Objective. To review systematically and to summarize the existing literature regarding performance of rapid diagnostic tests for urinary tract infection (UTI) in children.
Design. Systematic review and meta-analysis.
Methods. Published articles reporting the performance of urine dipstick tests (leukocyte esterase [LE] and/or nitrite), Gram stain, or microscopic analysis of spun or unspun urine in the diagnosis of UTI in children ≤12 years of age. Articles were identified through a comprehensive MEDLINE search, and those articles meeting a priori inclusion criteria were selected. Eligibility criteria included the use of urine culture as the reference standard, independent comparison of urine culture with the results of one of the screening tests, definition of positive screening test results provided, only pediatric patients included or evaluable separately, and both gold standard and screening test performed on all patients. For each test, heterogeneity of reported sensitivity and specificity of all studies was determined. The subgroups of studies with similar definitions of UTI and age of study subjects were analyzed separately to account for some of the differences in reported results. When significant unexplained heterogeneity among studies precluded simple combining of results, a summary receiver–operator characteristic curve was fitted for each screening test, from which pooled estimates of true-positive rate (TPR; ie, sensitivity) and false-positive rate (FPR; 1-specificity) were calculated.
Primary Results. A total of 1489 titles were identified by the MEDLINE search; 26 articles met all criteria for inclusion. There was significant heterogeneity among studies for nearly all tests for both TPR and FPR, which was explained only partially by the stringency of the definition of UTI or age of subjects studied. Based on the pooled estimates, the presence of any bacteria on Gram stain on an uncentrifuged urine specimen had the best combination of sensitivity (0.93) and FPR (0.05). Urine dipstick tests performed nearly as well, with a sensitivity of 0.88 for the the presence of either LE or nitrite and an FPR of 0.04 for the presence of both LE and nitrite. Pyuria had lower TPR and higher FPR: for presence of >5 white blood cells/high-power field in a centrifuged urine sample, the TPR was 0.67 and the FPR was 0.21, whereas for >10 white blood cells per mm3 in uncentrifuged urine, the TPR was 0.77 and the FPR was 0.11.
Conclusions. Both Gram stain and dipstick analysis for nitrite and LE perform similarly in detecting UTI in children and are superior to microscopic analysis for pyuria.
- UTI =
- urinary tract infection •
- LE =
- leukocyte esterase •
- WBC =
- white blood cells •
- hpf =
- high-power field •
- TPR =
- true-positive rate •
- FPR =
- false-positive rate •
- ROC =
- receiver–operator characteristic •
- LR =
- likelihood ratio
Urinary tract infection (UTI) is recognized increasingly as a common cause of fever in young children.1–3 However, clinical findings indicative of UTI in this group are often subtle and nonspecific, with fever often the only finding. On the one hand, identification of very low risk children would be desirable to reduce the cost of unnecessary urine culture. Conversely, clinicians would like to be able to identify those children with a sufficiently high likelihood of UTI to begin presumptive treatment while waiting for the results of the urine culture.
Several rapid screening tests are used commonly to make a presumptive diagnosis of UTI, including dipstick biochemical analysis of urine for nitrites or leukocyte esterase (LE), as well as microscopic examination of urine for formed elements including white blood cells (WBC) or bacteria. Numerous studies have been published concerning the usefulness of these tests in diagnosing UTI.4 ,5 Many of these studies have been performed in adult patients, in whom the diagnostic utility of the tests may differ.5 ,6 Moreover, even among pediatric studies the results differ frequently.5 ,7 Therefore, the clinician is left with little guidance regarding the optimal choice of diagnostic tests for diagnosing UTI in children. In this situation, meta-analysis, the structured review and statistical combination of the results of existing research, may be useful for identifying sources of variability among studies and for providing an overall estimate of diagnostic accuracy.
Hurlbut et al4 published a meta-analysis of studies of rapid dipstick tests in adult patients. Regarding pediatric studies, although informal reviews of diagnostic tests exist,5 ,6 we are aware of no previous attempt to synthesize the existing data in a structured, critical manner. The purpose of this study was to perform a systematic review of the published literature to identify studies of rapid diagnostic tests for UTI in children. We then used the techniques of meta-analysis to derive overall estimates of test characteristics (sensitivity and specificity) from the existing studies where possible and to investigate the sources of discrepancies among studies.8 ,9
Identification of Relevant Literature
We used the National Library of Medicine's PUBMED system to conduct a MEDLINE search for the years 1966 to 1998 for articles published in English, concerning the use of rapid diagnostic tests for UTI in children. The search strategy used was:
(urine [mh] or urinalysis [mh] or pyuria [mh] or reagent strips [mh] or bacteriuria [mh])
and ((urinary tract infections [mh] or pyelonephritis [mh]) not schistosomiasis [mh])
and (english [la])
and (infant, newborn [mh] or infant [mh] or child, preschool [mh] or child [mh]), where [mh] designates a Medical Subjects Heading term. The age qualifiers chosen identify entries dealing with children from birth through 12 years of age. In addition to the MEDLINE search, the reference lists of those studies selected for inclusion and of review articles known to the investigators were searched for other relevant references. The investigators also hand-searched their files and contacted experts in the field to inquire about other studies not identified by the MEDLINE search.
The titles and abstracts were reviewed by both authors for possible relevance, with discrepancies resolved by discussion and consensus. Those articles deemed relevant were retrieved and reviewed. Inclusion criteria included:
• primary data (not review of existing studies)
• use of one or more of the following rapid tests:
urine dipstick (LE, nitrite, or both) microscopic analysis of centrifuged urine sample for WBC, reported as number of WBC per high powered field (WBC/hpf) microscopic analysis of uncentrifuged urine sample for WBC, reported as number of WBC per mm3
Gram stain of uncentrifuged urine
enhanced urinalysis10 (cell count and Gram stain on uncentrifuged urine)
quantitative or semiquantitative urine culture as reference standard
both screening test and urine culture performed on all subjects
results of screening test not included in definition of UTI
data presented in format allowing crosstabulation of results of screening test and reference standard
if both children and adults were included in study, results for children evaluable separately
data not included in another published report (in cases in which multiple publications used the same subjects, we included the one publication with the greatest number of subjects).
Each included article then was read in detail, and the results were abstracted. For each article, information was abstracted on the age of subjects, colony count used to define UTI, whether a general or a special population (eg, urology clinic, only boys, etc) was used, and any possible methodologic concerns (eg, nonconsecutive patient enrollment or nonblinded assessment).
For each test, a 2 × 2 table was abstracted as follows:
In the case of a test with multiple possible thresholds (eg, for microscopic analysis of uncentrifuged urine, ≥10 WBC/mm3, ≥50 WBC/mm3, and ≥100 WBC/mm3), separate tables were abstracted for each cutoff described in the study. Separate tables also were constructed if test results for different age subgroups or with different definitions of UTI were available in a single study.
Combining Results of Studies
For each test and at each cutoff of positivity, the true-positive rate (TPR) and false-positive rate (FPR; 1-specificity) were determined individually from each included study. In meta-analysis, the goal is to combine results from different studies to obtain a more precise estimate than is possible from any of the individual studies. The simplest means of combing results is simply to pool raw data from all the included studies, yielding a weighted average for TPR and FPR.9 However, an important methodologic consideration is whether differences in results among studies of a given test are deemed to represent simply random differences attributable to sampling variation or whether such differences arise because of underlying heterogeneity in the studies.11 In the presence of residual unexplained heterogeneity of results among studies, simple summarization violates the underlying assumption that differences among studies are attributable simply to random variation and may produce misleading or inaccurate results.12 Thus, for each diagnostic test of interest, separate comparisons of the TPRs and FPRs of all studies were performed using the Pearson χ2test.9
To explore the possible sources of any heterogeneity among studies (as indicated by P < .05 for the χ2 test), the studies were divided into subgroups. The first subgroup was defined based on the definition of UTI used. Studies that defined UTI based on a colony count of ≥105 colonies/mL for clean catch specimens, ≥104 colonies/mL for catheterized specimens, or only specimens obtained by suprapubic bladder aspiration were classified as using a more stringent definition of UTI; those using lower thresholds for positive culture results were classified as less stringent. If the criterion for positive culture was not specified, the study was included in the less stringent subgroup. In some cases, a single study used two alternative cutoffs for positive culture results, allowing separate analyses for both definitions. In these instances, the study was included twice, once in each subgroup. (These studies are indicated in Table 1.) To examine whether the TPR of a given screening test depends on the definition of UTI used, a pooled estimate of TPR for each subgroup of studies examining that screening test was calculated. These pooled TPRs then were compared with a χ2 test. Similarly, pooled FPRs for each subgroup were calculated and compared.
A second subgroup analysis was based on the age of the patients included: all ages of children or only children <2 years of age. When the age of patients was not reported, it was assumed that all ages were included. Again, if a single study provided data allowing separate analyses of all children and younger children, it was included in both subgroups. For a given diagnostic test, subgroup analyses were performed if at least three studies were included in each subgroup.
It has been noted that, in the case of diagnostic tests, differences among studies are characterized frequently by a trade-off between sensitivity and specificity.8 ,9 ,13 When the results of individual studies are plotted graphically, they assume the shape of a receiver–operator characteristic (ROC) curve. One possible source of heterogeneity leading to this trade-off is the implicit use of different threshold values for positive results in different studies. This may occur even when the same cutoff ostensibly is used. For instance, in two studies of LE, even though both may consider any result ≥1+ to be positive, differences in the interpretation of the color on the stick may lead to inherent variation in the positive results cutoff. Differences in the patient populations with regard to the spectrum or prevalence of disease also may lead to such an observed trade-off.14 To look for evidence of such a trade-off, the TPR and FPR of each study were plotted against each other in a scatter plot, and the Spearman correlation coefficient was calculated. A summary ROC curve of all the studies then was created.13 The TPR and FPR were converted to their logits, and the sum and difference of the logits were calculated. Equally weighted least squares linear regression with the difference as the dependent variable and sum as the independent variable was performed. The resulting equation represents the logit form of the summary ROC curve from which a pooled estimate of TPR and FPR can be obtained. We computed a weighted average of the FPR from all included studies by simply combining the raw data for FPRs and TPRs, following the suggestion of Moses et al.13 From this pooled estimate of the FPR, the TPR was calculated from the summary ROC curve equation. Positive and negative likelihood ratios (LRs) were derived from the summary TPRs and FPRs.
All statistical analyses were performed using Stata version 5.0 (Stata Corp, College Park, TX). Unless otherwise noted, P < .05 was considered to indicate statistical significance.
The MEDLINE search yielded 1489 titles. Of these, 57 articles were reviewed in detail, and 26 met all inclusion criteria (Table 1)7 15–39. No relevant studies were identified from other sources. The range of TPRs and FPRs for each of the tests reported by the studies is shown in Table 2. If a given published study provided results separately for different tests, these were included as independent entries.
As indicated in Table 2, most of the tests showed significant heterogeneity among studies in both TPR and FPR. Figure 1A illustrates the performance of the tests when grouped by UTI definition used in the study (more stringent or less stringent) for those tests with enough studies with different definitions of UTI to allow subgroup analyses. Studies with a more stringent definition of UTI reported higher sensitivity (TPR) for LE alone and nitrite alone, but not for microscopic urinalysis with ≥5 WBC/hpf. Similarly, the FPR also differed by definition of UTI for LE alone, nitrite alone, and the combination of any nitrite or LE on dipstick test. However, except for the TPR for the LE test, the use of different definitions did not explain all the heterogeneity of results among studies, shown by the wide range of values and statistically significant heterogeneity within each subgroup (more stringent vs less stringent definition of UTI) depicted in Fig 1A. Thus, differences in the gold standard used by different studies explain some, but not all, of the observed heterogeneity in results.
Analysis by age of patients is shown in Fig 1B. Sensitivity results for the Gram stain and combined dipstick tests were not different among age groups, but the TPR of the presence of ≥10 WBC/mm3 was significantly higher in the studies including only children <2 years of age. The pooled FPR was significantly lower for all three tests in studies of children <2 years of age. With respect to both TPR and FPR, all three diagnostic tests demonstrated significant heterogeneity within age groups. As with UTI definition, differences in the age of the study subjects do not explain adequately the variation in reported results.
Because the variability among studies could not be explained completely by differences in the age of patients studied or by differences in the definition of the reference standard, summary ROC curves were constructed for each of the diagnostic tests. The summary ROC curves for the four dipstick tests are illustrated in Fig 2. Unlike conventional ROC curves, which demonstrate the effect of altering the cutoff for a positive test, these ROC curves show the trade-off between TPRs and FPRs among different studies of the same test with the same definition of a positive test result. The ROC analysis also allowed us to explore the possible effect of disease prevalence on the performance of the screening test. When the proportion of patients with positive culture results was included as a covariate in the ROC regression equations for nitrite, LE, Gram stain, and cell count, the coefficients for the prevalence covariate were not statistically significant (P > .3 in all cases).
Because the gold standard used to define UTI in various studies did seem to contribute to the observed variability of results, we used an ROC curve that excluded those few studies using low colony counts for the definition of UTI to derive a summary point estimate of TPR and FPR for each screening test. Table 2 shows the pooled estimate for the FPR for each test, and the summary estimate for the TPR derived from the equation for the ROC curve at that FPR. For microscopic analysis of centrifuged urine (≥5 WBC/hpf), there was no apparent relationship between TPR and FPR (Spearman R = −0.17); therefore, a summary ROC curve could not be constructed. In this case, a simple pooled estimate of TPR was calculated by combining the results of the studies into a single table as for FPR.
The test with the best combination of sensitivity and specificity was Gram stain (positive LR: 18.5; negative LR: 0.07). Urine dipstick tests performed nearly as well; the presence of both LE and nitrites had a positive LR of 12.6, whereas the absence of both LE and nitrite had a negative LR of 0.13
Only two studies were found that examined the combination of Gram stain and cell count on uncentrifuged urine, referred to as enhanced urinalysis.7 ,39 Two other studies used the combination of cell count and bacteria, but on an unstained specimen;16 ,40 of these, one included only the disjunctive combination (≥10 WBC/mm3 or any bacteria on Gram stain). For the combination of ≥10 WBC/mm3 and any bacteria, the TPR was significantly lower and the FPR was significantly higher in the study that used an unstained specimen to detect bacteriuria than in the two studies incorporating Gram stain. The pooled FPR for the latter two studies was 0.01 and the TPR was 0.85, giving a positive LR of 85. For the combination of ≥10 WBC/mm3 or any bacteria, the TPR was similar in all four studies, but the FPR was significantly higher for the two studies that used unstained specimens. Therefore, the pooled estimates were derived from the two studies using Gram stain, yielding a summary TPR of 0.95, FPR of 0.11, and negative LR of 0.06.
This systematic review of the literature identified a number of studies reporting on the performance of various screening tests to detect UTI in children. Even among the most methodologically sound papers, there is substantial variability in the reported sensitivity and specificity. Using meta-analytic methods, we were able to examine some of the reasons for this variability and to summarize the results of different studies to guide clinicians in interpreting the available tests.
For predicting a positive urine culture, the presence of any bacteria on a Gram-stained urine specimen offers the best combination of sensitivity and specificity, with the highest sensitivity of the tests evaluated. However, the dipstick test performs nearly as well, with a slightly lower sensitivity for the presence of any nitrite or LE and a slightly better FPR for the presence of both LE and nitrite. The technically more demanding microscopic analysis of either centrifuged or uncentrifuged urine offers no advantage over the dipstick test or Gram stain, with lower sensitivity and specificity. The enhanced urinalysis, a combination of Gram stain and cell count performed in a counting chamber using an uncentrifuged urine specimen, also seems to provide an excellent combination of sensitivity and specificity. However, this is based on only two studies, both of which were performed in children <2 years of age. Moreover, in these two studies the accuracy of the cell count component alone was substantially better than for the other studies using counting chamber cell count on uncentrifuged urine. Therefore, the results of these two studies should be interpreted with caution.
Some of the differences in results among studies were explained by identifiable differences in the methods used by the individual studies. For example, when we analyzed studies separately according to the definition of UTI used, we found that studies using a more stringent reference standard generally had different results than those using a lower colony count to define a positive culture. The use of more stringent criteria probably eliminates some contaminated cultures that are less likely to have pyuria or to contain pathogenic species that produce nitrites. Both of these factors would reduce the number of false-negative results when a more stringent gold standard is applied, which may explain the improved sensitivity of the LE and nitrite tests in this group of studies. However, using a higher colony count cutoff also will lead to misclassifying some patients who actually have UTI (with pyuria and nitrites) as not having the disease. Such misclassification can explain the higher FPRs for nitrite and LE alone among studies with more stringent definitions. Because the choice of gold standard affects the reported accuracy of the tests, we chose to calculate our summary estimates of sensitivity and specificity from only those studies using the more stringent colony counts used to define UTI; this corresponds with the definition of UTI that is accepted most widely.
The age of the patients was also a factor in the performance of the screening tests. There are several potential explanations for the observed differences. More frequent voiding in non-toilet-trained infants leads to decreased time for production of nitrites by nitrate-reducing organisms, while such infants might tend to have a less vigorous inflammatory response to infection, which may explain the lower sensitivity of the dipstick test for LE or nitrites. Differences in the usual means of specimen collection (clean catch for older children vs urine bag or urethral catheterization for infants) that would lead to differences in the potential for contamination of the specimen with perineal cells and bacteria may contribute to the lower FPR observed in the younger children. Similarly, proportionately fewer contaminated specimens in these younger children would tend to exclude those with a false-positive culture and no pyuria, and lead to a higher TPR for tests that detect small amounts of pyuria, such as the microscopic examination of urine using a counting chamber.
Even accounting for differences in the gold standard and age of the study populations, there remained significant differences among studies in the reported sensitivity and specificity of most of the screening tests. We were able to generate summary ROC curves for most of the tests, demonstrating that different studies of a given screening test using the same threshold for a positive result show the same trade-off between true- and false-positive results as would be expected if the various studies actually used different thresholds. Other factors such as differences in the techniques of the test performance, characteristics of the patient populations, and differences in disease prevalence can cause such a relationship between TPRs and FPRs. We did not find that UTI prevalence had an effect on the results. However, it should be noted that almost all the studies used some form of convenience sampling; thus, the proportion of positive cultures in the study sample that we used in our analysis is not necessarily a reflection of the disease prevalence in the population. Although we were not able to examine this and the other factors above directly because of incomplete information in the studies, by using the ROC curves, it was possible to obtain summary estimates of the TPR and FPR, adjusting for at least some of these unmeasured differences among studies.
Although the resulting summary ROC curves allow for a more meaningful comparison of different tests across a range of implicit thresholds which will be useful to the clinician, a single point on the curve must be chosen to select a value for sensitivity and specificity to be used in decision making. We used the simple pooled estimate of FPR from all studies to define the operating point of interest on the curve and derived the summary TPR from the ROC curve at that point. For many of the diagnostic tests we evaluated, such as nitrite alone or Gram stain, although the differences in FPRs among studies were statistically significant, the range reported was actually quite small and the observed variability is likely to have little clinical importance. In such cases, the use of a simple weighted average for the summary FPR estimate is justifiable despite the heterogeneity. For the other tests (eg, LE alone), although the range of FPR was substantial, the summary ROC curve proved quite flat over the range of interest. The choice of FPR in this case is unlikely to have an important impact on the calculated value of TPR. Nevertheless, the pooled FPR should be interpreted with some caution.
Several other potential limitations to this study are common to many meta-analyses. Although we used a comprehensive search strategy to identify relevant articles, some publications may have been missed, particularly those published before 1966. Moreover, only those studies actually published could be identified. Because studies with negative results (eg, those showing poor diagnostic performance) are less likely to be published, a phenomenon known as publication bias,12the resulting meta-analysis may overestimate the accuracy of the tests in question. A bias may be introduced by limiting searches to English publications, although the importance of this bias is unclear.12
Based on this systematic review of the existing literature, we conclude that the urine dipstick test and Gram stain perform similarly in detecting UTI in children, with high sensitivity and a low FPR. The enhanced urinalysis may offer a better combination of test performance characteristics but has not yet been well studied. All these tests have better sensitivity and specificity than the presence of pyuria in either centrifuged or uncentrifuged specimen. Indeed, the TPRs and FPRs of the presence of >5 WBC/hpf in a centrifuged urine specimen (standard urinalysis) is sufficiently poor that it cannot be recommended for making a presumptive diagnosis of UTI. The clinical usefulness of any of these tests will depend on the clinical setting in which they are applied, and, specifically, on the prevalence of the disease, which will affect the predictive value of a positive or negative result. Moreover, the choice of a testing strategy must be based on balancing a number of other considerations such as availability and cost of the tests themselves,7 as well as the relative costs of failing to diagnose a UTI in a child with a false-negative test result versus unnecessary treatment of an otherwise healthy child with a false-positive test result. Currently, we are undertaking a decision analysis to define further the optimal testing strategy for diagnosing UTI in children.
This work was supported by Grant MCJ-420648 from the Maternal and Child Health Bureau (Title V, Social Security Act), Health Resources and Services Administration, Department of Health and Human Services.
- Received October 12, 1998.
- Accepted May 17, 1999.
Reprint requests to (M.H.G.) the Division of Emergency Medicine, A. I. duPont Hospital for Children, 1600 Rockland Rd, Wilmington, DE 19899. E-mail:
- ↵Shaw KN, Gorelick MG, McGowan KL, McDaniel Yakscoe M, Schwartz JS. Prevalence of urinary tract infection in febrile young children in the emergency department. Pediatrics.1998;102(2). URL: http://www.pediactrics.org/cgi/content/full/102/2/e16
- ↵Shaw KN, McGowan KL, Gorelick MH, Schwartz JS. Screening for urinary tract infection in infants in the emergency department: which test is best? Pediatrics 1998;101(6). URL: http://www.pediatrics.org/cgi/content/full/101/6/e1
- Hoberman A,
- Wald ER,
- Penchasky L,
- Reynolds EA,
- Young S
- ↵Shapiro DE. Issues in combining independent estimates of the sensitivity and specificity of a diagnostic test. Acad Radiol.1995;2:37–47. Supplement
- ↵Pettitti DB. Meta-analysis, Decision Analysis, and Cost-effectiveness Analysis: Methods for Quantitative Synthesis in Medicine. New York, NY: Oxford University Press; 1994:52–53, 90–95
- Corman LI,
- Foshee WS,
- Kotchmar GS,
- Harbison RW
- Boreland PC,
- Stoker M
- Powell HR,
- McCredie DA,
- Ritchie MA
- Crain EF,
- Gershel JC
- Lejeune B,
- Baron R,
- Guillois B,
- Mayeux D
- Woodward MN,
- Griffiths DM
- Molyneux EM,
- Robson WJ
- Lockhart GR,
- Lewander WJ,
- Cimini DM,
- Josephson SL,
- Linakis JG
- Copyright © 1999 American Academy of Pediatrics