Cost-Effectiveness of CT Angiography and Perfusion Imaging for Delayed Cerebral Ischemia and Vasospasm in Aneurysmal Subarachnoid Hemorrhage

Model Design

We developed a decision model to perform comparative effectiveness and cost-effectiveness analyses by using a decision analytic framework in the TreeAge software program (TreeAge Pro Software, Williamstown, Massachusetts) to compare imaging strategies in SAH from a health care payer perspective. The detailed model structure is provided in On-line Fig 1. CTAP was considered the “new imaging strategy” and was compared with TCD as the “standard imaging strategy.” Clinical examination was included in the model before CTAP or TCD imaging to assess symptoms of vasospasm and/or DCI, as would typically occur in the clinical setting. The model was designed to assess patients during the typical time point at which vasospasm and DCI occur, between 4 and 14 days after SAH.

In the standard strategy, positive concordance between the clinical examination and TCD results led to induced hypertensive therapy, and negative concordance of the clinical examination and TCD led to observation, whereas discordance between the clinical examination and TCD results (such as symptoms but negative TCD findings or no symptoms but positive TCD findings) led to further testing with DSA. In the new strategy, CTA and CTP were evaluated as a single examination termed CTAP because these examinations are typically performed concurrently in clinical practice. If a patient was symptomatic, a positive CTAP finding was defined as a positive result on either CTA or CTP, whereas a negative CTAP finding was defined as a negative result on both CTA and CTP. The rationale is that given the presence of symptoms, a positive result on either examination provides sufficient supportive evidence to prompt treatment decisions. If a patient was asymptomatic, a positive CTAP finding was defined as a positive result on both CTA and CTP, whereas a negative CTAP finding was defined as a negative result on either CTA or CTP. The rationale is that in the absence of symptoms, a single positive result on either examination does not provide sufficient evidence to prompt treatment decisions. All management and treatment decisions are based on the clinical examination and TCD or CTAP combination tests.

The subsequent management and treatment decisions in the clinical pathway incorporated in the model were based on the most recent recommendations in the “Guidelines for the Management of Aneurysmal Subarachnoid Hemorrhage: A Guideline for Healthcare Professionals from the American Heart Association/American Stroke Association” (2012).¹³ Oral nimodipine was administered to all patients with aneurysmal subarachnoid hemorrhage (class I recommendation, level A evidence) because it has been shown to improve neurologic outcomes after SAH. Treatment of DCI is recommended with induction of hypertension (class I recommendation, level B evidence) and cerebral angioplasty and/or selective intra-arterial vasodilator therapy in symptomatic patients, particularly those who are not rapidly responding to hypertensive therapy (class IIa recommendation, level B evidence).

Each clinical pathway terminated in an outcome health state reflecting the downstream morbidity and mortality. A patient was assigned to only 1 of the 3 following health states: recovered, disability with dependence, or death. The long-term complications of testing or treatment affecting mortality and morbidity (as defined by the health states) were included in the model. Short-term complications such as mild allergic reactions were not included because patients recovered from these temporary ailments; this outcome would not imply significant reductions in quality of life contributing to these health states.

Input Parameter Probabilities

An outcomes-based approach was used in developing the decision model to directly assess the effectiveness of CTAP compared with TCD on clinical outcomes instead of primarily modeling the operating characteristics of these imaging examinations. We modeled probabilities of test results combined with distributions across clinical outcomes, both determined conditional on the clinical examination and imaging results. The probabilities assigned for the occurrence of symptoms and test results of CTAP, TCD, and DSA were directly derived from an SAH cohort enrolled in a diagnostic accuracy study.¹⁰ In this cohort, patients had both CTAP and TCD performed, which allowed direct comparison of these imaging strategies in the same patients. According to the study protocol, CTAP was performed at days 6–8 in asymptomatic patients and the same day that symptoms occurred in symptomatic patients with clinical deterioration. Day 7 was the median day CTAP was performed, at which time 45% (44/97) of patients had developed evidence of clinical deterioration.¹⁰ Clinical deterioration may manifest by alterations in consciousness, worsening on the Glasgow Coma Scale, or new neurologic deficits. For patients with limited clinical examinations, particularly patients who are comatose or mechanically ventilated, continuous monitoring of laboratory data and neurologic and systemic parameters was used. The details of the CTP scanning protocol and postprocessing methods are provided in the On-line Appendix. TCD was performed daily at the patient's bedside with comparisons between TCD examinations to evaluate changes in blood flow velocity measurements.

Clinical outcomes of functional disability were also assessed in the same SAH cohort¹⁰ and obtained from medical record review by a neurologist blinded to the hospitalization course and imaging data. Outcomes were based on the modified Rankin Scale and were categorized into 3 main health states as recovered (mRS 0–2), disability with dependence (mRS 3–5), and death (mRS 6). A multinomial prediction model, incorporating conditional dependence of serial imaging with CTAP and DSA or TCD and DSA, was developed and fitted to determine the predictive probabilities for each outcome based on the test results and on being symptomatic or asymptomatic. The 95% simultaneous confidence intervals for predictive probabilities were computed by using an asymptotic multivariate normal theory and were based on χ² tests.¹⁴ Statistical analysis was performed with “R: A Language and Environment for Statistical Computing” (Version 2.13.0; http://www.r-project.org/).¹⁵

Literature sources were used to determine the probabilities of long-term complications from testing (CTAP and DSA) and/or treatment with induced hypertension and intra-arterial (IA) therapy. On-line Table 1 demonstrates these input probabilities and their literature sources.

Assessment of Health Benefits and Costs

Health benefits were measured in quality-adjusted life years (QALYs). For each health state, we assigned a QALY score ranging between 0 and 1.0. Lifetime QALYs were calculated by multiplying the sum of the number of years spent in each health state by the utility associated with that state. The life expectancies were estimated from the literature as 28 years for “recovered”^16,17 and 10.8 years for “disability.”^18,19 The utilities were calculated as a weighted average from a systematic review of utilities assigned according to mRS scores.²⁰ The utility of “recovered” (mRS 0–2) was 0.80, and it was 0.22 for “disability” (mRS 3–5). By convention, “death” (mRS 6) was assigned a value of zero for both life years and utility.

The economic costs included in the model were those associated with imaging from CTAP, TCD, and DSA; treatment with induced hypertension and IA-therapy; complications from imaging and treatment; and disability after SAH, including both short- and long-term care. The costs for the imaging and treatment were based on the 2012 Medicare reimbursement rates, including both technical and professional fees. The costs for long-term care for patients with stroke have been reported in the literature.²¹ The estimated cost for recovered patients was calculated as the cumulative annual cost for the expected life years in this health state. The estimated cost for the disability patients was calculated as the first-year rehabilitation costs added to the cumulative annual cost for the remaining expected life years in this health state. By convention, death was assigned no additional cost.

Benefits and costs were discounted at a rate of 3% per year as recommended for cost-effectiveness analyses in the United States.^22,23

Model Validation

Internal validity (sometimes referred to as model verification) examines the extent to which the model calculations produced expected outcomes based on the data that were used to derive the model inputs.²⁴ We assessed the internal validity of the model by comparing the predicted probabilities of the health states (recovered, disability, and death) for each imaging strategy in the decision model with the observed outcome data from the SAH cohort. The percentage difference for the model results compared with the observed outcomes was calculated as a measure of deviation for the probabilities of each health state for the standard and new imaging strategies separately.

External validity is performed in a similar fashion but uses outcome data that were not used to develop the model as benchmarks for the predictive performance of the model.²⁴ To evaluate external validity of the decision model, we compared the overall predicted probabilities of the health states for both imaging strategies from the decision model with the published outcome probabilities from the International Subarachnoid Aneurysm Trial (ISAT).¹ The percentage difference was also calculated as a measure of deviation for model results relative to published outcomes.

Cost-Effectiveness and Sensitivity Analyses

We calculated the expected benefits and costs associated with each imaging strategy from the perspective of the health care payer. In the primary analysis for policy decision-making, the symptomatic and asymptomatic subgroups were combined in each imaging strategy, weighted by the frequency of symptoms. Additional subgroup analyses were performed to compare the symptomatic and asymptomatic groups separately. An imaging strategy that yielded the greatest quality-adjusted life years and lowest costs would be considered the imaging strategy of choice (dominant strategy).

Sensitivity analyses were performed to test the robustness of the results of the model, given the uncertainty in the input values for the parameters. One-way sensitivity analyses were performed for each parameter by altering the input values across the entire range of possible values (0–1) to identify the key drivers of the model results. In addition, 2-way sensitivity analyses were performed by simultaneously altering the input values for 2 parameters together to assess their combined effects on the results of the model. A willingness-to-pay (WTP) threshold of $100,000 per QALY was used in the 1-way and 2-way sensitivity analyses.

Probabilistic sensitivity analysis was performed to assess the uncertainty in the model results, incorporating the uncertainty of all parameter values together. Distributions for the key model parameters were derived by nonparametric bootstrapping of the SAH cohort,²⁵ focusing attention on the probability of symptoms and the probabilities of imaging (CTAP, TCD, and DSA) conditional on symptoms. Nonparametric bootstrapping was performed by resampling analytic datasets with replacement to evaluate the precision around point estimates without making assumptions regarding the distribution of these variables.²⁶ Parameters that were not directly derived from the cohort (n = 97), such as all other probability inputs, costs, and utilities, were assigned distributions (such as β, uniform, and γ) and varied on the basis of the SD estimates around the mean values. β distribution is a flexible one, which is bounded by 0 and 1, which makes it useful for varying probability inputs in probabilistic sensitivity analysis.²⁶ Uniform distributions assume equal probability of possible random values between 0 and 1.²⁶ γ distributions are rightward skewed with a lower bound at zero, which makes them useful for varying cost inputs in probabilistic sensitivity analysis.²⁶ On-line Table 1 demonstrates the input values and distributions used for all parameters in the model. Results of the probabilistic analysis (10,000 iterations) were used to generate a cost-effective scatterplot and cost-effectiveness acceptability curves to demonstrate the probability that the new strategy is cost-effective for varying cost per QALY thresholds.