Globally, approximately 50 million people are living with dementia, and the number is expected to reach 152.8 million by 2050 (1). Unfortunately, it has been found that a surprising number of individuals living with dementia do not receive any post-diagnosis support beyond the information they received at diagnosis (2). Indeed, in addition to the symptom of cognitive decline, individuals with dementia frequently manifest behavioral and psychological disturbances like agitation, depression, anxiety, apathy, and delusions, among others (3–5). It is estimated that 80–90% of patients with dementia develop behavioral and psychological symptoms of dementia (BPSD) during their illness (6). BPSD can have serious consequences, including prolonged hospitalization, increased health care costs and mortality (7–9), resulting in great suffering to patients as well as their families, putting a huge burden on society. The increased prevalence and incidence of BPSD underscores the need for effective treatment strategies. Therefore, it is imperative to explore effective pharmacological and non-pharmacological approaches to address BPSD.

Antipsychotics and anticonvulsants have demonstrated significant efficacy in the treatment of certain aspects of BPSD but are also associated with a higher rate of adverse events (10). In this context, pharmacological treatments are often used as a last resort in contrast to non-pharmacological therapies (11). Numerous non-pharmacological interventions have been used in managing BPSD, including physical exercise, cognitive training, music therapy, non-invasive brain stimulation, and acupuncture therapy, to name a few. A non-pharmacological approach could potentially reduce the severity of BPSD with comparable effectiveness to pharmacological treatment (12). Meanwhile, a recent study has shown that reminiscence, cognitive stimulation/rehabilitation, and other non-pharmacological approaches can reduce symptoms of depression in individuals with dementia (13). To date, most dementia management guidelines still recommend non-pharmacological intervention as the preferred first line of treatment. There is, however, a lack of consensus on the most effective non-pharmacological approach for BPSD, resulting in significant challenges in developing a definitive treatment protocol.

Network meta-analysis (NMA) is a statistical method that can synthesize direct and indirect evidence to rank different interventions from multiple trials in a network (14, 15). This type of analysis can provide more precise evidence for decision-makers to choose optimal treatment in clinical practice. To facilitate the selection of the most appropriate non-pharmacological treatment for BPSD patients, this study aims to perform an NMA that compares and ranks various non-pharmacological treatments for BPSD patients.

The purpose of this study is to evaluate the efficacy of non-pharmacological interventions in the treatment of BPSD and provide a clinical reference for health policy decision-makers.

This systematic review (SR) protocol has been registered on PROSPERO (CRD42022352095). The protocol will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocol (PRISMA-P) guidelines (16), according to the PRISMA-P checklist (Supplementary Appendix 1). SR will be conducted based on PRISMA-NMA and A Measure Tool to Assess Systematic Reviews-2 guidelines (17, 18). It is scheduled to begin on 1 October 2022 and end on 1 December 2022.

The study will not involve direct participant participation. Only randomized controlled trial (RCT) data from the databases will be used.

Studies published since inception to 1 October 2022 will be retrieved from the following databases: CNKI, WF, VIP Database, SinoMed, Web of Science (WOS), Embase, PubMed, PsycINFO, the CENTRAL, and AMED. To minimize publication bias, we will also retrieve data from clinical registries (WHO ICTRP, Clinical trials.gov, and ChiCTR). Furthermore, we will manually search the included papers’ references and similar published SRs for related studies.

Each database search strategy is shown in Supplementary Appendix 2. The following search terms will be used: (1) disease: dementia, behavioral and psychological symptoms of dementia, etc.; (2) non-pharmacological intervention: physical exercise, music therapy, reminiscence therapy, cognitive stimulation, etc.; and (3) study types: randomized controlled trials, or RCTs. Similar retrieval strategies will be used for other electronic databases. The terms will be used alone or in combination with “and” and “or.”

Two methodologically trained reviewers (Y-QL and Z-HY) will screen titles, keywords, and abstracts for relevance. Duplicate and ineligible studies will be excluded from the review process. All remaining studies will be examined for inclusion once the full text has been reviewed. In case of disagreements, a third reviewer (F-RL or L-XJ) will countercheck and arbitrate.

Two independent investigators (Y-QL and Z-HY) will extract data using a pre-designed extraction form including (1) identification information (i.e., year, country, and authors); (2) general information (i.e., sample size, study design, and allocation ratio); (3) participants (i.e., age, sex, race/ethnicity, trial inclusion and exclusion criteria, dementia type and severity, and comorbidity); (4) interventions and control; (5) outcomes; and (6) main results. We will contact respective authors concerning missing data, if any. Figure 1 depicts the entire study selection process.

Two assessors (Y-QL and Z-HY) will independently appraise the quality of the selected RCTs using the Risk of Bias Assessment Tool (Rob 2.0) from the Cochrane Collaboration (36). There are five types of bias covered by this tool: (1) the randomization process; (2) deviations from intended interventions; (3) missing outcomes data; (4) outcome measurement; and (5) selection of the reported results. The included trials will be rated and classified as low risk, high risk, or some concerns. If all domains were at low risk, we judged the study results to have an overall low risk of bias, some concerns if there were some concerns in any domain, and a high risk of bias if we judged any domain to be at high risk. A third reviewer (F-RL or L-XJ) will be consulted during the final decision-making process. Graphs will be generated using the Shiny app.¹

The comparison-adjusted funnel plot will be used if more than ten trials are included to visualize potential publication bias. Ideally, the data from these studies would be represented by a symmetrical inverted funnel plot. An asymmetrical funnel plot indicates the presence of publication bias.

Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system will be used to assess the quality of evidence (42). Evidence quality will be ranked as “high,” “moderate,” “low,” or “critically low.” It may be downgraded due to the risk of bias, inconsistency, indirectness, imprecision, and publication bias. Evidence may be upgraded due to a large magnitude of effect, a dose-response gradient, or attenuation by plausible confounding.

PROSPERO registration will include protocol amendments. We will document and publish any modifications to this protocol along with the results of the systematic review.