Implementation and impact of a business intelligence system on catheter-related bloodstream infections: a multicenter experience Authors: Po-Yen Huang, Chien-long Shih, Yun-Ting Chung, Ming-Wei Lin, Shian-Sen Shie, Ching-Tai Huang Published: 2025 ABSTRACT AbstractBackgroundCentral-line-associated bloodstream infections (CLABSI) and catheter-related bloodstream infections (CRBSI) are associated with increased mortality and healthcare costs. We describe implementing a hospital-wide business intelligence (BI) system coupled with infectious disease (ID) specialist intervention to support proactive CRBSI surveillance.MethodsWe conducted a multicenter, retrospective analysis of a quality improvement initiative from January 2019 to September 2024. A BI system used predefined criteria (leukocytosis, elevated C-reactive protein, prolonged catheter duration, and fever) to identify high-risk patients. Starting in July 2022, ID specialists reviewed these cases and provided advice on catheter reassessment. We employed interrupted time series (ITS) analysis to evaluate quarterly trends in CRBSI incidence and mean central venous catheter (CVC) duration, adjusting for seasonality and conducting a stratified analysis for ICU versus non-ICU settings.ResultsA total of 5090 patients with 6,750 CVC placements were included. While the overall trend for CRBSI incidence was not significantly affected by the intervention, the stratified analysis showed a statistically significant decrease in the CRBSI incidence trend within the ICU setting (slope change Δ = − 0.737/quarter, P = 0.045). No significant change was seen in the non-ICU setting. The intervention’s effect on average CVC duration was less evident, with varied and often non-significant changes across different sites.ConclusionA hospital-wide BI system, combined with ID specialist expertise, was associated with a significant reduction in CRBSI incidence over time, especially in the high-risk ICU setting. This technology-focused, expert-supported approach offers a promising approach to improving patient safety and infection control outcomes.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13756-025-01673-9.Keywords: Catheter-related bloodstream infections, Central line-associated bloodstream infections, Business intelligence system, Infection prevention, Interrupted time series analysis, Electronic surveillance Background Catheter-related bloodstream infections (CRBSI) and central line-associated bloodstream infections (CLABSI) are important healthcare-associated infections (HAI) linked to increased mortality, longer hospital stays, and higher medical costs [1, 2]. CLABSI is often used for surveillance to monitor infection rates but does not definitively identify the catheter as the infection source. Conversely, diagnosing CRBSI requires microbiological proof, offering greater clinical specificity [1, 3]. The financial impact of these infections is significant, with CLABSIs alone estimated to cost up to USD $55,000 per incident and an additional USD $129,000 for critically ill patients [4, 5]. Manual surveillance methods for CLABSI are labor-intensive and prone to delays, limiting timely interventions. They require retrospective review of medical records, lab data, and clinical documentation, often causing detection delays of days to weeks. This lag limits the opportunity for prompt catheter reassessment and removal when medically necessary. Manual processes may also miss cases due to incomplete documentation, potentially underestimating the true infection burden and hindering quality improvement efforts [6]. As a result, healthcare facilities are increasingly adopting automated electronic systems, such as business intelligence (BI) tools, to improve the accuracy, timeliness, and clinical relevance of surveillance [6, 7]. BI systems utilize real-time electronic health record data to identify at-risk patients, generate targeted alerts, and encourage proactive catheter reassessment, thereby overcoming limitations of manual surveillance [8]. These systems can reduce surveillance workload while maintaining or improving the sensitivity and specificity of infection recognition, enabling more efficient allocation of infection prevention resources [6–9]. Beyond the infection surveillance, BI platforms have shown broader applications, such as real-time monitoring of hand hygiene compliance, where interactive dashboards have sustained compliance rates above 95% by providing immediate feedback and enabling timely interventions [10]. However, evidence on the effectiveness of integrating BI systems with infectious disease (ID) specialist input remains limited, especially in large, multicenter settings. While previous studies have indicated that BI tools can improve the efficiency of case detection [6–9], most have evaluated them as surveillance tools, making it unclear whether these automated alerts consistently translate into meaningful clinical actions and better patient outcomes [7, 9]. Furthermore, the best way to combine technological capabilities with clinical judgment from specialists remains uncertain. Consistent surveillance methods are essential for meaningful comparisons with benchmark data, underscoring the importance of infection prevention staff and information technology support within multidisciplinary collaborations [11]. To address these gaps, this study aimed to evaluate the impact of a multicenter BI system combined with the engagement of ID specialists to optimize clinical interventions. By examining outcomes before and after adding ID specialist involvement, we sought to determine whether this technology-driven, specialist-supported infection prevention strategy could improve patient outcomes in a large healthcare system. Methods Study design and setting We conducted a retrospective, multicenter observational study to assess the impact of a BI system as part of a hospital-wide quality improvement initiative. The study took place across five sites of the Chang Gung Memorial Hospital System (Keelung, Tucheng, Linkou, Chiayi, and Kaohsiung) from January 2019 to September 2024. The BI system was launched in January 2019 to identify high-risk patients using predefined criteria: white blood cell (WBC) count > 10,000 cells/µL, C-reactive protein (CRP) level > 50 mg/L, central line placement > 7 days, and body temperature > 38 °C within the past 7 days. These parameters were selected based on clinical experience to identify patients with long-term central venous catheters (CVCs) and inflammatory responses, which indicate a higher risk of infection. Although the BI system was launched in 2019, its proactive use was not immediately established, partly due to the operational pressures of the COVID-19 pandemic. During this initial phase, clinicians had passive access to the BI alerts, but it was not integrated into their daily workflow, leading to infrequent use. Therefore, starting in July 2022, infectious disease (ID) specialists reviewed suspected cases flagged by the BI system. They advised clinicians via text or phone on reassessing catheter necessity when a potential benefit was anticipated. Definition of CRBSI Catheter-related bloodstream infection (CRBSI) was defined based on modified criteria adapted from the German Society of Hematology and Medical Oncology [12], and met one of the following conditions: (1) growth of the same pathogen from both peripheral blood culture and catheter tip, (2) growth of the same organism from blood cultures drawn from both a central line and a peripheral vein, or (3) isolation of pathogens in blood cultures that are typically associated with catheter-related infections (e.g., Staphylococcus spp., Candida spp., Bacillus spp., Corynebacterium spp., Lactobacillus spp., Enterococcus spp.). These criteria were selected because they offer a practical operational definition that is ideal for large-scale surveillance, especially in retrospective settings where strict microbiological evidence may not be available. Outcome measures and data analysis The primary outcomes were quarterly trends in CRBSI incidence per 1,000 catheter days and the mean duration of CVC placement. The impact of ID specialist involvement was evaluated using interrupted time series (ITS) analysis, with adjustments for seasonality. Autocorrelation was assessed using the Durbin-Watson statistic, and Heteroskedasticity and Autocorrelation-Consistent standard errors were applied when significant autocorrelation was detected. Additionally, a stratified ITS analysis was conducted to compare outcomes between patients with CVCs in the intensive care unit (ICU) versus non-ICU settings at the time of BI alert generation. A p-value < 0.05 was considered statistically significant. All analyses were performed using Python (version 3.12) in a Google Colab environment, with ITS analysis conducted using the statsmodels package (version 0.14.5). Figures were plotted using Matplotlib (version 3.10.0) and Seaborn (version 0.13.2). Results A total of 5,090 patients underwent 6750 CVC placements, accumulating 114,316 catheter days during the study period. The mean age was 63.2 years, and 64.2% of the patients were male. The median time to alert was 8 days (Interquartile range, 10–14 days). BI-generated alerts gradually increased across the hospital sites, with some sites showing increases while others remained stable or declined (Fig. 1). BI-Generated Alerts Across Hospital Sites. Quarterly trends in alerts produced by the business intelligence (BI) system across Chang Gung Memorial Hospital System from Q2 2019 to Q3 2024. Data for the Tucheng site starts in Q2 2020, corresponding with its operational launch. Tucheng experienced the most significant increase, while other sites remained stable or declined ITS analysis of CRBSI incidence across all sites combined indicated a downward trend following the intervention, although this change did not reach statistical significance (P = 0.173). However, as detailed in Fig. 2, the direction and magnitude of the post-intervention trend varied between individual sites. Interrupted Time Series Analysis of CRBSI Incidence by Site. Seasonality-adjusted interrupted time series analysis of catheter-related bloodstream infection (CRBSI) incidence per 1,000 CVC days across hospital sites. Green circles represent observed quarterly rates; dashed blue lines show pre-intervention fitted trends; solid orange lines show post-intervention fitted trends; dotted orange lines represent counterfactual projections; light blue shading indicates 95% confidence intervals. The vertical dashed red line marks the intervention point (2022 Q3). Overall analysis demonstrated non-significant downward trend (level Δ = +0.363, P = 0.818; slope Δ = −0.384, P = 0.173; R² = 0.290). Among sites, Chiayi exhibited a non-significant slope reduction (Δ = -1.621/quarter, P = 0.093) The overall mean CVC duration declined consistently throughout the study period, a trend that was not altered by the intervention (Fig. 3). Site-specific analyses revealed considerable variation in CVC duration trends, with some sites showing a decrease while others experienced a significant increase after the intervention. Interrupted Time Series Analysis of Mean CVC Duration by Site. Seasonality-adjusted interrupted time series analysis of mean central venous catheter (CVC) duration across hospital sites. Green circles represent observed quarterly means; dashed blue lines show pre-intervention fitted trends; solid orange lines show post-intervention fitted trends; dotted orange lines represent counterfactual projections (expected trends without intervention); light blue shading indicates 95% confidence intervals. The vertical dashed red line marks the intervention point (2022 Q3) when infectious disease specialist involvement began. Overall analysis showed minimal change (R² = 0.775). Tucheng exhibited a non-significant reduction in mean CVC placement duration (Δ = − 0.710 days/quarter, P = 0.077), while Kaohsiung showed a paradoxical significant increase (Δ = +0.434 days/quarter, P = 0.044) A stratified ITS analysis compared outcomes between ICU and non-ICU settings at the time of the BI alert. For CRBSI incidence, ICU patients showed a statistically significant decline in the post-intervention slope (Δ = − 0.737/quarter, P = 0.045), while non-ICU patients showed no significant change. For the mean CVC duration, both ICU and non-ICU groups showed a decreasing trend without a significant change after the intervention. An interaction analysis confirmed no significant differences between the ICU and non-ICU groups in either level or slope changes for both outcomes, suggesting that the intervention’s effect was similar across settings, despite a notable trend toward reduction within the ICU group. (Fig. 4). Interrupted Time Series Analysis of CRBSI Incidence and Mean CVC Duration Stratified by Patient Setting at Alert Generation. Seasonality-adjusted interrupted time series analysis comparing outcomes between ICU and non-ICU settings. Top panels show CRBSI incidence per 1,000 CVC days; bottom panels show mean CVC duration in days. Green circles represent observed quarterly values; dashed blue lines show pre-intervention fitted trends; solid orange lines show post-intervention fitted trends; dotted orange lines represent counterfactual projections (expected trends without intervention); light blue shading indicates 95% confidence intervals. The vertical dashed red line marks the intervention point (2022 Q3) when infectious disease specialist involvement began. Regarding CRBSI incidence, ICU patients demonstrated a significant decline in the post-intervention slope (Δ = − 0.737/quarter, P = 0.045; R² = 0.338), while non-ICU patients showed minimal change in slope. For the mean CVC duration, both ICU and non-ICU groups showed a decreasing trend before the intervention, with little change afterward Discussion The BI system effectively facilitated early identification of high-risk patients using consistent clinical criteria. Its implementation increased alert coverage and prompted a timely reassessment of catheter necessity, a key strategy in infection prevention, as prior studies have shown that early catheter removal can reduce mortality and complications [13–16]. Importantly, through the study period, our hospital maintained consistent infection control policies, with no new measures introduced to reduce CLABSI incidence. The established CVC bundle care strategies have been continuously applied since 2009 (supplementary Fig. 1). Most sites exhibited an increase in alerts over time, likely reflecting higher CVC utilization rather than shifts in detection sensitivity. These results align with previous findings that electronic surveillance systems improve infection monitoring and resource allocation compared to manual methods [6–9]. In the study, the reduction in CRBSI incidence was more noticeable in the ICU. Although the overall CRBSI rate across all sites did not show a statistically significant change, stratified ITS analysis indicated that ICU patients experienced a significant decrease in the infection trend after the intervention (Fig. 4). This suggests that the combination of automated BI alerts and ID review is particularly effective for critically ill patients. In contrast, the intervention’s impact on CVC duration was less clear, with results varying across different sites. Daily catheter reassessment is a core component of CVC bundle care and is widely endorsed as a practice for reducing catheter-related infections. This principle is typically operationalized through daily checklists, nursing reminders, or automated healthcare information system (HIS) alerts. However, in real-world settings, the volume and repetition of such reminders may contribute to alert fatigue among clinicians, diminishing their impact over time [17–19]. Our targeted alerting system addressed this challenge by selectively notifying clinicians about patients with fever, elevated inflammatory markers, and prolonged catheter duration—clinical scenarios more likely to be associated with CVC-related infection. By increasing the specificity of alerts, our system preserved clinician attention and engagement. While this approach may miss some infections, it fostered greater alert credibility and clinical engagement around catheter management. In July 2022, ID specialists were integrated into the BI analytics process through expert screening, personalized communication, and tailored suggestions. Their review filtered alerts and added interpretative value, ensuring the high-risk cases received timely attention. The subsequent reversal of CRBSI’s increasing trends in the ICU highlights the importance of this clinical collaboration (Fig. 4). Despite the longstanding implementation of CVC bundle care, including daily HIS-based reminders and nursing protocols, the addition of expert screening input enabled system-wide monitoring and more effective triage of patients at risk. This multidisciplinary teamwork may help reduce notification fatigue while maintaining focus on high-priority cases, ultimately improving the clinical relevance and adoption of the intervention [11]. The BI system was deployed simultaneously across all five hospital sites, with training embedded under the quality management center. ID specialists, who have access rights, participate in regular cross-site meetings to support its use. The alert criteria for the BI system are consistently defined across all the sites, although there may be individual variations in clinician reminders. Despite this standardized implementation, the intervention’s impact varied. The most pronounced effect was a decrease in CRBSI incidence in ICU settings (P = 0.045; Fig. 4). In contrast, the effects at the individual site level were inconsistent for both CRBSI incidence and CVC duration. Some sites showed non-significant trends toward improvement, while others showed no change or even a paradoxical increase in CVC duration. Differences in baseline practices, patient characteristics, or adherence to intervention protocols may explain this variation [20]. Institutional factors such as healthcare workers’ training, resource availability, and organizational culture can also influence the implementation and success of infection control measures. These findings align with previous studies noting variability in infection prevention outcomes across institutions [21]. Our focus on CRBSI, rather than CLABSI, aimed to accurately measure BI’s impact by using microbiological criteria. While CLABSI, mainly used for surveillance, detects bloodstream infections in patients with a central line in place for more than two consecutive days, it does not require verification that the catheter is the actual source. In comparison, CRBSI demands stricter evidence—such as matching culture results from catheter tips and peripheral blood—to confirm the catheter’s role. This difference is important because bloodstream infections can originate from other sources besides the catheter, and confirmation often happens only after removing the catheter. Therefore, using CRBSI provides a more precise evaluation of infection outcomes related to BI alerts, even if the diagnosis is not made at the time of the alert. This method helps connect real-time surveillance with meaningful infection attribution, improving the interpretation of BI performance. This study has several strengths, including a large, multicenter dataset that enhances generalizability, standardized CRBSI definitions that improve diagnostic accuracy, and an extended observation period covering both pre- and post-intervention phases, allowing for the analysis of temporal trends and long-term effects. Our methodology also facilitated precise tracking of each catheter episode in patients identified by the BI system, and we used microbiological criteria to diagnose CRBSI, reducing bias in outcome assessment. However, several limitations should be acknowledged. First, we focused exclusively on patients who triggered BI alerts, which may have excluded cases with catheter-related infections that did not present with fever or leukocytosis. Second, other intravascular devices such as PICCs, Port-A catheters, arterial lines, or dialysis catheters were not included, potentially underestimating the overall burden of catheter-related infections. Third, we did not incorporate diagnostic criteria like differential time-to-positivity, which could have improved specificity. Since the BI system was implemented simultaneously across all sites, there was no control group. The BI alert criteria were based on local experience rather than the literature or a multivariable analysis. Specific frequency data are unavailable due to the study’s retrospective design and variability in clinical workflows across sites. Regarding adherence, we did not systematically track the proportion of cases where their recommendations were followed. Further studies are needed to quantify adherence rates and better assess the feasibility and reproducibility of the model. Although the observed trends are promising, reductions in CVC placement duration and CRBSI incidence could be influenced by other concurrent changes in clinical practices. Lastly, while CRBSI classification improves accuracy in estimating infection outcomes, it is retrospective and may not fully reflect decision-making at the point of care. Future studies should explore real-time prediction and microbiological outcomes in these high-risk patients. Conclusion Our study demonstrates that a hospital-wide BI system coupled with ID specialist involvement was associated with a temporal decrease in CRBSI incidence, particularly in the ICU. These findings represent a promising strategy for infection control in hospital settings. As healthcare systems increasingly adopt digital transformation, our experience provides valuable insights into effectively integrating technology and clinical expertise to improve patient outcomes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors sincerely thank the infectious disease specialists, members of the Infection Control Department, and the nursing staff for their longstanding dedication and effective implementation of the bundle care strategies. We also express our gratitude to the Medical Quality Department and the Information Technology (MIS) Department at Chang Gung Memorial Hospital for their valuable support and collaboration throughout this study. The authors also wish to thank Peiyi Cheng and Ko Cheng for their statistical consultation, as well as the Center for Big Data Analytics and Statistics at Chang Gung Memorial Hospital, Linkou, for assistance with data analysis and interpretation. Abbreviations Business intelligence Central line-associated bloodstream infections Catheter-related bloodstream infections Central venous catheter White blood cell C-reactive protein Healthcare-associated infections intensive care unit Infectious disease Interrupted time series Healthcare information system Peripherally inserted central catheter Author contributions PYH and CTH conceived and designed the study. PYH, CLS, YTC, MWL, and SSS contributed to the implementation and data collection. PYH analyzed the data. PYH and CTH drafted and critically revised the manuscript. All authors reviewed and approved the final version of the manuscript. Funding No funding was received for this research project. Data availability All data supporting the findings of this study are available from the corresponding authors upon reasonable request. Declarations Ethics approval and consent to participate was obtained from the Institutional Review Board of Chang Gung Memorial Hospital, Taoyuan, Taiwan (IRB No. 202500451B0). The ethics committee waived informed consent due to this study’s retrospective observational nature. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Po-Yen Huang, Email: pyhuang@gmail.com. Ching-Tai Huang, Email: chingtaihuang@gmail.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data supporting the findings of this study are available from the corresponding authors upon reasonable request.