Markov Model Description
We constructed a Markov model from a healthcare payer perspective to examine cost-effectiveness of a hypothetical future blood-based diagnostic test using TreeAge Pro v 2022 (Williamstown, Massachusetts, USA). A Markov Model was chosen to mimic the real-world disease states of patients with known CA. Per national guidelines, costs and QALYs were discounted at a 3% rate.[6] In our model, patients exist in mutually exclusive statesand may transition between states at the end of the cycle. By default, the patients underwent the standard of care (SOC) which consists of three states of patients with known CA: active surveillance, treatment, or rupture. Terminal nodes of the model describe if the patient is alive and if alive what their health state is graded by the modified Rankin Scale (mRS). During active surveillance, patients’ CAs would then be treated, grow, rupture, or remain stable. Following treatment, patients would either be stable and receive further active surveillance or would progress to death. Following rupture, patients would either die or be hospitalized and treated. Patients that are hospitalized following rupture can experience different health states: healthy (mrs 0), mild disability (mRS 1-2), moderate to severe disability (mRS 3-5), or death (mRS 6). Similar pathway was constructed for the scenario model for a blood based diagnostic test. For this hypothetical pathway we assumed tri-annual blood testing. An illustration of the model can be found in Figure 1A. Average patient age was set at 30 years to reflect when CAs become more common as reported by the National Institute of Neurological Disorders and Stroke. More than 88% of all multiple CAs are detected in patients ≥30 years old.[7] Analysis was then performed over 50 year-cycles. Both values for patient age and number of cycles were specifically chosen in lieu of a typical average diagnosis of CA of 55 years. This was done to obtain more comparable results within the high risk-group analysis. We analyzed outcomes based on costs and quality-adjusted life years (QALYs).
Costs and Outcomes
The costs of DSA and other imaging included in the SOC were obtained from the 2022 National Medicare reimbursement rates. In agreement with prior literature, DSA was assumed to be a 6-vessel angiogram (bilateral internal carotid arteries, bilateral external carotid arteries, and bilateral vertebral arteries).[8] In order to allow for easier comparisons, cost of the hypothetical blood-test was assumed to be equivalent to the cost of DSA. The costs of complications, different health states, and vasospasm were obtained from available literature and adjusted for inflation to 2023. All costs and their descriptions are listed in Supplementary Table 1.
Model Parameters
Sensitivity and specificity of the SOC versus the blood test were used as input parameters in the decision model. Face validity of the model parameters was assessed by neurosurgery experts and prompted our inclusion of health states by modified Rankin Scale (mRS). We accounted for patients with no disability (mRS 0), mild disability (mRS 1-2), moderate to severe disability (mRS 3-5), and death (mRS 6). Associated probabilities of these health states were determined through the literature.[9-12] Associated costs were also obtained from the literature.[13-19] A weighted-average cost-probability value was then synthesized based on reported mRS grades for ruptured CA patients and used in the final analysis. We assumed all identified CAs were treated appropriately, because we could not account for individual case-specific management decisions. We assumed that 50% of the watchful waiting population for both model arms would be treated in the first year only. This assumption was made based on literature showing that 50% of patients end up in a watchful waiting state with 30%, 50%, and 90% intention-to-treat rates for radiologists, neurologists, and neurosurgeons, respectively, for small CAs.[20-22] Every year following (cycles 2-50) aneurysms in the watchful waiting population were treated at a rate of 6.7% based on the literature.[9]
Probabilities
All probabilities were found from past cohort studies or meta-analyses. Aubertin et al. reported on the initial probabilities of treatment and stable CAs as a result of watchful waiting.[9] The probability of rupture was reported by Cagnazzo et al. in a large systematic review.[10] The probability of sudden death following rupture was found by Huang and van Gelder in a large meta-analysis studying the probability of death following aneurysmal rupture.[11] The probability of different health states (death, no disability, mild disability, and moderate to severe disability) following treatment for CA rupture was reported in a systematic review by Nieuwkamp et al.[12] The probability of vasospasm was reported in a chapter by Singh et al.[23] The probabilities of complications from DSA were reported by a separate cost-effectiveness analysis by Jethwa et al., 2013.[13] All probabilities and their descriptions can be found in Supplementary Table 1.
Statistical Analysis
Incremental cost-effectiveness ratios (ICER) for our model illustrate the cost per QALY of each arm’s intervention. ICERs were calculated annually for each respective model arm and compared to determine the cost per QALY gained by use of a blood-based diagnostic test. One-way sensitivity analyses were run to determine the price point where a blood-based diagnostic test is dominant (cost saving with superior outcomes) compared to SOC. Further 1-way sensitivity analyses were run to generate a tornado diagram and determine which variables significantly influenced our model’s output.
High-Risk Markov Model
The blood test arm separates CAs by size (small <7mm, medium 7-12mm, Large 13+mm). Since previous studies have shown that CA growth is associated with increased risk of rupture, this model assumes that growing CAs are treated. CA growth rates based on size were obtained from existing literature.
A second Markov Model was created to determine the cost-effectiveness of an annual blood test for CA identification in high-risk groups. The high-risk Markov model (Figure 1B) illustrates the hypothetical screening utility of a blood based diagnostic test to detect CAs. This model is similar to the base model described above, except for annual rates of treatment and rupture for both arms (blood test vs SOC). Additionally, the blood test arm separates CAs by size (small [<7mm], medium [7-12mm], and Large [>13mm]). Previous studies have shown that CA growth is associated with increased risk of rupture, this model assumes that growing CAs are treated. Aneurysm growth rates based on size were obtained from existing literature.[24] Annual SOC cost was $0 based on the assumption that there would be no consistent active surveillance efforts in these populations. As in the previous model, it was assumed that these patients would either be undiscovered, present with rupture, or be discovered incidentally (probabilities and costs are presented in Supplementary Table 1). Hypothetical blood test sensitivity was assumed to be 95% with a specificity of 100%. These values were chosen based on similar test accuracies reported for high-sensitivity qualitative troponin (98%, 100%, respectively)[25] and Cologuard test for stage I colorectal cancer (98.2%, 95%, respectively).[26] Markov model analysis was also performed for three high-risk groups based on annual CA development rate. Annual rates of CA development were obtained from the literature for various high-risk groups.[27, 28] Markov model analyses were performed at annual CA development rates of 4%, 5%, and 9% to correspond with patients with one family member with an CA, patients who smoke, and patients with 2 or more family members with an CA, respectively. Price per test was calculated with a WTP of $50,000 per QALY, which is standard for cost effectiveness analyses.[29] One-way sensitivity analyses, cost-effectiveness analyses, and respective graphical output were performed similar to the base case.
Detection Rate of Aneurysms in High-Risk Populations
To estimate the average number of family members of patients with known aneurysms who could be screened, and who also happen to harbor an aneurysm, we used published demographic data and analysis available from the NIS database. We used the total number of treated aneurysm patients between 2004-2014 = 243,754 patients[30], and the average number of family members in the US (3.13) to arrive at 762,953 family members of which 30,518 (4%) could also harbor an aneurysm.[30] To estimate the number of new aneurysm diagnoses in smokers, we used published data for the prevalence of tobacco use in the US (12.5%) and CA rate in smokers (5%). The number of new aneurysm diagnoses was then calculated as the percent of screened population was varied.