Comparative effectiveness of multiple interventions for Alzheimer’s disease on ABC syndromes and QoL: A Bayesian network meta-analysis

Summary

Evidence comparing directly pharmacological and non-pharmacological treatments for Alzheimer’s disease (AD) is scarce. The ABC symptoms—activities of daily living (ADLs) (A), behavioral and psychological symptoms (BPSs) (B), cognitive function (C), and quality of life (QoL)—are critical for diagnosing and assessing treatment effects in AD. This study aims to evaluate five interventions for AD: pharmacological therapy (PT), photobiomodulation (PBM), cognitive therapy (CT), exercise therapy (ET), and repetitive transcranial magnetic stimulation (rTMS) using a Bayesian network meta-analysis approach. Among 6,450 records screened, 91 randomized controlled trials (RCTs) involving 12,242 participants met the inclusion criteria. PBM and rTMS showed significant benefits for cognitive function in AD patients. PT and CT showed notable effectiveness in enhancing activities of daily living. For QoL, CT and ET were identified as more favorable outcomes. Collectively, each intervention exerts unique merits targeting ABC symptoms and QoL, implying combined pharmacological and non-pharmacological regimens could optimize therapeutic gains for AD management.

Graphical abstract

Highlights

• •A Bayesian network meta-analysis of 91 RCTs assessed five AD interventions
• •PBM and rTMS excelled at improving cognitive and behavioral symptoms
• •PTs and CT aided ADL, and cognitive/physical therapy (CT/ET) boosted QoL
• •Findings reveal therapy-specific benefits for tailored AD treatment strategies

Health sciences; Neurology; Clinical neuroscience

Introduction

Alzheimer’s disease (AD) is the predominant neurodegenerative disorder characterized by a progressive decline in cognitive function and behavioral health in the elderly.¹ The core clinical manifestations of AD can be summarized as the ABC syndrome, encompassing the declines in daily activities (activity, A) and cognition (C), alongside behavioral and psychiatric changes (behavior, B),^2,3,4,5,6 which subsequently affect the quality of life (QoL) of patients and their caregivers. Moreover, increasing research has unveiled alterations in the brain and body that can occur up to 20 years before the onset of AD symptoms,⁷ illustrating the interconnected nature of these changes and shedding light on the complex etiology and mechanisms of AD.⁸ As AD progresses, worsening ABC symptoms and declining QoL increase care dependency and impose substantial social and economic burdens on families and healthcare systems.⁹ Therefore, assessments of ABC symptoms alongside QoL are pivotal in ensuring early, accurate diagnosis, measuring the effectiveness of treatments, and providing comprehensive care for individuals with AD.

The advancements in AD biomarkers and the emergence of disease-modifying therapies have significantly transformed the treatment landscape for this condition. The anti-amyloid immunotherapy agent aducanumab, approved by the Food and Drug Administration (FDA), has been mired by controversy regarding its clinical efficacy.¹⁰ In contrast, newly launched lecanemab and donanemab have demonstrated efficacy in improving cognitive function among AD patients, warranting notable recognition.^11,12,13 These developments further highlight the necessity of early diagnosis and intervention in the context of disease-modifying treatment of AD. For patients with moderate and severe AD, symptomatic management continues to rely on antipsychotic medications, including cholinesterase inhibitors (ChEIs) and memantine, which remain the primary therapeutic options.¹⁴ Nevertheless, the overall assessment of efficacy for these FDA-approved drugs and those currently in clinical trials—particularly regarding their impact on ABC symptoms and QoL—remains uncertain.

In addition to pharmacological treatment, non-pharmacological interventions, such as structured physical exercise and cognitive stimulation, have been employed to improve cognitive impairment in patients with early-stage AD.¹⁵ These non-pharmacological approaches serve as effective supplements to pharmacological therapy (PT), helping to improve ABC symptoms and enhance the QoL for patients with AD. They are increasingly accepted by both patients and their families as viable options.^16,17,18,19 Additionally, meta-analysis reveals that a combination of pharmacological and non-PTs can effectively slow the disease progression, alleviate symptoms, and improve the overall QoL for individuals with AD.^20,21,22

According to the expert consensus on the comprehensive management of AD patients from the Academy of Cognitive Disorder of China (ACDC), an active and comprehensive management of AD clinical symptoms is essential. QoL, influenced by socio-economic and cultural factors, as well as personal values, reflects individuals’ physical, psychological, social, and overall well-being.²³ Therefore, the well-being and satisfaction of patients should play a crucial role in guiding treatment decisions. This highlights the necessity of incorporating QoL assessments when evaluating treatment outcomes for AD. It is recommended that a thorough assessment of both the ABC syndrome and QoL is essential for evaluating the therapeutic efficacy of AD treatments, irrespective of the specific interventions employed.

The development of precise and personalized treatment plans for AD is grounded in the new diagnostic framework for the condition.²⁴ The six-stage framework has significantly advanced the treatment of AD. However, there are still challenges, such as the patient acceptance of the treatment methods, potential side effects, and accessibility of treatments. In this study, we initially incorporated five categories of clinical intervention trials: PT, photobiomodulation (PBM), cognitive therapy (CT), exercise therapy (ET), and repetitive transcranial magnetic stimulation (rTMS). Our objective was to evaluate the effectiveness of these interventions through a Bayesian network meta-analysis of randomized controlled trials (RCTs) that examine ABC syndrome and QoL in patients with AD. By doing so, we aim to provide robust, evidence-based medical insights that will help clinicians select appropriate treatment modalities tailored to the specific conditions and needs of individuals with AD.

Results

Study characteristics

The literature screening process is illustrated in Figure 1. Initially, a total of 6,450 articles were identified through preliminary screening based on the predefined inclusion and exclusion criteria. Following this, duplicates and irrelevant articles were excluded based on their titles and abstracts, resulting in 451 articles being selected for a full review. After a comprehensive evaluation, 91 articles were ultimately included in this study.

Figure 1.

PRISMA Flow diagram of the study selection

RCTs, randomized controlled trials; PBM, photobiomodulation; rTMS, repetitive transcranial magnetic stimulation; PT, pharmacological therapy; ET, exercise therapy; CT, cognitive therapy.

This study encompassed all eligible RCTs conducted between 1996 and 2024, involving a total of 12,242 participants (6,456 in the intervention group and 5,786 in the control group). Detailed baseline characteristics of the participants are shown in Data S1. The mean age of the participants varied from 69.50 to 88.25 years, while the duration of the intervention ranged from 1 to 76 weeks, with an average duration of 18 weeks. The mean Mini-Mental State Examination (MMSE) scores across the included studies ranged from 4.70 to 26.90. In addition, a significant portion of the studies (46.4%) was conducted in countries located in Europe and the Americas, with approximately 66.5% of the participants being women.

Among the 91 articles included in the study, specific data reporting gaps were identified. Notably, three studies omitted the documentation of baseline cognitive function values of the subjects.^25,26,27 Another three studies failed to provide precise baseline age data.^28,29,30 Moreover, three studies did not specify the exact frequency of treatment.^31,32,33 Despite these omissions, all nine articles still met the inclusion criteria for the study.

Quality of evidence assessment

The quality of the studies included in our analysis was assessed using the Cochrane Collaboration Tool. Figure 2 presents an overview chart of the risk of bias 2.0 assessment, while Figure 3 provides details on the risk of bias assessment for individual studies. Among the 91 included studies, 2 (2.2%) were “high risk,” 52 (57.1%) were “low risk,” and the remaining 37 (40.7%) had “some concerns.”

Figure 2.

Summary of risk of bias 2.0 assessment of data included in the network meta-analysis

The Cochrane Collaboration tool was used to assess the quality of the included studies, including five aspects of the randomization process: deviation from intended interventions, missing outcome data, measurement of the outcome, randomization process, and selection of the reported results. Each item was assessed by “low risk,” “high risk,” and “some concerns.” The percentages represent the proportion of included randomized controlled trials classified into each risk-of-bias category.

Figure 3.

Risk of bias 2.0 assessment within the individual studies at outcome level

Each study was considered as one unit of analysis for risk-of-bias assessment. The results showed that 2 studies (2.2%) were “high risk,” 52 studies (57.1%) were “low risk,” and 37 studies (40.7%) were “some concerns.”

Additionally, the overall quality of evidence for each major outcome was appraised using the GRADE approach. As shown in Table S2, the certainty of evidence was rated as moderate for cognitive function and behavioral and psychological symptoms (BPSs), and low for activities of daily living (ADLs) and QoL. Downgrading was primarily attributed to inconsistency and imprecision, particularly for ADLs and QoL.

Heterogeneity analysis, inconsistency test, and publication bias

The included studies exhibited varying degrees of heterogeneity, ranging from mild to severe, as indicated in Table S3. An assessment of overall inconsistency was conducted to test for discrepancies between direct and indirect evidence from the studies. The results showed no significant inconsistencies between the direct and indirect comparisons of each intervention (p = 0.78). Additionally, analysis of the funnel plot revealed good symmetry, and Egger’s test found no significant publication bias across cognitive (p = 0.26), activity (p = 0.33), behavior (p = 0.49), and QoL (p = 0.67) domains (Figure 4), suggesting an absence of substantial publication bias in the included studies.

Figure 4.

Network funnel plot of included studies

(A-D) Cognitive function (A), activities of daily living (B), behavioral and psychological symptoms (C), and quality of life (D).

The funnel plot was used to evaluate publication bias. Each point represents an individual study effect estimate, and the horizontal and vertical axes indicate the treatment effect size and its corresponding standard error, respectively. Publication bias was considered unlikely when the funnel plot showed a visually symmetrical distribution. Potential asymmetry was further assessed using Egger’s regression test. A two-sided p value < 0.05 was considered statistically significant.

Pairwise meta-analysis

Prior to conducting the Bayesian network meta-analysis, conventional pairwise meta-analyses were performed to summarize the direct comparisons between interventions. Pooled estimates were calculated using a random-effects model and expressed as standardized mean differences (SMDs) with 95% confidence interval (CI). As shown in Figure S1, non-pharmacological interventions, including PBM, rTMS, CT, and ET, demonstrated superior effects compared with control conditions across cognitive, ADLs, BPSs, and QoL. Among primary outcomes, PBM demonstrated the better improvements in cognitive performance (SMD = 0.72, 95%CI: 0.29–1.12), followed by rTMS (SMD = 0.34, 95% CI: 0.05–0.73) and ET (SMD = 0.29, 95% CI: 0.10–0.71). PT was associated with a modest improvement in cognition (SMD = 0.17, 95% CI: 0.00–0.35), whereas CT showed no significant benefit compared with the control group (SMD = 0.09, 95% CI: −0.37 to 0.55).

Network meta-analysis

Cognitive function (C)

Figure 5A presents a detailed network diagram that illustrates the effects of five distinct interventions on cognitive function in patients with AD. Each type of intervention underwent at least one direct comparison with a control group. Simultaneously, CT and ET, PT and CT, and PT and rTMS each created a closed loop with the control group, showcasing clear evidence for incorporating both direct and indirect comparisons within this study. Among the 84 RCTs encompassing a total of 11,305 participants, it was found that PBM, CT, PT, and rTMS exhibited significant therapeutic effects on cognitive efficacy when compared with the control group, while no significant difference was observed between ET and the control group. The specific SMDs were as follows: PBM (SMD = 0.74, 95% credible interval [CrI]: 0.19–1.29), rTMS (SMD = 0.71, 95% CrI: 0.29–1.14), PT (SMD = 0.61, 95% CrI: 0.33–0.90), CT (SMD = 0.34, 95% CrI: 0.08–0.60), and ET (SMD = 0.35, 95% CrI: −0.03 to 0.74). Furthermore, Figure 6A showed that PBM (surface under the cumulative ranking curve [SUCRA] = 0.79) demonstrated a superior therapeutic effect on improving cognitive function (C) in AD patients compared with the other interventions, which were ranked accordingly: rTMS (SUCRA = 0.78), PT (SUCRA = 0.69), ET (SUCRA = 0.37), and CT (SUCRA = 0.34). However, the head-to-head comparisons illustrated in Figure 7A revealed no significant difference among the various intervention comparisons.

Figure 5.

Network plot of primary and secondary outcomes

(A–D) Cognitive function (A), activities of daily living (B), behavioral and psychological symptoms (C), and quality of life (D).

The size of the nodes corresponds to the number of participants randomized to each treatment type, and the node size represents the total number of participants included in the corresponding intervention category. Treatment type with direct comparisons is linked with a line, its thickness corresponds to the number of trials comparing every pair of treatments.

Figure 6.

Forest plots change in primary and secondary outcomes

(A) Effects of different interventions on cognitive function, corresponding to the serial numbers in Data S1 are as follows: PBM, 1–8; rTMS, 40–51 and 53; PT, 68, 71, 72, and 74–90; ET, 54–66; CT, 9–30, 32–34, 36, and 38.

(B) Effects of different interventions on activities of daily living included in the studies, corresponding to the serial numbers in Data S1 are as follows: PBM, 6; rTMs, 40 and 42–45; ET, 54 and 59; CT, 11, 12, 17, 23, 29, 30, and 33; PT, 68, 73, 75, 76, 78–80, 84–86, 88, and 89.

(C) Effects of different interventions on behavioral and psychological symptoms included in the studies, corresponding to the serial numbers in Data S1 are as follows: PBM, 2; rTMS, 45, 48, 48, 51, and 52; ET, 54, 57, 58, 62, and 65; PT, 72, 73, 75, 76, 78–80, 85, 87, and 88; CT, 17, 29–31, 34, and 38.

(D) Effects of different interventions on quality of life included in the studies, corresponding to the serial numbers in Data S1 are as follows: CT, 16–18, 22, 25, 26, 28, and 31; ET, 54, 57–58, 62, 66, and 67; PBM, 5.

Treatment effects are expressed as SMDs with 95% CrIs. The n value represents the number of participants included in the original randomized controlled trials contributing to each treatment comparison. Comparisons were considered statistically significant when the 95% CrI excluded zero; otherwise, differences were considered not statistically significant. The surface under the SUCRA values was used to rank the relative efficacy of each intervention. SMD, standardized mean difference; CrI, credible interval; IG, intervention group; CON, control group; PBM, photobiomodulation; rTMS, repetitive transcranial magnetic stimulation; PT, pharmacological therapy; ET, exercise therapy; CT, cognitive therapy.

Figure 7.

League table of different interventions

(A) Network meta-analysis of outcomes about cognitive function and quality of life.

(B) Network meta-analysis of outcomes about activities of daily living and behavioral and psychological symptoms.

All results are presented as SMD with 95% CrI. The results of the activities of daily living and cognitive function are shown in the lower left triangle, whereas results for behavioral and psychological symptom and quality of life are shown in the upper right triangle. The n value represents the number of participants included in the original randomized controlled trials contributing to each comparison. Cells shown in bold indicate significant results. CON, control group; PBM, photobiomodulation; rTMS, repetitive transcranial magnetic stimulation; PT, pharmacological therapy; ET, exercise therapy; CT, cognitive therapy; NA, not available; SMD, standardized mean difference; CrI, credible interval.

ADLs (A)

Figure 5B illustrates the impact of various interventions on the ADLs of patients with AD. A total of 28 articles involving 6,699 participants were included in this analysis. The results presented in Figure 6B indicated that both PT (SMD = 0.17, 95% CrI: 0.00–0.36) and CT (SMD = 0.18, 95% CrI: 0.08–0.46) exhibited more favorable therapeutic effects in improving ADL compared with the control group. In contrast, the other interventions, such as PBM (SMD = 0.92, 95% CrI: −0.30–2.14), rTMS (SMD = 0.34, 95% CrI: −0.55–0.72), or ET (SMD = 0.27, 95% CrI: −0.20–0.78), did not show significant improvements in ADL of AD patients. Furthermore, the head-to-head comparisons shown in Figure 7B revealed no significant differences in the improvement of ADL among the various interventions for patients with AD.

BPSs (B)

Next, the network plot and the efficacy of different interventions on BPSs were analyzed. Figure 5C displayed the network plot derived from 27 studies involving a total of 4,885 participants. The forest map presented in Figure 6C details the effectiveness of five interventions on BPS. Both PBM and rTMS interventions showed significant improvements compared to the control group, with PBM yielding an SMD of 2.00 (95% CrI: 0.38–2.60) and rTMS showing an SMD of 0.89 (95% CrI: 0.25–1.50). In contrast, other interventions, such as ET, PT, and CT, did not exhibit significant differences in BPS compared with the control group. Notably, a direct comparison highlighted the superiority of PBM over CT, with an SMD of 2.83 (95% CrI: 0.06–5.45), indicating a statistically significant greater effect of PBM on BPS, as shown in Figure 7B.

QoL

To evaluate the impact of different interventions on QoL in AD patients, the network plots based on 16 studies including 1,619 participants were assessed, as shown in Figure 5D. The analysis included data for only three of the five interventions in comparison with the control group regarding their effects on the QoL. The results indicated that CT (SMD = 0.59, 95% CrI: 0.15–1.10, SUCRA = 0.72) and ET (SMD = 0.52, 95% CrI: 0.04–1.00, SUCRA = 0.65) demonstrated significant improvements in QoL, as depicted in the forest plot. While PBM intervention showed an SMD of 0.47 (95% CrI: −0.84 to 1.70, SUCRA = 0.55), indicating no significant improvement in QoL, as shown in Figure 6D. Furthermore, the head-to-head comparisons revealed no significant differences among the various interventions, as illustrated in Figure 7A.

Meta-regression analysis

Network meta-regression results indicated that intervention duration and baseline MMSE score significantly moderated the treatment effects on cognitive outcomes, whereas mean age did not show a significant effect (Methods S2 and Figures S2). Specifically, longer intervention duration was positively associated with greater effect size (p = 0.005), as was a higher baseline MMSE score (p < 0.001). Mean age showed no significant association with treatment efficacy (p = 0.854).

Sensitivity analysis

The sensitivity analyses showed that the pooled estimates remained largely consistent after sequentially excluding studies identified as having a high risk of bias, those with small sample sizes, trials with exceptionally short or long intervention durations, and studies conducted in Chinese language. No significant changes were observed in either the direction or magnitude of the effect estimates. This suggests that the overall findings were robust and not unduly influenced by any single study or subgroup (Methods S2 and Figures S3–S6).

Discussion

The Bayesian network meta-analysis employed in our study presents several advantages. It effectively integrates data from diverse sources, allowing flexible modeling that accommodates various study designs. By harnessing posterior probabilities, this approach provides more accurate and scientifically rigorous estimates compared with traditional frequentist methods.^34,35 Moreover, the combination of direct and indirect evidence through our analysis offers insightful comparisons and rankings of the effectiveness of different interventions in addressing ABC syndrome and QoL in AD patients. This comparative evaluation is quantitatively expressed through measures such as relative effect sizes (SMDs) and cumulative ranking curves (SUCRA), thereby providing innovative avenues for selecting optimal treatment strategies for AD. However, it is crucial to approach the cumulative probability rankings generated by Bayesian network meta-analysis with skepticism. While these rankings offer valuable insights into the relative performance of interventions, they may not always denote significant differences in treatment effects. For instance, although PBM, rTMS, and ET are ranked as the top interventions for improving daily living capabilities according to the SUCRA analysis, the differences in efficacy among these treatments are not statistically substantial. Therefore, while SUCRA rankings can be useful for assessing treatment options, they should be interpreted carefully, taking into account both their statistical and clinical significance.

In our investigation of treatments for AD, we examined both pharmacological and non-pharmacological approaches, categorizing them into five principal interventions for meta-analysis: PBM, CT, ET, rTMS, and PT. Currently, the drugs approved by the FDA for AD primarily target cognitive and behavioral symptoms (symptomatic treatment) or the underlying biological processes of the disease (disease-modifying treatment). The former mainly includes ChEIs , such as donepezil and galantamine,³⁶ along with NMDA receptor antagonists like the representative drug memantine.^37,38 In contrast, disease-modifying drugs primarily target β-amyloid proteins and include aducanumab, lecanemab, and donanemab. Notably, aducanumab was the first of these drugs to receive FDA approval in 2021 and has undergone extensive research.^39,40 Additionally, GV971 targets the gut-brain axis and is the only AD treatment that has been approved for market use, representing a shift in the treatment paradigm for AD, particularly for sporadic AD.⁴¹ Given this context, our study selects four representative AD treatment drugs—donepezil, memantine, aducanumab, and GV971—to comprehensively evaluate their therapeutic effects from the perspectives of ABC syndrome and QoL.

The interventions of PBM, ET, and rTMS are relatively singular and rigid, whereas CT offers a wider variety of options.^{42,43,44,45,46} Initially, our exploration of non-pharmacological measures, particularly CT, was based on methodologies established in our earlier research.⁴⁷ This process involved considering various forms of CT as distinct entities, in addition to analyzing the four medication-based treatments separately. However, as our review progressed, involving meticulous literature filtering and data collection, it became apparent that only a few studies met our inclusion standards to assess ABC syndrome and the QoL thoroughly. Moreover, the multifaceted nature of the outcome metrics identified increasingly complicated the interpretability of our findings. This complexity persisted despite attempts to clarify insights through regression analyses and subgroup evaluations, ultimately leading to concerns regarding the reliability of our conclusions. Consequently, to address these challenges and enhance the coherence of our analysis, we opted to consolidate the diverse CT approaches into a single category. Specifically, we categorized CT into three types: cognitive training, cognitive stimulation, and cognitive rehabilitation. The literature reviewed in this study encompasses music therapy, enriched environmental therapy, individualized cognitive stimulation, learning and memory training, computerized cognitive training, multimedia cognitive stimulation, and virtual reality (VR) cognitive stimulation. Regarding PT, the four pharmacological treatments are similarly amalgamated as CT intervention. We selected two types of drugs, “improving cognitive symptoms” and “disease-modifying therapy,” and finally incorporated four representative drugs: ChEIs Donepezil, the NMDA receptor antagonist Memantine, Aducanumab (Aβ-monoclonal antibody), and GV-971, a mediator of the cerebro-intestinal axis. Although not exhaustive, the study encompassed classical and widely used drugs, offering a comprehensive representation of the current state of drug therapy for AD in practice.³⁴ We assert that this consolidation approach is methodologically sound and serves the purpose of our study effectively.

It is well recognized that evaluating ABC symptoms contributes significantly to the accurate diagnosis, comprehensive treatment, and prognosis prediction of AD. Moreover, QoL has become an increasingly important focus in dementia research, representing the most effective metric for assessing therapeutic effects in individuals with AD.^48,49 With this consideration, we systematically assessed the effectiveness of the five types (PBM, CT, ET, rTMS, and PT) of pharmacological or non-pharmacological interventions via assessment of ABC symptoms and QoL in AD patients. Our findings suggest that each of the five interventions significantly enhances cognitive function (C) in individuals with AD, with PBM exhibiting the most pronounced improvement, followed by rTMS. Additionally, PT and CT are found to be more effective in improving the ADLs (A). Moreover, PBM and rTMS significantly mitigate BPSs (B). In terms of improving QoL for AD patients, CT and ET exhibited superior outcomes. Notably, our network meta-regression analyses revealed that longer intervention durations and higher baseline MMSE scores were associated with greater treatment effects. This suggests that both the timing and the total duration of treatment are critical moderators of intervention efficacy, potentially explaining the variability observed across studies. In the treatment of AD, pharmacological treatment, particularly ChEIs, may cause gastrointestinal or cardiovascular AEs,⁵⁰ whereas memantine is typically well tolerated.^51,52 Treatment with monoclonal antibodies such as aducanumab requires MRI monitoring to detect imaging abnormalities related to amyloid protein, brain edema, and hippocampal atrophy. In contrast, non-pharmacological interventions are widely utilized, and existing evidence indicates that these measures are generally safe and well-accepted by AD patients.^53,54 Although reports of adverse events (AEs) are inconsistent in various trials, such as rTMS and PBM, they typically only involve mild and transient discomfort.^18,55 Additionally, structured exercise and cognitive training rarely lead to intervention-related AEs. Therefore, it should be weighed the safety and tolerability of different methods when selecting an appropriate treatment for AD.

The pronounced efficacy of PBM on ABC symptoms is closely linked to its molecular mechanism, which utilizes the biological effects of visible and near-infrared light within the spectrum. PBM initiates biological responses, drives neuroprotective actions, and stimulates neurogenesis, as evidenced by multiple studies.^56,57 It has been demonstrated that PBM could improve cognitive functions and alleviate anger, anxiety, and wandering symptoms among AD patients.^16,58 As a non-invasive neuroregulatory tool, rTMS has been found to enhance neural microcirculation, notably by reducing Aβ production and its clearance.⁵⁹ Additionally, it plays a role in neuroprotection by influencing neuronal apoptosis and regeneration.⁶⁰ The function of rTMS in regulating the level of neurotransmitters is closely related to the improvement of psychological and behavioral symptoms.^61,62 Regarding CT for AD, the traditional methodologies include, but are not limited to, music therapy, reminiscence therapy, creative storytelling simulated scene tasks, etc. In recent years, advancements in technology and research have paved the way for innovative approaches to CT for AD. Noteworthy among these is computerized cognitive training, which employs software and digital platforms to deliver structured cognitive exercises to enhance cognitive functions. This method has gained traction, evidenced by its exploration in various studies,^44,45,46 highlighting its effectiveness and potential benefits for individuals with AD. Another cutting-edge approach is VR-based cognitive stimulation, which utilizes VR technology to create immersive environments for cognitive training and stimulation.⁴² PT, targeting specific symptoms of AD, effectively improves daily living abilities, matching treatment goals. Multiple meta-analyses support PT benefits on cognitive and mental functions.^21,34,35 However, the combination of drugs targeting multiple symptoms requires careful consideration of potential side effects. But the practical application should consider the side effects caused by multi-target drug combinations.

In terms of improving the QoL of patients, CT and ET showed a better role. It is worth noting that the effect of cognitive nursing in improving the QoL of patients with AD should not be ignored. Giving patients respect and careful care can make them relaxed and confident, and face life positively, thus improving the symptoms of dementia and improving the QoL.^63,64,65 Exercise itself can make people feel happy. In terms of ET for AD patients, many systematic reviews and meta-analyses have shown that exercise is beneficial for improving the overall cognition of the elderly.^66,67,68 In AD patients who are unable or unwilling to accept drug therapy, CT, and ET are ideal options for improving cognition and QoL.

Limitations of the study

This study has several notable strengths. It comprehensively integrated both pharmacological and non-pharmacological interventions within a Bayesian network meta-analysis, enabling indirect comparisons across diverse treatment modalities. The multidimensional evaluation framework based on the ABC syndrome and QoL provided a clinically meaningful and holistic perspective on AD management. Furthermore, rigorous quality assessment using ROB 2.0 and GRADE strengthened the methodological reliability of our findings.

Despite these advantages, several limitations should be acknowledged. First, the evidence base was limited for some intervention nodes. In particular, PBM was supported by only one RCT for several outcomes, with a relatively small sample size for ADLs, BPSs, and QoL. Therefore, estimates for these outcomes should be interpreted cautiously. Second, there was substantial clinical and methodological heterogeneity among the included studies, stemming from differences in intervention protocols and assessment tools. However, network meta-regression analyses demonstrated that intervention duration and baseline MMSE scores were significantly associated with between-study heterogeneity, suggesting that the clinical context and treatment intensity are critical factors in determining intervention success. Third, similar interventions were combined into broader treatment categories to enable network comparisons, although differences in their specific components, delivery methods, and mechanisms of action may have introduced within-category heterogeneity and limited the precision of treatment-specific estimates. Finally, the lack of caregiver-reported outcomes is an important evidence gap because potential reductions in caregiver burden related to improvements in patient BPS could not be evaluated.

Future studies should adopt standardized intervention durations and protocols to better capture the therapeutic potential of each modality. Larger, multicenter RCTs with extended follow-up periods are warranted to verify the durability and generalizability of the findings. For subsequent meta-analyses, assuming sufficient data become available, subgroup and dose-response analyses adjusting for treatment intensity, frequency, and duration will be essential to enhance the precision and clinical interpretability of future results.

Conclusion

Different interventions offer distinct clinical advantages for patients with AD across cognition, ADLs, behavioral symptoms, and QoL. These findings emphasize the importance of individualized multi-domain treatment strategies and support the integration of pharmacological and non-pharmacological approaches to optimize patient-centered care. However, due to the limited evidence base for certain interventions and the heterogeneity across studies, their comparative rankings should be considered preliminary and suggestive. Further large-scale, standardized RCTs are necessitated to confirm these findings and refine clinical guidelines.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Ruimin Wang (ruimin-wang@ncst.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •This study is a meta-analysis based on randomized controlled trials. The data analyzed in this study were all derived from previously published research and publicly available resources.
• •No custom code was generated; all analyses were performed using publicly available software as detailed in the key resources table.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We would like to thank Dr. Darrell Brann (Augusta University) for his critical reading and proofreading of the manuscript. This study was supported by the National Natural Science Foundation of China (82471467 to R.W.), the Central Guiding Local Technology Development (246Z7717G to R.W.), the Youth Talent Promotion Program of School of Public Health, North China University of Science and Technology (20237591 to F.G.), the Ministry of Education Industry-university Cooperative Education Projects (22107139135230 to X.Q.) and the Tangshan Science and Technology Program Projects (21130236C to J.B.).

Author contributions

R.W. played a guiding role in the manuscript writing, revision, and data analysis. F.G. contributed to the conception or design of the work, the writing, and revision of the manuscript, and data processing. G.L. contributed to the data analysis and processing. B.L., X.Q., and J.B. contributed to the literature screening and data analysis. All authors contributed to the article and approved the submitted version.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
China National Knowledge Infrastructure (CNKI)	CNKI	https://www.cnki.net
WanFang Database	WanFang Data	https://www.wanfangdata.com.cn
VIP Database	Chongqing VIP Information Co., Ltd.	https://www.cqvip.com
PubMed	National Library of Medicine (NLM)	https://pubmed.ncbi.nlm.nih.gov
MEDLINE	National Library of Medicine (NLM)	https://www.nlm.nih.gov
Embase	Elsevier	https://www.embase.com
Cochrane Central Register of Controlled Trials (CENTRAL)	Cochrane Library	https://www.cochranelibrary.com
Web of Science Core Collection	Clarivate Analytics	https://www.webofscience.com
PROSPERO Registration Record	International Prospective Register of Systematic Reviews	CRD42024581072
Software and algorithms
EndNote ×8	Clarivate Analytics	Version ×8
Risk of Bias 2.0 (ROB 2.0)	Cochrane Collaboration	https://www.riskofbias.info
GRADE framework	GRADE Working Group	https://www.gradeworkinggroup.org
Stata Statistical Software	StataCorp LLC	Version 16.0
R Statistical Software	R Foundation for Statistical Computing	Version 3.5.1
gemtc package	R package	https://cran.r-project.org/package=gemtc
netmeta package	R package	https://cran.r-project.org/package=netmeta
forestplot package	R package	https://cran.r-project.org/package=forestplot
Other
PRISMA Statement	Liberati et al.69	https://doi.org/10.1371/journal.pmed.1000100
PRISMA-NMA Extension	Hutton et al.70	https://doi.org/10.7326/M14-2385
Cochrane Handbook for Systematic Reviews of Interventions	Cochrane Collaboration	Version 6.3
Brooks–Gelman–Rubin Diagnostic Method	Brooks and Gelman, 1998	Statistical convergence diagnostic

Experimental model and study participant details

This study did not conduct any original experiments involving animals, human subjects, plants, microbial strains, cell lines or primary cell cultures. Therefore, no experimental models were established, and no research participants were directly recruited for this study.

Method details

We conducted a rigorous systematic review and network meta-analysis in accordance with the Cochrane Handbook for Systematic Reviews of Interventions version 6.3 and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.⁶⁹ This study has been registered with PROSPERO (CRD42024581072) without any amendments to the provided information.

Literature search strategies

We systematically searched three Chinese databases (China National Knowledge Infrastructure Database, Wan Fang Database, and VIP Database) and five English databases (PubMed, MEDLINE, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), and Web of Science) for all relevant studies published from their inception until 30 December 2024, without imposing language restrictions. We used the search term “Alzheimer’s disease” in combination with the Medical Subject Headings (MeSH) for five interventions: “photobiomodulation”, “cognitive therapy”, “exercise therapy”, “repetitive transcranial magnetic stimulation”, and “pharmacological therapies”, alongside “randomized controlled trials” to identify relevant literature. To ensure comprehensiveness and reduce the possibility of missing crucial studies, we also conducted a supplementary manual literature search, which included sourcing documents from review articles and meta-analyses within this field. Detailed search strategies for all databases are provided in Methods S1.

Eligibility criteria

The included RCTs conformed to the PICOS (Population, Intervention, Comparison, Outcome, Study design) principles established by the Cochrane Collaboration.⁷⁰ Firstly, the studies incorporated subjects who were older than 65 years, as identified by the National Institute on Aging and Alzheimer’s Association (NIA-AA) guidelines and the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR). Patients were diagnosed with AD based on the criteria from the Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA). Furthermore, all participating patients were confirmed to have no other primary or secondary diseases. Secondly, for the intervention options, the study included PBM, CT, ET, rTMS, and PT. CT encompassed cognitive training, cognitive stimulation, and cognitive rehabilitation. Meanwhile, PT covered four key drugs: the ChEIs such as donepezil, the N-methyl-D-aspartate (NMDA) antagonist receptor like memantine, the Aβ monoclonal antibodies aducanumab, and the gut microbiota-balancing drugs including GV-971. A brief description of the pharmacological and non-pharmacological interventions (PBM, rTMS, CT, ET, and PT) included in the analysis is provided in Table S1. Thirdly, the comparators consisted of either standard care or placebo; in “head-to-head” studies, any assessed interventions (such as PBM, CT, ET, rTMS, and PT) could serve as comparators. Fourth, the research was required to integrate at least one pertinent outcome measure. The principal outcome focused on cognitive changes in Alzheimer’s patients, gauged through evaluations like the Mini-Mental State Examination (MMSE), the Montreal Cognitive Assessment (MOCA), and the Alzheimer’s Disease Assessment Scale (ADAS-cog). Secondary outcomes included improvements in daily activities, psychiatric symptoms, behaviors, and overall quality of life. These were measured using the Activities of Daily Living (ADL), Neuropsychiatric Inventory (NPI), and the Quality of Life in Alzheimer’s Disease (QoL-AD) scale.

The following exclusion criteria were applied: subjects with cognitive impairment caused by other related diseases, the use of various interventions in combination, studies with missing or incomplete data, conference papers or abstracts without full texts, studies lacking clear inclusion and exclusion criteria, and duplicate published literature.

Data extraction and quality assessment

The literature was organized using EndNote X 8 reference manager for efficient induction and sorting. Initially, screening articles involved reviewing titles and abstracts to eliminate clearly irrelevant studies. Subsequently, the full texts of the remaining articles were examined in detail. Any articles inaccessible in full text or that failed to meet the inclusion criteria were excluded. A standardized data extraction form was employed by data extractors to collect information from the selected literature, which included details such as the first author, publication year, country, study subjects’ type, age, sample size, male-to-female ratio, dropout rate, type of intervention, duration, frequency, and outcome measures including Mean difference (MD) and Standard Deviation (SD). Throughout the screening process, two researchers independently reviewed the articles, extracted data, and cross-verified the information. In instances of disagreement, a third researcher was brought in to facilitate a resolution.

This study utilized the Risk of Bias 2.0 (ROB 2.0) tool,⁷¹ as recommended by the Cochrane Handbook for Systematic Reviews of Interventions (2019), to evaluate the bias risk in RCTs. The assessment focused on five key aspects of the randomization process: Deviations from the intended intervention, Missing outcome data, Measurement of the outcome, the Randomization process, and Selection of the reported results. Each aspect was judged as being at “High risk”, “Low risk”, or having “Some concerns”. In cases where evaluators disagreed on a study’s assessment, a consensus was sought through discussion with a third investigator. Additionally, the overall certainty of evidence for each major outcome (Cognition, Activity, Behavior, Quality of life) was evaluated using the GRADE approach, considering risk of bias, inconsistency, indirectness, imprecision, and publication bias.⁷²

Statistical analysis

To ascertain the similarity among the included studies, a balance test was utilized.⁷³ Further, heterogeneity between studies was measured using the I² statistic, which ranges from 0 to 100%, with higher values indicating greater heterogeneity. Heterogeneity was categorized into mild (25%), moderate (50%), and high (75%) levels.⁷⁴ Additionally, the overall inconsistency was employed to examine the discrepancies between direct and indirect evidence for each intervention, generally represented as p. When p > 0.05, the consistency model was adopted, suggesting no significant inconsistencies.⁷⁵

Within the R statistical environment (version 3.5.1), the “gemtc” and “netmeta” packages were applied for conducting network meta-analysis, while the “forestplot” package was utilized for creating forest plots. Conventional pairwise meta-analyses were first conducted to summarize the direct evidence between interventions. Random-effects models were applied to calculate pooled standardized mean differences (SMDs) with 95% confidence intervals (CIs) for each comparison. Subsequently, the Bayesian hierarchical random-effects framework was implemented using the gemtc package, which applies Markov Chain Monte Carlo (MCMC) to integrate direct and indirect data between the intervention and control groups.⁷⁶ Model convergence was assessed using the Brooks - Gelman - Rubin diagnostic (R-hat < 1.05) to ensure stability of the posterior estimates. As the outcome measures in this study were continuous variables, network estimates for all outcome measures were presented as SMDs with 95% credible intervals (CrI).⁷⁷ For handling missing data, such as absent means, standard deviations, or reported only p values, calculations were conducted using the Cochrane Collaboration’s method outlined in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 to derive SMDs and uncertainty estimates. p < 0.05 was considered indicative of statistical significance.

STATA 16.0 was employed to generate a network plot that visualizes the network of direct versus indirect comparisons among all interventions. In the network plot, each node (dot) corresponds to an intervention, with the size of the dot reflecting the sample size. Lines connecting the dots represent the direct comparisons between two interventions, where the thickness of the line indicates the number of studies comparing those interventions. Furthermore, to assess publication bias, funnel plots were created using Stata 16.0.⁷⁸ The risk of publication bias was also quantitatively examined using Egger’s linear regression test,^79,80 which assesses the symmetry of funnel plots. Ranking indicators were demonstrated using SUCRA (Surface under the cumulative ranking curve) values, which range from 0 to 1-higher values suggest potentially better outcomes.⁸¹

In addition, we performed network meta-regression to explore potential sources of heterogeneity and evaluate the robustness of the treatment effects for the primary outcome, using intervention duration, mean age, and baseline MMSE score as study-level covariates. To evaluate the robustness of the pooled estimates, we conducted several sensitivity analyses by: (1) excluding studies rated as having a high risk of bias according to the ROB 2.0 assessment tool; (2) excluding studies with small sample sizes (<20 participants per group); (3) excluding studies with extremely short or long intervention durations (<8 weeks or >48 weeks); and (4) restricting the analysis to English-language publications by excluding Chinese-language studies. The stability of the findings was assessed by comparing the consistency of results before and after each exclusion.

Quantification and statistical analysis

All statistical analyses were performed using R software (version 3.5.1) and Stata software (version 16.0). Pairwise meta-analyses and Bayesian network meta-analyses were conducted to evaluate the comparative effects of interventions on ABC syndromes and QoL. Statistical analyses included random-effects models, heterogeneity assessment using the I² statistic, node-splitting analysis for network inconsistency, Brooks–Gelman–Rubin diagnostic for Bayesian model convergence, SUCRA ranking, sensitivity analyses, and publication bias assessment using funnel plots and Egger’s test.

The sample size (n) represents the number of participants included in the original randomized controlled trials. The exact numbers of studies and participants for each outcome are reported in the results section, Supplementary Tables, and figure legends. Cognitive function analyses included 84 trials (11,305 participants), activities of daily living analyses included 28 trials (6,699 participants), behavioral and psychological symptom analyses included 27 trials (4,885 participants), and quality-of-life analyses included 16 trials (1,619 participants).Treatment effects were summarized as SMDs with 95% CIs or 95% CrIs. Statistical significance was defined as p < 0.05 for frequentist analyses, whereas Bayesian comparisons were considered statistically significant when the 95% CrI excluded zero.

Detailed statistical information, including statistical models, effect estimates, sample sizes, uncertainty measures, network plots, treatment rankings, heterogeneity assessments, risk-of-bias evaluations, and sensitivity analyses, is provided in the results section, Figures 1, 2, 3, 4, 5, 6, 7, and S1–S6, Data S1, Tables S1–S3, and Methods S1 and S2.