Clinical and economic benefits of advanced microplegia delivery system in cardiac surgery: evidence from 250 hospitals

The Premier Healthcare Database was used as the data source for this study. This database contains data from more than 580 million patient visits, or one in every five discharges in the nation [7]. The Premier database contains data from standard hospital discharge files, including a patient's demographic and disease state, and information on billed services, including medications, laboratory, diagnostics and therapeutic services in de-identified patient daily service records. In addition, information on hospital characteristics, including geographic location, bed size and teaching status, is also available. Comparisons between patient and hospital characteristics for the hospitals that submit data to Premier and those of the probability sample of hospitals and patients selected for the National Hospital Discharge Survey suggested that the patient populations are similar with regard to patient age, gender, length of stay (LOS),  mortality, primary discharge diagnosis and primary procedure groups [8].

All data used to perform this analysis were de-identified and accessed in compliance with the Health Insurance Portability and Accountability Act. As a retrospective analysis of a de-identified database, the research was exempt from IRB review under 45 CFR 46.101(b)(4).

The study population for this analysis was any inpatient hospital visit in the Premier database from 1 January 2009 through 31 March 2015 with a record of an International Classification of Diseases (ICD-9) primary procedure code for one or more of the following cardiac surgeries: coronary artery bypass graft (CABG), aortic valve replacement (AVR) and mitral valve replacement (MVR) surgeries. CABG, AVR and MVR were analyzed together, as these procedures are the principal indications for use of cardiopulmonary bypass and cardioplegia in adult patients [9]. See Supplementary File A for a listing of all ICD-9 procedure codes of interest. It was possible for patient visits to have more than one procedure at a time; the first procedure code of interest was taken for each patient.

Patient visits meeting this inclusion criteria were then divided into two separate cohorts: those patient visits at hospitals that employ the MPS2 system (treatment) versus those that do not (control). For ‘treatment’ hospitals, all patient visits were further subdivided into visits that occurred before and after the corresponding hospital employed the MPS2 system. Patient visits were excluded from ‘treatment’ hospitals if they occur during a 3 month ‘wash out’ period when a hospital was initiating the MPS2 system into their cardiac surgical protocol.

The unit of analysis for this research was the inpatient visit. This database does not allow for longitudinal follow-up of patients unless the visits occurred at the same hospital. Both clinical (composite adverse event [AE] end point) and economic outcomes (cost and LOS) were analyzed.

Clinical events were identified as an AE if there was a record of the event during the hospitalization and the event was not recorded as present on admission. The composite AE outcome was defined as one or more of the following events recorded using ICD-9 diagnosis codes: major cardiac or renal event (including: acute kidney injury with and without dialysis, acute myocardial infarction, angina, stent occlusion/thrombosis, stroke and transient ischemic attack), sepsis or other infection, wound complication, abdominal complication, pulmonary complication, cardiogenic shock, bleeding or death. A complete variable listing with corresponding ICD-9 diagnosis codes for each clinical event included in the AE composite outcome is located in Supplementary File B. A composite clinical outcome was chosen as the primary safety end point for this analysis as a patient-level measure of the serious AE rate. Combining events into a single patient-level end point helps bolster the statistical power for demonstrating any significant difference in the overall safety signal between cohorts.

Economic outcome variables included: total cost in dollars to the hospital for the inpatient visit of interest, total cost in dollars to the hospital for all medications administered during the inpatient visit of interest LOS in days and days spent in the intensive care unit (ICU days).

Key independent variables of interest included hospital use of the MPS2 system and the type of cardiac surgery being performed: CABG, AVR or MVR. The choice to make type of surgery an independent variable in the model as opposed to running separate models for each surgery type allows us to explore the relationship between type of surgery and the MPS2 system while maintaining statistical power. Covariates of interest included: patient demographics (age, race, marital status, region and insurance); patient comorbidities (diabetes, hypertension, etc.) including the Charlson Comorbidity Index; primary diagnosis or reason for hospitalization; and patient risk scores (All Patient Refined Diagnosis Related Group [DRG] severity of illness [SOI] and risk of mortality [ROM] scores).

The Charlson Comorbidity Index is the most common used prognostic measure of illness burden in clinical research. It has been cited in nearly 7000 studies [10]. The Charlson Comorbidity Index assigns a score based on comorbid condition and age. Another measure of risk used in this analysis was the risk of severity and mortality scores based on the 3M All Patient Refined DRG (All Patent Refined DRGs) Classification System [11]. This index is a widely adopted proprietary risk adjustment classification tool, which uses information from routine claims data to produce valid and reliable severity measurement and risk adjustment scores. It is used to account for differences related to the patients’ SOI or ROM in large datasets. SOI refers to the extent of physiologic decompensation or organ system loss of function. ROM refers to the likelihood of dying. The SOI and ROM indices each categorize patients into minor, moderate, major or extreme [12].

Prior to multivariable modeling, summary statistics were generated by treatment (hospitals before and after the MPS2 system) and control (hospitals that never employed the MPS2 system) for the following variable categories: patient visit demographics, comorbidities and surgical characteristics.

The methodology for this research was the difference-in-differences (DID) technique, frequently used in economics and health services research, to measure the change induced by a particular treatment or event, in this case, the hospital's use of the MPS2 system. A multivariable DID analysis using fixed effects was performed for each outcome of interest (overall AE composite, total costs, medication costs, LOS and ICU days). The hospital fixed-effect specification was chosen because it controls for time-invariant hospital characteristics that are otherwise unobservable, such as specific hospital protocols. The DID allowed for the examination of the effect of hospitals’ usage of the MPS2 system by comparing patient visits undergoing cardiac surgeries using MPS2 with patient visits undergoing these surgeries in the same hospitals before the introduction of the system, as well as to patient visits undergoing these surgeries with TC. Since hospitals adopted the system in different calendar years, the ‘control’ group consisted of patient visits from hospitals who never adopted the technology and patient visits from hospitals who eventually adopted it in subsequent years. Model results are shown with progressive saturation, the final models adjusted for hospital fix effects, patient demographics, surgical characteristics (type of surgery and primary diagnosis) and comorbid conditions. All models were run using Stata Version 12.

A total of 258,479 patient visits met the initial inclusion criteria across 275 hospitals. There were 43 hospitals contributing inpatient visits that employed the MPS2 and 232 TC hospitals. Final sample size for MPS2 hospitals before implementation was 16,049 patient visits and after implementation were 29,242 patient visits. For TC hospitals the total number of patient visits was 211,523. Figure 1 displays the attrition diagram for this analysis.

Demographics were consistent across all three groups (Table 1): MPS2 hospitals before and after implementation and TC hospitals. Average age was 66 years with 68% of the patients being male. Majority of patients across all three groups were married, had Medicare as their primary insurance and were having nonurgent elective procedures.

The distribution of severity and mortality scores were also fairly consistent across the groups with the control group (TC hospitals) having the lower percentage of patients with extreme mortality (13.6%) and severity (17.9%) scores (Table 1). Charlson Comorbidity Index was consistent across the three groups with an overall average of 5.2 (Table 2). As expected among patients having CABG, AVR and MVR procedures, high rates of diabetes, heart disease, lung disease and renal disease are present across all three groups (Table 2).

The most frequently occurring primary diagnosis was coronary atherosclerosis which occurred 52.2% at MPS2 hospitals before and 46.8% after implementation and 49.0% for TC hospitals (Table 3). The distribution of CABG, AVR and MVR procedures were also consistent across the board with CABG being the most frequently occurring procedure, keeping in mind that any patient could be having more than one procedure per visit.

Figure 2 displays a pie chart showing the distribution of events that make up the clinical AE event composite. Pulmonary events were the most frequently occurring category making up 48.6% of the distribution of AEs. Since patients can have more than one AE during a visit, in order to display the distribution of AEs in the pie chart, AEs are ordered by severity and only counted once.

Multivariable results of the fixed effect DID analysis for of LOS show an average of 9.75 hospital days with 2.94 ICU days for TC and 9.56 and 2.82 days, respectively for MPS2. This equates to a 0.18 reduction in overall LOS and a 0.12 reduction in ICU days for hospitals using MPS2 compared with controls (Supplementary File D). Therefore, approximately one in every five patients experienced 1 day reduction in LOS and one in every eight had a 1 day reduction in ICU days.

Modeling results from fully saturated models estimate an average total cost and medication cost for each cohort as follows: US$45,807.50 and 3571.01 for TC and US$44,576.50 and 3378.91 for MPS2. After controlling for all covariates, the differences in cost were statistically significant with a US$1231 reduction in overall hospital costs and a US$192 reduction for medication costs for MPS2 hospitals (Supplementary File E). This equates to a 2.5% absolute reduction in total costs and a 5.38% reduction in medication costs. All costs were adjusted to 2015 dollars based on consumer price index.

In the fully saturated model, Multivariable results of the fixed effect DID analysis for the AE event composite demonstrated a 2.25% reduction in overall AEs in MPS2 hospitals compared with controls. Figure 3 displays the relative risk of the AE composite at each stage of model saturation. The first bar includes only fixed effects and each subsequent bar shows the effect of additional variables. The final bar shows that after controlling for within hospital variation and all covariates (final model), there was an approximately 5.25% relative risk reduction in overall AEs.