Caregiver positivity as a psychosocial modifier of lecanemab‐related outcomes in Alzheimer's disease: A multicenter longitudinal study in China

Abstract

INTRODUCTION

Real‐world modifiers of lecanemab benefit in Alzheimer's disease (AD) remain elusive. We examined associations between caregiver positivity and longitudinal outcomes in AD.

METHODS

This seven‐center longitudinal study followed 127 biomarker‐supported AD patient–caregiver dyads over 18 months. Mixed‐effects, spline, and propensity‐score analyses evaluated continuous Positive Aspects of Caregiving (PAC) scores.

RESULTS

Higher baseline PAC scale scores were associated with more favorable 18‐month outcomes in cognition (Mini‐Mental State Examination (MMSE): β = 0.42, 95% confidence interval [CI]: 0.24–0.59; p < 0.001), daily function (Alzheimer's Disease Cooperative Study–Activities of Daily Living [ADCS‐ADL]: β = 0.68, 95% CI: 0.42–0.93; p < 0.001), disease severity (Clinical Dementia Rating–Sum of Boxes [CDR‐SB]: β = −0.26, 95% CI: −0.37 to −0.14; P < 0.001), patient quality of life (Quality of Life in Alzheimer's Disease [QOL‐AD]: β  = 0.53, 95% CI: 0.34–0.71; p < 0.001), and lower apathy ( β  = −0.09, 95% CI: −0.16 to −0.03; p = 0.003); these associations remained significant after multiplicity correction (all q’s ≤ 0.004). Patient QOL‐AD showed the most consistent sensitivity‐analysis association. Apathy was associated with caregiver burden and caregiver quality of life, but mediation was not established.

DISCUSSION

Caregiver positivity may stratify lecanemab efficacy in AD, supporting psychosocial assessment in personalized, caregiver‐inclusive management pathways.

Highlights

• This multicenter Chinese real‐world study examined caregiver positivity, measured by the Positive Aspects of Caregiving (PAC) scale, in Alzheimer's disease (AD) patient−caregiver dyads receiving lecanemab.

• Higher baseline PAC score was associated with more favorable cognitive, functional, and disease‐severity trajectories.

• The PAC scale may help characterize dyads with greater potential benefit from lecanemab and inform collaborative patient−caregiver management.

• Apathy may be associated with the caregiver's psychological state and the patient's and caregiver's benefits.

1. BACKGROUND

Alzheimer's disease (AD) is the leading cause of dementia and a major contributor to disability, dependency, and long‐term care use worldwide. The number of people living with dementia was estimated at 57.4 million in 2019, and is projected to reach 152.8 million by 2050. ¹ This growing burden is particularly pressing in China, where rapid population aging highlights the need for interventions that preserve cognition, function, and quality of life (QOL) during the early symptomatic stages of AD.

The emergence of disease‐modifying therapies (DMTs) has changed the therapeutic landscape of early AD. In the Phase 3 Clarity AD trial, lecanemab slowed clinical decline over 18 months, with a between‐group difference of −0.45 points on the Clinical Dementia Rating–Sum of Boxes (CDR‐SB; 95% confidence interval [CI]: −0.67 to −0.23), corresponding to a 27% reduction in decline relative to placebo. These findings indicate that anti‐amyloid therapy can alter the clinical trajectory of early AD, rather than provide symptomatic benefit alone. ² Evidence from other anti‐amyloid trials further supports the clinical relevance of DMTs. In the phase 3 TRAILBLAZER‐ALZ 2 randomized clinical trial, donanemab slowed integrated disease progression among participants with low or medium tau burden, with a 3.25‐point between‐group difference on the integrated Alzheimer's Disease Rating Scale (iADRS) at 76 weeks and a 35.1% slowing of decline. ³ Together, these data indicate that amyloid‐targeting therapies may provide measurable benefits across cognition, function, and global disease progression. However, real‐world effectiveness of DMTs depends on factors beyond pharmacology alone. In AD care, caregivers play a central role in treatment decision‐making, infusion adherence, safety monitoring, maintenance of daily routines, and behavioral engagement. Evidence from China further suggests that caregiver characteristics are associated with perceptions of lecanemab and burden, ⁴ underscoring the role of caregivers in the implementation of DMTs. This caregiver‐centered perspective is consistent with China's evolving dementia care policy. The National Action Plan for Addressing Dementia in Older Adults (2024–2030) calls for an integrated continuum spanning prevention, screening, diagnosis, rehabilitation, follow‐up, and care, while strengthening community support, caregiver training, and family assistance. ⁵ Earlier national initiatives similarly highlighted patient self‐management, family‐based care, community services, hospital‐based management, caregiver education, and multidisciplinary service teams. ⁶ Thus, evaluating caregiver positivity during lecanemab treatment is relevant to a policy framework that recognizes caregivers as integral participants in dementia care.

Most AD caregiver research has focused on burden, depression, and distress, whereas adaptive psychological resources remain less well characterized. The PAC framework captures self‐affirmation and positive meaning. However, whether PAC is associated with clinical benefit during lecanemab treatment remains unclear. ⁷ , ⁸ This gap is particularly relevant because neuropsychiatric symptoms often contribute to caregiver burden beyond the effects of cognitive impairment alone. Apathy, agitation, emotional lability, and aberrant motor behavior may reduce caregiver QOL and increase supervision needs. ⁹ In the Aging, Demographics, and Memory study (ADAMS), patients with one to two neuropsychiatric symptoms required 10.0 additional hours of active help and 12.4 additional hours of supervision per week, whereas those with three or more symptoms required 18.2 and 28.7 additional hours, respectively. ¹⁰ Annual dementia‐related costs have been estimated at $56,290 or $41,689 per person, depending on how informal care is valued, ¹¹ and U.S. analyses have further estimated informal caregiving costs of $36,667 using replacement‐cost approach and $15,792 using opportunity‐cost approach. ¹² These findings underscore that neuropsychiatric symptoms are important contributors to caregiver burden, time demands, and economic cost.

Against this background, we hypothesized that caregiver positivity may serve as a clinically relevant psychosocial marker of lecanemab‐related benefit, rather than simply reflecting milder disease. We examined whether baseline PAC identified a patient–caregiver profile associated with more favorable cognitive, functional, behavioral, and quality‐of‐life trajectories, and whether these associations were partly explained by changes in neuropsychiatric symptoms and caregiver outcomes. By incorporating positive aspects of caregiving into DMT research, this multicenter longitudinal study aimed to extend the biopsychosocial framework of AD treatment and evaluate PAC as a measurable caregiver‐related factor associated with clinical benefit during lecanemab treatment.

RESEARCH IN CONTEXT

1. Systematic review: Although disease‐modifying therapies have transformed early Alzheimer's disease (AD) treatment, research primarily emphasizes biological outcomes rather than caregiver psychological factors. The Hilgeman study suggests that baseline caregiver positivity (Positive Aspects of Caregiving [PAC]) influences the responsiveness to behavioral support; however, its association with lecanemab's real‐world effectiveness and dyadic outcomes remains unclear.

2. Interpretation: In this multicenter study, higher caregiver positivity was consistently correlated with superior patient outcomes. There was an improvement in neuropsychiatric symptoms, especially apathy, a reduction in caregiver burden, and an enhancement in the quality of life. The findings support a caregiver‐inclusive view of lecanemab, positioning caregivers as essential participants in treatment implementation.

3. Future directions: Future research should develop predictive models that integrate biological markers, clinical status, and caregiver psychosocial factors to identify dyads likely to experience sustained lecanemab benefit. These models will support caregiver‐inclusive treatment pathways by coordinating pharmacologic therapy with multidisciplinary support services.

2. METHODS

2.1. Study design

We conducted a multicenter, prospective longitudinal study in China to investigate whether caregiver positivity was associated with clinical benefit during lecanemab treatment in AD. The study was conducted across seven cognitive disorder centers in northern, eastern, southern, and central China, providing broad geographic representation. Data were collected through structured outpatient assessments and online questionnaires, allowing concurrent assessment of patient clinical outcomes and caregiver‐reported psychosocial measures under real‐world treatment conditions.

Participants were enrolled on a rolling basis across centers according to lecanemab initiation and routine follow‐up schedules. Because treatment initiation was not synchronized, baseline and follow‐up assessments were accumulated longitudinally (July 2024 to April 2026). The analytic dataset was locked in April 2026, when statistical analyses were initiated after most available follow‐up data had been collected. A total of 127 patients receiving lecanemab and 127 informal caregivers were enrolled. Patients were eligible if they were 50–85 years of age, met the National Institute on Aging–Alzheimer's Association (NIA‐AA) core clinical criteria for AD dementia, and had biomarker evidence consistent with AD pathology, defined by either positivity on amyloid positron emission tomography (PET) findings or internationally approved plasma biomarkers (phosphorylated tau 217 [p‐tau217] and p‐tau217/amyloid beta 42 [Aβ42], and so on). Informal caregivers were eligible if they were 18 years of age or older, had lived with the patient and been closely involved in caregiving for at least 1 year, and spent at least 1 hour per day in face‐to‐face contact with the patient. Caregiving involvement included meeting daily needs, supervising medication, accompanying patients to medical appointments, and maintaining communication with hospital staff during hospitalization when necessary. Caregivers were excluded if they had a diagnosed psychiatric disorder, except nicotine dependence, or were unable to communicate effectively. All participants were enrolled after providing written informed consent.

Participants were followed for up to 18 months, with assessments scheduled at baseline, 3, 6, 12, and 18 months. The 3‐month visit was regarded as an interim assessment, whereas the 6‐month visit served as the principal outcome assessment; 12‐ and 18‐month visits were used to describe longer‐term trajectories. Follow‐up counts represented visit‐specific analyzable observations rather than monotonic retention counts. Thus, some dyads could miss the 3‐month window or enter observation after treatment initiation but still contribute 6‐month data. Overall, 76 dyads contributed data at 3 months, 104 at 6 months, 58 at 12 months, and 22 at 18 months.

2.2. Study measurements

2.2.1. Patient demographic, clinical, and outcome assessments

Baseline data collection included comprehensive demographic and clinical characterization of both patients and caregivers. Patient characteristics comprised age, sex, marital status, educational attainment, disease duration, comorbid conditions, and apolipoprotein E (APOE) genotype. Core patient outcome measures were selected based on prior literature and Chinese expert recommendations for the standardized assessment of cognitive impairment and dementia in clinical practice. ¹³ The patient assessment battery included the Mini‐Mental State Examination (MMSE) for global cognition, the Alzheimer's Disease Cooperative Study–Activities of Daily Living (ADCS‐ADL) inventory for functional ability, the Neuropsychiatric Inventory (NPI) for behavioral and psychological symptoms of dementia (BPSD), the CDR‐SB for disease severity, and the Quality of Life in Alzheimer's Disease (QOL‐AD) scale for patient‐reported quality of life.

2.2.2. Caregiver characteristics and psychosocial assessments

Caregiver characteristics included age, sex, marital status, educational level, caregiving duration, daily caregiving time, geographic region, underlying comorbidities, and relationship to the patient. Caregiver outcomes included the Positive Aspects of Caregiving (PAC) scale, caregiver QOL‐AD, and the Zarit Burden Interview (ZBI).

The PAC scale was originally developed to assess positive appraisals of caregiving and comprises two core dimensions: self‐affirmation and outlook on life. ¹⁴ In the present study, self‐affirmation was used to reflect self‐efficacy, coping capacity, and psychological resilience, whereas outlook on life was used to capture positive expectations and future orientation.

Caregiver QOL‐AD was assessed using the caregiver‐rated QOL‐AD, a 13‐item instrument covering physical health, energy, mood, living environment, memory, family, intimate relationships, friendships, self‐evaluation, ability to perform household activities, capacity for enjoyment, financial status, and overall life quality. ¹⁵ Caregiving burden was measured using the Zarit Burden Interview (ZBI). The ZBI assessed multidimensional caregiving burden, including practical demands, emotional strain, social support, self‐efficacy, and health‐related impact. ¹⁶

2.3. Statistics

Descriptive statistics were used to summarize participant characteristics. Categorical variables were presented as frequencies and percentages, whereas continuous variables were expressed as means ± standard deviations (SDs) or medians with interquartile ranges (IQRs), according to distributional properties. Normality was assessed using the Shapiro–Wilk test. Group comparisons between high‐ and low‐PAC strata were performed using the Pearson chi‐square test or Fisher's exact test for categorical variables, and the independent‐samples t‐test or Mann–Whitney U test for continuous variables, as appropriate.

The pre‐specified primary analyses focused on baseline PAC scale total score as a continuous exposure and its association with longitudinal patient outcomes, including MMSE, ADCS‐ADL, CDR‐SB, NPI, and patient‐reported QOL‐AD. Exploratory analyses assessed PAC‐by‐time interactions, grouped comparisons, dimension‐specific associations of PAC self‐efficacy and outlook on life, and symptom‐level relationships of neuropsychiatric symptoms with caregiver burden, caregiver QOL‐AD, and caregiving time.

Longitudinal associations were examined using linear mixed‐effects models (LMMs) with random intercepts for participants. Models included fixed effects for PAC, follow‐up time, and the PAC‐by‐time interaction. Models controlled for clinically relevant baseline covariates, including patient age, sex, education, disease duration, APOE ε4 carrier status, baseline values of the corresponding outcome, baseline MMSE, baseline ADCS‐ADL, baseline CDR‐SB, baseline NPI total score, patient baseline QOL‐AD, baseline ZBI, and pretreatment daily caregiving time, caregiver age, and caregiver sex. Mixed‐effects models incorporated all available repeated observations under a missing‐at‐random assumption conditional on observed covariates.

To address multiplicity, p‐values were adjusted within separately defined analysis families. The primary outcome family comprised MMSE, ADCS‐ADL, CDR‐SB, NPI total, NPI‐apathy, and patient QOL‐AD. The Benjamini–Hochberg (BH) false discovery rate (FDR) was used as the primary correction because the outcomes were clinically correlated. Holm and Bonferroni corrections were additionally calculated as conservative sensitivity analyses; Bonferroni‐adjusted p‐values were derived by multiplying each raw p‐value by the number of tests within the corresponding analysis family and truncating values at 1.00. Standardized β coefficients with 95% CIs were obtained by rescaling the reported unstandardized β coefficients and 95% CIs by SD(predictor)/SD(outcome) within the analytic sample. Partial R² was calculated as t² /(t²+df) from Satterthwaite t statistics in refitted standardized mixed‐effects models. Incremental marginal R² was calculated as the difference between the full‐model Nakagawa‐style marginal R² and the marginal R² from a reduced model omitting the corresponding PAC term; negative differences were truncated at 0 and results were expressed as percentage points. Full correction results are in Table S1.

Grouped PAC comparisons were first explored using adjusted mixed‐effects models, and associations with continuous PAC were further examined using restricted cubic spline (RCS) models. RCS models were fitted with knots at the 10th, 50th, and 90th percentiles of PAC. Likelihood‐ratio tests assessed the overall PAC association, nonlinearity, and the PAC‐by‐time interaction, and model‐predicted contrasts between the 90th and 10th percentiles of PAC were estimated.

Propensity‐score matching (PSM) was performed to address measured baseline imbalance between high‐ and low‐PAC strata. The propensity‐score model included patient age, sex, education, disease duration, APOE ε4 carrier status, baseline MMSE, ADCS‐ADL, CDR‐SB, NPI total score, patient QOL‐AD, ZBI, pretreatment daily caregiving time, caregiver age, and caregiver sex. One‐to‐one nearest‐neighbor matching used the logit propensity score, the average treatment effect on the treated (ATT) estimand, no replacement, and a caliper of 0.2 SD of the logit propensity score; covariate balance was assessed using standardized mean differences, with absolute values < 0.10 indicating acceptable balance. Inverse probability of treatment weighting (IPTW) was conducted as an additional sensitivity analysis using stabilized propensity‐score weights targeting the average treatment effect (ATE) estimand. Matched and weighted longitudinal models compared high‐ and low‐PAC strata across follow‐up.

Associations among neuropsychiatric symptoms, caregiver burden, and caregiver QOL‐AD were examined using Pearson correlation analyses and mixed‐effects models. Exploratory regression‐based mediation analyses evaluated whether apathy might represent a candidate symptom‐level pathway linking caregiver positivity with caregiver outcomes. Mediation was considered supported only when bootstrap 95% CIs for indirect effects excluded zero. A hypothesized conceptual framework linking baseline dyadic context, caregiver positivity, neuropsychiatric symptoms, caregiving demands, treatment implementation, and longitudinal outcomes is shown in Figure 1.

FIGURE 1.

Caregiver positivity and lecanemab outcomes: conceptual framework. Note: This diagram illustrates hypothesized relationships among baseline clinical–dyadic context, caregiver positivity, patient apathy, caregiver burden, and patient outcomes during treatment. Solid green arrows denote primary adjusted associations; solid brown arrow indicates the symptom‐burden pathway; gray dashed arrow represents candidate influencing baseline factors. ADCS‐ADL, Alzheimer's Disease Cooperative Study–Activities of Daily Living; BPSD, behavioral and psychological symptoms of dementia; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; QOL‐AD, Quality of Life in Alzheimer's Disease; ZBI, Zarit Burden Interview.

3. RESULTS

3.1. Patient characteristics

A total of 127 patient–caregiver dyads from seven Chinese memory centers were included in this multicenter longitudinal cohort. Patients had a mean age of 68.0 years (SD 8.41), with a median age of 68 years (IQR 61,74), and 59.1% were female. Most patients were married (78.0%), and APOE ε4 carriage was common (74.8%), consistent with a biomarker‐enriched early AD population. Median educational attainment was 12 years (IQR 9,15), median disease duration was 3 years (IQR 1,5), and baseline cognitive status was consistent with mild impairment overall, with a median MMSE score of 21 (IQR 16,23). Baseline functional performance, global clinical severity, and behavioral and psychological symptoms were characterized by median ADCS‐ADL, CDR‐SB, and NPI scores of 46 (IQR 40,50), 4 (IQR 2.5,6), and 4 (IQR 0,11), respectively. Baseline patient QOL‐AD was in the moderate range, with a median QOL‐AD score of 33 (IQR 28,36). Together, these data suggest a cohort comprising primarily patients in the early symptomatic stages of AD, with relatively preserved but vulnerable functional status. 

TABLE 1.

Baseline characteristics of patients and caregivers.

Variable	Overall
Patient characteristics	n = 127
Gender, n (%)
Male	52 (40.9%)
Female	75 (59.1%)
Age, years
Mean ± SD	68 ± 8.41
Median (IQR)	68 (61, 74)
Marital status, n(%)
Married	99 (78.0%)
Unmarried	28 (22.0%)
APOE ε4 carrier, n(%)
Carrier	95 (74.8%)
Uncarrier	32 (25.2%)
Education, years	12 (9, 15)
Disease duration, years	3 (1, 5)
Baseline MMSE	21 (16, 23)
Baseline ADCS‐ADL	46 (40, 50)
Baseline CDR‐SB	4 (2.5, 6)
Baseline NPI	4 (0, 11)
Baseline QOL‐AD (patient)	33 (28, 36)
Caregiver characteristics	n = 127
Gender, n(%)
Male	55 (43.3%)
Female	72 (56.7%)
Age, years
Mean ± SD	58.95 ± 12.87
Median (IQR)	60 (50, 68)
Relationship to patient, n(%)
Spouse	81 (63.8%)
Children	44 (34.6%)
Others	2 (1.6%)
Marital status, n(%)
Married	123 (96.9%)
Unmarried	4 (3.1%)
Underlying disease, n(%)
With underlying comorbidities	77 (60.6%)
Without underlying comorbidities	50 (39.4%)
Education, years	12 (12, 16)
Caregiving years	2 (1, 3.13)
Pre‐treatment caregiving hours/day	3 (0, 6)
Post‐treatment caregiving hours/day	2 (0, 5.25)
Baseline QOL‐AD (caregiver)	37 (32.75, 40)
Baseline ZBI	9 (5, 17.25)
PAC scores
PAC total score	34 (31.75, 38)
PAC‐Self‐efficacy	19 (16, 21)
PAC‐Outlook on life	15.5 (14, 17.25)

Note: Continuous variables were assessed for distributional characteristics before tabulation. Variables with approximately normal distributions are presented as mean ± SD, whereas skewed variables are presented as median (IQR). Categorical variables are summarized as n (%). Percentages were calculated using the number of participants with available data for each variable as the denominator. Pre‐treatment and post‐treatment caregiving hours refer to average daily caregiving time before and after lecanemab initiation, respectively, and are reported as median (IQR). APOE ε4 carrier status was defined as the presence of at least one ε4 allele.

Abbreviations: ADCS‐ADL, Alzheimer's Disease Cooperative Study–Activities of Daily Living; APOE ε4, apolipoprotein E ε4 allele; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; IQR, interquartile range; MMSE, Mini‐Mental State Examination; n, number; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; QOL‐AD, Quality of Life in Alzheimer's Disease; SD, standard deviation; ZBI, Zarit Burden Interview.

3.2. Caregiver characteristics

Caregiver data were available for 127 participants. Caregivers were younger than patients, with a mean age of 58.95 years (SD 12.87) and a median age of 60 years (IQR 50,68). Women accounted for 56.7% of caregivers, indicating a modest female predominance. Most caregivers were spouses (63.8%), followed by adult children (34.6%), highlighting the central role of family‐based care in the Chinese context. Nearly all caregivers were married (96.9%), and 60.6% reported at least one underlying medical comorbidity.

Caregiving burden was variable across families. Median caregiving duration was 2 years (IQR 1,3.13), with a median of 3 h/day before treatment (IQR 0,6), which decreased to 2 h/day after treatment initiation (IQR 0,5.25). Median caregiver QOL‐AD score was 37 (IQR 32.75,40), whereas median ZBI score was 9 (IQR 5,17.25), suggesting generally mild burden, with substantial between‐family variability.

Caregiver positivity was relatively high overall: median PAC total score was 34 (IQR 31.75,38), with subscale scores of 19 (IQR 16,21) for self‐efficacy and 15.5 (IQR 14,17.25) for outlook on life. These data supported subsequent analyses of caregiver positivity as a psychosocial factor associated with patient and caregiver outcomes. Baseline demographic, clinical, and caregiving characteristics of patients and caregivers are summarized in Table 1. 

FIGURE 2.

Standardized effect sizes for the primary associations between baseline PAC total score and patient outcomes. Note: This figure presents standardized beta coefficients and 95% CIs for the primary continuous PAC associations across patient outcomes, allowing comparison of effect magnitudes across scales. ADCS‐ADL, Alzheimer's Disease Cooperative Study‐Activities of Daily Living; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; CI, confidence interval; FDR q, false discovery rate‐adjusted q value; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; QOL‐AD, Quality of Life in Alzheimer's Disease; Std beta, standardized beta coefficient.

3.3. Association between PAC total score and clinical outcomes

After multivariable adjustment, each 1‐point higher baseline PAC total score was associated with higher MMSE (β = 0.42; 95% CI: 0.24–0.59; p < 0.001; standardized β = 0.424), higher ADCS‐ADL (β = 0.68; 95% CI: 0.42–0.93; p < 0.001; standardized β = 0.506), lower CDR‐SB (β = –0.26; 95% CI: –0.37 to –0.14; p < 0.001; standardized β = –0.376), and higher patient QOL‐AD (β = 0.53; 95% CI: 0.34–0.71; p < 0.001, standardized β = 0.471). These associations remained significant after BH‐FDR correction (all q’s = 0.002). Although the standardized coefficients were moderate, the incremental marginal R² associated with inclusion of the PAC term was modest, ranging from 0.03% to 1.26% across these outcomes.

PAC total score was not associated with the NPI total score (β = −0.18; 95% CI: −0.58 to 0.22; p = 0.386; q = 0.386; standardized β = −0.086), but was associated with lower apathy severity (β = −0.09; 95% CI: −0.16 to −0.03; p = 0.003; q = 0.004; standardized β = −0.335). The incremental marginal R² for the PAC term in the apathy model was 1.41%, indicating a modest additional contribution beyond the measured covariates.

PAC‐by‐time interaction terms did not remain significant after multiplicity correction for MMSE, ADCS‐ADL, CDR‐SB, NPI total, apathy, or patient QOL‐AD (all q’s ≥ 0.340). Thus, the data did not provide statistical evidence that the PAC‐outcome associations differed across scheduled visits. Although the ADCS‐ADL point estimates were numerically larger at later assessments, the overall interaction was not significant; these post hoc estimates were therefore not interpreted as evidence of a cumulative or progressively increasing association. Full estimates are presented in Table 2 and Figure 2.

TABLE 2.

Integrated primary longitudinal associations between baseline PAC total score and patient outcomes.

Outcome/term	LMM β (95% CI)	p	BH‐FDR q	Std β (95% CI)	Partial R 2	Incremental marginal R 2, %	RCS overall q	RCS nonlinear q	RCS P90‐P10 contrast
MMSE	0.42 (0.24, 0.59)	<0.001	0.002	0.424 (0.242, 0.596)	0.0027	0.10	0.813	0.640	0.46 (−1.14, 2.06), q = 0.656
MMSE*TIME		0.229	0.344
ADCS‐ADL	0.68 (0.42, 0.93)	<0.001	0.002	0.506 (0.312, 0.692)	0.0036	0.03	0.813	0.973	−0.58 (−2.34, 1.18), q = 0.656
ADCS‐ADL*TIME		0.082	0.340
3 months	4.63 (−0.44, 9.69)	0.073	0.117
6 months	5.59 (0.31, 10.88)	0.038	0.076
12 months	8.39 (2.47, 14.31)	0.006	0.048
18 months	9.72 (1.59, 17.86)	0.019	0.076
CDR‐SB	−0.26 (−0.37, −0.14)	<0.001	0.002	−0.376 (−0.535, −0.202)	0.0019	0.04	0.813	0.640	−0.18 (−0.92, 0.56), q = 0.656
CDR‐SB*TIME		0.162	0.340
NPI
NPI total	−0.18 (−0.58, 0.22)	0.386	0.386	−0.086 (−0.277, 0.105)	0.0027	0.03	0.463	0.640	1.94 (−6.60, 10.47), q = 0.656
NPI total*TIME		0.999	0.999
NPI‐apathy	−0.09 (−0.16, −0.03)	0.003	0.004	−0.335 (−0.596, −0.112)	0.0405	1.41	0.182	0.640	−0.65 (−1.26, −0.03), q = 0.118
NPI‐apathy*TIME		0.380	0.456
QOL‐AD (patient)	0.53 (0.34, 0.71)	<0.001	0.002	0.471 (0.302, 0.631)	0.0906	1.26	0.036	0.640	2.40 (0.90, 3.90), q = 0.010
QOL‐AD (patient)*TIME		0.170	0.340
3 months	0.12 (0.01, 0.25)	0.029	0.076
6 months	0.09 (−0.04, 0.21)	0.166	0.190
12 months	0.01 (−0.03, 0.24)	0.134	0.179
18 months	−0.001 (−0.18, 0.18)	0.989	0.989

Note: This table combines the primary adjusted linear mixed‐effects model, multiplicity‐adjusted p‐values, standardized β coefficients, term‐level R² indices, and restricted cubic spline sensitivity analyses for continuous PAC total score.

Abbreviations: ADCS‐ADL, Alzheimer’s Disease Cooperative Study–Activities of Daily Living; BH‐FDR, Benjamini‐Hochberg false discovery rate; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; CI, confidence interval; LMM, linear mixed‐effects model; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; P90‐P10, predicted contrast comparing the 90th and 10th percentiles of PAC total score; PAC, Positive Aspects of Caregiving; q, BH‐FDR‐adjusted p value; QOL‐AD, Quality of Life in Alzheimer’s Disease; RCS, restricted cubic spline; R², coefficient of determination; Std β, standardized beta coefficient; TIME, follow‐up time; β, unstandardized regression coefficient from linear mixed‐effects models.

3.4. Exploratory dose–response and robustness analyses

Following the original median‐split strategy, PAC was first categorized into high‐ and low‐PAC strata using the sample median. In exploratory stratified mixed‐effects models before propensity‐score adjustment, compared with the low‐PAC group, the high‐PAC group showed higher MMSE (β = 2.48; 95% CI: 0.44–4.52; p = 0.018; q’s = 0.027), higher ADCS‐ADL (β = 5.24; 95% CI: 2.32–8.16; p = 0.001; q = 0.002), lower CDR‐SB (β = −2.30; 95% CI: −3.58 to −1.03; p < 0.001; q = 0.002), higher patient QOL‐AD (β = 4.19; 95% CI: 2.03–6.32; p < 0.001; q = 0.002), and lower NPI‐apathy (β = −0.79; 95% CI: −1.49 to −0.09; p = 0.027; q = 0.032). The NPI total score difference was smaller and its 95% CI crossed zero (β = −3.69; 95% CI: −8.02 to 0.65; p = 0.095; q = 0.095). Median‐split PAC‐by‐time interaction terms did not remain significant after correction (all interaction p ≥ 0.196; all q ≥ 0.502), suggesting that these high‐versus‐low average differences did not show a clearly different temporal pattern across follow‐up.

Because median‐split comparisons may be influenced by baseline differences between high‐ and low‐PAC dyads, propensity‐score analyses were then used as robustness checks for the grouped high‐versus‐low PAC results rather than as replacements for the primary continuous PAC models. Matching reduced many measured imbalances, but residual imbalance remained in this small real‐world cohort because high‐ and low‐PAC dyads differed across several baseline clinical and caregiver domains; disease duration, baseline apathy, baseline patient QOL‐AD, and caregiver age, therefore, still showed absolute standardized mean differences > 0.10 after matching. After 1:1 matching and application of stabilized IPTW, the high‐versus‐low PAC estimates were attenuated for MMSE (PSM: 0.31, 95% CI: −0.86 to 1.47; p = 0.599; IPTW: 0.63, 95% CI: −0.39 to 1.66; p = 0.224), ADCS‐ADL (PSM: −0.09, 95% CI: −1.33 to 1.15; p = 0.883; IPTW: −0.31, 95% CI: −1.47 to 0.84; p = 0.592), CDR‐SB (PSM: −0.18, 95% CI: −0.64 to 0.27; p = 0.423; IPTW: −0.36, 95% CI: −0.83 to 0.11; p = 0.131), NPI total (PSM: −0.60, 95% CI: −2.11 to 0.90; p = 0.427; IPTW: −3.25, 95% CI: −8.41 to 1.92; p = 0.216), and NPI‐apathy (PSM: −0.21, 95% CI: −0.61 to 0.19; p = 0.305; IPTW: −0.31, 95% CI: −0.67 to 0.05; p = 0.088). Patient QOL‐AD remained higher in the high‐PAC group after matching (difference = 1.25, 95% CI: 0.12–2.37; p = 0.031) and stabilized weighting (difference = 1.20, 95% CI: 0.09–2.32; p = 0.035).

PAC was subsequently categorized into tertiles for exploratory comparisons across ordered PAC strata, and continuous PAC was further examined using RCS models. In tertile‐defined mixed‐effects models, compared with the lowest PAC stratum, the highest stratum showed higher MMSE (β = 4.48, 95% CI: 1.89–7.07; p = 0.001; q = 0.002), higher ADCS‐ADL (β = 7.13, 95% CI: 3.33–10.93; p < 0.001; q = 0.002), lower CDR‐SB (β = −2.89, 95% CI: −4.55 to −1.22; p = 0.001; q = 0.002), and higher patient QOL‐AD (β = 6.30, 95% CI: 3.57–9.04; p < 0.001; q = 0.002). The NPI total score did not show a clear highest‐versus‐lowest tertile difference (β = −1.36, 95% CI: −7.07 to 4.35; p = 0.639; q = 0.639), and the apathy contrast did not remain significant after correction (β = −0.93, 95% CI: −1.84 to −0.17; p = 0.046; q = 0.055). Intermediate‐versus‐lowest tertile estimates were smaller, with corrected associations retained for MMSE (β = 3.57, 95% CI: 1.18–5.96; p = 0.004; q = 0.014), ADCS‐ADL (β = 4.85, 95% CI: 1.36–8.35; p = 0.007; q = 0.014), and CDR‐SB (β = −2.13, 95% CI: −3.67 to −0.60; p = 0.007; q = 0.014), whereas the highest‐versus‐intermediate contrast was retained only for patient QOL‐AD (β = 4.64, 95% CI: 1.52–7.75; p = 0.001; q = 0.006). RCS analyses provided no evidence of nonlinearity for any patient outcome by likelihood‐ratio testing (raw nonlinearity p‐values ranged from 0.108 to 0.973; all q’s ≥ 0.640). After multiplicity correction, the overall spline‐based PAC association by likelihood‐ratio testing remained significant only for patient QOL‐AD (raw p = 0.006; q = 0.036), with a model‐predicted 90th‐versus‐10th percentile contrast of 2.40 points (95% CI: 0.90–3.90; p = 0.002; q = 0.010).

The PSM/IPTW analyses relied on a grouped high‐versus‐low PAC contrast, and the tertile analyses further categorized continuous PAC into three strata; both approaches discard within‐group variation and may reduce statistical power compared with the primary continuous PAC analysis. Accordingly, the matched and weighted estimates were interpreted as sensitivity analyses used to assess robustness, with their precision considered when interpreting the findings. Full estimates are presented in Tables 2, 3 and S2A–S5D, and in Figures 3 and S1.

TABLE 3.

Summary of robustness analyses for primary PAC associations.

Outcome	Primary LMM BH‐FDR result	RCS overall PAC association	PSM high‐vs‐low PAC average effect	IPTW high‐vs‐low PAC average effect
MMSE	q = 0.002	q = 0.813	0.31 (−0.86, 1.47), p = 0.599	0.63 (−0.39, 1.66), p = 0.224
ADCS‐ADL	q = 0.002	q = 0.813	−0.09 (−1.33, 1.15), p = 0.883	−0.31 (−1.47, 0.84), p = 0.592
CDR‐SB	q = 0.002	q = 0.813	−0.18 (−0.64, 0.27), p = 0.423	−0.36 (−0.83, 0.11), p = 0.131
QOL‐AD (patient)	q = 0.002	q = 0.036	1.25 (0.12, 2.37), p = 0.031	1.20 (0.09, 2.32), p = 0.035
NPI‐apathy	q = 0.004	q = 0.182	−0.21 (−0.61, 0.19), p = 0.305	−0.31 (−0.67, 0.05), p = 0.088
NPI total	q = 0.386	q = 0.463	−0.60 (−2.11, 0.90), p = 0.427	−3.25 (−8.41, 1.92), p = 0.216

Note: This table summarizes how the primary continuous PAC findings compared with restricted cubic spline, propensity‐score matched, inverse probability weighted, and multiplicity‐adjusted sensitivity analyses.

PSM and IPTW estimates are presented as average effect (95% CI), p value.

Abbreviations: ADCS‐ADL, Alzheimer’s Disease Cooperative Study–Activities of Daily Living; BH‐FDR, Benjamini‐Hochberg false discovery rate; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; CI, confidence interval; IPTW, inverse probability of treatment weighting; LMM, linear mixed‐effects model; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; p, nominal p value; PSM, propensity‐score matching; q, BH‐FDR‐adjusted p value; QOL‐AD, Quality of Life in Alzheimer’s Disease; RCS, restricted cubic spline.

FIGURE 3.

Restricted cubic spline analyses of baseline caregiver positivity and longitudinal clinical outcomes. Note: Restricted cubic spline models were used to examine potential non‐linear associations between baseline PAC total score and model‐predicted outcomes during follow‐up. The solid line represents the estimated model‐predicted outcome, and the shaded area represents the 95% confidence interval. Higher scores indicate better performance for MMSE, ADCS‐ADL, and patient QOL‐AD, but worse status for CDR‐SB, NPI total, and NPI‐apathy. These analyses treated PAC as a continuous measure and were conducted to complement exploratory grouped analyses; they were not intended to define clinical thresholds. ADCS‐ADL, Alzheimer's Disease Cooperative Study–Activities of Daily Living; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; QOL‐AD, Quality of Life in Alzheimer's disease.

3.5. Exploratory dimension‐specific associations of PAC

In exploratory dimension‐specific models, higher self‐efficacy was associated with higher MMSE (β = 0.70, 95% CI: 0.41–0.99; p < 0.001; q = 0.002), higher ADCS‐ADL (β = 1.02, 95% CI: 0.59–1.45; p < 0.001; q = 0.002), lower CDR‐SB (β = −0.39, 95% CI: −0.57 to −0.19; p < 0.001; q = 0.002), lower apathy severity (β = −0.15, 95% CI: −0.25 to −0.05; p = 0.004; q = 0.005), and higher patient QOL‐AD (β = 0.82, 95% CI: 0.51–1.12; p < 0.001; q = 0.002), whereas self‐efficacy was not associated with NPI total (β = −0.15, 95% CI: −0.81 to 0.51; p = 0.655; q = 0.655). Higher outlook on life showed associations in the same directions for MMSE (β = 0.69, 95% CI: 0.31–1.07; p < 0.001; q = 0.002), ADCS‐ADL (β = 1.29, 95% CI: 0.75–1.83; p < 0.001; q = 0.002), CDR‐SB (β = −0.49, 95% CI: −0.74 to −0.25; p < 0.001; q = 0.002), apathy (β = −0.17, 95% CI: −0.31 to −0.04; p = 0.010; q = 0.012), and patient QOL‐AD (β = 0.97, 95% CI: 0.58–1.36; p = 0.002; q = 0.003), but not NPI total (β = −0.53, 95% CI: −1.37 to 0.29; p = 0.207; q = 0.207).

The absolute standardized β coefficients ranged from 0.305 to 0.462, whereas incremental marginal R² values ranged from 0.00% to 2.16%, indicating modest additional explanatory contributions. No dimension‐by‐time interaction remained significant after correction. In particular, the self‐efficacy‐by‐time interaction for ADCS‐ADL was attenuated after BH‐FDR correction (raw p = 0.012; q = 0.072), and the outlook‐on‐life‐by‐time interaction for patient QOL‐AD was not significant (raw p = 0.055; q = 0.330). Accordingly, the dimension‐specific results do not establish cumulative associations or distinct temporal patterns. Full estimates are presented in Table 4.

TABLE 4.

Integrated exploratory dimension‐specific associations of PAC with patient outcomes.

PAC dimension	Outcome/term	β (95% CI)	p	BH‐FDR q	Std β (95% CI)	Partial R 2	Incremental marginal R 2, %
Self‐efficacy	MMSE	0.70 (0.41, 0.99)	<0.001	0.002	0.420 (0.246, 0.594)	0.0001	0.01
	MMSE*TIME		0.365	0.596
	ADCS‐ADL	1.02 (0.59, 1.45)	<0.001	0.002	0.450 (0.260, 0.640)	0.0165	0.13
	ADCS‐ADL*TIME		0.012	0.072
	3 months	−0.21 (−0.46, 0.03)	0.088	0.088
	6 months	−0.29 (−0.52, −0.05)	0.019	0.036
	12 months	−0.46 (−0.74, −0.17)	0.002	0.008
	18 months	−0.43 (−0.80, −0.05)	0.027	0.036
	CDR‐SB	−0.39 (−0.57, −0.19)	<0.001	0.002	−0.335 (−0.489, −0.163)	0.0079	0.11
	CDR‐SB*TIME		0.136	0.408
	NPI
	NPI total	−0.15 (−0.81, 0.51)	0.655	0.655	−0.043 (−0.230, 0.145)	0.0133	0.61
	NPI total*TIME		0.954	0.954
	NPI‐apathy	−0.15 (−0.25, −0.05)	0.004	0.005	−0.332 (−0.553, −0.111)	0.0618	2.16
	NPI‐apathy*TIME		0.497	0.596
	QOL‐AD (patient)	0.82 (0.51, 1.12)	<0.001	0.002	0.433 (0.269, 0.591)	0.0483	0.71
	QOL‐AD (patient)*TIME		0.406	0.596
Outlook on life	MMSE	0.69 (0.31, 1.07)	<0.001	0.002	0.336 (0.151, 0.521)	0.0080	0.24
	MMSE*TIME		0.117	0.351
	ADCS‐ADL	1.29 (0.75, 1.83)	<0.001	0.002	0.462 (0.269, 0.656)	0.0014	0.01
	ADCS‐ADL*TIME		0.560	0.672
	CDR‐SB	−0.49 (−0.74, −0.25)	<0.001	0.002	−0.341 (−0.516, −0.174)	0.0004	0.00
	CDR‐SB*TIME		0.347	0.520
	NPI
	NPI total	−0.53 (−1.37, 0.29)	0.207	0.207	−0.122 (−0.315, 0.067)	0.0013	0.18
	NPI total*TIME		0.761	0.761
	NPI‐apathy	−0.17 (−0.31, −0.04)	0.010	0.012	−0.305 (−0.557, −0.072)	0.0094	0.34
	NPI‐apathy*TIME		0.293	0.520
	QOL‐AD (patient)	0.97 (0.58, 1.36)	0.002	0.003	0.416 (0.248, 0.583)	0.1113	1.51
	QOL‐AD (patient)*TIME		0.055	0.330
	3 months	0.33 (0.09, 0.57)	0.008	0.032
	6 months	0.23 (−0.02, 0.48)	0.068	0.091
	12 months	0.29 (0.01, 0.56)	0.042	0.084
	18 months	0.08 (−0.31, 0.46)	0.700	0.700

Note: This table combines the original dimension‐specific mixed‐effects estimates with multiplicity‐adjusted q values, standardized β coefficients, and term‐level R² indices for PAC self‐efficacy and outlook on life.

Abbreviations: ADCS‐ADL, Alzheimer’s Disease Cooperative Study–Activities of Daily Living; BH‐FDR, Benjamini‐Hochberg false discovery rate; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; CI, confidence interval; Incremental marginal R², increase in fixed‐effect explained variance after adding the corresponding PAC dimension term to the model, expressed as percentage; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; PAC, Positive Aspects of Caregiving; Partial R², variance uniquely attributable to the corresponding PAC dimension term after accounting for other fixed effects; q, BH‐FDR‐adjusted p value; QOL‐AD, Quality of Life in Alzheimer’s Disease; R², coefficient of determination; Std β, standardized regression coefficient after standardization of the exposure and outcome; TIME, follow‐up time; β, unstandardized regression coefficient from linear mixed‐effects models, representing the estimated change in outcome per 1‐point increase in the corresponding PAC dimension score.

3.6. Neuropsychiatric symptoms and caregiver outcomes: Exploratory association and mediation analyses

Given that higher PAC score was associated with lower patient apathy in the primary models, we further explored how neuropsychiatric symptoms were related to caregiver burden, caregiver QOL‐AD, and caregiving demands. Greater overall neuropsychiatric symptom burden was strongly associated with higher caregiver burden (correlation between NPI total and ZBI: r = 0.73; 95% CI: 0.71–0.74; p < 0.001, q = 0.001) and modestly inversely correlated with caregiver QOL‐AD (r = −0.25; 95% CI: −0.29 to −0.22; p < 0.001, q = 0.001), supporting a clinically coherent symptom–burden relationship.

Symptom‐specific mixed‐effects models further suggested heterogeneity across neuropsychiatric domains. Emotional elevation (β = 3.71; 95% CI: 1.69, 5.73; p < 0.001, q = 0.002), apathy (β = 0.79; 95% CI: 0.21–1.38; p = 0.008, q = 0.015), and aberrant motor behavior (β = 1.26; 95% CI: 0.55–1.96; p < 0.001, q = 0.002) were associated with higher ZBI scores. Apathy also showed a nominal inverse association with caregiver QOL‐AD (β = −0.27; 95% CI: −0.54 to −0.06, p = 0.049), although this did not remain significant after BH‐FDR correction (q = 0.069). The mean post‐versus‐pretreatment difference in daily caregiving time was 0.31 h/day (95% CI: −0.51 to −0.10; p = 0.003; q = 0.003), corresponding to a small standardized effect (Cohen's d = −0.27).

Formal mediation analyses at 3, 6, and 12 months did not demonstrate significant indirect effects through apathy. Therefore, apathy was interpreted as a candidate symptom‐level correlate associated with both PAC and caregiver burden, rather than as a confirmed mediator. Detailed estimates are shown in Table S6 and in Figures 4 and S2.

FIGURE 4.

Treatment‐associated reduction in daily caregiving time. Note: (A) Shows daily caregiving time before and after treatment. Points represent mean caregiving time, and vertical error bars indicate 95% confidence intervals. (B) Shows the within‐dyad change in daily caregiving time, calculated as post‐treatment minus pre‐treatment time. Each dot represents one patient–caregiver dyad, and the dotted horizontal line indicates no change. Overall, daily caregiving time decreased from 5.22 to 4.91 h/day after treatment, corresponding to a mean reduction of 0.31 h/day. The paired‐samples t‐test showed a statistically significant decrease in caregiving time after treatment (mean difference [post‐treatment − pre‐treatment] = −0.31 h/day, 95% CI: −0.51 to −0.10; t = −2.988, p = 0.003), with a small effect size (Cohen's d = −0.265, 95% CI: −0.442 to −0.088). CI, confidence interval.

3.7. Baseline characteristics associated with high versus low caregiver positivity: Identifying profiles of caregivers more likely to exhibit higher positivity

At baseline, the high‐PAC group had higher MMSE scores (23 [IQR 19, 24.5] vs 19 [IQR 15, 22]; p = 0.003; q = 0.015), higher patient QOL‐AD (34 [IQR 31, 39] vs 31 [IQR 26, 35]; p < 0.001; q = 0.010), and higher caregiver QOL‐AD (38 [IQR 33.25, 40] vs 35.5 [IQR 30, 39]; p = 0.011; q = 0.037). Daily caregiving time before and after treatment initiation was numerically lower in the high‐PAC group, but these differences did not remain significant after multiplicity correction (pre‐treatment: p = 0.042, q = 0.084; post‐treatment: p = 0.035, q = 0.084). Baseline ZBI, patient disease duration, APOE ε4 status, and the measured caregiver demographic characteristics did not differ after correction (raw p‐values for the remaining listed comparisons were ≥0.109).

In logistic regression models, each 1‐point higher baseline patient QOL‐AD score was associated with greater odds of membership in the high‐PAC total‐score and high‐self‐efficacy groups (both odds ratios [OR] = 1.142; 95% CI: 1.064–1.227; P < 0.001; q = 0.002), as well as the high‐outlook‐on‐life group (OR = 1.073; 95% CI: 1.009–1.142; p = 0.025; q = 0.025). Each 1‐point higher baseline caregiver QOL‐AD score was also associated with greater odds of membership in the high‐PAC total‐score group (OR = 1.099; 95% CI: 1.024–1.178; p = 0.009; q = 0.011) and the high‐self‐efficacy group (OR = 1.112; 95% CI: 1.034–1.196; p = 0.004; q = 0.007).

These baseline associations indicate that higher caregiver positivity co‐occurred with more favorable patient and caregiver quality‐of‐life profiles and better‐preserved cognition. They also underscore that PAC may partly reflect the pre‐existing clinical and caregiving context. These findings do not establish causal effects or individual‐level predictive utility. Full baseline comparisons and regression estimates are presented in Table S3A–C.

4. DISCUSSION

In this multicenter Chinese cohort, baseline PAC was associated with more favorable lecanemab‐related outcomes over 18 months, including cognition, daily function, and patient QOL‐AD outcomes. These graded associations suggest that psychological resources represent a meaningful psychosocial dimension of treatment‐related outcomes, not merely background.

4.1. Caregiver positivity as an independent caregiver‐related factor associated with cognition, daily function, and behavior

The primary continuous analyses showed that higher PAC was associated with more favorable MMSE, ADCS‐ADL, CDR‐SB, apathy, and patient QOL‐AD estimates after adjustment for measured demographic and clinical covariates. ¹⁷ Although these associations remained after multiplicity correction, the incremental variance explained by the PAC term was modest. In sensitivity analyses, PSM and IPTW attenuated the high‐versus‐low PAC differences in cognition, daily function, disease severity, and apathy. In contrast, the association with patient QOL‐AD remained statistically supported after both PSM (95% CI: 0.12–2.37; p = 0.031) and IPTW (95% CI: 0.09–2.32; p = 0.035), making patient‐reported quality of life the most consistent sensitivity‐analysis finding across analytical specifications. PAC may therefore capture aspects of the caregiving context related to perceived patient well‐being while also reflecting baseline disease status and dyadic circumstances. This interpretation accords with recent evidence that QOL‐AD captures early‐AD experience beyond cognition and function and declines with disease progression, supporting its value in patient‐centered therapeutic evaluation. ¹⁸ , ¹⁹

This interpretation is particularly relevant in the era of DMTs. Although lecanemab targets amyloid pathology, its implementation in routine practice depends on behavioral and environmental processes, ⁴ including attendance at infusion visits, safety monitoring, maintenance of daily routines, encouragement of activity, and timely responses to emerging symptoms. ²⁰ Greater caregiver positivity could plausibly facilitate these care processes; however, the present study did not directly measure treatment implementation or demonstrate an interaction with the pharmacologic effect of lecanemab. The findings should therefore be interpreted as longitudinal associations rather than evidence that PAC alters lecanemab efficacy.

4.2. Reframing caregivers: from burden recipients to participants in treatment implementation

The previous AD literature has predominantly conceptualized caregivers as individuals who experience substantial burden, psychological distress, and sleep disturbances. ²¹ , ²² , ²³ This perspective has generated important evidence on depression, caregiver strain, and unmet needs, but it does not fully capture the role of caregivers in treatment implementation. Our study extends this framework by positioning caregivers not only as affected individuals, but also as active participants shaping patient outcomes through daily support.

This conceptual shift broadens the traditional biomedical model toward a more integrated biopsychosocial framework. In AD, pathology unfolds within a patient–caregiver context in which caregiver attitudes, coping styles, and emotional resources may shape the daily implementation of treatment. ²⁴ By incorporating positive aspects of caregiving into therapeutic outcome research, our findings provide empirical support for the notion that successful AD management depends not only on biological targets and pharmacologic treatment, but also on caregiver engagement and psychosocial context. ²⁵

In practical terms, this means that optimizing outcomes with anti‐amyloid therapy may require attention not only to imaging biomarkers and infusion schedules, but also to caregiver readiness, confidence, and psychological well‐being. This caregiver‐inclusive perspective is consistent with the policy direction of China's national dementia initiatives, which emphasize caregiver training, effective patient–caregiver communication, emotional support, and the development of integrated service networks. ⁵ , ⁶ In this sense, PAC may provide one measurable dimension of the caregiving context, without serving as a criterion for treatment eligibility or an established predictor of response.

4.3. Exploratory relationships among neuropsychiatric symptoms, caregiver burden, and care demands

Exploratory analyses identified a clinically coherent pattern of association among BPSD, caregiver burden, and caregiver quality of life. Greater overall NPI burden strongly correlated with higher ZBI scores and modestly correlated with lower caregiver QOL‐AD. Emotional elevation, apathy, and aberrant motor behavior were associated with greater caregiver burden. As prior dementia caregiving studies have shown, neuropsychiatric symptoms are associated with higher caregiver burden, sleep disturbance, and affective distress across dementia syndromes. ²⁶ Although BPSD remains a major contributor to family stress and caregiver vulnerability, apathy may be particularly relevant in the present context. ²⁷ As an early‐emerging negative symptom characterized by diminished motivation and emotional disengagement, ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² apathy may impose a chronic, less‐visible caregiving demand and reduce reciprocal emotional feedback for caregivers. ³³ , ³⁴

We observed that higher PAC was associated with lower apathy, which was associated with caregiver burden and caregiver QOL; however, formal mediation was not demonstrated. Apathy should therefore be viewed as a candidate symptom‐level correlate rather than a confirmed pathway. ³⁵ , ³⁶ The mean post‐versus‐pretreatment difference in reported daily caregiving time was −0.31 h/day (95% CI: –0.51 to –0.10; p = 0.003; q = 0.003), corresponding to a small standardized effect. If the observed difference were sustained unchanged, it would amount to 168.25 h over 18 months. Given the substantial population affected by AD in China (≈12 million), even modest per‐dyad changes warrant prospective evaluation, but the present data do not support national or global extrapolation. ³⁷

Together, these findings extend the interpretation of lecanemab benefit beyond conventional cognitive endpoints and are consistent with prior evidence that DMT‐related benefits may also be reflected in patient and care‐partner quality‐of‐life outcomes. ³⁸ Nevertheless, the symptom‐burden‐care‐demand pattern observed here remains associative and hypothesis‐generating. Future studies should prospectively collect repeated neuropsychiatric, time‐use, informal care cost, and caregiver well‐being measures to determine whether these domains change together and whether targeted behavioral or psychosocial interventions provide additional benefit. ³⁹

4.4. Caregiver positivity as an indicator of dyadic care context

Baseline analyses showed that higher PAC co‐occurred with better preserved patient cognition and higher patient and caregiver QOL‐AD scores. Baseline patient QOL‐AD score was also consistently associated with higher PAC in logistic models. These findings suggest that PAC reflects not only caregiver psychological resources, but also the functional and emotional context in which caregiving occurs. ⁴⁰ The attenuation of most grouped estimates after propensity‐score adjustment may reflect both improved control of measured baseline imbalance and the information loss and reduced precision inherent in dichotomizing PAC, particularly in the smaller matched sample; it therefore tempers, rather than negates, the primary continuous‐score findings.

A bidirectional relationship is plausible. Earlier disease stages and more manageable care demands may help preserve positive caregiving appraisals, ⁴¹ whereas greater caregiver confidence and engagement may facilitate adherence to complex care routines. However, the present observational data cannot establish the direction of these relationships or determine whether PAC contributes to subsequent outcomes independently of all measured and unmeasured baseline factors. ⁴² Accordingly, PAC should not yet be considered a psychosocial enrichment marker for selecting patients for lecanemab or predicting durable treatment benefit. A more appropriate interpretation is that PAC may help characterize the dyadic care context and identify caregivers who may require additional support.

4.5. Integrating caregiver positivity into policy‐concordant dementia care pathways

These findings support consideration of routine PAC assessment in dementia care settings where lecanemab is prescribed. Although current precision medicine in AD primarily relies on amyloid and tau biomarkers, caregiver positivity may provide a complementary psychosocial dimension. ⁴³ One potential clinical application would be to incorporate PAC as one dimension of a broader caregiver assessment. In memory clinics, trained assessors, nurses, psychologists or other qualified clinical personnel may administer PAC at lecanemab initiation and during routine follow‐up, alongside measures of caregiver burden, caregiver quality of life, daily caregiving time, psychological distress, and social support. PAC assessment may help formally integrate caregivers into the treatment pathway and guide tailored education or support when PAC scores are low or declining.

This approach also aligns with China's national dementia policy priorities. The 2024–2030 National Action Plan emphasizes a full‐course care pathway spanning screening, diagnosis, rehabilitation, follow‐up, and care, while calling for expanded caregiver training, community support, and psychological assistance for family caregivers. ⁵ The 2023–2025 dementia prevention and treatment promotion action further specifies caregiver coaching in cognitive activation, exercise rehabilitation, daily care, emotional management, communication, stress relief, and confidence building. ⁴⁴ Within this framework, PAC may serve as a practical component of multidimensional caregiver assessment and help direct supportive resources without being used to determine pharmacologic eligibility.

Caregiver positivity may be modifiable. Interventions such as strengths‐based counseling, positive psychology exercises, coping‐skills coaching, peer support, and psychoeducation may enhance caregiver resilience and confidence. ⁴⁵ If future trials confirm causality, integrating PAC‐informed caregiver support into lecanemab care pathways could become a scalable adjunct for improving the real‐world implementation and sustainability of DMT. ⁴⁶ This is especially relevant in resource‐constrained healthcare systems, where low‐cost psychosocial strategies may provide incremental value alongside pharmacologic therapy. ⁴⁷

4.6. Limitations

Several limitations merit consideration. PAC was self‐reported, and the observational design precludes causal inference and cannot exclude residual confounding or reverse causality. Unequal visit‐specific sample sizes, particularly at 18 months, may have limited the power for longitudinal interaction analyses. Predictive performance was not assessed because prediction‐model development was beyond the scope of this association‐focused study. Future dedicated studies should expand the cohort, incorporate broader patient psychological and caregiver‐resource measures, and develop and externally validate prespecified prediction models.

5. CONCLUSIONS

In conclusion, higher baseline caregiver positivity was associated with more favorable longitudinal clinical and patient quality‐of‐life outcomes during lecanemab treatment, with patient QOL‐AD showing the most consistent signal across sensitivity analyses. Apathy was related to PAC and caregiver burden, but not confirmed as a mediator. PAC may complement caregiver assessment, whereas its predictive value requires validation in larger prospective cohorts.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no financial or non‐financial competing interests relevant to this work. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Tianjin Huanhu Hospital (approval number: JH‐ERB‐2021‐012). Written informed consent was obtained from all participants prior to their enrollment. For patients with cognitive impairment who lacked the capacity to provide consent directly, consent was obtained from their legally authorized representatives (spouses or adult children). All participants or their legal guardians were provided with detailed information about the study procedures, potential risks, and benefits before signing the consent form. A copy of the signed consent form has been retained in the study records.

Supporting information

Supporting Information: alz71894‐sup‐0001‐SuppMat.pdf

Supporting Information: alz71894‐sup‐0002‐ICMJE.pdf