Length of stay, cost, and mortality of healthcare-acquired bloodstream infections in children and neonates: A systematic review and meta-analysis

Abstract Objective: To estimate the attributable mortality, length of stay (LOS), and healthcare cost of pediatric and neonatal healthcare-acquired bloodstream infections (HA-BSIs). Design: A systematic review and meta-analysis. Methods: A systematic search (January 2000–September 2018) was conducted in PubMed, Cochrane, and CINAHL databases. Reference lists of selected articles were screened to identify additional studies. Case–control or cohort studies were eligible for inclusion when full text was available in English and data for at least 1 of the following criteria were provided: attributable or excess LOS, healthcare cost, or mortality rate due to HA-BSI. Study quality was evaluated using the Critical Appraisal Skills Programme Tool (CASP). Study selection and quality assessment were conducted by 2 independent researchers, and a third researcher was consulted to resolve any disagreements. Fixed- or random-effect models, as appropriate, were used to synthesize data. Heterogeneity and publication bias were evaluated. Results: In total, 21 studies were included in the systematic review and 13 studies were included in the meta-analysis. Attributable mean LOS ranged between 4 and 27.8 days; healthcare cost ranged between $1,642.16 and $160,804 (2019 USD) per patient with HA-BSI; and mortality rate ranged between 1.43% and 24%. The pooled mean attributable hospital LOS was 16.91 days (95% confidence interval [CI], 13.70–20.11) and the pooled attributable mortality rate was 8% (95% CI, 6–9). A meta-analysis was not conducted for cost due to lack of eligible studies. Conclusions: Pediatric HA-BSIs have a significant impact on mortality, LOS, and healthcare cost, further highlighting the need for implementation of HA-BSI prevention strategies.

Healthcare-associated infections (HAIs) have a significant impact on mortality, length of stay (LOS), and healthcare cost worldwide. [1][2][3] In the United States, ∼2 million patients suffer from HAIs annually, nearly 90,000 are estimated to die due to HAIs, and the overall annual direct cost of HAIs to hospitals ranges from $28 billion to $45 billion. 4 The 2008 Annual Epidemiological Report on Communicable Diseases in Europe of the European Centre for Disease Prevention and Control (ECDC) declared that HAIs caused 16 million extra days of hospital stay and 37,000 attributable deaths annually, while the associated annual cost reached 7 billion euros (2019 USD, 7.75 billion). 5 Healthcare-acquired bloodstream infections (HA-BSIs) are the most common HAIs in critically ill pediatric patients of all age groups because central venous catheters (CVCs) are commonly used in these patients. [6][7][8][9] In the United States, HA-BSIs are associated with the highest number of preventable deaths among HAIs, as well as the highest costs, ranging from $960 million to $18.2 billion annually. 10 Although strong evidence indicates that most HA-BSIs are preventable, 9,[11][12][13][14][15][16] pediatric and neonatal HA-BSI rates remain far above zero in many countries worldwide, highlighting the need to initiate further national and targeted prevention strategies. [17][18][19] Published data regarding pediatric and neonatal HA-BSI outcomes vary significantly, depending on the country of origin, year of publication, and study design. 20 Accurate estimates of LOS, cost, and mortality attributed to HA-BSIs are essential for developing cost-effective infection prevention and infection control measures. 21 In the present study, we systematically reviewed the available evidence and estimate attributable LOS, healthcare costs, and mortality rates for pediatric and neonatal HA-BSIs.

Methods
This study was conducted in accordance with the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement. 22 The systematic review protocol is not registered.

Literature search strategy
A systematic search from January 2000 to September 2018 of the PubMed, Cochrane, and CINAHL databases was conducted by 1 researcher (S.K.) using 3 groups of key words related to the terms "bloodstream," "population," and "outcome." These 3 categories were combined using the Boolean "AND" and "OR." Appendix 1 (online) presents the full search strategy used for MEDLINE, which was adapted for the other databases. Cited references from selected articles were screened to identify additional studies that were not retrieved in the initial search. Conference abstracts were not searched because they do not contain sufficient data for quality assessment.

Selection criteria
Following the literature search, identified studies were checked to exclude duplicates. The remaining articles were independently screened by 2 researchers (S.K. and C.T.) to identify studies that met the predetermined inclusion criteria. The selection process was performed in 2 steps. In the first step, titles and abstracts were evaluated for eligibility against the predetermined criteria. In the second step, the full-text articles were assessed when the information provided in titles/abstracts was insufficient to decide on inclusion or exclusion. Any disagreements between the 2 researchers were discussed and resolved by a third researcher (G.K.).
The study eligibility criteria were selected by applying the PICOS (population, intervention, comparison, outcomes, and setting) question format: • Population: Studies referring to neonates and children <18 years of age with HA-BSI were eligible, and those that included both adult and pediatric populations were eligible only if stratified results by age group were presented. • Interventions and comparators: Studies including both a group of children with HA-BSI (BSIs) and a group of children without HA-BSI (non-BSIs) were eligible. • Outcomes: Studies that provided data on at least 1 of the following factors were included: attributable or excess LOS, or cost, or mortality due to HA-BSIs. • Study design: Cohorts or case-control studies were selected.
Cohort or case-control study design was set as a criterion because these study types are reported to measure LOS, cost, and mortality more accurately. 21 Articles that investigated only HA-BSIs caused by specific microorganisms were excluded, as were articles in which the primary outcomes were not the evaluation of attributable or excess LOS, cost, or mortality. Studies that presented mean or median values of the aforementioned attributable outcomes were included in the systematic review; their results are presented separately. However, in the meta-analysis, studies that reported median values of the attributable outcomes were excluded. Finally, only studies with their full text published in English were included.

Data extraction
Data extraction was performed by 1 researcher (S.K.), and the information was recorded in Microsoft Excel tables (Microsoft, Redmond, WA). The following data were noted: first author, year of publication, country, study design, hospital unit type, definition of HA-BSI, number of children with and without infection (BSIs and non-BSIs), and matching criteria (if used). Moreover, LOS, cost, and mortality (separately for each group), as well as attributable LOS, cost, or mortality, along with the corresponding 95% confidence intervals, were recorded. In studies where confidence intervals were not provided, we followed the recommendations of the Cochrane collaboration for calculating them. 23 In cases in which these calculations were not feasible, the relevant studies were excluded from the meta-analysis.
For studies that provided separate estimates for >1 hospital unit (eg, the neonatal and the pediatric intensive care units, NICU and PICU), estimates were recorded separately. In 3 studies, the number of children without infection (non-BSI) was calculated by subtracting the BSIs from the total number of pediatric patients included in the study 24,25 or by applying the matching ratio. 26 Quality assessment Study quality was evaluated by 2 researchers (S.K. and C.T.) using the Critical Appraisal Skills Programme Tool (CASP) for casecontrol and prospective cohort and retrospective cohort studies. 27 Disagreements were discussed with a third researcher (G.K.), and all 3 researchers ultimately reached consensus.

Statistical analysis
A meta-analysis was conducted using the STATA commands "metaan" and "metan" to estimate the pooled effect sizes with 95% confidence intervals (CIs) for attributable LOS and attributable mortality, respectively, as well as to graphically present the results in forest plots. The I 2 statistic was used to assess heterogeneity among the included studies. An I 2 > 75% indicates high heterogeneity among studies, and in such case, a random-effects model was used to obtain the pooled effect sizes. Moreover, sensitivity analysis was conducted by removing 1 study each time to identify the study that most influenced the results. Finally, the Egger test and funnel plots were used to evaluate potential publication bias.

HA-BSI attributable cost
Attributable healthcare cost was presented in 8 studies, 6,25,29,31,32,36,43,44 and ranged from $1,642.16 to $160,804 (2019 USD) per patient with HA-BSI. This range refers to the 7 studies that estimated mean (and not median) attributable healthcare cost (Table 3). At this point, providing specific data around cost per hospital unit would be inaccurate because of the small number of referring studies: only 3 studies measured PICU HA-BSI cost, 31,36,43 2 studies estimated NICU costs, 29,36 and 4 studies assessed costs for hematology-oncology patients. 6,25,36,44 All of these studies differ with regard to the corresponding currency and year.
Meta-analysis was not conducted for the attributable HA-BSI cost due to the lack of eligible studies; only 3 of the participating studies estimated standard error of attributable HA-BSI cost, and their study populations presented heterogeneity. 32,36,44 HA-BSI attributable mortality Attributable mortality was reported in 8 studies 30,34,35,[39][40][41][42][43] and was calculated in 1 study, using estimates that were provided separately for BSI and non-BSI patients. 33 The attributable mortality rate ranged between 0.01 and 0.24 (Table 4). The attributable mortality rate for the NICU was between 0.01 and 0.24, and for the PICU it was between 0.11 and 0.24.

Discussion
The goal of this systematic review and meta-analysis was to provide evidence around HA-BSI attributable LOS, cost, and mortality among pediatric and neonatal patients, targeting the design and implementation of appropriate and cost-effective prevention strategies. As far as we know, this is the first attempt to synthesize all existing data around HA-BSI outcomes in the pediatric and neonatal population.
The HA-BSI mean attributable LOS ranged from 4 to 27.80 days, and the pooled mean attributable hospital LOS was 16.90 days (95% CI, 13.70-20.11). Stratified results by type of unit revealed a higher impact of HA-BSIs in PICUs (pooled mean attributable LOS, 16.40 days) compared to NICUs (pooled mean attributable LOS, 11.40 days). The results were more consistent in PICUs, with the mean attributable LOS ranging from 11.40 to 21.10 days, whereas in NICUs it ranged between 4 and    27.80 days. However, no statistically significant heterogeneity was observed in PICUs or in NICUs. As described in the previous section, 7 of the studies that participated in the meta-analysis presented time-matched outcomes of BSI and non-BSI patients with regard to LOS, and only one used a multistate modeling approach to estimate attributable HA-BSI LOS. Previous work by Manoukian et al 21 has suggested that excess LOS associated with HA-BSIs in adult populations presents significant variations according to the statistical method employed. More specifically, Manoukian et al suggest that studies using time-fixed methods overestimate the attributable LOS compared to timevarying methods, 21 because they do not take into account the time-dependent bias. Multistate modeling is considered the most accurate statistical method of attributable LOS estimation. 20,50 This hypothesis can only be partially confirmed by our study because the only study using multistate modeling among those included in our review provided the lower estimation of excess LOS (1.57 days). 33 Attributable mean healthcare cost ranged from $1,642.16 to $160,804 (2019 USD) per patient with HA-BSI. Previous work by Umscheid et al 10 assessed attributable cost of catheterassociated BSIs (CA-BSIs) in the adult ICU and reported costs from $41,900 to $123,600 (2009 USD).
In general, the large difference observed between reported attributable cost estimates for several HAI types is due to differences in the perspective of cost analysis (ie, hospital or societal), the costing methodology (ie, microcosting approach or not), and the year of costing, as well as differences in clinical practice patterns and healthcare systems among countries (ie, use of novel and expensive technologies in high-income countries, etc). 20 Attributable mortality rate ranged between 0.01 and 0.24, and pooled mortality was 0.08 (95% CI, 0.06-0.09). Stratified analysis by type of unit revealed that the pooled mortality rate was higher in PICUs (0.13) compared to NICUs (0.08).
Previous systematic reviews and meta-analyses have presented data indicating that the odds ratio for in-hospital death associated with HA-BSIs in adult patients ranges between 1.96 and 2.75. 51,52 However, attributable HA-BSI mortality rates present significant variations, according to several causative microorganisms and susceptibility patterns. 51,53 Quantifying excess HA-BSI outcomes is essential for both healthcare providers and policy makers. Improving efficiency with regard to resources and bed days by implementing targeted prevention strategies is crucial to increasing a hospital's capacity to provide high-quality care to the highest number of patients. Precise measurements of HA-BSI outcomes could guide decision making around investments in infection control.
This study has several limitations. First, we acknowledge the possibility of language bias due to the fact that only studies written in English were incorporated in this review. Practical reasons, namely the difficulty of translating from a variety of languages, led us to the decision to include only English-language studies. Moreover, restricting the search strategy to only electronic databases may have introduced publication bias because this approach is unlikely to identify studies that have not been published in peerreviewed journals. Because we did not include unpublished studies in our review, it was impossible to assess the potential publication bias by comparing the results of published and unpublished studies. However, we applied the Egger test, which revealed that no publication bias exists in HA-BSI-attributable LOS and mortality studies. We should underscore that the Egger test is inappropriate where there is heterogeneity; the test has low power and is of little use in analyses with few studies.
Finally, another important limitation of this study is the inclusion of studies that used a variety of HA-BSI definitions, and    although all of these BSIs were nosocomial, this could result in outcome differences. We tried to overcome this problem by conducting a subanalysis in studies assessing the attributable LOS and mortality of CLABSIs, and the outcomes were presented separately. The heterogeneity in the HAI definitions used by several authors in the literature is a major problem when trying to conduct a qualitative or quantitative synthesis of the available literature data on HAI outcomes.
In conclusion, HA-BSIs in children and neonates are associated with higher mortality, LOS, and healthcare costs than in children and neonates without HA-BSIs. This finding justifies and may enhance efforts to implement HA-BSI prevention strategies. Future research efforts could make better use of existing HAI definitions and evolving statistical methodologies, presenting more accurate, high-quality, and comparable outcome results globally. Financial support. This study was part of the doctoral thesis of the physician K.S., which was financed through a scholarship by the General Secretariat for Research and Technology (GSRT) and the Hellenic Foundation for Research and Innovation (HFRI).

Conflicts of interest.
All authors report no conflicts of interest relevant to this article.