Selection of trials

On august 14th 2023, we searched the literature for published phase 3 trials of monoclonal antibodies against β-amyloid that met their primary outcome for reducing cognitive and functional decline. Trials included were EMERGE (the high-dose arm of aducanumab), CLARITY-AD (lecanemab), and TRAILBLAZER-ALZ2 (donanemab). Details of included trials are presented in Supplementary Table 1. Trial inclusion and exclusion criteria were obtained from ClinicalTrials.gov and the original trial publications [3,4,5]. In case of discrepancy between ClinicalTrials.gov and the published report, we selected criteria from the latter.

Study population to assess generalisability of trial findings

The Rotterdam Study is an ongoing population-based cohort study investigating determinants and occurrence of disease in persons aged 40 years and older. The study started in 1990 and now comprises 17,931 individuals living in the Ommoord suburb of Rotterdam, the Netherlands. The design of the Rotterdam Study has been described in detail previously [18]. In brief, participants are invited for interview and extensive in-person examination at a dedicated research centre every 3–6 years. Routine examinations as part of the Rotterdam Study protocol include standardised questionnaires (e.g., social support, mental health, medication use), cognitive assessment, cardiovascular exams (e.g., blood pressure measurement, electrocardiogram and cardiac ultrasound), psychiatric assessment, venepuncture, and genotyping. From 2005 onward, brain MRI became part of the core study protocol of the Rotterdam Study. The fourth examination cycle of the study (2002–2005) included 6,052 participants and the fifth cycle (2009–2013) included 7,162 participants (Supplementary Fig. 1). For the current study, we included all 968 participants who were diagnosed with MCI (n = 779) or early dementia due to clinical manifestation of AD (n = 189)(Supplementary Fig. 1). To ensure that the study population aligned with the severity criteria of MCI or dementia used in clinical trials, we applied the MMSE-based inclusion criteria for each drug. Based on these criteria 751 participants met the inclusion criteria for aducanumab (MMSE score ≥ 24), 832 for lecanemab (MMSE score ≥ 22), and 762 for donanemab (MMSE score 22–28)(Supplementary Fig. 1). Participants with dementia had to have a dementia diagnosis within 1 year from their study centre visit. From 2005 onwards MRI of the brain was implemented into the core protocol of the Rotterdam Study. In total, 255/454 participants with MCI or early dementia due to AD from the fifth examination round had brain MRI scan available. Based on the MMSE criteria for each trial, 221 participants with brain MRI were included for aducanumab, 239 for lecanemab, and 209 for donanemab.

The Rotterdam Study has been approved by the Medical Ethics Committee of the Erasmus Medical Centre and by the Ministry of Health, Welfare and Sport of the Netherlands, implementing the Population Studies Act for the Rotterdam Study. All participants provided written informed consent for participation in the study, including permission to obtain information from their treating physicians.

Ascertainment of mild cognitive impairment and dementia

Participants were screened for dementia at baseline and every 3–6 years during follow-up examinations using the Mini-Mental State Examination (MMSE) and the Geriatric Mental Schedule (GMS) organic level. Those with a MMSE score of < 26 or a GMS organic level score of > 0 were further examined using the Cambridge Examination for Mental Disorders of the Elderly. Additionally, participants were continuously under surveillance for incident dementia through the electronic linkage between the study database and medical records from general practitioners and the Regional Institute of Outpatients Mental Health Care. The general practitioner functions as a gatekeeper within the Dutch healthcare system, receiving written information of any medical specialists’ consultations of their patients. The final diagnosis of dementia and its most common subtypes was made by a consensus panel led by a neurologist based on the standard criteria for all-cause dementia (DSM-III-R) and clinical Alzheimer’s disease (NINCDS-ADRDA). Follow-up for dementia was completed until January 1, 2020.

Criteria for MCI were based on the Petersen criteria, and included the presence of subjective cognitive complaints and objective cognitive impairment, in the absence of dementia [19]. A detailed description of our MCI assessment has been published previously [20]. Subjective cognitive complaints were evaluated by interview, including questions on memory and everyday functioning. We assessed objective cognitive impairment using a cognitive test battery comprising the letter-digit substitution task, Stroop test, verbal fluency test (animal categories), and 15-word verbal learning test based on Rey’s recall of words [21]. Compound scores were constructed for various cognitive domains, including memory function, information-processing speed, and executive function [21, 22]. We classified participants from the Rotterdam Study as objectively cognitively impaired if in any of the cognitive domains, the compound score was at least 1.5 standard deviation lower than expected based on age and education adjusted means, derived from the study population.

Operationalization of trial eligibility criteria

For each of the trial inclusion and exclusion criteria, we defined an operationalization in the Rotterdam Study. This was done through consensus discussion between two authors (JJC and FJW). A detailed description of the eligibility criteria for each of the three trials, and their operationalization, is provided in Supplementary Tables 2, 3 and 4. For example, history of cardiovascular disease was based on a combination in-person assessment and continuous linkage with medical records for various comorbidities, use of medication was assessed based on pharmacy dispensary data (ATC-codes), and we had in-person questionnaires for social support and screening for anxiety and depression.

Brain magnetic resonance imaging

Participants underwent scanning on a 1.5-T MRI scanner (GE Healthcare) using a multisequence protocol consisting of T1-weighted, proton density-weighted, fluid-attenuated inversion recovery, and T2-weighted sequences. For brain volumetry, T1-weighted (voxel size 0.49 × 0.49 × 1.6 mm3), proton density–weighted (voxel size 0.6 × 0.98 × 1.6 mm3), and the fluid-attenuated inversion recovery (FLAIR) (voxel size 0.78 × 1.12 × 2.5 mm3) scans were used for automated segmentation of brain tissues, including white matter hyperintensities [23, 24]. In short, a k-nearest neighbor tissue classification algorithm was implemented for quantification of white matter hyperintensities (WMH) [24]. All segmentations were visually inspected, and manually corrected if needed. Severe WMH were defined as a volume of 16.1 milliliter or more, corresponding to a Fazekas score of 3 [25, 26]. All scans were appraised by trained research physicians for the presence of cerebral microbleeds (i.e., small round to ovoid hypointense areas on T2*-weighted images), lacunes (i.e., focal cavitating lesions ≥ 3 and < 15 mm), and cortical infarcts [27]. These ratings were done blinded to clinical data. Furthermore, abnormalities including superficial siderosis, intracranial aneurysms, meningioma, suspected gliomas, cavernous angioma, arachnoid cysts, arteriovenous malformations and dural fistulas were rated by research physicians, and subsequently evaluated by a consultant neuroradiologist.

Prediction of amyloid status

To assess amyloid positivity, we used a prediction model based on age and APOE ε4 allele count [28]. This model was developed within the Anti-Amyloid Treatment in Asymptomatic Alzheimer’s trial population (A4 Study), and previously showed high discriminative ability to differentiate people with and without brain amyloid as measured by 18F-florbetaben PET in the Rotterdam Study population (area under the curve: 0.84 [0.79–0.88]) [28].

Statistical analyses

Missing data, most notably for social support (17.9%), Parkinson’s disease (due to incomplete assessment of the Unified Parkinson Disease Rating Scale, 7.0%) and anxiety disorder (9.1%), were imputed using 5-fold multiple imputations (“mice” package in R). Details of all missing and imputed variables are described in supplementary Table 7.

For each trial, we determined the number of participants who met the eligibility criteria, both overall and per criterion separately. We compared the subset of trial eligible individuals to all individuals with MCI and early AD dementia in the population on several key demographics and participant characteristics, including, age, sex, educational attainment, MMSE, APOE ε4 carrier status, MCI or dementia status, cardiovascular disease and antithrombotic medication use. We repeated all analyses in the subsample (n = 255) with brain MRI.

Finally, among all individuals with MCI in the primary analyses, we determined 5-year progression to dementia stratified by trial eligibility, by computing hazard ratios (HR) using Cox proportional hazard models.

All analyses were performed in R (version 4.2.1). This study is reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [29].

Role of funding sources

The funding sources that contributed to the Rotterdam Study or to authors had no role in study design, data collection, data analysis, data interpretation, writing of the report, and in the decision to submit the paper for publication.