In 2013, the US Centers for Disease Control and Prevention (CDC) released ventilator-associated event (VAE) definitions. To replace traditional concepts for surveillance of ventilator-associated pneumonia (VAP), the new definition was designed to identify a broad range of complications in patients receiving mechanical ventilation. 1,Reference Klein Klouwenberg, van Mourik and Ong2 In 2019 in the United States, 2,028 hospitals were conducting VAE surveillance. 3,Reference Klompas4 Recently, the new definition has been also adopted outside the United States. Reference Zhu, Cai and Ma5-Reference Chao, Chang, Wu and Chan7 Studies have suggested that VAE is common among patients receiving mechanical ventilation Reference Zhu, Cai and Ma5 and that a VAE may result in an adverse outcome, including death. Reference Zhu, Cai and Ma5,Reference Agrafiotis, Siempos, Ntaidou and Falagas8,Reference Boyer, Schoenberg, Babcock, McMullen, Micek and Kollef9
However, sound evidence to support the management of VAE is lacking. Reference Klompas4,Reference Klompas, Li, Kleinman, Szumita and Massaro10 In 2015, Klompas Reference Klompas11 has proposed several potential strategies to prevent and control VAE. Some proposed interventions have effective Reference Klompas, Li, Kleinman, Szumita and Massaro10,Reference Klompas11 ; however, the impact of other interventions, such as blood transfusion, is not yet fully understood. Reference Klompas11 Previous studies have reached conflicting conclusions. Reference Villanueva, Colomo and Bosch12-Reference Rohde, Dimcheff and Blumberg16 Furthermore, only a limited number of studies have focused on VAE, Reference Lewis, Li, Murphy and Klompas15,Reference Cocoros, Priebe and Gray17 which typically involved small sample sizes and reached inconsistent conclusions. For instance, a nested case-control study including 192 pairs of children with ventilator-associated complication (VAC) and controls has demonstrated that blood product use was associated with a higher risk of VAC among patients admitted to the neonatal intensive care unit (ICU). Reference Cocoros, Priebe and Gray17 However, another case-control study has exhibited no difference between VAC and controls regardless of type or volume of blood transfusion. Reference Lewis, Li, Murphy and Klompas15
In 2015, West China Hospital (WCH) developed a prospective surveillance system for healthcare-associated infections (HAIs) in ICUs. As of December 31, 2018, nearly 30,000 ICU admissions had been monitored, and 1,800 were judged as VAE cases. Based on the surveillance system at WCH, we conducted a nested case-control study with a relatively large sample size to evaluate the association between blood transfusion and risk of VAE.
Methods
This study followed the reporting standards according to Reporting of Studies Conducted Using Observational Routinely Collected Health Data Reference Benchimol, Smeeth and Guttmann18 This study was reviewed and was approved by the West China Hospital Institutional Review Board (WCH2018-409).
Data source
We conducted a nested case-control study using ICU-HAI registry in the WCH system, which includes 3 independent healthcare institutions and is a national critical care center. The system provides tertiary care for Sichuan and other provinces. 19
The ICU-HAI registry was developed by a integrating multisource database, including the ICU-HAI system, ICU system, and electronic medical record (EMR) system. Reference Wang, Zhu and He20 The ICU-HAI system is a prospective surveillance system that monitors all patients admitted to ICUs. Information regarding catheterization, HAI, and prevention and control measures is actively collected by experienced infection control experts. This unique system for routine VAE surveillance in China monitors >5,000 cases of VAE every year. The ICU system contains information regarding vital signs, notes, life support, and risk assessments, which is electronically recorded by well-trained nurses. The EMR system contains patient-level clinical care information including demographics, laboratory test results, prescriptions, and diagnoses. Reference Li, Yu and Li21 By integrating the 3 databases using a unique identifier, the ICU-HAI registry is a valuable data source for management of HAIs in the ICU setting. Reference Wang, Zhu and He20,Reference Zhu, Kang, Wang, Cai, Sun and Zong22
The ICU-HAI registry includes patients admitted to 6 ICUs in the WCH system starting April 1, 2015. As of December 31, 2018, the registry included 28,848 patients and 29,480 admissions. Previous studies showed that the registry had a high level of quality and comprehensiveness. Reference Wang, Zhu and He20,Reference Li, Yu and Li21,Reference Wang, Liu and Yu23,Reference He, Wang and Zhu24
Study cohort and design
Patients who consistently received mechanical ventilation for at least 4 days between April 1, 2015, and December 31, 2018, were identified from the registry. We excluded patients with incomplete information, those aged <18 years, those with extremely long hospital stays (>365 days), and those who did not pay for hospitalization because of a medical dispute. The clinical characteristics differed among non-VAE patients with and without consecutive stable or improved respiratory status. Therefore, we also excluded patients with consecutively unstable or increased minimum daily fraction of inspired oxygen (FiO2) or positive end-expiratory pressure (PEEP) during the mechanical ventilation period. We additionally excluded 363 non-Chinese patients because many of them were transferred elsewhere and, therefore, the information regarding their clinical outcomes was incomplete.
We chose a case-control design nested within the cohort, to ensure identical duration of mechanical ventilation between patients with VAE and those without and to improve the efficiency of estimation of treatment effects. Reference Essebag, Genest, Suissa and Pilote25
Case identification and control selection
We defined cases as patients with VAEs. VAE cases were judged according to the CDC criteria, defined as ≥2 days of increased FiO2 or PEEP after ≥2 days of stable or improving ventilator settings. 1,Reference Klein Klouwenberg, van Mourik and Ong2 All patients with a VAE were prospectively and actively identified from the ICU-HAI system. ICU nurses electronically recorded the value of FiO2 and PEEP on an hourly basis. A threshold-based warning system was implemented in the ICU-HAI system. Once the value of FiO2 or PEEP reached the threshold, an alarm was triggered. Thereafter, a team of infection control practitioners judged whether the patient had a VAE, and if so, classified VAE as VAC, infection-related ventilator-associated complication (IVAC), or possible ventilator-associated pneumonia (PVAP). Validation of PVAP showed that the consistency of this classification between 2 infection control practitioners was 96.2%. In this study, VAE cases were furtherly classified as (1) VAC-plus, defined as all patients with VAC, including those also met criteria for IVAC and PVAP; (2) VAC-only, defined as patients who met criteria for VAC, but not for IVAC and PVAP; (3) IVAC-plus, defined as patients with IVAC, including those also met criteria for PVAP; (4) PVAP.
For each VAE case, we randomly selected up to 2 controls from those still at risk for VAE at time of the case occurrence. We used incidence density sampling approach to match cases and controls with the same number of mechanical ventilation days (the number of days from the initiation of mechanical ventilation to VAE). We also matched cases and controls for age (<45, 45–59, 60–75, 75–89, and ≥90 years) and the number of days between ICU admission and initiation of mechanical ventilation. For VAE cases, only the first episode of VAE was measured if patient developed >1 episode of VAE; for non-VAE patients, only the first episode of mechanical ventilation was measured if patients had multiple mechanical ventilation treatments.
Blood transfusion and covariates
Blood transfusion was defined as administration of red blood cells (RBCs), platelet, or plasma. Information regarding blood transfusion was extracted from the physician’s order of hospital information system. We measured blood transfusion as a time-varying variable. Daily blood transfusion from initiation of mechanical ventilation to the event of interest were calculated. We also measured daily RBC transfusion and transfusion volume from initiation of mechanical ventilation to the event of interest.
We identified potentially confounders from the ICU-HAI registry, including demographic characteristics, ICU type at admission, acute conditions, chronic comorbidities, APACHE II score at admission, daily fluid balance, medications, operations, and other treatments. Acute conditions, fluid balance, medications, and other treatments were measured from initiation of mechanical ventilation to the event of interest. Detailed description of covariates was listed in Table 2.
Information regarding medication was identified from prescription data, while chronic comorbidity information was identified from discharge diagnosis codes using the International Classification of Diseases, Tenth edition. With respect to acute comorbidities at ICU admission, we performed text mining by structuring the text information from the transfer records.
Statistical analysis
We assessed associations between blood transfusion (ie, RBC transfusion, plasma transfusion, platelet transfusion, and all blood transfusion types combined) and risk of VAE using weighted Cox proportional hazards model. The weighted Cox proportional hazard is a method that can estimate the effects of complex time-varying exposures in case–control studies. Reference Leffondre, Wynant, Cao, Abrahamowicz, Heinze and Siemiatycki26 Hazard ratios (HRs) for VAC-plus, VAC-only, IVAC-plus, and PVAP were calculated. We also measured the associations between RBC transfusion volume and the risk of all tiers of VAEs. Hazard ratios of RBC transfusions of 0–3 units and RBC transfusions of >3 units were also calculated.
All analyses were adjusted using a combination of fixed and time-varying covariates (Table 2). Time-varying variables including daily exposure to medication and process were measured with respect to whether each treatment was performed on each day from the initiation of mechanical ventilation to events of interest. We used multiple imputation to handle missing data of APACHE II score and fluid balance. The proportions of missing data for APACHE II score and fluid balance were 10.2% and 7.0%, respectively. Compared with single imputation (eg, mean or regression imputations), multiple imputation considers uncertainty behind missing value estimations, which is considered superior to single imputation. Reference Zhang27
Sensitivity analyses
We conducted several sensitivity analyses to evaluate the robustness of effect estimates: (1) alternative definition of blood transfusion: daily blood transfusion within 4 days prior to the event of interest; (2) alternative approach for missing data: complete case analysis without imputation; and (3) alternative inclusion and exclusion criteria: excluding patients who developed VAE within 4 days after mechanical ventilation began.
Results
In total, 29,480 patients were assessed for eligibility. After exclusion, 6,176 patients receiving mechanical ventilation consistently for at least 4 days and with at least consecutive 2 days of stable or improved respiratory function were included in the initial cohort. After performing incidence density sampling, 1,657 patients with VAE and 3,293 matched controls were finally included. Among patients with VAE, the median day from initiation of mechanical ventilation to VAE was 4 days (interquartile range [IQR], 3–8), and the median day from first blood transfusion to VAE was 3 days (IQR, 2–6).
The median age was 59 years (IQR, 46–69), and 3,122 patients (63.1%) were male. The distribution of ICUs differed among cases and controls. In total, 27.7% patients with VAE and 16.2% of controls were admitted to the surgical ICU. The most common acute conditions were pneumonia (9.6%) and shock (5.7%), and the most common comorbidities were hypertension (21.1%), heart failure (9.6%), and malignant tumor (8.5%). The proportions of comorbidities and acute conditions were similar in cases and controls, except for heart failure. The proportion of heart failure was higher in controls than in patients with VAEs (10.4% vs 7.9%; P = .006). The median hemoglobin at ICU admission was 101 g/L (IQR, 84–120), which was similar in cases and controls (101 g/L; [IQR, 84–121] and 102 g/L [IQR, 84–120], respectively; P = .94) (Table 1).
Characteristics of Cases and Matched Controls
Note. IQR, interquartile range; ICU, intensive care unit; ARDS, acute respiratory distress syndrome; MV, mechanical ventilation.
Compared with controls, the median hospitalization stays (27 days [IQR, 17–41] vs 24 days [IQR, 16–36] days; P < .001), ICU stays (18 days [IQR, 11–30] vs 14 days [IQR, 9–23]; P < .001), and mechanical ventilation days (13 days [IQR, 8–21] vs 9 days [IQR, 6–15]; P < .001) were longer among patients with VAE. The crude mortality rate among patients with VAEs were also higher than those without a VAE (20.3% vs 14.4%; P < .001) (Table 1).
Blood transfusion among included patients
Among 1,675 patients with VAEs, 802 (48.4%) received blood transfusions. Of these, 652 (39.3%) received RBC transfusions, 487 (29.4%) received plasma transfusions, and 168 (10.1%) received platelet transfusions. The median number of units of transfused RBC was 0 (IQR, 0–3.5) among patients with VAEs (Supplementary Table 1 online). The proportion of patients receiving blood transfusion among controls were similar to those in the VAE group. Among those who did not experience a VAE, 1,216 (36.9%) patients received RBC transfusions, 921 (28.0%) received plasma transfusions, and 404 (12.3%) received platelet transfusions (Supplementary Table 1 online).
Associations between blood transfusion and VAE
We summarized associations between blood transfusion and risk of VAE (Table 2). Compared to no blood transfusion, blood transfusions from initiation of mechanical ventilation to the events of interest were associated with significantly higher hazards for VAC-plus (HR, 1.47; 95% CI, 1.22–1.77; P < .001), VAC-only (HR, 1.29; 95% CI, 1.01–1.65; P = .038), IVAC-plus (HR, 1.78; 95% CI, 1.33–2.39; P < .001), and PVAP (HR, 2.10; 95% CI, 1.10–3.99; P = .024). RBC transfusions were associated with increased risk of VAC-plus (HR, 1.34; 95% CI, 1.09–1.65; P = .006), IVAC-plus (HR, 1.70; 95% CI, 1.22–2.36; P = .002), and PVAP (HR, 2.49; 95% CI, 1.17–5.28; P = .018), but not with significantly increased risk of VAC-only (HR, 1.14; 95% CI, 0.86–1.50; P = .36). Plasma transfusions were associated with significantly higher risk of VAC-plus (HR, 1.39; 95% CI, 1.10–1.75; P = .006) and IVAC-plus (HR, 1.82; 95% CI, 1.22–2.70; P = .003); however, no statistically significant increase in VAC-only (HR, 1.21; 95% CI, 0.90–1.63; P = .211) or PVAP (HR, 1.82; 95% CI, 0.76–4.37; P =.181) was detected. We detected no statistically significant increase in all tiers of VAE for platelet transfusions (Table 2).
Associations Between Blood Transfusion and Risk of VAE a
Note. VAE, ventilator-associated events; IVAC, infection-related ventilator associated complication, PVAP, possible ventilator-associated pneumonia; RBC, red blood cell.
a Model adjusted for ICU type, acute conditions (ie, acute respiratory distress syndrome, shock, gastrointestinal bleeding, and pneumonia), chronicity comorbidities (ie, malignant tumor, chronic lung disease, cardiovascular disease, congestive heart failure, liver and renal failure, and diabetes), APACHE II score at admission, daily fluid balance, medications (ie, neuroleptic agents, opioids, sedative, thromboembolism prophylaxis, acid inhibitors, neuromuscular blockers, intestinal probiotics, expectorant, antibiotics, immunosuppressive agent, and vasopressors), operation and other treatment (ie, cranial or cardiac surgery, fiberoptic bronchoscopy examination, gastrointestinal decompression, tracheotomy, mandatory ventilation, enteral nutrition, esophagogastroduodenoscopy, and head-of-bed elevation).
Compared to patients without RBC transfusions, patients with RBC transfusions of >3 units had significantly higher hazards for VAC-plus (HR, 1.73; 95% CI, 1.25–2.40; P = .001) than those without RBC transfusions. However, RBC transfusion of 0–3 units was not associated with significantly higher hazard for VAC-plus compared to treatment without RBC transfusion (HR, 1.18; 95% CI, 0.92–1.50; P = .193). Compared to those without blood transfusion, the risk of IVAC-plus was significantly higher both in patients with RBC transfusions of >3 units (HR, 2.11; 95% CI, 1.24–3.58; P = .006) and those with RBC transfusions of 0–3 units (HR, 1.52; 95% CI, 1.03–2.24; P = .037). The risk of PVAP was significantly higher in patients with RBC transfusions of >3 units (HR, 3.37; 95% CI, 1.08–10.47; P = .036) but not in those with RBC transfusions of 0–3 units (HR, 2.05; 95% CI, 0.82–5.17; P = .126) (Table 2).
Sensitivity analyses
Sensitivity analyses using alternative definition of exposure showed similar results. The results were consistent among the subset of patients with complete data (N = 3,501) for all transfusion types combined, but the confidence interval crossed 1 for RBCs alone (HR, 1.30; 95% CI, 0.97–1.74; P = .083). The sensitivity analysis excluding patients who developed VAE within 4 days of mechanical ventilation initiation (N = 2,439), the hazard ratios for all blood transfusion types combined was similar to the primary analysis but with wider confidence intervals that crossed one (HR, 1.25; 95% CI, 0.93–1.67; P = .138) (Supplementary Table 2 online).
Discussion
The main finding of this study was that blood transfusion, especially RBC transfusion, was common among patients in the ICU who received mechanical ventilation. Compared to those who did not receive blood transfusions, patients who received blood transfusions had significantly higher risks for all tiers of VAEs. The risks were significantly higher among patients who received RBC transfusions of >3 units than among those who did not.
Conservative blood transfusion has been proposed as a potential strategy to prevent VAEs. Reference Klompas11 The findings of this study lend further support to this strategy. The reason for the increased risk of VAE may have been increased risk of infection, and in part, volume overload. Our study showed that blood transfusion increased the risk of both infection-related and non–infection-related complications, which supports the hypothesis. Blood transfusions have been shown to have important effects on immunologic function, increasing the risk of infection. Reference Lannan, Sahler, Spinelli, Phipps and Blumberg28,Reference Buddeberg, Schimmer and Spahn29 In vitro studies have shown that allogeneic transfusion may induce dysregulation of inflammation, downregulation of cellular host defenses, and enhanced B-cell function. Reference Buddeberg, Schimmer and Spahn29,Reference Baumgartner, Silliman, Moore, Banerjee and McCarter30 Indeed, several clinical studies have demonstrated a higher risk of infections among transfused patients versus those without. Reference Buddeberg, Schimmer and Spahn29,Reference Taylor, Manganaro, O’Brien, Trottier, Parkar and Veremakis31,Reference Taylor, O’Brien and Trottier32 A study involving 1,717 ICU patients found that RBC transfusion was associated with a higher risk of nosocomial infections; the infection rates in the transfusion and the nontransfusion group were 15.38% and 2.92%, respectively. Reference Taylor, Manganaro, O’Brien, Trottier, Parkar and Veremakis31 Infection, especially pulmonary infection, may further lead to respiratory deterioration and VAEs. Moreover, volume overload may be another potential mechanism for VAEs. Patients receiving a blood transfusion have a higher risk of circulatory overload. Reference Klompas11,Reference Villanueva, Colomo and Bosch12 A randomized controlled trial suggested that, compared with a liberal strategy, a restrictive transfusion strategy was associated with decreased risk of cardiac complications, such as pulmonary edema. Reference Villanueva, Colomo and Bosch12 A systematic review also showed that a restrictive transfusion strategy was associated with a significantly lower hazard for pulmonary edema compared to a liberal strategy. Reference Salpeter, Buckley and Chatterjee13
Some studies, however, have not shown significant associations between blood transfusions and risk of VAE. Reference Lewis, Li, Murphy and Klompas15,Reference Cocoros, Priebe and Gray17 The potential reason for the inconsistent conclusion may be partly due to the heterogeneity of included patients. The effects of blood transfusion on adverse outcomes varied among patients with various comorbidities. Reference Lewis, Li, Murphy and Klompas15,Reference Rohde, Dimcheff and Blumberg16 Moreover, the number of RBC units transfused may also be associated with adverse outcomes. A study showed that the risk of infection increased 9.7% for every unit increase in RBCs transfused. Reference Taylor, O’Brien and Trottier32 Our study also showed that patients transfused with >3 RBC units has an increased risk of all 3 tiers of VAEs; however, no significant differences were detected for the risk of VAE and PVAP among patients who received transfusion of <3 RBC units.
This study has several strengths. The sample size was relatively large, and it included thousands of patients with VAE. We identified VAE cases from a prospective system that routinely underwent VAE surveillance for all patients admitted to ICUs. The accuracy of PVAP has been shown to be high. We considered blood transfusions as time-varying variates and adjusted for an extensive array of potential confounders.
However, our study has several limitations. First, this was an observational study. Although we performed a number of statistical models, residual confounding factors remained. Second, due to the observational nature of the study, blood transfusion was not randomly performed but was left to the discretion of physicians. Information regarding the indication of blood transfusion was lacking, which may have led to indication bias. Third, transfusion strategies vary among individual clinicians, and we were unable to determine the optimal transfusion trigger, and assess the association between the rate of transfusion and VAE. Finally, we collected data from a homogeneous healthcare system; therefore, our findings may not be generalizable.
Acknowledgments
Financial support
W.W. received grant support from the National Key Research and Development Program of China (grant no. 2020YFC2009003). X.S. received grant support from Sichuan Youth Science and Technology Innovation Research Team (grant no. 2020JDTD0015) and the 1·3·5 Project for Disciplines of Excellence, West China Hospital, Sichuan University (grant no. ZYYC08003). Z.Y.Z. received grant support the 1·3·5 Project for Disciplines of Excellence, West China Hospital, Sichuan University (grant no. ZYYC08006).
Conflicts of interest
All authors report no conflicts of interest relevant to this article.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/ice.2021.178
