Types of studies

We included longitudinal studies that had prospectively defined cohorts with any accepted definition of mild cognitive impairment (MCI), as outlined below, at time of performing the ¹⁸F‐florbetapir Aβ scan and a reference standard (see Index tests and Reference standards below). We obtained the results at the follow‐up of the studies. These studies had to employ delayed verification of progression to dementia and were sometimes labelled as 'delayed verification cross‐sectional studies' (Bossuyt 2008; Knottnerus 2002). We included case‐control studies when they incorporated a delayed verification design. This occurred in the context of a cohort study, so these studies were invariably diagnostic‐nested case‐control studies.

Participants

Participants recruited and clinically classified as having MCI at time of performing the test were eligible for inclusion. We established the diagnosis of MCI using the Petersen criteria or revised Petersen criteria (Petersen 1999; Petersen 2004; Winblad 2004), the criteria included in the Matthews study (Matthews 2008), CDR = 0.5 (CDR structured interviews collects information from both the collateral source and the subject regarding memory, orientation, judgment and problem solving, community affairs, home and hobbies, and personal care, where the range of possible scores varies from none = 0 point to severe = 3 points) (Morris 1993), the National Institute on Aging‐Alzheimer's Association (NIA‐AA) core clinical criteria (Albert 2011), or a combination.

We excluded studies that included people with MCI possibly caused by any of the following:

• Current or a history of alcohol or drug abuse.

• Central nervous system (CNS) trauma (e.g. subdural hematoma), tumour, or infection.

• Other neurological conditions (e.g. Parkinson’s or Huntington’s diseases). Regarding Parkinson's disease, many of the studies specifically excluded Parkison's disease patients from the group with mild cognitive impairment. This specific group of patients is complex in both regards to defining neuropathology and in determination of functional decline. For these reasons, this group of patients needs to be addressed in specific studies.

Index tests

The index test of this SR was the ¹⁸F‐florbetapir biomarker test. We used the criteria and cut‐off values for test positivity as reported in the included studies. We considered positivity for ¹⁸F‐florbetapir Aβ scan uptake and retention exceeding a certain threshold.

Target conditions

There were three target conditions in this SR:

• Alzheimer’s disease dementia (ADD) (progression from MCI to ADD).

• Any other forms of dementia (progression from MCI to any other forms of non‐ADD).

• Any form of dementia (progression from MCI to any form of dementia).

Reference standards

The reference standard was the progression to the target conditions evaluated by a physician with expertise in the dementia field (preferably, a geriatrician, psychiatrist, or neurologist). For the purpose of this SR, we accepted several definitions of ADD. We included studies that applied the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS‐ADRDA) criteria (McKhann 1984), the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria (APA 1987; APA 1994), and the International Classification of Diseases (ICD) (ICD‐10) criteria for ADD. Notably, different iterations of these standards may not be directly comparable over time (e.g. APA 1987 versus APA 1994). Moreover, the validity of the diagnoses may vary with the degree or manner in which the criteria have been operationalised (e.g. individual clinician versus algorithm versus consensus determination). We considered all these issues when we interpreted the results.

Similarly, we accepted differing clinical definitions of other dementias. For Lewy body dementia, the reference standard is the McKeith criteria (McKeith 1996; McKeith 2005); for frontotemporal dementia, the Lund criteria (Boxer 2005; Brun 1994; Neary 1998), the DSM criteria (APA 1987; APA 1994), the ICD criteria (ICD‐10), or the International Behavioural Variant FTD Criteria Consortium (Rascovsky 2011); and for vascular dementia, the National Institute of Neurological Disorders and Stroke and Association Internationale pour la Recherché et l'Enseignement en Neurosciences (NINDS‐AIREN) criteria (Román 1993), the DSM criteria (APA 1987; APA 1994), or the ICD criteria (ICD‐10).

The time interval over which progression from MCI to ADD (or other forms of dementia) occurs is very important. We used one year as the minimum period of delay in the verification of the diagnosis (the time between the assessment at which a diagnosis of MCI is made and the assessment at which the diagnosis of dementia is made).

Electronic searches

We searched MEDLINE (Ovid SP) from 1946 to May 2017; Embase (Ovid SP) from 1974 to May 2017; PsycINFO (Ovid SP) from 1806 to May 2017; BIOSIS Citation Index (Thomson Reuters Web of Science) from 1922 to May 2017; Web of Science Core Collection, including the Science Citation Index (Thomson Reuters Web of Science) and the Conference Proceedings Citation Index (Thomson Reuters Web of Science) from 1946 to May 2017; LILACS (Bireme); CINAHL (EBSCOhost) from 1980 to May 2017; ClinicalTrials.gov (https://clinicaltrials.gov); and the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (http://www.who.int/ictrp/search/en/). We also searched ALOIS, the Cochrane Dementia & Cognitive Improvement Group’s specialized register of dementia studies (http://www.medicine.ox.ac.uk/alois/).

We used two approaches in designing the search. One focused solely on the specifically named index test (including a range of synonyms); the second, run in parallel, covered a more general search, linking broader terms for the index test, focused by terms describing its diagnostic use to terms for the target condition to try to capture the more difficult to locate studies of a more general nature, where these particular radioligands were included in diagnostic accuracy research but not named specifically in the parts of the electronic bibliographic record that are searchable and therefore would be missed.

See Appendix 2 for details of the sources and search strategies that we used. No language or date restrictions were applied to the electronic searches.

Searching other resources

We examined the reference lists of all relevant studies for additional studies. We also searched the Database of Abstracts of Reviews of Effects (DARE) via the Cochrane Library: www.cochranelibrary.com), the National Institute for Health Research ‐ Health Technology Assessment Database (NIHR‐HTA) (via the Cochrane Library: www.cochranelibrary.com), the Aggressive Research Intelligence Facility (ARIF) database (www.arif.bham.ac.uk) for other related systematic diagnostic accuracy reviews, and the International Federation of Clinical Chemistry and Laboratory Medicine Committee for Evidence‐based Laboratory Medicine database (C‐EBLM) (http://www.ifcc.org/ifcc‐education‐division/emd‐committees/c‐eblm/evidence‐based‐laboratory‐medicine‐c‐eblm‐base).

We checked the reference lists of any relevant studies and systematic reviews, and performed citation tracking using the Science Citation Index to identify any additional relevant studies.

Selection of studies

Two review authors (GM, RV) independently screened the retrieved titles and abstracts for potentially eligible studies. A third review author (PF) resolved any disagreements between the two review authors. The two review authors (GM, RV) then independently assessed the full‐text articles of the selected studies with the inclusion criteria. They resolved any disagreements through discussion or, where necessary, consulted a third review author (PF) who acted as an arbitrator. When a study did not present all relevant data for creating a 2 × 2 table, we contacted the study authors directly to request further information. When more than one article presented data on the same population, we included the primary article, which was the article with the largest number of people or with the most informative data (e.g. longest time of follow‐up in the primary outcome).

Data extraction and management

We planned to extract the following data regarding the study characteristics.

• Bibliographic details of primary paper:

  • author, title of study, year, and journal.

• Basic clinical and demographic details:

  • number of participants;

  • clinical diagnosis;

  • MCI clinical criteria;

  • age;

  • gender;

  • sources of referral;

  • participant recruitment;

  • sampling procedures.

• Details of the index test:

  • method of the ¹⁸F‐florbetapir test administration, including those who administered the test;

  • thresholds used to define positive and negative tests;

  • other technical aspects as seemed relevant to the review, e.g. brain areas.

• Details of the reference standard:

  • definition of ADD and other dementias used in the reference standard;

  • duration of follow‐up from time of index test performed to defining ADD and other dementias by the reference standard: one year to less than two years; two years to less than four years; and more than four years. If participants had been followed for varied amounts of time, we recorded a mean follow‐up period for each included study. If possible, we grouped those data into minimum, maximum, and median follow‐up periods, which could then become the subject of subgroup analyses;

  • prevalence or proportion of population developing ADD and other dementias, with severity, if described.

We created 2 × 2 tables (cross‐relating index test results of the reference standards) as shown in Appendix 3. For each included study, we recorded the number of people lost to follow‐up. We also extracted data necessary for the assessment of quality, as defined below. Two review authors (GM, RV) independently performed data extraction. We resolved any disagreements regarding data extraction by discussion, or by consulting a third review author (PF).

Assessment of methodological quality

We assessed the methodological quality of each included study using the Quality Assessment of Diagnostic Accuracy Studies 2 tool (QUADAS‐2) (Whiting 2011), as recommended by Cochrane (Davis 2013). This tool is comprised of four domains: participant selection, index test, reference standard, and participant flow.

Two review authors (GM, RV), who were blinded to each other’s scores, independently performed the QUADAS‐2 assessment. We resolved any disagreements by discussion or, if necessary, consulted a third review author (PF) who acted as an arbitrator. We assessed each domain in terms of risk of bias, and also considered the first three domains in terms of applicability concerns. In Appendix 4, we have detailed the components of each of these domains and provided a rubric that shows how we made judgements concerning risk of bias. Key areas important to quality assessment are participant selection, blinding, and missing data.

We included three additional signalling questions on our checklist.

• Was the PET scan interpretation done by a trained reader physician? (We included this under the ’Index test’ domain).

• Was there a clear definition of a positive result? (We included this under the ’Index test’ domain).

• Was the study free of commercial funding? (We included this under the ’flow and timing’ domain).

We included the item pertaining to the PET scan interpretation and the definition of positive results to take into account the subjective nature of ¹⁸F‐florbetapir Aβ scan image interpretation, which may be based on a variety of different criteria, such as extensive clinical experience, different standardised uptake values (SUV), different morphological features, or a combination of the aforementioned. We included the third additional item in order to record any potential bias resulting from commercial interest in the results due to the potential risk by the manufacturing company leading to more favourable results and conclusions than sponsorship by other sources (Lundh 2017).

We did not use QUADAS‐2 data to form a summary quality score. We produced a narrative summary that described the numbers of included studies that were at high, low, or unclear risk of bias as well as concerns regarding applicability, which we have described in Appendix 5.

Statistical analysis and data synthesis

We applied the DTA framework for the analysis of a single test and extracted the data from each included study into a 2 × 2 table, showing the binary test results cross‐classified with the binary reference standard, and we ignored any censoring that might have occurred. We acknowledge that such a reduction in the data may represent a significant oversimplification.

We used data from the 2 × 2 tables abstracted from the included studies: true positive (TP), false negative (FN), false positive (FP), true negative (TN), and entered these into Review Manager 5 (RevMan 5) (Review Manager 2014) to calculate the sensitivities, specificities, and their 95% confidence intervals. We also presented individual study results graphically by plotting estimates of sensitivities and specificities in both a forest plot and a receiver operating characteristic (ROC) space. If an individual included study published more than one threshold, we presented the graphical findings for all reported thresholds.

We planned to segment analyses into separate follow‐up mean periods for the delay in verification: one year to less than two years; two to less than four years; and greater than four years. We planned to clearly note where the same included studies contributed to the analysis for more than one reference standard follow‐up interval.

However, due to lack of data, we conducted no meta‐analyses, but we prepared a 'Summary of findings' table regardless.

Investigations of heterogeneity

We were able to include only three studies, therefore issues of heterogeneity did not arise.

Sensitivity analyses

We found insufficient data to conduct any sensitivity analyses.

Assessment of reporting bias

We did not investigate reporting bias.