Preoperative atrial fibrillation/flutter impact on risk-adjusted repeat aortic intervention patients

Database description and ethical approval

This retrospective, observational cohort study was conducted to compare risk-adjusted r-SAVR and viv-TAVR outcomes in patients with preoperative AF/AFL (AF/AFL+) and without pre-operative AF/AFL (AF/AFL-). The New York State Statewide Planning and Research Cooperative System (SPARCS) database was used for data collection. Developed in 1979, SPARCS is an all-payer reporting database that compiles information from hospitals throughout New York State regarding demographics, diagnoses, therapeutic interventions, and outcomes. Coordinated via the Department of Surgery (Dr. Pryor-Principal Investigator), r-AVR patients’ records were extracted from the SPARCS database. For the 2005-2018 New York State SPARCS database records, de-identified reports were generated by the Biostatistical Consulting Core Lab; using only de-identified reports for this study, a “not human subjects research” written exemption for these analyses was received by the Stony Brook University Committee on Research in Human Subjects (IRB 2021-00563). The study’s protocol is available online at: https://commons.library.stonybrook.edu/dos-articles/1/.

Patient population

This study’s analyses relied upon a comprehensive list of billing codes (see Supplementary Table 1; using coding manuals, these billing codes were identified by expert coders, the billing code details were extracted by the study data analytics team. All billing codes used were validated by the study’s clinician team. Using these codes, the inpatient records for adults (age > 18) undergoing a non-emergent AVR procedure from January 2005 through November 2018 were extracted. Duplicate records (N = 23), records with unknown gender (N = 1), and records missing unique personal identifiers (UPID; N = 193) were excluded. Patients who had an r-AVR at least 30 days after the first AVR were identified. Due to the increased risk for an adverse post-AVR event, patients with concomitant or prior coronary artery bypass graft or percutaneous coronary intervention, prior thoracic aortic aneurysm, aortic dissection, active endocarditis, solid tumor without metastasis, or metastatic cancer were also excluded. Demographics, baseline health conditions, and Elixhauser comorbidity score (See Supplementary Table 2) at r-AVR were considered. The Elixhauser score described patients’ risk of mortality and readmission based on pre-existing diagnoses^[13].

Study outcomes

Co-primary study endpoints included the Society of Thoracic Surgeons’ (STS) composite endpoint [major morbidity (MM); a composite comprised of 5 major complications and/or 30-day operative mortality] and 30-day readmission (READMIT). As secondary study endpoints, the STS composite’s individual 30-day operative mortality and 5 major morbidity sub-components were separately examined; these included STS major complications of a periprocedural permanent stroke, renal failure, prolonged ventilation, deep sternal wound infection, and/or repeat procedure within 30 days of the first r-AVR procedure^[14]. The STS 30-day operative mortality definition included both in-hospital death and all deaths within 30 days of the procedure. Additional secondary outcomes were evaluated, including length of hospital stay [length of stay (LOS); time from admission date to discharge date], post-procedure LOS (time from procedure date to discharge date), as well as non-STS post-procedural clinically relevant SAVR/TAVR complications including myocardial infarction, cardiac arrest, acute kidney injury, prosthetic valve endocarditis, major stroke, transient ischemic attack, major bleeding, and vascular complications. As comorbidities are commonly difficult to differentiate from post-procedural complications by exclusively using billing codes, major complications were identified only if there had been no prior evidence of that condition for the two years preceding the procedure. Following October 2015, new ICD-10 complication codes were also used to differentiate complications from comorbidities^[15,16].

Statistical analysis

Statistical analyses were performed by an institutional biostatistical consulting core lab team member with SPARCS data analytics expertise and SAS 9.4 (SAS Institute Inc., Cary, NC) biostatistics/database programming experience; data extraction, analysis and manuscript writing tasks occurred from January 2021 to February 2022. Demographics, baseline health conditions, and Elixhauser comorbidity score (See Supplementary Table 2) at r-AVR were considered. The Elixhauser score describes patients’ risk of mortality and readmission based on pre-existing diagnoses. Chi-square tests with exact P-values based on Monte Carlo simulation were utilized to examine the marginal association between categorical variables (patients’ characteristics, risk factors, specific Elixhauser comorbidities) and AF/AFL+, as well as between categorical variables and binary outcomes (e.g., MM and READMIT)^[17]. Welch’s t-tests were used to compare the unadjusted marginal differences in continuous variables (age, Elixhauser readmission score, Elixhauser mortality score) by AF/AFL+ or by binary outcomes (MM and READMIT).

As the first step, the cardiac literature was carefully reviewed to identify the patients’ risk factors to be considered as model-eligible for the MM and READMIT multivariable logistic regression models. Based on these initial literature-based conceptual variable lists, endpoint-specific bivariate screenings (P-value < 0.10) were evaluated to identify potential associations; additionally, these variables’ effect sizes were ordered to identify the optimal multivariable model-eligible variables. Importantly, variables were removed from model eligibility consideration for any coding completeness issues, clinical interpretability challenges, or collinearity with well-established AVR risk factors. Given inherent sample size (n = 334) limitations, the number of MM and READMIT endpoints was divided by 10 to identify the maximum number of multivariable risk model-eligible variables; thus, there were 6 MM model-eligible variables and 5 READMIT model-eligible variables pre-screened to be included in multivariable models^[18,19]. Similarly, a multivariable logistic regression model reported the other patient characteristics that were most commonly associated with pre-operative AF/AFL (n = 152 AF/AFL patients; 15 AF/AFL pre-screened variables were included as AF/AFL model-eligible).

Based on nested c-index comparisons, the multivariable models containing the Elixhauser score performed better than models utilizing Elixhauser-related comorbidities; thus, the Elixhauser score’s weighted sub-components were not further considered as model-eligible variables^[20]. In each logistic regression analysis, an odds ratio (OR) > 1.00 indicated an adverse outcome impact, while an OR < 1.00 indicated a protective effect. Observed/expected (O/E) ratios were calculated using baseline regression models without the key variables of interest compared; however, final study regression models directly assessed the impact of these key variables of interest. For all analyses performed, the protocol-driven statistical significance threshold was set at P < 0.05; however, all unadjusted p-values are reported for independent review. Statistical analysis was performed using SAS 9.4 (SAS Institute Inc., Cary, NC).

Baseline patient characteristics

As described in Table 1, r-SAVR AF/AFL+ patients were significantly older (mean ± standard deviation: 69.92 ± 11.06 vs. 59.17 ± 14.00, P < 0.001) and less often Hispanic (1.2% vs. 8.5%, P = 0.029) compared AF/AFL-patients. For r-SAVR patients with AF/AFL+ versus AF/AFL-, cerebrovascular disease (19.5% vs. 8.5%, P = 0.021), permanent pacemaker or implantable cardiac defibrillator (16.1% vs. 4.2%, P = 0.004), hyperlipidemia (63.2% vs. 46.6%, P = 0.018), rheumatic heart disease (11.5% vs. 3.4%, P = 0.028), fluid and electrolyte disorders (4.6% vs. 0.0%, P = 0.032), and pulmonary hypertension (26.4% vs. 11.9%, P = 0.007) were more frequently reported. Although viv-TAVR AF/AL+ versus AF/AFL- patients had similar distributions for demographics and comorbidities, there were no statistically significant risk factor differences other than AF/AFL+ viv-TAVR patients had lower rates of chronic obstructive pulmonary disease (15.4% vs. 31.3%, P = 0.033).

As overall comorbidity complexity scores, the Elixhauser summary indices were not different for r-SAVR patients with and without pre-operative AF/AFL (Elixhauser Mortality Index: 15.14 ± 11.02 vs. 12.47 ± 11.16, P = 0.090; Elixhauser Readmission Index: 23.67 ± 14.68 vs. 21.13 ± 14.49, P = 0.220). Similarly, viv-TAVR patients with AF/AFL+ versus AF/AFL- had no difference in their Elixhauser summary indices (Elixhauser Mortality Index: 15.43 ± 10.61 vs. 13.45 ± 10.48, P = 0.289; Elixhauser Readmission Index: 30.43 ± 15.45 vs. 30.61 ± 15.87, P = 0.949).

Outcomes following repeat AVR

Repeat AVR outcomes are described in Table 2. Study endpoints were evaluated in the r-SAVR and viv-TAVR cohorts [Figure 2]. Rates of the main study endpoints were similar for patients with AF/AFL compared to those without after r-SAVR (STS composite endpoint: 24.1% vs. 14.4%, P = 0.076; 30-day readmission: 19.5% vs. 14.4%, P = 0.329) and viv-TAVR (STS composite endpoint: 18.5% vs. 15.6%, P = 0.668; 30-day readmission: 15.4% vs. 9.4%, P = 0.301). Of all r-AVR procedures, the AF/AFL+ versus AF/AFL-patients had no differences identified for the MM composite (21.7% vs. 14.8%, P = 0.103) and for 30-day READMIT (17.8% vs. 12.6%, P = 0.191). Across all r-SAVR outcomes evaluated, the AF/AFL+ patients had increased rates of prolonged ventilation (10.3% vs. 2.5%, P = 0.031) and cardiac arrest (14.9% vs. 5.1%, P = 0.016). Additionally, r-SAVR patients with AF/AFL had significantly increased post-operative days (12.74 ± 11.23 vs. 9.31 ± 8.05, P = 0.017). Viv-TAVR patients with and without AF/AFL had similar rates of clinical and resource utilization outcomes. Univariate analysis results for 30-day readmission and the composite endpoint are described in Supplementary Tables 3 and 4.

Figure 2. Endpoint outcomes in patients with and without pre-operative AF/AFL undergoing (A) r-SAVR and (B) viv-TAVR. r-SAVR: Redo surgical aortic valve replacement; AF/AFL: atrial fibrillation/flutter; viv-TAVR: valve-in-valve transcatheter aortic valve replacement.

Multivariable modeling for preoperative atrial fibrillation

To understand the nature of the r-AVR patient population presenting with AF/AFL, a multivariable model was built to identify the other patient characteristics associated with preoperative AF/AFL risk. The multivariable model’s patient characteristics identified to be associated with presence of preoperative AF/AFL included older age (OR, 95%CI: 1.03, 1.01-1.05, P = 0.017), along with no documented history of cerebrovascular disease (OR, 95%CI: 0.44, 0.23-0.86, P = 0.017) or prior pacemaker or implantable cardiac defibrillator (OR, 95%CI: 0.45, 0.22-0.92, P = 0.029); this model’s c-index was 0.729 and Hosmer-Lemeshow test statistic P-value = 0.0617 (indicating no lack of model fit).

Multivariable modeling for primary study outcomes

The impact of AF/AFL on the STS composite endpoint was evaluated while holding other variables constant [Table 3]. This final model had a C-index of 0.753. Presence of pre-operative AF/AFL did not significantly impact odds of the composite (OR, 95%CI: 1.23, 0.66-2.28, P = 0.512). Other predictors of the composite endpoint included black race (OR, 95%CI: 2.89, 1.01-8.32, P = 0.049) and Elixhauser mortality score (OR, 95%CI: 1.07, 1.04-1.10, P < 0.0001).

For 30-day readmission, the impact of AF/AFL was evaluated holding other model-eligible variables constant [Table 4]. For this 30-day readmission model, the model had a c-index of 0.682. Pre-operative AF/AFL did not affect odds of 30-day readmission (OR, 95%CI: 1.15, 0.60-2.19, P = 0.681). However, history of cerebrovascular disease predicted 30-day readmission (OR, 95%CI: 2.54, 1.23-5.25, P = 0.012); additionally, viv-TAVR compared to r-SAVR procedures were protective against 30-day readmission (OR, 95%CI: 0.44, 0.21-0.91, P = 0.027).

As sensitivity analyses, the multivariable model built predicting the likelihood of patients incurring preoperative AF/AFL was added to the co-primary endpoint models built for the composite endpoint and for 30-day readmission [Supplementary Tables 5 and 6]. As these propensity scores did not substantially alter this study’s co-primary models’ findings, the current study conclusions should be considered robust.

Limitations

As an observational database analysis, this study has several inherent limitations. As with any retrospective cohort study, there may have been unknown confounding factors impacting the MM composite and/or READMIT endpoints. To address this limitation, however, all literature-based risk factors that were previously identified with a potential association with the endpoints were evaluated in this r-AVR study.

Additionally, this study was limited by a small sample size (n = 334). As a follow-up to this study, future analyses of a much larger r-AVR database should be planned. Given these preliminary findings, future research should utilize a database containing at least 15,518 r-AVR records to detect a difference in the READMIT endpoint; for the MM endpoint, a future study should plan to utilize a database of at least 5857 r-AVR records. These future sample size projections were based on a power of 80% and a significance level of 0.05. Given these findings, however, national databases (e.g., the MEDPAR or Cerner national database) will be required to address this question more rigorously. Thus, these preliminary findings based on the New York State AF/AFL patients’ experience should be re-verified by testing these same hypotheses in a larger r-AVR population.

Across SPARCS hospitals’ r-AVR procedures reported, this study focused on the population of adult New York State residents; thus, the post-procedural follow-up endpoints (i.e., 30-day readmissions and 30-day operative death) might be most accurate. Given that children (under age 18) may have other complex congenital cardiac abnormalities requiring phased sets of cardiac procedures, these were removed. As New York residents may have differential risks, moreover, their findings may not be reliably generalized to other populations with a different risk profile.

As this SPARCS billing database was used to drive hospital reimbursement, the billing codes have been assumed to be reasonably accurate. However, SPARCS administrative billing errors in coding may exist, particularly for the subgroup of billing codes not directly tied to differential reimbursement. As hospital billing codes transitioned from ICD-9 to ICD-10 in October 2015, moreover, there may have been transition-related coding-related inconsistency challenges for patient risk factors, treatment, and outcome codes. As the transition of ICD-9 codes to the newer ICD-10 codes may have been imperfect, a historical “look back” period of 2-years was used to augment the new ICD-10 complication classification codes; this approach assured that comorbidities (i.e., important risk factor diagnoses existing pre-procedure, such as the patient having a prior stroke pre-procedure) could be differentiated from complications (i.e., new diagnoses arising post-procedure, such as a patient having a perioperative stroke). Given the SPARCS database does not contain current procedural terminology codes (i.e., cardiac interventionalist billing codes) regarding all of the cardiac procedure’s details (e.g., type and size of prosthetic valve implant) or echocardiographic information (e.g., aortic valve area, gradient across the valve, velocity across the valve, and dimensionless index), this clinical information was not available for endpoint risk-adjustment. Based on billing codes alone, continuous arrhythmias could not be distinguished from intermittent arrhythmias to analyze for a differential effect on outcomes. Additionally, SPARCS does not include medication details (i.e., national drug codes); thus, the impact of different medical management approaches using specific drug therapies either pre-procedure or perioperatively could not be further investigated.

In conclusion, Pre-operative AF/AFL was not associated with 30-day readmission or the composite endpoint, consisting of major complications and/or operative mortality, following r-SAVR and viv-TAVR. However, this study identified other predictors of these adverse outcomes, including black race, Elixhauser mortality score, and cerebrovascular disease, which may allow clinicians to identify patients at elevated r-AVR risk.