Research diagnostic criteria for Alzheimer’s disease: findings from the LipiDiDiet randomized controlled trial

LipiDiDiet is a double-blind proof-of-concept RCT conducted at 11 sites in Finland, Sweden, Germany, and the Netherlands [17]. The 2-year core trial, completed in 2015, was followed by up to four 1-year double-blind extension studies. LipiDiDiet included 311 individuals aged 55–85 years, recruited mainly from memory clinics. For eligibility, participants were required to meet the IWG-1 criteria for prodromal AD [1], i.e., have an underlying AD pathology and mild episodic memory impairment, defined in LipiDiDiet as a performance below one standard deviation in at least two cognitive tests (at least one memory test). The following cognitive screening tests were used: Free and Cued Selective reminding test, free and delayed recall, Wechsler Memory Scale-revised (WMS-r) story, delayed recall, WMS-r delayed recall of figures, Trail Making Tests A and B, symbol digit substitution test, and category fluency test. Sufficient evidence for AD pathology was defined as abnormality in at least one of the following biomarkers: cerebrospinal fluid (CSF) Aβ (Aβ42 and/or Aβ42/40 ratio), CSF total tau (t-tau), CSF phosphorylated tau (p-tau), fluoro-deoxy-glucose-positron emission tomography (FDG-PET), or medial temporal lobe atrophy (MTA) on magnetic resonance imaging (MRI). Due to feasibility, MTA was often the primary biomarker used to assess eligibility in LipiDiDiet.

Exclusion criteria were dementia diagnosis or substantial cognitive impairment (Mini-Mental State Examination (MMSE) < 24 or < 20 if ≤ 6 years of education), cholinesterase inhibitor or memantine use, neuroimaging abnormalities (stroke, intracranial bleeding, mass lesion, normal pressure hydrocephalus), and conditions potentially interfering with participation (e.g., alcohol/drug abuse). Individuals taking omega-3 products or vitamin B6, B12, C, E, or folic acid (> 200% of the recommended daily intake) were also excluded due to the nature of the intervention.

LipiDiDiet is registered at Netherlands Trial Register (identifier NL1620). Ethical approval was granted by local ethics committees, and written informed consent was obtained from all participants and study partners prior to enrollment.

Participants were randomized in a 1:1 ratio to the intervention and control groups. The intervention group consumed once a day a 125-ml medical food drink (Souvenaid®) containing the multinutrient Fortasyn Connect™ (Nutricia; Zoetermeer, the Netherlands). Fortasyn Connect™ consists of a combination of nutrients which AD patients might be deficient in [18], including omega-3 fatty acids, vitamins, folic acid, phospholipids, and antioxidants. In short-term RCTs in mild AD, this multinutrient showed beneficial effects on cognitive performance and brain functional connectivity [19, 20]. The control group consumed once a day an iso-caloric placebo product similar in appearance, flavor, and composition, but without the multinutrient. The participants, clinical team, and outcome evaluators were blinded to group assignment. Main study parameters and outcomes were assessed at baseline, 6, 12, and 24 months. For motivation and compliance, additional visits with the study nurse or physician were organized at 3, 9, and 18 months, and phone calls were conducted monthly for the first 6 months and every 2 months after that. Protocol details are described elsewhere [17].

The primary trial outcome was change in cognitive performance measured with a Neuropsychological Test Battery (NTB) composite Z score including five tests: Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) 10-word list learning, delayed recall, and recognition, category fluency test, and letter digit substitution test. Secondary cognitive outcomes included the domain-specific Z scores for memory (three tests; CERAD 10-word list learning, delayed recall, and recognition) and executive functioning (four tests; category fluency test, letter digit substitution test, concept shifting test condition C, and WMS-r digit span), as well as the NTB total Z score. The total score was based on 16 tests (the above-mentioned tests and additionally WMS-r logical verbal memory, immediate and delayed recall, WMS-r visual paired associates, immediate and delayed recall, 30-item Boston Naming Test, CERAD constructional praxis, copy and recall, and concept shifting test conditions A and B). Test scores were calculated as standardized Z scores with higher scores demonstrating better performance. Cognition was assessed by study psychologists at baseline, 6, 12, and 24 months.

Other secondary outcomes, including the Clinical Dementia Rating-Sum of Boxes (CDR-SB) score reflecting global cognitive-functional performance, were assessed at baseline, 12, and 24 months. Dementia and AD were diagnosed using the Diagnostic and Statistical Manual of Mental Disorders 4th edition (DSM-IV) [21] and the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria [22].

At each study site, structural MRI was performed according to local scanning protocols. To assess MTA, coronal reconstructions of 3D T1-weighted scans were visually rated according to the Scheltens scale ranging from 0 (no atrophy) to 4 (severe atrophy). The right and left hemispheres were rated separately, and the sum was calculated. Scans were assessed locally to determine eligibility at screening. For the present study, results of the centralized analysis were used to assess biomarker profiles. Central analysis was conducted at the VU University Medical Center in Amsterdam, the Netherlands. MTA was defined as a score of ≥ 1.

In the present study, all participants (both the intervention and control group participants) with centrally analyzed baseline biomarkers (CSF and/or MTA; N = 287) were classified according to the A/T/N biomarker scheme [23] and the IWG-2 [5], NIA-AA 2011 [3], and NIA-AA 2018 [7] research diagnostic criteria (Table 1). Biomarkers considered in the classification were MTA (imaging marker for neuronal injury) and CSF biomarkers reflecting Aβ deposition (Aβ42, Aβ42/40 ratio) and neuronal injury (p-tau, t-tau). Aβ or tau positron emission tomography (PET) scans were not performed in LipiDiDiet.

Participants were considered to have prodromal AD (IWG-2) if they had abnormal Aβ and at least one abnormal CSF neuronal injury marker. High AD likelihood (NIA-AA 2011) was defined as abnormal Aβ and at least one (any) abnormal neuronal injury marker. Participants with conflicting biomarkers were classified either in the isolate amyloid pathology group (abnormal Aβ but normal neuronal injury markers) or in the suspected non-AD pathology (SNAP) group (normal Aβ but at least one abnormal neuronal injury marker) [24]. Participants with only one biomarker available (in LipiDiDiet MTA) were classified as having either an intermediate AD likelihood or an inconclusive/uninformative status, depending on whether MTA was present. According to the NIA-AA 2018 criteria, participants with abnormal Aβ and p-tau were considered to have AD (A+T+N± profile). Participants with abnormal Aβ and neuronal injury markers but normal p-tau were classified in the Alzheimer’s and concomitant suspected non-Alzheimer’s pathologic change group (A+T−N+). In case of conflicting biomarkers, participants were classified as having an Alzheimer’s (A+T−N−) or non-Alzheimer’s pathologic change (A−T+N±, A−T−N+).

Between-group differences in participant baseline characteristics were analyzed with t-tests and chi-square tests, as appropriate. Z scores for NTB composite, total, memory, and executive functioning were calculated as previously described [17]. We analyzed change from baseline in the NTB and CDR-SB scores using linear mixed models for repeated measures as previously reported, with baseline score, randomized treatment, time, treatment × time interaction, and baseline MMSE as fixed effects [17]. We additionally included the biomarker profile (research criteria) and criteria × time interaction, as well as sex and baseline age, as fixed effects in the present study. All trial participants regardless of randomized treatment were included in the analyses (intervention effects were accounted for by including randomized treatment and treatment × time interaction in the models). A random intercept with a variance components covariance structure was used within sites and a random intercept and slope for time with an unstructured covariance structure within subjects. Least-squares means for change from baseline were estimated from the linear mixed model. p-values are shown for the difference in least-squares means over 2 years between each group and the respective reference group.

Associations between the biomarker profile (research criteria) and 2-year risk of progression to dementia were analyzed with Cox proportional hazards models adjusted for randomized treatment (to account for intervention effects), baseline MMSE, sex, baseline age, and study site. Results are presented as hazard ratios (HR) and 95% CI. Results of all longitudinal analyses are reported for the modified intention-to-treat (mITT) population (all randomized participants with at least one post-baseline assessment, excluding visit data after progression to dementia and start of AD medication and/or open-label Souvenaid). SAS software version 9.4 was used in the analyses, and level of statistical significance was < 0.05. All analyses were post hoc.

Baseline characteristics of the participants with centrally assessed biomarkers (MTA on MRI and/or CSF) are shown in Table 2. Out of 287 participants with at least one centrally analyzed baseline biomarker, 180 (62.7%) had only MTA assessment available; CSF was analyzed for 107 participants (37.3%). Participants with CSF available were younger and had lower CDR-SB. In total, 62.3% of the participants with available apolipoprotein E (APOE) data were ε4 carriers. MTA score was at least 1 in 86.4% (241 out of 279) of the participants. CSF Aβ, t-tau, and p-tau levels were abnormal in 87.9% (94 out of 107), 86.0% (92 out of 107), and 69.2% (74 out of 107) of the participants, respectively. Classification according to the A/T/N biomarker scheme is shown in Fig. 1. The majority, 63.6%, had an A+T+N+ profile.

Figure 2 illustrates the classification according to the different research diagnostic criteria. In total, 75.7% (81 out of 107) of the participants had IWG-2 prodromal AD. Approximately half of the participants were classified in the NIA-AA 2011 intermediate AD likelihood group (54.0%, 155 out of 287) and 8.7% (25 out of 287) in the inconclusive/uninformative group. A third had a high AD likelihood (32.4%, 93 out of 287) and 4.5% (13 out of 287) were categorized in the SNAP group. One had an isolate amyloid pathology; no one had a low AD likelihood. With respect to the NIA-AA 2018 criteria, 63.6% of the participants (68 out of 107) were classified in the AD group, 23.4% (25 out of 107) in the Alzheimer’s and concomitant suspected non-Alzheimer’s pathologic change group, and 12.1% (13 out of 107) in the non-Alzheimer’s pathologic change group. One had an Alzheimer’s pathologic change. However, even a slight adjustment of the p-tau cut-off (> 55 pg/ml instead of > 60 pg/ml) changed the classification: in this case, more participants were assigned to the AD group (72.0%, 77 out of 107) and fewer participants to the Alzheimer’s and concomitant suspected non-Alzheimer’s pathologic change group (15.0%, 16 out of 107). The proportion of APOE ε4 carriers was high in groups with stronger evidence for AD pathology, and the percentage seemed to increase when applying the more recent criteria (NIA-AA 2011 intermediate AD likelihood 64.5%, high AD likelihood 72.2%, IWG-2 prodromal AD 76.9%, NIA-AA 2018 AD 78.5%; data not shown).

Table 3 shows per each set of criteria the estimates for 2-year change in cognitive and cognitive-functional performance, progression rates to dementia, and hazard ratios (HR) for dementia risk.

The changes in NTB scores were modest overall in both IWG-2 groups, and no differences were observed between the groups (Table 3). For CDR-SB, there was more worsening in the prodromal AD group (estimate for change 1.18 points vs. 0.42 points; p = 0.03). This group also had a higher risk of progression to dementia (42.0% vs. 26.9%; HR 4.6, 95% CI 1.6–13.7).

The changes in NTB scores were fairly small in all NIA-AA 2011 groups. No statistically significant differences were observed between the reference group SNAP and the other groups, but there was a trend indicating somewhat higher rate of decline in the intermediate and high AD likelihood groups (Table 3). The pattern of decline in these two groups was however similar. Individuals with a high AD likelihood were more likely to progress to dementia than those with SNAP (43.0% vs. 7.7%; HR 7.4, 95% CI 1.0–54.7). The risk was similar in the high and intermediate AD likelihood groups.

The changes in NTB were small across all NIA-AA 2018 groups. The AD group showed consistently the highest rate of decline, but the changes did not differ from those of the reference group non-AD pathologic change (Table 3). CDR-SB scores worsened more in the AD group than in the reference group (estimate for change 1.43 points vs. 0.48 points; p = 0.03) and in the Alzheimer’s and concomitant suspected non-Alzheimer’s pathologic change group (1.43 points vs. 0.43 points; p = 0.01). The AD group was also more likely to progress to dementia than the reference group (44.1% vs. 7.7%, HR 9.4, 95% CI 1.2–72.7) and the Alzheimer’s and concomitant suspected non-Alzheimer’s pathologic change group (44.1% vs. 40.0%, HR 2.7, 95% CI 1.0–7.0).

The study population was largely Caucasian and a homogenous group of RCT participants recruited primarily from memory clinics. Our findings may thus not be fully generalizable to other populations or settings, and further investigation will be needed in ethnically and geographically more diverse cohorts, ideally with a longer timeframe to assess disease progression. In the LipiDiDiet RCT, the IWG-1 criteria for prodromal AD were used to recruit participants, and the eligibility criteria included evidence for underlying AD pathology based on either imaging or CSF markers (i.e., participants were not required to have both assessments). Therefore, CSF was not available for the whole study population, and the classification according to certain criteria (IWG-2, NIA-AA 2018) was possible only in a subset of participants. Furthermore, as the sample was overall small, and only few participants were classified in the groups reflecting lower certainty of AD (e.g., low AD likelihood, SNAP), we lacked statistical power to detect between-group differences in the longitudinal analyses of disease progression. The smaller than expected cognitive changes over a rather short follow-up period of 2 years could have also affected our results. LipiDiDiet extension studies will shed light on the longer-term cognitive and cognitive-functional trajectories and prognosis in prodromal AD, which could improve the operationalization of the research diagnostic criteria. Another limitation of our study is that the Aβ and tau assessment was based only on CSF. CSF and PET reflect different aspects of pathology [43] and incorporating PET could have affected the classification. Nevertheless, Aβ status was determined based on both Aβ42 and Aβ42/40 ratio, which is an advantage as the latter parameter is potentially a more accurate measure of Aβ pathology [44]. Another major strength of our study was the possibility to investigate the research diagnostic criteria in a real-life RCT setting. This is because all biomarkers were centrally analyzed, and the sampling and assessment protocols were standardized. In multicenter studies, variability introduced by different laboratory procedures and cut-offs is a common challenge. Cognitive testing was also standardized in our study.