Prevalence of Bacterial Meningitis Among Febrile Infants Aged 29-60 Days With Positive Urinalysis Results

Key Points Question Are well-appearing febrile infants 29 to 60 days of age with positive urinalysis results at increased risk for bacterial meningitis? Findings In a systematic review and meta-analysis of 17 distinct international data sets (25 374 infants), the pooled prevalence of bacterial meningitis among infants with positive urinalysis results ranged from 0.25% (by cerebrospinal fluid [CSF] cultures or clinical definition) to 0.44% (by CSF cultures only). Prevalence estimates were not higher compared with infants who had negative urinalysis results (0.28% by CSF cultures or clinical definition and 0.50% by CSF cultures only). Meaning These findings suggest that, contrary to all published guidelines, invasive CSF testing in well-appearing febrile infants in the second month of life based on a positive urinalysis result alone is not supported by differential risk ratios.


Introduction
Fever among infants in the first months of life remains among the most common problems in pediatric health care. 1 These infants are at increased risk for potentially life-threatening serious bacterial infections (SBIs), specifically urinary tract infections (UTIs), bacteremia, and bacterial meningitis. 2 Approximately 10% of febrile infants 60 days or younger have underlying UTIs, 2 which presents a theoretical risk of hematogenous spread to the meninges. Consequently, infants with UTIs have historically been considered at increased risk for bacterial meningitis. To avoid missing 1 case of bacterial meningitis, nearly 400 infants will routinely undergo invasive cerebrospinal fluid (CSF) testing by lumbar puncture (LP), hospitalization, and broad-spectrum antibiotic therapy. 3 Failure to detect concomitant meningitis among infants with UTIs is associated with serious sequalae. Modern urinalyses accurately predict UTIs among young infants. 4 A presumptive diagnosis of UTI relies entirely on urinalysis results at initial evaluation before urine culture results are available.
Given the changing epidemiology of SBIs 17 and risks that decrease with infant age, 2,18 the necessity of LP for infants older than 28 days with a presumptive UTI has been questioned for decades. 19,20 Previous studies 21,22 suggest that a presumptive UTI is not associated with an increased risk of bacterial meningitis among well-appearing infants older than 28 days and that urinalysis results should not alter decisions regarding CSF testing. However, given the low overall prevalence of bacterial meningitis in this age group (approximately 0.4%), 18 no single study has been powered to determine the true risk of meningitis among well infants with a positive urinalysis result. The objectives of this study were to estimate the prevalence of bacterial meningitis among wellappearing febrile infants 29 to 60 days of age with positive urinalysis results and to compare this prevalence with that of infants with negative urinalysis results to inform whether routine LP is required.
The search strategy was conceptualized by all study authors with the assistance of a medical librarian guided by published Medical Subject Heading terms and keywords. 25 Broadly, the search combined the terms fever AND (urinary tract infection OR lumbar puncture OR meningitis) AND infant (eFigure 1 in the Supplement). Additional studies were identified through searching references of qualifying studies and systematic reviews. We included both prospective and retrospective studies with primary data. Narrative reviews, case reports, editorials, and guidelines were excluded.

Study Selection and Definitions
Studies were eligible if they reported on previously healthy, full-term (Ն37 weeks' gestation), wellappearing (documented by unstructured physician assessment or validated observation score 26 ) infants 29 to 60 days of age and evaluated for fever (documented rectal temperature of Ն38°C).
Results of urinalysis by microscopy or dipstick were eligible for inclusion. 27 A positive urinalysis result was defined as any finding of leukocyte esterase or nitrites, a white blood cell count of 10/μL (0.01 × 10 9 /L) or higher on an uncentrifuged specimen (or Ն5/μL [0.01 × 10 9 /L] per high-power field on a centrifuged specimen), or a positive Gram stain result. 28 Studies enrolling only infants with proven infections (viral or bacterial) or abnormal laboratory test results (eg, only infants with a positive urinalysis result) were excluded.
We contacted authors for patient-level data from studies that reported aggregate data only, those that reported UTI prevalence rather than urinalysis data specifically, and those enrolling with broader inclusion criteria. We included only studies from which we could ascertain both urinalysis results and meningitis status and only if meningitis status was determined by CSF testing or clinical follow-up when CSF was not obtained. We excluded studies in which we could not ascertain the number of infants with positive urinalysis results who underwent LP or whether meningitis status was based solely on CSF pleocytosis. To avoid double counting individual infants, publications that originated from the same source population were consolidated and analyzed as a single data set and cited by the most recent publication.
The primary outcome measure was the prevalence of definite meningitis among infants with positive urinalysis results, proven by CSF culture yielding pathogenic bacteria. The secondary outcome measure was the prevalence of meningitis among infants with positive urinalysis results, using a pragmatic clinical definition: a positive CSF culture result, bacteremia with CSF pleocytosis, or a suggestive history at clinical follow-up if CSF was not obtained.

Identification and Data Extraction
Titles and abstracts were screened independently by 3 study investigators (V.S., F.F.M., and G.D.M.), and potentially eligible studies were evaluated for inclusion by full-text review (B.B. and F.F.M. or V.S. and G.D.M.) using standardized criteria determined a priori, with discrepancies resolved by an additional study investigator. Interreviewer agreement was tested using the Light or Cohen κ. Data extraction was performed by 1 study investigator (B.B.) using standardized data extraction criteria, and 2 additional investigators (V.S. and G.D.M) reviewed data extraction to validate accuracy.
Extracted data included study characteristics (country, methodological design, method of recruitment, study years, publication year, study definitions, and follow-up duration when applicable) and participant characteristics (age, urinalysis results, and meningitis status).

Appraisal of Methodological Quality
Studies were assessed for methodological quality and risk of bias using the Newcastle-Ottawa Scale (NOS) checklist for nonrandomized cohort studies, 29 as recommended by the Cochrane Collaborative. 30 The NOS was adapted for this appraisal, and studies were considered at high risk for bias if they received less than 4 of 10 points (eFigure 2 in the Supplement). 31 Because the data of interest were often not the primary outcomes of the included studies, items on the scale were scored on the quality of the design relative to the outcomes of this analysis rather than that of the original study objective. Two study investigators (V.S. and G.D.M.) independently appraised all data sets for methods, and discrepancies were resolved by a third (B.B.).

Statistical Analysis
Meta-analyses were performed to assess the pooled prevalence of bacterial meningitis and to estimate odds ratios (ORs) comparing infants with positive urinalysis results and infants with negative urinalysis results. Meta-analyses were conducted using a random-effects model because substantial heterogeneity was anticipated based on differing study designs. Pooled proportions and ORs were estimated after being transformed with the logit function using generalized logistic regression models, which perform well for sparse data. 32 Results were summarized in forest plots. To assess the robustness of pooled estimates, sensitivity analyses were planned a priori: (1) excluding studies at high risk for bias (NOS score <4), (2) analyzing only prospectively collected data sets, and (3) considering only studies with clinical follow-up of 7 days or more and those with follow-up of 30 days or more. Heterogeneity was estimated using I 2 statistics and prediction intervals. 33 A continuity correction of 0.05 was added to studies with 0 events to allow inclusion in funnel plots. Pooled proportions were assessed for publication bias by graphical inspection of funnel plots and by using the Egger test for OR estimates because there were few events per study. All analyses were performed using the meta package in R software, version 3.5.3 (R Foundation for Statistical Computing). 34

Results
Search of the electronic databases yielded 3227 unique publications, with an additional 12 identified by searching references of relevant studies. After removing 588 duplicates, screening by title and abstract identified 134 studies for full-text review; 34 did not meet inclusion, and 100 required author contact for patient-level data. Following contact with primary authors, 52 additional studies were excluded, 36 because authors could not be reached or were unable to provide required data (description of excluded studies in eTable 1 in the Supplement). In total, 48 individual studies 1,4,11,12,14,15,17,21,22,26, were included for meta-analysis, with all data confirmed by the original authors for accuracy. Studies that originated from the same cohort were consolidated into a single data set for analysis. Ultimately, 17 distinct data sets were included in the analysis (Figure 1). Table 1, including years, study design, setting, follow-up duration, and number of eligible infants. Complete urinalysis data and meningitis status were available for 25 374 previously healthy, well-appearing full-term infants 29 to 60 days of age. All studies were published in peer-reviewed journals. Eight of 17 data sets enrolled infants in the US. 14,21,22,53,54,56,57,70 Nine data sets were prospectively collected, 12,14,35,58,61,64,67,70,72 and 1 study 53 used a pre/post intervention design in which only infants from the postintervention period were analyzed prospectively. Two additional data sets were retrospective analyses 22,54 of infants managed in the context of quality improvement initiatives. These 2 data sets were included among the sensitivity analysis of prospective studies because all relevant covariates were collected prospectively. Methodological quality of included data sets varied widely, with NOS scores ranging from 3 to 8, and 1 study 66 at high risk for bias (eTable 2 in the Supplement).

Descriptive characteristics of included studies are presented in
Among included data sets, meningitis status was determined by CSF testing for 12 735 infants and by CSF testing or clinical follow-up for 25 374 infants. For both the primary and secondary outcomes, the unweighted proportion of infants with bacterial meningitis was higher among infants with negative urinalysis results in 9 data sets, 14,21,22,35,53,54,58,66,70 higher among infants with positive urinalysis results in a single data set, 61 and 0 in both groups in 7 data sets. 12,56,57,62,64,67,72 The 7 data sets reporting 0 cases of meningitis in both groups did not contribute to the pooled OR, and several data sets had estimated ORs greater than 1 despite 0 events in the group with positive urinalysis results because of imbalanced sample sizes.    sensitivity analyses did not differ significantly, lending confidence to findings of the primary analyses ( Table 2).

Discussion
This systematic review and meta-analysis is the largest and most comprehensive study, to our knowledge, to evaluate the risk among well-appearing febrile infants older than 28 days with a positive urinalysis result. Accurate prevalence estimates are essential for practitioners to quantify the risk of concomitant bacterial meningitis and to inform clinical decision-making. The present analysis combined data from 17 unique data sets with more than 25 000 infants from geographically diverse populations. The results suggest that infants with a positive urinalysis result are at no higher risk for bacterial meningitis than infants with a negative urinalysis result. This finding is contrary to the dogma held for nearly 30 years. 6 Historically, all risk-stratification criteria, including very recently derived clinical decision rules, 5-14 categorize a positive urinalysis result as a high-risk feature, prompting invasive testing, broad-spectrum antibiotic exposure, and hospitalization. Although this practice has been controversial for decades, the very low overall prevalence of bacterial meningitis has meant that no single study could reliably answer this clinical question. [20][21][22]31,66 This large metaanalysis provides compelling evidence that decisions regarding LP for this subgroup of infants should not be guided by urinalysis results alone.
The evaluation of febrile young infants is invasive, anxiety-provoking for parents, and associated with iatrogenic risk and significant system-wide resource use. 75 Since 2016, improved care for febrile young infants has become the largest US-wide quality initiative ever endorsed by the American Academy of Pediatrics, including 124 independent hospitals across 38 US states. 15 Widely disseminated clinical pathways and electronic decision-support tools have been developed, which classify infants 29 to 60 days of age with a positive urinalysis result at increased risk for bacterial meningitis and recommend LP; conversely, infants with a negative urinalysis result are classified as low risk, and LP is not required. 16 Findings from this analysis are in contrast to these recommendations.

Positive urinalysis results
For the pooled odds ratio (OR) analysis (C), the arrow indicates that the upper confidence limit falls beyond the x-axis; diamond, the overall estimate from the meta-analysis and its confidence interval, with the center of the diamond representing the pooled estimate; and the bar below the diamond, the prediction interval.

Negative urinalysis results
For the pooled odds ratio (OR) analysis (C), the arrow indicates that the upper confidence limit falls beyond the x-axis; diamond, the overall estimate from the meta-analysis and its confidence interval, with the center of the diamond representing the pooled estimate; and the bar below the diamond, the prediction interval.

Strengths and Limitations
One strength of the current analysis is the calculation of pooled ORs rather than only prevalence estimates, 31 which must be compared with historical controls. 18 This analysis allows a direct comparison of prevalences and an estimation of odds among infants with positive and negative urinalysis results within included studies. Of note, the pooled ORs for the primary and secondary outcomes were below 1, supporting the conclusion that infants with positive urinalysis results are not at higher risk for bacterial meningitis. In fact, the prevalence of meningitis was higher among infants with positive urinalysis results in just a single medium-sized data set. 61 Moreover, another large data set 74 has since enrolled many additional infants with an even lower relative risk among those with positive urinalysis results than was available at the time of analysis. Furthermore, 3 data sets reported 0 cases of meningitis in the urinalysis-positive group and 1 case or more in the urinalysis-negative group but generated ORs greater than 1. 35,66,70 Paradoxical ORs such as these are possible when there is an imbalance in sample sizes, and the smaller sample has 0 events. Inclusion of these studies is known to bias pooled ORs toward the null. 76 Taken together, it is likely that pooled OR point estimates reported are, if anything, an overestimate.
A novel contribution of this analysis was the purposeful exclusion of studies before the year 2000 to account for changing SBI epidemiology. 17,18 The current analysis differs importantly from a recent small meta-analysis, 31 which consisted primarily of retrospective studies using a cultureproven UTI case definition. Similar to estimates reported here, Nugent et al 31 reported 11 cases of bacterial meningitis among 3868 infants 29 to 90 days of age with an abnormal urinalysis result or culture-confirmed UTI who underwent LP (pooled prevalence, 0.25%). However, their analysis did not estimate the prevalence among infants with negative urinalysis results; thus, no direct comparison could be made or OR calculated. In addition, results of this analysis were driven largely by a single study 77 that contributed 1609 infants selected on the basis of a culture-confirmed UTI not urinalysis results. Of importance, clinical decisions about LP and hospitalization rely on initial urinalysis results not urine culture results. A urinalysis is highly sensitive (0.94; 95% CI, 0.91-0.97) and specific (0.91; 95% CI, 0.90-0.91) for UTIs in febrile young infants. 4 A urinalysis is also the most universally used diagnostic test for risk stratification, 2,48 and an abnormal urinalysis result is among the most frequent reasons that infants do not meet low-risk criteria. 57 There are several additional strengths of the current analysis. The study population does not include infants from studies that selected infants on the basis of a clear focus of infection and thus addresses the most common and challenging clinical conundrum when evaluating well-appearing febrile young infants. Authors were contacted to obtain patient-level data and to accurately consolidate overlapping studies to prevent repeat counting of individual infants. Sensitivity analyses were selected a priori and increase the confidence in the results of the main analysis. Measures were Abbreviation: OR, odds ratio.
a Nine data sets were prospectively collected, 1 data set had prospective postintervention data, and 2 retrospective data sets were analyzed with prospective studies because of prospective collection of all relevant covariates. taken to assess the possibility of publication bias, which does not appear to have influenced the results to any significant degree.
This analysis also has limitations. It is possible that the pooled prevalence of meningitis among infants with negative urinalysis results reported is falsely elevated because infants with a normal urinalysis result who did not undergo CSF testing would not be included in the denominator.
However, studies with clinical follow-up of at least 7 and 30 days would capture these infants, and sensitivity analyses reveal OR point estimates that are still not higher among infants with positive urinalysis results. The secondary outcome used a pragmatic clinical definition of meningitis, and neither a threshold for CSF pleocytosis [78][79][80] nor a definition of history suggestive of meningitis at follow-up was prespecified; rather, the outcomes were reported as classified by the primary study authors. Fifty-two studies were excluded (several with overlapping data sets), including 36 for which patient-level data were not available. Bias introduced by their exclusion is theoretically possible; however, it is unlikely that these infants were systematically different from those analyzed. In all studies, the decision to perform an LP was at the discretion of the treating physician; however, the sensitivity analysis limited to studies with follow-up of 30 days or more with the lowest pooled ORs mitigates the risk of missing cases of bacterial meningitis among infants without CSF testing or a short clinical follow-up. Although every attempt has been made to not double count infants, the possibility cannot be completely excluded given that several included studies were large national or multinational studies, although most data sets did not overlap temporally or geographically. Only studies published in English or French were included, and most were conducted in emergency departments; therefore, estimates may not be generalizable to other settings (ie, ambulatory clinics or unrepresented countries). In addition, pooled prevalence estimates are associated with the urinalysis status in isolation, and the risk when other diagnostic biomarkers are also within normal limits was not assessed (ie, C-reactive protein and procalcitonin). Studies with 0 events provide challenges in estimation with traditional meta-analytic methods; as such, generalized linear mixedeffects models were used to compensate. High heterogeneity across studies was expected, and qualitatively this was true based on study methods. Despite this, for some outcomes, the I 2 statistics could not be accurately estimated because of the small (and 0) event rates in many studies. 76 For these estimates, the prediction intervals must be relied on to provide a relevant alternative measure of the heterogeneity and are reported for all analyses.

Conclusions
Invasive CSF testing, hospitalization, and empirical antibiotic treatment of well-appearing febrile infants older than 28 days with a positive urinalysis result have been predicated for decades on the assumption of an increased risk of bacterial meningitis. Despite fever in young infants being a common clinical problem, no single study to date has been adequately large to reliably determine the true relative risk among infants with positive urinalysis results. Findings from this large metaanalysis suggest that well-appearing febrile infants 29 to 60 days of age with a positive urinalysis result are not at an elevated risk for bacterial meningitis compared with infants with negative urinalysis results. Overall, these results suggest that the rate of concomitant bacterial meningitis in this population is low, and LP should not be undertaken on the basis of a positive urinalysis result alone.