Transcatheter Aortic Valve Implantation Wait‐Time Management: Derivation and Validation of the Canadian TAVI Triage Tool (CAN3T)

Abstract

Background

Transcatheter aortic valve implantation (TAVI) has seen indication expansion and thus exponential growth in demand over the past decade. In many jurisdictions, the growing demand has outpaced capacity, increasing wait times and preprocedural adverse events. In this study, we derived prediction models that estimate the risk of adverse events on the waitlist and developed a triage tool to identify patients who should be prioritized for TAVI.

Methods and Results

We included adult patients in Ontario, Canada referred for TAVI and followed up until one of the following events first occurred: death, TAVI procedure, removal from waitlist, or end of the observation period. We used subdistribution hazards models to find significant predictors for each of the following outcomes: (1) all‐cause death while on the waitlist; (2) all‐cause hospitalization while on the waitlist; (3) receipt of urgent TAVI; and (4) a composite outcome. The median predicted risk at 12 weeks was chosen as a threshold for a maximum acceptable risk while on the waitlist and incorporated in the triage tool to recommend individualized wait times. Of 13 128 patients, 586 died while on the waitlist, and 4343 had at least 1 hospitalization. A total of 6854 TAVIs were completed, of which 1135 were urgent procedures. We were able to create parsimonious models for each outcome that included clinically relevant predictors.

Conclusions

The Canadian TAVI Triage Tool (CAN3T) is a triage tool to assist clinicians in the prioritization of patients who should have timely access to TAVI. We anticipate that the CAN3T will be a valuable tool as it may improve equity in access to care, reduce preventable adverse events, and improve system efficiency.

Nonstandard Abbreviations and Acronyms

CAN3T

Canadian TAVI Triage Tool

CIHI‐DAD

Canadian Institute for Health Information Discharge Abstract Database

TAVI

transcatheter aortic valve implantation

Clinical Perspective.

What Is New?

• The Canadian TAVI Triage Tool (CAN3T), is a tool to assist clinicians in identifying and prioritizing patients referred for transcatheter aortic valve implantation who are at higher risk for adverse events.

What Are the Clinical Implications?

• The CAN3T has the potential to improve waitlist management and equity in access to care and reduce the number of unplanned procedures and adverse events.

The past decade has seen a paradigm shift in the therapeutic options for the management of severe aortic stenosis, with the emergence of transcatheter aortic valve implantation (TAVI) as an alternative to surgical aortic valve replacement. As evidence has accumulated, TAVI has become the standard of care for inoperable and high–surgical risk patients, and a reasonable alternative for intermediate– and low–surgical risk patients. ¹ As a result, the demand for TAVI has grown exponentially worldwide, outpacing capacity in many jurisdictions, resulting in substantial wait times. ² , ³ For example, in Ontario, Canada, despite a 5‐fold improvement in the penetration of TAVI in the health system, ² , ⁴ median wait times worsened from 80 days in 2012 to 110 days in 2018. ⁵

Previous research has found that longer wait times for TAVI are associated with important adverse events such as mortality and hospitalization. ⁶ Not only this, the deterioration in health status that occurs from waiting translates to reduced recovery and greater mortality in the postprocedural period. ⁷ This is particularly true if deterioration results in a hospitalization and the need for an urgent, unplanned procedure. ⁸ , ⁹ Long wait times have detrimental health system impacts, including increases in health care costs and reductions in the availability for elective procedures. ¹⁰

The approach to long wait times requires both an increase in overall system capacity as well as proper waitlist management. Expanding system capacity increases the procedural volumes, reducing the average time patients are at risk for adverse events. ¹¹ This, however, has been particularly challenging with system disruptions such as the COVID‐19 pandemic, which in many jurisdictions forced hospitals to halt or reduce the availability of elective procedures. ¹² Risk‐stratification tools have been considered important approaches to waitlist management and can assist clinicians in triaging patients by identifying and prioritizing those at higher risk of preprocedural adverse events to reduce the negative consequences of longer wait times. ¹¹ To date, there is a paucity of such evidence‐based tools or recommendations for waitlist management of patients referred for TAVI.

Accordingly, our goal was to address this gap in knowledge by identifying characteristics that are associated with an increased risk of adverse events while on the waitlist, derive models to predict the risk of these adverse events at different time points, and finally develop a tool (the Canadian TAVI Triage Tool [CAN3T]) that can be used by clinicians to identify high‐risk patients who should be prioritized for TAVI.

METHODS

Study Design and Setting

We conducted an observational retrospective cohort study using population‐based administrative data from Ontario, Canada, held at ICES (previously known as the Institute for Clinical Evaluative Sciences). The use of anonymized administrative data held at ICES without patient consent is permitted in Ontario based on section 45 of Ontario's Personal Health Information Protection Act, which does not require review by a research ethics board. Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols may be sent to ICES at ices.on.ca.

Data Sources

This study used the CorHealth Ontario Clinical Registry as the primary data source. Reporting to the database is mandatory in Ontario, and a requirement for provincial procedural funding. ² 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. ¹³ The CorHealth registry is linked to administrative databases held at ICES using unique, encoded identifiers. 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, ¹⁴ , ¹⁵ heart failure, ¹⁶ hypertension, ¹⁷ , ¹⁸ chronic obstructive pulmonary disease, ¹⁹ and dementia. ²⁰ Medical frailty was determined using the Hospital Frailty Risk Score. ²¹ Comorbidities and previous cardiac procedures were identified during the period of up to 20 years before referral. Mortality was ascertained via the Registered Persons Database as were additional demographic variables such as quintile of median neighborhood income and rural residence. Details on the variable definitions can be found in Table S1.

Study Population

All Ontario residents older than 18 years were eligible for the study. The index event for inclusion was referral for TAVI, from April 1, 2012, to March 31, 2020. Patients were followed up until the first of death on the waitlist, removal from the waitlist (off‐listing), receipt of a TAVI procedure, or March 31, 2020 (the time at which patients who still remained in the waitlist were censored). For patients who had repeat TAVI 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. Patients who remained on the TAVI waitlist for longer than 1 year were also excluded (≈1.6% of patients); the rationale for this exclusion is that such patients with excessive wait times were likely referred for TAVI but did not have an indication for the procedure after evaluation from the heart team and remained on the waitlist for monitoring their clinical evolution.

Outcome Measures

The 4 primary outcomes of interest were: (1) death from any causes while on the TAVI waitlist; (2) all‐cause hospitalization while on the waitlist; (3) receipt of an urgent TAVI (defined as a TAVI procedure during an acute hospitalization); and (4) a composite outcome defined as the first of all‐cause waitlist hospitalization, death, or urgent TAVI.

Statistical Analysis

To identify predictors of adverse events on the waitlist, we used the subdistribution hazards model, also known as the Fine‐Gray model. This model allows for the estimation of the cumulative incidence function while accounting for competing risks that would preclude the occurrence of the outcome of interest. ²² , ²³ Competing risks are relevant in our study as we were only interested in events during the preprocedural period, and therefore any events that resulted in removal of the patient from the waitlist must be considered in the analysis. For all‐cause death, removal from the waitlist (off‐listing) or receipt of a TAVI procedure were considered competing risks; for urgent TAVI, death or off‐listing; for all‐cause hospitalization, off‐listing or death or receipt of TAVI; and for the composite outcome, off‐listing was the only competing risk.

Model Building

To achieve a parsimonious and clinically useful model, we performed variable selection in three steps. In the first step, certain candidate variables were removed from further consideration based on an assessment of their correlations and practical considerations. Stepwise backward selection was then performed until all variables in the model had a significance level <0.05. ²⁴ Finally, significant variables with negative coefficients were removed due to potential collider bias and no biological plausibility. ²⁵ We did not assess linearity of the effects of continuous covariates nor did we include any interactions between covariates.

Model Performance

We assessed the discrimination and calibration of each model. To quantify the discriminative ability, we calculated the concordance statistic (C‐statistic) and the area under the receiver operating characteristic curve. At each time point of interest, random pairs of patients were compared, with one with the outcome and the other without the outcome. We used the Brier score to assess overall predictive accuracy, estimating the mean difference between the observed status and predicted probability of event at each time point; therefore, values closer to 0 indicate better performance of the model. ²⁶ We also calculated the calibration slope. A slope closer to 1 indicates optimal calibration. ²⁷

All performance measures and predicted risks were assessed at 14, 30, 60, and 84 days. We defined 84 days from referral (12 weeks) as a maximum wait time for any patient, as per the current Canadian Cardiovascular Society guidelines. ²⁸

Maximum Acceptable Risk

The maximum acceptable risk is somewhat arbitrary and based on clinical expertise. Incorporating the 84‐day threshold from the guidelines, based on the distribution of predicted risks in our cohort, we defined the median risk at 84 days as a threshold for a maximum acceptable risk. Therefore, our tool will recommend that a patient have a TAVI performed in a maximum of 84 days if the predicted risk at this time is equal or lower than the threshold for maximum acceptable risk; if the predicted risk is higher than the threshold at 84 days, the tool will provide a maximum individualized wait time that should be <84 days, and is defined as the number of days at which the predicted risk reaches the threshold.

Validation

We performed internal validation of the models using bootstrapping to calculate optimism‐corrected performance measures, following the procedures described by Steyerberg. ²⁴ A schematic of the internal validation summarizes the validation process and can be found in Figure S1.

Missing Data

Our data were mostly complete, with the exception of New York Heart Association class, of which 28.39% was missing. To utilize all available data, we performed multiple imputation using the multivariate imputation using a chained equations algorithm over 5 iterations, generating 10 complete data sets. Estimates were combined using Rubin rules. For the validation procedures, the multiple imputation was nested in the bootstrapping, ie, for each bootstrapped data set, 10 complete data sets were generated and statistics were summarized accordingly, yielding 1 estimate per bootstrapped sample. Figure S1 provides an analytical schematic of the model derivation and validation.

All analyses were conducted in the R statistical environment, version 3.6.1 (R Foundation for Statistical Computing), ²⁹ using the packages survival, rms, riskRegression, pec, and mice. Statistical significance was defined by a 2‐sided P value <0.05.

RESULTS

Cohort

We identified 13 809 referrals for TAVI in Ontario from April 1, 2012 to March 31, 2020 in 13 380 unique patients. After excluding 41 patients with invalid identifiers, index or removal dates, and procedures performed elsewhere, 13 339 remained. Demographic and clinical characteristics of the patients are summarized in the Table. An additional 211 patients were excluded for having follow‐up times >1 year, yielding a total of 13 128 patients included in the analysis (Figure 1). Throughout the period of the study, 586 (4.46%) patients died while on the waitlist and 4343 (33.08%) had at least 1 all‐cause hospitalization. A total of 6854 TAVIs were completed (52.21% of all referrals), from which 1135 (16.56%) were urgent procedures. From the 4872 (37.1%) patients who were off‐listed, 1029 (7.84%) were rereferred for surgical aortic valve replacement and 3843 (29.27%) were off‐listed for other reasons. Possible causes for off‐listing include cancellation of the procedure, patient or physician decision to not treat the disease, decision to manage disease without invasive intervention, and decision to monitor the patient (who may be reconsidered for the procedure at a later time).

Table .

Baseline Patient Characteristics (N=13 128)

Age, y
Mean±SD	81.0±8.3
Median (IQR)	82 (76–87)
Women	5906 (45.0)
Rural residence	1598 (12.2)
Income quintile
1	2631 (20.0)
2	2832 (21.6)
3	2668 (20.3)
4	2424 (18.5)
5	2539 (19.3)
Charlson Comorbidity Index
Mean±SD	1.4±1.8
Median (IQR)	1 (0–2)
NYHA class
I	1094 (11.6)
II	2942 (31.3)
III	4605 (49.0)
IV	760 (8.1)
Congestive heart failure	7096 (54.0)
Cardiac/atrial arrhythmia	2321 (17.7)
Peripheral vascular disease	342 (2.6)
Cerebrovascular disease	545 (4.1)
COPD	4528 (34.5)
Cognitive impairment/dementia	989 (7.5)
Cancer	864 (6.6)
Dialysis	442 (3.4)
Interstitial lung disease*	133 (1.0)
Liver disease	231 (1.8)
Renal disease	883 (6.7)
Diabetes	5736 (43.7)
Hypertension	11 945 (91.0)
Dyslipidemia	7426 (56.6)
Hospital Frailty Risk Score
Mean±SD	2.6±4.6
Median (IQR)	0 (0–3.4)
Prior stroke	306 (2.3)
Coronary artery/ischemic heart disease	4931 (37.6)
Previous CABG	2025 (15.4)
Previous PCI	2241 (17.1)
Previous valve surgery†	1140 (8.7)
Previous pacemaker (includes CRT‐P)	1076 (8.2)
Previous ICD (includes CRT‐D)	201 (1.5)
Fiscal year of referral
2012	408 (3.1)
2013	883 (6.7)
2014	1252 (9.5)
2015	1492 (11.4)
2016	1679 (12.8)
2017	1945 (14.8)
2018	2265 (17.2)
2019	3204 (24.41)
Wait time, d
Mean±SD	97.1±79.7
Median (IQR)	78 (32–143)
Death	586 (4.5)
All‐cause hospitalization	4343 (33.1)
Urgent TAVI	1135 (8.6)
Elective TAVI	5719 (43.6)
Off‐listed	4872 (37.1)
Censored	816 (6.2)

Values are expressed as number (percentage) unless otherwise indicated. CABG indicates coronary artery bypass graft; COPD, chronic obstructive pulmonary disease; CRT‐D, cardiac resynchronization therapy defibrillator; CRT‐P, cardiac resynchronization therapy pacemaker; IQR, interquartile range; NYHA, New York Heart Association; PCI, percutaneous coronary intervention; and TAVI, transcatheter aortic valve implantation.

^*Includes pulmonary fibrosis.

^† Includes aortic, mitral, tricuspid, and pulmonary valves.

Figure 1. Cohort flow diagram.

TAVI indicates transcatheter aortic valve implantation.

Table S2 summarizes patients' characteristics by waitlist outcomes. In general, patients who were off‐listed and rereferred to surgical aortic valve replacement were the healthiest of the cohort. Patients who died on the waitlist or underwent urgent TAVI were more likely to have congestive heart failure, those who died on the waitlist were more likely to be undergoing dialysis, and those who underwent urgent TAVI had more likely undergone previous valve surgery.

Model Building

After an initial assessment, we removed the Charlson Comorbidity Index as it was correlated with many other variables. Prior stroke and dialysis were also removed because they had similar definitions and therefore correlated with cerebrovascular disease and renal disease, respectively. The remaining variables were included for stepwise backward selection (Figure S3; Table S3). Dyslipidemia and previous pacemaker were removed for having negative coefficients. The final models are presented in Figure 2, and baseline cumulative incidence function for selected time points are shown in Table S4.

Figure 2. Final subdistribution hazards models.

COPD indicates chronic obstructive pulmonary disease; NYHA, New York Heart Association; PCI, percutaneous coronary intervention; and TAVI, transcatheter aortic valve implantation.

For all‐cause death, the final model included frailty, sex, age, congestive heart failure, chronic obstructive pulmonary disease, liver disease, renal disease, and interstitial lung disease. The urgent TAVI model included New York Heart Association class, congestive heart failure, coronary artery or ischemic heart disease, previous percutaneous coronary intervention, and valve disease. The all‐cause hospitalization model included frailty, sex, New York Heart Association class, age, diabetes, congestive heart failure, chronic obstructive pulmonary disease, liver disease, coronary artery or ischemic heart disease, and previous valve surgery. Finally, for the composite outcome, the same variables were used as in the hospitalization, with the addition of renal disease.

All variables retained in the final models were also present in >50% of the models derived in bootstrap samples (Figures S2 and S3).

Model Performance

Apparent and corrected performance measures can be found in Table S5. At 84 days, optimism‐corrected C statistics varied from 0.63 (95% CI, 0.62–0.64) focomposite outcome) to 0.71 (95% CI, 0.69–0.74) (urgent TAVI) (Figure S4), area under the curve varied from 0.66 (95% CI, 0.63–0.69) (death) to 0.71 (95% CI, 0.69–0.73) (urgent TAVI), and Brier scores varied from 0.02 (death) to 0.19 (composite outcome) (Figure S5). The calibration slopes varied from 0.79 (death) to 0.99 (urgent TAVI).

Predicted Risks

The distribution and quantiles of predicted risks at 84 days in our cohort can be found in Table S6 and Figure S6. In the full cohort, 586 patients (4.46%) died on the waitlist, with predicted risks at 84 days ranging from 0.58% to 27.55% and a median of 2.5%; 1135 patients (8.65%) underwent an urgent TAVI within 84 days, with predicted risks ranging from 1.82% to 46.93% and a median of 7.77%; and 4343 patients (33.08%) were hospitalized within 84 days on the waitlist, with predicted risks ranging from 9.26% to 69.33% and a median of 28.25%.

In terms of the composite outcome at 84 days, 4559 patients (34.73%) experienced the composite outcome, with predicted risks ranging from 8.93% to 73.16% and a median of 27.92%.

Based on the median predicted risk at 84 days in our cohort, the thresholds for a maximum acceptable risk were as follows: for all‐cause death it was 2.5%, for urgent TAVI, 7.77%; for all‐cause hospitalization, 28.25%; and for the composite outcome, 27.92%. As an example, if a patient has a predicted risk of death that reaches this maximum acceptable risk of 2.5% at 60 days, the tool will recommend that the patient should wait no longer than 60 days to undergo TAVI. If, in contrast, a patient has a predicted risk of death that does not reach 2.5% within 84 days, the tool will recommend that the patient wait no longer than 12 weeks for the procedure, in accordance with current Canadian guidelines.

DISCUSSION

This study described the development and validation of the CAN3T, an online tool that can be used to assist waitlist management for patients referred for TAVI in the identification of patients at high risk for preprocedural adverse events. We followed best practices recommended for the development and validation of clinical prediction models, ³⁰ using a representative registry, including a thorough assessment of the data, incorporating subject matter in the model specification process, and using validated methods to deal with missing data and for internal validation.

To date, there is no prioritization system developed specifically for TAVI. Similar clinical tools designed for prioritization of patients on waitlists have successfully improved waitlist management and patient outcomes. A clinical priority system designed in New Zealand, prioritizing patients for valve surgery and coronary artery bypass graft by severity of symptoms, showed an 80% reduction in preprocedural deaths, ³¹ and a recently implemented prioritization framework for elective procedures in England has shown improvements in data quality and management of the waitlists. ³² Notwithstanding, only few studies have systematically assessed the effectiveness of prioritization tools thus far. ³³ The CAN3T has the potential to improve patient outcomes, avoiding the occurrence of preventable preprocedural hospitalizations and deaths, and consequently improving system efficiency with fewer urgent procedures and shorter procedural length of stay.

Many characteristics have emerged as important predictors of adverse events on the TAVI waitlist. Unsurprisingly, a history of congestive heart failure was a significant risk factor for all outcomes with a substantial impact. Among the candidate comorbidities, frailty, chronic obstructive pulmonary disease, liver disease, renal disease, and coronary/ischemic heart disease were predictors of most of the adverse events; from the previous cardiac procedures, valve surgery was the most commonly significant predictor. Only a few studies have also investigated risk factors for preprocedural outcomes in different cohorts of patients referred for TAVI: Popovic and colleagues concluded that EuroSCORE, body mass index, and frailty indices do have prognostic value both for the pre‐ and the post‐TAVI period, ³⁴ and Shimura et al identified New York Heart Association class, frailty, prior coronary artery bypass graft, pulmonary disease, albumin, creatinine, and hemoglobin as predictors of mortality after referral for TAVI. ³⁵ Our study is broadly consistent with this previous work; however, it is important to emphasize that our models were developed for clinical predictions and not to assess causal relationships between the risk factors and outcomes. Therefore, our findings should not be used as evidence to confirm or exclude any risk factors for the events we were interested in making predictions for.

Most available risk stratification tools developed to triage patients classify patients into risk categories. We expand on this approach by presenting a novel paradigm for waitlist management that does not attempt to classify patients but instead incorporates the deterioration in health that occurs while patients wait for a procedure. The CAN3T was designed to be used by the heart team during the postreferral assessment, and answers the question of how long a patient referred for TAVI can wait until their risk of an adverse event becomes excessive, in a very flexible way as patients nor the maximum wait times are categorized. Acknowledging that there is no consensus on what can be considered an excessive risk of an adverse event, and that this definition may change within a specific setting, we incorporated this as a modifiable parameter in our tool. We also provide the predicted risk at different meaningful time points so the clinician can understand better how these risks increase with time (Figure 3). Another feature of the CAN3T is the inclusion of different adverse events, allowing the user to choose 1 outcome they may find more suitable or relevant to their setting. Alternatively, all outcomes may also be used to inform prioritization; in this case, the maximum recommended wait time for TAVI would be the earliest time when any of the outcomes reach the risk thresholds.

Figure 3. The Canadian TAVI Triage Tool (CAN3T) online tool.

TAVI indicates transcatheter aortic valve implantation.

We believe this flexibility of the tool is an attractive feature as different programs may have a different patient profile or different eligibility criteria for the procedure, and these particularities can be considered when defining which outcomes or threshold values will be used to triage patients. Indeed, the lack of flexibility in prioritization tools has been cited as a barrier to their use and acceptance by clinicians. ³³

Our study must be interpreted in the context of some limitations that merit discussion. First, despite the use of a large data set, our models build on data from Ontario only, which performs ≈45% of all TAVIs in Canada, ² thus limiting the generalizability of the tool to other contexts. Another important limitation of the CAN3T tool was its moderate performance, with C statistics that ranged from 0.63 to 0.75, and which will inevitably be lower in any external population. Dynamic models incorporating changes after referral and time‐covariate interactions could possibly have improved performance; however, they would require much more data and constant retriage of the patient after the initial triage, which may be a barrier for implementation of the tool.

The difficulty in achieving optimal performance has been an issue in other prediction models derived in TAVI cohorts, and 1 of the possible causes is the fact that this population is older and with many comorbidities, which makes it difficult to distinguish adverse events caused by aortic stenosis itself. ³⁶ On the other hand, the complex relationship between comorbidities, symptoms, and severity of aortic stenosis justifies the potential value of less‐specific outcomes in clinical decision‐making for the patient with AS. In the current study, we only collected data for all‐cause hospitalization and death, and we acknowledge that not having cause‐specific outcomes is a limitation of this study as some admissions and deaths may not be related to AS and thus not preventable by earlier TAVI. However, we believe that it is possible that models with cause‐specific and all‐cause outcomes have comparable performances, and we plan to explore this in future studies.

Arguably, the clinical utility of a prediction model (its ability to change clinical decision‐making) is more important than an optimal performance, and we believe this is the case for the CAN3T. Despite its moderate performance, the tool can be an important ally to identify patients who would benefit from being prioritized for TAVI, specifically when we consider that there is no such tool yet available for this purpose. Such value is also greater in a period when many health systems face unprecedented long wait times after the COVID‐19 pandemic, ³⁷ during which many of health care resources for elective procedures such as TAVI had to be reallocated for the management of patients with COVID‐19.

An additional issue that may impact the performance of the CAN3T is the dynamism of the TAVI landscape. During the past decade we have seen an expansion of indication for TAVI to lower surgical risk in patients, as well as improvements in the technology and procedure, consequently affecting the patient profile and risk prediction. ³⁶ , ³⁸ Our models were built on data collected over 8 years, a period that encompasses much of this evolution. To explore possible temporal changes in our cohort throughout these 8 years, we observed the median time to event per fiscal year (Figure S7) and applied our models to calculate, for each outcome, the median risk per fiscal year of referral (Figure S8). For time to event, no clear trend was observed from 2012 to 2016, followed by a decreasing trend between 2017 and 2019. As for the median predicted risks, we did observe a reduction after 2014, which is compatible with the expansion of TAVI from higher to lower surgical risk groups. Although we acknowledge that including a more diverse population to derive our models may affect their internal validity, incorporating this diversity may be beneficial for their external validity. These temporal changes, added to the effects of the COVID‐19 pandemic, may impact the performance of our models in the future and will be the object of examination in future studies.

The use of data from a pre‐COVID period may also be seen as a limitation, but this decision was based on research findings demonstrating that disruptions caused by the pandemic led to reductions in detection, referrals, and cardiovascular procedures, affecting wait times, patient profile, and incidence of adverse events ³⁹ , ⁴⁰ ; we therefore believe that data from the pre‐COVID period is more representative of the population and generalizable to other settings.

The CAN3T is publicly available and can be accessed at https://sunnybrook.ca/CAN3T. As future steps of this work, our group plans on performing studies to assess different aspects of the clinical utility of the tool, such as cluster randomized trials, prospective observational studies, and model‐based simulations. These studies will allow us to investigate some of the complexities involved in the adoption of triage tools into routine clinical practice. ⁴¹ With dissemination of the tool, experiences from institutions will be used to inform future updates. We also plan on externally validating the models using TAVI registries from other Canadian provinces.

CONCLUSIONS

The CAN3T is a triage tool developed to assist the clinician in the prioritization of patients who should have timely access to TAVI. It was derived from a population‐level registry, using validated methods, and has moderate discrimination. User feedback and external validation of the models in a different population may not only improve its performance and generalizability but also incorporate changes in patient profile that occur within the dynamic TAVI landscape. We anticipate the CAN3T to be a valuable support tool for patients and clinicians, with the potential of bringing meaningful changes to the health system as it improves equity in access to care, reduces preventable adverse events, and improves system efficiency.

Sources of Funding

Dr Wijeysundera is supported by the Canada Research Chairs Program. Dr Madan is supported by the Heart and Stroke Foundation of Canada.

Disclosures

None.

Supporting information

Tables S1–S6.

Figures S1–S8.

References 42, 43, 44, 45, 46