Objective: To determine public health and cost consequences of time delays to EVT for patients, healthcare systems, and society, we estimated quality-adjusted life years (QALY) of EVT-treated patients and associated costs based on times to treatment.

Methods: The Markov model analysis was performed from United States healthcare and societal perspectives over a lifetime horizon. Contemporary data from seven trials within the HERMES collaboration served as data source. Aside from cumulative lifetime costs, we calculated the net monetary benefit (NMB) to determine the economic value of care. We used a contemporary willingness-to-pay threshold of $100,000 per QALY for NMB calculations.

Results: Every 10 minutes of earlier treatment resulted in an average gain of 39 days (95% prediction interval: 23-53 days) of disability-free life. Overall, the cumulative lifetime costs for patients with earlier or later treatment were similar. Patients with later treatment had higher morbidity-related costs yet over a shorter time span due to their shorter life expectancy, resulting in similar lifetime costs as in patients with early treatment. Regarding the economic value of care, every 10 minutes of earlier treatment increased the NMB by $10,593 (95% prediction interval: $5,549-$14,847) and by $10,915 (95% prediction interval: $5,928-$15,356) taking healthcare and societal perspectives, respectively.

Conclusions: Any time delay to EVT reduces QALYs and decreases the economic value of care provided by this intervention. Healthcare policies to implement efficient pre-hospital triage and accelerate in-hospital workflow are urgently needed.

METHODS

In support of the Transparency and Openness Promotion Guidelines, the authors offer cost calculations for healthcare systems and societies aside from the United States. Requests that provide a country-specific cost framework are welcomed.

 

Standard Protocol Approvals, Registrations, and Patient Consents

No IRB approval was necessary; all seven randomized trials of the HERMES (Highly Effective Reperfusion evaluated in Multiple Endovascular Stroke trials) collaboration have been approved by the local ethics committees.

 

Model Structure

We performed quality-adjusted life year (QALY) and cost estimations for the United States adopting the healthcare and societal perspective. The healthcare perspective only includes costs related to healthcare providers. We added an analysis from a societal perspective as relevant amounts of the lifetime costs of stroke are a result of premature mortality, reduced productivity of stroke survivors, and informal care by next of kin; these are included in the societal perspective. We adhered to the recommendations by the Second Panel on Cost-Effectiveness in Health and Medicine²³ and the CHEERS (Consolidated Health Economic Evaluation Reporting Standards) statement for the healthcare perspective (Data available from Dryad [Tables e-3 and e-4] https://doi.org/10.5061/dryad.6m905qfwk). As we were not able to capture the complexity of all items that go into the societal cost calculation, the recommendations set out by the Second Panel on Cost-Effectiveness in Health and Medicine for analysis form the societal perspective were not fully satisfied

A Markov model was developed using decision-analytic software (TreeAge Pro 2017, version 17.1.1.0; TreeAge, Williamstown/MA, USA). A short-run model was created to analyze costs and functional outcomes within the initial 90 days after the index stroke. Patients enter the model on admission to the hospital for acute ischemic stroke at different times from symptom onset, are EVT-eligible or EVT-ineligible due to time delays, and afterwards enter one of the seven health states according to the degree of disability as assessed by the modified Rankin Scale (mRS). The reference case analysis was performed for a stroke onset at the age of 65 years. The model does not compare the treatments of EVT versus intravenous thrombolysis nor does it apply to patients that only receive intravenous thrombolysis. A long-run Markov state transition model estimated the expected costs and outcomes over the lifetime of the patient, using a cycle length of one year. During each cycle, patients could either remain in the same health state, suffer a recurrent stroke and recover or transit to worse mRS states, or die.^{24, 25} Death resulted from age-related mortality rates and the excess mortality rates of stroke survivors. The model structure is shown in Figure 1 (Data available from Dryad [Figure e-1] https://doi.org/10.5061/dryad.6m905qfwk). All simulations were carried out over a lifetime horizon.

 

Model Input Parameters

The input parameters for the model were based on contemporary EVT trial collaboration data and the most recently published literature providing the best available level of evidence (Table 1). ADDIN EN.CITE.DATA ^{11, 26-37} We accounted for the impact of patients' age on all input parameters based on a recent systematic review.¹⁵

 

Probabilities of Clinical Events

The initial probabilities (i.e. the probability of entering a mRS health state at the end of the initial 90 days) were derived from patient subgroups of symptom onset times to arterial puncture based on seven EVT trials within the HERMES collaboration. ADDIN EN.CITE.DATA ³⁸ The patient outcomes were grouped by hours of symptom onset to arterial puncture as this yielded well-balanced and appropriate sample sizes for application of the Dirichlet distribution for uncertainty modelling. Treatment eligibility for EVT was assumed to be 100% within the first two hours of onset to arterial puncture based on expert consensus. After two hours, we assumed a conservative 3% decrease in eligibility per 30 minutes based on a recent study on infarct progression over time²⁷ and expert consensus. Overall, these model assumptions result in rates that are comparable to estimates provided by a study that reported the collective treatment eligibility for an all-comer patient population within the first six hours of symptom onset.³⁹ The outcomes of EVT-ineligible patients were taken from the control arm subgroup with established ischemic changes in line with current guidelines.¹⁰ The transition probabilities during each annual cycle of the long-run model accounted for remaining in the same health state, the annual recurrent stroke rate, the probability of reentering the same or a lower health state following a recurrent stroke, and the annual death rate. The age-specific annual death rate of the general population was drawn from the United States Life Table.²⁹ The excess death rate of stroke survivors was calculated according to hazard rate ratios by mRS health states as reported by contemporary cohort studies.³⁰

 

 

 

Health Benefits

Therapy effectiveness was calculated using QALYs according to current recommendations.²³ QALYs were calculated by multiplying years spent in mRS health states by assigned utility weights. Utility weights were based on a recent consensus analysis of reported patient-centered and clinician-centered utility values.³³ Values range from 0·0 to 1·0, with 0·0 representing no, and 1·0 representing perfect quality of life. All QALYs were discounted by 3% each year according to current recommendations.²³ Days of disability-free life were calculated by dividing QALYs through 365. The patients' functional independence was defined for the time patients stayed in the health states mRS 0-2. Days of life in functional independence gained by earlier treatment were calculated based on the cumulative time the simulations spent in the health states mRS 0-2 across the lifetime projections, using the average change across the above-defined subgroups of time from symptom onset to expected arterial puncture.

 

Costs

Aside from the cumulative lifetime costs, we also calculated the net monetary benefit (NMB) to determine the economic value of care. The NMB combines weighted QALYs and costs into one composite outcome:²³ NMB = [(lifetime QALYs x willingness-to-pay) - lifetime costs].²³ A higher NMB indicates a better economic value of care. NMBs were calculated using a willingness-to-pay (WTP) threshold of $100,000 per QALY.²³ For the healthcare perspective, direct costs within the first 90 days after stroke and direct annual long-term costs were based on contemporary data and stratified for each of the seven mRS health states.^{11, 31} The costs for EVT were taken from the SWIFT PRIME trial.¹¹ The costs for intravenous thrombolysis were taken from the National Inpatient Sample. All costs were adjusted to 2017 United States Dollars according to the medical care component of the Consumer Price Index and discounted by 3% each year in line with current recommendations.²³ For the societal perspective, we accounted for indirect costs caused by stroke, which were assessed based on the human capital approach. The amount of the societal losses are measured based on the lost productivity due to premature mortality in stroke patients, the reduced productivity that is caused by the morbidity of stroke survivors, and the costs for informal care given by family members.²³ The detailed methods for societal cost calculations are reported; Data available from Dryad (Additional Methods https://doi.org/10.5061/dryad.6m905qfwk).

 

Sensitivity Analyses

To test the robustness of the model prediction, we conducted probabilistic sensitivity analyses, allowing for simultaneous alteration of multiple input parameters. All input values in the model were varied using distributions that reflect each input parameter's uncertainty as derived from HERMES collaboration outcome data or the literature. Distributions were calculated using probability density functions appropriate to each parameter as shown in Table 1 (sensitivity ranges for each parameter are provided; Data available from Dryad [Table e-1] https://doi.org/10.5061/dryad.6m905qfwk). The probabilistic sensitivity analysis was conducted using 10,000 2^nd order Monte Carlo simulation runs. As this analysis accounts for all uncertainties related to the model, the results are reported as median estimates with 95% prediction intervals. Additional one-way sensitivity analyses, varying only individual parameters, were conducted and are available (Data available from Dryad [Figure e-3] https://doi.org/10.5061/dryad.6m905qfwk). External model validation was performed using longitudinal data as reported by the REVASCAT trial⁴⁰ and MR CLEAN trial⁴¹ (Data available from Dryad [Figure e-4] https://doi.org/10.5061/dryad.6m905qfwk).

 

Estimation of Patient Level and Population Level Effects

Results from the probabilistic sensitivity analysis, which are based on HERMES collaboration patient outcome data grouped by hours of symptom onset to arterial puncture, indicated a near-linear relationship between additional hours and loss of effectiveness as well as NMB. Under this assumption, the results were used to estimate patient level effects of 10-minute time delays. For population level effect estimation, the most recently reported annual rate of EVT-treated stroke patients in the United States population (3·3% of all 692,000 ischemic strokes) was considered to extrapolate patient level results to national estimates.⁴²

 

Data Availability Statement

All the data from all the trials has been submitted and is available through the VISTA-ENDOVASCULAR repository (http://www.virtualtrialsarchives.org/vista-endovascular/).

There are no missing values. No additional data is needed to use this dataset for the analysis.