Evaluation of dementia: A systematic study of the usefulness of the American Academy of Neurology's Practice Parameters

Dementia is the most prevalent neurologic disorder in elderly persons, affecting an estimated 4 to 6% of persons over age 65 at costs in the United States of $50 to $100 billion per year.¹ Accurate diagnosis of the underlying disorder responsible for the dementia syndrome has important implications for prognosis and anticipatory guidance of the caregiver, as well as treatment and management of the patient. The typical cost of evaluating a patient with dementia according to routine guidelines recommended by the National Institutes of Health² is approximately $1,000 to $1,200.³ In the United States, approximately 266,400 individuals develop dementia each year (i.e., 12% of the population in the United States is older than 65 years, and each year approximately 1% of persons older than 65 develop dementia).⁴ Accordingly, the cost of evaluating new patients with dementia is approximately $2.5 to $3.2 billion each year. In today's climate of economic austerity and managed care, the challenge is to develop guidelines for selective use of diagnostic studies to reduce costs while maintaining accuracy.

The frequency of abnormal findings in various diagnostic tests has been reported in several previous studies. In 200 outpatients evaluated for dementia, Larson et al.⁵ considered 3 to 10% of laboratory findings to be of therapeutic importance. Depending upon the referral sample, 4 to 19% of CT head scans have shown abnormalities amenable to surgical treatment (e.g., subdural hematoma, normal-pressure hydrocephalus, or tumor).^5-12 Several investigators have proposed strategies to improve the diagnostic yield of the more expensive neuroimaging studies.^5,10,13 These types of recommendations are relevant to practicing clinicians (e.g., physicians, nurse practitioners, physician assistants) and may have important ramifications for the policy and delivery of health care.

The Quality Standards Subcommittee of the American Academy of Neurology(AAN) is charged with developing practical recommendations for neurologists for diagnostic procedures and treatment modalities that are common, costly, or controversial.¹⁴ Based upon a critical review of the strength of clinical evidence, practice parameters are divided into three levels of certainty: standards (strong certainty), guidelines (moderate certainty), and options (unclear certainty). Clinical evidence is divided into three classes: class I evidence is provided by one or more well-designed randomized controlled clinical trials; class II evidence is provided by one or more well-designed clinical studies such as case-control studies, cohort studies, and so forth; and class III evidence is provided by expert opinion, nonrandomized historical controls, or one or more case reports. Standards are defined as "generally accepted principles for patient management that reflect a high degree of clinical certainty (i.e., based on class I evidence or, when circumstances preclude randomized clinical trials, overwhelming evidence from class II studies that directly addresses the question at hand or from decision analysis that directly addresses all the issues)." Guidelines represent "recommendations for patient management that may identify a particular strategy or range of management strategies and that reflect moderate clinical certainty (i.e., based on class II evidence that directly addresses the issue, decision analyses that directly address the issue, or strong consensus of class III evidence)." Finally, options refer to management strategies for which there is unclear clinical certainty.¹⁴

In 1994, the Quality Standards Subcommittee published recommendations for the diagnostic evaluation of persons suspected to have dementia(figure 1).¹⁴ The neurologic history and examination (including mental status examination) were recommended as essential or standard components of the diagnostic workup. Certain laboratory tests were suggested as guidelines, including complete blood count, electrolytes, calcium, glucose, urea nitrogen, creatinine, liver function tests, thyroid function tests (free thyroid index and thyroid stimulating hormone (TSH)) vitamin B12, and syphilis serology. Diagnostic studies beyond those listed above were suggested as options. The latter include other laboratory tests, neuropsychological assessment, electroencephalography, lumbar puncture, and neuroimaging studies. Relevant clinical material is reviewed by Corey-Bloom et al.,³ but neither article^3,14 explicitly describes the analyses leading to the final stratification of the Practice Parameters.

Perhaps the most contentious of the Practice Parameters is the placement of neuroimaging studies in the optional category.^15,16 Here, the wording of the report of the Quality Standards Subcommittee¹⁴ differs in an important way from that of the background paper.³ According to the Subcommittee,"neuroimaging should be considered in every patient,"¹⁴ whereas the background paper recommends more strongly that a "neuroimaging study should be performed once in all cases of dementia."³ The Subcommittee statement indicates that there is "no consensus on the need for such studies in the evaluation of patients with insidious onset after age 60 without focal signs or symptoms, seizures, or gait disturbances."¹⁴ This statement suggests that there is at least some consensus on the need for neuroimaging in patients with dementia who have symptom onset before age 60 years, noninsidious course, focal signs or symptoms, and gait disturbance. To our knowledge, however, the value of this particular combination of indicators has not been tested in practice.

In the present study, the usefulness of the AAN Practice Parameters¹⁴ is systematically analyzed in a convenience sample. Consecutive cases referred to our diagnostic and treatment center were reviewed. First, diagnoses of dementia and type of dementia were made based on the results of the history, mental status, and physical examination. Subsequent assessment focused on the added value of laboratory, neuropsychological, and neuroimaging studies for either the diagnosis or management of dementia. Finally, the sensitivity and specificity of several indicators for selective neuroimaging were assessed, using the results of the neuroimaging studies as the gold standard.

Methods. Sample. The study sample comprised all cases referred for evaluation of memory impairment to the Southern California Alzheimer Disease Diagnostic and Treatment Center (ADDTC) between January 1, 1994 and December 31, 1994. This clinic is affiliated with the University of Southern California and is staffed by an interdisciplinary team, including two board-certified neurologists with subspecialty training in behavioral neurology and a physician assistant with 7 years of experience working in the clinic. Five cases were excluded either because neuroimaging studies were not done (n = 4) or the results were unavailable (n = 1). The final sample of 119 cases was 35% men, 65% women, 50% Caucasian, 30% Hispanic, 11% African American, and 8% Asian. The mean age of the overall sample was 69.9 ± 8.4 years. The men were significantly younger than the women (69.9 ± 8.4 years versus 75.5 ± 10.1 years; p < 0.01). The average number of years of education was 10.2 ± 5.4 years.

Evaluation. The evaluation of patients consisted of customary and optional components. Customary procedures included: C1) a complete medical history, which was obtained from collateral sources, and to the extent possible, from the patient him/herself; C2) description of the patient's instrumental and basic activities of daily living (IADL and ADL) by a collateral source using the Blessed Dementia Rating Scale¹⁷; C3) physical and neurologic examination of the patient; C4) mental status examination of the patient using the Mini-Mental State Examination (MMSE)¹⁸ and the Blessed Memory Information Concentration Test¹⁷; C5) blood tests including a complete blood count, sedimentation rate, electrolytes, glucose, liver panel, calcium, phosphorus, thyrotropin (T₄), thyroid stimulating hormone (TSH), vitamin B12, folate, microhemagglutinin assay for treponemal pallidum (MHATP); and C6) neuroimaging studies. Of the 119 patients, 79 (66%) underwent MRI, 58 (49%) had CT, and 18 (15%) had both MRI and CT. Other diagnostic procedures were ordered on an individual basis: O1) neuropsychological testing was performed whenever possible for patients with MMSE > 10; and O2) lumbar punctate, EEG, and SPECT studies were ordered if the neurologist thought these studies would contribute to diagnosis or management. When the results of the customary and optional diagnostic tests were available, a team conference was convened to make a consensus diagnosis.

Added value of diagnostic tests for diagnosis of dementia. In order to assess the usefulness of AAN practice parameters algorithm (see figure 1), the 119 cases were reviewed retrospectively by one of the board-certified neurologists. The results of neuroimages were reviewed preferably visually (n = 99) or by report (n = 20). Established methods were used whenever possible to rate the severity and pattern of atrophy, the severity of periventricular and deep white matter changes,^19,20 as well as to localize focal abnormalities.²¹ A dementia diagnosis was made repeatedly at several steps along the diagnostic work-up process. Dx-S was based on information obtained from the "standard" AAN procedures in the evaluation of dementia, namely, the history and physical, neurologic, and mental status examinations (C1-C4). Dx-NP was made based on the standard information plus neuropsychological test results (C1-C4 plus O1). Dx-L was based on the standard information plus the laboratory studies (C1-C4 plus C5). Dx-I was based on the standard information plus the neuroimaging studies(C1-C4 plus C6). Finally, Dx-F was based on all of the information available(C1-C6, O1). In order to assess the added value of each type of procedure for diagnosis, cases were tabulated where the information obtained from the procedure changed the diagnosis (e.g., when neuroimaging led to a Dx-I that differed from Dx-S).

Several standardized definitions were employed for the diagnosis of dementia and for specific types of dementing illnesses. For a diagnosis of dementia, we required fulfillment of criteria specified in the Diagnostic and Statistical Manual (DSM III-R)²² and a Clinical Dementia Rating (CDR) of at least mild dementia.²³ A diagnosis of probable or possible Alzheimer's disease (AD) was based on NINCDS-ADRDA criteria.²⁴ A diagnosis of probable or possible ischemic vascular dementia (IVD) was based on criteria suggested by the State of California ADDTC.²⁵ The other categories of diagnoses used in this study are described in the operations manual of the State of California Minimum Uniform Data set (MUDS).

The ADDTC criteria for probable IVD require: 1) dementia; 2) evidence of two or more ischemic strokes by history, neurologic signs, and/or neuroimaging studies or occurrence of a single stroke with a clearly documented temporal relationship to the onset of dementia; and 3) evidence of at least one infarct outside the cerebellum by CT or T1-weighted MRI.²⁵ Because of the third requirement for neuroimaging evidence, a diagnosis of probable IVD cannot be made based on the standard history and examinations alone (i.e., a Dx-S, based on C1-C4, cannot be probable IVD, but only possible IVD). Thus, evidence of an infarct outside the cerebellum in a neuroimaging study would often suffice to change the diagnosis (e.g., Dx-S of possible IVD could be changed to Dx-I of probable IVD). In effect, the technical requirement for neuroimaging together with our study design tends to overestimate the added value of neuroimaging for differential diagnosis. Hence, for the purpose of this study, we also modified the ADDTC criteria for probable IVD* to exclude the requirement for neuroimaging. That is, if a patient had evidence of stroke based on history and physical examination alone and met criteria for modified probable IVD*, presence of an infarct outside the cerebellum per neuroimaging was not considered to have changed the diagnosis of dementia. The added value of neuroimaging is reported using for Dx-S both the original criteria for probable IVD²³ and the modified criteria for probable IVD*.

Several types of situations arose where two or more conditions were considered to be possible causes or contributors to dementia. A diagnosis of"A and B" (mixed dementia) was used when the two conditions were thought to be independent causes of dementia. The most common example of mixed dementia was a combination of AD and vascular dementia (VaD). A diagnosis of "A with B" was used when the primary cause of dementia was considered to be A (e.g., AD), but B (e.g., depression or Lewy body variant) was considered to be a significant comorbid factor. Finally, a diagnosis of "A or B" was used when there remained uncertainty about which one of two disorders was thought to be more likely. For example, in a case with disproportionate or markedly asymmetric frontotemporal atrophy, the diagnoses of possible AD or Pick's disease was considered.

Added value for management. We also considered whether a diagnostic test would have significantly influenced the clinical management of disorders with recognized relevance for cognitive well-being. Here, a certain degree of individual judgment was employed. Several situations were considered to be significant: 1) the diagnostic test led to the addition of a new treatment (e.g., the discovery of a brain tumor led to surgical resection, or the identification of hypothyroidism led to the new administration of medication), 2) the diagnostic test was required to monitor the adequacy of current treatment (e.g., thyroid function tests in a patient who is already taking synthroid), 3) the diagnostic test was necessary to confirm or exclude an entity, which was suggested by findings in the history or physical examination, and which would have required distinct clinical management (e.g., B12 level required to rule out subacute combined degeneration, which was suggested by loss of position and vibratory sensation in the toes, and which would have required administration of B12). Instances where diagnostic tests were considered to have affected clinical management were tabulated in each of these three categories.

Sensitivity and specificity of indicators for neuroimaging. We examined the sensitivity and specificity of several indicators for obtaining a neuroimaging study. These indicators included: 1) symptom onset before age 60 years, 2) noninsidious course, 3) focal signs or symptoms, or 4) gait disturbance. In this study, focal symptoms and signs were defined as focal weakness or sensory loss, cogwheel rigidity, reflex asymmetry, or Babinski sign. Gait disturbance was defined as a change in walking, not attributed to a peripheral musculoskeletal disorder. In this study, a "significant finding" in a neuroimaging study was defined in several ways: 1) disclosure of a surgically treatable lesion (i.e., normal-pressure hydrocephalus, brain tumor, or subdural hematoma), 2) disclosure of a focal lesion other than diffuse atrophy (e.g., including focal atrophy, focal encephalomalacia, disproportionate enlargement of the ventricles compared with the sulci), and 3) information that changed the diagnosis of dementia. The latter could include findings that either confirm or rule out possible diagnostic considerations.

In a true-positive case, the indicators recommended an imaging study and the study did show a significant finding. In a true-negative case, the indicators recommended no imaging study and the study showed no significant finding. In a false-positive case, the indicators recommended a study, but no significant finding resulted. In a false-negative case, the indicators recommended a study, but no significant finding resulted. Sensitivity was calculated as the number of true-positive cases divided by the sum of the true-positive and false-negative cases. Specificity was calculated as the number of true-negative cases divided by the sum of the true-negative and false-positive cases. Sensitivity and specificity were calculated for the total sample cases (n = 119) and for all cases with dementia (i.e., CDR ≥ 1 [n = 98]) for each of the three definitions of "significant" neuroimaging findings. Chi-square analyses were also performed to assess the strength of the relationship between the guidelines and the frequency of positive imaging studies, defined in three different ways.

Results. Distribution of diagnoses. The most common dementia diagnosis at all stages of the evaluation process was AD, which decreased by 9% between the initial and final evaluations(table 1). Following the standard history and physical examination, Dx-S comprised 44% probable AD and 6% possible AD (total = 50%); following the final evaluation, Dx-F comprised 41% probable AD and 0% possible AD (total AD = 41%). The second largest change in diagnostic category between Dx-S and Dx-F was a 6% increase in the diagnosis of mixed dementia (from 16% to 22%), due largely to information obtained from laboratory testing. The third significant change was a 4% increase in the diagnosis of vascular cognitive impairment, due to information obtained by neuroimaging. At Dx-F, this included 13 cases of probable IVD, one case of possible IVD, and three cases with cognitive impairment due to vascular disease. The group of mixed dementias at Dx-F (n = 26) comprised a heterogeneous sample: six cases of AD and VaD, 12 cases of AD with another diagnosis, five cases of VaD with another diagnosis, and three cases with a combination of non-AD and non-VaD diagnoses.

Added value of diagnostic tests for diagnosis of dementia. After adding information obtained from laboratory, neuropsychological, and neuroimaging studies, diagnoses for each case changed in several ways(tables 1 through 4). Laboratory studies were obtained in all subjects and changed the diagnosis in 11 cases(9%) (see tables 2 and 3). There were four cases with a positive MHATP, four cases of hypothyroidism, one case of B12 deficiency, and two cases where metabolic diagnoses were suggested from the standard evaluation but were ruled out by laboratory testing. The addition of laboratory data was largely responsible for the overall increase between Dx-F and Dx-S in the frequency of the diagnosis of mixed dementia. Neuropsychological testing was performed in 59% of the patients. In this subset of 70 patients, the results of neuropsychological studies changed the diagnosis in eight cases (11%). In five cases, neuropsychological testing altered the diagnosis between dementia and cognitive impairment not meeting criteria for dementia (in three cases, from dementia to cognitive impairment; in two cases, from cognitive impairment to dementia). In two cases, the neuropsychologist identified the presence of significant depression that changed a diagnosis from pure AD to AD with depression. Finally, in one case, the neuropsychologist noted a pattern of cognitive performance highly atypical of AD, which changed the diagnosis from probable to possible AD.

The greatest number of changed diagnoses followed neuroimaging studies (n= 23 or 19% using the modified definition of probable IVD* for Dx-S; n = 33 or 28% using the original definition of probable IVD²⁵ for Dx-S) (see tables 2 and 4). This included three cases with potentially surgically treatable lesions: one patient with a right temporal meningioma compressing the brainstem (figure 2), and two with ventricular enlargement disproportionate to sulcal enlargement (raising the possibility of normal-pressure hydrocephalus). There were 24 cases where neuroimaging disclosed other focal lesions: 22 with infarcts (10 suspected based on the history and physical examination, eight suspected but clinically undiagnosed, and four unsuspected) and two with severe focal atrophy of the temporal lobes (figure 3), raising the possibility of Pick's disease. There were six cases where neuroimaging changed the diagnosis by excluding focal lesions that had been suspected based on the history and physical examination: four cases where no evidence of stroke was found, and two cases where subdural hematomas were ruled out. The presence of deep white matter changes by themselves were not considered sufficient to change the diagnosis of dementia (moderate to severe deep white matter changes were found in 32% of vascular dementia and 14% of AD cases). However, a diagnosis of possible IVD was made in one case, where neuroimaging showed severe leukoaraiosis and history and physical examination indicated vascular risk factors and early urinary incontinence or gait disturbance.

Added value of diagnostic tests for patient management. Laboratory studies were considered to have influenced patient management in 13% of subjects. This included nine subjects who were being treated with thyroid supplementation for a history of hypothyroidism. We believe that inappropriate treatment of hypothyroidism could affect cognition and that the results of T₄ and TSH testing were needed to determine whether to continue or to alter the dose of thyroid hormone. In two subjects (see table 3, cases 3 and 10), loss of position as well as vibratory sensation were noted on the neurologic examination, raising the possible diagnosis of B12 deficiency or neurosyphilis. In these patients, we felt that the results of laboratory studies, one way or the other, would have altered patient management (e.g., possible administration of parenteral B12 or penicillin). In one case, neuropsychological testing was considered to have modified the careplan by identifying the coexistence of depression. The results of neuroimaging studies were considered to have changed patient management in 18 (15%) of cases. In five cases, the results of the imaging influenced possible surgical management (one patient with meningioma underwent surgical resection, two patients had possible normal-pressure hydrocephalus but did not have shunts placed, in two cases a suspected subdural hematoma was ruled out). In 13 cases, the imaging studies were thought to have affected the medical evaluation and treatment of vascular risk factors and vascular disease (eight patients with previously undiagnosed infarcts but with either noninsidious course, focal signs or symptoms, or gait disturbance; four clinically silent infarcts; and one case in which a suspected infarct was ruled out). We did not consider imaging confirmation of a clinically diagnosed stroke (n = 10), or changes in diagnosis from one neurodegenerative condition to another, or change in the certainty of the diagnosis of AD (e.g., from possible to probable AD) to have changed the clinical management.

The sensitivity and specificity of indicators for neuroimaging. Sensitivity and specificity of several indicators for selective neuroimaging were calculated using each of the three definitions of "significant" neuroimaging findings as the gold standard for the total sample (n = 119) and the subset with dementia (n = 98) (table 5). Sensitivity ranged from 67 to 87%, improving as the number of positive cases increased(i.e., lowest when a positive imaging study was defined conservatively by a surgically treatable lesion [n = 3] and highest when a positive study was defined as one that changed the diagnosis of dementia when the original definition of IVD was used for Dx-S [n = 33]). Specificity was significantly less than sensitivity, but similar for all three definitions of a positive study, ranging from 41 to 54%. The highest levels of sensitivity (87%) and specificity (54%) were observed in predicting imaging studies that changed the diagnosis in the patients with dementia (n = 98), using the original definition of probable IVD. Similar levels of sensitivity and specificity were associated with the prediction of focal lesions in the total sample(sensitivity 78%, specificity 47%) or with changing the diagnosis of dementia when IVD* was used for Dx-S (sensitivity 78%, specificity 45%).

When a positive imaging study was defined as either a focal lesion or one that changed the diagnosis in the sample as a whole, sensitivity was limited by six false-negative cases, where the indicators suggested no need for brain imaging, but significant neuroimaging abnormalities were found (see table 4). The lesions missed included: 1) infarcts that changed the diagnosis from "probable AD" to "possible AD and possible IVD" (n= 1), from "probable AD" to "probable IVD" (n = 1), and from "cognitive impairment" to "cognitive impairment due to cerebrovascular disease" (n = 2); 2) disproportionate focal atrophy of the temporal lobe, which changed the diagnosis from "probable AD" to "possible AD or possible Pick's disease" (n = 1); and 3) ventricular enlargement disproportionate to the degree of sulcal enlargement, which changed the diagnosis from "probable AD" to "possible AD or possible normal-pressure hydrocephalus" (n = 1).

Specificity was significantly decreased by a high false-positive rate (36% in the total sample) where there was an indication for brain imaging, but no significant finding other than diffuse atrophy or mild leukoaraiosis was found. Fifty-eight percent of the false-positive cases had more than one indication for brain imaging (e.g., history of focal symptom and gait disturbance). The presence of focal signs and symptoms was the most frequent false-positive indicator (36%), followed by gait disturbance (24%), early symptom onset (12%), and noninsidious course (5%). Systematic elimination of any one of these indicators resulted in a slight improvement in specificity but a significant loss of sensitivity.

When a positive imaging study was defined by the presence of a focal lesion or by changing the diagnosis of dementia, several highly statistically significant associations were found with the indicators (see table 5). For focal lesions and the indicators, chi-square values ranged from 5.2 (p = 0.02) to 8.3 (p= 0.004) for the total sample and the dementia subsample, respectively. For changing the diagnosis of dementia, the chi-square values ranged from 3.8(n.s.) to 14.4 (p = 0.0001), highest when the original definition of probable IVD was employed in the dementia subsample. No statistically significant associations were found between the indicators and the rare surgically treatable lesions.

Discussion. This study addresses the added value of several diagnostic tests recommended by the Quality Standards Subcommittee of the American Academy of Neurology (AAN) as guidelines (laboratory studies) or options (neuropsychological testing and structural neuroimaging) in the diagnosis and evaluation of dementia.¹⁴ We analyzed how specific test results would affect clinical decision-making related to diagnosis and management. In addition, we calculated the sensitivity and specificity of several indicators for neuroimaging compared with the actual results of these studies. Since the cost-effectiveness of diagnostic studies does not lend itself to randomized clinical trials or the acquisition of class I evidence, our study was designed to provide class II evidence.

The convenience sample comprised patients with memory impairment referred to a university-affiliated interdisciplinary clinic. The most common diagnosis was AD (41% AD and 15% AD-plus at final diagnosis). The proportion of subjects with cerebrovascular disease (14% VaD and 9% VaD-plus), though similar to other geriatric outpatient series,^10,12 was higher than in others,^5,9 possibly reflecting our particular interest in vascular dementia. The history and examinations were performed by two board-certified neurologists with subspecialty training in behavioral neurology or by an experienced physician assistant. Considering the nature of subject referral and the specialization of the investigators, generalization of the results of this study to the general community or to a primary-care setting warrant caution.

The added value of laboratory testing was comparable to other reports in the literature.^5,9,26 Laboratory studies changed the diagnosis in 9% of cases (see table 3), largely by increasing the number of mixed diagnoses (e.g., positive MHATP, low T₄, high TSH, or low B12). Even in the absence of primary treatment for the predominant cause of dementia (e.g., AD), treatment of co-morbid disorders reduces excess disability.^9,26,27 Laboratory results were pertinent to patient management in 13% of cases. Even in cases where hypothyroidism has been previously diagnosed and treated, current knowledge of thyroid function is necessary for patient management. The AAN Quality Standards Subcommittee recommends laboratory studies as a guideline¹⁴ but does not suggest when they should be ordered. In our study, it was often difficult to predict abnormal laboratory findings from the history and examination alone. Based on their low cost (currently about $165 for the recommended panel) and significant yield, this study supports the inclusion of laboratory studies as a cost-effective component of the standard dementia evaluation.

Neuropsychological studies were more likely to change diagnosis (11%) than management (1%). Criteria for AD and IVD require the presence of dementia.^24,25 Hence, in differentiating mild cognitive impairment from early dementia, the results of neuropsychological testing often affect diagnosis. The pattern of neuropsychological testing may also influence differential diagnosis. For example, relative preservation of secondary memory suggests non-AD dementia; predominant impairment of concentration may support a diagnosis of depression. Indeed, depression was diagnosed in 12% of subjects, although this was often suspected after the standard history and examination. Taking into account their moderate cost(currently about $200 to $300) and unique, but selective, contribution, this study supports neuropsychological testing as an option (as recommended by the AAN Quality Standards Subcommittee) in cases of mild cognitive impairment or when secondary memory impairment is not a predominant feature.

Neuroimaging studies frequently changed clinical diagnosis (19 to 28%) and management (15%). Surgically treatable lesions (e.g., normal-pressure hydrocephalus, subdural hematoma, or brain tumor) were found in three cases. In one case, memory and cognition improved after surgical resection of a brain tumor. Other focal findings (i.e., cerebral infarction [n = 22] and focal temporal atrophy [n = 2] altered diagnosis or management. Arguably, confirmation of an infarct or hemorrhage should lead to careful evaluation and treatment of cardiovascular risk factors. Finally, neuroimaging was useful in ruling out lesions mistakenly suspected from the clinical history and examination (n = 6). Thus, although relatively expensive (approximately$500 for CT and $700 for MRI), neuroimaging studies contribute important information either not suspected or wrongly suspected from the standard evaluation.

We explored the utility of several indicators for neuroimaging studies, namely: symptom onset before age 60 years, noninsidious course, focal neurologic signs or symptoms, and gait disturbance. Sensitivity was lower when a positive imaging study was defined as a surgically treatable lesion(67%), and higher when defined as changing the diagnosis of dementia (87%). Specificity was lower than sensitivity, but similar for all three definitions of a positive study (41 to 54%). These figures fall within the broad range of 12 to 88% for sensitivity and 37 to 78% for specificity reported previously in the literature,¹⁰ using a variety of other clinical indicators for neuroimaging.

The best values for sensitivity and specificity were obtained when we defined a significant neuroimaging finding as one that changed the diagnosis of dementia and used the published definition²⁵ for a diagnosis of IVD (see table 5). Sensitivity is calculated as the true-positive rate divided by the sum of the true-positive and false-negative rates. When positive cases are infrequent, each false-negative case will significantly reduce sensitivity. The positive case rate was 10-fold higher when a positive imaging study was defined by change in diagnosis (n = 33) rather than a surgically treatable lesion (n = 3). When a positive study is defined as changing the diagnosis of dementia, the absence, as well as the presence, of findings may be counted as "true positives" and thereby increase sensitivity. Three of the indicators (i.e., noninsidious course, focal signs and symptoms, and gait disturbance) are associated with an initial diagnosis other than probable AD. Neuroimaging, regardless of the actual finding, would be likely to change the final diagnosis: if a focal finding were confirmed, the diagnosis might change from possible AD to non-AD; if the imaging showed no local finding, the diagnosis might change from possible to probable AD. Finally, the published criteria for a diagnosis of probable IVD²⁵ require a neuroimaging study to confirm the presence of an infarct. Thus, the finding of an ischemic lesion on neuroimaging, even if already suspected from the clinical history and examination, would change the diagnosis from possible to probable IVD. In summary, a more liberal definition of true positives, a confound between indicators and initial diagnosis, and a definitional requirement for neuroimaging all strengthen the association between the clinical indicators and imaging studies that are likely to change the diagnosis of dementia.

For good clinical care, high sensitivity is more important than high specificity. It is preferable to overuse imaging than to underuse it and miss something treatable. For the sake of economy, high specificity is more important than high sensitivity. It is "cheaper" (at least initially) to underuse and reduce the number of the false positives. While imperfect, the indicators studied here offer a step in the right direction. The results favor good clinical care, while still reducing costs: sensitivity was higher than specificity; false positives (36%) outnumber false negatives (5%). Imaging would have been obviated in one-third of cases, but at the clinical expense to misdiagnosing 5% of cases. These false-negative cases included four infarcts, one possible Pick's disease, and one possible normal-pressure hydrocephalus. Judgment about what constitutes "acceptable" false negatives is open to debate.^15,16 Given the inherent trade-off between sensitivity and specificity and between clinical and economic goals, the identification of an optimal strategy for selective use of neuroimaging studies remains an elusive challenge. Additional studies are needed in larger community samples.