Early- versus late-onset Alzheimer’s disease in clinical practice: cognitive and global outcomes over 3 years

The SATS is a 3-year, prospective, open-label, non-randomised, multicentre study that was started to investigate the long-term effectiveness of ChEI therapy (donepezil, rivastigmine and galantamine) in clinical practice. Different findings from the SATS have been presented in several publications (e.g., [12, 13, 22‐24]). In total, 1258 participants with AD were recruited from 14 memory clinics located in different geographical areas across Sweden. All 1021 patients exhibiting a baseline Mini Mental State Examination (MMSE) [25] score ranging from 10 to 26 and an age at onset of AD (4 participants had missing data) were included in the current study. Of these, 143 individuals were defined as having EOAD (age at onset of AD < 65 years), and 874 were defined as having LOAD (age at onset of AD ≥ 65 years); hence, 1017 participants were enrolled.

Considered for inclusion in the SATS were outpatients aged ≥ 40 years who met the criteria for the clinical diagnosis of dementia, as defined in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition [26], and for possible or probable AD according to the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association [27]. All patients were diagnosed by physicians who specialise in dementia disorders. The dementia specialist estimated the age at onset on the basis of an interview with the caregiver (usually the spouse or an adult child) regarding observations of early symptoms of AD. Moreover, the selected individuals had to live at their own home at the time of AD diagnosis, to have a responsible caregiver and to be assessable with the MMSE at the start of the ChEI treatment (baseline). The exclusion criteria were not fulfilling the diagnostic criteria for AD, already receiving active ChEI therapy or having contra-indications to ChEIs.

After inclusion in the study and the baseline evaluations, the participants were prescribed ChEI treatment as part of the ordinary Swedish health-care system and in accordance with the approved product labelling. All patients started with donepezil 5 mg, rivastigmine 3 mg, or galantamine 8 mg, as in routine clinical practice. The SATS is an observational study, and the choice of drug type and all decisions regarding dosage were left entirely up to the dementia specialist’s discretion and professional judgement. Most individuals received an increased dose after 4–8 weeks of treatment, and we aimed at further dose increases depending on the chosen ChEI agent. However, for some participants, the dose was reduced because of side effects. The ChEI dose was recorded after 2 months of therapy and then every 6 months after baseline. Medications other than ChEIs were documented at baseline and allowed during the study, with the exception of memantine. If the patient stopped taking the ChEI or if memantine was initiated, the individual discontinued the SATS at that time point. The date of and reason for any drop-out from the SATS were recorded.

The SATS patients were investigated in a well-structured follow-up programme in which researchers evaluated cognitive, global and ADL performance at the start of ChEI treatment, after 2 months (MMSE and global rating only) and semi-annually over 3 years. Cognitive status was assessed using the MMSE, with scores ranging from 0 to 30 (a lower score indicating more impaired cognition), and the Alzheimer’s Disease Assessment Scale–Cognitive subscale (ADAS-Cog) [28], with a total range of 0 to 70 (a higher score indicating more impaired cognition). The Clinician Interview-Based Impression of Change (CIBIC) [29] was used as a global rating of ‘change from the initiation of ChEI therapy’. The evaluations were performed at all intervals using a 7-point scale ranging from 1 (very much improved) to 7 (marked worsening). Three groups of response were defined at each CIBIC interval: 1–3 indicated improvement, 4 indicated no change and 5–7 indicated worsening. No guidelines or descriptors were provided to define the individual ratings. The classification between, for example, minimally improved or very much improved was left to the dementia specialist’s clinical judgement.

The Instrumental Activities of Daily Living (IADL) scale [30] consists of eight different items: ability to use the telephone, shopping, food preparation, housekeeping, laundry, mode of transportation, responsibility for one’s own medications and ability to handle finances. Each item was scored from 1 (no impairment) to 3–5 (severe impairment), which yielded a total range of 8–31 points. The Physical Self-Maintenance Scale [30] consists of six different items: toileting, feeding, dressing, grooming, physical ambulation and bathing. Each item was scored from 1 (no impairment) to 5 (severe impairment), which allowed a total range of 6–30 points. Trained dementia nurses assessed the ADL performance on the basis of interviews with the caregiver. To facilitate the comparison of rates in MMSE and ADAS-Cog scores, changes in score were converted to positive values, which were indicative of improvement, and negative values, which were indicative of deterioration.

IBM SPSS Statistics for Windows version 22.0 software (IBM Corporation, Armonk, NY, USA) was used to perform the statistical analyses. The level of significance was defined as p < 0.05 if not otherwise specified, and all tests were two-tailed. Observed-case analyses were used to avoid overestimation of the treatment effect by imputing higher previous outcome scores in a longitudinal study of a progressively deteriorating disease. Parametric tests were used because of the large sample size and the approximately normally distributed continuous potential predictors. Independent-sample t tests were used to compare the differences between the means obtained for two groups, such as EOAD and LOAD, and χ² tests were used to analyse categorical variables.

Mixed, linear and non-linear fixed and random coefficient regression models using the participant as a hierarchical variable (i.e., to allow intra-individual correlation) were performed. The mixed-effects models method also takes into account the varying number of evaluations available for each patient and unequal time intervals between the follow-up visits, which are common statistical limitations found in long-term studies [31]. The non-completers contributed information during the time of participation; hence, we considered the trajectories of all SATS patients.

The dependent variables were the cognitive scores assigned at the second and subsequent visits for each individual; that is, the mixed-effects models do not intend to predict the scores at the initiation of ChEI therapy. The following described independent variables were included in the models. First, the initial cognitive scores for each participant (to adjust for baseline differences) and their interaction with linear and quadratic terms for time in the study (to enable a non-linear rate of change in the models) were included as fixed effects; that is, time in months (and time in months²) × MMSE (or ADAS-Cog) baseline score. Time was defined as the exact number of months between the start of ChEI treatment and each visit, thus using all data points at the actual time intervals. Secondly, several possible socio-demographic and clinical predictors were included as fixed effects in the models, such as sex; age at the start of ChEI therapy; clinician’s estimated duration of AD; years of education; presence of the APOE ε4 allele (no/yes); solitary living (no/yes); IADL and basic ADL capacity; number of medications at baseline; and specific concomitant medications (no/yes for each group), including antihypertensive/cardiac therapy, antidiabetics, asthma medications, thyroid therapy, lipid-lowering agents, oestrogens, NSAIDs/acetylsalicylic acid, antidepressants, antipsychotics and anxiolytics/sedatives/hypnotics. Thirdly, the effect of the different ChEI agents was analysed using the type of drug (coded as a set of dummy variables) and dosages. The terms ‘ChEI agent × dose’ and ‘age × ChEI dose’ were also included in the models. The ChEI dose could vary during the treatment period for an individual patient and between patients; therefore, the mean dose used during the entire follow-up period was calculated for each participant. In cases of drop-out, the mean dose used during the individual’s time of participation in the SATS was calculated. To obtain a similar metric for percentage maximum dosage for each of the three ChEIs, the mean dose was divided by the maximum recommended dose for each drug, namely 10 mg for donepezil, 12 mg for rivastigmine (oral administration) and 24 mg for galantamine. Lastly, some potential interactions (sex, age or education) with cognitive severity at baseline or with time in the study were included in the models. The random terms were an intercept and time in months, with a variance components covariance matrix. Non-significant variables (p > 0.05) were eliminated in a backward stepwise manner. The hierarchical principle was applied in the mixed-effects models; variables that appeared in significant interactions were not considered for elimination.

The 1017 SATS participants were divided into two cohorts according to their age at onset of AD: EOAD (< 65 years, n = 143 [14%]) and LOAD (≥ 65 years, n = 874 [86%]). The socio-demographic and clinical characteristics of the two cohorts are presented in Table 1. The presence of two APOE ε4 alleles was more frequent, and the proportion of one or no ε4 alleles was less frequent, among the patients with EOAD than in those with LOAD [χ²(2) = 23.98, p < 0.001]. The mean duration of illness was longer [t ₍₁₀₁₅₎ = 4.36, p < 0.001], and the level of education higher [t ₍₁₀₁₅₎ = 2.93, p = 0.004], among the younger than the older individuals. Cognitive ability at the initiation of ChEI therapy did not differ between the onset groups. Among the participants with EOAD, a lower percentage received donepezil and a higher percentage received rivastigmine or galantamine [χ²(2) = 8.09, p = 0.017]. The mean dose of donepezil during the study was higher in the younger cohort [t ₍₅₁₄₎ = 2.32, p = 0.020], but it was similar between the groups for the other two ChEI agents.

Regarding the MMSE score, 71% of the patients with EOAD and 64% of those with LOAD showed improvement/no change (≥ 0-point change) after 6 months of ChEI treatment [χ²(1) = 2.60, p = 0.107]. Improvement/no change (≥ 0-point change) in ADAS-Cog score was observed for 50% of the younger and 56% of the older cohort after 6 months [χ²(1) = 1.60, p = 0.206].

The mean (95% CI) MMSE and ADAS-Cog scores and the changes from baseline scores over the 3-year study by age-at-onset group are shown in Table 2. Using the ADAS-Cog scale, the rate of cognitive decline was faster among the EOAD participants at 12, 18 and 30 months after the initiation of ChEI therapy (Fig. 1). The mean (95% CI) semi-annual rates of change in the ADAS-Cog score at each time point for the younger and older cohorts, respectively, were, for 6–12 months, –2.4 (–3.8, –1.0) versus –1.9 (–2.4, –1.4) points (p = 0.483); for 12–18 months, –2.9 (–4.0, –1.8) versus –2.0 (–2.6, –1.5) points (p = 0.207); for 18–24 months, –1.5 (–2.9, –0.1) versus –1.9 (–2.5, –1.3) points (p = 0.643); for 24–30 months, –2.9 (–4.7, –1.2) versus –2.1 (–2.7, –1.5) points (p = 0.325); and for 30–36 months, –2.6 (–4.6, –0.6) versus –2.7 (–3.5, –1.8) points (p = 0.945). No significant difference in disease progression over time between the onset groups was detected when the MMSE score was used.

The percentages of the EOAD versus LOAD cohorts according to changes in global performance (CIBIC) after 2 months and semi-annually over 3 years from the start of ChEIs are illustrated in Fig. 2. The proportions of the remaining younger and older patients who exhibited improvement or no change in CIBIC score at each visit are presented in Table 2; however, no differences between the two groups were found. The individuals with EOAD and LOAD, respectively, were further divided into APOE genotypes. No significant differences in changes in cognitive or global capacities after 3 years of ChEI treatment were observed between these groups. The mean (95% CI) percentage of maximum ChEI dose was higher in EOAD than in LOAD participants at all visits after 12 months of therapy (Table 2).

In total, 86 patients (60%) with EOAD and 556 (64%) with LOAD did not complete the 3-year SATS [χ²(1) = 0.64, p = 0.425]. The reasons for drop-out in this cohort have been reported previously [24]. In the EOAD group, the completers received a greater percentage (mean ± SD) of the maximum recommended ChEI dose during the study [76 ± 14% versus 60 ± 19%, t ₍₁₄₁₎ = 5.66, p < 0.001]. No significant differences between the younger completers and drop-outs were found regarding cognitive or global status at the start of treatment. In the LOAD group, the completers also received a greater mean percentage of the maximum recommended ChEI dose during the study [69 ± 18% versus 59 ± 18%, t ₍₈₇₂₎ = 8.01, p < 0.001]. The older completers exhibited significantly better cognitive [mean ± SD, MMSE score 22.4 ± 3.4 versus 20.9 ± 3.8 points, t ₍₈₇₂₎ = 6.06, p < 0.001; ADAS-Cog score 18.6 ± 8.4 versus 22.4 ± 8.7 points, t ₍₈₅₈₎ = −6.30, p < 0.001] and global (CIBIC median score 3 versus 4 points, median test, p = 0.017) performance at baseline than the drop-outs. In addition, antipsychotic use was less frequent among the completers with LOAD [6 (2%) versus 37 (7%) patients, χ²(1) = 9.83, p = 0.002]. The other variables of interest in this study, such as sex, APOE genotype, age at baseline, duration of AD, years of education, number of concomitant medications and other specific medications received, did not differ between the completers and those who discontinued the study in any of the age-at-onset groups. Fifteen (10%) of the younger and 66 (8%) of the older individuals dropped out because of initiation of concomitant memantine therapy [χ²(1) = 1.45, p = 0.229]; their mean length (95% CI) of participation in the study was 23.1 (20.1–26.2) versus 19.7 (17.8–21.5) months [t ₍₇₉₎ = 1.66, p = 0.101], respectively.

In the mixed-effects models, only participants with three or more assessments were included to enable analyses of a non-linear rate of cognitive change (EOAD n = 128 [90%]; LOAD n = 774 [89%]). The models were performed to identify the socio-demographic and clinical predictors that affected the patients’ longitudinal trajectories (EOAD 667 data points, LOAD 3733 data points). The percentages of variance that accounted for the dependent variable regarding all fixed factors were 54.7% for MMSE and 53.6% for ADAS-Cog in the EOAD group and 51.1% for MMSE and 55.3% for ADAS-Cog in the LOAD group. This indicates a good fit of the models (p < 0.001 for all models). The mixed-effects models, significant predictors and unstandardised β coefficients with 95% CIs are presented in Tables 3 and 4.

Better cognitive status at the initiation of ChEI therapy implied a slower rate of progression over time. A higher mean ChEI dose during the study (regardless of drug agent) and a lower level of education were independent predictors of a more positive cognitive long-term outcome in both LOAD models and using the ADAS-Cog model in EOAD. No differences in cognitive trajectories were found among the three ChEI agents in the EOAD or LOAD cohorts. An interaction effect showed that a higher level of education led to increased cognitive impairment over the course of the disease. Female sex, younger age and less-preserved IADL capacity at the start of ChEI treatment were factors that implied a higher rate of cognitive deterioration in LOAD. Interaction effects between age (or sex in the ADAS-Cog mixed-effects model) and cognitive ability at baseline demonstrated that this difference in cognitive performance between age groups (or sex) was more pronounced among the older individuals who were more cognitively impaired. The interaction term of age with ChEI dose was not significant in any of the models. According to the MMSE model, the use of antipsychotics in participants with LOAD implied worse prognosis.

Below, we provide non-linear regression models for calculation of the predicted ADAS-Cog score for a cohort of ChEI-treated patients with EOAD and LOAD, based on the respective baseline score. These equations are intended to predict the scores at subsequent evaluations over a 3-year period. The ADAS-Cog models explained a substantial degree of variance in the dataset; that is, they displayed a good fit (EOAD R ² = 0.577, R = 0.760, p < 0.001; LOAD R ² = 0.537, R = 0.733, p < 0.001).

Predicted ADAS-Cog score in EOAD was calculated as follows:

Predicted ADAS-Cog score in LOAD was calculated as follows:

where t is the time in months between baseline and the actual visit and x _i is the baseline ADAS-Cog score.