Regional Differences in Outcomes for patients undergoing Transcatheter Aortic Valve Replacement in New York State and Ontario

Abstract

Background:

TAVR has become the standard of care for a wide spectrum of patients with severe aortic stenosis. However, there are wide variations in access to TAVR between jurisdictions. It is unknown if such variation is associated with differences in post-procedural outcomes. Our objective was to determine whether differences in health care delivery in jurisdictions with high versus low access of care to TAVR translate to differences in post-procedural outcomes.

Methods:

In this observational, retrospective cohort study, we identified all Ontario and New York State residents greater than 18 years of age who received TAVR from January 1^st, 2012, to December 31^st, 2018. Our primary outcomes were post-TAVR 30 day in-hospital mortality and all cause readmissions. Using indirect standardization, we calculated the observed versus expected outcomes for New York patients, had they been treated in Ontario.

Results:

Our cohort consisted of 16,814 TAVR patients at 36 hospitals in New York State and 5,007 TAVR patients at 11 hospitals in Ontario. In Ontario, TAVR access rates increased from ~18.2 TAVR/million in 2012 to 87.4 TAVR/million in 2018, while for New York State, the rates increased from 31.9 to 220.4 TAVR/million. For 30-day mortality, 3.1% of Ontario TAVR patients had an in-hospital death, compared to 2.5% of New York patients. With adjustment, this translated to an observed/expected ratio of 0.70 (95% CI 0.54–0.92) for New York patients.

Conclusions:

Having greater access to TAVR maybe associated with improved outcomes, potentially due to an intervention earlier in the disease trajectory.

Graphical Abstract

INTRODUCTION

Transcatheter aortic valve replacement (TAVR) has revolutionized the therapeutic options for the patients with symptomatic severe aortic stenosis over the last two decades. Contemporary clinical practice guidelines recommend TAVR as the treatment of choice for patients who are deemed at prohibitive or high risk for surgical aortic valve replacement (SAVR), and a reasonable alternative for patients at intermediate or low risk ^1–3. As such, there has been an exponential growth in demand for TAVR worldwide ⁴.

In many jurisdictions, the demand for TAVR has exceeded capacity, resulting in poor access, with potentially a higher threshold for offering therapy, and/or longer wait-times and substantial wait-time morbidity and mortality ^5–9. These harms must be weighed against possible benefits of centralization, specifically the referral to specialized centers with potentially higher procedural volumes ¹⁰. In other jurisdictions, rapid dissemination of TAVR centers have ensured adequate capacity at a population level, albeit with relatively low volumes at some institutions. Given the relationship between operator/hospital volume and outcomes seen in TAVR, this has raised concerns of poorer post-procedural outcomes being a possible undesirable clinical consequence of more widespread availability of TAVR ^11–13 There is a paucity of literature on how these two opposing scenarios compare – (1) potentially sicker patients pre-procedurally, but with potentially improved post-procedural outcomes given the higher operator/hospital experience, versus (2) less sick patients with shorter wait-times, but potentially poorer post-procedural outcomes due to lower operator-volume experience. Previous cross-country comparison between Canada and the United States has provided unique insights into how different health care systems funding and delivery may translate into different outcomes for patients with comparable conditions^14–17. Such literature is limited in TAVR ^{18, 19}.

Accordingly, to address this gap in knowledge, we compared outcomes between Ontario, Canada and New York State, USA, as a natural experiment comparing two healthcare systems with substantially different capacities to perform TAVR. Our objective was to determine whether differences in health care delivery in jurisdictions with high versus low access of care to TAVR translate to differences in post-procedural mortality and readmissions.

METHODS

Study design and setting

We conducted an observational retrospective cohort study using population based administrative data. Due to privacy restrictions that limit the transfer of data out of each jurisdiction, each dataset was analyzed separately without merging. Canadian data were held at ICES, Ontario (previously known as the Institute for Clinical Evaluative Sciences). The use of ICES data in this retrospective cohort study was authorized under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a Research Ethics Board. The use of anonymized administrative data without patient consent at ICES is allowed in Ontario based on provincial privacy legislation. New York State Department of Health approved the analysis of state discharge records by Weill Cornell Medicine. The Weill Cornell Medicine Institutional Review Board approved the study and waived informed consent.

We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement for reporting of observational studies ²⁰. The data underlying this article cannot be shared as it is based on administrative data and governed by the privacy regulations in Ontario, Canada.

Context

The study was conducted in New York State, USA and Ontario, Canada. Ontario is Canada’s largest province with a population of approximately 14.8 million. All residents receive universal health care provided by a single third-party payer, the Ministry of Health and Long-Term Care. TAVR has been available in Ontario since 2012 and is currently available at all 11 hospitals in Ontario that provide SAVR. In contrast, TAVR has been available in the US since 2011. New York State has a population of 19 million and TAVR is offered at 36 hospitals. Insurance coverage at these hospitals is a combination of either Medicare, Medicaid, private health insurance, or self-pay.

Data sources

Ontario data were obtained from the CorHealth Ontario Clinical Registry. This database receives demographic, comorbidity, and procedural data on all invasive cardiac procedures in Ontario, and its accuracy has been previously validated by means of retrospective chart review and comparisons with other databases ²¹. We linked the data using unique, encoded identifiers to administrative databases held at ICES. The Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD) provided data on acute hospitalizations, complications, as well as supplemented baseline comorbidity and procedural data. Validated ICES-derived databases were used to identify diabetes, heart failure, hypertension, chronic obstructive pulmonary disease (COPD), and dementia. In-hospital and out-of-hospital mortality was ascertained via the Registered Persons Database.

New York State data were obtained from New York State Department of Health Statewide Planning and Research Cooperative System (SPARCS), an all-age group, all payer database that collects patient and treatment information for every hospital discharge, outpatient and ambulatory surgery, and emergency department visit in the state. A crosswalk was designed to ensure that ICD 10 codes used to define comorbidities in the Ontario data corresponded to comparable ICD 9/10 codes used in the New York data. In-hospital death and readmission data were ascertained from the SPARCS dataset.

Study population

We included all Ontario and New York State residents greater than 18 years of age who received TAVR from January 1^st, 2012, to December 31^st, 2018. For patients who had repeated procedures during the period of the study, only the first procedure was included. Patients were excluded from the analysis if they had invalid birth, death dates or identifiers. The maximum follow-up date was for 30 days post discharge from the index TAVR hospitalization.

Outcomes

The primary endpoint was 30-day in-hospital mortality post TAVR procedure. We restricted our analysis to in-hospital mortality as mortality data in New York was limited to these settings. The secondary outcome was hospital readmission within 30 days after the index TAVR acute care hospital discharge.

Statistical Analysis

The access to TAVR was determined by the number of TAVR/million of total population, obtained from Census data for each jurisdiction by calendar year.

To understand differences in post-procedural outcomes in each jurisdiction we applied indirect standardization to estimate the mortality for New York patients, had they been treated in Ontario. We first constructed models in Ontario. For 30-day mortality, we built a logistic regression model and for 30-day readmission, we build a Fine-Gray model, to account for the competing risk of death. Selection of candidate variables was based on clinical expertise. Variables included in the multivariable models were age, sex, TAVR access site (transfemoral versus non-transfemoral), urgent/elective procedure, previous coronary artery bypass grafting (CABG), previous percutaneous coronary intervention (PCI), previous valve surgery, coronary artery disease, cancer, heart failure, COPD, cerebrovascular disease, dementia, dialysis, arrhythmia, diabetes, hypertension and year of procedure. Model discrimination was assessed using the c-statistic.

Parameter estimates (and the baseline cumulative incidence function when modeling readmission) with the variance-covariance matrices from the models established based on Ontario data, were applied to New York data to obtain the expected mortality/readmission for the New York patients. We then calculated mean observed/expected (O/E) ratio for New York. If the O/E ratio was <1, then patients in New York had an improved outcome compared to Ontario. To account for uncertainty and estimate a 95% confidence interval for the O/E ratio, we used 2000 bootstrap samples of the Ontario data to build 2000 sets of model coefficients. We then applied these coefficients to 2000 bootstrap samples of the New York data for predicted outcomes. For each bootstrap, we performed sampling with replacement and used the same sample size as the original full cohort for each jurisdiction. The 95% CI was derived based on the 2.5^th percentile and the 97.5^th percentile of the O/E ratios estimated from these bootstrap samples. In the sensitivity analysis, we restricted to elective (i.e. non-urgent TAVR) transfemoral access only patients and repeated this analysis.

We used SAS version 9.4 (SAS Institute Inc. Cary, North Carolina). Statistical significance was two-sided p-values of ≤0.05.

RESULTS

Study Populations

Between January 1^st, 2012, and December 31^st, 2018, there were 18,584 and 5,049 TAVR patients identified in New York State and Ontario respectively. After applying exclusion criteria, we had 16,814 TAVRs performed in the 36 hospitals in New York State and 5,007 TAVRs performed in the 11 hospitals in Ontario (Figure 1). Over this period, the median volume/hospital in New York was 268 TAVR, ranging from 1 to 2559. In Ontario, the median volume/hospital was 528 TAVR, ranging from 13 to 839.

Figure 1:

Cohort Selection

From 2012 to 2018, there was a dramatic increase in access to TAVR for both jurisdictions (Figure 2). In Ontario, rates increased from ~18.2 TAVR/million in 2012 to 87.4 TAVR/million in 2018. For New York State, the rates increased from 31.9 to 220.4 TAVR/million over the same period. There was almost 3-fold higher utilization of TAVR in New York State compared to Ontario over this time span. There were important differences between the two cohorts (Table 1). Specifically, there were more transfemoral and urgent cases in New York. Although more patients in New York had a history of coronary artery disease, more patients in Ontario had previous PCI or CABG.

Figure 2:

Rates of TAVR Access in New York State and Ontario

Table 1:

Baseline Characteristics

Baseline Characteristics (N - %)	New York State (16,814)	Ontario (5,007)
Mean (SD)	81.9 (8.3)	82.05 ± 7.41
Female	8477 (50.4%)	2,228 (44.5%)
Congestive heart failure	13384 (79.6%)	3,620 (72.3%)
Coronary artery disease	14594 (86.8%)	3,499 (69.9%)
Cardiac arrhythmia	7741 (46.0%)	1,257 (25.1%)
Cerebrovascular disease	2712 (16.1%)	251 (5.0%)
COPD	5473 (32.6%)	1,777 (35.5%)
Dementia	1382 (8.2%)	365 (7.3%)
Cancer	1740 (10.3%)	343 (6.9%)
Dialysis	752 (4.5%)	174 (3.5%)
Interstitial lung disease	945 (5.6%)	64 (1.3%)
Liver disease	892 (5.3%)	84 (1.7%)
Diabetes	6708 (39.9%)	2,199 (43.9%)
Hypertension	15943 (94.8%)	4,737 (94.6%)
Previous Cardiac Procedures
Previous CABG procedure	2310 (13.7%)	1,022 (20.4%)
Previous PCI procedure	836 (5.0%)	1,720 (34.4%)
Previous valve surgery	809 (4.8%)	652 (13.0%)
Urgent	4752 (28.3%)	831 (16.6%)
Elective	12062 (71.7%)	4,176 (83.4%)
Transfemoral	15936 (94.8%)	4,280 (85.5%)
TAVR Valve in Valve	42 (0.2%)	550 (11.0%)
Total TAVRs - Calendar Year
2012	624 (3.7%)	244 (4.9%)
2013	1187 (7.1%)	442 (8.8%)
2014	1646 (9.8%)	600 (12.0%)
2015	2165 (12.9%)	692 (13.8%)
2016	3064 (18.2%)	805 (16.1%)
2017	3820 (22.7%)	962 (19.2%)
2018	4308 (25.6%)	1262 (25.2%)

SD- standard deviation; COPD: chronic obstructive lung disease; CABG: coronary artery bypass grafting; PCI: percutaneous coronary intervention; TAVR: trascatheter aortic valve replacement

Unadjusted outcomes

As seen in Supplemental Appendix S1, 3.1% of Ontario TAVR patients had an inhospital death within 30 days of their procedure, compared to 2.5% of New York patients. In contrast, a similar proportion of patients had repeat readmissions within 30 days of discharge (14.6% and 14.1% in Ontario and New York respectively).

Model

The final models for mortality and readmission are shown in Tables 2 and 3 respectively. The most important drivers of mortality were access site (odds ratio [OR] of 0.42 for transfemoral), urgency status (OR 2.31 for urgent) and year of procedure, with worse mortality in early years. In contrast, the strongest drivers of readmissions were comorbidities, such as heart failure (hazard ratio [HR] 1.38), atrial arrhythmia (HR 1.48) and a history of coronary artery disease (HR 1.27).

Table 2:

30 Day Mortality Model

Variable	P-value	Odds Ratio	Lower CL	Upper CL
TAVR Access site: Transfemoral	<.0001	0.479	0.340	0.676
TAVR procedure status: Urgent	<.0001	2.018	1.470	2.770
Age	0.440	1.008	0.987	1.030
Female	0.093	1.291	0.959	1.739
Previous Cardiovascular conditions: Heart failure (CHF)	0.002	1.977	1.297	3.013
Comorbidities:
Arrhythmia/Atrial Arrhythmia	0.333	1.169	0.852	1.603
CAD/Ischemic Heart Disease	0.087	1.370	0.956	1.964
Cerebrovascular Disease	0.835	1.066	0.585	1.943
COPD	0.200	1.212	0.903	1.626
Cognitive Impairment/Dementia	0.942	0.980	0.566	1.695
Cancer	0.907	0.967	0.548	1.706
Dialysis	0.105	1.672	0.899	3.108
Hypertension	0.686	0.876	0.460	1.666
Diabetes	0.608	0.924	0.685	1.248
CABG	0.003	0.497	0.316	0.782
PCI	0.606	1.087	0.791	1.494
Valve Surgery (including mitral/tricuspid and/or pulmonary)	0.386	0.817	0.518	1.290
Year:
2012	0.010	2.480	1.242	4.949
2013	0.001	2.605	1.473	4.605
2014	<.0001	3.204	1.928	5.326
2015	0.000	2.738	1.641	4.568
2016	0.125	1.553	0.885	2.727
2017	0.050	1.708	1.000	2.919

c-statistic: 0.75 (95% CI 0.71–0.78) COPD: chronic obstructive lung disease; CABG: coronary artery bypass grafting; PCI: percutaneous coronary intervention; TAVR: trascatheter aortic valve replacement

Table 3:

30 Day Readmission Model

Variable	P-value	Hazard Ratio	HR Lower CL	HR Upper CL
TAVR Access site: Transfemoral	<.0001	0.603	0.501	0.726
TAVR procedure status: Urgent	0.132	1.157	0.957	1.399
Age	0.032	1.013	1.001	1.024
Female	0.305	1.087	0.927	1.275
Previous Cardiovascular conditions: Heart failure (CHF)	0.001	1.381	1.136	1.678
Comorbidities:
Arrhythmia/Atrial Arrhythmia	<.0001	1.475	1.254	1.734
CAD/Ischemic Heart Disease	0.013	1.265	1.051	1.522
Cerebrovascular Disease (CVD)	0.249	1.189	0.886	1.596
COPD	0.000	1.317	1.134	1.529
Cognitive Impairment/Dementia	0.130	1.223	0.942	1.587
Cancer	0.137	1.223	0.938	1.594
Dialysis	0.084	1.354	0.960	1.909
Hypertension	0.539	1.126	0.771	1.644
Diabetes	0.010	1.218	1.048	1.417
Coronary Artery Bypass Surgery	0.853	0.982	0.805	1.196
Coronary Intervention	0.615	0.959	0.814	1.130
Valve Surgery (including mitral/tricuspid and/or pulmonary)	0.056	0.788	0.617	1.006
Year:
2012	0.629	1.091	0.767	1.553
2013	0.496	1.106	0.827	1.479
2014	0.583	0.927	0.708	1.215
2015	0.006	1.402	1.101	1.785
2016	0.146	1.195	0.940	1.521
2017	0.611	1.063	0.839	1.348

c-statistic: 0.62 (95% CI 0.60–0.64); COPD: chronic obstructive lung disease; CABG: coronary artery bypass grafting; PCI: percutaneous coronary intervention; TAVR: trascatheter aortic valve replacement

In Figure 3, indirect standardization was applied to the New York cohort. The observed mortality was 2.5% and the expected mortality was 3.6%, resulting in an O/E ratio of 0.70 (95% CI 0.54–0.92). Given that this ratio was < 1, patients in New York had reduced mortality compared to in Ontario. The O/E ratio for readmission did not show any statistically significant difference (O/E of 0.95 ;95%CI: 0.85–1.07).

Figure 3:

Observed versus Expected Outcomes

In a sensitivity analysis, we included only patients undergoing elective, transfemoral TAVR (Supplemental Appendix S2). In contrast to the primary analysis, there was no longer any significant difference between the 2 cohorts with regards to either mortality or readmission. The observed mortality in Ontario was 2.9%, while that in New York State was 1.9%. The O/E ratio for mortality was 0.91 (95% CI 0.62–1.40).

DISCUSSION

In this international comparison of TAVR in New York State, US and Ontario, Canada, we found substantially different penetration of TAVR, with ~3 fold greater access in New York State. We found that the median number of TAVR done per hospital in each jurisdiction was higher in Ontario, but both jurisdictions had a substantial range in volume per institution. In addition, we found that in-hospital 30-day mortality, was lower in New York State, the jurisdiction with the greater access. This difference was mitigated when we restricted the analysis to patients who were elective, transfemoral cases.

There is a rich data on the variation of TAVR penetration or access. The impact of inadequate access is both a higher threshold for patients to be offered TAVR, as well as potentially lengthy wait-times. Indeed, Ontario restricted TAVR to inoperable and high risk patients due to capacity concerns, despite evidence for lower risk patients and approval for such patients in other jurisdictions. In so far as wait-times, work from Ontario have shown that wait-times have increased over time as demand has grown, and with it, TAVR wait-time mortality and hospitalization has increased ²². In addition, the impact of wait-times on post-procedural events have been studied, with evidence suggesting that a deterioration in health status that leads to an urgent procedure translates to poorer outcomes, including length of stay, hospital related costs, and 30-day mortality ^{6, 23, 24}. This has prompted calls for both approaches to better wait-time management and triage, as well as investments to improve TAVR capacity.

A parallel rich literature has focused on the presence and strength of an operator as well as institutional volume-outcome relationship ^{11, 25, 26}. These have broadly found that outcomes are more variable in the lowest volume quartile, with a relative risk reduction of ~20% in short-term mortality between a high-volume and volume centers ¹¹. This has led to minimal volume requirements in some jurisdictions for both operators and hospitals ^{27, 28}.

In this study, we compared TAVRs performed under two health systems with substantial differences in access as well as volume of procedures/hospital. We found differences in short-term outcomes, specifically that the jurisdiction with higher overall access in terms of TAVR/population had improved outcomes, despite lower volume/hospital on average. Unfortunately, wait-time data for New York State is lacking, but is anticipated to be much shorter than Ontario, given the increased capacity. Somewhat counterintuitively, we found that despite the higher capacity in New York State, it had a higher proportion of urgent cases. This is despite Ontario likely having sicker patients, given that Ontario had restricted TAVR access to inoperable and high-risk patients during this period. The differences we found in short term outcomes were mitigated when we excluded urgent patients. This suggests that there is potentially a lower threshold to admit patients from the wait-list who are deteriorating and perform an urgent procedure in New York, as compared to Ontario, where due to the lack of capacity, such patients may have died on the wait-list. This reinforces the notion that with greater capacity, there is a more liberal approach to patient selection. This hypothesis requires further evaluation, with longer term outcome data beyond 30 days.

Work done by our group and others has found that the strongest predictor of short-term mortality post-TAVR is peri-procedural complications, as compared to longer term mortality which is driven by co-morbidities ²⁹. It is an important area for further work to understand if the differences we observed between our cohorts are indeed due to fewer peri-procedural issues. If so, this may be due to the New York cohort being earlier in their disease progression by having greater access and therefore, a shorter wait-time. This would provide insight as to the optimal capacity in terms of TAVR/population to minimize wait-times while nonetheless ensuring adequate operator and institutional experience.

Our results must be interpreted in the context of several limitations that merit discussion. First, we did not have information on wait-time duration or events in the New York cohort. As such, we could not exclude other reasons for the differences we observed in outcomes – instead, we can state that broadly the differing health systems in New York State and Ontario are associated with different outcomes. That said, wait-time information represents an important data gap that exists for most international TAVR registries. This paucity of data on the wait-time period is an area for quality improvement. A TAVR team assumes care for a patient once a referral is received. Similar to care in ST-segment elevation myocardial infarction, one should initiate the “clock” once this medical contact is made. Our study is another call for improvement in data collection in the period from referral to procedure, both in terms of wait-times as well as clinical events that occur in the wait-time period. Second, we were limited in our ability to control for potential confounders by the data collected in each jurisdiction. We did not have information on parameters such as left ventricular function or STS score, both of which are known to influence short-term outcomes. Indeed, the fact that our models have only moderate C-statistics may be indicative of this issue. Although we performed risk adjustment, we cannot exclude the possibility that some of the differences we observed were due to unmeasured patient-level confounders between the 2 cohorts. Finally, and most importantly, our study was ecologic – in other words, done at the aggregate level and cannot be extended to the potential outcomes for the individual patient. As such, our work should be considered hypothesis-generating, not conclusive.

In conclusion, we found substantial differences in access between New York State and Ontario, Canada, with the jurisdiction with improved access associated with improved outcomes. This calls for further research to understand the optimal balance between overall TAVR capacity as well as individual operator and institution volume.