A New Risk Assessment Model for Hospital-Acquired Venous Thromboembolism in Critically Ill Children: A Report from the CHAT Consortium

Abstract

Objective:

To create a risk model for hospital acquired venous thromboembolism (HA-VTE) in critically ill children upon admission to an intensive care unit (ICU).

Design:

Case-control study.

Setting:

Intensive care units from eight children’s hospitals throughout the United States.

Patients:

Critically ill children with HA-VTE (cases) aged 0-21 years and similar children without HA-VTE (controls) from January 2012 to December 2016. Children with a recent cardiac surgery, asymptomatic VTE or a VTE diagnosed before ICU admission were excluded.

Interventions:

None.

Measurements:

The multi-institutional Children’s Hospital Acquired Thrombosis (CHAT) Registry was used to identify cases and controls. Multivariable logistic regression was used to determine the association between HA-VTE and putative risk factors present at or within 24 hours of ICU admission to develop the final model.

Main Results:

A total of 548 HA-VTE cases (median age 0.8 years, interquartile range [IQR]=0.1-10.2) and 187 controls (median age 2.4 years, IQR=0.2-8.3) were analyzed. In the multivariable model, recent central venous catheter placement (odds ratio [OR] 4.4, 95% confidence interval [CI] =2.7-7.1), immobility (OR 3.6, 95% CI=2.1-6.2), congenital heart disease (OR 2.9, 95% CI=1.7-4.7), length of hospital stay prior to ICU admission ≥3 days (OR 2.5, 95% CI=1.1-5.6), and history of autoimmune/inflammatory condition or current infection (OR 2.1, 95% CI=1.2-3.4) were each independently associated with HA-VTE. The risk model had an area under the receiver operating characteristic curve of 0.79 (95% CI=0.73-0.84).

Conclusions:

Using the multicenter CHAT Registry, we identified five independent risk factors for HA-VTE in critically ill children, deriving a new HA-VTE risk assessment model. A prospective validation study is underway to define a high-risk group for risk-stratified interventional trials investigating the efficacy and safety of prophylactic anticoagulation in critically ill children.

INTRODUCTION

Pharmacological prophylaxis is the standard of care for hospitalized adults at high-risk for venous thromboembolism (VTE), but at low-risk of bleeding, based on findings from randomized controlled trials (RCTs) ^1,2. These RCTs, and their application in hospital settings, have typically relied upon risk stratification criteria, which are derived from hospital acquired VTE (HA-VTE) risk assessment models (RAMs) ^3-5. A HA-VTE RAM in particular assists in targeting high-risk patients for VTE prevention measures such as early ambulation, mechanical prophylaxis or pharmacological prophylaxis.

In hospitalized children, there is a paucity of both RCTs and the RAMs to inform the design of such VTE prevention measures. Progress in this area has been limited by the lower incidence of HA-VTE in children than in adults, and also by the scarcity of multicenter cooperative efforts to develop and validate pediatric-specific HA-VTE RAMs ^6,7. Although the overall incidence is low, pediatric HA-VTE has been identified as the second leading cause of preventable harm in hospitalized children ⁸. In recent years, evidence has accumulated that critically ill children are at especially high risk for HA-VTE, with an estimated incidence of 2% in the general (non-cardiac post-operative) pediatric intensive care unit (ICU) population and up to 18% with a central venous catheter (CVC) ^9-14. Single-institutional studies have identified immobility, inflammatory conditions, presence of a CVC and a prolonged hospital stay to be specific HA-VTE risk factors in critically ill children ^9,11,12.

Pharmacological prophylaxis is routinely given to critically ill adult patients without implementing a RAM due to their high HA-VTE risk, but for critically ill children with a lower HA-VTE risk, not all would benefit from prophylaxis. In order to limit the risk of bleeding in critically ill children from unneeded pharmacological prophylaxis, HA-VTE RAMs specific for children admitted to the ICU are needed. The Children’s Hospital Acquired Thrombosis (CHAT) Consortium was established to address this need for multicenter efforts to identify high-risk variables and create pediatric-specific RAMs in order to conduct risk-stratified interventional trials of HA-VTE prevention in children. Due to critically ill children having higher HA-VTE risk than non-critically ill hospitalized children, and potentially distinct HA-VTE risk factors, two pediatric RAMs have been developed ¹⁵. Accordingly, the objective of this analysis was to derive a HA-VTE RAM specifically applicable to critically ill children, using the CHAT Registry.

MATERIALS AND METHODS

Study design

We conducted a case-control study of critically ill participants within the CHAT Registry. The institutional review board reviewed and approved the study, and granted a waiver of consent, at each participating hospital (Children’s Hospital Los Angeles Institutional Review Board approval CHLA-14-00388). Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) reporting guidelines were followed in the drafting of this manuscript ¹⁶.

Registry and study participants

Full methodological details of the CHAT Registry have been previously published ¹⁰. In brief, the CHAT Registry consists of participants from eight hospitals around the U.S. (Akron Children’s Hospital, Boston Children’s Hospital, Children’s Hospital Colorado, Children’s Hospital Los Angeles, Children’s Mercy Kansas City, CHOC Children’s Hospital, Children’s Hospital of Pittsburgh and Indiana Hemophilia and Thrombosis Center/Peyton Manning Children’s Hospital) aged 0-21 years-old diagnosed with an imaging confirmed HA-VTE as well as hospitalized non-HA-VTE controls. Cases in the Registry were defined as consecutive participants admitted to one of the study institutions without symptoms of a VTE and were later diagnosed with a VTE confirmed by radiology imaging, which consisted of Doppler ultrasonography, computed tomography scan, venography, echocardiogram or magnetic resonance imaging. Control participants in the CHAT Registry were hospitalized patients without a diagnosis of VTE on admission and throughout their hospitalization and were matched by institution and admission year to HA-VTE cases.

Unmatched HA-VTE cases and non-VTE controls within the Registry who were admitted or transferred to an ICU (defined as any unit that cared for critically ill children—e.g., pediatric ICU, cardiac ICU, neonatal ICU) at any point during their hospitalization from January 1, 2012 through December 31, 2016 were eligible for the present analysis. Cases and controls were excluded if they had a cardiac surgery within two weeks prior to ICU admission/transfer. Cases were excluded if they had an asymptomatic VTE identified during radiographical imaging for other reasons or if their VTE was diagnosed prior to ICU admission/transfer. Signs and symptoms of VTE used to define symptomatic VTE included: painful limb swelling or change in limb temperature or color (in the presence of deep venous thrombosis of the limb); hypoxemia or shortness of breath (in the presence of pulmonary embolism); unexplained fever or thrombocytopenia, CVC malfunction or occlusion requiring local instillation of tissue plasminogen activator (tPA) into a CVC (in the presence of CVC-related venous thrombosis); change in mental status, headache and/or vomiting (in the presence of cerebral sinovenous thrombosis); and face/neck swelling, dyspnea, headache, upper extremity swelling and/or distended chest veins (in the presence of superior vena cava thrombosis leading to superior vena cava syndrome).

Data collection, management and quality assurance

Data collected for both HA-VTE cases and controls included demographics, past medical history, clinical management and laboratory result details, such as placement of a central venous catheter, surgeries and other procedures, complete blood count values, infectious disease diagnoses, medications and mobility status using the Braden Q mobility score ¹⁷ based on previous studies evaluating VTE risk in children ^{9,11,13,14,18-21}. For HA-VTE cases, additional details regarding the anatomic distribution and treatment of VTE were also collected. A full description of data elements within the CHAT Registry has been published previously ¹⁰.

To serve as putative risk factors, characteristics must have been present prior to, or within 24 hours of, ICU admission/transfer. Obesity was defined as BMI-for-age-weight that is ≥95% percentile per the Centers for Disease Control and Prevention ²². A platelet count >350 K/uL was used as a marker of VTE risk based on previous work from the CHAT Consortium group ¹⁵. In order to standardize mobility and quantitatively asses this variable, the mobility portion of the Braden Q score was used. Immobility was defined as a Braden Q mobility score of 1 or 2 (completely immobile or very limited immobility) ¹⁷. Intravascular procedures included cardiac catheterization, dialysis, stent placement, coiling and plasmapheresis. Autoimmune/inflammatory disorder included inflammatory bowel disease, juvenile rheumatoid arthritis, lupus and autoimmune/inflammatory disorder not otherwise specified based on participant past medical history. Infection included bacteremia/sepsis, infections of the gastrointestinal tract, meningitis, osteomyelitis, pneumonia, upper respiratory tract, and urinary tract infections. A high-risk condition was defined as protein losing enteropathy, hemoglobinopathies, chronic use of parental nutrition, personal history of thrombophilia (inherited and acquired), or past personal history of VTE. For a recently placed CVC to be included in the analysis, the CVC had to be placed 30 days prior or within 24 hours of ICU admission/transfer.

Data were collected using standardized electronic case report forms (eCRFs) within Research Electronic Data Capture (REDCap) ^23,24. Access to the eCRFs were provided to each participating center along with detailed data dictionaries, in order to ensure reproducible data collection. Automated monthly data monitoring was in place to identify missing data fields, outliers and incorrect dates and sent to each site’s research staff and principle investigator for review and verification.

Statistical Analysis

Patient demographic, clinical, and past medical history were summarized across case-control status using median and interquartile range (IQR) for continuous variables and counts and percentages for categorical variables. Associations between variables and case-control status were assessed by univariate and multivariable logistic regression models. Variables with a p-value ≤0.10 in the univariate analyses were included in a multivariable model for further evaluation, according to our a priori analysis plan. Variables with p-values >0.05 in the multivariable models were removed from the models. Thus, the final models included only variables with a p-value <0.05 from the adjusted models. A complete set analysis included participants that had information on all variables in the model. Due to high frequency of missing values (assumed to be missing at random) for some variables (platelet count, Braden Q mobility score, recent hospitalization), the multivariable analyses were conducted with the multiple imputation method, which became the primary analysis set, in addition to complete-case analysis (excluded observations with missing values). Twenty imputed datasets were created using fully conditional specification (due to non-monotone missing data). Predictive mean matching was used to impute missing platelet count (19% missing) and logistic regression was used for Braden Q mobility score (46% missing) and recent hospitalization (2% missing), with all other variables as predictors in the model (primary analytical dataset). Model discrimination was assessed using the area under the receiver operating characteristic (AUROC) curve with the corresponding 95% confidence interval (CI). Model calibration was assessed by comparing the expected number of HA-VTE cases calculated from the RAM with the observed number of HA-VTE cases using deciles. The number of eligible participants in the CHAT Registry determined the sample size for the study. All analyses were conducted using SAS version 9.4 (Cary, NC).

RESULTS

Characteristics of Participants

The CHAT Registry contained 2111 participants (1164 HA-VTE cases and 947 controls). For risk variable analysis and CHAT-ICU RAM development (i.e., the present analysis), 548 HA-VTE cases and 187 controls were eligible for inclusion during the January 2012 to December 2016 study period (Figure 1).

Figure 1.

Participant inclusion and exclusion justification from the Children’s Hospital Acquired Thrombosis Registry for risk variable analysis and risk assessment model development.

*HA-VTE, Hospital acquired venous thromboembolism; ICU, intensive care unit

The demographic and clinical characteristics of the study population are presented in Supplemental Table 1, along with odds ratios, 95% CIs and p-values from univariate analyses. The median age of HA-VTE cases was 0.8 years [interquartile range (IQR)=0.1-10.2 years] compared to a median age of 2.4 years (IQR 0.2-8.3 years) for control subjects. The distribution of males was 58% (n=318) among cases and 50% (n=94) for control subjects. The most common past medical history was congenital heart disease for all participants. The median time from hospital admission to HA-VTE diagnosis for cases was 10 days (IQR =5-25 days), and the median time from ICU admission/transfer to HA-VTE diagnosis was 8 days (IQR =4-21 days).

HA-VTE risk factor assessment and RAM development

Twenty-six factors were assessed as potential predictive variables for HA-VTE, which included participant demographics, clinical factors during admission (e.g. Braden Q mobility score, extracorporeal membrane oxygenation), and past medical history (e.g. history of cancer, congenital heart disease). As shown in Supplemental Table 1, 11 putative predictive variables had unadjusted p-values ≤0.10, including: male sex; race; immobility (Braden Q mobility score of 1 or 2); length of hospital stay prior to ICU admission/transfer ≥3 days; hospitalization 30 days prior to current hospital admission; procedure with intravascular instrument within 30 days; a recently placed CVC; history of congenital heart disease; history of autoimmune/inflammatory disorder or infection; mechanical ventilation; and presence of high-risk conditions such as protein losing enteropathy, hemoglobinopathies, chronic use of parental nutrition, thrombophilia (inherited and acquired), or past personal history of VTE. Race met the criterion for inclusion in multivariate analyses, but was not further considered due to a high proportion of missing/unknown data.

The results of multivariable analysis are shown in Table 1. The final CHAT-ICU RAM using both complete-case and multiple imputation analysis included only variables with p-value <0.05 in the multivariable models. In the complete case (i.e., no missing data) analysis, the final RAM included four variables that significantly increased the risk of HA-VTE with point estimates ranging from an OR of 2.48 (95% CI=1.24-4.94) for history of autoimmune/inflammatory disorder/infection to an OR of 4.54 (95% CI=2.45-9.17) for a CVC placed within 30 days prior to or within 24 hours of ICU admission. Risk estimates remained similar with imputed analysis (final model), (i.e., including cases with missing data); however, length of stay prior to ICU admission of ≥3 days was an additional statistically-significant risk factor (OR 2.48, 95% CI=1.10-5.62). The AUROC curve was 0.78 (95% CI=0.73-0.84) for the complete-case model and 0.79 (95% CI=0.73-0.84) for the imputed model (Figure 2).

Table 1.

Final Children’s Hospital-Acquired Thrombosis (CHAT) intensive care unit risk assessment model.

Variable	Complete Case (N=395)*		Imputed (N=735)*
	OR (95% CI)	p-value	OR (95% CI)	p-value
Braden Q score ≤2 within 24 hours of ICU admission (reference= slight/no limitations)	3.40 (1.92-6.04)	<0.001	3.65 (2.14-6.24)	<0.001
Length of stay prior to ICU admission ≥3 days	Not Applicable		2.48 (1.10-5.62)	0.03
Central venous catheter placed 30 days prior to or on ICU admission	4.54 (2.45-9.17)	<0.001	4.37 (2.69-7.07)	<0.001
Past medical history of congenital heart disease	2.95 (1.47-5.92)	0.002	2.87 (1.74-4.73)	<0.001
Past history of autoimmune/inflammatory disorder or infection during admission	2.48 (1.24-4.94)	0.01	2.06 (1.23-3.44)	0.01

^*Complete case analysis included participants with no missing values for all variables in the model; imputed analysis included all participants in the original data set (N 735) with missing information on all variables imputedNote: Area under the curve (AUC) was 0.78 (95% CI= 0.73-0.84) for the complete-case and 0.79 (95% CI= 0.73-0.84) for the imputed model.

Figure 2a and 2b.

Children’s Hospital Acquired Thrombosis Registry-intensive care unit risk assessment model receiver operating characteristic curve for the final model (a) and imputed final model (b).

DISCUSSION

Critically ill children are at a particularly high risk of HA-VTE. Few single-institutional retrospectively derived RAMs have been reported for this at-risk pediatric patient population. Using data from almost 550 critically ill pediatric HA-VTE cases and over 180 critically ill controls from the CHAT Registry, we identified CVC, immobility, congenital heart disease, length of hospital stay ≥3 days prior to the ICU admission/transfer and autoimmune or inflammatory disorder or a current infection as independent risk factors for HA-VTE in critically ill children, giving rise to a RAM with good discriminatory accuracy (AUROC, 0.79).

Several prior retrospective case-control studies have reported pediatric HA-VTE RAMs in either non-critically ill or non-selective hospitalized pediatric populations ^{11,14,19,25,26}. These studies have been reviewed in a recent meta-analysis ¹¹ and generally have identified similar risk factors to those identified in the present multicenter analysis in critically ill children as well as prior single-center studies in critically ill children such as inflammatory conditions, the presence of a CVC and congenital heart disease.

Our work substantiates that of the other two previous retrospective case-control studies evaluating risk of HA-VTE in critically ill children by Arlikar, Goldenberg and colleagues at Johns Hopkins All Children’s in Florida ⁹ and by Branchford, Goldenberg and co-workers at Children’s Hospital Colorado ¹⁹. Both studies identified similar independent risk factors including serious infection or systemic inflammatory condition, presence of a CVC, and prolonged hospitalization. The Arlikar study ⁹ identified mechanical ventilation as an additional risk factor, which did not remain a significant independent risk factor after multivariate adjustment in the present work consistent with the Branchford report ¹⁹. However, the present analysis of the CHAT Registry extends this prior work, via a multicenter retrospective case-control study in which immobility was also able to be discerned as an independent HA-VTE risk factor.

In some of the non-ICU-selective studies, estrogen exposure (from oral contraceptive pills) has been identified as a pediatric HA-VTE risk factor among females. The relatively younger age of our critically ill study population may, in part, explain this difference; alternatively, or in addition, the potency of other risk factors in critically ill children (e.g. infection/systemic inflammatory conditions) may render the influence of estrogen on HA-VTE risk relatively less substantive.

HA-VTE risk factors identified in critically ill adult medical and surgical patients are similar to the high-risk variables identified in our study, including immobility, mechanical ventilation and CVCs ^27,28. Laboratory based variables, such as a low hemoglobin and elevated platelet count have also been identified as high-risk variables in critically ill adults. Although we did not find thrombocytosis as a significant VTE risk factor for our CHAT-ICU RAM, our RAM developed for children who are both critically ill and non-critically ill did find this variable as a significant risk factor ¹⁵. The lack of significance for an elevated platelet count may have been due to the variable missing in 19% of the participants. A posthoc power analysis revealed there was an 87% power to detect an OR of 0.71 for platelet count.

Our study contains a few noteworthy limitations. First, our RAM was derived in critically ill children but there was no differentiation among those admitted to a pediatric, cardiac or neonatal ICU. This study design was intentional in order to recognize the variations across hospitals nationally and world-wide in regard to a single ICU for all pediatric patients versus multiple critical care units in which children are cohorted by neonatal age (i.e., neonatal ICU), congenital and acquired cardiac disease (including perioperative management [i.e., cardiac ICU]), and all other (i.e., pediatric ICU). While this renders our findings more generalizable, it may mean that subpopulation-specific HA-VTE risk factors may not have been discerned. For this reason, the recent study by Atchison, Goldenberg and co-workers in critically ill children in a cardiac ICU setting ¹² is worthy of consideration in tandem with the present findings for critically ill children more generally. Nonetheless, the present work did exclude participants who had a cardiothoracic surgery within two weeks prior to ICU admission/transfer from our RAM development, but not those who may have undergone such surgery during their ICU stay. Second, we do not have the ability to detect risk factors for asymptomatic VTE within the present study design; however, recent studies have questioned the clinical significance of asymptomatic VTE ^14,29.

Another limitation are differences in variable definition when comparing variables used to create the previous CHAT RAM. For instance, immobility in the original RAM was based on a low Braden Q score within the first 24 hours of hospital admission, compared to the first 24 hours of ICU admission for the current RAM. In addition, age was a categorical variable in the original RAM compared to a continuous variable in the ICU-RAM. To evaluate variables included only in the original RAM with our current RAM, a posthoc power analysis was performed and revealed an 87% power to detect an OR of 0.71 for platelet count and >90% power to detect an OR of 0.73 for surgery (observed in the present study) at the two-sided alpha level 0.05. However, for steroid and past medical history of cancer there was a 9% and 22% power to detect ORs of 1.06 and 1.18 respectively (observed in the present study). Thus power could potentially be an issue for steroid and cancer in the present analysis. Lastly, the present RAM was developed via a retrospective case-control design with missing data points, such as Braden Q mobility scores and platelet counts, and hence warrants prospective validation. Accordingly, the present authors and CHAT Consortium colleagues have launched a multicenter prospective cohort study in critically ill children, for independent validation and optimization of the present CHAT-ICU RAM.

CONCLUSIONS

In conclusion, hospitalized, critically ill children have been previously identified as having an increased risk of VTE. The present work addresses a critical knowledge gap in regard to identifying risk factors for HA-VTE in critically ill children, as a basis for future risk-stratified interventional trials of HA-VTE prevention in this population. Through this analysis of the multicenter CHAT Registry, we have developed a RAM comprised of five statistically-significant risk factors for HA-VTE in critically ill children, with favorable preliminary discriminatory performance. Substantiation and optimization of the CHAT-ICU RAM is underway via a multicenter prospective cohort study recently launched through the CHAT Consortium.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)

Supplemental Table 1. Demographics and venous thromboembolism risk factors assessed.