microalbuminuria in patients with DM. Because the diagnostic performance of the UAC is comparable to that of the ACR, our findings indicate that the UAC of random urine samples may become the screening tool of choice for the population with DM, considering the rising incidence of DM and the constrained health care resources in many countries.
Accepted for Publication: December 21, 2013.
Corresponding Authors: Yu-Kang Tu, DDS, PhD, and Kuo-Liong Chien, MD, PhD, Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University, 17 Hsu Chow Rd, Taipei, Taiwan 10055 (firstname.lastname@example.org and email@example.com).
Published Online: May 5, 2014. doi:10.1001/jamainternmed.2014.1363.
Author Contributions: Drs Tu and Chien had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: H.-Y. Wu, Chiang, Hung, K.-D. Wu, Tu, Chien.
Acquisition, analysis, or interpretation of data: H-Y. Wu, Peng, Chiang, Huang, Hung, Chien.
Drafting of the manuscript: H.-Y. Wu, Peng, Chiang, Tu.
Critical revision of the manuscript for important intellectual content: Huang, Hung, K.-D. Wu, Tu, Chien.
Statistical analysis: H.-Y. Wu, Peng, Chiang, Tu, Chien.
Administrative, technical, or material support: Huang, Hung, Tu, Chien.
Study supervision: Hung, K.-D. Wu, Tu, Chien.
Conflict of Interest Disclosures: None reported.