Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Biol Psychiatry. Author manuscript; available in PMC 2009 November 15.
Published in final edited form as:
PMCID: PMC2613777

Combining early markers strongly predicts conversion from mild cognitive impairment to Alzheimer’s disease



The utility of combining early markers to predict conversion from mild cognitive impairment (MCI) to Alzheimer’s Disease (AD) remains uncertain.


148 outpatients with MCI, broadly defined, were followed at 6-month intervals. Hypothesized baseline predictors for follow-up conversion to AD (entire sample: 39/148 converters) were cognitive test performance, informant report of functional impairment, apolipoprotein E genotype, olfactory identification deficit, MRI hippocampal and entorhinal cortex volumes.


In the 3-year follow-up patient sample (33/126 converters), five of eight hypothesized predictors were selected by backward and stepwise logistic regression: FAQ (informant report of functioning), UPSIT (olfactory identification), SRT immediate recall (verbal memory), MRI hippocampal volume, MRI entorhinal cortex volume. For 10% false positives (90% specificity), this five-predictor combination showed 85.2% sensitivity, combining age and MMSE showed 39.4% sensitivity, and combining age, MMSE, and the three clinical predictors (SRT immediate recall, FAQ, and UPSIT) showed 81.3% sensitivity. Area under ROC curve was greater for the five-predictor combination (0.948) than age plus MMSE (0.821; p =.0009), and remained high in sub-samples with MMSE ≥ 27/30 and amnestic MCI. For the entire patient sample, based on dichotomizing estimated risk at 0.5, positive likelihood ratio was 16.8 (95% CI 6.4, 44.3) and negative likelihood ratio was 0.2 (95% CI 0.1, 0.4).


The five-predictor combination strongly predicted conversion to AD and was markedly superior to combining age and MMSE. Combining only clinically administered measures also led to strong predictive accuracy. If independently replicated, the findings have potential utility for early detection of AD.


A large number of elderly people have cognitive impairment, which increases with age (1,2). Mild cognitive impairment (MCI) often represents a transition between age-related memory decline and Alzheimer’s disease (AD) (3,4). The annual conversion rate from MCI to AD is 8–15%, and most conversions occur within 3 years of presentation (3,5). Identifying predictors of conversion to AD will aid in estimating prognosis and targeting early treatment in likely converters.

In a single-site, long-term study of cognitively impaired outpatients without dementia, specific baseline measures across several domains were hypothesized to predict conversion to AD. Among these measures, our original predictive findings on olfactory identification deficits (6,7), and informant report of functional deficits (8), are consistent with related studies in cognitively impaired subjects (9,10). Our published predictive accuracy of specific neuropsychological deficits (11) and MRI hippocampal and entorhinal cortex atrophy (12) are consistent with the literature (13,14). Each measure significantly predicted conversion to AD after adjusting for age, sex, and education. However, as in other studies, no individual measure was strong enough to use as the sole predictor. Therefore, the value of combining predictors, and their added value to the widely used combination of age and MMSE, was assessed. To aid clinicians in early detection, an initial algorithm to predict conversion to AD was developed.

Methods And Materials


Patients presented with memory complaints to a Memory Disorders Clinic jointly run by Psychiatry and Neurology at New York State Psychiatric Institute/Columbia University, and met study criteria for cognitive impairment without dementia and without a specific identifiable cause. These patients, and age and sex-matched healthy control subjects, were followed systematically. Healthy controls were recruited primarily by advertisement, had normative MMSE and SRT test scores, met all other inclusion/exclusion criteria, and were group-matched to patients on age and sex. All subjects signed informed consent in this IRB-approved protocol.

Eligibility Criteria

Inclusion criteria were age 41–85 years, cognitive impairment ≥ 6 months and ≤ 10 years, and Folstein MMSE score ≥ 22/30 (15). Neuropsychological screening inclusion guidelines were Folstein MMSE recall ≤ 2/3 objects at 5 minutes, or Selective Reminding Test (SRT) delayed recall score > 1 SD below norms, or Wechsler Adult Intelligence Scale-Revised (WAIS-R) performance IQ score ≥ 10 points below WAIS-R verbal IQ score. Patients without these neuropsychological deficits were considered if they met three criteria: subjective complaint of memory decline, informant’s confirmation of memory decline, and modified Blessed Functional Activity Scale score ≥ 1 on the first 8 items that cover memory-related cognitive and functional problems (16). Final determination for inclusion was based on a consensus diagnosis between two expert raters (D.P.D., Y.S.) who reviewed clinical, functional and neuropsychological information, laboratory test results, and MRI clinical report, while remaining blind to all other data.

These inclusion criteria identified a broad group of cognitively impaired non-demented patients. This study began before mild cognitive impairment (MCI) criteria were published (17). Since all patients had subjective memory complaints (diagnostic criterion for MCI) and completed an extensive neuropsychological test battery, baseline MCI subtype was determined post-hoc by using age, education, and sex-based regression norms derived from 83 healthy control subjects (63 from this study and 20 from other studies) who received the same test battery, as previously described (11). Based on this approach, 73% of patients met criteria for amnestic MCI with or without other cognitive domain deficits, 13.5% had non-amnestic MCI, and 13.5% had cognitive scores < 1.5 SD below norms, i.e., insufficient to meet MCI criteria (11).

Exclusion criteria were a diagnosis of dementia, schizophrenia, current major affective disorder, alcohol/substance dependence, history of stroke, cortical stroke or infarct ≥ 2 cm in diameter based on MRI, cognitive impairment entirely caused by medications, or other major neurological illness, e.g., Parkinson’s disease.


At baseline, the study neurologist/psychiatrist completed history, physical, neurological, and psychiatric examination. Laboratory tests included complete blood count, serum electrolytes, thyroid, liver and renal function tests, rapid plasma reagin (RPR), serum vitamin B12 and folate levels, and MRI brain scan.

All subjects received an annual neuropsychological test battery, reviewed by an experienced neuropsychologist (Y.S.). Two expert raters (D.P.D. and Y.S.) made a consensus diagnosis at each follow-up, while remaining blind to data from prior visits. The diagnosis of dementia was based on DSM-IV criteria and possible or probable AD on NINCDS-ADRDA criteria (18). The endpoint of conversion to AD required this diagnosis at two consecutive annual assessments during follow-up. The Clinical Dementia Rating (CDR) was the secondary outcome criterion. At each visit, the physician rated an initial CDR and the final CDR was scored at the consensus diagnosis meeting based on the physician’s CDR plus neuropsychological test data.

Baseline predictors of future conversion to AD

At study inception, predictors were chosen based on the existing scientific literature. In this sample, among five neuropsychological variables that were examined, SRT immediate recall (12-item, 6-trial sum score) and WAIS-R digit symbol were the strongest predictors (11). Apolipoprotein E genotype was assessed and patients were classified as ε4 carriers (ε3/ε4 or ε2/ε4 or ε4/4) or ε4 non-carriers (19). The University of Pennsylvania Smell Identification Test (UPSIT) employs 40 microencapsulated common odorants, each presented with four written response alternatives (scratch and sniff forced choice format). The test score ranges 0 to 40 (20). The Pfeffer Functional Activities Questionnaire (FAQ) (21) assesses 10 common activities that focus on complex cognitive/social functioning. The informant report’s total score (range 0 to 10) was obtained (8).

Brain MRI was acquired on a single GE 1.5T unit. Volumetric ratings utilized a 3D coronal volume spoiled gradient recalled echo (SPGR) sequence, 2 mm thick contiguous slices perpendicular to the temporal horns (12). After establishing intra-rater reliability and inter-rater reliability with experts (ICCs 0.90–0.99) (12), a single rater (G.P.) quantified hippocampal and entorhinal cortex volume by published methods (22), blind to all other information.

Statistical Analyses

Demographic, clinical and predictor variables were compared in converters to AD, non-converters to AD, and controls by χ2 and Kruskal-Wallis test (several variables were non-normally distributed) for categorical and continuous variables, respectively. ANCOVA was used for group differences in hippocampal and entorhinal cortex volume, adjusting for intracranial volume.

The hypothesized predictor variables were age, SRT immediate recall, WAIS-R Digit Symbol, UPSIT, FAQ, apolipoprotein E ε4 carrier status, MRI hippocampal and entorhinal cortex volumes (adjusting for intracranial volume). Bivariate associations among these variables and baseline age and education were examined by Spearman correlation coefficients.

Since the UPSIT score had a skewed distribution, the log (41-UPSIT score) transformation (23) was applied in regression analyses.

Logistic regression analyses were conducted to assess the incremental effect of predictors (each predictor’s individual effect controlled by other predictors’ effects) on probability of conversion to AD by 3-year follow-up. The equal follow-up duration for all subjects permitted assessment of sensitivity, specificity, positive and negative predictive values, and likelihood ratios. The 3-year duration was chosen because most conversions occurred within this time frame, consistent with the literature (3), and to evaluate clinically relevant short-term to intermediate-term conversion to AD. For early converters who did not reach 3-year follow-up, the diagnosis was assumed to have remained AD.

From the eight hypothesized predictors, stepwise and backward procedures selected predictors for a final logistic regression model in which all remaining predictors were significantly associated with conversion to AD. For these selected predictors, sensitivity and false positive rate (1-specificity) were calculated for Receiver Operating Characteristic (ROC) curves. Area under the ROC curve was used for classification accuracy. Based on the fitted logistic model and dichotomizing estimated risk at 0.5, likelihood ratios (LR) with 95%CI were calculated for LR(+) as the ratio of true-positive percent to false-positive percent, and LR(−) as the ratio of false-negative percent to true-negative percent. The added value of the predictors to the combination of age and MMSE (age+MMSE) in predicting conversion to AD, based on logistic regression, was examined. These analyses were repeated in subsamples with amnestic MCI (11), MMSE ≥ 27/30, age > 55 years, and the 5-year follow-up sample.

From the final logistic regression model, a predictor algorithm based on dichotomizing the selected predictors was developed. The optimal dichotomized cutoff scores for the selected predictors were derived from their individual ROC curves with each cutoff point minimizing the distance between perfect and actual estimates of sensitivity and specificity. Likelihood ratios (percent of number of abnormal predictors among AD converters to percent among non-converters) and their 95% confidence intervals were calculated. For internal validation, the prediction error of this algorithm was estimated by a five-fold cross-validation procedure (24). The cross-validation procedure randomly partitioned the sample into 5 roughly equal size samples. In each of 5 runs, one of these random sub-samples was the test sample and the rest (combined 4 random sub-samples) was the training sample (24). The training sample was used to estimate parameters of the logistic regression model with the selected predictors to predict the outcome in the test sample. Combining misclassification on the test sample and averaged across the 5 runs, the cross-validation procedure gave the misclassification error rate of the predictor algorithm.


Demographic and Clinical Features

Based on initial statistical power estimation for the neuropsychogical and MRI predictors derived from the literature, 150 consecutive eligible patients were recruited. Two patients diagnosed with other neurological disorders (corticobasal degeneration, and amyotrophic lateral sclerosis presenting with frontal lobe deficits) within 6 months of presentation were excluded, leaving 148 patients in the sample.

Future converters were older and scored lower on baseline MMSE than non-converters or controls, but the groups did not differ significantly in sex distribution and apolipoprotein E ε4 carrier frequency (Table 1). Baseline SRT immediate recall, WAIS-R Digit Symbol, UPSIT, hippocampal and entorhinal cortex volumes decreased from controls to non-converters to converters. FAQ informant scores were greater in converters than non-converters (controls did not have informant interviews). There were no sex differences for any predictor. Table 2 illustrates baseline correlations among the predictors and age, education and MMSE.

Table 1
Baseline features of 148 MCI patients (converters and non-converters to AD during 1–9 years of follow-up) and 63 healthy control subjects.
Table 2
Baseline correlations in MCI patients among age, education, MMSE and specific hypothesized predictors.

Of the 148 patients, 128 met the neuropsychological inclusion guidelines and 20 patients had neuropsychological deficits not severe enough to meet these guidelines but they met the other criteria (subjective memory complaints, informant’s confirmation of memory decline, and Blessed rating criteria). These two groups did not differ significantly in age, sex, years of education, or conversion rates to AD. Patients who met neuropsychological inclusion guidelines had lower baseline MMSE scores compared to those who did not (mean 27.3±2.1 versus mean 28.5±2.6, t=2.3, p < .03).

Follow-up sample

In all instances when dementia was diagnosed, the CDR rating was confirmatory (≥ 1, indicating dementia), and hence the analyses were not repeated for this secondary outcome. Follow-up duration averaged 5 years in controls (Table 1). In patients, converters had shorter follow-up because they exited the study after two consecutive annual AD diagnoses (Table 1). MMSE showed greater annual percent decline in converters (mean ± SD: 3.08±5.8) compared to non-converters (0.11±2.52; p < .0001). Of the 39 converters, 31 had probable AD and 8 had possible AD, i.e., AD with concomitant conditions that developed during follow-up (depression n=6, prescription drug abuse n=1, Parkinsonian features without meeting criteria for Lewy Body Dementia n=1). Decline in cognitive scores during follow-up did not differ between probable and possible AD converters who were combined for data analysis. All converters were greater than 55 years old at baseline evaluation.

Of nineteen deaths during follow-up, there were four autopsies, all of which confirmed the clinical diagnoses (3 AD, 1 healthy control).

3-year follow-up sample: variable selection

Thirty-three of 126 patients converted to AD by 3-year follow-up (incidence 9.2% per person year). Twenty-two patients dropped out before the 3-year follow-up time-point. At baseline, they did not differ from the 126 patients who were followed for at least 3 years in age, education, MMSE or any hypothesized predictor (all p’s > 0.3).

All eight a priori hypothesized predictors were entered into a logistic regression model with the outcome of AD/no AD by 3-year follow-up. Using either stepwise or backward selection in the model, age, WAIS-R Digit Symbol and apolipoprotein E ε4 did not meet the criterion for significance (alpha=0.1) to be in the final model. Five predictors remained significant and entered the final model: FAQ, UPSIT, SRT immediate recall, hippocampal volume, entorhinal cortex volume.

Sensitivity and specificity

In the 3-year follow-up sample, at 80% specificity (20% false positives), age, MMSE and the five predictors showed individual sensitivities ranging 41–72%. Since a low false positive rate is clinically important (a patient should not be told erroneously that he/she has AD), the individual sensitivities were also calculated at the 10% false positive rate (90% specificity). These sensitivities ranged 28–51% for age, MMSE and the five selected predictors individually (Table 3).

Table 3
Effects of individual and combined predictors for conversion to Alzheimer’s Disease (AD) in the 3-year follow-up patient sample.

For the combination of age and MMSE (age+MMSE), sensitivity declined from 72.7% for 20% false positives to 39.4% for 10% false positives. For the five-predictor combination, sensitivity was 92.6% for 20% false positives and remained high at 85.2% for 10% false positives (Table 3). For 10% false positives, combining age, MMSE, and the five predictors showed 85.2% sensitivity (Table 3).

Further analyses focused on clinically relevant combinations of the five predictors, together with the widely used measures of age and MMSE. For 10% false positives, the clinically administered measures, SRT immediate recall, UPSIT, and FAQ informant report, each added markedly to the sensitivity and positive predictive value of age+MMSE (Table 3). Combining age, MMSE, SRT immediate recall, FAQ, and UPSIT showed 81.3% sensitivity and 83.3% positive predictive value. Based on estimated risk of 0.5, this combination led to a positive likelihood ratio of 13.0 and negative likelihood ratio of 0.2. Although UPSIT and FAQ informant report are brief, inexpensive and easy to administer, they are not routinely used clinically. Therefore, the combined effect of age, MMSE, SRT immediate recall, and MRI hippocampal and entorhinal cortex atrophy was evaluated. For 10% false positives, this combination showed 70.4% sensitivity and 82.6% positive predictive value, and based on estimated risk of 0.5, positive likelihood ratio of 15.5 and negative likelihood ratio of 0.3 (Table 3). The positive likelihood and negative likelihood ratios (16.8, 0.2, respectively) were strongest for the five predictors plus age and MMSE, but with fairly wide confidence intervals (Table 3).

Both hippocampal and entorhinal cortex volume statistically contributed to prediction. Removing either measure lowered sensitivity for 10% false positives to some degree: all five predictors sensitivity=85.2% (95% CI 63.4–93.8), four predictors excluding entorhinal cortex volume sensitivity=78.6% (95% CI 71.8–98.6), four predictors excluding hippocampal volume sensitivity=81.5% (95% CI 66.8–96.1).

For the five-predictor combination, in ROC analyses the area under the curve (AUC) was high (0.948) and greater than for age+MMSE (0.821; p =.0009; Figure 1 Panel A). AUC was higher for the five-predictor combination added to age+MMSE compared to age+MMSE (p=.0012, Figure 1 Panel B). The five-predictor combination also led to strong prediction in amnestic MCI patients (32/90 converters by 3 years, AUC 0.934), patients with baseline MMSE ≥ 27/30 (19/92 converters by 3 years, AUC 0.941), patients followed for 5 years (37/90 converters, AUC 0.938), and patients over 55 years old (33/112 converters by 3 years, AUC 0.942).

Figure 1Figure 1
Comparisons of Receiver Operating Characteristic (ROC) curves in the 3-year follow-up patient sample.

Including education with age and MMSE in similar models, and replacing SRT immediate recall by SRT delayed recall, led to virtually identical results.

Predictor algorithm

Using the 3-year follow-up sample, a predictor algorithm from the logistic regression model was developed. The optimal dichotomized cutoff scores for the 5 predictors were derived from their individual ROC curves with each cutoff point minimizing the distance between perfect and actual estimates of sensitivity and specificity. Dichotomized cutoffs for “abnormality” were SRT immediate recall < 40, UPSIT < 32, FAQ > 1, hippocampal volume < 4.1 cc, entorhinal cortex volume < 0.4 cc (Table 4). These dichotomized variables were then entered in the logistic model, and the estimated risk of AD conversion tabulated for every predictor combination.

Table 4
Probability of MCI conversion to AD in the 3-year follow-up sample (25.2% converters) by the number of abnormal predictors in the predictor algorithm.

In the five-predictor algorithm, the 3-year probability of conversion was very low if only one predictor was abnormal (1.4 to 3%) and low (7.5 to 24%) if only two of five predictors were abnormal. For three abnormal predictors the probability of conversion ranged 34–67%, for four abnormal predictors this probability ranged 84–92%, and if all five predictors were abnormal this probability was 98% (Table 4). The likelihood ratio for predicting conversion to AD increased with increasing number of abnormal predictors, particularly for three or more abnormal predictors (Table 4).

For internal validation of the five-predictor algorithm, the five-fold cross-validation procedure estimated the misclassification error of the predictor algorithm as 14.95%.


In this broadly defined clinical sample of cognitively impaired, non-demented patients presenting with memory complaints, each predictor’s effect was consistent with other studies that examined predictors of MCI conversion to AD (25,26), with the exception of apolipoprotein E ε4 carrier status that was not significant when included with other predictors in logistic regression analyses. Sampling differences may account for discrepancies across studies; apolipoprotein E ε4 carrier status is a significant predictor in several studies (2528) but some recent reports indicate weak prediction (29,30).

In the 3-year follow-up sample, individual baseline measures significantly predicted future conversion to AD but the five-predictor combination markedly improved prediction. The five-predictor combination led to very strong prediction at the MCI stage, exceeding consensus threshold criteria for markers of AD of sensitivity >80% and specificity > 80% (31). The five-predictor combination markedly added predictive value above the combination of age and MMSE at a 10% false positive rate, and positive and negative likelihood ratios confirmed strong prediction. A low false positive rate is necessary when predicting likely conversion to AD, i.e., predicting AD when the patient does not develop AD can lead to adverse clinical consequences. The findings indicate that knowing the patient’s age and administering the MMSE, a combination commonly used by practicing physicians, are together insufficient to accurately predict conversion to AD.

The diagnosis of MCI requires lack of impairment in activities of daily living (17). This is the first study to explicitly examine informant report of impairment in complex social/cognitive abilities, which can be impaired in MCI patients, and odor identification deficits together as predictors of conversion to AD. These inexpensive measures do not require prior training to administer: FAQ 2–3 minutes (can be self-report), UPSIT 12–15 min (multiple choice, can be self-administered). In the predictor models, for 10% false positives the combination of SRT immediate recall, FAQ and UPSIT, together with age and MMSE, led to strong predictive accuracy.

The combination of SRT immediate recall with age, MMSE and the MRI variables also led to fairly strong prediction. Neuropsychological testing of episodic verbal memory and other cognitive abilities is time-consuming and can be expensive. MRI volumetric evaluation of hippocampal and entorhinal cortex volumes is expensive and currently limited to research/academic centers. Nonetheless, neuropsychological and MRI assessment improve prediction of conversion to AD, provide other useful diagnostic information, and their serial assessment may clarify the MCI/AD diagnosis. A trained MRI rater, after completing image alignment, can assess hippocampal volume in 30–45 minutes and entorhinal cortex volume in 15–20 minutes using this study’s methods (22,32). Since MRI hippocampal and entorhinal cortex volumetric assessment techniques vary by research group, their identified algorithm cutoffs may not apply for other assessment methods. In contrast, the SRT, UPSIT and FAQ employ uniform administration and scoring.

In other studies, combining MMSE/verbal recall measures with MRI hippocampal volume/medial temporal lobe atrophy or apolipoprotein E genotype led to moderately strong accuracy (65–84% correctly classified) in predicting conversion to AD (25,26). We found that adding olfactory identification deficits and informant report of functional impairment, and entorhinal cortex volume, led to strong predictive accuracy (92.5% correctly classified) that was appreciably higher than prior reports (25,26). The predictor algorithm developed from the data obtained in our study emphasizes the need to examine multiple predictors.

For the five-predictor combination, similar findings were obtained in the amnestic MCI, high baseline MMSE, and 5-year follow-up sub-samples. Strong prediction in the high MMSE (≥ 27/30) group is clinically important because diagnosis and estimation of prognosis is difficult in patients with minimal global cognitive impairment.

The individual effects of olfactory identification deficits and informant report of functional impairment were consistent with related reports (9,10), and the combined effects of episodic memory deficits and MRI volumetric measures were highly consistent with the literature (14,32). For the algorithm, to reduce the risk of over-fitting, the cutoff scores for the 5 predictors were identified from their individual ROC curves and not from their combined ROC curve. Since SRT immediate recall, UPSIT and hippocampal volume declined with increasing age (5,22,33), and SRT immediate recall correlated with education in years, the optimal algorithm cutoffs for the five predictor variables may need alteration when applied in older, less educated samples. A complete independent external validation in a larger sample is needed, but to date no other study has evaluated informant report of functional impairment and odor identification deficits together with neuropsychological and MRI volumetric measures. Further, using the five-predictor combination does not preclude the need for proper subsequent verification of diagnostic classification.

The MRI, olfactory and verbal recall measures tapped into early AD pathology: neurofibrillary tangles begin in trans-entorhinal cortex and progress through limbic areas to neocortex (34). Impairment in complex, integrated brain functions were sensitive to incipient AD: short + long-term verbal memory (SRT immediate recall incorporates encoding with recognition, comparison to a bank of memories, retrieval over 6 consecutive trials), odor identification (UPSIT incorporates odor detection, odor memory, odor naming) and complex cognitive/social functional abilities (FAQ informant report captures the behavioral output of multiple brain regions acting together).

Gender, obesity, vascular risk factors, and family history predict dementia in large-scale epidemiologic studies. Given the known small effect size of these variables, they were not hypothesized predictors in this clinical sample. In secondary analyses, we examined gender, body mass index and family history, and none were significant predictors (data not shown).

This study had some limitations. The study was conducted in a University academic medical center; the results need validation in broader clinical settings and cannot be directly applied to the general elderly population. The sample inclusion/exclusion criteria were broader than the narrowly defined amnestic MCI samples studied by other groups (3,4), but they did identify a clinically relevant, heterogenous group of patients. The findings in our large amnestic MCI subsample were consistent with the results obtained in the entire sample, thus permitting direct comparison to other studies that restricted their samples to amnestic MCI patients (3,4). The raters were not blind to one of the selected predictors, SRT immediate recall, which was part of the neuropsychological test battery. However, the expert raters employed strict diagnostic methods, and the diagnoses were indirectly supported by differences in global cognitive decline in converters versus non-converters. Patients with stroke or MRI cortical or large lacunar infarctions were excluded, reducing the chance of cerebrovascular disease influencing MCI conversion to AD (35).

Other measures that strongly distinguish AD from controls and potentially predict MCI conversion to AD include decreased temporoparietal metabolism with FDG PET (36) and increased uptake of ligands that label amyloid using PET (37,38). PET has high costs and limited availability. PET amyloid imaging may not show disease progression because maximal brain amyloid saturation occurs early in AD (39). Cerebrospinal fluid (CSF) elevated tau or phosphorylated tau levels, and decreased amyloid ßeta42 levels, and their ratio, have been shown to strongly predict MCI conversion to AD (40). Future studies will need to assess these CSF measures when combined with other predictors. Lack of patient acceptability of lumbar puncture can be a limitation. A recent report showed that combining immune and related plasma markers strongly separated AD from controls and predicted MCI conversion to AD in a small MCI sample, but independent replication in a larger sample is needed (41).

In our study, a combination of five measures led to strong predictive accuracy for MCI conversion to AD, and markedly added value to the combination of age and MMSE. These findings clearly need independent replication in larger samples before widespread clinical use is justified. Another potential use is to enrich patient selection in treatment trials.


Dr. Devanand had full access to all of the data in this study and takes responsibility for the integrity of the data and accuracy of the data analysis. This study was supported by National Institute of Aging Grant No. R01AG17761; by Grants R01AG12101, P50 AG08702, and P30 AG08051 from the National Institute of Aging; and by Grant Nos. MH55735, MH35636, and MH55646 from the National Institute of Mental Health. This study was presented in part at the 43rd Annual American College of Neuropsychopharmacology Meeting, December 12, 2004; San Juan, Puerto Rico.

Dr. Richard Mayeux's contribution to the interpretation of data is acknowledged.

Dr. Doty is a major shareholder in Sensonics, Inc., the manufacturer of the University of Pennsylvania Smell Identification Test. Dr. Devanand received research funding from Eli Lilly, Novartis, and Forest Labs, and consultancy fees from GlaxoSmithKline. Dr. Tabert is now a full-time employee of International Flavors and Fragrances. Dr. Bell is a speaker for Pfizer/Eisai, Novartis, and Bayer, GE, and Merck. Dr. Stern received research funding from Elan, Wyeth, and Eli Lilly, and consultancy fees from Elan, Wyeth, and Bayer. Dr. Pelton received research funding from Forest and Novartis. Drs. Liu, Pradhaban, and Cuasay report no biomedical financial interests or potential conflicts of interest.


Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


1. Di Carlo A, Lamassa M, Baldereschi M, et al. CIND and MCI in the Italian elderly: frequency, vascular risk factors, progression to dementia. Neurology. 2007;68:1909–1916. [PubMed]
2. Larrieu S, Letenneur L, Orgogozo JM, et al. Incidence and outcome of mild cognitive impairment in a population-based prospective cohort. Neurology. 2002;59:1594–1599. [PubMed]
3. Petersen RC. Mild cognitive impairment: current research and clinical implications. Semin Neurol. 2007;27:22–31. [PubMed]
4. Lopez OL, Jagust WJ, DeKosky ST, et al. Prevalence and classification of mild cognitive impairment in the Cardiovascular Health Study Cognition Study. Part 1. Arch Neurol. 2003;60:1385–1389. [PubMed]
5. Winblad B, Palmer K, Kivipelto M, et al. Mild cognitive impairment--beyond controversies, towards a consensus: report of the International Working Group on Mild Cognitive Impairment. J Intern Med. 2004;256:240–246. [PubMed]
6. Devanand DP, Michaels-Marston KS, Liu X, et al. Olfactory deficits in patients with mild cognitive impairment predict Alzheimer’s disease at follow-up. Am J Psychiatry. 2000;157:1399–1405. [PubMed]
7. Tabert MH, Liu X, Doty RL, et al. A 10-item smell identification scale related to risk for Alzheimer’s disease. Ann Neurol. 2005;58:155–160. [PubMed]
8. Tabert MH, Albert SM, Borukhova-Milov L, et al. Functional deficits in patients with mild cognitive impairment: prediction of AD. Neurology. 2002;58:758–764. [PubMed]
9. Graves AB, Bowen JD, Rajaram L, et al. Impaired olfaction as a marker for cognitive decline: interaction with apolipoprotein E epsilon4 status. Neurology. 1999;53:1480–1487. [PubMed]
10. Portet F, Ousset PJ, Visser PJ, et al. Mild cognitive impairment (MCI) in medical practice: a critical review of the concept and new diagnostic procedure. Report of the MCI Working Group of the European Consortium on Alzheimer’s Disease. J Neurol Neurosurg Psychiatry. 2006;77:714–718. [PMC free article] [PubMed]
11. Tabert MH, Manly JJ, Liu X, et al. Neuropsychological Prediction of Conversion to Alzheimer’s Disease in Patients with Mild Cognitive Impairment. Arch Gen Psychiatry. 2006;63:916–924. [PubMed]
12. Devanand DP, Pradhaban G, Khandji A, et al. Hippocampal and entorhinal atrophy in the prediction of MCI conversion to Alzheimer’s disease. Neurology. 2007;68:828–836. [PubMed]
13. Fleisher AS, Sowell BB, Taylor C, et al. Clinical predictors of progression to Alzheimer disease in amnestic mild cognitive impairment. Neurology. 2007;68:1588–1595. [PubMed]
14. Stoub TR, Bulgakova M, Leurgans S, et al. MRI predictors of risk of incident Alzheimer disease: a longitudinal study. Neurology. 2005;64:1520–1524. [PubMed]
15. Folstein MF, Folstein SE, McHugh PR. “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12:189–198. [PubMed]
16. Stern Y, Hesdorffer D, Sano M, Mayeux R. Measurement and prediction of functional capacity in Alzheimer’s disease. Neurology. 1990;40:8–14. [PubMed]
17. Petersen RC, Smith GE, Waring SC, et al. Mild cognitive impairment: clinical characterization and outcome. Arch Neurol. 1999;56:303–308. [PubMed]
18. McKhann G, Drachman D, Folstein M, et al. Clinical diagnosis of Alzheimer disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer Disease. Neurology. 1984;34:939–944. [PubMed]
19. Devanand DP, Pelton GH, Zamora D, et al. Predictive utility of apolipoprotein E genotype for Alzheimer disease in outpatients with mild cognitive impairment. Arch Neurol. 2005;62:975–980. [PubMed]
20. Doty RL, Shaman P, Dann M. Development of the University of Pennsylvania Smell Identification Test: a standardized microencapsulated test of olfactory function. Physiology and Behavior (Monograph) 1984;32:489–502. [PubMed]
21. Pfeffer RI, Kurosaki TT, Harrah CH, Jr, Chance JM, Filos S. Measurement of functional activities in older adults in the community. J Gerontol. 1982;37:323–329. [PubMed]
22. De Santi S, de Leon MJ, Rusinek H, et al. Hippocampal formation glucose metabolism and volume losses in MCI and AD. Neurobiol Aging. 2001;22:529–539. [PubMed]
23. Frye RE, Schwartz BS, Doty R. Dose-Related Effects of Cigarette Smoking on Olfactory Function. JAMA. 1990;263:1233–1236. [PubMed]
24. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. Springer; New York: 2001.
25. Tierney MC, Szalai JP, Snow WG, et al. A prospective study of the clinical utility of ApoE genotype in the prediction of outcome in patients with memory impairment. Neurology. 1996;46:149–154. [PubMed]
26. Visser PJ, Verhey FR, Scheltens P, et al. Diagnostic accuracy of the Preclinical AD Scale (PAS) in cognitively mildly impaired subjects. J Neurol. 2002;249:312–319. [PubMed]
27. Mayeux R, Saunders AM, Shea S, et al. Utility of the apolipoprotein E genotype in the diagnosis of Alzheimer’s disease. Alzheimer’s Disease Centers Consortium on Apolipoprotein E and Alzheimer’s Disease. N Engl J Med. 1998;338:506–511. [PubMed]
28. Tervo S, Kivipelto M, Hänninen T, et al. Incidence and risk factors for mild cognitive impairment: a population-based three-year follow-up study of cognitively healthy elderly subjects. Dement Geriatr Cogn Disord. 2004;17(3):196–203. [PubMed]
29. Blacker D, Lee H, Muzikansky A, et al. Neuropsychological measures in normal individuals that predict subsequent cognitive decline. Arch Neurol. 2007;64:862–871. [PubMed]
30. Tyas SL, Salazar JC, Snowdon DA, et al. Transitions to mild cognitive impairments, dementia, and death: findings from the Nun Study. Am J Epidemiol. 2007;165:1231–8. [PMC free article] [PubMed]
31. Frank RA, Galasko D, Hampel H, et al. National Institute on Aging Biological Markers Working Group. Biological markers for therapeutic trials in Alzheimer’s disease. Proceedings of the biological markers working group; NIA initiative on neuroimaging in Alzheimer’s disease. Neurobiol Aging. 2003;24:521–536. [PubMed]
32. Killiany RJ, Hyman BT, Gomez-Isla T, et al. MRI measures of entorhinal cortex vs hippocampus in preclinical AD. Neurology. 2002;58:1188–1196. [PubMed]
33. Doty RL, Shaman P, Applebaum SL, Giberson R, Sikorsky L, Rosenberg L. Smell identification ability: changes with age. Science. 1984;226:1441–1443. [PubMed]
34. Braak H, Braak E. Neuropathological staging of Alzheimer-related changes. Acta Neuropathol (Berl) 1991;82:239–259. [PubMed]
35. van de Pol LA, Korf ES, van der Flier WM, et al. Magnetic resonance imaging predictors of cognition in mild cognitive impairment. Arch Neurol. 2007;64:1023–1028. [PubMed]
36. Silverman DH, Small GW, Chang CY, et al. Positron emission tomography in evaluation of dementia: Regional brain metabolism and long-term outcome. JAMA. 2001;286:2120–2127. [PubMed]
37. Small GW, Kepe V, Ercoli LM, et al. PET of brain amyloid and tau in mild cognitive impairment. N Engl J Med. 2006;355:2652–2663. [PubMed]
38. Klunk WE, Engler H, Nordberg A, et al. Imaging brain amyloid in Alzheimer’s disease with Pittsburgh Compound-B. Ann Neurol. 2004;55:306–319. [PubMed]
39. Engler H, Forsberg A, Almkvist O, Blomquist G, Larsson E, Savitcheva I, Wall A, Ringheim A, Långström B, Nordberg A. Two-year follow-up of amyloid deposition in patients with Alzheimer’s disease. Brain. 2006;129:2856–66. [PubMed]
40. Hansson O, Zetterberg H, Bucchave P, et al. Association between CSF biomarkers and incipient Alzheimer’s disease in patients with mild cognitive impairment: a follow-up study. Lancet Neurology. 2006;5:228–234. [PubMed]
41. Ray S, Britschgi M, Herbert C, et al. Classification and prediction of clinical Alzheimer’s diagnosis based on plasma signaling proteins. Nat Med. 2007;13(11):1359–62. [PubMed]