The SLR was designed to identify relevant evidence in populations with AD (both mild cognitive impairment [MCI] due to AD and AD dementia), on the relationship between CDR-SB or CDR Global scores and outcomes. All study types, including randomized controlled trials (RCTs), observational studies, and case studies and reports, were considered relevant. Following consultation with a panel of clinical experts, outcomes of specific interest were NPS, cardiovascular disease (CVD), epilepsy, bedsores, infections, incontinence, obstructive sleep apnea, falls, ADLs, dependence, nursing home placement, economic consequences, and healthcare resource use.

Both journal and congress publications were considered relevant. Primary publications and SLRs or meta-analyses were included, but narrative reviews were not. Only English language publications were included, but there was no restriction by date or geography.

The study protocol was designed and conducted in line with the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (14), and registered with PROSPERO (registration number: CRD42023392801). Searches of Embase, MEDLINE, and the Cochrane Library were conducted on September 13, 2022. Titles and abstracts were screened by one reviewer to determine whether they met the eligibility criteria (Table 1). All publications meeting the criteria were obtained as full articles and reassessed against the eligibility criteria.

Detailed data, including study setting, methods, patient characteristics, and study results, were entered into a data extraction table, and quality checked by an independent reviewer.

Some studies reported longitudinal relationships between CDR-SB/CDR score progression and the progression of outcomes, whereas others reported cross-sectional data examining outcomes in different stages of AD severity. A third type of study reported how disease severity at baseline affected subsequent progression or how baseline characteristics affected AD progression.

The CDR scale comprises six domains: memory, orientation, judgement and problem solving, community affairs, home and hobbies, and personal care (12), each of which is assigned a score of 0, 0.5, 1.0, 2.0, or 3.0.

CDR Global scores (referred to hereafter throughout this review as CDR scores for brevity) reflecting dementia severity, are calculated from the domain scores using an algorithm, which generates the following categories of severity:

The CDR-SB score is derived by adding the scores of the six domains; the total score ranges between 0 and 18 (12).

The original 10-item version of the NPI (15) includes delusions, hallucinations, agitation/aggression, depression, anxiety, euphoria, apathy, disinhibition, irritability, and motor disturbances (13); night-time behavior disturbances, and appetite and eating abnormalities were later added (16). Each of the 12 items is scored on scales for frequency (1–4), severity (0–3), and distress (0–5). Individual scores for each domain are calculated as frequency × severity, and the total NPI score is the sum of the domain scores (13).

Various measures to assess ADLs and IADLs are used in AD research and clinical practice. Table 2 summarizes the scales included in this manuscript.

In total, 925 references were included for screening by abstract and title, resulting in 125 references included for full paper review (see Figure 1 for PRISMA flow diagram). In total, 57 references met the inclusion criteria at full paper review and were included for data extraction, along with one additional relevant reference that was identified in supplementary searches, resulting in 58 studies in total. Overall, 42 studies reported data on comorbidities (33–74), Twelve studies reported data on ADLs (21, 38, 49, 53, 62, 65, 67, 68, 75–78), two reported data on dependence (79, 80), five reported data on nursing home placement (68, 81–84), and six reported data on economic outcomes or resource use (68, 85–89). No studies reported data on bedsores, infections, obstructive sleep apnea, or falls.

Nineteen studies reported data from Asia (China, India, Japan, Singapore, South Korea, Taiwan, and Thailand) (33, 38, 40–43, 57, 58, 60, 61, 63, 67, 73, 74, 76, 78, 82, 83, 87), 17 reported data from Europe (Finland, France, Germany, Italy, Norway, Spain, Turkey, and the UK) (34–37, 39, 44, 48–50, 53, 54, 56, 62, 80, 85, 88, 89), and 12 reported data from North America, all of which were from the USA (21, 51, 52, 55, 59, 64, 65, 68–72). Five studies reported data from multiple countries (75, 77, 79, 81, 86), three reported data from South America (Brazil) (45–47), and two publications did not report the country (66, 84). More than two-thirds of the studies used primary data (39 studies); the remainder were conducted using secondary data, with the exception of one study with an unclear design (80). Individuals with AD were typically recruited from memory clinics, neurology departments, or hospitals, or as part of cohort studies of ageing, dementia, or AD. A small number of studies analyzed data collected in RCTs.

Most studies diagnosed AD according to standard clinical criteria, typically the National Institute of Neurological and Communicative Disorders and Stroke−Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria (90). However, only five studies (34, 49, 72, 75, 77) confirmed the diagnosis of AD using biomarkers. Four studies measured β-amyloid and/or tau in cerebrospinal fluid (34, 49, 75, 77), one of which detected amyloid deposits via positron emission tomography scans as an alternative method (75). The fifth study detected amyloid deposits using immunohistochemistry post mortem (72).

Twelve studies assessed ADLs (Table 3), and these data demonstrated, not surprisingly, that CDR and CDR-SB scores were reliably correlated with decline in ADLs and IADLs, irrespective of the ADL measurement instrument used in the study.

Five studies assessed AD progression and ADLs longitudinally. Hotta et al. found that the ability to perform ADLs, assessed using the Physical Self-Maintenance Scale (PSMS) and the IADL Scale, decreased with increasing CDR score over time (76). Eldholm et al. (49), Hallikainen et al. (53), and McDougall et al. (77) found correlations between changes in CDR-SB scores and decline in ADLs as measured using the IADL Scale, the Alzheimer’s Disease Cooperative Study (ADCS)-ADL scale, and the Functional Activities Questionnaire (FAQ), respectively. Tay et al. found that IADL scores were significantly worse at both baseline and follow-up for people who progressed by 2 CDR-SB points or more from baseline over 1 year than for those who did not progress, although no significant differences were detected in terms of the extent of IADL score change (67).

Four studies used cross-sectional data. Royall et al. (65), Dubbelman et al. (75), and Ton et al. (68) found that IADL scores were correlated with CDR or CDR-SB scores; the latter study also found that ADL scores were correlated with CDR-SB scores. Na et al. found that both CDR and CDR-SB scores were correlated with Functional Assessment Staging Tool (FAST) stage (78).

Five studies assessed correlations at baseline and subsequent progression. Hallikainen et al. found that patients with a CDR score of 0.5 at baseline maintained better ADCS-ADL scores each year over 3 years of follow-up than those with a baseline CDR score of 1.0 (53). Naurhashemi et al. found that individuals with AD who had a CDR score of 1.0 or greater at baseline experienced significantly greater ADL decrements over the following 2 years than those with a baseline score of CDR 0.5; however, there was no significant difference between the groups in terms of changes in the ability to perform IADLs (62). Similarly, Eldholm et al. found no significant differences between annual IADL score changes in patients with baseline CDR scores of 0.5, 1.0, or 2.0, which correspond to MCI due to AD, mild AD dementia, and moderate AD dementia, respectively (49). Caroline et al. found that baseline IADLs were a predictor of fast CDR-SB score progression (38), but Park et al. classified patients as having impaired or unimpaired ADLs at baseline using the PSMS and IADL Scale, and found no difference between groups in terms of CDR-SB score decline over the following year (21).

Two studies reported a correlation between scores on the Dependence Scale and CDR-SB scores, using cross-sectional data. Cohen et al. found that a 1-point change in CDR-SB score corresponded to a change of 0.68 points on the Dependence Scale in a multinational cohort with MCI or mild-to-moderate AD dementia (79), whereas Jones et al. found a Dependence Scale score increase of 0.47 for a 1-point increase in CDR-SB score in a UK population with AD dementia (80). For context, total score ranges are broadly similar for the CDR-SB and the Dependence Scale (0–18 and 0–15, respectively; Table 2).

Five studies examined associations between the risk of nursing home placement and CDR-SB or CDR scores (68, 81–84). AD severity, denoted using CDR or CDR-SB scores, was linked to nursing home placement in studies reporting cross-sectional and short-term longitudinal data.

Park et al. assessed the risk of nursing home placement over approximately 3 years’ follow-up in people recently diagnosed with AD in South Korea (n = 2,470 with baseline data; n = 816 with follow-up data) (83). According to the results of Cox proportional hazards models to identify predictors of nursing home placement, baseline CDR-SB score was not a significant predictor (hazard ratio [HR], 1.04; 95% CI, 0.99–1.09; p = 0.05), but annual CDR-SB score change was a predictor (HR, 1.15; 95% CI, 1.06–1.23; p < 0.01) (21). Knopman et al. analyzed data from the multinational ADCS trial of tocopherol and selegiline, comparing people with AD dementia who moved to a nursing home (n = 114) with those who did not (n = 227) (81). While baseline CDR-SB scores were not linked to nursing home placement, there was a significant link between reaching CDR 3.0 over 2 years’ follow-up, versus not reaching CDR 3.0, and nursing home placement (OR adjusted for baseline Mini Mental State Examination [MMSE] and total Behavior Rating Scale for Dementia scores, 7.0; 95% CI, 3.99–12.37; HR adjusted for baseline MMSE, 4.8; p < 0.001) (81). Ota et al. analyzed data from 633 people with MCI (etiology not specified) or AD dementia in Japan (82). Baseline CDR score was linked to the risk of nursing home placement during follow-up, although the duration of follow-up was not stated. Relative to a CDR score of 0.5, HRs were 1.40 (95% CI, 0.78–2.53; p = 0.269) for CDR 1.0, 2.82 (1.52–5.22; p = 0.001) for CDR 2.0, and 6.13 (2.47–15.24; p < 0.000) for CDR 3.0 (82). However, Rountree et al. followed 500 people with AD dementia (country not reported) for 20 years, and found that change in CDR-SB score was not related to the time to nursing home placement (84).

One study included only cross-sectional data. Ton et al. analyzed secondary data from 121 people with amnestic MCI and 174 with mild, moderate, or severe AD in the US Ageing, Demographics and Memory Study (ADAMS), with normal cognition as the reference group. Higher baseline disease severity, as determined via CDR-SB score category, was associated with use of nursing homes during the preceding 2 years (OR from linear trend tests across all five cognitive states, 2.28; 95% CI, 1.64–3.17; p < 0.001) (68).

Six studies reported data on costs and/or healthcare resource use (68, 85–89) (Supplementary Table 4). These data clearly demonstrated a strong relationship between the increasing requirement for formal and informal care with AD progression as assessed by CDR/CDR-SB scores. Direct medical costs also increased in the later stages of AD and were linked to the cost of residential care.

Two analyses used data from the Alzheimer’s disease Follow-Up Study (ALSOVA) study in Finland. Ruokostenpohja et al. assessed the likelihood of receiving the Finnish care partner’ allowance (89), which is a fee paid to family care partners providing care at home to a person with reduced functional capacity, illness, or disability. The allowance is not an automatic entitlement, and the value varies according to region (91). CDR-SB score was significantly associated with the likelihood of receiving the care partners’ allowance (89). Declining ADCS-ADL score was also a significant predictor of receiving the care partners’ allowance, but increasing NPI score was not a predictor (89). Jetsonen et al. found that people in higher baseline CDR-SB score categories incurred significantly higher annual costs for both formal and informal care during 5 years’ of follow-up, relative to those with lower CDR-SB scores (88).

In the analysis of data from the US ADAMS study reported by Ton et al. individuals in higher CDR-SB score categories had a relatively higher chance of using home care, as well as higher direct medical costs and lower household income (68). In an analysis of baseline multinational RCT data by Gustavsson et al., CDR-SB score was correlated with the following cost categories: total informal care (value of lost productivity for care partners younger than 65 years, and value of lost leisure time for those 65 years or older, not including time spent on supervision), patient accommodation, and community services (86). In this study, declining ADCS-ADL score was also correlated with these cost categories, and increasing total NPI score was correlated with use of informal care and community services. ADLs were identified as the largest cost driver in the analyses, with a 1-point decrease in ADCS-ADL score translating into a 3.6% increase in total costs of care (86).

Two studies reported cross-sectional data by CDR score. Darba et al. found that direct medical care costs increased with CDR score in a Spanish population, particularly at CDR 3.0. There were also notable increases in informal care costs (contributed by time spent on ADLs, IADLs and supervision) at CDR 2.0 and 3.0, compared with earlier stages. Social care costs and indirect costs arising from lost productivity of the care partner were highest at CDR 0.5, and lowest at CDR 1.0 (Figures 4A,B) (85). Ikeda et al. conducted a targeted literature review in Japan, finding that higher public long-term care costs and total medical costs (excluding AD medications) were linked to higher CDR scores (Figure 4C) (87).

Data on resource use were mixed. Gustavsson et al. found no significant association between CDR-SB score and hospitalization (86); however Ton et al. reported that increasing CDR-SB score was associated with a higher likelihood of hospitalization but a decreasing chance of outpatient visits (68). There was a linear but less pronounced relationship between increasing CDR-SB score and visits to a doctor and the same was true of drug utilization (68).

Figures 2–4 and Table 4 summarize the 12 studies that reported outcomes by CDR score category (37, 41, 49, 53, 54, 57, 62, 65, 72, 82, 85, 87). These studies suggest that NPI score increases with progression from MCI due to AD to mild AD dementia (CDR 0.5 to CDR 1.0) (53, 54). Both the number of NPS and the frequency of specific NPS increased across CDR score categories (37, 41, 57, 62, 65, 72). Caputo et al. found that hallucinations, and apathy were significantly more common at CDR 2.0 and CDR 3.0 than at CDR 1.0, and disinhibition and aberrant motor behavior were significantly more common at CDR 3.0 than at CDR 1.0 (37). Most CVD comorbidities had no clear relationship with CDR score, but people with a score of CDR 2.0 were slightly more likely to have experienced previous hypertension or heart disease than those at CDR 0.5 or CDR 1.0 (50). A steady decline in ADL scores was observed across CDR score categories, including between CDR 0.5 and CDR 1.0 (49, 53, 62). The likelihood of nursing home placement increased across CDR score categories (82). Accordingly, both Darba et al. (85) and Ikeda et al. (87) found that care costs increased with CDR score, with the largest increases being seen at CDR 3.0. However, there was no clear linear relationship between CDR score and healthcare costs in these studies. Darba et al. found that direct medical costs were higher at CDR 0.5 than at CDR 1.0, and increased again at CDR 2.0 and CDR 3.0, and in Ikeda et al. total healthcare costs did not increase notably until CDR 3.0.