Effectiveness of Diagnostic Strategies in Suspected Delayed Cerebral Ischemia
A Decision Analysis
Background and Purpose—Delayed cerebral ischemia (DCI) is a serious complication after aneurysmal subarachnoid hemorrhage. If DCI is suspected clinically, imaging methods designed to detect angiographic vasospasm or regional hypoperfusion are often used before instituting therapy. Uncertainty in the strength of the relationship between imaged vasospasm or perfusion deficits and DCI-related outcomes raises the question of whether imaging to select patients for therapy improves outcomes in clinical DCI.
Methods—Decision analysis was performed using Markov models. Strategies were either to treat all patients immediately or to first undergo diagnostic testing by digital subtraction angiography or computed tomography angiography to assess for angiographic vasospasm, or computed tomography perfusion to assess for perfusion deficits. According to current practice guidelines, treatment consisted of induced hypertension. Outcomes were survival in terms of life-years and quality-adjusted life-years.
Results—When treatment was assumed to be ineffective in nonvasospasm patients, Treat All and digital subtraction angiography were equivalent strategies; when a moderate treatment effect was assumed in nonvasospasm patients, Treat All became the superior strategy. Treating all patients was also superior to selecting patients for treatment via computed tomography perfusion. One-way sensitivity analyses demonstrated that the models were robust; 2- and 3-way sensitivity analyses with variation of disease and treatment parameters reinforced dominance of the Treat All strategy.
Conclusions—Imaging studies to test for the presence of angiographic vasospasm or perfusion deficits in patients with clinical DCI do not seem helpful in selecting which patients should undergo treatment and may not improve outcomes. Future directions include validating these results in prospective cohort studies.
- brain ischemia
- decision support techniques
- diagnostic imaging
- perfusion imaging
- subarachnoid hemorrhage
- vasospasm, intracranial
An important contributor to poor outcome after aneurysmal subarachnoid hemorrhage (ASAH) is delayed cerebral ischemia (DCI).1 DCI occurs in ≈30% of patients within the first 2 weeks after ASAH. It is defined clinically as evidence of a new neurological deficit or decrease in level of consciousness, which may be reversible or may progress to infarction, and potentially even to death.2 Efforts to identify patients at risk for DCI have focused on the concept of vasospasm.3 However, not all patients with angiographic vasospasm develop DCI, and not all patients with DCI have preceding angiographic vasospasm.4 Current AHA practice guidelines advocate monitoring for the presence of angiographic vasospasm, and support hemodynamic augmentation with medical therapy before consideration of intra-arterial therapies. However, the best methods of prediction, diagnosis, and management of DCI remain uncertain.5 Current practice in many centres is to perform imaging studies to diagnose or exclude angiographic vasospasm or perfusion deficits in patients suspected to have DCI. Substantial debate exists on the best imaging method for angiographic vasospasm or perfusion deficits, and whether such studies should be performed before initiating treatment to prevent the development of infarction. Conventional angiography is still considered the gold standard for angiographic vasospasm, but its prediction for DCI is not better than moderate,6 and it carries a small complication rate related to the interventional aspect of the procedure, which may in itself result in permanent neurological deficit.7 Alternative approaches include computed tomography angiography (CTA), which can noninvasively image intracranial vessel calibre but is less accurate in distal vessels than conventional angiography,8 and computed tomography perfusion (CTP), which delineates perfusion deficits within the brain parenchyma, and thus more directly identifies areas at risk of infarction.9
We used decision analysis models to identify the best strategy for preventing cerebral infarction in an ASAH patient with clinical diagnosis of DCI. Options included initiating medical therapy in all patients with clinical diagnosis of DCI, or reserving management for those with imaging-based confirmation of angiographic vasospasm or perfusion deficits.
We used TreeAge software to create Markov decision models with 3-month cycles, allowing lifetime follow up of the study cohort. Test characteristics, treatment characteristics, probabilities of adverse events, and probabilities and utilities of outcome states were derived from the literature (online-only Data Supplement); where no data existed, assumptions were made based on expert consensus (G.J.E.R. and T.K.). All data obtained for the model were independent of patient sex.
The base case subject was a 55-year-old patient with ASAH and clinical diagnosis of DCI. The latter was defined by the presence of new neurological deficit or reduced state of consciousness, with exclusion of other possible causes (including hydrocephalus, rebleeding, or metabolic causes). The base case scenario assumed the presence of a single aneurysm that was properly occluded without remaining risk of rebleeding.
The strategies under consideration were either to treat all patients immediately or to first perform digital subtraction angiography (DSA), CTA, or CTP to confirm the presence of angiographic vasospasm or perfusion deficits. Separate models were created for analysis of DSA/CTA and CTP, as these modalities test for different phenomena (angiographic vasospasm versus perfusion deficits intended to reflect measurement of DCI directly; Figure 1A and 1B). If diagnostic tests were performed first, only those patients that tested positive were treated. DSA is considered the gold standard for angiographic vasospasm diagnosis, and so therefore sensitivity and specificity were assumed to be 99% in the base case scenario. CTA was included as an alternative, noninvasive option for imaging angiographic vasospasm. CTP is currently the only perfusion imaging modality used in the clinical setting as a means to monitor for or diagnose DCI. Transcranial Doppler ultrasound was not included as an imaging option because it is generally used as a screening tool for angiographic vasospasm in ASAH patients, and subsequently confirmed with another diagnostic test before instituting therapy.5
Treatment was standardized across all models, and consisted of induced hypertension, as per current AHA practice guidelines for ASAH management.5,10 Patients were assumed to be free of treatment contraindications (baseline hypertension or cardiac vulnerability). The use of angioplasty as an alternative or second-line management was not considered in the model, as the evidence supporting its implementation is currently not as strong as that for induced hypertension.5
The entire cohort began in the baseline health state, defined as post-ASAH with clinical diagnosis of DCI. At this point, in the model testing for angiographic vasospasm, patients could either be treated with induced hypertension immediately, or first undergo DSA or CTA for diagnosis of angiographic vasospasm (Figure 1A); in the model testing for perfusion deficits (or DCI directly), patients could either be treated immediately or first undergo CTP for diagnosis of perfusion deficits (Figure 1B). In the diagnostic testing arms, only true and false positives were treated with induced hypertension. Adverse effects from the test or treatment were incorporated at their respective stages (Figure 2).
Within the angiographic vasospasm testing scenario, 2 models were created to account for potential differences in treatment efficacy. The first model was based on the assumption of no benefit from induced hypertension in patients without angiographic vasospasm. This is a worst-case scenario model, where in the diagnostic arms, false positives were subject to adverse events, but not benefits, of treatment. This was also true for the patients without angiographic vasospasm who were immediately treated. The second model allowed for some efficacy of induced hypertension in patients without angiographic vasospasm; we assumed 50% efficacy relative to that in patients with angiographic vasospasm, based on the fact that induced hypertension might theoretically target and improve microcirculatory dysfunction.
In the scenario on testing for perfusion deficits, only one model was created with respect to treatment efficacy. In the absence of DCI, we assumed no treatment benefit of induced hypertension. Efficacy of hypertensive treatment in patients with perfusion deficits (ie, preventing progression to infarction) was assumed to be similar to that in angiographic vasospasm because there is no existing data about the efficacy of induced hypertension specifically in patients with perfusion deficits.
Health States and Markov Subtrees
Health state possibilities were determined by the outcome of the disease (infarction or no infarction), and delineated on the basis of neurological disability (well/mild–moderate disability/severe disability) or death. Superimposed on the disability health states were the possibilities of having experienced permanent complications from the test or treatment that resulted in different utilities and mortality rates. From the testing stage, the important complication resulting in long-term implications was chronic renal insufficiency; for the treatment, this was myocardial infarction.
From the second cycle onwards, the cohort subgroups transitioned within a Markov subtree related to their specific combination of heath states (Figure 3). For simplicity, we assumed that patients could only improve in terms of disability, and not worsen, because the ASAH with DCI was considered to be a 1-time insult that did not recur after the initial 3 months. On the basis of the literature, if improvement occurred, this was most likely to take place in the early cycles postinsult. In addition, patients in all health states could die, and the probability of death per cycle was derived from reported standardized mortality ratios (SMRs) for the specific health states, applied to the baseline risk of mortality in the general population. SMRs were assumed to be multiplicative for any combination of health states (eg, in a patient with ASAH with severe disability in the context of chronic renal insufficiency, the SMR was derived from multiplying the SMR for ASAH with severe disability and the SMR for chronic renal insufficiency).
The outcomes assessed were survival in terms of life-years and quality-adjusted life-years (QALYs). Utilities were assigned on a range of 0 to 1 according to convention, with 0 being dead and 1 being perfect health. Utilities related to long-term test or treatment complications, specifically chronic renal insufficiency and myocardial infarction, were also obtained from the literature and incorporated into the analysis. Utilities were assumed to be multiplicative for any combination of health states.
One-way sensitivity analyses were performed to assess the robustness of all models, as well as to determine how the model outputs varied within clinically plausible ranges for the variables included. Two- and three-way sensitivity analyses were also done to assess the relationship between different variables and the outcome. The most clinically relevant sensitivity analyses were established a priori.
For the angiographic vasospasm testing scenario, where the underlying relationship between vasospasm and DCI and the effectiveness of medical therapy is unknown, the key sensitivity analyses focused on modeling the uncertainty about the efficacy of treatment in preventing infarction in patients with and without angiographic vasospasm. This was considered clinically relevant because a wide difference in efficacy between patients with and without angiographic vasospasm with increased benefit in the vasospasm group—as assumed in the base case for both model variations—would theoretically favor testing for angiographic vasospasm. As a result, the analyses thought to be most relevant included 1-way sensitivity analyses of the treatment efficacy in each group (vasospasm and nonvasospasm) modeled by altering the probability of infarction post-treatment in each group; a 2-way sensitivity analysis evaluating the probability of infarction post-treatment in vasospasm versus nonvasospasm; and a 3-way sensitivity analysis superimposing the range of angiographic vasospasm prevalence on the 2-way sensitivity analysis of post-treatment infarction probability in each group. The last analysis was felt to be relevant because altering the angiographic vasospasm prevalence is equivalent to changing the pretest probability of angiographic vasospasm detection and reflects the uncertainty in the relationship between clinical symptoms of DCI and presence of angiographic vasospasm.
For the perfusion deficit testing scenario, the relevant sensitivity analyses focused on modeling the efficacy of treatment in DCI, as well as the uncertainty about DCI prevalence in symptomatic patients (pretest probability). When testing directly for DCI via assessment for perfusion deficits, altering the pretest probability reflects the uncertainty in the accuracy of clinically diagnosing DCI, which is particularly relevant in sicker patients with more severe SAH.
Base Case Analysis
In the first angiographic vasospasm model, which assumed no treatment benefit of induced hypertension in nonvasospasm patients, Treat All and DSA were equivalent strategies, both slightly superior in QALYs and life-years to CTA (Table). In the second angiographic vasospasm model, which assumed half the benefit of treatment in nonvasospasm patients relative to those with angiographic vasospasm, Treat All became the superior strategy (Treat All, 9.81 QALYs; DSA, 9.78 QALYs), dominating over diagnostic testing options.
In the perfusion deficits model Treat All was the dominant strategy (Treat All, 10.08 QALYs; CTP, 9.93 QALYs).
Using Treat All as the preferred strategy, the outcome probability distributions for all models were within the reported outcome ranges of published series of ASAH patients (online-only Data Supplement).11–14
Sensitivity Analyses: Angiographic Vasospasm Models
Extensive 1-way sensitivity analyses were performed for both models, and tornado diagrams of 1-way sensitivity analyses were built using QALYs as outcomes (online-only Data Supplement). Only a few variables had threshold values that influenced the decision. In the first angiographic vasospasm model, thresholds were found in 10 variables, however, the difference between decision arms remained <0.1 QALYs for all variables, thus the leading strategies Treat All and DSA remained equivalent. In the second angiographic vasospasm model, thresholds were found for 2 variables, but there was a significant difference between strategies only with 1: the probability of infarction with treatment in no vasospasm. In this case, Treat All was the dominating strategy over the majority of the range, with more evident dominance with lower rate of infarction (higher treatment efficacy); however, when the probability of infarction increased to 0.23 in nonvasospasm (lower treatment efficacy), Treat All and DSA became equivalent (0.23 threshold value, Figure 4A). Changing the efficacy of treatment in patients with angiographic vasospasm did not change the preferred strategy (Figure 4B).
A 2-way sensitivity analysis comparing the efficacy of treatment in vasospasm and no vasospasm also demonstrated that Treat All was the dominant strategy over the range of possibilities for treatment effectiveness, specifically as long as a slight effect of treatment in both vasospasm and nonvasospasm was present (Figure 4C). In 3-way sensitivity analysis, with increasing prevalence or pretest probability of angiographic vasospasm, the dominance of Treat All over the diagnostic testing strategies increased (Figure 4D). When varying the probability of adverse events from treatment, Treat All was always dominant; however, it became more dominant with lower probability of complications (data not shown).
Sensitivity Analysis: Perfusion Deficits Model
Extensive 1-way sensitivity analyses were performed for this model, as represented in the tornado diagram (online-only Data Supplement). Only one variable had a threshold value that influenced the decision—the probability of death without infarction with DCI—however, the difference in QALYs between treatment arms remained low (≈0.2 QALYs; Figure 5A). A 2-way sensitivity analysis altering the efficacy of treatment (probability of infarction in treated versus untreated DCI) demonstrated that Treat All remained dominant over the majority of the range of input values, with CTP only preferred if treatment was extremely ineffective in DCI (Figure 5B). Superimposing a variable prevalence of DCI on the latter analysis demonstrated that Treat All became more dominant as the prevalence (or pretest probability) of DCI increased (Figure 5C). When varying the probability of adverse events from treatment, Treat All became more dominant with lower probability of complications (data not shown).
Our modeling study shows that in patients with ASAH and clinical suspicion of DCI, treating all patients without preceding diagnostic test and treatment allocation according to results of DSA are equivalent strategies under the assumption that treatment is ineffective in patients without angiographic vasospasm. However, when a moderate treatment effect was assumed in patients without angiographic vasospasm, treating all patients without preceding diagnostic test became the superior strategy. Finally, as long as some efficacy of treatment was present in DCI, treating all patients is superior to reserving treatment for those with documented perfusion deficits on CTP. Extensive 1-way sensitivity analyses demonstrated that the models were robust within clinically plausible variable ranges, and selected 2- and 3-way sensitivity analyses reinforced dominance of the Treat All strategy with variation of clinically important disease and treatment parameters. Thus, in patients who on the basis of clinical assessment are suspected of having DCI, testing for the presence of angiographic vasospasm or perfusion deficits before initiating hypertensive treatment did not lead to improved outcomes, and therefore did not add value to the management of these patients.
Currently, the effectiveness of imaging studies in patients suspected to have DCI is uncertain. A review discussing the utility of clinical assessment and various imaging modalities for diagnosis and monitoring of angiographic vasospasm and DCI offered several conclusions, such as (1) clinical assessment alone was suboptimal because of the poor neurological status of many patients with ASAH, (2) transcranial Doppler was a valuable method for diagnosis of angiographic vasospasm involving the middle cerebral artery, however, less useful in assessment of other vessels, (3) CTA compared favorably with the gold standard of DSA for angiographic vasospasm diagnosis, and (4) early CTP deficits may predict development of DCI, whereas late CTP deficits may diagnose DCI.15 An important limitation of this review is that it studied methods to monitor angiographic vasospasm, whereas angiographic vasospasm is poorly related to DCI. To our knowledge, only one study has examined the ability of CTA imaging to diagnose DCI—rather than angiographic vasospasm—in symptomatic patients post-ASAH.16 In that study, unenhanced computed tomography had lower sensitivity but higher specificity than CTA, but CTP had the best sensitivity (0.84) and specificity (0.79) for diagnosis of DCI defined retrospectively on the basis of clinical criteria. The use of imaging studies to predict development of DCI remains unclear. A meta-analysis on the use of CTP demonstrated that while it was useful in diagnosis of DCI in symptomatic patients, it was not useful to predict development of DCI when performed on admission in ASAH patients.17
Most imaging techniques are designed to identify the presence or absence of angiographic vasospasm, which may be justified on the basis of a proposed relationship between angiographic vasospasm, DCI, and poor outcome. However, the strength of association between the presence of angiographic vasospasm and the development of DCI and poor outcome is moderate. Although the presence of angiographic vasospasm is associated with infarction,6,18 it is neither a necessary nor a sufficient factor for predicting DCI.6 Our model findings are supported by the fact that treatment strategies aimed at reducing angiographic vasospasm have failed to improve clinical outcome19; as such, a management strategy whereby treatment is given to patients selected by presence of angiographic vasospasm would be expected to be ineffective. Given the lack of a therapy that improves outcome by treating angiographic vasospasm, there is in our view currently no indication to perform angiographic imaging in patients with suspected DCI after ASAH.
Given the uncertainty about angiographic vasospasm as an appropriate imaging marker for DCI, emphasis has shifted toward evaluating the utility of CTP in this context. Although CTP is a theoretically promising technique given its ability to image effects on perfusion at the level of the microcirculation, according to 2 recent meta-analyses, its sensitivity for the diagnosis of DCI in the symptomatic period remains in the range of 70% to 95% and its specificity in the range of 63% to 82%.17,20 The diagnostic performance of CTP therefore remains suboptimal for detection of DCI; this likely contributes to its inability to improve outcomes when used to dictate management decisions, as demonstrated by our modeling. Further work on standardizing CTP acquisition and interpretation may be helpful to optimize diagnostic accuracy of this technique.
Several limitations to the model interpretation must be addressed. First, limitations to external generalizability exist on the basis of the selected baseline cohort characteristics (eg, exclusion of multiple aneurysms and of rebleeding events). Second, although most model data were derived from the literature, several probability values had to be estimated because of lack of published evidence. In these instances, estimates were obtained on expert consensus (G.J.E.R. and T.K.), and wide ranges were used in sensitivity analysis to offset the uncertainty, which was not found to affect the conclusions. A third limitation is that we did not explicitly address the subgroup of SAH patients who cannot be assessed clinically because of a poor clinical condition (World Federation of Neurologic Surgeons SAH grade V) or cardiopulmonary dysfunction necessitating ventilation and sedation. These patients pose a different question, and a modeling approach to assess the value of imaging to select patients for treatment in this subgroup would be particularly interesting. However, currently there is insufficient data available in the published literature to create a robust model specific to this subgroup. A final limitation is that we did not formally include costs in our analysis. However, as DSA, CTA, and CTP incur direct costs and do not lead to improved outcome, adding costs would not alter the clinical implication of the study.
In patients with a clinical diagnosis of DCI, imaging studies to test for the presence of angiographic vasospasm or perfusion deficits do not seem helpful in selecting which patients should undergo treatment and may not improve outcomes. Further research should focus on validating these conclusions in prospective cohort studies.
Presented in part at the Eastern Neuroradiological Society meeting, White Sulphur Springs, WV, August 15–18, 2013.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.005916/-/DC1.
- Received April 22, 2014.
- Revision received October 26, 2014.
- Accepted October 28, 2014.
- © 2014 American Heart Association, Inc.
- Vergouwen MD,
- Vermeulen M,
- van Gijn J,
- Rinkel GJ,
- Wijdicks EF,
- Muizelaar JP,
- et al
- Vergouwen MD,
- Ilodigwe D,
- Macdonald RL
- Connolly ES,
- Rabinstein AA,
- Carhuapoma JR,
- Derdeyn CP,
- Dion J,
- Higashida RT,
- et al
- Rabinstein AA,
- Friedman JA,
- Weigand SD,
- McClelland RL,
- Fulgham JR,
- Manno EM,
- et al
- Greenberg ED,
- Gold R,
- Reichman M,
- John M,
- Ivanidze J,
- Edwards AM,
- et al
- Molyneux A,
- Kerr R,
- Stratton I,
- Sandercock P,
- Clarke M,
- Shrimpton J,
- et al
- Washington CW,
- Zipfel GJ
- Dankbaar JW,
- de Rooij NK,
- Velthuis BK,
- Frijns CJ,
- Rinkel GJ,
- van der Schaaf IC
- Etminan N,
- Vergouwen MD,
- Ilodigwe D,
- Macdonald RL
- Mir DI,
- Gupta A,
- Dunning A,
- Puchi L,
- Robinson CL,
- Epstein HA,
- et al