Patiënts with mild cognitive impairment (MCI) are a heterogeneous group, both in terms of etiology and prognosis. Sometimes an early stage of a neurodegenerative or other brain disease will eventually lead to dementia, but some patiënts with MCI never develop dementia. Better prediction of whether and when patiënts with MCI develop dementia is relevant. Several techniques may possibly contribute to a better prediction of which patiënts with MCI will and will not develop dementia:

• magnetic resonance imaging (MRI), including volumetry of gray and white matter and of specific brain structures;
• nuclear imaging of the brain, such as positron emission tomography (PET) of amyloid plaques and glucose metabolism;
• liquor cerebrospinalis (hereafter: CSF).

Summary of Findings

Summary of Findings table index-1: Magnetic Resonance Imaging (MRI) markers for predicting progression from MCI to Alzheimer dementia

Population: Patiënts with Mild Cognitive Impairment (MCI)

Intervention: MRI

Comparator: No marker

Reference: Pathological anatomical (PA) diagnosis or the clinical consensus diagnosis of dementia or diagnostic criteria based on DSM IV or NINCDS-ADRDA with an average minimum follow-up duration of 2 years

_{* Summary estimates are based on the summary operating point from a hierarchical summary ROC curve (HSROC) model. CI = confidence interval.}

_{^1. Risk of bias: serious. Due to lack of blinding, participant selection (registry data) or lack of reporting on the index test (unclear pre-specified definition for a ‘positive’ result).}

_{^2. Imprecision: serious. Wide 95% confidence intervals.}

_{^3. Inconsistency: serious. Due to sparse and inconsistent data.}

Summary of Findings table index-2: Cerebrospinal fluid (CSF) markers for predicting progression from MCI to Alzheimer dementia

Population: Patiënts with Mild Cognitive Impairment (MCI)

Intervention: CSF markers (amyloidβ1-42, total-tau (t-tau), phosporylated-tau181 (p-tau), ratio t-tau/aβ, ratio p-tau/aβ, neurofilament light)

Comparator: No CSF marker

Reference: Pathological anatomical (PA) diagnosis or the clinical consensus diagnosis of dementia or diagnostic criteria based on DSM IV or NINCDS-ADRDA with an average minimum follow-up duration of 2 years.

_{* The studies used various thresholds for calculating specificity and sensitivity, therefore summary estimates were not computed but derived from the HSROC model at the median value of specificity computed from the included studies. CI = confidence interval.}

_{^1. Risk of bias: very serious. Due to lack of reporting on enrolment or lacking consecutive or random enrolment of patiënts; no pre-set cut-off specified for the index test and not all patiënts were accounted for in the analysis or the time interval between the index test and reference standard was not appropriate (duration of follow-up was shorter than one year).}

_{^2. Inconsistency: serious. Due to sparse and inconsistent data with conflicting results.}

_{^3. Imprecision: very serious. Due to the optimal information size was not achieved.}

Summary of Findings table index-3: Positron Emitted Tomography (PET) markers for predicting progression from MCI to Alzheimer dementia

Population: Patiënts with Mild Cognitive Impairment (MCI)

Intervention: Positron Emitted Tomography (PET) markers

Comparator: No marker

Reference: Pathological anatomical (PA) diagnosis or the clinical consensus diagnosis of dementia or diagnostic criteria based on DSM IV or NINCDS-ADRDA with an average minimum follow-up duration of 2 years.

_{* Summary estimates of sensitivity and specificity were not computed because the studies used various thresholds for calculating specificity and sensitivity.}

_{^1. Risk of bias: very serious. Due to lack of blinding or lack of reporting on the index test (for Zhang 2014 and Smailagic 2015 also unclear pre-specified definition for a ‘positive’ result) or reference test as well as the variation of methods used.}

_{^2. Inconsistency: serious. Due to sparse and inconsistent data with conflicting results.}

Description of studies

A total of seven studies were included in the analysis of the literature. One study for index-1 MRI (Lombardi, 2020), two studies for index-2 CSF markers (Ritchie, 2014; Ritchie, 2017) and four studies for index-3 PET (Cotta Ramusino, 2024; Martinez, 2017; Smailagic, 2015 and Zhang, 2014). Important study characteristics and results are summarized in table 2. The assessment of the risk of bias is summarized in the risk of bias tables (under the tab ‘Evidence tabellen’).

Index 1 Magnetic Resonance Imaging (MRI)

Lombardi (2020) performed a systematic review and meta-analysis to determine the diagnostic accuracy of MRI for detecting mild cognitive impairment (MCI) patiënts who convert to Alzheimer's disease dementia (ADD) over time. A systematic literature search was performed up until 29 January 2019. Studies were included if they had 1) prospective cohorts with a clinical follow-up diagnosis of Alzheimer’s disease dementia as a reference standard (delayed verification), 2) baseline MRI documented at or around the time the MCI diagnosis was made, 3) sufficient data to construct two by two tables expressing MRI results by disease status and 4) used either quantitative volumetric measurements or qualitative visual assessment of MRI to detect atrophy in specific brain regions or the whole brain. Case series, case-control and retrospective studies were excluded. 33 studies with 3935 participants evaluated the use of MRI to detect conversion to ADD. Of the 3935 participants, 1341 (34%) developed Alzheimer’s dementia. In the absence of specified thresholds, the authors used a hierarchical summary ROC curve (HSROC) model to estimate pooled accuracy measures as well as to investigate relative diagnostic odds ratios in subgroup analyses (assuming parallel ROC curves in logits). Studies yielded heterogeneous estimates and no clear threshold effects were apparent both graphically and statistically in analyses with more data. The metadas user-written command in Statistical Analysis System (SAS) (version 9.4. SAS Institute Inc., Cary, NC, USA) statistical package was used for the analyses.

Index 2 Cerebrospinal fluid (CSF)

Ritchie (2014) performed a systematic review and meta-analysis to determine the diagnostic accuracy of plasma and CSF Aß levels for detecting mild cognitive impairment (MCI) patiënts who would convert to Alzheimer's disease dementia (ADD) or other forms of dementia over time. A systematic literature search was performed up until 3 December 2012. Studies were included if they had 1) prospectively well-defined cohorts, 2) any accepted definition of MCI but no dementia, 3) baseline CSF or plasma Aß levels, or both, documented at or around the time the MCI diagnosis was made, 4) reference standard for Alzheimer’s dementia diagnosis based on the National Institute for National Institute of Neurological and Communicative Diseases and Stroke/ Alzheimer's Disease and Related Disorders Association (NINCDS- ADRDA) or the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria, 5) sufficient data to construct two by two tables expressing biomarker results by disease status. Studies were excluded if they included patiënts with possible other causes for dementia: psychiatric, neurological, metabolic, immunological, hormonal or cerebrovascular disorders, genetic cause or early onset ADD (if studies included patiënts below the age of 50 years, they were excluded). Although Ritchie 2014 was not limited to studies reporting amyloidß42, no studies were identified that reported on diagnostic accuracy for ADD for CSF amyloidß40 or CSF amyloidß42/amyloidß40 ratio. Fourteen studies with 1349 participants evaluated the use of CSF amyloidß42 to detect conversion to ADD. Of the 1349 participants, 436 developed Alzheimer’s dementia. Positive cases were determined based on the amyloidß42 thresholds employed in the respective primary studies. Because of variation in assay thresholds, the authors did not estimate summary sensitivity and specificity. Instead, the authors derived estimates of sensitivity at fixed values of specificity from the model fitted to produce a summary receiver operating characteristic (ROC) curve. The authors used a hierarchical summary receiver operating characteristic (HSROC) model that accounted for between study variability through the inclusion of random effects and derived sensitivities with 95% confidence intervals (CI) at median, lower and upper quartile values of the specificities from the included studies.

Ritchie (2017) performed a systematic review and meta-analysis to determine the diagnostic accuracy of 1) CSF t-tau, 2) CSF p-tau, 3) the CSF t-tau/Aβ ratio and 4) the CSF p-tau/Aβ ratio index tests for detecting MCI patiënts who would convert to ADD or other forms of dementia over time. A systematic literature search was performed in January 2013. Studies were included if they had 1) prospectively well-defined cohorts, 2) any accepted definition of MCI but no dementia, 3) CSF t-tau or p-tau and CSF tau (t-tau or p-tau)/Aβ ratio documented at or around the time the MCI diagnosis was made, 4) reference standard for Alzheimer’s dementia diagnosis based on the NINCDS- ADRDA or DSM-IV criteria, 5) sufficient data to construct two by two tables expressing biomarker results by disease status. Fifteen studies were included with 1282 participants with MCI at baseline. 1172 participants had analysable data, 430 participants developed Alzheimer’s dementia and 130 participants other forms of dementia. Follow-up ranged from less than one year to over four years, but in the majority of studies was in the range of one to three years. Diagnostic accuracy was evaluated for CSF t-tau in seven studies (291 cases and 418 non-cases), CSF p-tau in six studies (164 cases and 328 non-cases), CSF p-tau/Aβ ratio in five studies (140 cases and 293 non-cases) and CSF t-tau/Aβ ratio in only two studies. Because of variation in assay thresholds, the authors did not estimate summary sensitivity and specificity. Instead, the authors derived estimates of sensitivity at fixed values of specificity from the model fitted to produce a summary receiver operating characteristic (ROC) curve. The authors used a hierarchical summary receiver operating characteristic (HSROC) model that accounted for between study variability through the inclusion of random effects and derived sensitivities with 95% confidence intervals (CI) at median, lower and upper quartile values of the specificities from the included studies.

Index 3 Positron Emitted Tomography (PET)

Cotta Ramusino (2024) performed a systematic review to determine the diagnostic accuracy of molecular imaging markers for detecting mild cognitive impairment (MCI) patiënts who convert to Alzheimer's disease dementia (ADD) over time. The intention was to update existing reviews up until 2017. A systematic literature search was performed from 1 January 2017 up until 28 February 2022. Studies were included if they had 1) prospective cohorts with a clinical follow-up diagnosis of Alzheimer’s disease dementia as a reference standard (delayed verification), 2) sample size of at least 50 patiënts with MCI, 3) follow-up of at least 3 years, 4) reported on critical outcome measures, e.g., sensitivity, specificity, accuracy, area under the receiver, or operating characteristic curve (ROC or AUC) and 5) used molecular brain imaging techniques (amyloid-, tau-, [18F] FDG-PETs, DaT-SPECT, and cardiac [123I]-MIBG scintigraphy). No studies on tau-PET were eligible for inclusion. Eight studies with 1806 participants evaluated the use of 18F PET with florbetapir (n=7) or flutametamol (n=1) to detect conversion to ADD. Of the 1806 participants, 549 (30%) developed Alzheimer’s dementia. 25 studies with 6803 participants evaluated the use of 18F-FDG PET to detect conversion to ADD. Of the 6803 participants, 2572 (38%) developed Alzheimer’s dementia.

Martinez (2017) performed a systematic review and meta-analysis to determine the diagnostic accuracy of 18F PET with florbetapir for detecting mild cognitive impairment (MCI) patiënts who convert to Alzheimer's disease dementia (ADD) over time. A systematic literature search was performed up until May 2017. Studies were included if they had 1) prospective cohorts with a clinical follow-up diagnosis of Alzheimer’s disease dementia as a reference standard (delayed verification), 2) baseline PET documented at or around the time the PET diagnosis was made, 3) sufficient data to construct two by two tables expressing PET results by disease status and 4) used the 18F-florbetapir PET scan. A case control study with a delayed verification design was included, this occurred in the context of a cohort study so is a diagnostic nested case-control study (Doraiswamy, 2014). Two studies with 448 participants evaluated the use of 18F-florbetapir PET to detect conversion to ADD (Doraiswamy 2014, Schreiber 2015). Of the 448 participants, 69 (15,4%) developed Alzheimer’s dementia.

Smailagic (2015) performed a systematic review and meta-analysis to determine the diagnostic accuracy of 18F-FDG uptake by brain tissue as measured by PET for detecting mild cognitive impairment (MCI) patiënts who convert to Alzheimer's disease dementia (ADD) over time. A systematic literature search was performed up until January 2013. Studies were included if they had 1) prospective cohorts with a clinical follow-up diagnosis of Alzheimer’s disease dementia as a reference standard (delayed verification), 2) baseline PET documented at or around the time the PET diagnosis was made, 3) sufficient data to construct two by two tables expressing PET results by disease status and 4) used PET to detect 18F-FDFG uptake in the brain. Case control studies with a delayed verification design were included, these occurred in the context of a cohort study so are diagnostic nested case-control studies. 14 studies with 421 participants evaluated the use of 18F-FDG PET to detect conversion to ADD. Of the 421 participants, 150 (36%) developed Alzheimer’s dementia. Different brain regions were examined, all studies included the temporo-parietal lobes, 12 also included the posterior cingulate and part of the frontal lobes. Because of between study variation in thresholds and measures of 18F-FDG uptake, the authors did not estimate summary sensitivity and specificity. Instead, the authors derived estimates of sensitivity at fixed values of specificity from the model fitted to produce a summary receiver operating characteristic (ROC) curve. The authors used a hierarchical summary receiver operating characteristic (HSROC) model that accounted for between study variability through the inclusion of random effects and derived sensitivities with 95% confidence intervals (CI) at median values of the specificities from the included studies.

Zhang (2014) performed a systematic review and meta-analysis to determine the diagnostic accuracy of PET with the 11C-labelled Pittsburgh Compound-B (11C-PIB) ligand for detecting mild cognitive impairment (MCI) patiënts who convert to Alzheimer's disease dementia (ADD) over time. A systematic literature search was performed up until 12 January 2013. Studies were included if they had 1) prospective cohorts with a clinical follow-up diagnosis of Alzheimer’s disease dementia as a reference standard (delayed verification with a minimum of 1 year between time of MCI and time of dementia diagnosis), 2) baseline PET documented at or around the time the MCI diagnosis was made, 3) sufficient data to construct two by two tables expressing PET results by disease status and 4) used PET with the 11C-labelled Pittsburgh Compound-B (11C-PIB) ligand. Case control studies with a delayed verification design were included, these occurred in the context of a cohort study so are diagnostic nested case-control studies. Nine studies with 274 participants evaluated the use of PET to detect conversion to ADD. Of the 274 participants, 112 (41%) developed Alzheimer’s dementia. Because of between study variation in thresholds and measures of 11C-PIB amyloid retention, the authors did not estimate summary sensitivity and specificity. Instead, the authors derived estimates of sensitivity at fixed values of specificity from the model fitted to produce a summary receiver operating characteristic (ROC) curve. The authors used a hierarchical summary receiver operating characteristic (HSROC) model that accounted for between study variability through the inclusion of random effects and derived sensitivities with 95% confidence intervals (CI) at median, lower and upper quartile values of the specificities from the included studies.

Table 2. Characteristics of included systematic reviews

^{* For further details, see risk of bias table in the appendix}

Results

Index 1 Magnetic Resonance Imaging (MRI)

1.1 Total hippocampal volume

1.1.1 Specificity and sensitivity

The systematic review and meta-analysis of Lombardi (2020) included 22 studies with 2209 participants to assess the accuracy of total hippocampus volume measured by MRI to detect conversion from MCI to ADD compared to other outcomes. Of the 2209 participants, 687 were diagnosed with ADD. Individual study estimates of sensitivity were between 28% and 100% while the specificities were between 43% and 94%. Results are shown in Figure 1.

1.1.2 Positive predictive value

Lombardi (2020) did not report on the outcome positive predictive value.

1.1.3 Negative predictive value

Lombardi (2020) did not report on the outcome negative predictive value.

1.1.4 Positive likelihood ratio

Lombardi (2020) reported a positive likelihood ratio of 2.53 (95% CI 2.09 to 3.06).

1.1.5 Negative likelihood ratio

Lombardi (2020) reported a negative likelihood ratio of 0.38 (95% CI 0.29 to 0.50).

Figure 1. Forest plot of total hippocampal volume measured by structural MRI for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Lombardi 2020)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. Studies are ordered according to the estimates of sensitivity. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

1.2 Atrophy medial temporal lobe

1.2.1 Specificity and sensitivity

The systematic review and meta-analysis of Lombardi (2020) included seven studies with 1077 participants to assess the accuracy of medial temporal lobe volume measured by MRI to detect conversion from MCI to ADD compared to other outcomes. Of the 1077 participants, 330 were diagnosed with ADD. Individual study estimates of sensitivity were between 40% and 86% while the specificities were between 44% and 85%. Results are shown in Figure 2.

1.2.2 Positive predictive value

Lombardi (2020) did not report on the outcome positive predictive value.

1.2.3 Negative predictive value

Lombardi (2020) did not report on the outcome negative predictive value.

1.2.4 Positive likelihood ratio

Lombardi (2020) reported a positive likelihood ratio of 1.81 (95% CI 1.41 to 2.32).

1.2.5 Negative likelihood ratio

Lombardi (2020) reported a negative likelihood ratio of 0.56 (95% CI 0.46 to 0.67).

Figure 2. Forest plot of medial temporal lobe volume measured by structural MRI for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Lombardi 2020)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

1.3 Lateral ventricles volume

1.3.1 Specificity and sensitivity

The systematic review and meta-analysis of Lombardi (2020)  included five studies with 1077 participants to assess the accuracy of MRI t-tau to detect conversion from MCI to ADD compared to other outcomes. Of the 1077 participants, 371 were diagnosed with ADD.  Individual study estimates of sensitivity were between 51% and 75% while the specificities were between 47% and 73%. Results are shown in Figure 3.

1.3.2 Positive predictive value

Lombardi (2020) did not report on the outcome positive predictive value.

1.3.3 Negative predictive value

Lombardi (2020) did not report on the outcome negative predictive value.

1.3.4 Positive likelihood ratio

Lombardi (2020) reported a positive likelihood ratio of 1.61 (95% CI 1.39 to 1.87).

1.3.5 Negative likelihood ratio

Lombardi (2020) reported a negative likelihood ratio of 0.66 (95% CI 0.57 to 0.78).

Figure 3. Forest plot of lateral ventricles volume measured by structural MRI for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Lombardi 2020)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

1.4 Total entrorhinal cortex volume

1.4.1 Specificity and sensitivity

The systematic review and meta-analysis of Lombardi (2020) included four studies with 529 participants to assess the accuracy of total entorhinal cortex volume measured by MRI to detect conversion from MCI to ADD compared to other outcomes. Of the 529 participants, 229 were diagnosed with ADD. Individual study estimates of sensitivity were between 50% and 88% while the specificities were between 60% and 100%. Results are shown in Figure 4.

1.4.2 Positive predictive value

Lombardi (2020) did not report on the outcome positive predictive value.

1.4.3 Negative predictive value

Lombardi (2020) did not report on the outcome negative predictive value.

1.4.4 Positive likelihood ratio

Lombardi (2020) did not report a positive likelihood ratio. Meta-analyses were not conducted due to sparse and heterogeneous data

1.4.5 Negative likelihood ratio

Lombardi (2020) did not report a negative likelihood ratio. Meta-analyses were not conducted due to sparse and heterogeneous data.

Figure 4. Forest plot of total entrorhinal cortex volume measured by structural MRI for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Lombardi 2020)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

1.5 Whole brain volume

1.5.1 Specificity and sensitivity

The systematic review and meta-analysis of Lombardi (2020) included four studies with 424 participants to assess the accuracy of whole brain volume measured by MRI to detect conversion from MCI to ADD compared to other outcomes. Of the 424 participants, 220 were diagnosed with ADD. Individual study estimates of sensitivity were between 33% and 92% while the specificities were between 41% and 100%. Results are shown in Figure 5.

1.5.2 Positive predictive value

Lombardi (2020) did not report on the outcome positive predictive value.

1.5.3 Negative predictive value

Lombardi (2020) did not report on the outcome negative predictive value.

1.5.4 Positive likelihood ratio

Lombardi (2020) did not report a positive likelihood ratio. Meta-analyses were not conducted due to sparse and heterogeneous data

1.5.5 Negative likelihood ratio

Lombardi (2020) did not report a negative likelihood ratio. Meta-analyses were not conducted due to sparse and heterogeneous data.

Figure 5. Forest plot of whole brain volume measured by structural MRI for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Lombardi 2020)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

Index 2 Cerebrospinal fluid (CSF)

2.1 Amyloidβ42

2.1.1 Specificity and sensitivity

The systematic review and meta-analysis of Ritchie (2014) included fourteen studies with 1349 participants to assess the accuracy of CSF amyloidß42 to detect conversion from MCI to ADD compared to other outcomes. Of the 1349 participants, 436 were diagnosed with ADD. Individual study estimates of sensitivity were between 36% and 100% while the specificities were between 29% and 91%. Results are shown in Figure 6.

2.1.2 Positive predictive value

Ritchie (2014) did not report on the outcome positive predictive value.

2.1.3 Negative predictive value

Ritchie (2014) did not report on the outcome negative predictive value.

Figure 6. Forest plot of CSF amyloidß42 for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Ritchie 2014)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative}

2.2. Total-tau (CSF t-tau)

2.2.1 Specificity and sensitivity

The systematic review and meta-analysis of Ritchie (2017) included seven studies with 709 participants to assess the accuracy of CSF t-tau to detect conversion from MCI to ADD compared to other outcomes. Of the 709 participants, 291 were diagnosed with ADD. Individual study estimates of sensitivity were between 51% and 95% while the specificities were between 48% and 88%. Results are shown in Figure 7.

2.2.2 Positive predictive value

Ritchie (2017) did not report on the outcome positive predictive value.

2.2.3 Negative predictive value

Ritchie (2017) did not report on the outcome negative predictive value.

Figure 7. Forest plot of CSF t-tau for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Ritchie 2017)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative; CI = confidence interval}

2.3. Phosphorylated tau (CSF p-tau)

2.3.1 Specificity and sensitivity

The systematic review and meta-analysis of Ritchie (2017) included six studies with 492 participants to assess the accuracy of CSF p-tau to detect conversion from MCI to ADD compared to other outcomes. Of the 492 participants, 164 were diagnosed with ADD. Individual study estimates of sensitivity were between 40% and 100% while the specificities were between 22% and 86%. Results are shown in Figure 8.

2.3.2 Positive predictive value

Ritchie (2017) did not report on the outcome positive predictive value.

2.3.3 Negative predictive value

Ritchie (2017) did not report on the outcome negative predictive value.

Figure 8. Forest plot of CSF p-tau for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Ritchie, 2017)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative; CI = confidence interval}

2.4. CSF t-tau/Aß ratio

The systematic review and meta-analysis of Ritchie (2017) included two studies with 251 participants to assess the accuracy of CSF t-tau/Aß to detect conversion from MCI to ADD compared to other outcomes. Of the 251 participants, 102 were diagnosed with ADD. One study (n=37) reported a sensitivity of 91% and a specificity of 50% based on TP = 10, FP = 13, FN = 1, TN = 13. The other study (n=214) reported a sensitivity of 96% and a specificity of 51% based on TP = 87, FP = 60, FN = 4, TN = 63.

2.5. CSF p-tau/Aß ratio

2.5.1 Specificity and sensitivity

The systematic review and meta-analysis of Ritchie (2017) included five studies with 433 participants to assess the accuracy of CSF p-tau/Aß to detect conversion from MCI to ADD compared to other outcomes. Of the 433 participants, 140 were diagnosed with ADD. Individual study estimates of sensitivity were between 80% and 96% while the specificities were between 33% and 95%. Results are shown in Figure 9.

2.5.2 Positive predictive value

Ritchie (2017) did not report on the outcome positive predictive value.

2.5.3 Negative predictive value

Ritchie (2017) did not report on the outcome negative predictive value.

Figure 9. Forest plot of CSF p-tau/ Aß ratio for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Ritchie 2017)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative; CI = confidence interval}

2.6. Neurofilament light

None of the included studies (Ritchie 2014, Ritchie 2017) reported on the diagnostic accuracy (specificity, sensitivity, positive predictive value or negative predictive value) of neurofilament light.

Index 3 Positron Emitted Tomography (PET)

3.1 Amyloid 11C-PIB-PET

3.1.1 Specificity and sensitivity

The systematic review and meta-analysis of Zhang (2014) included nine studies with 274 participants to assess the accuracy of 11C-PIB-PET to detect conversion from MCI to ADD compared. Of the 274 participants, 112 were diagnosed with Alzheimer’s dementia. Individual study estimates of sensitivity were between 83% and 100% while the specificities were between 46% and 88%. Results are shown in Figure 10.

3.1.2 Positive predictive value

Zhang (2014) did not report on the outcome positive predictive value.

3.1.3 Negative predictive value

Zhang (2014) did not report on the outcome negative predictive value.

3.1.4 Positive likelihood ratio

Zhang (2014) did not report on the outcome positive likelihood ratio.

3.1.5 Negative likelihood ratio

Zhang (2014) did not report on the outcome negative likelihood ratio.

Figure 10. Forest plot of 11C-PIB-PET for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Zhang 2014)

_{Plot shows study-specific estimates of sensitivity and specificity (squares) with 95% confidence interval (black line) and study. TP: true positive; FP: false positive; FN: false negative; TN: true negative; CI = confidence interval}

3.2 Amyloid 18F-florbetapir

3.2.1 Specificity and sensitivity

The systematic review and meta-analysis of Martinez (2017) included two studies with 448 participants to assess the accuracy of 18F-florbetapir PET to detect conversion from MCI to ADD compared. Of the 448 participants, 69 were diagnosed with Alzheimer’s dementia. One study (n = 401) examined follow-up between one to less than two years and reported a sensitivity of 89% (95% CI 78 to 95) and a specificity of 58% (95% CI 53 to 64) by visual assessment, and a sensitivity of 87% (95% CI 76 to 94) and a specificity of 51% (95% CI 45 to 56) by quantitative assessment by the standardised uptake value ratio (SUVR). The other study (n=47) examined follow-up between two years to less than four years and reported a sensitivity of 67% (95% CI 30 to 93) and a specificity of 71% (95% CI 54 to 85) by visual assessment.

The systematic review of Cotta Ramusino (2024) included 8 studies with 1806 participants to assess the accuracy of 8F-florbetapir PET to detect conversion from MCI to ADD compared to other outcomes. Of the 1806 participants, 549 (30%) developed Alzheimer’s dementia. Individual study estimates of sensitivity were between 0,64 and 0,94 while the specificities were between 0,48 and 0,93.

Results are shown in table 4.

3.2.2 Positive predictive value

Martinez (2017) and Cotta Ramusino (2024) did not report on the outcome positive predictive value.

3.2.3 Negative predictive value

Martinez (2017) and Cotta Ramusino (2024) did not report on the outcome negative predictive value.

3.2.4 Positive likelihood ratio

Martinez (2017) and Cotta Ramusino (2024) did not report on the outcome positive likelihood ratio.

3.2.5 Negative likelihood ratio

Martinez (2017) and Cotta Ramusino (2024) did not report on the outcome negative likelihood ratio.

Table 4. Sensitivity and specificity of 18F-florbetapir PET for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Martinez 2017, Cotta Ramusino 2024)

_{* Some studies used multiple methods to assess the PET scan and thus establish the diagnosis. When applicable this is indicated by giving the name of the method used}

3.3 Glucose metabolism ([18F]FDG-PET)

3.3.1 Specificity and sensitivity

The systematic review and meta-analysis of Smailagic (2015) included 14 studies with 421 participants to assess the accuracy of [18F]FDG-PET to detect conversion from MCI to ADD compared to other outcomes. Of the 421 participants, 150 were diagnosed with ADD. Individual study estimates of sensitivity were between 0,25 and 1,00 while the specificities were between 0,15 and 1,00.

The systematic review of Cotta Ramusino (2024) included 25 studies with 6803 participants to assess the accuracy of 18F-FDG PET to detect conversion from MCI to ADD compared to other outcomes. Of the 6803 participants, 2572 (38%) developed Alzheimer’s dementia. Individual study estimates of sensitivity were between 0,25 and 1,00 while the specificities were between 0,15 and 1,00.

Results are shown in table 5.

3.3.2 Positive predictive value

Neither Smailagic (2015) or Cotta Ramusino (2024) reported on the outcome positive predictive value.

3.3.3 Negative predictive value

Neither Smailagic (2015) or Cotta Ramusino (2024) reported on the outcome negative predictive value.

Table 5.  Sensitivity and specificity of [18F]FDG-PET for early diagnosis of dementia due to Alzheimer's disease in people with mild cognitive impairment (Smailagic 2015, Cotta Ramusino 2024)

_{* Some studies used multiple methods to assess the PET scan and thus establish the diagnosis. When applicable this is indicated by giving the name of the method used}

The ideal body of evidence would include studies that directly compare the test strategies under consideration (i.e., randomized trials) and the resulting interventions and consequences for patiënts (i.e., patiënt-important outcomes). Such studies would, by design, address all of the issues in the analytical framework and allow guideline panelists to apply the familiar GRADE approach for interventions’ (Schünemann, 2019). If this direct evidence does not exist, a search question on test accuracy outcomes is formulated. Connecting test accuracy to downstream consequences is required for decision-making.

A systematic review of the literature was performed to answer the following question(s):

1. What is the value of MRI in predicting progression from MCI to dementia?
2. What is the value of CSF testing in predicting progression from MCI to dementia?
3. What is the value of PET in predicting progression from MCI to dementia?

Table PICRO

Relevant outcome measures

The guideline panel considered specificity as critical outcome measure for decision making; and sensitivity, positive predictive value, negative predictive value as important outcome measures for decision making. From a clinical perspective high specificity is more important than high sensitivity. In other words, avoiding false positive diagnoses is more important than avoiding false negative diagnoses, due to the absence of available disease-modifying treatments (see table 1).

Table 1. Consequences of diagnostic test characteristics

A priori, the guideline panel did not define the outcome measures listed above but used the definitions used in the studies.

The guideline panel deliberately did not define minimal clinically (patiënt) important differences. Rather, evaluated progression to a clinical dementia syndrome. From a clinical perspective, progression to dementia is most clear and unambiguous. Using a measure of cognitive decline may still result in uncertainty, because the magnitude of relevant cognitive decline is uncertain and may be different for each person, and cognitive test results tend to fluctuate over time in persons with MCI, which may lead to improvement after an initial deterioration.

Search and select (Methods)

The databases Medline (via OVID) and Embase (via Embase.com) were searched with relevant search terms (biomarkers (MRI/PET/CSF) for mild cognitive impairment) from 2010 until 20 August 2024. The detailed search strategy is listed under the tab ‘Literature search strategy’. The systematic literature search resulted in 595 hits. Studies were selected based on the following criteria:

• Systematic reviews (searched in at least two databases, detailed search strategy with search date, in- and exclusion criteria, exclusion table, risk of bias assessment and results of individual studies available).
• Full-text English language publication and
• Studies according to the PICROTS.

Because of the large body of literature and the knowledge on beforehand that several high-quality systematic reviews exist, the guideline committee decided to initially search for systematic reviews only, with the option to extend the search to individual studies if the results from the systematic reviews would be deemed insufficient.

Initially, 22 studies were selected based on title and abstract screening. After reading the full text, 15 studies were excluded (see the exclusion table under the tab ‘Evidence tabellen’) and seven studies were included.