br Study population br All patients
All patients age 18 years old who sustained IHCA during an index admission were analyzed. Cardiac arrests were identified using ICD-9 codes 427.41 or 427.5 (ventricular fibrillation and cardiac arrest, respectively). Since DX1 represents the diagnosis code for the primary reason for hospitalization, only those cases that had
ICD-9 codes of 427.41 or 427.5 ascribed as a secondary diagnosis were included in the analysis. Conversely, if either code appeared as a primary diagnosis, these were considered to be hospitalizations for an out-of-hospital cardiac arrest and were therefore excluded from the analysis.13–15
Discharge diagnoses and procedures were recoded using the Clinical Classification of Diseases Software (“DXCCS” ) into broad categories, available as separate variables within NIS. We identified cancer hospitalizations using ICD-9 and DXCCS codes (DXCCS 1–
30) or the presence of an indicator of cancer in the comorbid condition files. DXCCS codes indicating cancer are 11–45. The comorbidity file included in the NIS lists 29 comorbidities (also known as Elixhauser’s Comorbidity measures) based on ICD-9 CM diagnoses and the diagnosis-related group in effect on the discharge date. These comorbidities are not directly related to the principal diagnosis or the
main reason for admission and are likely to have originated before the hospital stay.16 In 2015, the Healthcare Cost and Utilization Project (HCUP) State Inpatient Database was used to create two indices
based on 29 co-morbidity measures designed to predict in-hospital mortality and 30-day readmission.17 Those indices were calculated for our cohort as well.
For this BMS 986120 analysis, all hospitalizations that lacked an ICD-9 or DXCCS code indicating a diagnosis of cancer were considered non-cancer hospitalizations. Admissions with metastatic cancer were excluded due to the metabolic derangements and systemic inflamma-
tion associated with advanced malignancy, and the fact that such diagnoses tend to portend limited survival.18,19
Data elements utilized from NIS were medical comorbidities, demographic characteristics (age, sex, and race), income quartile, insurance status, hospital level characteristics, and procedures performed. The procedures of interest were cardiopulmonary resuscitation (CPR), coronary angiography and percutaneous coro-nary intervention (PCI), intra-aortic balloon pump (IABP) placement, defibrillator (ICD) implantation, and targeted temperature manage-ment (TTM). Procedural utilization was identified using ICD-9 CM procedures codes, as listed in Supplemental Table 1.
The primary outcome of this study was IHCA survival rates. Secondary outcomes included utilization of aforementioned post-arrest procedures, length of stay, discharge disposition, and adjusted cost of hospitalization. The adjusted cost was obtained by multiplying hospital charges with the cost-to-charge ratios and wage index for each hospital for each year, then adjusting for inflation to 2017 dollars.19,20 The wage index helps correct for geographic variations in costs among hospitals.
We utilized propensity score matching to assess the impact of a cancer diagnosis on our stated primary and secondary endpoints. Three propensity-matched cohorts, using various matching schemes, were employed to study our specific outcomes. We applied the 8 !1 Digit Match algorithm, which matched a case to a control whose score equaled the 8th decimal point followed by 7th decimal point followed by 6th decimal point and so on using a greedy matching algorithm.21 This schema afforded 2:1 matching of non-cancer to cancer admissions with IHCA in all three cohorts. Complete details of propensity matching are provided in
Supplemental methods. The consort diagram of the cohorts is detailed in Supplemental Fig. 1.
Sample weight, stratification, cluster and domain information included in the NIS were used to derive national estimates from sample data. Continuous variables were analyzed using Student’s t-test if organs were normally distributed, as determined by the Anderson–Darling test. Non-parametric continuous variables were analyzed using the Kruskal–Wallis test. Categorical variables were analyzed using Chi-Square test. Trends were evaluated using the Cochrane Armitage test and linear regression models for categorical and continuous variables, respectively. Hospitalization cost was logarithmically transformed to yield normally-distributed values for analysis.
Moreover, in order to better understand if procedure utilization differed based on general prognosis, a sensitivity analysis was performed on a subgroup of subjects with cancers that portended a relatively favorable survival. This was defined as a 5-year survival >90% and included subjects with non-advanced thyroid, breast, prostate, and testicular cancers, or non-Hodgkin lymphomas.22 Another sensitivity analysis was undertaken to classify a sicker group of patients in the ICU. However, since NIS does not have a dedicated ICU field, a sub-cohort of patients was created with the All Patient Refined Diagnosis Related Groups (APR-DRGs) severity variable.23 APR-DRG severity is classified based on the severity of illness and the risk of mortality into 4 subclasses, namely, minor (1), moderate (2), major (3) and extreme (4). APRDRG of 3 and 4 was used to create a group of these possible ICU patients.