Which unruptured cerebral aneurysms should be treated?

The model.

Treatment of unruptured aneurysms creates a short-term risk of procedural complications followed by a reduction in the risk of SAH over the remaining lifetime of the patient. To model the protracted effect of elevated risk on the average life of a patient given competing mortality risks, we used Markov models.17,18 We designed these models to track a hypothetical cohort of patients until all are dead, placing them into one of four health states: healthy, mild disability (independent), moderate–severe disability (dependent), or dead. Yearly rates of SAH and all-cause mortality were applied to the cohort, and the number of patients in each health state recorded for each year. Similar models were used for each treatment option, and the total cost and quality-adjusted life years (QALYs) per hypothetical patient were totaled for each model. Although standard practice in cost–utility analysis is to compare each treatment to the next best option, we chose to compare both treatments to no intervention as coiling is only practiced in large centers and data comparing clipping to coiling are sparse.

We assumed that aneurysms were discovered prior to entry into the model, so we did not consider screening as part of the clinical scenarios. The three treatment options compared were no treatment, surgical clipping, and endovascular coil embolization (coiling). Models for each of these options were constructed from decision trees (figure 1). The no-treatment option does not involve initial procedure-related risk, but there is an ongoing risk of SAH. Surgical clipping is associated with an initial risk of death or disability due to surgical complications, but if the patient survives, we assume there is no subsequent risk of SAH because the aneurysm is fully treated. For endovascular coiling, we have included both an early, procedure-related risk and a delayed risk of SAH; lack of long-term follow-up raises the possibility that coil embolization reduces but does not eliminate risk of SAH. Surgical and endovascular complications produce focal and diffuse brain injury resulting in neurologic deficits similar to those seen after SAH, so we assumed that the disability states were interchangeable.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Decision trees for no treatment, clipping, and coiling. Each decision tree was converted to a Markov model. SAH = subarachnoid hemorrhage.

We modeled all direct medical costs associated with the clipping and coiling procedures and with SAH (hospital, physician fees, outpatient, rehabilitation, and nursing home/home care). We have taken a societal perspective; i.e., costs are included regardless of who pays.19 A discount rate of 3% was applied to all costs and benefits.19

Model inputs were derived from the literature and a study of costs at our institution (table 1). Rupture rates were taken from the ISUIA. For other inputs, the following hierarchy was followed according to availability of each type of data: the most recent meta-analysis; prospective and cohort studies; case-control and case-series data; estimations based on consultation with experts in the field. For sensitivity analysis, the 95% confidence intervals (CI) for ISUIA results, meta-analyses, and cohort studies were used to establish the input range tested. For other inputs, conservatively large CIs were estimated.

For multivariable Monte Carlo sensitivity analysis it was also important to define the distribution of the range. We assumed a distribution was normal if an input was derived from a study that reported normal statistics. For other variables, we assumed that each input was a best estimate of a true population value, and values more distant from this best estimate would be less likely to represent the true population mean, but no value would be impossible unless it was precluded by physical reality (e.g., rate of rupture less than zero). To reflect this, we modeled input variability using the normal distribution if the lower and upper estimates of the ranges were symmetrical around the best estimate and the log-normal distribution if skewed.20,21

Scenarios.

Aneurysm rupture rates vary depending on the size of the aneurysm and whether there is a history of SAH from a different aneurysm.16 To account for variable aneurysm characteristics and include the possibility that an untreated aneurysm may produce disutility from symptoms, we modeled 8 different clinical scenarios (table 2). We chose scenario A, a 50-year-old woman with an asymptomatic unruptured aneurysm smaller than 10 mm and no history of SAH, for the primary analysis (the base case) because it was most common in the ISUIA16 and asymptomatic aneurysms had been the focus of prior cost–utility analyses.13,14

Risks.

Cohort age was taken as 50 years because this approximated the average age of treated patients in the ISUIA (see table 1). All-cause mortality rates were taken from US mortality figures from 1990 through 1995, specific for age and sex.22 Age-specific mortality was determined within 10-year categories and after age 85 years by geometric interpolation and extrapolation, respectively. For disabled patients, the relative risks of mortality were taken from rates in disabled head trauma patients23; similar to victims of procedural complication and SAH, their excess mortality should be defined by neurologic deficits rather than underlying risk factors (such as hypertension and atherosclerosis in ischemic stroke patients). These relative risks were similar to those reported for minor and severe disability after stroke compared to no disability.24,25

The yearly risk of SAH for untreated aneurysms was taken from the ISUIA for each clinical scenario.16 Case fatality and permanent dependent disability rates for SAH were taken from a recent meta-analysis.26 Rates of mild disability were calculated from 1-year follow-up data on SAH survivors.27

Mortality and permanent complication rates for surgical clipping were obtained from a recent meta-analysis (clipping, 2.6% mortality and 10.9% permanent morbidity).2 In this meta-analysis, 53% of patients with a morbid complication were dependent on others for care. These results were very similar to those of the ISUIA, in which surgery-related death was 3.2% and morbidity was 12% at 1 year.16

Published reports of results of endovascular coiling are limited. Therefore, we relied on a single, large study of coiling outcomes in the University Health Systems Consortium in which in-hospital mortality was 0.4% (see table 1).28 This study did not measure long-term outcomes, but compared rates of discharge to nursing homes/rehabilitation hospitals for surgically and endovascularly treated patients. Because this short-term outcome is likely to correlate well with permanent disability,28 the multivariable odds ratio (OR) of nursing home/rehabilitation hospital discharge (clipping to coiling: OR 2.1, 95% CI 1.4 to 3.3) was used to estimate the rate of mild and moderate/severe complications after coiling using surgical outcomes as a reference.2 We assumed that coiled aneurysms provided 90% protection against SAH based on results from treatment of ruptured aneurysms.29-33 We relied on data for previously ruptured aneurysms for this model input because rupture rates of untreated, unruptured aneurysms are too small to expect useful information from available data.16

Costs.

We determined costs for clipping and coiling of unruptured aneurysms at the hospital of the University of California, San Francisco (UCSF) from 1994 to 1997. Because of the need to maintain confidentiality in bargaining for health contracts, UCSF asked that we publish only aggregate costs. Thus, all reported cost figures combine hospital costs and physician professional fees, converted to 1997 dollars using the Medical Care category of the Consumer Price Index.34

Our inclusion criteria were: age greater than 18 years, no other aneurysm treatment within 2 months, no SAH from another aneurysm within 6 months, and no coexistent arteriovenous malformation. Thirty-three clipped patients and 70 coiled patients met these criteria. Hospital costs were determined for each unit of activity and consumption using a state-of-the-art cost accounting system implemented throughout UCSF (Transition Systems Incorporated, Boston, MA). Costs of treatment were determined for the entire inpatient course of therapy, including follow-up admissions for further treatment, through December 1997. Inclusion of subsequent admissions was important because it accounted for 20% of coiling hospital costs and 8% of clipping costs.

We estimated physician professional fees for clipping and coiling and associated diagnostic procedures using the Medicare fee schedule, which is also used by Medicaid and many health maintenance organizations to set reimbursement rates for procedures. We used the 1997 fee schedule for San Francisco. We assumed that aneurysms were already identified in all cases and that angiograms were not performed in untreated patients as further aneurysm characterization would not be required. We assumed that two complete 3-vessel angiograms were performed for coiled aneurysms (one before and one follow-up). For clipped cases, we assumed that a 3-vessel angiogram was performed before treatment and a single vessel angiogram after. Professional fees for inpatient care were determined from mean lengths of stay in the University Health Systems Consortium comparative study (9.1 days for clipping and 4.3 days for coiling)28 using high-acuity initial and follow-up visit reimbursement rates.

To determine the cost of subarachnoid hemorrhage, we used an estimate of 12% out-of-hospital mortality from a population-based retrospective cohort study and assumed these patients accrued no cost.35 For patients arriving at the hospital, we estimated hospital costs by determining the average cost of all 70 SAH patients treated at UCSF during fiscal year 1997. Professional fees were determined as described above except we assumed that all patients had two complete angiograms. For costs after the acute hospitalization, we used published figures for 1991 Medicare costs during the first 6 months after a stroke,36 escalated to 1997 dollars. The annual direct costs for caring for disabled patients were taken from data on long-term care in stroke patients.37

Utility.

Benefits were measured in QALYs. Death was assigned a utility of 0, and no disability from the procedures or SAH a utility of 1. Individuals without aneurysm-related problems, of course, may have other medical problems and associated disutilities, upon which our disutilities would be additive. We could find no empirical studies of the disutility of disability due to SAH or the procedures, but we did find studies of disability due to brain injury. Gage and colleagues used time tradeoff and standard gamble methods to obtain relative utility valuations for permanent disability states after stroke for 83 patients with atrial fibrillation.38 Median utilities for mild strokes and an average of moderate and major strokes were used to define the utilities for mild and moderate–severe disability health states, respectively.

The psychological disutility associated with concern that an aneurysm could rupture at any time is important but unmeasured. Prior cost-effectiveness analyses of unruptured aneurysm treatment have assumed that untreated patients have a utility of 0.95 due to worry.13,14 Because this burden should depend on rupture rates as long as physicians are appropriately reassuring, we assumed that the utility for untreated patients would vary with the rupture rates as follows: RU = 1 − 5 × [rupture rate]. The factor 5 is an estimate of the emotional dimension of living with a low-risk, high-consequence condition.39 For an aneurysm with 1% yearly rupture rate, this reproduces the prior estimate of 0.95 and is very similar to the mildly impaired emotional state (“occasionally fretful, angry, irritable, anxious, depressed, or suffering” = 0.93) in the Health Utilities Index Mark 2.40

For pain or compressive symptoms produced by an unruptured aneurysm, we estimated a utility of 0.84 from valuation of the state with frequent pain relieved by medications described in Health Utilities Index Mark 2.40 We assumed that treatment completely relieved symptoms, as has been shown for the majority of surgical and endovascular cases.41-44

Sensitivity analysis.

Univariate sensitivity analysis was performed by varying each input variable throughout its full range while holding other variables constant. Multivariable sensitivity analysis was performed using a Monte Carlo simulation with all input variables (except age and discount rate) and 10,000 trials (Crystal Ball, version 4.0, Decisioneering, Denver, CO). Ninety-five percent CIs were obtained nonparametrically from the output distribution. We considered an upper limit of the 95% CI less than $100,000/QALY to be strong evidence of cost-effectiveness. The $100,000/QALY criterion was chosen as a reasonable cut-point to define moderate evidence for adoption and appropriate utilization of a medical technology.45