Mechanical Ventilation and Hospital-Acquired Venous Thromboembolism Among Critically Ill Children Free

We conducted a multicenter, retrospective cohort study utilizing the PHIS registry (Children’s Hospital Association, Lenexa, KS) from January 2018 through December 2019. The PHIS registry is an administrative database from 49 children’s hospitals in the United States in which each encounter encompasses demographic data and up to 41 International Classification of Diseases, 10th revisions, Clinical Modification Diagnostic codes (ICD-10). The registry includes Clinical Transaction Classification (CTC) codes used to identify pharmaceutical, supply, laboratory, radiographic, and procedural data during each unique encounter. Encounters are characterized by type (observation, outpatient, inpatient) and level of service (eg, PICU). This study was reviewed and approved by our local institutional review board (IRB 00306446).

A complete list of ICD-10, CTC, and clinical service billing codes used to identify HA-VTE and MV are provided in Supplemental Table 5. Inclusion criteria were age <18 years and PICU admission date falling within the study period. We excluded admissions for immediate postnatal care (ie, NICU), observational status hospitalizations, and patients with a principal admission diagnosis of VTE (a discrete field within PHIS). The primary outcome was the proportion of HA-VTE observed through hospital discharge, as defined by extremity DVT, PE, and organ-specific DVT (eg, splanchnic venous thrombosis, cerebral sinovenous thrombosis) among encounters with and without MV exposure. General patient and encounter characteristics extracted for study including patient age, gender, ethnicity, race, PHIS pediatric medical complexity algorithm (PMCA, avalidated index for medical complexity)³⁹  score, prematurity, comorbidities (grouped by organ system), CVC insertion, length of stay (LOS), MV exposure, and index mortality. The MV variable only includes invasive ventilation via an endotracheal tube and not noninvasive modalities, such as mask or nasal bilevel positive airway pressure. Diagnostic comorbidities by organ system and medical technology dependency (ie, those with tracheostomy +/− ventilator assistance, renal dialysis, gastrostomy tube, cerebrospinal fluid shunt) are collated by PHIS into discrete variables available for aggregate analyses.³⁹ 

In addition to assessing for differences in HA-VTE occurrence among children with and without MV exposure, we isolated cases and controls from the study sample with and without HA-VTE. Then, hospital characteristics, demographics, and other exposure variables were assessed to explore for associations with the development of HA-VTE. As potential exposures may vary by study sample features, we further explored exposure variables among cases and controls with and without HA-VTE among children who underwent MV during their hospital stay.

Descriptive statistics were used to summarize patient characteristics using proportions with percentages, means ± SDs, or medians (interquartile range, IQR) depending on data distribution and type. For HA-VTE proportions among encounters with and without MV exposure, 95% confidence intervals (CI) for estimates are reported. Quantitative data distribution were assessed for normality using Shapiro-Wilk tests. Comparative analyses between 2 independent groups were employed including χ² for categorical variables and Wilcoxon rank-sum for continuous variables. To evaluate for potential relationships with HA-VTE for MV and other salient covariates, multivariable mixed-effects logistic regression with the random-effect parameter set as participating hospital center were employed, yielding odds ratios (ORs) with corresponding Wald 95% CIs. Covariates were selected a priori informed from prior publications regarding HA-VTE among critically ill children, including patient age on admission, CVC presence, PMCA, comorbid malignancy, comorbid infectious disease, and technology dependence. A goodness-of-fit analysis was performed for the logistic model (Pearson χ²). Type I error was set at 0.05. Encounters were used as the unit of analysis and assumed independent. Analyses used Stata© v15.1 (Statacorp, College Station, TX).

After study criteria were applied, we identified 205 231 unique encounters from 47 participating centers, including children with admission dates between January 2018 through December of 2019. General sample characteristics and clinical outcomes for the study sample and for cohorts defined by encounter exposure to MV are shown in Table 1. Median age was 3.5 years (interquartile range [IQR]: 0.9–10.9), median LOS was 5 days (IQR: 3–10), 11.8% (n = 24 158) had a CVC inserted during hospitalization, 34.5% underwent MV (n = 70 829), and mortality rate was 2.4% (n = 5000). The overall proportion of HA-VTE was 2.2% (n = 4530, 95% CI: 2.1%–2.3%) and greater among those with exposure to MV (4.4%, 95% CI: 4.2%–4.4%) as compared with those without MV (1.1%, 95% CI: 1%–1.1%, P < .001). The mean hospital center HA-VTE occurrence rates was 2.2±0.9% and ranged from 0.9% to 4.9% (Fig 1).

Encounter characteristics by presence or absence of HA-VTE are provided in Table 2. Compared with those without HA-VTE, encounters with HA-VTE had a higher frequency of MV (68.6% versus 33.7%), CVC placement (51.4% versus 10.9%), index mortality (9.4% versus 2.3%), greater PMCA (3 [IQR: 3–3] versus 3 [IQR: 2–3]) and longer LOS (25 [IQR: 11–55] versus 3 [IQR: 3–9] days (all P < .001). Unadjusted logistic regression evaluating the association between MV and HA-VTE yielded an OR of 4.3 (95% CI, 4.0–4.6; P < .001). In a multivariate logistic regression (Table 3 ), MV exposure remained an independent risk factor for HA-VTE, increasing odds of HA-VTE 2.5-fold (OR 2.51, 95% CI, 2.33–2.69; P < .001). Observed HA-VTE events were not associated with participating hospital center.

Compared with those without MV exposure, patients undergoing MV were younger (2.3 years [IQR:0.5–8.8] versus 4.3 [IQR:1–11.8]), more frequently had a CVC (22.2% versus 6.3%), had a longer LOS (9 [IQR:4–20] versus 4 [IQR:2–7] days), and higher mortality (6.5% versus 0.3%). General characteristics and clinical outcomes for encounters receiving MV with and without HA-VTE are summarized in Table 4. Children in this MV subgroup who developed HA-VTE were younger (1.7 [IQR:0.3–9.4] versus 2.3 [IQR 0.5–8.8] years), had greater PMCA scores (3 [IQR:3–3] versus 3 [IQR:2–3]), experienced a longer LOS (37 [IQR:19–73] versus 8 [IQR:4–18]), had a higher frequency of CVC placement (59.5% vs 20.5%, all P < .001), and a greater mortality (13.3% versus 6.2%) than those who did not develop HA-VTE. Comorbid diagnoses varied in this subgroup, with a greater frequency of cardiac, hematologic or immunologic, malignancy, metabolic, infectious diseases, renal or urologic, and prematurity diagnoses among those with, as compared with without, HA-VTE (all P < .001).

In this multicenter, retrospective cohort study assessing occurrence of (and risk factors for) HA-VTE from over 200 000 PICU encounters among 47 participating centers over a 2-year period, we found that the frequency of HA-VTE was significantly greater among those with, as compared with without, MV exposure (4.4% versus 1.1%). In a multivariable logistic regression model with adjustment for risk factors identified previously in smaller retrospective studies, MV exposure was a statistically significant, independent risk factor for HA-VTE among children hospitalized in the PICU, with a magnitude of increased odds (OR: 2.51, 95% CI: 2.33–2.69). These data are consistent with, and extend the findings of, univariate analyses published by Jaffray et al²⁶  and the Children’s Hospital Acquired Thrombosis Consortium that found MV within 24-hours of PICU admission to be associated with subsequent HA-VTE. Our findings highlight the importance of future inquiry regarding anticoagulant-based, phased thromboprophylaxis trials for hospitalized at-risk critically-ill children undergoing MV.

The purpose of MV is to optimize gas exchange, limit metabolic stress, and rest a critically ill patient’s fatigued muscular system. Yet, the provision of MV is a delicate balance between applying adequate respiratory support with limited acquired physical and biologic trauma. For example, regional lung overdistention during positive pressure MV is known to result in local and systemic inflammation.^31–34  A spectrum of MV-related toxicities, including high-volume overdistention (ie, “volutrauma”), high-pressure ventilation (ie, “barotrauma”), repetitive low volume ventilation (ie, “atelectotrauma”), and low ratios of tidal volume to barometric force required to achieve goal minute ventilation (ie, “compliance-trauma”) have been described.^40–43  Toxic physiologic interactions secondary to MV are characterized by pulmonary inflammatory infiltrates, hyaline membrane development, increased endothelial permeability, and pulmonary edema. In animal models assessing MV, elevated plasma levels of proinflammatory cytokines (interleukins-1β, 6, 8, and tumor necrosis factor-α) and polymorphisms in their regulatory genes can induce hypercoagulability.^35–38  Our study findings, in the context of the unavoidable nature of invasive therapies for children with respiratory failure, warrant further evaluation in prospective multicenter cohort studies through which risk models can be derived and optimized, to inform the design of next-step risk-stratified clinical trials of anticoagulant thromboprophylaxis in critically ill children undergoing MV. A theoretical basis for ventilator-associated hypercoagulopathy is proposed in Figure 2 linking MV physiologic extremes, inflammation, and subsequent hypercoagulability. Such models are likely to be enhanced via the incorporation of candidate molecular biomarkers of ventilator-associated coagulopathy, a phenomenon that remains poorly defined to date.

Adults develop HA-VTE at an estimated frequency of 0.4% to 1.3% with 40% of cases having 3 or more risk factors, such as greater severity of illness, postoperative care, oncologic disease, traumatic illness, acute immobilization, CVC presence, prior VTE, greater patient age, and obesity.^44–47  Adult HA-VTE rates are far greater among those with medical-surgical critical illness ranging between 13% to 31%.⁴⁸  Similar to our data, observational studies describing risk factors for HA-VTE have revealed a greater frequencies of PE, limb DVT, and nonlimb DVT among critically all adults undergoing MV.^49,50  Yet, as with pediatrics, risk analyses for HA-VTE among critically ill adults have focus on validated risk scores and have yet to investigate MV as an independent risk factor.⁵¹  A systemic review and meta-analysis by Alhazzani and colleagues showed the presence of any heparin-based thromboprophylaxis compared with placebo among critically ill adults pooled risk of DVT and PE without increasing the risk of clinically relevant major bleeding or mortality.⁵²  These data, along with other supporting evidence in adult literature, back the current recommendations for pharmacologic and mechanical thromboprophylaxis for all critically ill adults without contraindication.^7,53 

Although we speculate an acquired prothrombotic state related to MV plays an integral role in the development of HA-VTE, several dependent clinical factors coexist among children with critical illness undergoing MV that may intensify HA-VTE risk in our study population. For example, 59.5% of children undergoing MV with a HA-VTE in our study also had a CVC present during hospitalization. Children undergoing MV typically are iatrogenically immobilized for safety or to enhance patient comfort, which may add to venous stasis, altered blood flow states, and prothrombotic risk. Ultimately, many potential covariates were accounted for in our model to control for confounding, but their presence should inspire future investigation to determine their contributing impact to HA-VTE risk within the context of children undergoing MV.

This study has several limitations. First, PHIS registry data lack validated indices of acute severity of illness and the presence of billing does not ensure that a patient had a specific diagnosis, received a prescribed medication, or experienced a procedure during hospitalization. Erroneous coding regarding HA-VTE or MV could lead to misclassification, sampling bias, and misinterpretation of exploratory analyses. Similarly, absent coding would result in similar bias. Dates of service for billing codes are based upon calendar days without an exact time stamp. As a result, there is a theoretical ∼48-hour window for any event, outcome, procedure, or prescribed medication to have occurred. Because of this, we were unable to comment upon temporal or sequential associations. Second, encounters were viewed as independent; however, given the chronic nature of PICU admissions, it is possible that individual patients are represented in multiple hospitalizations. Third, because of limitations in documentation of mechanical or anticoagulant based thromboprophylaxis, we cannot account for institution-specific variability in preventive practices that may play a role as effect modifiers for HA-VTE. Lastly, the influence of other factors from the triad of Virchow (eg, venous stasis, and immobility as 1 surrogate thereof) could not be measured, as these factors are not recorded as distinct variables within the registry.

In this multicenter, retrospective, registry-based study assessing the occurrence of HA-VTE through hospital discharge among children <18 years old hospitalized in the PICU from January 2018 through December 2019 from 47 participating centers in the PHIS registry, we found the proportion of HA-VTE before discharge was higher among those with, versus without, MV exposure (4.4% versus 1.1%). After adjustment for known HA-VTE risk factors in multivariable logistic regression, MV remained an independent risk factor for HA-VTE (OR 2.51, 95% CI: 2.33–2.69).