In this study, data were used from a longitudinal cohort of consecutive patients referred by a tertiary memory clinic (Centre Mémoire de Ressources et de Recherche (CMRR)) for a brain 18F-FDG PET scan at the Centre Hospitalier Régional Universitaire (CHRU; Regional University Hospital) of Nancy to determine whether evidence of a neurodegenerative pathology was present. Patients who obtained scans between January 1, 2010, and January 1, 2019, were retrospectively included in this study (ClinicalTrials.gov number NCT04804722). The prescription of a brain 18F-FDG PET scan by the tertiary memory clinic followed the national French recommendations for the diagnosis of AD and related disorders in force at this time period [10].

All included patients presented with cognitive complaints. A consultation with > 5 years of experience neurologist/geriatrician and neuropsychological assessment at the tertiary memory clinic, as well as brain morphological imaging (MRI or tomodensitometry), were performed before the brain 18F-FDG PET scan. Patients were followed longitudinally for at least three years until January 1, 2022.

This study was granted approval by the National Comité Ethique et Scientifique pour les Recherches, les Etudes et les Evaluation en Santé (CESREES, file no. 4,611,320 Bis) and by the Commission Nationale Informatique et Libertés (CNIL, Decision number: DR-2022-090) and was conducted following the principles of the Declaration of Helsinki. This trial was also reported in the Clinical Trials database (NCT04804722). The STROBE statement [17] was used as a reporting guide for the present article.

Prior to 2018, brain 18F-FDG PET images were acquired on an analog system (Biograph 6, Siemens®); thereafter, a digital system (Vereos, Philips®) was used.

Following an injection of 4.5 MBq/kg (for the analog device) or 2 MBq/kg (for the digital device) of 18F-FDG, the patient underwent neurosensory rest for 30 min. Patients were scanned in a supine position with a single bed position for an acquisition time of 10 min with the analog system and 15 min with the digital system. All patients were instructed to fast for at least six hours before the injection and had a blood glucose level < 10 mmol/L.

Image reconstruction was performed using the Ordered Subset Expectation Maximization (OSEM) iterative reconstruction algorithm. The reconstruction was conducted over two iterations with 21 subsets, a 256 × 256 matrix, 2.7 × 2.7 × 2.7 mm3 voxel spacing and postfiltering with a 4-mm Gaussian filter for the analog PET device [18] and over two iterations with 10 subsets, a 256 × 256 matrix, 1 × 1 × 1 mm3 voxel spacing and point spread function (PSF) correction for the digital PET device [19]. Corrections for attenuation, random coincidences and dispersion were applied for the images obtained with both devices.

A combined visual and semiquantitative analysis of the brain 18F-FDG PET images was performed by two experienced physicians (S.H. and A.V.) who were blinded to all clinical data, with final consensus in case of discordances. This combined analysis was performed according to the latest guidelines of the European Association of Nuclear Medicine [9]. First, the brain 18F-FDG PET scans were reviewed visually based on the typical regions of hypometabolism associated with neurodegenerative pathologies accurately defined in the latest European Association of Nuclear Medicine guidelines [9]. The semiquantitative analysis of the brain 18F-FDG PET images was performed by an experienced engineer (M.D.) using the Statistical Parametric Mapping software, SPM 12, run on MATLAB 2020b (MathWorks, Inc., Sherborn, MA). Brain 18F-FDG PET images were spatially normalized using a dementia-specific 18F-FDG-PET template [20] and intensity-normalized with proportional scaling. All images were subsequently smoothed with an 8 mm full-width at half-maximum (FWHM) Gaussian filter. Statistical analyses of this semi-quantitative analysis were performed individually at the voxel-to-voxel level according to an optimized procedure validated by Perani et al. [21] via comparison to a dataset of healthy participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort (n = 75; age (mean ± SD) 72.25 ± 5.25; 42 women). Age was treated as a continuous variable and used as a covariate. A Z score map of individual hypometabolism was calculated for each scan via the two-sample Student’s t test. A probability threshold of p < 0.05 was considered to indicate statistical significance, and the extent threshold (k extent) was set at 100 voxels. An illustrative example of this process is available in Fig. 1.

Example of the combined visual (upper panel, axial and coronal slices of brain 18F-FDG PET images) and semiquantitative (lower panel, with hypometabolisms in green projected on axial MR slices on the left and hypometabolisms in blue on a 3D volume rendered on the right) analysis for a 65-year-old male patient (initially diagnosed with MCI); these results were consistent with Alzheimer’s disease. The final diagnosis in the National Alzheimer’s database was Alzheimer’s disease and conversion to dementia after 31.5 months

In both the visual and semiquantitative analyses, the brain 18F-FDG PET scans were classified as normal, abnormal but incompatible with a neurodegenerative disease diagnosis, or abnormal and compatible with a neurodegenerative disease diagnosis. Brain 18F-FDG PET scans deemed compatible with a neurodegenerative pathology were further divided into two categories: those compatible with an AD pattern and those that were not according to the hypometabolic areas described in the pattern of AD in the latest European Nuclear Medicine Association guidelines [9].

The patient cohort was matched with three databases representing patient data obtained under real-life conditions: the NHDS, a database to which any general practitioner, whether practicing in or outside a hospital setting, can submit data and therefore includes the largest cohort among the three databases; the NAB, a more specific database to which only neurologists and geriatric specialists can submit data; and, finally, data from objective CSF AD biomarker tests. The French NHDS makes possible, since 2016, to link health insurance data, hospital data, and more recently the medical causes of death, disability-related data, and a sample of data from complementary health insurance organizations. The purpose of the NHDS is to make these data available in order to promote studies, research or evaluations of a nature in the public interest. The NAB is a French central information system of the memory clinic network that coordinates the management of patients with neurodegenerative disorders according to the 2009 national guidelines. This national database includes longitudinal follow-up data for patients who attended a tertiary memory clinic for cognitive complaints. These data include information obtained through initial and follow-up visits and syndromic and etiological diagnoses. Each participating center is required to provide information on patients seen for cognitive complaints through a computer file with limited space for data entry to facilitate and improve participation in the national database. The CSF AD biomarker tests were retrieved from medical records of the Nancy University Hospital. These CSF biomarkers allow to define AD biologically according to the A/T/N classification [22, 23].

The patient cohort was first matched with the patients in the NHDS to identify those who experienced a conversion to dementia within the three years after the brain 18F-FDG PET scan. As the NHDS does not collect identifiable patient information (i.e., surname, family name, health insurance identification number, date of birth), a deterministic indirect data linkage method was applied between data from the patient cohort and those from the anonymized NHDS based on common variables (sex, month and year of birth, identification number in the Regional University Hospital of Nancy, date of brain 18F-FDG PET scan, and date of lumbar puncture (LP), when available). This method could result in potential incorrect matches. However, incorrect matches were limited as we retained only patients with a 1:1 match, i.e., patients with multiple correspondence in one or the other database were excluded. Patients designated as having a “long-term condition” (LTC) of dementia at the time the brain 18F-FDG PET scan was performed were secondarily excluded.

Data for patients in the cohort who matched with those in the NHDS and presented without dementia at the time of brain 18F-FDG PET scanning were then matched with the clinicobiological data obtained in real-life conditions from the NAB and with the medical records of the Nancy University Hospital. These data could be directly linked by matching patient name, surname, sex, and date of birth.

The main outcome was conversion to dementia according to the data in the NHDS and NAB within three years of the brain 18F-FDG PET scan, verifying that this information was concordant between these two databases. The secondary outcomes were a diagnosis of a neurodegenerative disease according to the NAB data and medical records, death according to the NHDS data, and factors associated with a conversion to dementia and brain 18F-FDG PET results according to the NHDS and NAB data.

The criterion for dementia conversion in the NHDS was the designation of an LTC for dementia by any general practitioner within the three years after the brain 18F-FDG PET scan. Aside from age, sex and educational level, which were extracted from the brain 18F-FDG PET scan records when available and are known factors influencing cognitive decline, data collected from the NHDS including the designation of an LTC for dementia, and other factors influencing conversion to dementia such as hospitalization and death within the three years after the PET scan, and other LTC designations and medications potentially related to neurodegenerative pathologies.

The criterion for conversion to dementia in the NAB was the identification of a dementia stage of cognitive impairment within the three years after the brain 18F-FDG PET scan. The criterion for the diagnosis of a neurodegenerative disease was a final diagnosis at the last follow-up consultation in the NAB within the three years after the brain 18F-FDG PET scan. Data concerning the stage of cognitive impairment (initial and during follow-up) and syndromic (dysexecutive, amnesic, linguistic, and diffuse) and etiological diagnoses (grouped into five classes: neurodegenerative diseases, psychiatric pathology, vascular damage, other causes (i.e., epilepsy, encephalopathy/encephalitis, intracranial tumor, posttraumatic sequelae and other organic causes), and subjective memory complaints) at the last follow-up consultation were also collected from the NAB.

The criterion for the identification of a neurodegenerative disease in the medical records was the determination of LP biomarker levels according to the A/T/N classification [22, 23] according to laboratory reference values, i.e., A + if Aß42 < 700 pg/mL or Aß42/Aß40 < 0.06, T + if pTau > 60 pg/mL and N + if tTau > 350 pg/mL. All CSF biomarker assays were performed in the same laboratory at Nancy University Hospital using the manual Innotest ELISA technique with 10 mL polypropylene sampling tubes (Ref: TP 10 − 03 GOSSELIN (CML)).

Categorical variables are expressed as numbers and percentages, and continuous variables are expressed as means and standard deviations or medians and quartiles. According to the NHDS, the sensitivity (Se), specificity (Sp), accuracy (Acc), positive predictive value (PPV), and negative predictive value (NPV) of brain 18F-FDG PET scans classified as supporting or not supporting the diagnosis of a neurodegenerative disease were calculated for the risk of dementia conversion within three years. Because all-cause deaths might interfere with the identification of dementia conversion, we identified three groups of patients: patients who did not progress to dementia and were still alive during the three years following the PET scan, patients who progressed to dementia before dying or were still alive during the follow-up period, and patients who died without having been documented as converting to dementia.

We conducted bivariate and multivariable analyses with stepwise selection to identify predictive factors for dementia conversion within the three years following the PET scan using the Fine and Gray model for competing risk, with all-cause death as the competing risk for dementia conversion. We also performed a subanalysis to identify predictive factors only for patients whose brain 18F-FDG PET scans did not support the diagnosis of a neurodegenerative disease using bivariate and multivariable logistic regressions. Variables with a p value < 0.1 in bivariate analyses were selected as candidates for the multivariable model. Spearman correlation was assessed between candidate variables with a threshold of R > 0.75. Variables occurring fewer than five times were not included in the multivariable analyses. The diagnostic performance metrics for brain 18F-FDG PET scans for the risk of death within three years were also calculated. Kaplan‒Meier analysis with log rank tests was conducted to determine the prognostic value of brain 18F-FDG PET for dementia-free survival and overall survival.

The Se, Sp, Acc, PPV and NPV of brain 18F-FDG PET scans classified as supporting or not supporting the diagnosis of a neurodegenerative disease were also calculated for the risk of dementia conversion within three years with respect to the NAB and the medical records from the Nancy University Hospital. In addition, a bivariate logistic regression analysis was performed to assess the value of the brain 18F-FDG PET scan interpretation as well as age, sex and educational level for the risk of conversion to dementia within three years according to the NAB data. These PET diagnostic performance metrics were also calculated for the diagnosis of any neurodegenerative pathology, for the specific diagnosis of AD (according to the NAB data), and for detecting different combinations of CSF biomarkers (according to the medical records of Nancy University Hospital). The NAB data and medical records were also subjected to subanalyses involving PET scans demonstrating a pattern that favored or did not favor an AD diagnosis and only in the MCI patient group. Chi-square tests for the association of brain 18F-FDG PET scan interpretations with the initial stage of impairment and syndromic and etiological diagnoses were performed according to the NAB data.

Unless otherwise indicated, a p value < 0.05 was considered to indicate statistical significance. Missing data were not included in the analyses. All the statistical tests were performed with IBM SPSS 25.0 software and SAS Enterprise Guide version 8.2 (SAS Institute, Inc., Cary, N.C., USA).