Outcomes following emergent open repair for thoracic aortic dissection are improved at higher volume centers in direct admissions and transfers

The Nationwide Inpatient Sample (NIS), which is sponsored by the Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project, was used to identify patient discharges related to emergent aortic dissection repair that occurred between January 1, 2004 and December 31, 2008. The total sample size included 1,507 patients. Because the NIS provides only de-identified patient claims data, this analysis qualified for Institutional Review Board exemption.

The NIS is a 20 % sampling of abstracted discharge data from a national survey of all non-federal acute-care hospitals in the United States, and contains discharge records from over 1,000 hospitals [12]. The database contains up to 15 procedure codes per patient using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) procedure code index. Acute aortic dissection surgery was abstracted using diagnostic (441.01: Dissection of aorta, thoracic; 441.03: Dissection of aorta, thoracoabdominal) and procedural (38.35: Resection of vessel with anastomosis, other thoracic vessels; 38.45: Resection of vessel with replacement, thoracic vessels) ICD-9-CM codes, in either the first, second, or third code positions. Only patients with admissions coded as emergent, undergoing open repairs, and ≥ 18 years-old were included in the analysis. Of note, while the diagnostic and procedure codes as well as the emergent status strongly select for Type A acute aortic dissections, aortic aneurysms with rupture and complicated Type B aortic dissections are not excluded from the study population. Further, the NIS does not provide data on repeat procedures or mortality beyond the index hospital admission.

The average annual distribution of acute aortic dissection repairs across all NIS centers over the five year study period demonstrated three general clusters [Fig. 1]. Based on this distribution, and in keeping with other studies [11], the study population was stratified by annual center volume into low volume (≤5 cases/year) (n = 963; 63.9 %), intermediate volume (6–10 cases/year) (n = 370; 24.5 %), and high volume (≥11 cases/year) (n = 174; 11.6 %) groups. The analysis was further stratified by admission type: direct admission (DA), transfer admission (TA), and other. The other category included patients who were referred from outpatient clinics, health maintenance organization patients and patients for whom admission source data could not be obtained. The primary outcome was mortality.

To account for potential differences in comorbidity burden between center volume groups, the All Patient Refined-Diagnosis Related Group (APR-DRG) Severity of Illness classification was used to compare severity of comorbid illness. The score, first developed by 3 M, allows analysis of outcomes across large cohorts for a given diagnostic group [13]. Severity of illness is defined as the extent of organ system derangement or physiologic decompensation for a patient. The APR-DRG severity of illness subclasses are determined by using discharge billing codes and are based on primary and secondary discharge diagnosis, age, and preexisting medical conditions; codes reflecting in-hospital complications are excluded. The four subclasses of disease severity are: minor (1 point), moderate (2 points), major (3 points), and extreme (4 points). APR-DRG risk of mortality classification was used to adjust outcomes for their risk of mortality. This classification is similar to abovementioned APR-DRG severity of illness classification.

Continuous variables were reported as mean ± standard error and were compared using the Student’s t test. Categorical variables were reported as percentages and compared using the chi-squared test. Univariate analysis was performed using student T-test comparing two groups. Multivariate logistic regression analysis was performed using variables shown in Table 1, including patient co-morbidities, demographics and APRGRG for mortality risk as well as severity of illness. For all analyses, the conventional p-value of 0.05 or less was used to determine the level of statistical significance. All reported p-values are two-sided. All statistical analyses were performed using Stata 13 (Stata Corp, College Station, TX).

From January 1, 2004 to December 31, 2008 there were a total of 1,507 patients in the NIS registry who underwent emergent open repair of an acute aortic dissection. When categorized by center volume, 63.9 % of cases (n = 963) were performed in low volume centers, 24.5 % (n = 370) were performed in intermediate volume centers, and 11.6 % (n = 174) were performed in high volume centers. Baseline demographics and co-morbidities by center volume category are shown in Table 1. Significant differences between groups included primary payer distribution and APR-DRG Severity of Illness scores, with scores indicating a greater comorbidity burden in the intermediate volume category and low volume category. However, when APR-DRG scores were compared between DA and transfers among different volume groups, there was no statistical difference, as shown in Table 2. Common co-morbidities included hypertension, obesity, peripheral vascular disorder, and history of substance abuse.

Average number of admissions in high volume centers was 17.40 ± 9.59 (mean ± SD), intermediate volume center was 9.74 ± 4.71 and in low volume centers was 2.83 + 2.29. Distribution of admission types across the three volume strata are represented in Fig. 2. As expected, high volume centers had a greater proportion of TA (TA) (28.2 %) than intermediate (17.3 %) and low volume (7.1 %) centers. High volume centers, conversely, had a smaller percentage of DA (38.0 %) than intermediate (55.7 %) and low volume (66.6 %) centers.

Mortality data was missing for one patient in the intermediate volume group. Mortality for the entire patient sample was 21.8 % (n = 328). Univariate analysis suggested that mortality at high volume centers was significantly lower (22 out of 174 (12.6 %)) than at medium (76 out of 369 (20.6 %), p = 0.001) and low volume (230 out of 963 (23.9 %), p = 0.025) centers. For DA patients, mortality was 10.6 % (7 out of 66), 21.4 % (44 out of 206), and 24.0 % (154 out of 641) for high, medium, and low volume centers respectively. This demonstrated a statistically significant difference between high and low volume centers (p = 0.013), and a difference approaching statistical significance between high and intermediate volume centers (p = 0.051). For TA patients, mortality was 10.2 % (5 out of 49), 12.7 % (8 out of 64), and 23.5 % (16 out of 68) for high, medium, and low volume centers, respectively. Differences in mortality between volume strata in transfer patients did not reach statistical significance, due to small sample size. Mortality was lowest at high volume centers for every admission type [Table 3].

Multivariate logistic regression analysis showed that compared to high volume centers, being in low volume centers was an independent risk of mortality with Odds ratio (OR) of 2.06, 95 % Confidence Interval (CI) 1.25 – 3.38, p = 0.004. Direct and transfer admissions were not independent risk factors mortality. Other admissions when taken as a variable in multivariate analysis, was collinear with mortality and was omitted from analysis. Other independent risk factors for increased mortality were increased age and increasing APR-DRG mortality risk. Intermediate volume centers were not independent risk factor for increased mortality. See Table 4 for full details.

Low volume centers had similar mortality rates for both DA and TA (154 out of 641 patients (24.0 %) and 16 out of 68 patients (23.5 %) for DA and TA). A similar observation was made for high volume centers (7 out of 66 patients (10.6 %) and 5 out of 49 patients (10.2 %) for DA and TA). These findings suggest that center volume may be a more important factor in determining outcome than admission type. DAs, despite being potentially more emergent cases, had the same mortality as TAs at high volume centers. Similarly, TAs at low volume centers did not have improved mortality over DAs, despite potentially having a less emergent status. Determining exactly what characteristics of high volume centers confer this mortality advantage in both DAs and TAs, would be useful for improving ATAD repair outcomes across all volume strata. Nonetheless, a large portion of aortic dissections are still repaired in low volume centers with higher mortality. There needs to be a change in mindset of clinicians to transfer patients with acute aortic dissection to high volume centers where experienced surgeons, state of art operative technology and excellent post-operative patient care is available.

The above findings are further confirmed by another important observation: DAs at high volume centers have improved outcomes compared to DAs at low volume centers. A majority of the comorbidities (hypertension, history of substance abuse, diabetes, and peripheral vascular disease), were evenly distributed across groups with similar APR-DRG scores. Furthermore, comparing DAs at high volume centers and DAs at low volume centers accounts for the survivorship bias associated with transfer admission, as both patient groups are directly admitted to the treating hospital. One possible explanation for the inferior outcomes at low volume centers, is that low volume centers may have a greater diagnostic delay before identifying the ATAD. One study even demonstrates that the delay in diagnosis at a non-tertiary care center is more significant in causing delay for treatment than the delay associated with transferring the patient to a tertiary care center [18]. It is conceivable that the difference we observed in mortality across volume strata may best be explained by a difference in admission to diagnosis time, rather than a difference in diagnosis to open repair time related to patient transfer.

Additionally, TAs at high volume centers were observed to have improved mortality over DAs at low volume centers. While, survival bias is likely a confounding factor, this observation suggests that patients who are able to survive a transfer, may have improved mortality, if transferred to a high volume center, than if they stay at their original hospital.

The evidence from this analysis supports efforts for the regionalization of ATAD repairs. Recently, Harris and colleagues conducted a prospective analysis of a standardized, quality-improvement protocol for the regional treatment of acute aortic dissection [19]. The authors demonstrated a reduction in the length of time to diagnosis and surgical repair, increased use of beta-blockers on arrival and discharge, increased use of intraoperative transthoracic echocardiography, and a small decrease in short-term mortality. Despite this study, and the growing body of evidence demonstrating superior outcomes at high volume centers, nearly 65 % of emergent aortic dissection repairs in our analysis were done at centers performing 5 or fewer of such cases each year. As such, the regionalization of care for high-risk aortic surgery, represents an opportunity to improve outcomes for ATAD.

Regionalization has already been proven to be an effective method for improving outcomes, the most prominent example being ST-elevation myocardial infarction (STEMI) networks. Studies have demonstrated improved outcomes after regionalization [20, 21], despite significant barriers to implementation (competition between providers, lack of funding, lack of data collection and unified transfer center protocols, and lack of bed availability) [22, 23]. Today, in the United States, a STEMI center exists for every 585,135 persons [24], and the successes of STEMI networks have generated similar regionalization initiatives for stroke care [25, 26].

However, there is still too few data to expand regionalization practices of more common pathologies, such as STEMIs, to rarer events, such as ATADs. Recent studies analyzing rare and acute events may be beneficial in developing future regionalization networks specific to ATAD. For example, after reports of improved outcomes following subarachnoid hemorrhage repair at high volume centers [27], one group established a critical center volume threshold for subarachnoid hemorrhage repair of 6 cases/year [28]. A similar threshold could be determined in an initial phase of regionalization of ATAD care. Furthermore, these types of analyses could be further extended to other acute events, such as cardiogenic shock, acute respiratory failure, and pulmonary embolus. The regionalization model for each pathology would vary depending on where the pathology fell on the incidence spectrum of rare event (ATAD) to relatively common event (STEMI).

There are several limitations to this analysis. Firstly, the NIS is an administrative database and does not contain detailed patient characteristics or operative variables that may influence outcomes. Specifically, as previously stated, aortic aneurysms with rupture and complicated Type B aortic dissections cannot be excluded from the analysis. Secondly, the categorization of low, intermediate, and high volume centers was based on breakpoints in the distribution of annual center volume, as there are currently no guidelines defining what constitutes a low versus high volume center. Thirdly, APR-DRG severity scores cannot be compared to clinical risk stratification scores such as the STS score, as they are not calculated preoperatively; rather, they are based on ICD-9 billing codes generated after hospital discharge. Nevertheless, APR-DRG has become a preferred method for assessing severity of illness in the analysis of administrative data [29‐33]. Finally, although other studies have shown that surgeon-specific volume often drives the volume-outcomes relationship at an institutional level [34], the NIS does not contain data on surgeon-specific volume. This analysis is, therefore, not able to determine whether institutional experience, surgeon experience, or both explain the lower mortality at high volume centers. Our analysis only focused on open aortic aneurysm repairs and therefore its results are not applicable to newly emerged endovascular and hybrid approaches for treatment of aortic dissections.