Drivers and outcomes of variation in surgical versus transcatheter aortic valve replacement in Ontario, Canada: a population-based study

Study design and setting

We conducted an observational retrospective cohort study using population based administrative data held at ICES, Ontario (previously known as the Institute for Clinical Evaluative Sciences). We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology statement for reporting of observational studies.13

Context

The study was conducted in Ontario, which is Canada’s largest province with a population of approximately 14.8 million.14 All residents receive universal healthcare provided by a single third-party payer, the Ministry of Health. TAVR has been available in Ontario since 2012, and is currently funded for inoperable, high-risk, intermediate-risk and low-risk patients with AS. TAVR is currently available at all 11 hospitals in Ontario that provide SAVR.

Data sources

Our study used the CorHealth Ontario Clinical Registry as our primary data source. The database receives demographic, comorbidity and procedural data and its accuracy has been previously validated by means of retrospective chart review and comparisons with other databases.15 We linked the data using unique, encoded identifiers to administrative databases held at ICES. The Canadian Institute for Health Information Discharge Abstract Database provided data on acute hospitalisations, complications, as well as supplemented baseline comorbidity and procedural data. Validated ICES-derived databases were used to identify diabetes,16 17 heart failure,18 hypertension,19 20 chronic obstructive pulmonary disease,21 and dementia.22 Medical frailty was determined using the Hospital Frailty Risk Score.23 Mortality was ascertained via the Registered Persons Database as were additional demographic variables such as quintile of median neighbourhood income and rural residence.

Study population

We included all Ontario residents greater than 18 years of age who received TAVR or SAVR from 1 April 2016 to 31 March 2020. This time frame was selected to capture a contemporary AS population. Patients who received SAVR with or without coronary artery bypass graft (CABG) were included only if the main diagnosis was predominantly AS (ICD-10 I350/I352), and those who underwent other concomitant cardiac procedures were excluded from this study. For patients who had repeated procedures during the period of the study, only the first procedure was included. Patients with a previous SAVR performed prior to 2016 were included in our primary cohort. Patients were excluded from the analysis if they had invalid birth, death dates or identifiers. The maximum follow-up date was 30 June 2020.

Outcome measures

For objective 1, the primary outcome was a completed TAVR or SAVR, treated as a dichotomous variable. We defined a TAVR/SAVR ratio for each hospital and categorised hospitals into tertiles of low, medium and high TAVR/SAVR ratio. We also treated TAVR/SAVR ratio as a continuous variable. For objective 2, our primary clinical outcomes were all-cause mortality and all-cause readmission until maximum follow-up.

Statistical analysis

We used a two-level hierarchical logistic regression model to investigate the between-hospital variability in the management of AS; we included hospital-specific random intercepts to account for clustering of patients by procedural hospital. We developed four models: (1) an unconditional/null model that contained the hospital-specific random effects only; models adjusted for; (2) patient characteristics; (3) patient characteristics and hospital factors; and finally (4) all variables, including the year of the procedure. For patient characteristics, we did not include the Charlson score, or ischaemic heart disease in the models, as it was colinear with other factors. We calculated two metrics of variation: the median OR (MOR) and variance partition coefficients (VPC) for each model. The MOR is defined as the median value of the OR for the outcome between individuals with identical covariates but randomly selected from different clusters (ie, hospitals), with clusters being ordered by TAVR/SAVR ratio.24 We calculated 95% crediblility intervals for the MOR. The VPC is the share of the total individual variance that is attributable to the cluster (ie, hospital) level; the higher the VPC, the more relevant is the influence of the cluster.24 As we add covariates to each model (null, vs 2, vs 3, etc), if those covariates explained the variation between hospitals, we would expect the MOR and VPC to decrease. To provide a more comprehensive interpretation of the hospital-level effects, we calculated the 80% interval OR (IOR-80), which incorporates the magnitude of cluster-level residual variations to the cluster-level covariates fixed effects. The IOR-80 is the interval centred on the median of the distribution between the 10^th and 90^th percentiles of the values of the OR for each cluster-level covariate. If the interval includes the value one, the effect of the covariate is not as relevant when compared with the remaining cluster-level heterogeneity.25

To determine the consequences of variation in care, we developed Cox proportional hazards models with random intercepts (also known as frailty models) adjusted for patient characteristics, year of procedure and hospital factors, as well as wait-time and in-hospital complications, with the main factor being hospital TAVR/SAVR ratio. For readmission, we developed a cause-specific Cox proportional hazards model to account for the competing risk of death. In these models, the main exposure was the TAVR/SAVR ratio. We created two models: in the first, the TAVR/SAVR ratio was treated as a 3-level categorical variable; in the second, it was treated as continuous. We assessed if the TAVR/SAVR was non-linear and if so, applied a restricted spline analysis. We repeated each of these models, excluded patients’ admission status, hospital length of stay (LOS) and in-hospital periprocedural complications (bleeding, stroke, acute kidney injury and pacemaker/implantable cardioverter-defibrillator (ICD) insertion) to understand if these were the drivers of any outcome differences between hospitals in high, medium and low ratios of TAVR/SAVR.

We also repeated the models with an interaction term between the TAVR/SAVR ratios and receipt of either procedure to determine if the type of procedure (SAVR or TAVR) was an effect modifier of the clinical impact of variation.

Sensitivity analysis

We conducted a number of sensitivity analyses to determine the robustness of our models. We repeated our analyses, restricted to patients who underwent isolated SAVR and TAVR. We restricted our analyses to patients without a previous open-heart operation. We used SAS V.9.4 (SAS Institute). Statistical significance was considered to be two-sided p<0.05.

Cohort

We identified 5051 TAVRs performed in Ontario between 1 April 2016 and 31 March 2020; 24 of these were repeated procedures, yielding a total of 5027 unique TAVR patients. During the period of our study, of the 11 855 SAVRs with or without CABG, 7436 had a primary diagnosis of AS. After excluding patients with concomitant multiple procedures, invalid codes and multiple procedures, 5577 patients who underwent SAVR were included in the analysis (figure 1).

The baseline characteristics of the patients are reported in table 1. TAVR patients were older (mean age 81.6 vs 70.4), more likely female (43.2% vs 32.3%) and had greater comorbidity burden, as well as prior procedures. As for the hospital factors, TAVR was more likely to be performed in hospitals with more acute care beds (mean 472.3 vs 440.6) but a smaller overall AS volume, based on total number of AS procedures (SAVR or TAVR) (mean 569.8 vs 581.3) over the study period. TAVR patients had greater wait times (median days 114 vs 57 days), while SAVR patients were more likely to undergo an urgent procedure, defined as a procedure occurring during an acute hospitalisation (14.7% vs 18.4%).

Unadjusted outcomes are found on table 2. The only in-hospital complication that differed significantly between the two groups was pacemaker implantation, which was higher for TAVR (10.2% vs 4.7%). TAVR patients had a shorter procedural LOS (median days 3 vs 7). There were no differences in crude 30-day all-cause mortality or readmission. However, TAVR was associated with worse unadjusted outcomes at 90 days and maximum follow-up.

Variation in procedures

The overall number of procedures increased over the years, with 2393 being performed in 2016 and 2944 in 2019. This corresponds to an increase in procedures rates per million population from 171 to 201 (figure 2). TAVR rates increased from 61 to 122, whereas SAVR rates decreased from 110 to 79. TAVR/SAVR ratios differed substantially between hospitals, from 0.21 to 3.27 (figure 3). Mean TAVR/SAVR ratios for the entire province increased from 0.55 in 2016 to 1.53 in 2019 (online supplemental figure 1). Baseline patient and hospital characteristics by TAVR/SAVR ratio tertiles are found in online supplemental table S1.

Drivers of AS care

Figure 4 illustrates the drivers of receiving TAVR instead of SAVR. In general, increased comorbidity was associated with an increased likelihood of TAVR. For example, age (OR 1.22, 95% CI 1.21 to 1.23), frailty score (OR 1.08, 95% CI 1.06 to 1.1) and congestive heart failure (OR 2.14, 95% CI 1.89 to 2.44). In addition, females were more likely to receive TAVR (OR 1.95, 95% CI 1.72 to 2.2), as were patients who had previous procedures including PCI, CABG or pacemakers. As for the hospital factors, hospitals with a higher overall AS procedural volume over the entire study period were associated with a lower likelihood of TAVR (OR 0.66, 95% CI 0.63 to 0.7 for every 100 procedures). There was a strong temporal effect, with TAVR more likely in the more recent years.

Patient variables, hospital factors and year of procedure all increased the MOR and VPC compared with the null model (online supplemental table S2). This indicates that neither patient, hospital factors nor year of procedure explained the between-hospital variation in receiving TAVR instead of SAVR. The IOR-80 for the hospital factors in the full models were 0.15–8.44 for acute care beds, 0.09–4.93 for number of procedures, and 0.57–31.65 for teaching hospitals (online supplemental table S3). These wide intervals indicate that the residual variation between hospitals is substantial when compared with the hospital factors effects.

Consequences of AS care

When analysed as a categorical variable, The TAVR/SAVR ratio showed a significant association with all-cause readmissions (figure 5) with a gradient of improved outcomes with a higher ratio. Compared with high tertile hospitals, low and medium tertiles were associated with greater hazards of readmissions (HR 1.24, 95% CI 1.08 to 1.43 and HR 1.13, 95% CI 1.03 to 1.24 for low and medium tertiles, respectively). For all-cause mortality, the association was non-significant.

When analysed as a continuous variable, we did not find any evidence of non-linearity in the TAVR/SAVR ratio. When modelled as a linear variable, findings were not consistent compared with the categorical models, and the association was significant for all-cause mortality, whereas for readmissions it was not (HR 0.87 with each unit increase in TAVR/SAVR ratio, 95% CI 0.77 to 0.98 for mortality; HR 0.95 with each unit increase in TAVR/SAVR ratio, 95% CI 0.88 to 1.01 for readmission). The interaction term for procedures and TAVR/SAVR ratio was not statistically significant in any of our models. When we repeated our models, removing in-hospital complications and LOS, the results for all our models were consistent.

Sensitivity analysis

In the analyses restricting SAVR patients to isolated SAVR only and to patients with no previous open heart surgery, our overall results were qualitatively similar (online supplemental tables 4–11).

Limitations

This study must be interpreted in the context of limitations that merit discussion. First, our study focused on variation of either SAVR or TAVR. Importantly, this requires that a patient with AS is diagnosed and referred for invasive therapies. As such, it is critical to acknowledge that our study may underestimate the degree of disparity, as there may be marked differences in the proportion of AS patients who are treated medically/conservatively. Given the substantial differences in outcomes between conservative therapy vs TAVR/SAVR, this disparity may translate to substantial variations in prognosis. Indeed, previous work in the USA suggests that there is geographic discordance between AS incidence and referral to therapy.29 It is an ongoing area of research in our group to understand if there are geographic, ethic and socioeconomic gradients in the referral for AVR therapies vs medical therapy in Ontario. Second, we were only able to include a few hospital-level factors due to data availability; however there may be additional hospital factors also relevant, such as number of trained professionals, diagnostic infrastructure, and availability of hospital facilities.5 8 Third, only mortality and readmission were measured as outcomes in the study, and variation in care may also have effects on additional outcomes such as functional capacity and quality of life.8 30 Finally, given the observation nature of the study, our findings should be considered hypothesis generating and not conclusive.