ORIGINAL RESEARCHFree Access

Minor Head Injury: CT-based Strategies for Management—A Cost-effectiveness Analysis

Abstract

Purpose

To compare the cost-effectiveness of using selective computed tomographic (CT) strategies with that of performing CT in all patients with minor head injury (MHI).

Materials and methods

The internal review board approved the study; written informed consent was obtained from all interviewed patients. Five strategies were evaluated, with CT performed in all patients with MHI; selectively according to the New Orleans criteria (NOC), Canadian CT head rule (CCHR), or CT in head injury patients (CHIP) rule; or in no patients. A decision tree was used to analyze short-term costs and effectiveness, and a Markov model was used to analyze long-term costs and effectiveness. n-Way and probabilistic sensitivity analyses and value-of-information (VOI) analysis were performed. Data from the multicenter CHIP Study involving 3181 patients with MHI were used. Outcome measures were first-year and lifetime costs, quality-adjusted life-years, and incremental cost-effectiveness ratios.

Results

Study results showed that performing CT selectively according to the CCHR or the CHIP rule could lead to substantial U.S. cost savings ($120 million and $71 million, respectively), and the CCHR was the most cost-effective at reference-case analysis. When the prediction rule had lower than 97% sensitivity for the identification of patients who required neurosurgery, performing CT in all patients was cost-effective. The CHIP rule was most likely to be cost-effective. At VOI analysis, the expected value of perfect information was $7 billion, mainly because of uncertainty about long-term functional outcomes.

Conclusion

Selecting patients with MHI for CT renders cost savings and may be cost-effective, provided the sensitivity for the identification of patients who require neurosurgery is extremely high. Uncertainty regarding long-term functional outcomes after MHI justifies the routine use of CT in all patients with these injuries.

© RSNA, 2010

Supplemental material: http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.2541081672/-/DC1

Introduction

Head computed tomography (CT) is increasingly being performed routinely in patients with minor head injury (MHI) for the rapid and reliable diagnosis of trauma-related complications after the injury (1). Although such complications are relatively infrequent after MHI (in 6%–10% of cases), they may require neurosurgical intervention (in 0.4%–1.0% of cases) (24). MHI is defined as blunt head injury with a Glasgow Coma Scale score of 13–15 at presentation. Because of the high population incidence of MHI (100–300 in 100000 people) (5), this injury places a substantial economic burden on society and the health care industry.

Compared with clinical observation, CT is less costly and equally effective for the clinical management of patients with MHI (68). By using prediction rules, one could reduce the use of head CT in patients with MHI by selecting only those patients at elevated risk of trauma-related complications to undergo CT (9). Although the published prediction rules—specifically, the New Orleans criteria (NOC) (3), Canadian CT head rule (CCHR) (4), and CT in head injury patients (CHIP) rule (10) (Table 1)—are reported to be 100% sensitive for the detection of trauma-related complications that require neurosurgery, associated 95% CIs are wide (83%, 100%) (11,12). The NOC and the CCHR have been substantially validated (11,12), and the CHIP rule is an adaptation of these two rules based on external validation findings. The use of these rules could reduce the use of CT by up to 37% (11,12), which would seem to be cost saving (13). However, given the wide 95% CIs for the sensitivity of these rules, the selective use of CT appears to introduce the risk of missing patients with trauma-related complications. Although this risk is expected to be small, the consequences in terms of lost quality of life may be substantial, potentially outweighing the cost savings of selective CT use. In this study, our purpose was to assess the cost-effectiveness of using selective CT strategies compared with performing CT in all patients with MHI. Furthermore, we evaluated whether further research to reduce uncertainty regarding patient outcomes would be required and/or justified.

Table 1 Performance Parameters and CT Criteria for Three Prediction Rules

Table 1

Note.—Variables were entered into the model as beta distributions, with the exception of diagnostic odds ratio, which was entered as a log-normal distribution. GCS = Glasgow Coma Scale. Sources: reference 10 for CHIP rule and reference 11 for NOC and CCHR.

*Data are point estimates, with 95% confidence intervals (CIs) in parentheses.

CT-detected lesion was defined as any intracranial traumatic finding on CT images, including depressed skull fracture but excluding isolated linear skull fracture.

In the model, specificity was modeled as a function of the sensitivity and the diagnostic odds ratio to reflect the inverse relationship between sensitivity and specificity in the (probabilistic) sensitivity analyses.

§The 95% CIs reported in the literature are wider than those used in the current study model. In studies to evaluate clinical prediction rules for selective CT, any intracranial trauma-induced lesion detected at CT is considered a proxy for a lesion that requires neurosurgery, because occurrence of the latter lesion is rare. Therefore, in the (probabilistic) sensitivity analyses, we linked the sensitivity for the detection of any intracranial trauma-induced lesion and the sensitivity for the detection of lesions requiring neurosurgery by using proportional odds ratios (Appendix E2 [online]), allowing these sensitivities to change simultaneously. We modeled the value of the sensitivity for the detection of lesions requiring neurosurgery to be at least that of the sensitivity for the detection of any intracranial trauma-induced lesion. In a secondary analysis, we repeated the (probabilistic) sensitivity analysis, allowing these two sensitivities to vary independently.

In patients with high risk, CT scanning is deemed mandatory; in patients with medium risk, CT scanning is recommended, but close clinical observation is an alternative.

#CT scanning is indicated in patients who meet one major criterion or two minor criteria.

Materials and Methods

Study Population

No industry provided support for this retrospective study, and the authors’ work was independent of the funding organizations. The internal review board of Erasmus MC-University Medical Center approved this study, and written informed consent was obtained from all interviewed patients. We used data from the CHIP Study, in which data on 3364 consecutive patients with MHI at four Dutch university hospitals were prospectively collected between February 11, 2002, and August 31, 2004 (10,11). All patients had undergone head CT. Data on the clinical course and follow-up of all patients with intracranial trauma-related findings were retrieved from medical records (14). Long-term functional outcomes, in terms of Glasgow Outcome Scale (GOS) (15) and quality-of-life (EuroQOL-5D) (16,17) scores, were assessed by means of telephone interviews with a subset of these patients (14).

Decision Model

We developed a decision model to evaluate the use of CT in the setting of MHI (Fig 1). The following strategies were considered: (a) CT performed in all patients with MHI, (b) CT performed selectively according to the NOC, (c) CT performed selectively according to the CCHR, (d) CT performed selectively according to the CHIP rule, and (e) CT not performed (reference strategy). Short-term costs and effectiveness were modeled with a decision tree, and a Markov model was used to assess long-term (≥1 year after the injury) costs and effectiveness in terms of quality-adjusted life-years (QALYs). A glossary of the parameters and analyses used in our model is provided in the Appendix. All evaluations were started after the patients were clinically examined in the emergency department.

Simplified version of decision tree. In the first strategy, all patients undergo head CT, which does or does not depict intracranial trauma-induced lesion, which does or does not require neurosurgery. In selective CT strategies, CT is performed only if indicated by the prediction rule (NOC, CCHR, or CHIP). At CT, patients are identified as having (true-positive prediction) or not having (false-positive prediction) intracranial trauma-induced lesion. Patients with true-positive prediction are admitted for clinical observation (non-neurosurgery lesion) or neurosurgery (neurosurgery lesion). If CT is not indicated with prediction rule, patient is discharged without observation, in some cases incorrectly (false-negative prediction), depending on prediction rule sensitivity. Strategy 5 is reference strategy: All patients, with or without neurosurgery lesion, are discharged without CT. Markov model (M, 1-year cycle length) consists of four health states: death (GOS = 1), severe disability (GOS = 3), minor disability (GOS = 4), or full recovery (GOS = 5). Vegetative state (GOS = 2) was not modeled as it is very rare after MHI (1822) and patients infrequently survive first year (23). After first year, functional outcome was assumed to be stable (21,24); thus, there were no transitions between health states, other than from alive to dead.

Data Sources and Assumptions

We based our model on data from the CHIP Study ( tables 13Appendix E1 [online], Table E2 [online]) (10,11). We searched the literature (PubMed, English-language only) for data that were not available from the CHIP Study and for data to be used in the sensitivity analyses. All variables were entered into the model as distributions.

Table 2 GOS Score–based Probabilities of Long-term Functional Outcome

Table 2

Note.—Data are point estimates, with 95% CIs in parentheses. All variables were entered into the model as beta distributions. Sources: reference 14 for lesions not requiring neurosurgery and identified neurosurgery lesions and references 14 and 25 for missed neurosurgery lesions.

*Data on GOS score–based long-term (≥1 year after the injury) outcome were available for 92 patients, four of whom had a neurosurgery lesion. GOS scores: 1 = death, 3 = severe disability, 4 = moderate disability, 5 = full recovery. GOS score of 2 (vegetative state) was not observed in the CHIP Study and was therefore not modeled.

Estimates of probability of outcome after delayed diagnosis of a neurosurgery lesion were based on our estimates for an identified neurosurgery lesion and were adjusted for the delay in diagnosis on the basis of the data of Cordobes et al (25) by calculating the ratio of delayed diagnosis to correct identification of a neurosurgery lesion for each outcome.

Table 3 Quality-of-Life Estimates Determined in Current Study Population

Table 3

Note.—Quality-of-life data for 87 patients were available.

*GOS scores: 1 = death, 3 = severe disability, 4 = moderate disability, 5 = full recovery. GOS score of 2 (vegetative state) was not observed in the CHIP Study and was therefore not modeled.

Patient is well, with no lesion seen at CT.

CT was considered to be 100% sensitive for the identification of patients with a lesion requiring neurosurgery (7,2729) (ie, neurosurgery lesion). Patients with missed neurosurgery lesions were assumed to have returned to the emergency department (by ambulance) and to have undergone delayed CT and delayed neurosurgery. Their functional outcomes were estimated to be worse than those of the patients who were assigned the correct diagnosis and treated without delay (Table 2) (25,3035).

Patients with missed intracranial trauma-induced lesions that did not require neurosurgery were assumed to have been discharged home without clinical observation but otherwise to have had the same clinical course and recovery as those who were assigned the correct diagnosis with use of the prediction rule. A proportion of these patients were expected to present for medical evaluation again owing to prolonged or worsening complaints and undergo CT after being reassessed. We assumed that this proportion of patients was equal to the proportion of patients in our study population who underwent repeat CT for worsening symptoms (36). Patients without intracranial trauma-related CT findings were considered to have fully recovered with a subsequent maximal quality of life (18,26).

EuroQOL-5D (16) results were converted to quality-of-life utilities (17) and used to calculate the QALYs for each health state in the Markov model (Table 3). Outcome was discounted at 3% per year (3739).

Cost Data

We included direct health care and direct non–health care costs (Table E2 [online]) (3739). A cost analysis was performed to estimate the cost of head CT (40). All other costs were provided by the Dutch Health Care Insurance Board (41), converted to year 2006 costs based on the Dutch consumer price index (42), and reported in 2006 U.S. dollars ($1.256 = €1.00 [43]). All future costs were discounted at 3% per year (3739).

Outcome Measures

We calculated total costs for the first year after the injury, lifetime costs, QALYs, incremental cost-effectiveness ratios, and gains in net health benefit as compared with these values yielded with use of the reference strategy. The willingness-to-pay threshold was assumed to be $75000 per QALY gained.

Reference-Case Analysis

Data analysis was performed (M.S., M.G.M.H.; 1 year and 25 years experience, respectively) from a societal perspective with a lifetime horizon by using DATA TreeAge Pro 2007 Suite software (TreeAge Software, Williamstown, Mass). At reference-case cohort analysis, we evaluated the data of 10000 41-year-old men with MHI, who were representative of the typical patient in the CHIP Study. The prior probability of an intracranial trauma-induced lesion not requiring neurosurgery was 7.6%, and that of an intracranial trauma-induced lesion requiring neurosurgery was 0.5% (11,14).

Sensitivity Analysis

We repeated our analysis by using United Kingdom and Dutch recommendations for cost-effectiveness modeling (Appendix E2 [online]). We used one-way, two-way, and three-way sensitivity analyses to assess the effect of varying parameters across their 95% CIs (Appendix E2 [online]). Finally, we performed probabilistic sensitivity analyses by using two alternative modeling approaches (Appendix E2 [online]), whereby we drew from all variable distributions (Appendix E1 [online], Table E2 [online]) by using Monte Carlo simulations of 10 000 samples. We calculated the probability that performing CT in all patients—as compared with performing CT selectively according to each prediction rule for varying willingness-to-pay thresholds by using acceptability curves—was cost-effective. The EVPI (simulation with 20 000 samples), population EVPI (U.S. population, 5 years, discount rate of 3%), and partial EVPI (two-level simulation with 500 × 500 samples) were determined to assess the value of performing further research to decrease uncertainty related to the model parameters (44,45).

Results

Study Population and Parameter Estimates

Data on 3181 of the 3364 patients in the CHIP Study (11,14) were included. Parameter estimates, costs, and outcome data from the CHIP Study are summarized in tables 13, Appendix E1 (online), and Table E2 (online).

First-Year Costs and Outcomes

Total costs in the first year after the injury were highest for patients with a neurosurgery lesion who had not initially undergone CT ($44509 [Table 4]). Among the patients with a lesion that did not require neurosurgery, total costs in the first year were lower among those who did not undergo CT ($10108) than among those who did ($18890) (Table 4). The outcome 1 year after the injury was worst for the patients with a neurosurgery lesion who had not undergone CT owing to delayed diagnosis and surgery (71% survival, 0.47 QALY [Table 4]). The outcome of patients with a lesion that did not require neurosurgery was the same regardless of whether CT was performed (96% survival, 0.72 QALY [Table 4]).

Table 4 Estimated Total Costs, Survival, and Effectiveness First Year after MHI

Table 4

Note.—Difference in costs between patients with a neurosurgery lesion who had not initially undergone CT and those with a neurosurgery lesion who had undergone CT was due to the patient’s return to the hospital by ambulance ($588), reexamination in the emergency department ($138), an increased proportion of patients admitted for intensive care observation for an average of 5 days (100% versus 59%, $2234 per day), and an increased proportion (10% versus 6%) of patients discharged from the hospital to a nursing home as a result of poorer outcome for the rest of the year ($274 per day for 353 days). The difference in costs between patients with a non-neurosurgery lesion who underwent CT and those who did not was due to the reduced number of CT examinations performed (100% versus 32%, $77 per examination) and the fact that patients in whom a non-neurosurgery trauma-induced lesion was detected at CT were admitted for clinical observation: 84% in a normal ward ($539 per day) for an average of 7 days and 16% in an intensive care ward ($2234 per day) for an average of 5 days and an additional 14 days in a normal ward. Patients who did not undergo CT—and consequently in whom a non-neurosurgery lesion was not identified—were discharged home without being admitted to the hospital. Only 20% of these patients were seen for outpatient follow-up, while all patients in whom a trauma-induced lesion was identified at CT made an average of two outpatient follow-up visits ($104 per visit). No costs were involved for the patients without a traumatic lesion who did not undergo CT, while for those patients without a trauma-induced lesion who underwent CT, the cost of CT, including time costs, was the only cost that contributed to the total costs.

Reference-Case Analysis

At reference-case analysis, all CT strategies were almost equally effective (22.464 QALYs [table 5]). The differences were negligible because they resulted from poorer outcomes in patients with a missed neurosurgery lesion, with CT strategies assumed to have not led to the misclassification of patients, and because neurosurgery lesions were very rare. Of the CT strategies, use of the CCHR was the least costly (lifetime cost, $8800) and thus the most cost-effective (Table 5). The use of any of the prediction rules would lead to cost savings compared with the use of CT in all patients, but the cost savings rendered by using the CCHR and the CHIP rule would be substantially greater than those rendered by using the NOC.

Table 5 Results of Reference-Case Cohort Analysis

Table 5

*The incremental cost-effectiveness ratio (ICER), reported in 2006 U.S. dollars per QALY gained, was calculated as the incremental difference in cost divided by the incremental difference in QALY. Compared with the CCHR, the CHIP rule was $54 more expensive, with 0.00002 QALY gained. Thus, the incremental cost-effectiveness ratio for the CHIP rule, as compared with that for the CCHR, was $54 divided by 0.00002, which equals $3 million per QALY gained. A strategy is dominated if another strategy is equally or more effective and less costly. A strategy is superior if it is the least costly compared with all other strategies and is equally or more effective.

Gain in net health benefit (NHB) rendered with given strategy, as compared with that rendered with the reference strategy (no CT), was calculated as the difference in effectiveness between the given strategy and the reference strategy, minus the quotient of the difference in cost between the given strategy and the reference strategy divided by the willingness-to-pay threshold, at a willingness-to-pay threshold of $75000 per QALY.

Potential annual cost savings rendered by using the given strategy as compared with these savings rendered by performing CT in all of an estimated 900 000 patients who present with MHI in the United States annually. Estimated incidence of MHI in the United States was derived from an incidence of 300 in 100 000 individuals in a U.S. population of 300 million people. NA = not applicable.

Sensitivity Analyses

Repeating the analysis according to the United Kingdom and Dutch recommendations for cost-effectiveness analysis again revealed that selective CT yielded cost savings and that the CCHR and the CHIP rule yielded substantially higher cost savings than did the NOC.

Varying all parameters across their 95% CIs at one-way sensitivity analysis did not affect the model outcome, with the exception of when the sensitivity of the prediction rules for the identification of patients requiring neurosurgery was varied. At sensitivities lower than 97%, performing CT in all patients was more cost-effective, with an incremental cost-effectiveness ratio lower than the willingness-to-pay threshold. At sensitivities lower than 91%, performing CT in all patients became dominant, being both less costly and more effective than all the other strategies.

The model outcome was not affected by varying proportions of patients with poor outcomes after delayed neurosurgery, age, CT examination cost, or increased radiation-induced mortality.Assuming equal costs regardless of whether a non-neurosurgery trauma-related lesion was correctly identified or left undiagnosed did not affect the outcomes substantially. With variation of these parameters at two- and three-way sensitivity analyses and use of alternative modeling approaches (Appendix E2 [online]), the threshold sensitivity for the identification of patients who required neurosurgery ranged from 91% to 99%, depending on the combination of parameter values.

At probabilistic sensitivity analysis, the probability that performing selective CT was cost-effective compared with performing CT in all patients was 0.51–0.64, depending on the willingness-to-pay threshold. Performing CT selectively according to the CHIP rule was most likely to be cost-effective.

At value-of-information analysis, the EVPI for further research was $1759 per patient, which for the entire U.S. population (300 million people [U.S. Census Bureau, http://www.census.gov/popest/states/NST-ann-est.html]) over a period of 5 years amounted to $7 billion. Partial EVPI calculations revealed that this was mainly due to uncertainty about long-term functional outcomes ($1703 per patient), which in turn was due largely to uncertainty about the outcome of patients with a non-neurosurgery lesion ($1498 per patient) and to a lesser extent to uncertainty about the outcome of patients with a neurosurgery lesion ($187 per patient).

Discussion

Of the three models—involving the use of the NOC, the CCHR, or the CHIP rule—that involving the use of the CCHR to select patients for CT after MHI was the most cost-effective and, compared with scanning all patients with MHI, could lead to annual U.S. cost savings of $120 million. This finding, however, is valid only under the assumption that this prediction rule is highly sensitive for the identification of patients who require neurosurgical intervention. At lower sensitivities, the cost savings are lower, and at sensitivities lower than 91%–99%, performing CT in all patients is cost-effective compared with performing selective CT. Furthermore, uncertainty about the long-term functional outcomes after MHI currently precludes a definitive conclusion as to whether performing selective CT is more cost-effective than performing CT in all patients with MHI. Our value-of-information analysis revealed that further research is warranted to reduce uncertainty about long-term functional outcomes after MHI.

Stein et al (13) found using the CCHR to be more cost-effective than CT performed in all patients, skull radiography, clinical observation, and no treatment. Other selective CT strategies, such as use of the NOC or the CHIP rule, were not modeled. In the present study, we confirmed that selecting patients for CT on the basis of a prediction rule is cost-effective, with the CCHR yielding the largest cost savings.

Notwithstanding the capability to reduce the number of patients who undergo CT with selective CT strategies, the use of these strategies introduces the risk of misclassifying patients and consequently missing those who need neurosurgery. This issue has been left unaddressed in previous cost-effectiveness studies (13,46). We found that compared with CT performed in all patients with MHI, selective CT had a probability of being cost-effective of only 51%–64%. Furthermore, performing CT in all patients was the most cost-effective strategy when the sensitivity for identifying patients requiring neurosurgery was lower than 91%–99%. Thus, although trauma-related neurosurgery lesions after MHI are very rare, performing CT in all patients with MHI is more cost-effective than missing even a small proportion of these patients, because a delayed diagnosis in patients who require neurosurgery presumably leads to poorer outcome and associated higher costs due to disability and lost quality of life.

Because of the rare occurrence of neurosurgery lesions after MHI, the sensitivities of the prediction rules for the detection of such lesions have wide 95% CIs, with lower limits well below 90% (11,12). Given the rare occurrence of neurosurgery lesions, the prediction rules were originally designed such that any CT-depicted intracranial trauma-induced lesion was considered a proxy for a lesion requiring neurosurgery. Therefore, the sensitivity for the detection of neurosurgery lesions would not be lower than the overall sensitivity for the detection of any intracranial trauma-induced lesion. In our study, the lower limit of the 95% CI for sensitivity for the detection of intracranial trauma-induced lesions with use of the CCHR was well below 90%, indicating a real possibility that using the CCHR, despite its large potential for cost savings, is less cost-effective than performing CT in all patients with MHI. The sensitivity of the NOC for the detection of intracranial trauma-induced lesions was very high, indicating that using this rule could reduce the risk of missing patients who require neurosurgery; however, this rule yields very low cost savings owing to its low specificity. On the other hand, the CHIP rule yields considerable annual cost savings ($71 million) and is more likely to be cost-effective, given the lower limit of the 95% CI of 91% for sensitivity for the detection of intracranial trauma-induced lesions. This finding was confirmed in our probabilistic sensitivity analysis. An additional advantage of using the CHIP rule is its wide applicability: It can also be applied to patients who do not have a history of posttraumatic amnesia or loss of consciousness.

One strength of our study was that we based our model on our own data, with only a few parameter values estimated by using data from the literature. An important parameter was the cost of a CT examination, which we derived from our own hospital data, which include only true costs. Our cost estimate for CT was much lower than that reported in previous cost-effectiveness studies because the true costs are lower than the charges and reimbursements used in previous studies. All other costs included in the model were true costs, proportionate to the cost of a CT examination, and lower than those reported in other studies.

The most important uncertain parameters were functional outcome after a delayed diagnosis of neurosurgery lesion and functional outcome after a missed diagnosis of trauma-induced lesion not requiring neurosurgery. Data on these outcomes are scarce and, owing to medical-ethical issues, unlikely to become available. In our study, those patients with a false-negative prediction of trauma-induced non-neurosurgery lesion who were consequently discharged home without a diagnosis were assumed to have had the same functional outcome as the patients with a true-positive prediction, because the patients with a diagnosis of trauma-induced non-neurosurgery lesion did not undergo any intervention that affected their functional outcome (36). Whether undergoing clinical observation without subsequent neurosurgery or outpatient follow-up, as patients with a diagnosis of non-neurosurgery lesion generally do, or knowing the cause of a mild neurologic impairment affects functional outcome is entirely speculative. Such uncertainty about the long-term functional outcome after MHI was taken into account in our probabilistic sensitivity analysis and was found to be the most important parameter requiring further research in our value-of-information analysis.

The functional outcome after a delayed diagnosis of neurosurgery lesion is generally assumed to be poorer than that after timely neurosurgery is performed (25,3035,47), although the evidence to support this is minimal and indirect. The outcome estimates for our model were based on, to our knowledge the only published study in which the effect of CT on functional outcome in patients with trauma-induced neurosurgery lesions was assessed (25). Although that study was retrospective and is relatively old, it provides the most direct evidence of improved functional outcome with early identification of trauma-induced neurosurgery lesions.

With use of CT, radiation exposure is a concern (48) because patients with MHI tend to be young and thus have known increased radiation risks, and they may experience multiple MHIs that require repeated CT examinations during their lifetime. We therefore included radiation-induced mortality in our model and repeated the analyses with the assumption of extreme increases in radiation-related mortality combined with variation of patient age at exposure; however, we found that this did not affect the model outcome. With CT scanners being widely available and increasing pressure related to medicolegal issues, many clinicians would probably welcome the availability of a lenient CT strategy. Our study results suggest, given the currently available evidence and the remaining uncertainty concerning long-term functional outcomes, that performing CT in all patients with MHI is justified, even when radiation issues are considered, since performing selective CT is not unequivocally more cost-effective.

Uncertainty about model parameters was the main limitation of this study. Most parameter values were based on our own study data and were compared with and complemented by findings from the literature. We addressed these uncertainties by extensively varying all model parameters in our sensitivity analyses. The finding that the CCHR was the most cost-effective was insensitive to variations in most of the uncertain parameters, except the sensitivities of the prediction rules, as addressed earlier. This difference in results between the reference-case analysis and the probabilistic sensitivity analysis was due to the fact that distributions are different for different prediction rules and all distributions are asymmetric. Owing to a larger degree of uncertainty, the distribution of the CCHR is wider than that of the CHIP rule and thus leads to the confusing difference in results between the two analyses and indicates uncertainty regarding the selection of patients for CT.

Another limitation may have been the fact that our study was based on cost data from the Netherlands and thus yielded information about European rather than U.S. costs, the latter generally being higher. Although the cost reductions related to performing CT with prediction rules might therefore be expected to be greater in the United States when fewer examinations are performed, our sensitivity analyses revealed no effect of increasing the cost of a CT examination.

A final limitation may have been our assumption that patients without a CT-depicted intracranial trauma-induced lesion would fully recover. A proportion of such patients, however, may develop the postconcussion syndrome, which involves a variety of complaints that can negatively affect quality of life (49). Given that the postconcussion syndrome persists for more than 6 months in only a minority of patients, the negative effect on quality of life among the entire population of patients who did not have a CT-depicted intracranial trauma-induced lesion was expected to be small and was not likely to affect the outcomes of our study.

Performing CT selectively on the basis of prediction rules yields cost savings, but owing to the potential loss of effectiveness compared with the effectiveness of scanning all patients with MHI, it is not necessarily cost-effective. Of the three prediction rules, the CCHR is expected to yield the highest annual cost savings, although it can have a sensitivity for the identification of patients who require neurosurgery that is below the threshold at which CT performed in all patients with MHI is cost-effective. Compared with the CCHR, the CHIP rule is more sensitive, more likely to be cost-effective, and more widely applicable, and it has the potential to yield substantial cost savings. More research is warranted to increase certainty about long-term patient outcomes after MHI. Until such research is conducted, the routine use of CT in all patients with MHI is justified.

Advances in Knowledge
•. 

Selecting patients with minor head injury (MHI) for CT on the basis of a prediction rule such as the Canadian CT head rule (CCHR) or the CT in head injury patients (CHIP) rule could lead to substantial U.S. cost savings: $120 million or $71 million, respectively.

•. 

Because the use of selective CT strategies introduces the risk of patients being wrongly classified, according to the prediction rule, as without risk of trauma-related complications, the sensitivity of a prediction rule for the identification of patients who require neurosurgery needs to be extremely high for the use of such a rule to be cost-effective.

•. 

If the sensitivity of a prediction rule for the identification of patients who require neurosurgery is lower than 97%, it is cost-effective to perform CT in all patients.

Implication for Patient Care
•. 

If constrained resources necessitate the selective use of CT, then selecting patients on the basis of the CHIP rule is recommended, because of the three prediction rules (New Orleans criteria, CCHR, and CHIP rule), it is the most likely to be cost-effective and the most widely applicable and yields substantial cost savings.

•. 

The routine use of CT in all patients with MHI is justified owing to uncertainty regarding these patients’ long-term functional outcomes.

We thank Wibeke J. van Leeuwen, Carolien H. van Bavel-van Hamburg, and Belinda Tara-Prins, research nurses in the department of radiology at Erasmus MC–University Medical Center Rotterdam, and Jolanda Brauer, research nurse in the department of neurology at University Medical Center Nijmegen St. Radboud, for their invaluable contributions to patient data collection. We also thank Esther van der Heijden and Naziha el Ghannouti, research nurses in the department of neurology at Erasmus MC-University Medical Center, for training one of the investigators (D.A.v.R.) to assess functional outcomes after neurologic events by means of telephone interview.

Author Contributions

Author contributions: Guarantors of integrity of entire study, M.S., M.G.M.H.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, M.S., D.W.J.D., D.A.v.R.; clinical studies, D.W.J.D., P.J.N., H.M.D., P.E.V., D.R.K., D.A.v.R., P.A.M.H., A.T., H.L.J.T., M.G.M.H.; statistical analysis, M.S., M.G.M.H.; and manuscript editing, M.S., D.W.J.D., P.J.N., H.M.D., P.E.V., P.A.M.H., A.T., H.L.J.T., M.G.M.H.

Supported by grants from ZonMW (ZonMW DO: 945–06–309) and by College voor Zorgverzekeringen (CVZ: VAZ 01–104) and Radiologisch onderzoek Nederland (RADION).

Authors stated no financial relationship to disclose.

Appendix

Glossary of Terms

Acceptability curve: plot of the probability that a strategy is cost-effective compared with the alternative for varying willingness-to-pay thresholds

Decision tree: graphic representation of all possible options and consequences following each option

Expected value of perfect information (EVPI): expected outcome with perfect information minus expected outcome with available information. The partial expected value of perfect information is the expected value of learning the true value(s) of an individual or subset of parameters.

Net health benefit: health gain expected from a strategy (effectiveness) minus the health gain required to justify the cost incurred (cost/willingness-to-pay threshold)

Incremental cost-effectiveness ratio: incremental difference in cost divided by incremental difference in effectiveness

QALY: measure of a person’s expected life span that is weighted by the quality of life

Reference-case analysis: analysis involving the use of the best available point estimates for all parameters in a model

Sensitivity analysis: analysis of the uncertainty of the variable values in a model to test the stability of the conclusions over a range of probability estimates. At one-way sensitivity analysis, one variable value is varied, whereas at n-way sensitivity analysis, multiple variable values are varied at the same time.

Societal perspective: assumes that the entire society may be affected by the choice of strategy and therefore accounts for all effects and resources incurred to the society at large

Value-of-information analysis: decision analytic technique in which the benefit of collecting additional information to reduce or eliminate uncertainty concerning parameter values is evaluated

Willingness-to-pay threshold: maximal cost society is willing to pay for the gain of one unit of effectiveness (eg, 1 QALY)

References

  • 1 Borg J, Holm L, Cassidy JD et al.. Diagnostic procedures in mild traumatic brain injury: results of the WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury. J Rehabil Med 2004;61–75.. Crossref, MedlineGoogle Scholar
  • 2 Af Geijerstam JL, Britton M. Mild head injury: mortality and complication rate—meta-analysis of findings in a systematic literature review. Acta Neurochir (Wien) 2003;145:843–850, discussion 850. Crossref, MedlineGoogle Scholar
  • 3 Haydel MJ, Preston CA, Mills TJ, Luber S, Blaudeau E, DeBlieux PM. Indications for computed tomography in patients with minor head injury. N Engl J Med 2000;343:100–105. Crossref, MedlineGoogle Scholar
  • 4 Stiell IG, Wells GA, Vandemheen K et al.. The Canadian CT head rule for patients with minor head injury. Lancet 2001;357:1391–1396. Crossref, MedlineGoogle Scholar
  • 5 Cassidy J, Carroll L, Peloso P et al.. Incidence, risk factors and prevention of mild traumatic brain injury: results of the WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury. J Rehabil Med 200428–60. Crossref, MedlineGoogle Scholar
  • 6 Af Geijerstam JL, Britton M, Marke LA. Mild head injury: observation or computed tomography? economic aspects by literature review and decision analysis. Emerg Med J 2004;21:54–58.. Crossref, MedlineGoogle Scholar
  • 7 Af Geijerstam JL, Oredsson S, Britton M. Medical outcome after immediate computed tomography or admission for observation in patients with mild head injury: randomised controlled trial. BMJ 2006;333:465–470. Crossref, MedlineGoogle Scholar
  • 8 Norlund A, Marké LA, Af Geijerstam JL, Oredsson S, Britton M. OCTOPUS Study Immediate computed tomography or admission for observation after mild head injury: cost comparison in randomised controlled trial. BMJ 2006;333:469–472. Crossref, MedlineGoogle Scholar
  • 9 Haydel MJ. Clinical decision instruments for CT scanning in minor head injury. JAMA 2005;294:1551–1553. Crossref, MedlineGoogle Scholar
  • 10 Smits M, Dippel DW, Steyerberg EW et al.. Predicting intracranial traumatic findings on computed tomography in patients with minor head injury: the CHIP prediction rule. Ann Intern Med 2007;146:397–405. Crossref, MedlineGoogle Scholar
  • 11 Smits M, Dippel DW, de Haan GG et al.. External validation of the Canadian CT head rule and the New Orleans criteria for CT scanning in patients with minor head injury. JAMA 2005;294:1519–1525. Crossref, MedlineGoogle Scholar
  • 12 Stiell IG, Clement CM, Rowe BH et al.. Comparison of the Canadian CT head rule and the New Orleans criteria in patients with minor head injury. JAMA 2005;294:1511–1518. Crossref, MedlineGoogle Scholar
  • 13 Stein SC, Burnett MG, Glick HA. Indications for CT scanning in mild traumatic brain injury: a cost-effectiveness study. J Trauma 2006;61:558–566.. Crossref, MedlineGoogle Scholar
  • 14 Smits M, Hunink MG, van Rijssel DA et al.. Outcome after complicated minor head injury. AJNR Am J Neuroradiol 2008;29:506–513. Crossref, MedlineGoogle Scholar
  • 15 Jennett B, Bond M. Assessment of outcome after severe brain damage. Lancet 1975;1:480–484. Crossref, MedlineGoogle Scholar
  • 16 Rabin R, de Charro F. EQ-5D: a measure of health status from the EuroQol Group. Ann Med 2001;33:337–343. Crossref, MedlineGoogle Scholar
  • 17 Lamers LM, Stalmeier PF, McDonnell J, Krabbe PF, van Busschbach JJ. [Measuring the quality of life in economic evaluations: the Dutch EQ-5D tariff] Kwaliteit van leven meten in economische evaluaties: het Nederlands EQ-5D-tarief. Ned Tijdschr Geneeskd 2005;149:1574–1578. MedlineGoogle Scholar
  • 18 de Andrade AF, de Almeida AN, Bor-Seng-Shu E, Lourenco L, Mandel M, Marino R. The value of cranial computed tomography in high-risk, mildly head-injured patients. Surg Neurol 2006; 65(suppl 1):S1:10–S1:13. CrossrefGoogle Scholar
  • 19 Fabbri A, Servadei F, Marchesini G et al.. Clinical performance of NICE recommendations versus NCWFNS proposal in patients with mild head injury. J Neurotrauma 2005;22:1419–1427. Crossref, MedlineGoogle Scholar
  • 20 Vitaz TW, Jenks J, Raque GH, Shields CB. Outcome following moderate traumatic brain injury. Surg Neurol 2003;60:285–291, discussion 291. Crossref, MedlineGoogle Scholar
  • 21 Miller KJ, Schwab KA, Warden DL. Predictive value of an early Glasgow Outcome Scale score: 15-month score changes. J Neurosurg 2005;103:239–245. Crossref, MedlineGoogle Scholar
  • 22 Pal J, Brown R, Fleiszer D. The value of the Glasgow Coma Scale and Injury Severity Score: predicting outcome in multiple trauma patients with head injury. J Trauma 1989;29:746–748. Crossref, MedlineGoogle Scholar
  • 23 Engberg AW, Teasdale TW. A population-based study of survival and discharge status for survivors after head injury. Acta Neurol Scand 2004;110:281–290. Crossref, MedlineGoogle Scholar
  • 24 Rudehill A, Bellander BM, Weitzberg E, Bredbacka S, Backheden M, Gordon E. Outcome of traumatic brain injuries in 1,508 patients: impact of prehospital care. J Neurotrauma 2002;19:855–868. Crossref, MedlineGoogle Scholar
  • 25 Cordobes F, Lobato RD, Rivas JJ et al.. Observations on 82 patients with extradural hematoma: comparison of results before and after the advent of computerized tomography. J Neurosurg 1981;54:179–186. Crossref, MedlineGoogle Scholar
  • 26 Masson F, Maurette P, Salmi LR et al.. Prevalence of impairments 5 years after a head injury, and their relationship with disabilities and outcome. Brain Inj 1996;10:487–497. Crossref, MedlineGoogle Scholar
  • 27 Livingston DH, Lavery RF, Passannante MR et al.. Emergency department discharge of patients with a negative cranial computed tomography scan after minimal head injury. Ann Surg 2000;232:126–132. Crossref, MedlineGoogle Scholar
  • 28 Stein SC, Ross SE. Mild head injury: a plea for routine early CT scanning. J Trauma 1992;33:11–13. Crossref, MedlineGoogle Scholar
  • 29 Af Geijerstam JL, Britton M. Mild head injury: reliability of early computed tomographic findings in triage for admission. Emerg Med J 2005;22:103–107. Crossref, MedlineGoogle Scholar
  • 30 Mendelow AD, Gillingham FJ. Extradural haematoma: effect of delayed treatment. BMJ 1979;2:134. Crossref, MedlineGoogle Scholar
  • 31 Rockswold GL, Leonard PR, Nagib MG. Analysis of management in thirty-three closed head injury patients who “talked and deteriorated.”. Neurosurgery 1987;21:51–55.. Crossref, MedlineGoogle Scholar
  • 32 Miller JD, Tocher JL, Jones PA. Extradural haematoma: earlier detection, better results. Brain Inj 1988;2:83–86. Crossref, MedlineGoogle Scholar
  • 33 Massaro F, Lanotte M, Faccani G, Triolo C. One hundred and twenty-seven cases of acute subdural haematoma operated on: correlation between CT scan findings and outcome. Acta Neurochir (Wien) 1996;138:185–191. Crossref, MedlineGoogle Scholar
  • 34 Bricolo AP, Pasut LM. Extradural hematoma: toward zero mortality—a prospective study. Neurosurgery 1984;14:8–12. Crossref, MedlineGoogle Scholar
  • 35 Teasdale G, Galbraith S, Murray L, Ward P, Gentleman D, McKean M. Management of traumatic intracranial haematoma. Br Med J (Clin Res Ed) 1982;285:1695–1697. Crossref, MedlineGoogle Scholar
  • 36 Smits M, Dippel DW, de Haan GG et al.. Minor head injury: guidelines for the use of CT—a multicenter validation study. Radiology 2007;245:831–838. LinkGoogle Scholar
  • 37 Gold MR, Siegel JE, Russell LB, Weinstein MC. Cost-effectiveness in health and medicine New York, NY: Oxford University Press, 1996. Google Scholar
  • 38 Siegel JE, Weinstein MC, Russell LB, Gold MR. Recommendations for reporting cost-effectiveness analyses: Panel on Cost-Effectiveness in Health and Medicine. JAMA 1996;276:1339–1341. Crossref, MedlineGoogle Scholar
  • 39 Weinstein MC, Siegel JE, Gold MR, Kamlet MS, Russell LB. Recommendations of the Panel on Cost-effectiveness in Health and Medicine. JAMA 1996;276:1253–1258. Crossref, MedlineGoogle Scholar
  • 40 Ouwendijk R, de Vries M, Pattynama PM et al.. Imaging peripheral arterial disease: a randomized controlled trial comparing contrast-enhanced MR angiography and multi–detector row CT angiography. Radiology 2005;236:1094–1103.. LinkGoogle Scholar
  • 41 Guidelines for pharmaco-economic research [Richtlijnen voor farmaco-economisch onderzoek]. Diemen, the Netherlands: College voor Zorgverzekeringen, 2006. Google Scholar
  • 42 Centraal Bureau voor de Statistiek. StatLine databank. Centraal Bureau voor de Statistiek Web site. http://statline.cbs.nl/StatWeb/. Accessed December 11, 2007. Google Scholar
  • 43 Oanda. FXhistory: historical currency exchange rates. Oanda Web site. http://www.oanda.com/convert/fxhistory. Accessed December 11, 2007. Google Scholar
  • 44 Ades AE, Lu G, Claxton K. Expected value of sample information calculations in medical decision modeling. Med Decis Making 2004;24:207–227. Crossref, MedlineGoogle Scholar
  • 45 Claxton K, Sculpher M, Drummond M. A rational framework for decision making by the National Institute For Clinical Excellence (NICE). Lancet 2002;360:711–715. Crossref, MedlineGoogle Scholar
  • 46 Smits M, Dippel DW, Hunink MG. Cost effectiveness of using computed tomography (CT) for minor head injury compared with several other management strategies. J Trauma 2007;62:1314–1315.. Crossref, MedlineGoogle Scholar
  • 47 Galbraith S. Misdiagnosis and delayed diagnosis in traumatic intracranial haematoma. BMJ 1976;1:1438–1439. Crossref, MedlineGoogle Scholar
  • 48 Brenner DJ, Hall EJ. Computed tomography: an increasing source of radiation exposure. N Engl J Med 2007;357:2277–2284. Crossref, MedlineGoogle Scholar
  • 49 King NS. Post-concussion syndrome: clarity amid the controversy?. Br J Psychiatry 2003;183:276–278. Crossref, MedlineGoogle Scholar

Article History

Received September 18, 2008; revision requested December 12; revision received May 5, 2009; accepted May 27; final version accepted August 10.
Published online: Jan 6 2010
Published in print: Feb 2010