Longitudinal associations of domain-specific subjective cognitive decline with plasma Alzheimer’s biomarkers and cognitive change

INTRODUCTION

Global population aging is driving an increase in the prevalence of Alzheimer’s disease (AD) and other dementias^[1]. AD accounts for approximately 60%-80% of dementia cases^[2] and poses a critical public health threat. Current disease-modifying therapies demonstrate efficacy primarily in the mild cognitive impairment (MCI) and early dementia stages^[3-5], underscoring the need to shift focus toward pre- or early clinical detection using cost-effective biomarkers^[6]. Subjective cognitive decline (SCD) was formally defined in 2014^[7] and operationalized as Stage 2 in the AD continuum according to the 2024 revised Alzheimer’s Association criteria^[8]. Amyloid-positive SCD specifically marks this preclinical AD stage^[9], and epidemiological studies confirm that SCD is associated with an elevated risk of progression to AD, approximately 10% every three years^[10-13]. Neuroimaging studies further corroborate early pathology in SCD, including medial temporal atrophy^[14,15] and hippocampal-parietal degeneration^[16]. Moreover, the integrative (amyloid, tau, and neurodegeneration) ATN biomarker framework, incorporating plasma markers [amyloid-β (Aβ), p-tau181, (glial fibrillary acidic protein) GFAP, and (neurofilament light chain) NfL] and magnetic resonance imaging (MRI) indicators, has shown particular utility in predicting near-term cognitive decline^[17].

Despite these advances, significant challenges remain. Commonly used SCD assessments, such as the Everyday Cognition (ECog) scale, disproportionately emphasize memory and executive functions while underrepresenting language and attention domains^[18,19]. Although self-reported SCD correlates with objective memory and language deficits^[20], its longitudinal association with declines in specific cognitive subdomains has not been firmly established. Furthermore, following the 2018 AD biomarker framework^[21], the link between SCD and the AD continuum requires further strengthening. The DELCODE study linked minor objective cognitive deficits in SCD to lower cerebrospinal fluid (CSF) Aβ42 levels^[22], and Aβ(1-38) has been shown to predict conversion from SCD to MCI^[21]; however, classical plasma biomarker studies face inherent limitations. Small-sample studies suggest that GFAP and p-tau isoforms (p-tau181, p-tau217) can differentiate MCI from SCD^[22-24]; nevertheless, the multimodal relationships remain fragmented. Critically, although evidence from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) links ECog-assessed domains to amyloid-β positron emission tomography (Aβ-PET) and tau-PET^[25], and other studies connect multidomain SCD to abnormal Aβ42/40 and p-tau^[26], ECog subdomains demonstrate limited predictive utility for established biomarkers^[27]. Consequently, a systematic investigation of persistent domain-specific SCD in relation to key plasma biomarkers, - namely Aβ42/40, p-tau181, p-tau217, GFAP, and NfL - has emerged as an essential research priority.

This community-based cohort study therefore addresses three pivotal questions: first, the associations between domain-specific SCD and declines in objective cognitive domains (OCD); second, the relationships between SCD and plasma AD biomarkers; and third, the associations of SCD with the incidence of MCI and dementia. Figure 1 shows the analytical strategy. The findings are expected to advance early AD detection paradigms.

Figure 1. The study design flowchart. MCI: Mild cognitive impairment; SCD: subjective cognitive decline; OCD: objective cognitive domains; HMACS: Hubei Memory and Aging Cohort Study; NfL: neurofilament light chain; GFAP: glial fibrillary acidic protein.

METHODS

RESULTS

Association of SCD (baseline, endpoint, and changes during follow-up) with changes in OCD

We first examined the association between baseline domain-specific SCD and changes in OCD. No significant associations were observed between any baseline SCD domains and changes in their corresponding OCD (all P > 0.05) [Supplementary Table 2]. Next, we analyzed the association between SCD domains at the end of follow-up and changes in OCD. After adjustment for all covariates, a higher global SCD score was associated with a decline in global OCD (β = -0.10, P < 0.001). Specifically, a higher SCD memory score was associated with a decline in OCD memory (β = -0.06, P = 0.025), and a higher global SCD score was associated with declines in OCD memory (β = -0.08, P = 0.008) and attention (β = -0.08, P = 0.002). In addition, a higher SCD execution score was associated with a decline in global OCD (β = -0.09, P < 0.001; Figure 2 and Supplementary Table 3). Finally, we investigated the association between changes in SCD domains and OCD. An increase in global SCD score was significantly associated with a decline in global OCD (β = -0.08, P = 0.001). Increases in SCD language (β = -0.08, P = 0.001) and execution scores (β = -0.07, P = 0.001) were each associated with a decline in global OCD. Moreover, an increase in global SCD score was associated with a decline in OCD attention (β = -0.07, P = 0.010) [Figure 2 and Supplementary Table 4].

Figure 2. Association of endline SCD domains and their two-year change with changes in objective cognitive domains. Average marginal mean estimates for objective cognitive domains change (and their 95%CIs) as a function of SCD domains. Demographic adjustment includes age, sex, education, and residence location. Full covariate adjustment additionally accounts for social network, lifestyle, and comorbidities, that is, all 23 variables which are listed in Table 1. SCD: Subjective cognitive decline; OCD: objective cognitive domains; CI: confidence interval.

Association between SCD domains and incident MCI/dementia

After full covariate adjustment, a higher global SCD score at the end of follow-up was associated with an increased risk of both MCI (HR = 1.17, P = 0.043) and dementia (HR = 1.75, P = 0.009). Similarly, SCD language was associated with dementia risk (HR = 1.44, P = 0.005), while SCD execution was associated with both MCI (HR = 1.10, P = 0.038) and dementia (HR = 1.33, P = 0.017). SCD attention was also associated with MCI (HR = 1.11, P = 0.020) and dementia (HR = 1.49, P = 0.001). In addition, an increase in global SCD over time was associated with incident MCI (HR = 1.16, P = 0.017). Increases in SCD global (HR = 2.00, P = 0.002), language (HR = 1.40, P = 0.006), execution (HR = 1.29, P = 0.018), and attention (HR = 1.43, P = 0.003) were all significantly associated with the risk of incident dementia [Figure 3 and Supplementary Tables 5-7].

Figure 3. Association of baseline, endline, and change over two years of SCD domains with incident MCI and dementia. Fully adjusted for covariates (adjusted variables: sex, age, education, residence, marriage, stable income, satisfaction with income, living alone, number of living siblings, number of close friends, frequency of interacting with neighbors, personality, smoking, drinking, physical activities, intellectual activities, hypertension, diabetes, coronary heart disease, hyperlipidemia, cerebrovascular disease, insomnia, and GDS). SCD: Subjective cognitive decline; CI: confidence interval; MCI: mild cognitive impairment; GDS: geriatric depression scale.

DISCUSSION

This community-based cohort study provides the first evidence that domain-specific SCD (in memory, language, execution, and attention) among cognitively unimpaired older adults, measured at follow-up or as change over 2.05 ± 1.26 years, was associated with decline in objective cognitive domains and with the incidence of MCI and dementia. These findings indicate that domain-specific SCD not only coincides with objective decline but also predicts near-term clinical onset. Furthermore, higher global SCD at follow-up was correlated with lower baseline plasma levels of Aβ42 and Aβ40, suggesting that plasma biomarkers may reflect both the presence and progression of SCD. Nevertheless, the stage and domain-specific characteristics of these biomarkers require further validation in larger longitudinal studies. Collectively, our results establish the value of SCD for screening AD-related cognitive impairment risk in community settings by linking domain-specific SCD to objective cognitive decline, clinical progression, and relevant plasma AD biomarkers.

A key advance of this study lies in its systematic assessment of four SCD subdomains (memory, language, executive function, and attention), along with behavioral characteristics such as time of onset, help-seeking behavior, and self-comparison with peers. This approach distinguishes it from prior studies that relied primarily on memory complaints alone^[38,39]. We found that both SCD scores at the end of follow-up and their changes over approximately two years were significantly associated with objective cognitive decline. This finding extends previous work, including the Latino Neurocognitive Aging Study, which assessed only global SCD and memory/executive domains^[6], and the DELCODE study, which linked global SCD to subtle cognitive deficits in a small sample^[40]. Longitudinal changes in SCD may better capture the true trajectory of cognitive decline than single-time assessments, offering a more sensitive marker for early risk detection. Our results complement neuroimaging evidence showing that hippocampal and amygdala volumes predict five-year dementia risk in individuals with SCD or MCI^[41]. The clinical relevance of our results is underscored by our cohort’s mean age (70.87 ± 5.06 years), which is closer to the high-risk period than that of the SOLINCA study (56.4 ± 8.1 years). Together with evidence linking SCD to atherosclerosis, AD neuropathological change (ADNC), Lewy body disease (LBD)^[42], and neuroimaging biomarkers, our study reinforces SCD as a predictive target for dementia across the aging spectrum.

This study provides the first evidence that domain-specific SCD, particularly in non-memory domains such as language, executive function, and attention, is associated with the risk of MCI/dementia, whether measured at follow-up or as a change over time. This finding reinforces the view of SCD as the earliest symptomatic stage of AD^[10,11,16]. Given that both the DSM-5 and the MCI classifications recognize single-domain impairment^[43], tracking domain-specific changes in SCD has direct clinical relevance. Furthermore, the global burden of dementia largely stems from the combined effects of multiple co-occurring neuropathologies^[44]. The inherent heterogeneity of AD dementia is evidenced by distinct pathology-cognition relationships, such as the association of medial temporal tau with memory deficits versus neocortical tau with executive dysfunction^[45]. Co-occurrence of ADNC with limbic-predominant age-related TDP-43 encephalopathy neuropathologic change (LATE-NC), LBD, or cerebrovascular disease exacerbates global and domain-specific cognitive decline^[44], and interactions among these pathologies shape distinct cognitive trajectories^[46]. Against this complex and heterogeneous background, our study systematically links subjective and objective domain-specific decline, addressing a key gap in the current understanding of preclinical cognitive impairment.

We found that in this cognitively unimpaired cohort, a higher global SCD score at follow-up was associated with lower levels of Aβ42 and Aβ40, which is consistent with the results of another Sino-German study^[9]. In contrast, greater longitudinal increases in SCD were associated with lower p-tau181 in our study, whereas a previous report from the same SILCODE group found that higher plasma p-tau181 levels were associated with greater brain Aβ deposition and faster cognitive decline in the preclinical stage^[47]. This apparent discrepancy may be due to differences in outcome measures (cross-sectional plasma p-tau181 vs. longitudinal change) and the specific SCD assessment tools used. No other AD-related biomarkers (i.e., p-tau217, NfL, or GFAP) showed significant associations. These results indicate that aberrant changes in plasma AD biomarkers are already detectable at the SCD stage, distinguishing these individuals from those without SCD. However, the associations between SCD and plasma AD biomarkers are not consistent. Several factors may explain this inconsistency. First, biomarkers were measured only at baseline, whereas SCD was assessed longitudinally. This temporal misalignment means that baseline biomarkers reflect pathological status at study entry, whereas changes in SCD capture dynamic cognitive perceptions over the follow-up period. This design differs from studies that measure biomarkers concurrently with SCD assessment^[48], which may account for discordant results. Second, our exclusive focus on cognitively unimpaired individuals places participants at the earliest preclinical stage (SCD Stage 2, according to AA criteria^[8]). At this stage, Aβ pathology may be accumulating but has not yet triggered the secondary tau phosphorylation and neuroinflammatory cascades that become more prominent in MCI and dementia^[49]. A recent Korean study demonstrated that plasma p-tau181, p-tau217, and GFAP levels were significantly higher in participants with MCI than in those with SCD (P < 0.05)^[22], supporting the notion that these markers may not yet be elevated in pure SCD without objective impairment. Another study from the Sino-German CLoCODE cohort similarly reported that plasma p-tau181 levels added predictive value for cognitive decline among SCD participants^[50]. The association with lower Aβ42/40 may reflect early amyloid deposition, while the absence of positive p-tau/GFAP associations aligns with findings that plasma GFAP and p-tau isoforms effectively differentiate MCI from SCD but show weaker signals at the SCD stage alone^[50]. Third, the inverse association with p-tau181 (β = -0.05, P = 0.047) is counterintuitive but may reflect competing pathologies. In very early stages, individuals with greater SCD awareness might have better health literacy and engagement, potentially slowing p-tau accumulation. Alternatively, this could represent a statistical artifact due to the relatively small biomarker subcohort (n = 657) and multiple comparisons. Recent work using ultrasensitive assays has shown that plasma p-tau217/Aβ42 exhibits the highest performance in detecting Aβ pathology among cognitively unimpaired individuals area under the curve (AUC = 0.94)^[51], suggesting that combinations of biomarkers may provide more robust signals than individual markers alone. These divergent findings underscore that SCD-biomarker relationships vary across populations and disease stages, emphasizing the need for longitudinal biomarker studies, harmonized protocols, and consideration of population-specific factors.

Stratified analyses revealed notable sex- and age-related patterns that inform the clinical utility of SCD. The observed sex differences - specifically, women showed associations across global SCD, language, and attention domains, whereas men primarily exhibited associations in the executive domain - may reflect women’s greater health awareness and sensitivity to subtle cognitive fluctuations^[52,53], whereas men’s executive complaints could align with vascular risk factors more prevalent in men that impact frontal-executive functions^[45]. Age stratification further illuminated the evolving nature of SCD: in the younger subgroup (60-69 years), associations were strongest for language and executive domains, higher-order functions vulnerable to early AD pathology in prefrontal and parietal networks^[10,16,45], whereas in participants aged ≥ 70 years, memory complaints emerged as the dominant predictor of objective memory decline, consistent with the classic progression of AD pathology to hippocampal structures^[7]. This age-dependent shift from non-memory to memory complaints suggests that the predictive value of specific SCD domains evolves along the disease continuum, with implications for the timing and targeting of screening efforts.

The identification of 5.7% of participants exhibiting single-domain worsening provides direct evidence that SCD progression is not merely a reflection of scale properties but rather reflects genuine domain-specific neurobiological processes. The relatively low proportion of single-domain worsening may explain the modest effect sizes observed in group-level analyses, as most participants either remained stable or showed decline across multiple domains.

The key strengths of this study include the application of a multidimensional SCD-I- recommended framework, the novel examination of correspondences between domain-specific SCD and both plasma biomarkers and objective cognitive change, and a large sample size that enhances the reliability of the findings. Its limitations include a relatively short follow-up period, which hinders causal inference; the potential subjectivity inherent in self-reported questionnaires; the need for nationwide multi-center validation to ensure generalizability, the absence of an MCI or dementia comparator group to contextualize the magnitude of biomarker associations; and the lack of neuroimaging validation for the blood-based biomarkers.

In conclusion, this study demonstrates that domain-specific SCD scores are associated with objective cognitive decline over a mean follow-up of two years and that they indicate an increased risk of incident cognitive impairment. Women appear more sensitive to SCD. Plasma Aβ42 and Aβ40, but not p-tau181, p-tau217, GFAP, or NfL, were associated with the presence and progression of SCD. However, the inherent variability of SCD itself and the differences across its cognitive domains are reflected in their distinct associations with the AD biomarker profile. Therefore, the utility of plasma AD markers for screening at the SCD stage may require further exploration. We recommend incorporating domain-specific SCD assessment into community-based AD screening programs and monitoring domain-specific changes using comprehensive test batteries.