Rate of progression of mild cognitive impairment to dementia – meta-analysis of 41 robust inception cohort studies


Alex J Mitchell, Liaison Psychiatry, University Hospitals Leicester and University of Leicester, Leicester, UK.
E-mail: alex.mitchell@leicspart.nhs.uk


Objective:  To quantify the risk of developing dementia in those with mild cognitive impairment (MCI).

Method:  Meta-analysis of inception cohort studies.

Results:  Forty-one robust cohort studies were identified. To avoid heterogeneity clinical studies, population studies and clinical trials were analysed separately. Using Mayo defined MCI at baseline and adjusting for sample size, the cumulative proportion who progressed to dementia, to Alzheimer’s disease (AD) and to vascular dementia (VaD) was 39.2%, 33.6% and 6.2%, respectively in specialist settings and 21.9%, 28.9% and 5.2%, respectively in population studies. The adjusted annual conversion rate (ACR) from Mayo defined MCI to dementia, AD and VaD was 9.6%, 8.1% and 1.9%, respectively in specialist clinical settings and 4.9%, 6.8% and 1.6% in community studies. Figures from non-Mayo defined MCI and clinical trials are also reported.

Conclusion:  The ACR is approximately 5–10% and most people with MCI will not progress to dementia even after 10 years of follow-up.


  • • In studies to date, typically less than half of individuals with mild cognitive impairment (MCI) progress to dementia.
  • • The annual conversion rate is about 7% to dementia and AD, and 2% for VaD.
  • • Risk of progress is influenced by the definition and subtype of MCI and the setting.


  • • Studies of less than 3 years duration were not analysed.
  • • Only placebo data from trial settings was included.
  • • The significance of subject memory complaints was difficult to study.


From the time mild cognitive impairment (MCI) was first defined its definition and course have been subject to considerable debate (1). Petersen and colleagues at the Mayo clinic suggested diagnostic criteria for MCI, as well as introducing amnestic (aMCI), non-memory (nmMCI) and multi-domain (mMCI) subtypes although these concepts were not rigorously field tested (2, 3). Since then there has been further controversy about the optimal definition and utility of MCI (4–8). The Mayo clinic criteria are thought by some to be too restrictive, particularly in the requirement for subjective memory complaints (SMC) and objective memory performance deficits 1.5 standard deviations (SDs) below age-adjusted means. It has been suggested that these criteria might reduce the proportion diagnosed with MCI too severely (9). Several large studies have recently reported on the community prevalence of strictly defined MCI. In a door-to-door survey of 1600 residents of Tuscany, Italy, the prevalence of aMCI was 4.9%, lower than the prevalence of dementia itself (6.2%) (10). In a study of 806 subjects (60–76 years of age) from eastern Finland, 5.3% met the Mayo MCI criteria (11). A cross-sectional community study incorporating neuropsychological testing found a 14.9% prevalence amongst 745 subjects in Kolkata, India (12). In the Canadian Study of Health and Aging (CSHA) a strict definition of MCI yielded a 1% prevalence of MCI whereas a modified definition increased the prevalence to 3% (13). In the latter study, cases with dementia were specifically excluded from the denominator when calculating prevalence rates. Recently, simplified criteria for MCI have been proposed by the Working Group of the European Consortium on Alzheimer’s Disease and the MCI Working Group of the European Consortium on Alzheimer’s Disease (see box 1) (14, 15). However, others have suggested their own criteria, most commonly omitting the requirement for subjective cognitive (memory) complaints (6, 16).

Table Box 1..   Simplified Consensus Criteria for MCI from Two Working Groups
A. Moderate cognitive deficits, short of dementia
B. Self-reported and/or informant reported cognitive complaints
C. Impairment on objective clinical cognitive tests
D. Preserved basic activities of daily living and minimal impairment in complex instrumental functions
Adapted from Winblad et al. 2004 (14) and Portet et al. 2006 (15).

The clinical value of MCI is thought to be its ability to predict later dementia. Bruscoli and Lovestone (2004) reviewed 19 longitudinal studies linking MCI with subsequent dementia although only nine of these had duration of 3 years or longer (17). In this narrative review, the annual conversion rate (ACR) was approximately 10% but with large differences between studies that were unexplained. A 10–15% ACR has been very widely cited but rarely scrutinized using robust data (18). It is possible that a surplus of short-term studies has given a misleading impression of the ACR. If a 10% ACR held true then within 10 years of diagnosis almost all surviving MCI suffers would have dementia. Peterson suggested that as many as 90% might develop dementia if followed for long enough (19). He was not alone in this assertion as other groups documented a 44–64% conversion rate within just 2 years of initial observation (1, 20, 21). This has led many to suggest that MCI is an inescapable intermediate stage between normal ageing and dementia (18). Against this, it is well known that not all patients with MCI deteriorate. In fact some patients appear to improve over time (22, 23). For example, Wolf and colleagues (1998) were one of the first groups to show that over almost 3 years 19.5% of MCI sufferers had recovered and an additional 61% neither improved nor deteriorated (22). An alternate possibility therefore is that MCI may not be a homogenous condition but may comprise several disease groups united by propensity to cause modest cognitive impairment. If this hypothesis were true one might expect considerable variation in the long-term natural history and an adequate response to treatment in some but not all sufferers.

Aims of the study

Given this debate, the primary aim of this study was to systematically compile and analyse the rates of conversion from MCI to dementia but taking into account variations in sampling. Our hypothesis was that studies adhering to a strict MCI definition (particularly those requiring SMC) would show higher progression to dementia than other studies. Our secondary aim was to calculate the relative risk of progression when compared with elderly controls.

Material and methods

Inclusion criteria and definition of MCI

Two minimum standards of sample size and study duration were set to ensure that only robust studies were added (see exclusions below for further details). Where authors adhered to the original 1997 Petersen et al. criteria (2) for MCI or the 2001 revised Petersen et al. criteria (the latter now called aMCI) (18) the study was denoted as following a Mayo clinic definition. Where authors deviated from this, typically by omitting the requirement for subjective cognitive complaints, then the study was denoted as using non-Mayo clinic criteria. Regarding outcome measures, we allowed any operational (such as ICD10, DSMIIIR or DSMIV) or neuropsychologically based definition of dementia and further noted specific conditions of Alzheimer’s Disease (AD) or vascular dementia where defined by authors.

Excluded studies

Studies were excluded that defined cognitive impairment solely on the basis of Age-Associated Memory Impairment, Cognitive Impairment No Dementia (CIND) or Clinical Dementia Rating (CDR) definitions. We excluded studies that examined the progression of dementia alone (baseline CDR = 1.0) or those that examined risk in mixed samples comprised of both MCI individuals and healthy controls (CDR 0.5 + 0) (24, 25). We excluded short-term follow-up studies of less than 3 years mean duration, as these were most likely to lead to inaccurate conclusions as a short follow-up period reduces the power of a cohort study. Equally, we excluded studies where the sample size of cases or controls was less than 20 subjects (26, 27). We excluded studies of cognitive impairment that were dependent upon specific aetiological factors such as alcohol (Wernicke–Korsakoff dementia), vitamin or metabolic deficiencies or infections, unless other types of dementia were measured (28, 29). We excluded case–control studies, which had no longitudinal period of observation (30, 31). We made every effort to only include analysis of unique patients and controls. Several groups reported findings in multiple publications, particularly those from large databases such as the CSHA. Where possible we excluded studies reporting subsamples and included only the main analysis (13, 32–35). We included data from randomized trials but only using placebo data as this was considered more representative of the natural history of MCI. We excluded studies that presented inadequate data for inclusion in the analysis. For example, those in which no raw number was presented (or calculable), those that examined predictors of decline alone and those that looked at other aspects of prognosis such as functional decline, hospitalization and mortality.

Search criteria

A systematic literature search, critical appraisal of the collected studies and pooled analysis were conducted. The following abstract databases were searched: Medline/Pubmed 1966 to October 2008, PsycINFO 1887 to October 2008 and Embase 1980 to October 2008. Five full-text collections, Science Direct, Ingenta Select, Ovid Full text, Blackwell Online and Wiley Interscience were searched. The abstract database Web of Knowledge (ISI) was searched, using the above terms as a text word search, and using key papers in a reverse citation search. The following search terms were used ‘progress* or convert* or prognosis or deteriorate or develop or decline* or cohort or longitud* or prospective* [Abstract]’ AND ‘Dementia or Alzheimer* or mild cognitive [Abstract].’

Statistical pooled and meta-analysis

All primary data were extracted as raw numbers and pooled to calculate raw cumulative conversion and then corrected for years of observation and sample size. Chi-squared test of significance was used for relative risks. We calculated rates of progression as a proportion of those recruited at baseline (inception cohort method) rather than those that survived to follow-up, as this most closely resembles clinical practice when attempting to give estimates of prognosis. Further most studies did not present clear completer data. We also calculated person years of observation in each type of study. Weighted proportion meta-analysis was used to adjust for study size using the DerSimonian–Laird model to allow for heterogeneity.



Search results

Results of the initial search and exclusions are shown in Fig. 1. There were insufficient cases of mixed dementia, Lewy body dementia and frontotemporal dementia to analyse. We found 41 studies that met inclusion but not exclusion criteria (Tables 1 and 2). Twenty-five studies were conducted using strict Mayo clinic criteria, three of which were from randomized trials (36–38). Of the remaining 22, 14 were clinical studies conducted in specialist centres, largely memory clinics (39–52) and eight were population studies conducted in the community (53–60). Sixteen were conducted using modified (non-Mayo) criteria. Of these four were conducted in specialist centres (61–64) and 12 in the community (65–76). Of 25 studies using a Mayo clinic definition the mean observation period was 4.57 years (SD 2.14) corresponding to 22 872 person years of observation. In those conforming to a non-Mayo definition the mean observation period was 4.59 years (SD 1.1) corresponding to 16 632 person years of observation.

Figure 1.

 QUOROM flow diagram.

Table 1.   Methodological summary of rates of progression from MCI to AD
Observation periodStudy period (years, mean)SettingMCI definitionOutcome
  1. 3MS, Modified Mini-Mental State Examination; MCI, mild cognitive impairment; aMCI, amnestic variant of mild cognitive impairment; mMCI, multi-domain variant of mild cognitive impairment; nMCI, non-memory variant of mild cognitive impairment; HR, hazard ratio; RR, relative risk; MMSE, Mini-mental state examination; BSRT, Babcock Story Recall Test; ILSA, Italian Longitudinal Study on Aging; CDR, clinical dementia rating; SD, standard deviation.

Mayo clinic definition of MCI
 Visser et al. (2006) (39)10    Memory clinicMCI diagnostic criteria in memory clinic Amnestic MCI was defined as a score 1.5 SD below age-, sex-, and education-corrected means on the learning measure or delayed recall measure of the RAVLT or a 10-word listDementia
 Ganguli et al. (2004) (53)10    Pennsylvania Community sample: MoVIES-studyRetrospective application of The Petersen amnestic MCI criteria were operationalized as i) impaired memory: Word List Delayed Recall score of <1 SD below mean; ii) normal mental status: Mini-Mental State Examination score of 25+; iii) normal daily functioning: no instrumental impairments; iv) memory complaint: subjective response to standardized question; v) not demented: CDR Scale score of <1Dementia (DSMIIIR)
 Annerbo et al. (2006) (49)6   Specialist centreRevised Petersen crtieria
Subjective memory complaints, objective memory impairment (by neuropsychological tests), not demented/not normal, and no impairment in activity in daily living
Dementia by DSMIV and ICD10
 Busse et al. (2006) (54)6   CommunityMCI original criteria
Sample 75 years or older
Dementia (DSMIV)
 Hansson et al. (2007) (40)5.2Specialist centreOriginal Petersen crtieriaConversion to DSMIIIR Dementia and NINCDS-ADRDA AD
 Larrieu et al. (2002) (60)5   Personnes Ages QUID study (PAQUID) followed a community-based cohort of healthy elderly for 5 yearsMCI was defined as memory complaints with objective memory impairment according to Peterson et al.Dementia (DSMIIIR)
 Gabryelewicz et al. (2008) (51)5   Specialist centrePetersen crtieriaConversion to DSMIIIR dementia, AD
 Dickerson et al. (2007) (58)5   Community volunteersMCI defined by a. informant based complaints b. MMSE above 23 c. objective deficits d. no functional decline e. no dementiaNINCDS_ADRDA criteria AD
 Visser & Verhey (2008) (42)5   Maastricht memory clinicAmnestic MCI (Petersen et al. 1999) was defined as a score 1.5 SD below the mean of a reference population after correction for age, gender and education and with no functional impairment (a score below 4 on the GDS).
Single-domain and multidomain MCI considered together under aMCI
 Artero et al. (2008) (59)4   Community (three cities in France)Classical criteria (a) presence of a cognitive complaint from either the subject or a family member, (b) absence of dementia, (c) change from normal cognitive functioning, (d) decline in any area of cognitive functioning, and (e) preserved overall general functioning but maybe increasing difficulty in the performance of activities of daily living.dementia (DSMIV), AD and VaD
 Petersen et al. (1999) (41)4  Mayo clinicCDR 0.5 plus Mayo clinic criteriaOutcome AD
 Tabert et al. (2006) (42) 3.88Memory clinicPeterson criteriaConversion to AD (not generic dementia)
 Bombois et al. (2008) (52)3.8Lille memory clinicClassical criteriaDementia, AD, VaD, LBD
 Jack et al. (2004) (44)3.6Mayo clinic (community and specialist referral patients)Criteria for the diagnosis of MCI were as follows: i) memory complaint documented by the patient and collateral source, ii) normal general cognition, iii) normal activities of daily living, iv) no dementia v) memory impairment for age and educationAD
 Yaffe et al. (2006) (48)3.1Mayo clinic (community and specialist referral patients)MCI defined as cognitive impairment but did not meet criteria for dementia
MCI subtypes also studied
AD (NINCDS-ADRDA criteria), vascular (State of California criteria)
AD outcome not clearly reported
 Gabryelewicz et al. (2006) (45) 3.02Specialist centreThe diagnosis of MCI according to Mayo Clinic Petersen’s Criteria was conducted by a panel of specialistsDSMIII dementia
 Marcos et al. (2006) (46)3   Specialist centrePetersen et al. criteria for MCI (all had SMC)Progression to AD
 Herukka et al. (2005) (47)3   Specialist centrePetersen et al. criteria for MCI including subjective memory complaints.AD (method not stated)
 Palmer et al. (2007) (56)3   Community
Kungsholmen Project, a longitudinal population-based study of those aged 75 years living in the Kungsholmen area of Stockholm, Sweden.
Mayo clinic Petersen Criteria for baseline MCI.
aMCI and mMCI combined in data
Dementia (DSMIIIR)
AD by criteria similar to NINCDS_ADRDA criteria
 Ravaglia et al. (2006) (50)3   Specialist
Participants were recruited among the outpatients seeking medical advice for cognitive complaints at the Center for Physiopathology of Aging, University of Bologna.
MCI subtypes diagnosed according to standardized Petersen criteria (amnestic; impairment of memory plus other cognitive domains; non-amnestic)Dementia
 Ritchie et al. (2001) (57)3   Community (recruited from GP research network)Original Petersen criteria)Dementia (DSMIIIR)
 Zanetti et al. (2006) (55)3   CommunityMayo clinic Petersen Criteria for MCIDementia
Non-Mayo clinic definition of MCI
 Tyas et al. (2007) (65)8   Community
Analyses were based on seven annual examinations (1991–2002) of 470 Nun Study participants living in the United States
MCI had at least one specific area of impaired cognitive function, such as memory or naming, but had intact global cognitive ability and activities of daily livingDementia (clinical criteria)
 Aggarwal et al. (2005) (66)5.1    CommunityClinical diagnoses were performed using a three-stage process, including neuropsychological testsNINCDS_ADRDA
Criteria AD
 Devanand et al. (2007) (61)5   Specialist clinicalBroadly defined MCI
Selective Reminding Test (SRT)
Criteria AD
 Bozoki et al. (2001) (62)5   Specialist centre
Michigan Alzheimer’s disease research centre
Definition consisted of neuropsychological deficits plus MMSE over 23 plus not depressed
They assessed cognitive function and divided them into two groups of M− (=aMCI) and M+
 Hogan & Ebly (2000) (68)5   Community and institution
Canadian Study of Health and Aging (CSHA) randomly selected community- and institution-dwelling respondents, aged 65 years and older
MCI ICD10 type 2 (short or long-term memory impairment with intellectual decline but no functional impairment)Dementia (DSMIIIR)
 Grober et al. (2000) (67)5   Community
Elderly volunteers from the Einstein Aging Study
Blessed score without dementia 
 Ishikawa & Ikeda (2007) (69)5   CommunityMCI without SMC i) normal general cognitive function, with MMSE 24; ii) objective memory impairment, assessed by three-words recall in MMSE (delayed recall 0/3 or 1/3); iii) neuropsychiatric examination: absence of dementia or depression, diagnosed by geriatric neuropsychiatrists according to the DSM-III R and iv) no ADL impairmentDementia
 Heun et al. (2006) (70)4.7CommunityPeterson (1999) Mayo criteriaAD [interviewed using the Structured Interview for Diagnosis of Dementia of the Alzheimer Type, Multi-infarct Dementia, and Dementia of other Aetiology (SIDAM)]
 Bennett et al. (2002) (71)4.5Special community
All subjects were older Catholic nuns, priests, and brothers
Cognitive Impairment short of dementiaNINCDS_ADRDA
Criteria AD
 Storandt et al. (2002) (63)4.4Alzheimer Disease Research CenterFifty-nine people in the CDR 0.5 group received a global CDR of 0.5 but had impairments in fewer than four areas and thus did not meet our strict diagnostic criteria for AD
Group similar to MCI, as they had memory complaints, poor memory performance, CDR of 0.5, and preserved general cognitive abilities as defined by MMSE scores of _24 and do not meet criteria for AD
Dementia and AD by expert evaluation with some postmortem confirmation
 Lopez et al. (2007) (72)4.3Community (Cardiovascular Health Study)CHS Cognition Study diagnostic criteria for MCI
MCI-AT included subjects with impairments (defined as performance <1.5 SD from controls) and possible mild defects of IADL
MCI-MCDT, required impairments in at least one cognitive domain (other than memory), or one abnormal test in at least two domains (defined as performance <1.5 SD) from controls, without severe impairment, in activities of daily living
Dementia and AD
 Di Carlo et al. (2007) (76)3.9Community From the ILSAMCI using neuropsychological tests
Subjects not scoring 1 SD only on the BSRT and preserved attention were defined as MCI
Dementia (DSMIIIR)
 Solfrizzi et al. (2004) (74)3.5Community From the ILSAClinical criteria of MCI using neuropsychological tests including the BSRT and MMSEDementia (DSMIIIR)
 Solfrizzi et al. (2007) (75)3.5Community From the ILSAClinical criteria of MCI using neuropsychological tests including the BSRT and MMSEDementia (DSMIIIR)
 Alexopoulos et al. (2006) (64) 3.49Specialist
Outpatients register for cognitive disorders
Impaired cognition plus preserved basic ADL plus CDR = 0.5 without dementiaDementia (ICD10)
 Tschanz et al. (2006) (73)3   CommunityBaseline cognitive disorder as indicated by mild difficulty in daily functioning (based on informant report) or objective impairment on neuropsychological testing using 3MS and IQCODE
In addition those with those with early memory involvement and/or neuropsychological testing consisting of a predominance of memory impairment (1.5 or more SD below age-corrected means or percentile equivalent) with no or lesser impairment of other cognitive domains were sub-classified as prodromal AD whereas others were called other cognitive syndromes
Dementia (DSMIIIR)
NINCDS-ADRDA AD also examined
Randomized trial settings
 Thal et al. (2005) (36)4RCT in specialist centresCriteria: patient reports memory problem, or informant reports that patient has memory problem; informant reports that patient’s memory has declined in the past year; MMSE score over24 plus neuropsychological criteriaNINCDS_ADRDA
Criteria AD
 Feldman et al. (2007) (38)4RCT in specialist centresAll patients were identified at entry to have MCI operationally defined by having cognitive symptoms, a global CDR stage of 0.5 and a score of less than 9 on the New York University delayed paragraph recall testNINCDS_ADRDA
Criteria AD
 Petersen et al. (2005) (37)3RCT in specialist centresAmnestic mild cognitive impairment of a degenerative nature (insidious onset and gradual progression), seven impaired memory, a Logical Memory delayed-recall score approximately 1.5–2 SD below an education-adjusted norm, a CDR of 0.5, a score of 24 to 30 on MMSE, and an age of 55–90 yearsNINCDS_ADRDA
Criteria AD
Table 2.   Statistical summary of rates of progression from MCI to AD
StudyDuration (years, mean)SettingMCI sample sizeProportion developing dementiaProportion developing ADProportion developing VaDACR dementiaACR ADACR VaD
  1. MCI, mild cognitive impairment; AD, Alzheimer’s disease; VaD, vascular dementia; ACR, annual conversion rate.

Mayo clinic definition of MCI
 Visser et al. (2006) (39)10Specialist centre640.4380.438 0.0440.044 
 Annerbo et al. (2006) (49)6Specialist centre930.344  0.057  
 Hansson et al. (2007) (40)5.2Specialist centre1670.5630.4130.1080.1080.079 
 Visser & Verhey (2008) (42)5Specialist centre119 0.378  0.076 
 Gabryelewicz et al. (2008) (51)5Specialist centre1050.2950.2480.0380.0590.050 
 Petersen et al. (1999) (41)4Specialist centre76 0.474  0.118 
 Tabert et al. (2006) (42)3.88Specialist centre148 0.264  0.068 
 Bombois et al. (2008) (52)3.8Specialist centre1700.3940.1710.0410.1040.045 
 Jack et al. (2004) (44)3.6Specialist centre41 0.634  0.176 
 Gabryelewicz et al. (2006) (45)3.02Specialist centre1050.2100.1810.0290.0690.0600.009
 Yaffe et al. (2006) (48)3.1Specialist centre3270.609  0.196  
 Herukka et al. (2005) (47)3Specialist centre78 0.295  0.098 
 Marcos et al. (2006) (46)3Specialist centre82 0.463  0.154 
 Ravaglia et al. (2006) (50)3Specialist centre1650.2910.2060.0850.0970.0690.028
 Ganguli et al. (2004) (53)10Community400.2750.225 0.0280.023 
 Busse et al. (2006) (54)6Community910.385  0.064  
 Dickerson et al. (2007) (58)5Community42 0.405  0.081 
 Artero et al. (2008) (59)4Community28820.0660.0420.0070.0160.011 
 Palmer et al. (2007) (56)3Community47 0.574  0.191 
 Zanetti et al. (2006) (55)3Community650.3080.1690.1380.1030.0560.046
 Larrieu et al. (2002) (60)5Community58 0.448  0.090 
 Ritchie et al. (2001) (57)3Community270.111  0.037  
Non-Mayo clinic Definition of MCI
 Devanand et al. (2007) (61)5Specialist centre139 0.266  0.053 
 Bozoki et al. (2001) (62)5Specialist centre48 0.438  0.088 
 Storandt et al. (2002) (63)4.4Specialist centre59 0.322  0.073 
 Alexopoulos et al. (2006) (64)3.49Specialist centre810.407  0.117  
 Tyas et al. (2007) (65)8Community9490.075  0.009  
 Aggarwal et al. (2005) (66)5.1Community2180.3850.376 0.0760.074 
 Grober et al. (2000) (67)5Community680.294  0.059  
 Hogan & Ebly (2000) (68)5Community2100.1620.105 0.0320.021 
 Ishikawa & Ikeda (2007) (69)5Community980.2240.1120.0510.0450.0220.010
 Heun et al. (2006) (70)4.7Community6330.0880.0280.0140.0190.0060.003
 Bennett et al. (2002) (71)4.5Community188 0.340  0.076 
 Lopez et al. (2007) (72)4.3Community1360.515  0.120  
 Di Carlo et al. (2007) (76)3.9Community4450.052  0.013  
 Solfrizzi et al. (2004) (74)3.5Community1130.186  0.053  
 Solfrizzi et al. (2007) (75)3.5Community1210.256  0.073  
 Tschanz et al. (2006) (73)3Community1200.458  0.153  
Randomized trial settings
 Thal et al. (2005) (36)4Specialist centre732 0.112  0.028 
 Petersen et al. (2005) (37)3Specialist centre259 0.282  0.094 
 Feldman et al. (2007) (38)4Specialist centre510 0.214  0.053 

Progression of mayo clinic defined MCI

Cumulative risk of dementia and related conditions.  Across 13 studies involving a total sample of 4301 individuals with Mayo defined MCI 32.3% (95% CI = 17.9–48.5%) converted to generic dementia over the natural observation period. Meta-analysis by settings revealed a 39.2% conversion in specialist settings (95% CI = 28.7–50.3%) and 21.9% (95% CI = 7.3–41.6%) in community studies. Only two groups documented a total conversion rate over 50% and these were both from specialist centres (40, 48).

Eighteen studies involving a total sample of 4454 individuals with Mayo defined MCI found that 624 (unadjusted crude rate of 14.0%) converted to ‘probable AD’ over the natural observation period. Of these only two groups documented a total conversion rate over 50% (44, 56). Meta-analysis stratified by settings showed a 33.6% conversion in specialist settings (95% CI = 26.5–41.1%) and 28.9% (95% CI = 9.3–53.9%) in community studies.

Seven studies involving 3659 individuals with Mayo defined MCI at baseline found that 74 progressed to VaD (unadjusted crude rate of 2.0%). There were insufficient studies to examine other types of dementia. Using random effects proportion meta-analysis this was adjusted to 6.2% (95% CI = 3.6–9.4%) in specialist settings and 5.2% (95% CI = 0.5–25.6%) in community studies.

Annual risk of dementia and specific dementias.  In studies adhering to the Mayo clinic definition, correcting for sample size, and dementia type the ACR in specialist settings was 9.6% (95% CI = 6.3–13.4%) to dementia, 8.1% (95% CI = 6.3–10.0%) to AD and 1.9% (95% CI = 1.0–3.0%) to VaD. The adjusted ACR in community settings was 4.9% (95% CI = 1.6–9.9%) to dementia, 6.8% (95% CI = 1.9–14.5%) to AD and 1.6% (95% CI = 0.3–9.4%) to VaD.

Progression of MCI in studies using non-Mayo definition of MCI

Cumulative risk of dementia and related conditions.  Twelve studies involving a total sample of 4456 individuals with non-Mayo defined MCI found that 520 (unadjusted crude rate of 16.3%; adjusted 24.1%, 95% CI 15.4–33.9%) converted to broadly defined dementia. Of those conducted in community settings, the corrected meta-analytic proportion was 22.7% (95% CI = 14.2–32.6%).

Three studies conducted in specialist settings, found that 33.1% (95% CI = 23.9–43.0%) progressed to AD compared with 17.1% (95% CI = 4.5–35.6%) for population studies in the community. Only two studies involving 731 non-Mayo MCI defined patients documented VaD as an outcome, both in the community. The meta-analytic cumulative proportion was 2.9% (95% CI = 0.36–7.8%).

Annual risk of generic dementia and specific dementias.  There was only one study for generic dementia in specialist settings using non-Mayo definition with an 11.7% ACR. For community studies there equivalent corrected ACR was 5.2% (95% CI = 2.9–8.0%). For AD the corrected ACR was 6.9% (95% CI = 4.1–10.4%) for clinical studies and 3.6% for community (population) studies (95% CI = 1.0–7.7%). There were no VaD specialist studies of this type but in the community the ACR was VaD 0.6% (corrected 0.54%, 95% CI = 0.068–1.4%) Table 3.

Table 3.   Summary of rates of annual progression from MCI to dementia and related conditions
SettingMayo defined MCI (N = 24)Non-Mayo defined MCI (N = 17)
Progression to Dementia (any type)Progression to ADProgression to VaDProgression to dementia (any type)Progression to ADProgression to VaD
Specialist (clinical)9.6% (N = 8) (95% CI = 6.3–13.4%)8.1% (N = 12) (95% CI = 6.3–10.0%)1.9% (N = 5) (95% CI = 1.0–3.0%)11.7% (N = 1) (95% CI = 5.6–20.7%)6.9% (N = 3) (95% CI = 4.1–10.4%)No data
Community (population)4.9% (N = 5) (95% CI = 1.6–9.9%)6.8% (N = 6) (95% CI = 1.9–14.5%)1.6% (N = 2) (95% CI = 0.3–9.4%)5.2% (N = 11) (95% CI = 2.9–8.0%)3.6% (N = 5) (95% CI = 1.0–7.7%).0.54% (N = 2) (95% CI = 0.068–1.4%)
Randomized trial settingNo data5.5% (N = 3) (95% CI = 2.5–9.5%)No dataNo dataNo dataNo data

Comparative risk of dementia in healthy controls

Eight studies (three using Mayo definition and five not using Mayo definition) examined the risk of dementia or AD in healthy age comparable controls head-to-head with the risk in individuals with MCI, a total pooled sample of 6713 individuals (39, 44, 56, 65, 67, 71, 73). One addition study was excluded because of an unrealistically small sample of just 18 individuals (26). Across all but one study (71) there were no significant demographic differences with the comparator MCI population. In five head-to-head studies where dementia was the outcome, the ACR (mean follow-up 6.0 years) was 3.6% to dementia vs. 0.43% for healthy subjects a relative risk of 8.36. Correcting for sample size, meta-analysis showed a pooled relative risk of 13.8 (95% CI = 8.44–22.6) which was statistically significant (P < 0.01). In six head-to-head studies for AD, the ACR (mean follow-up 6.0 years) was 3.5% to AD vs. 0.78% for healthy subjects, a relative risk of 4.47. This was adjusted to 8.93 (95% CI = 4.18–19.07, P < 0.01) on random effects meta-analysis Fig. 2.

Figure 2.

 Relative risk meta-analysis of annual progression in mild cognitive impairment (MCI) vs. healthy subjects. (a) Relative risk of Alzheimer’s disease. (b) Relative risk of dementia. Axis indicates the relative risk of conversion to dementia in those with MCI vs. those without. The size of the boxes represent the size of the sample in the study.

Progression of dementia and/or AD in MCI subgroups

Nine studies examined the trajectories of subtypes of MCI, specifically aMCI and/or mMCI and/or nmMCI (42, 44, 48, 50, 55, 62, 64, 76, 77). The corrected ACR for aMCI was 11.7% (nine studies sample size = 646), mMCI 12.2% (eight studies, sample size = 446) and nmMCI 4.1% (five studies, sample size = 354). There was a significant difference in outcomes (aMCI vs. nmMCI χ2 = 35.1, P = 0.0001; mMCI vs. nmMCI χ2 = 35.0, P = 0.0001).

Progression of MCI in long-term clinical trials

Cumulative risk of dementia.  No studies were identified that defined generic dementia.

Cumulative risk of specific dementias.  Three studies involving placebo treatment of 1501 individuals with Mayo defined MCI found that 264 (18.4%) converted to AD over a mean observation period of 3.6 years. No studies examined outcomes of other dementias.

Annual risk of dementia and specific dementias.  The uncorrected ACR to AD was 4.1%, corrected to 5.5% (95% CI = 2.5–9.5%).


Across 41 robust cohort studies the overall ACR was 6.7% (95% CI = 4.6–9.1%) for progression to dementia, 6.5% (95% CI = 4.8–8.5%) for AD, and 1.6% (95% CI = 0.8–2.7%) for VaD all substantially less than previously thought. Nevertheless the relative risk of progression was 15.9 (MCI Mayo type) and 6.2 (MCI non-Mayo type) for dementia and 9.5 (MCI Mayo type) and 4.7 (MCI non-Mayo type) for AD when compared with healthy elderly individuals. The cumulative proportion that progressed to dementia rarely exceeded 50% even in long-term studies and averaged 32.3% for those with Mayo clinic defined MCI and 24.1% for those with non-Mayo criteria (30.2%, 22.9% for AD respectively). Even in six studies with an observation period of 5 years or more the cumulative conversion rate was only 38.2%. This suggests that many, perhaps most people with MCI do not deteriorate to dementia in the medium term (78, 79). A significant proportion improve and others do not survive long enough to allow dementia to develop. For those individuals with SMC fulfilling full criteria for MCI presenting to memory clinics then the risk of dementia is about 10% per year. Even then, in this high risk group the risk of progression appears to diminish with time (56, 71, 80, 81). For individuals who do not meet Mayo criteria for MCI (typically those without SMC) and for those seen in the community and annual risk of progression is usually about 5% with about one in five deteriorating to dementia over the next 5 years. Even in clinical trials from specialist centres the ACR is about 8%, demonstrating why large samples are needed to examine therapeutic effects. Related to this, the large AD Anti-inflammatory Prevention Trial of 2528 individuals recently showed that of those who developed incident dementia over 3 years only 63% (66% for AD) had a prodromal phase of CIND, MCI or a related condition (82). Taken together this suggests that MCI is neither necessary nor sufficient for later dementia. Certain types of cognitive impairment do not warrant a label of well-defined MCI, such as those with impaired global cognition but persevered memory, language, and visuospatial functioning but these individuals still carry a high risk of decline (58, 77).

From our data several key factors moderated rates of deterioration namely, the classification of MCI, the subtype of MCI and the setting. The relative risk of progression was approximately two-fold in those defined using strict criteria and in those recruited from specialist settings, but only about 1/3rd in those with nmMCI. The finding that risk was significantly higher in patients diagnosed using the Mayo clinic criteria may be partly because of the difference in requirement for SMC. It is to be expected that when omitting SMC, people with baseline cognitive deficits progress more slowly as SMC are a known risk factor for dementia (83). It is also likely that most of those presenting to specialist centres would have SMC whether SMC have been documented or not. There has been considerable interest in the proposed subtypes of MCI. We identified nine studies that examined differential rates of progression. There was a significantly higher ACR for mMCI and aMCI than nmMCI. Further studies are required before clear conclusions can be drawn about the validity of these MCI subtypes. Alternate classifications based on aetiology may be more successful.

This analysis has a number of limitations. Notably we were reliant on the quality and non-repetition of primary publications. We sought to ensure data integrity by excluding small studies of less than 3 years duration. However, long-term studies are subject to greater attrition and there is likely to be difference in risk when compared using an inception vs. completer cohort method [see Mitchell and Shiri-Feshki (84) for further discussion]. There is a possibility of double counting some individuals especially where several related publication arise from one group. We attempted to deal duplicate data from the same authors by excluding studies where the population characteristics were nearly identical to those already entered (see table). We included data from three RCTs but excluded others of less than 3 years duration and focused only on placebo arms. We also acknowledge considerable disagreement in the diagnostic definition of MCI, but attempted to allow for this by separating strict and modified criteria.

This meta-analysis supports the notion that MCI is not a unified disease but heterogeneous disorder with subtypes yet to be fully defined. Some underlying causes such as psychiatric disorders remain poorly described but are likely to have a distinct phenomenology and course. MCI can no longer be assumed to always be a simple transitional state between normal aging and dementia.


Many thanks to the helpful staff of the postgraduate medical library, Leicester General Hospital.

Declaration of interest