Methods

Data source

We conducted a retrospective analysis of data from 2017 to 2022 using the National Inpatient Sample (NIS). The NIS, a core component of the Healthcare Cost and Utilization Project (HCUP) sponsored by the Agency for Healthcare Research and Quality (AHRQ), is the largest publicly available all-payer inpatient database in the United States. It is designed to generate nationally and regionally representative estimates of hospitalizations, health care utilization, clinical outcomes, and associated costs. It provides detailed discharge-level data, including patient demographics, primary payer, hospital characteristics, diagnoses, and procedures. The NIS design uses a stratified, probability sampling approach based on hospital ownership, bed size, teaching status, US census region, and urban-rural location. A 20% sample of hospitals is drawn from each stratum, and all discharges from sampled hospitals are included. Sampling weights are applied to produce nationally representative estimates. Data are derived from state-level HCUP partners, with recent editions incorporating submissions from up to 48 states and the District of Columbia, representing >97% of the US population. The dataset is limited to community hospitals, defined as non-federal, short-term, general, and specialty facilities. This robust sampling methodology enables comprehensive, population-based analyses of inpatient care in the United States. Methodological details are available at https://www.hcup-us.ahrq.gov/nisoverview.jsp.

Study population

We identified adult hospitalizations (≥18 years) between 2017 and 2022 from the NIS. We identified patients with hematologic malignancies using International Classification of Diseases, Tenth Revision (ICD-10) diagnosis codes for acute lymphoblastic leukemia (ALL), multiple myeloma (MM), and non-Hodgkin lymphoma (NHL) recorded in either the primary diagnosis field (DX1) or secondary diagnosis fields (DX2-DX40). To ensure mutually exclusive disease categories, we excluded hospitalizations with >1 hematologic malignancy code across diagnosis fields, as well as admissions with overlapping hematologic malignancy codes appearing simultaneously in both primary and secondary diagnosis positions. Among the remaining hospitalizations, we identified patients who underwent CAR T-cell therapy during the same admission using ICD-10 procedure codes recorded in the procedure fields (PR1-PR25). We included only admissions with a single qualifying hematologic malignancy diagnosis and a documented CAR T-cell procedure in the final analytical cohort and subsequently categorized the cohort into patients with and without disseminated intravascular coagulation (DIC).

We identified ALL using ICD-10 diagnosis code C91.0, MM using code C90, and NHL using codes C82, C83, C84, C85, C86, and C88. We identified CAR T-cell therapy administration using the following ICD-10 procedure codes: XW033C3, XW043C3, XW24376, XW23346, XW24346, XW23376, XW033C7, XW033G7, XW033H7, XW033J7, XW033K7, XW033M7, XW033N7, XW043C7, XW043G7, XW043H7, XW043J7, XW043K7, XW043M7, XW043N7, XW033A7, and XW043A7. We identified DIC using ICD-10 diagnosis code D65. Where available, we cited published validation studies reporting positive predictive value (PPV), sensitivity, and specificity for these ICD-10 codes. Validation metrics for each outcome definition are summarized in ¹.

Patient and hospitalization characteristics

Our study variables encompassed patient demographics (age, sex, and race/ethnicity), insurance, and socioeconomic characteristics (insurance type/primary payer: Medicare, Medicaid, private insurance, or self-pay; and median household income quartile based on the patient’s ZIP code), and hospital teaching status/location (teaching vs nonteaching). Clinical variables included medical comorbidities, as well as both primary and secondary outcomes. We used the Charlson Comorbidity Index (CCI) to assess and stratify patients according to their comorbidity burden. As no patients had a CCI score of 0 or 1, CCI was categorized as 2 and ≥3. The CCI is a validated and widely used tool for estimating 1-year mortality risk by assigning weighted scores to specific comorbid conditions based on their prognostic impact. These conditions include myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease, peptic ulcer disease, mild and moderate-to-severe liver disease, diabetes mellitus (with and without chronic complications), hemiplegia or paraplegia, moderate-to-severe renal disease, solid malignancies, leukemia, lymphoma, metastatic solid tumors, and HIV/AIDS. The cumulative score reflects the overall burden of comorbidity, with higher scores corresponding to increased risk of mortality and poorer clinical outcomes. In addition, we identified and included medical comorbidities such as sepsis, CRS, SARS-CoV-2 infection, hypertension, dyslipidemia, and atrial fibrillation, and treatments such as long-term antiplatelet and anticoagulant use based on ICD-10 codes, as detailed in ¹.

Outcomes measured

Our primary outcome was all-cause in-hospital mortality among patients with ALL, MM, and NHL who received CAR T-cell therapy, comparing those with DIC with those without DIC. Secondary outcomes were categorized into 3 principal domains. The first domain, organ dysfunction and critical care utilization, encompassed acute kidney injury (AKI), renal replacement therapy (RRT), respiratory failure (RF), mechanical ventilation use (MV), all-cause shock (shock), and vasopressor use. The second domain, thrombotic and hemorrhagic complications and interventions included acute venous thromboembolism (VTE)—comprising deep vein thrombosis and pulmonary embolism—acute coronary syndrome (ACS), gastrointestinal hemorrhage (GIH), nontraumatic intracranial hemorrhage (ICH), and transfusion requirements, such as red blood cell transfusion (RBC), platelet transfusion, and cryoprecipitate and fresh frozen plasma (FFP). The third domain, overall health care resource utilization, was assessed using length of hospital stay (LOS) and total hospitalization charges (TOTCHG).

Statistical analysis

All analyses accounted for the complex survey design of the NIS, including stratification, clustering, and weighting, using appropriate survey procedures. We compared continuous variables using independent Student’s t-test and categorical variables using Pearson chi-squared test. For outcome analyses, we applied logistic regression for binary outcomes and linear regression for continuous outcomes. To account for potential confounders, we performed multivariable regression analyses. Multivariable regression models were adjusted for the following patient-level confounders: age, sex, race/ethnicity, CCI, patient zip code income quartile, primary payer/insurance type, and comorbid conditions including sepsis, CRS, SARS-CoV-2, dyslipidemia, hypertension, atrial fibrillation, and long-term anticoagulant or antiplatelet use. These models were also adjusted for the hospital-level confounder, specifically, and hospital teaching status/location (rural, urban nonteaching, and urban teaching). We excluded patients with missing data for any of the following variables: age, sex, race/ethnicity, CCI, patient zip code income quartile, primary payer/insurance type, and hospital teaching status/location. Moreover, some covariate categories were automatically omitted in outcome-specific regression models due to sparse observations or perfect prediction. Details of omitted covariate categories are provided in ¹. CAR T-cell therapy hospitalizations were predominantly concentrated in urban teaching hospitals, resulting in limited variability in hospital teaching status/location, which was not retained in all models. The CCI was presented categorically for descriptive analyses but modeled as an ordinal variable in regression analyses to preserve information and improve statistical efficiency. Additionally, we constructed a second logistic regression model that included only variables associated with the outcome of interest in univariable analysis at a significance level of P < .20. Unless otherwise specified, primary results are reported from the fully adjusted multivariable regression models. We conducted all statistical analyses using STATA version 18.0 Basic Edition (StataCorp) and considered 2-tailed P values of <.05 to be statistically significant.

Footnotes

1. The online version contains supplementary material available at https://doi.org/10.1016/j.rpth.2026.106886. ↩ ↩² ↩³