Association of pulmonary artery catheter with in-hospital outcomes after cardiac surgery in the United States: National Inpatient Sample 1999–2019 – Scientific Reports

Data source

Secondary analyses of existing data from the Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) NIS were performed for the time-period of 1999–2019. The NIS is the largest publicly available, all-payer inpatient care database of community (non-federal) hospitals in the U.S. It consists of 5–8 million hospital discharge records sampled from 1000 hospitals on an annual basis since 1988. Each year, a 20% stratified probability sample of hospitals (before 2012) or hospital discharge records (since 2012) were selected from all participating HCUP states, stratified by bed size, teaching status, urbanicity, and region. Within the NIS database, hospital discharge weights were provided that can be used to generate national estimates, for all years combined, and to examine trends over time taking into consideration sampling design changes in 2012. The NIS data elements included patient demographics, up to 15 diagnoses and 15 procedures, as well as hospital course and outcomes. All methods were carried out in accordance with relevant guidelines and regulations. The study was determined to be research not involving human subjects at Fort Belvoir Community Hospital. Since no experiments on humans and/or use of human tissue were performed as part of this study, a waiver of institutional review board approval was granted at Fort Belvoir Community Hospital. Due to the nature of the research study, informed consent was waived by the Institutional Review Board of Fort Belvoir Community Hospital. Supplemental Digital Content 1 provides a listing of all ICD-9-CM/ICD-10 diagnostic and procedure codes applied to define eligibility criteria and variables of interest. Of note, ICD-9-CM codes were used for the years 1999–2014 and the first three quarters of the year 2015. By contrast, ICD-10 codes were used for the fourth quarter of the year 2015 and the years 2016–2019.

Eligibility criteria

Variables pertaining to patient and hospital characteristics, diagnostic and procedure codes, and outcomes were defined using data elements from the NIS core and hospital databases, after applying pre-defined inclusion and exclusion criteria. The study population met the following inclusion criteria: (1) Age ≥ 18 years; (2) At least one of the first 15 procedure variables (PR1-PR15) comprise an ICD-9-CM/ICD-10 code that corresponds to cardiac surgery. A comprehensive list of ICD-9-CM/ICD-10 procedure codes that correspond to cardiac surgery was compiled and hospital discharge records were selected if the patient had at least one type of cardiac surgery. After excluding hospitals with < 50 cardiac surgeries performed overall between 1999 and 2019 and those with zero PACs per year, patients were excluded if they met at least one of the following criteria: (1) < 5 procedure codes; (2) missing data on patient and hospital characteristics described below. We chose to exclude hospital records with less than 5 procedure codes to maximize the total sample size while minimizing the likelihood of misclassifying patients who underwent multiple types of cardiac surgery. As show in Fig. 1, 132,916,882 of 158,971,760 1999–2019 NIS records corresponded to patients ≥ 18 years of age, and of those 1,477,041 records involved cardiac surgeries. After applying exclusion criteria, 969,034 records (68% male, mean age: 65 years) were kept (92,159 PAC recipients and 876,875 PAC non-recipients), of which 323,929 corresponded to patients that had one or more of the subgroup characteristics (32,539 PAC recipients and 291,390 PAC non-recipients) and 645,105 corresponded to patients who did not have any of the subgroup characteristics (59,620 PAC recipients and 585,485 PAC non-recipients).

Patient characteristics

Patient-level characteristics were defined as age (in ‘years’), sex (“Male”, “Female”), race/ethnicity (“White”, “African American”, “Hispanic”, “Other”), Charlson comorbidity index [CCI] (“0”, “1”, “2+”), elective admissions (“Yes”, “No”), admission quarter (“1st quarter”, “2nd quarter”, “3rd quarter”, “4th quarter”), weekend admission status (“Monday–Friday”, “Saturday–Sunday”) and primary payer (“Medicare”, “Medicaid”, “Private insurance”, “Self-pay”, “No pay”, “Other”). The CCI score reflects the cumulative increase in likelihood of one-year mortality due to the severity of the effect of comorbidities. In this study, the CCI was calculated using 15 ICD-9-CM/ICD-10 diagnostic codes by the Stata command charlson. Alternatively, the CCI was replaced with dichotomous (“Yes”, “No”) variables based on ICD-9-CM/ICD-10 diagnostic codes, representing each of the following comorbidities that are relevant to cardiovascular surgery patients: congestive heart failure, myocardial infarction, aortic valve disease, mitral valve disease, tricuspid valve disease, pulmonary valve disease, respiratory failure, chronic obstructive pulmonary disease, pulmonary hypertension, hypertension, pneumonia, atherosclerotic disease, stroke, diabetes, cancer, and chronic liver disease.

Hospital characteristics

Hospital-level characteristics were defined as hospital region (“Northeast”, “Midwest”, “South”, “West”), hospital control (“Government or Private”, “Government, non-federal”, “Private, not-for-profit”, “Private, investor-owned”, “Private”), hospital location and teaching status (“Rural”, “Urban—Non-Teaching”, “Urban—Teaching”) and hospital bed size (“Small”, “Medium”, “Large”).

Selected subgroups

ICD-9-CM/ICD-10 diagnostic and procedure codes were used to define specific subgroups: “heart failure”, “pulmonary hypertension”, “mitral or tricuspid valve disease” and “combined surgery”. The “combined surgery” subgroup consists of hospital records whereby at least two of the following procedures were applied: Coronary Artery Bypass Grafting (CABG), aortic valve surgery, mitral valve surgery, tricuspid valve surgery, pulmonary valve surgery. Furthermore, hospital discharge records were defined as “Any” versus “None” based on presence or absence of at least one of these characteristics.

PAC receipt

PAC receipt was defined as the study exposure of interest. A comprehensive list of ICD-9-CM/ICD-10 procedure codes that correspond to PAC receipt were generated for eligible hospitalizations in the 1999–2019 NIS database. These codes were obtained from primary and secondary procedure variables and, as such, hospital discharge records were classified into two categories: (1) PAC recipient; (2) PAC non-recipient. Furthermore, hospital-level PAC rates were calculated and categorized into quartiles for the purpose of sensitivity analyses.

Clinical outcomes

Data were extracted on multiple study outcomes, including in-hospital mortality (“deceased” vs. “alive”) and in-hospital LOS (days). Total LOS was defined as a continuous variable in the context of regression modeling and categorized as “≥ 7 days” vs. “< 7 days” in the context of regression and causal modeling.

Statistical analysis

All statistical analyses were conducted using Stata version 17 (StataCorp, College Station, TX), taking into consideration complex sampling design as well as specific recommendations¹⁶. Descriptive statistics included mean (± standard error) for continuous variables and frequencies with percentages for categorical variables. Bivariate associations were examined using uncorrected Chi-square and design-based F-tests, as appropriate. Linear and binary logistic regression models were constructed to estimate crude and adjusted beta coefficients as well as odds ratios (cOR and aOR) with their 95% confidence intervals (CI) for exposure variables as predictors of the selected health outcomes. Risk-adjustment and targeted maximum likelihood estimation (TMLE) with Super Learner algorithms were performed when comparing recipients and non-recipients of PAC on health outcomes, overall, as well as according to pre-specified subgroups. Multivariable models were adjusted for age, sex, race/ethnicity, CCI, elective admissions, admission quarter, weekend admission status, primary payer, hospital region, hospital control, hospital location and teaching status as well as hospital bed size. Sensitivity analyses were performed whereby CCI was replaced with the comorbidities described above as covariates in overall risk-adjusted models. Because regression methods are frequently biased if outcome models are mis-specified, causal inference methods incorporating propensity scores, the G-formula or TMLE are preferred¹⁷. Although propensity score methods necessitate exposure models to be correctly specified, double-robust methods such as TMLE require correct specification of either outcome or exposure models¹⁷. TMLE is a semiparametric estimator allowing use of machine learning algorithms to minimize model misspecification¹⁷. Unlike TMLE, classical regression methods for estimating the average treatment effect (ATE), or risk difference, assume that ATE is constant across confounder levels with no effect modification. We applied the eltmle package in Stata, while using Super Learner with tenfold cross-validation to evaluate the predictive performance for potential outcomes and weighted averages as a propensity score for distinct machine learning algorithms. The default Super Learner machine learning algorithm was applied as previously defined in an R v.1.2.0-5 package: (1) stepwise selection, (2) generalized linear modeling (GLM), (3) GLM variant that includes second order polynomials and two-by-two interactions of main terms included in the model. The ATE, causal risk ratio (CRR) and marginal odds ratio (MOR) were estimated with 95% CI for each hypothesized relationship using TMLE^{17,18,19,20,21}. Using risk-adjustment and TMLE analyses, we performed sensitivity analyses whereby the overall relationship between PAC use and the outcomes of interest were examined within quartiles of hospital-level PAC rates. Complete subject analyses were performed after examination of patterns of missingness. Two-sided statistical tests were conducted and P < 0.05 were considered statistically significant.

Ethical approval

Since the project was determined to be research not involving human subjects, a waiver of institutional review board approval was granted at Fort Belvoir Community Hospital. Due to the nature of the research study, informed consent was waived by Institutional Review Board of Fort Belvoir Community Hospital. The project adhered to relevant ethical guidelines/regulations in accordance with the Declaration of Helsinki.