Treatment modifiers and predictors of risperidone response in dementia: An individual participant data meta‐analysis of six randomized controlled trials

Abstract

INTRODUCTION

Risperidone is approved for behaviors and psychological symptoms of dementia (BPSD), despite modest efficacy and known risks. Identifying responsive symptoms, treatment modifiers, and predictors is crucial for personalized treatment.

METHOD

A one‐stage individual participant data meta‐analysis of six randomized controlled trials (risperidone: n = 1009; placebo: N = 712) was conducted. Mixed‐effects models assessed treatment effects, modifiers, and predictors, with BPSD measured via the Behavioral Pathology in Alzheimer's Disease scale.

RESULTS

Risperidone showed modest 8 week benefits for aggression (standardized mean difference [SMD]: −0.22; p < 0.001), psychosis (SMD: −0.23; p = 0.001), and anxiety/phobias (SMD: −0.19; p = 0.014), but not for activity, affective, or sleep disturbances. Pharmacokinetic/pharmacodynamic‐related factors (e.g., body mass index, endocrine disease, race/ethnicity) potentially modified treatment effects. Week 2 response predicted week 8 improvement (odds ratio: 4.46; p < 0.001).

DISCUSSION

Risperidone provided symptom‐specific benefits in reducing aggression, psychosis, and anxiety/phobias. Week 2 response predicted treatment outcomes, while certain patient characteristics may modify treatment response. Further research is needed to optimize the benefit–risk balance and individualize treatment.

Highlights

• Risperidone modestly reduces symptoms of psychosis, aggression, and anxiety/phobias.

• Risperidone shows no effect on activity, affective, or sleep disturbances.

• Patient factors (body mass index, endocrine disease, race/ethnicity) may affect response.

• Positive response by week 2 predicts significant improvement later.

1. INTRODUCTION

There are ≈ 57 million people living with dementia, ¹ and this number is projected to increase to nearly 158 million cases in 2050. ² Throughout the course of dementia, people often exhibit changed behaviors associated with cognitive decline, commonly referred to as behaviors and psychological symptoms of dementia (BPSD), or sometimes described as neuropsychiatric symptoms or responsive behaviors. ³ , ⁴ These include manifestations such as psychosis, aggression, agitation, anxiety, depression, changes in sleep or appetite, and apathy. ³ BPSD are prevalent among people with dementia, with estimates reaching 32% in community‐dwelling populations ⁵ and up to 90% in hospitalized patients and nursing home residents. ⁶ , ⁷

Antipsychotics are often used to manage BPSD in people living with dementia, with the prevalence reaching up to 37.5%. ⁸ , ⁹ These symptoms vary widely between individuals, and current evidence suggests that antipsychotics offer only modest benefits in treating psychosis and agitation in this population. ¹⁰ , ¹¹ , ¹² Meanwhile, these medications can increase the risk of adverse events, including cardiovascular disease, stroke, and mortality in older people. ¹⁰ , ¹¹ , ¹³ In light of these risks, antipsychotics should only be initiated for severe symptoms or cases in which non‐pharmacological approaches prove ineffective. ⁸ , ¹⁰ , ¹³ Therefore, identifying treatment modifiers, which are patient‐specific characteristics that define subgroups more likely to experience different benefits or risks from medication, can facilitate more targeted and effective treatment strategies. ¹⁴ , ¹⁵ Additionally, exploring predictors of therapeutic success can help clinicians personalize treatment decisions, guiding whether to continue the current therapy or consider alternative approaches.

Individual participant data meta‐analysis (IPD‐MA) currently offers the most robust evidence for detecting subgroup effects because it can standardize data across studies and prevent ecological bias common in aggregate analyses. ¹⁶ , ¹⁷ Furthermore, IPD‐MA can clarify whether observed effects apply uniformly or are modified by specific patient characteristics, thereby providing high‐quality evidence on treatment modifiers and predictors of an outcome. ¹⁷ , ¹⁸ , ¹⁹ Current research on antipsychotic use in dementia has mainly relied on aggregate data meta‐analyses, ³ , ¹¹ or single‐trial studies, ²⁰ , ²¹ with only one IPD‐MA exploring factors associated with adverse events. ³ , ²² To date, no IPD‐MA has been performed to investigate potential treatment modifiers or predictors of antipsychotic response across both overall and specific symptom domains of BPSD. Given that risperidone remains the only antipsychotic approved for managing BPSD in some countries, including Australia, Canada, the United Kingdom, and New Zealand, ³ , ¹³ our study aims to use IPD‐MA to identify treatment modifiers and predictors of risperidone response in individuals with dementia experiencing different types of BPSD.

2. METHODS

2.1. Data sources and access

Data were obtained from the Yale Open Data Access (YODA) Project, an independent platform that enables secure and transparent sharing of clinical trial data. YODA was selected for its rigorous scientific review process and its provision of harmonized datasets suitable for secondary analyses. ²³ , ²⁴ We searched this repository for studies involving individuals with dementia and Alzheimer's disease (AD). After identifying relevant trials, we reviewed the associated documentation to select datasets that met the following eligibility criteria: (1) adult participants diagnosed with dementia and exhibiting BPSD, (2) participants involved in double‐blind placebo‐controlled trials receiving risperidone treatment, and (3) IPD were available for the primary outcome. Once the potentially eligible studies were identified, we requested access to the participant‐level datasets through the standard procedure outlined on the website on May 1, 2024. ²⁴

2.2. Trial characteristics

A total of seven trials involving risperidone for the treatment of BPSD in individuals with dementia were identified. One trial was excluded because it was not placebo controlled, leaving six eligible trials for inclusion in the study. Among these, four trials—USA‐63 (ClinicalTrials.gov registration NCT00253123), ²⁵ INT‐24 (NCT00249145), ²⁶ AUS‐5 (NCT00249158), ²⁷ and USA‐232 (NCT00034762) ²⁸ —have had their primary results published. This IPD‐MA also includes two unpublished trials: BEL‐14 (n = 39) and INT‐83 (n = 18). We included all individuals receiving either placebo or risperidone in this analysis. Trial durations varied: 12 weeks for USA‐63, INT‐24, and AUS‐5; 8 weeks for USA‐232 and INT‐83; and 4 weeks for BEL‐14. Most trials involved men and women aged ≥ 55 diagnosed with AD, vascular dementia, or mixed type dementia based on Diagnostic and Statistical Manual of Mental Disorders Fourth Edition criteria, ²⁹ while BEL‐14 used Berg criteria for diagnosis of senile dementia of the AD type. ³⁰ All trials, except for BEL‐14, included participants with a Mini‐Mental State Examination (MMSE) score of ≤ 23. USA‐63 and INT‐24 required participants to have a Behavioral Pathology in Alzheimer's Disease (BEHAVE‐AD) total score of ≥ 8, along with a global rating of ≥ 1 on the same scale. For USA‐232 and INT‐83, an inclusion criterion was a score of at least 2 on any item of the BEHAVE‐AD psychosis subscale. Only AUS‐5 used the Cohen–Mansfield Agitation Inventory (CMAI) frequency score to include participants. ³¹ All trials excluded participants with a diagnosis of other psychiatric conditions. The USA‐63 studies used three fixed‐dose risperidone regimens (0.5 mg, 1 mg, or 2 mg daily), while the other studies used flexible dosing. Dosing ranges included 0.5 to 4 mg daily in INT‐24, 0.5 to 2 mg in AUS‐5, 1 to 4 mg in BEL‐14, and 1 to 1.5 mg in USA‐232 and INT‐83. All trials used BEHAVE‐AD score as the efficacy parameter, while AUS‐5, INT‐24, and USA‐63 also included the CMAI score as an additional parameter. Table S1 in supporting information provides a summary of the trial details.

2.3. Risk of bias assessment

Six trials, AUS‐5, BEL‐14, INT‐24, INT‐83, USA‐63, and USA‐232, were assessed for risk of bias (RoB) using “RoB 2 tool” proposed by the Cochrane group. ³² Two authors independently graded these trials as “low risk of bias,” “high risk of bias,” or “some concerns” in the following five domains: risk of bias arising from the randomization process, risk of bias due to deviations from the intended interventions, missing outcome data, risk of bias in measurement of the outcome, and risk of bias in selection of the reported result. Figure S1 in supporting information provides the RoB of these trials.

2.4. Outcome variables: the BEHAVE‐AD rating score

BPSD were assessed by the BEHAVE‐AD total scale and subscales at week 4 and week 8. ³³ The BEHAVE‐AD is a 25‐item scale consisting of seven subscales: (A) Paranoid and Delusional Ideation (7 items); (B) Hallucinations (5 items); (C) Activity Disturbances (3 items); (D) Aggressiveness (3 items); (E) Diurnal Rhythm Disturbances (1 item); (F) Affective Disturbances (2 items); and (G) Anxieties and Phobias (4 items). Each item of the BEHAVE‐AD scale was initially rated as absent (0) or present; if present, the severity was further classified into one of three categories, where 1 = present, 2 = present with emotional component, and 3 = present with both emotional and physical components. ³⁴ Additionally, a global rating assessed the overall impact of symptoms on patients and caregivers, scored from 0 (not at all) to 3 (severely). ³⁵ In this study, we defined psychosis as the combination of items in both subscale A and subscale B, consistent with the definition used in the original trial. ²⁵ , ²⁷ BEHAVE‐AD scores were captured at baseline, as well as at week 2, 4, and 8. We analyzed BEHAVE‐AD scores as both dichotomous and continuous outcomes. A clinically significant therapeutic response was defined as a ≥ 30% reduction in the BEHAVE‐AD total score from baseline, assessed as a binary outcome. ²⁵ , ²⁶ , ²⁷ , ²⁸ For continuous outcomes, we retained raw scores for the total scale, each individual subscale, and the global rating scale. An early response was defined as achieving the therapeutic response threshold by week 2, while any reduction in score from baseline at week 2 was classified as early improvement. Further details of the BEHAVE‐AD rating scale are provided in Appendix 1 in supporting information.

RESEARCH IN CONTEXT

1. Systematic review: We conducted a literature review using the Web of Science and PubMed databases. Existing research on antipsychotic use in dementia primarily consists of aggregate‐data meta‐analyses or single‐trial studies, with only one individual participant data meta‐analysis (IPD‐MA) addressing factors related to adverse events. No IPD‐MA has been conducted to explore potential treatment modifiers or predictors of antipsychotic response to specific behaviors and psychological symptoms of dementia.

2. Interpretation: Risperidone modestly improved specific symptoms, including aggression, psychosis, and anxiety/phobia. Subgroup analyses indicated that pharmacokinetic‐ and pharmacodynamic‐related factors may affect efficacy, and early response could predict later symptom improvement.

3. Future directionsbrk: Future studies should explore the potential treatment modifiers identified in this study and focus on strategies to balance the risks and benefits of risperidone use, ultimately fostering more personalized treatment decisions.

2.5. Covariate extraction

Covariates were gathered at baseline and during the 4‐ and 8‐week periods, including demographic information at baseline (age, sex, body mass index [BMI], race/ethnicity), clinical conditions at baseline (MMSE score, dementia diagnosis, comorbidities, BEHAVE‐AD score, presence of BPSD), laboratory results at baseline (creatinine levels), mean risperidone dose, and the use of concomitant psychotropic medications during 4 and 8 weeks. Psychotropic medications, with their corresponding Anatomical Therapeutic Chemical (ATC) code, ³⁶ were defined as: anxiolytics (N05B), hypnotics and sedatives (N05C), antidepressants (N06A), and anti‐dementia medications (N06D). ⁹ The presence of BPSD at baseline was classified as follows: psychosis (score ≥ 2 on any items in subscales A and B), activity disturbances (score ≥ 1 on any items in subscale C), aggression (score ≥ 1 on any items in subscale D), sleep disturbances (score ≥ 1 on any items in subscale E), affective disturbance (score ≥ 1 on any items in subscale F), and anxieties/phobias (score ≥ 1 on any items in subscale G). ²⁷ ,  

TABLE 1.

Patient characteristics.

	Characteristic of total population
Baseline characteristic	Placebo, N = 712 *	Risperidone, N = 1009 *
Trial (N, %)
AUS‐5	170 (23.8%)	167 (16.6%)
BEL‐14	19 (2.7%)	20 (2.0%)
INT‐24	114 (16.0%)	115 (11.4%)
INT‐83	8 (1.1%)	10 (1.0%)
USA‐063	163 (22.9%)	462 (45.8%)
USA‐232	238 (33.5%)	235 (23.3%)
Age (years; mean, SD)	83 (8)	83 (7)
Sex (N, %)
Female	498 (70%)	702 (70%)
Male	214 (30%)	307 (30%)
BMI	23.3 (4.7)	23.2 (4.6)
Unknown	52	81
MMSE score at baseline (median, IQR)	9 (12)	8 (12)
Unknown	30	31
Psychotropic medication use at baseline (N, %)
Any anxiolytic use at baseline	44 (6.2%)	53 (5.3%)
Any hypnotics and sedatives use at baseline	86 (12%)	99 (9.8%)
Any antidepressant use at baseline	33 (4.6%)	49 (4.9%)
Any anti‐dementia medication use at baseline	85 (12%)	71 (7.0%)
BEHAVE‐AD score at baseline (Mean, SD)
Total BEHAVE‐AD score at baseline	17 (9)	16 (8)
Unknown	23	29
BEHAVE‐AD psychosis subscale score at baseline	5.7 (4.9)	5.4 (4.3)
Unknown	23	29
BEHAVE‐AD activity disturbances subscale score at baseline	2.98 (2.27)	2.87 (2.16)
Unknown	0	3
BEHAVE‐AD aggressiveness subscale score at baseline	4.63 (2.79)	4.55 (2.92)
Unknown	0	3
BEHAVE‐AD sleep subscale score at baseline	0.74 (0.93)	0.70 (0.93)
Unknown	0	3
BEHAVE‐AD affective disturbance subscale score at baseline	1.04 (1.40)	0.87 (1.26)
Unknown	2	9
BEHAVE‐AD anxieties and phobias subscale score at baseline	1.92 (2.04)	1.79 (2.01)
Unknown	2	9
BEHAVE‐AD global rating score at baseline	1.82 (0.78)	1.83 (0.75)
Unknown	1	3
Presence of BPSD at baseline (N, %)
Presence of psychosis symptoms at baseline	498 (71%)	677 (68%)
Unknown	7	15
Presence of activity disturbances at baseline	622 (87%)	868 (86%)
Unknown	0	3
Presence of aggressiveness symptoms at baseline	655 (92%)	896 (89%)
Unknown	0	3
Presence of affective disturbance symptoms at baseline	338 (48%)	433 (43%)
Unknown	0	3
Presence of sleep disturbance symptoms at baseline	332 (47%)	438 (44%)
Unknown	0	3
Presence of anxieties and phobias symptoms at baseline	475 (67%)	650 (65%)
Unknown	0	3
Race (N, %) †
White	502 (88%)	762 (88%)
Other	33 (5.8%)	37 (4.3%)
Black	35 (6.1%)	65 (7.5%)
Dementia type (N, %) †
Alzheimer's disease	428 (76%)	629 (74%)
Mixed type dementia	41 (7.2%)	74 (8.7%)
Vascular dementia	97 (17%)	146 (17%)
Unknown	4	15
eGFR (ml/min; mean, SD) †	62 (18)	62 (17)
Unknown	4	2
Any currently active cardiovascular disease (N, %) †	381 (67%)	595 (69%)
Unknown	0	2
Any currently active endocrine disease (N, %) †	173 (30%)	241 (28%)
Unknown	0	1
Any currently active neurological disease (N, %) †	162 (28%)	290 (34%)
Unknown	1	1
Number of participants at week 4	640	888
Number of participants at week 8	543	768

Abbreviation: BEHAVE‐AD, Behavioral Pathology in Alzheimer's Disease; BMI, body mass index; BPSD, behaviors and psychological symptoms associated with dementia; eGFR, estimated glomerular filtration rate; IQR, interquartile range; MMSE, Mini‐Mental State Examination; SD, standard deviation.

^* n (%); Mean (SD)

^† Data obtained from 3 trials only (AUS‐5, USA‐063, and USA‐232). Denominators are the number of placebo (N = 570) and risperidone (N = 864) in this dataset.

Trial BEL‐14 did not provide participant‐level data on comorbidities, race/ethnicity, or dementia diagnosis; INT‐24 lacked participant‐level data on comorbidities; and INT‐83 did not collect baseline creatinine measurements. Additionally, trial AUS‐5 did not capture BEHAVE‐AD scores at week 2. To maintain consistency when integrating data across six trials, we constructed two master datasets. Dataset A included data from all available trials but had fewer variables: age, sex, BMI, MMSE scores, baseline BEHAVE‐AD total scores and each subcomponent score, presence of BPSD at baseline, risperidone mean dose, and concomitant psychotropic medications. Dataset B contained data from three trials only (AUS‐5, USA‐63, and USA‐232) but included all variables from Dataset A along with additional variables, including race/ethnicity, comorbidities, estimated glomerular filtration rate, and dementia diagnosis. A detailed summary of variables in each dataset is provided in Table S2 in supporting information.

2.6. Statistical analysis

2.6.1. Descriptive analyses

The baseline characteristics of participants were reported descriptively as proportions or as means/medians with corresponding standard deviations (SDs)/interquartile ranges (IQRs). We conducted a one‐stage IPD‐MA using a multivariable mixed‐effects regression with random intercepts. ¹⁷ This approach accounts for heterogeneity between studies in baseline risks (intercepts), while assuming uniform predictor effects across all studies. ¹⁹ Outcomes were reported at week 4 and 8, as odds ratios (ORs) for dichotomous outcomes and standard mean differences (SMDs) for continuous outcomes, with 95% confidence intervals (CIs). SMD values of 0.2, 0.5, and 0.8 correspond to small, medium, and large effects, respectively. ³⁷

2.6.2. Estimating risperidone effects in total population and subpopulation

We evaluated the effectiveness of risperidone compared to placebo in the overall study population and in subpopulations, which were defined as the presence of specific BPSD at baseline. Outcomes assessed in the total population were the BEHAVE‐AD total score and the global rating scale. In subpopulations defined by symptoms, the corresponding BEHAVE‐AD subscale scores were used as the assessment.

2.6.3. Exploring modifiers of risperidone treatment

Treatment modifiers are patient characteristics that may influence differential treatment responses among subgroups. To explore potential treatment modifiers for the BEHAVE‐AD total score and its subscales, we tested interaction terms between baseline covariates and treatment group. ³⁸ We applied random‐intercept models combined with an S‐learner strategy to ensure robustness in IPD‐MA. ¹⁹ All treatment–factor interactions were adjusted for other confounding variables using Dataset A. If the variable was exclusively present in Dataset B (e.g., comorbidities, dementia diagnosis), adjustments were made using Dataset B. After identifying interaction terms that were statistically significant, we performed subgroup analyses to further ensure the robustness of our findings. ³⁸ , ³⁹

2.6.4. Identifying predictors of treatment response

Predictors are baseline characteristics or treatment‐emergent factors that may help estimate the likelihood of a future outcome. For the analysis of predictors of therapeutic response, we restricted the sample to participants receiving risperidone only. Both logistic and linear mixed‐effects regression models were applied to examine treatment response at weeks 4 and 8. Variables included in the final models were based on clinical reasoning and previous literature, ²⁰ , ²¹ , ²² including age, sex, race/ethnicity, BMI, MMSE score, dementia diagnosis, presence of BPSD symptoms at baseline, total and subscale BEHAVE‐AD score, psychotropic medication use at baseline, concomitant psychotropic medications during the trials, risperidone dose, and early response to treatment. We developed two types of models: one that included only baseline variables and another that incorporated both baseline and treatment‐emergent variables.

The variance inflation factor (VIF) for each included variable was checked to ensure the absence of multicollinearity (VIF < 5). ⁴⁰ A p value < 0.05 was considered statistically significant. As for the multivariable regression analysis for the predictors of treatment response (eight outcomes at each endpoint), a Bonferroni corrected p value (< 0.05/44  =  0.00113) was considered statistically significant in either Dataset A or B. ⁴¹ Logistic and linear mixed effects models were fitted using the glmer and lmer functions in the package lme4. ⁴² Incomplete covariate data were accommodated under a missing‐at‐random (MAR) assumption through full‐information maximum‐likelihood estimation, as implemented in the mixed‐effects models. ⁴³ Statistical analyses were performed using R version 4.3.0.

3. RESULTS

3.1. Baseline patient characteristics

The analysis included 1009 patients receiving risperidone and 712 patients receiving a placebo. Both groups had a mean age [SD] of 83 (risperidone: 83 [8]; placebo: 83 [7]) years, with males comprising 30% of the total cohort. Participants were predominantly White (88% in both groups), with median (IQR) MMSE scores of 8 (12) for risperidone and 9 (12) for placebo. The most common BPSD in both groups were aggression, followed by activity disturbances and psychosis. Mean [SD] baseline BEHAVE‐AD total scores were similar between groups (risperidone: 16 [8]; placebo: 17 [9]). Cardiovascular conditions were the predominant comorbidities (69% risperidone; 67% placebo). Hypnotics and sedatives were the most frequently used psychotropic medications at baseline (risperidone: 9.8%; placebo: 12%). Additional details of the cohort characteristics are presented in Table 1.

3.2. Treatment effects of risperidone in total population and subpopulation

In the overall population, risperidone use was not statistically associated with achieving a therapeutic response at both week 4 (OR: 1.23; 95% CI: 0.97–1.56), and at week 8 (OR: 1.30; 95% CI: 1.01–1.67) after the Bonferroni correction (Table S3 in supporting information). Meanwhile, mean reductions in BEHAVE‐AD total scores and global rating scales from baseline were statistically significant at both time points (Figure 1). Beneficial effects of risperidone were seen for the aggression (Week 4 SMD: −0.17; 95% CI: −0.28 to −0.06; Week 8 SMD: −0.22; 95% CI: −0.34 to −0.10) and anxieties and phobias (Week 4 SMD: −0.16; 95% CI: −0.30 to −0.02; Week 8 SMD: −0.19; 95% CI: –0.35 to −0.04) subpopulations. By week 8, psychosis subgroups had lower BEHAVE‐AD psychosis scores (SMD: −0.23; 95% CI: −0.37 to −0.09). Interestingly, people without sleep disturbance at baseline showed a statistically higher score for sleep disturbance at week 4 (SMD: 0.10; 95% CI: 0.01–0.20).

FIGURE 1.

Risperidone's effects on the total population and subpopulation. BEHAVE‐AD, Behavioral Pathology in Alzheimer's Disease; CI, confidence interval; MD, mean difference.

3.3. Treatment modifiers of risperidone effect

We identified statistically significant interactions between baseline characteristics and treatment effect for several outcomes at weeks 4 and 8. These characteristics included BMI, sex, race/ethnicity, MMSE score, and the presence of active cardiovascular, endocrine, and neurological diseases. Only statistically significant interactions are reported in Table S4 in supporting information. Subgroup analysis of risperidone response to different outcomes is shown in Figure 2. Risperidone statistically significantly improved the BEHAVE‐AD total score at week 8 among participants with a normal BMI (SMD: −0.23; 95% CI: −0.38 to −0.09) and those with currently active neurological conditions (SMD: −0.36; 95% CI: −0.57 to −0.15). For the BEHAVE‐AD global rating scale, males receiving risperidone showed greater improvement compared to placebo at week 4 (SMD: −0.34; 95% CI: −0.54 to −0.15). Additionally, participants without endocrine diseases at baseline who used risperidone had lower BEHAVE‐AD Aggression score at week 8 (SMD: −0.32; 95% CI: −0.46 to −0.17). Among White participants, risperidone use was associated with a statistically significant reduction in the BEHAVE‐AD anxieties and phobias subscale at week 8 compared to non‐users (SMD: −0.20; 95% CI: −0.32 to −0.07). In contrast, only modest or non‐significant effects were observed across the remaining BEHAVE‐AD subscales in all subgroups.

FIGURE 2.

Subgroup analysis of Risperidone response in different outcomes. BEHAVE‐AD, Behavioral Pathology in Alzheimer's Disease; BMI, body mass index; CI, confidence interval; MMSE, Mini‐Mental State Examination.

3.4. Predictors of therapeutic response

Table 2 presents the results of the multivariate mixed‐effects logistic regression examining predictors of therapeutic response. No statistically significant factor was found in models including only baseline variables. Models with treatment‐emergent variables showed that early response at week 2 was strongly associated with achieving therapeutic response at both week 4 (OR: 9.04; 95% CI: 6.10–13.39) and week 8 (OR: 4.46; 95% CI: 3.01–6.61). A sensitivity analysis using Dataset B, which included additional adjustment for comorbidities, shared consistent results (Table S5 in supporting information). No other predictors were found to be significantly associated with therapeutic response.

TABLE 2.

Multivariable mixed‐effect logistic regression model of therapeutic response among risperidone users at week 4 and 8.

Predictors of therapeutic response	Week 4		Week 8
	OR (95% CI)	p	OR (95% CI)	p
Model 1
Age	1 (0.98–1.02)	0.8605	0.99 (0.96–1.01)	0.2852
Male (reference: female)	1.08 (0.76–1.53)	0.6677	0.93 (0.64–1.37)	0.7262
Body mass index	0.96 (0.93–1)	0.0274	0.99 (0.96–1.03)	0.7385
MMSE score	1 (0.98–1.03)	0.8663	1 (0.97–1.02)	0.7453
Use of psychotropic medications at baseline
Anxiolytics	1.09 (0.55–2.18)	0.7985	1.02 (0.46–2.25)	0.9601
Hypnotics and sedatives	0.96 (0.55–1.68)	0.8971	1.65 (0.91–2.99)	0.1012
Antidepressants	1.32 (0.62–2.8)	0.4739	0.42 (0.17–1.04)	0.0601
Anti‐dementia medications	1.82 (0.95–3.47)	0.0691	1.15 (0.59–2.27)	0.6753
BEHAVE‐AD total score at baseline	1.13 (0.9–1.41)	0.3027	0.75 (0.58–0.96)	0.0246
Presence of BPSD at baseline (reference: no symptom)
Psychosis	1.3 (0.88–1.92)	0.1834	0.98 (0.65–1.49)	0.9238
Activity disturbances	1.02 (0.65–1.61)	0.9221	1.17 (0.71–1.94)	0.5374
Aggression	0.86 (0.51–1.43)	0.5528	1.11 (0.64–1.94)	0.7132
Affective disturbance	0.96 (0.62–1.19)	0.3507	0.98 (0.73–1.49)	0.8273
Sleep disturbances	1.14 (0.82–1.58)	0.4436	0.93 (0.65–1.33)	0.6999
Anxieties and phobias	1.05 (0.75–1.47)	0.7785	0.96 (0.76–1.56)	0.6433
Number of observations	741		646
Number of trials	6		5
Model 2
Early response (Reference: No response)	9.04 (6.1–13.39)	<0.001 *	4.46 (3.01–6.61)	<0.001 *
Age	0.99 (0.97–1.02)	0.6301	1.01 (0.98–1.04)	0.4836
Male (reference: female)	0.93 (0.6–1.46)	0.7676	0.91 (0.58–1.43)	0.6807
Body mass index	0.96 (0.92–1)	0.0416	1.01 (0.97–1.06)	0.5379
MMSE score	1.01 (0.98–1.05)	0.5082	1.01 (0.98–1.04)	0.5825
Use of psychotropic medications at baseline
Anxiolytics	0.79 (0.28–2.26)	0.6645	2.34 (0.72–7.58)	0.1573
Hypnotic and sedatives	2.17 (0.58–8.02)	0.2475	0.5 (0.14–1.73)	0.272
Antidepressants	1.16 (0.45–2.99)	0.7621	1.67 (0.62–4.54)	0.313
Anti‐dementia medications	1.68 (0.79–3.57)	0.1745	0.67 (0.33–1.36)	0.2674
BEHAVE‐AD total score at baseline	1.00 (0.97–1.04)	0.8702	1.01 (0.97–1.05)	0.7103
Concomitant use of psychotropic medications
Anxiolytics	0.88 (0.45–1.75)	0.7192	0.58 (0.3–1.11)	0.1004
Hypnotics and sedatives	0.7 (0.24–2.03)	0.5087	1.42 (0.52–3.85)	0.4911
Mean dose of risperidone	0.91 (0.59–1.4)	0.6682	1.4 (0.93–2.11)	0.1039
Presence of BPSD at baseline (reference: no symptom)
Psychosis	1.41 (0.86–2.32)	0.1709	1.02 (0.62–1.67)	0.9359
Activity disturbances	0.79 (0.45–1.39)	0.4073	0.81 (0.45–1.46)	0.4759
Aggression	1.17 (0.63–2.15)	0.6193	1.27 (0.68–2.37)	0.4555
Affective disturbance	0.67 (0.45–1.01)	0.0579	1.23 (0.81–1.88)	0.3277
Sleep disturbances	1.11 (0.73–1.68)	0.6191	1.18 (0.77–1.81)	0.4551
Anxieties and phobias	1.19 (0.78–1.8)	0.4269	1.26 (0.83–1.92)	0.2862
Number of observations	620		536
Number of trials	5		4

Note: Model 1 includes baseline variables only. Model 2 includes baseline variables and treatment‐emergent variables. Model 2 does not include AUS‐5 trial as the BEHAVE‐AD total score was not captured at week 2.

Abbreviations: BEHAVE‐AD, Behavioral Pathology in Alzheimer's Disease; CI, confidence interval; BPSD, behaviors and psychological symptoms of dementia; MMSE, Mini‐Mental State Examination; OR, odds ratio.

^*Values are significant results after Bonferroni correction (p value < 0.05/44  =  0.00113).

3.5. Predictors of symptom improvement

Table 3 presents the statistically significant results of the linear mixed‐effects regression models, including treatment‐emergent factors, for the BEHAVE‐AD total score and its subscales at week 8. Across all models, early score reduction at week 2 was consistently associated with lower symptom scores after 8 weeks. A unit increase in the baseline MMSE score was modestly associated with greater reductions in the BEHAVE‐AD total score, aggression, and activity disturbances, but associated with increased psychosis and anxiety/phobia scores. In the BEHAVE‐AD total score model, concomitant use of anxiolytics was associated with higher scores (SMD: 0.23; 95% CI: 0.05–0.41). Baseline use of anti‐dementia medications was linked to a reduction in aggression scores (SMD: −0.31; 95% CI: −0.50 to −0.13). Results for full models of Dataset A at both weeks are provided in Table S6 in supporting information. A sensitivity analysis using Dataset B showed similar findings (Table S7 in supporting information). Models that included only baseline variables indicated that MMSE was statistically significant in relation to some of the subscales (Tables S8 and S9 in supporting information).

TABLE 3.

Mixed‐effect linear regression model of predictors of all outcomes at week 8.

Predictors of BPSD score at week 8	Total score		Psychosis		Activity disturbances		Aggressiveness		Sleep disturbances		Affective disturbances		Anxieties and phobias		Global rating
	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p	SMD (95% CI)	p
Early reduction in rating score
BEHAVE‐AD total score	−0.08 (‐0.09–0.08)	<0.001 *	−	−	−	−	−	−	−	−	−	−	−	−	−	−
BEHAVE‐AD Psychosis subscale score	−	−	−0.2 (‐0.21; ‐0.19)	<0.001 *	−	−	−	−	−	−	−	−	−	−	−	−
BEHAVE‐AD activity disturbances subscale	−	−	−	−	−0.31 (‐0.35; ‐0.28)	<0.001 *	−	−	−	−	−	−	−	−	−	−
BEHAVE‐AD Aggression subscale score	−	−	−	−	−	−	−0.27 (‐0.29; ‐0.25)	<0.001 *	−	−	−	−	−	−	−	−
BEHAVE‐AD sleep disturbance subscale	−	−	−	−	−	−	−	−	−0.99 (‐1.07; ‐0.92)	<0.001 *	−	−	−	−	−	−
BEHAVE‐AD affective disturbance subscale	−	−	−	−	−	−	−	−	−	−	−0.6 (‐0.65; ‐0.54)	<0.001 *	−	−	−	−
BEHAVE‐AD anxieties and phobias subscale	−	−	−	−	−	−	−	−	−	−	−	−	−0.34 (‐0.37; ‐0.3)	<0.001 *	−	−
BEHAVE‐AD Global rating subscale score	−	−	−	−	−	−	−	−	−	−	−	−	−	−	−0.77 (‐0.83; ‐0.71)	<0.001 *
Age	−	−	−	−	−	−	−0.01 (‐0.02; 0)	0.0357	−	−	−	−	−	−	−	−
Male (reference: female)	−	−									−0.14 (‐0.26; ‐0.03)	0.0153
BMI	−	−											−0.01 (‐0.03; 0)	0.0224	−	−
MMSE	−0.01 (‐0.02; ‐0.01)	0.0021 *	0.02 (0.01; 0.03)	<0.001 *	−0.01 (‐0.02; 0)	0.0046	−0.03 (‐0.04; ‐0.02)	<0.001 *	−	−	−	−	0.02 (0.01; 0.03)	<0.001 *	−0.01 (‐0.02; 0)	0.0091
Use of psychotropic medications at baseline
Anxiolytics	−	−	−	−	−0.39 (‐0.72; ‐0.05)	0.0253	−	−	−	−	−	−	−	−	−	−
Hypnotic and sedatives	−	−	−	−	−	−	0.37 (0.06; 0.69)	0.0211	−	−	−	−	−	−	−	−
Anti‐dementia medications	−	−	0.24 (0.08; 0.4)	0.0041	−	−	−0.31 (‐0.5; ‐0.13)	<0.001 *	−	−	−	−	−	−	−	−
Concomitant use of psychotropic medications
Anxiolytics	0.23 (0.05; 0.41)	0.0128	−	−	−	−	−	−	−	−	−	−	0.24 (0.05; 0.43)	0.0141	−	−
Hypnotics and sedatives	−	−	−	−	−	−	−0.35 (‐0.59; ‐0.11)	0.0051 *	−	−	−	−	−	−	−	−
Presence of BPSD symptom at baseline (reference: no symptom)
Psychosis	0.64 (0.51; 0.76)	<0.001 *	0.49 (0.38; 0.6)	<0.001 *	−0.25 (‐0.4; ‐0.1)	0.0011 *	−0.38 (‐0.51; ‐0.25)	<0.001 *	−	−	−0.14 (‐0.27; ‐0.01)	0.0351	−	−	−	−
Activity disturbances	0.42 (0.26; 0.57)	<0.001 *	−	−	0.91 (0.73; 1.09)	<0.001 *	−0.49 (‐0.65; ‐0.34)	<0.001 *	−	−	−	−	−	−	−	−
Aggression	0.6 (0.44; 0.76)	<0.001 *	−0.41 (‐0.54; ‐0.27)	<0.001 *	−	−	0.86 (0.7; 1.03)	<0.001 *	−	−	−0.22 (‐0.38; ‐0.06)	0.0081	−	−	−	−
Affective disturbance	0.36 (0.26; 0.47)	<0.001 *	−	−	−0.3 (‐0.43; ‐0.18)	<0.001 *	−0.21 (‐0.32; ‐0.1)	<0.001 *	−	−	1.33 (1.2; 1.45)	<0.001 *	−	−	−0.16 (‐0.28; ‐0.05)	0.005
Sleep disturbances	0.35 (0.24; 0.46)	<0.001 *	−0.15 (‐0.24; ‐0.05)	0.0017 *	−	−	−0.25 (‐0.36; ‐0.14)	<0.001 *	1.72 (1.58; 1.85)	<0.001 *	−	−	−	−	−	−
Anxieties and phobias	0.36 (0.25; 0.47)	<0.001 *	−	−	−	−	−0.38 (‐0.49; ‐0.27)	<0.001 *	−	−	−	−	0.92 (0.79; 1.04)	<0.001 *	−	−
BEHAVE‐AD total score at baseline	−	−	0.81 (0.74; 0.88)	<0.001 *	0.51 (0.41; 0.6)	<0.001 *	0.73 (0.65; 0.81)	<0.001 *	0.11 (0.03; 0.19)	0.0049	0.29 (0.21; 0.37)	<0.001 *	0.37 (0.28; 0.46)	<0.001 *	0.37 (0.28; 0.45)	<0.001 *
Number of observations	541		541		541		541		541		541		541		539
Number of trials	4		4		4		4		4		4		4		4

Abbreviations: BEHAVE‐AD, Behavioral Pathology in Alzheimer's Disease; BPSD, behaviors and psychological symptoms of dementia; CI, confidence interval; SMD, standardized mean difference.

^*Values are significant results after Bonferroni correction (p value < 0.05/44 =  0.00113)

4. DISCUSSION

4.1. Summary of main findings

To the best of our knowledge, this is the first study to apply an IPD‐MA approach to evaluate treatment response, treatment modifiers, and predictors of risperidone in people with dementia. Our findings show that risperidone provided a modest benefit in achieving a therapeutic response at week 8. Symptom‐specific analyses revealed that individuals with baseline psychosis experienced a reduction in psychosis scores by week 8, while those with aggression and anxiety showed improvement at week 4. No significant benefit was observed for other BEHAVE‐AD components. We also identified several demographic and clinical factors that modified risperidone's effect on the BEHAVE‐AD total score and its subcomponents. Notably, achieving an early therapeutic response at week 2 was a strong predictor of continued response at both week 4 and week 8.

4.2. Discussion of main results

Our study demonstrated that risperidone did not produce a clinically significant reduction in the BEHAVE‐AD total score but showed a modest effect in alleviating aggression, and anxieties/phobias symptoms after 4 weeks, and psychosis symptoms after 8 weeks in people with dementia. These findings align with those of a Cochrane review ¹¹ and a systematic review by Zhou et al., ⁴⁴ despite the use of mixed outcome measures, including BEHAVE‐AD and the Neuropsychiatric Inventory (NPI). However, while the Cochrane review reported non‐significant effects of risperidone for psychosis, both our analysis and Zhou et al.’s review observed statistically significant benefits. This discrepancy may be explained by differences in data sources: our analysis used IPD from six trials, and Zhou et al. included a larger number of studies, while the Cochrane review relied on a smaller set of aggregate‐level data. The improvement in aggression, anxieties, and phobias might be explained by risperidone's sedative and antidepressant effects by blocking serotonin 5‐HT2A, α1 and α2 adrenergic and H1 histaminergic receptors. ¹³ No statistically significant effects of risperidone on other neuropsychiatric symptoms was seen, supporting current clinical guidelines that recommend its use only for severe cases of psychosis or aggression when non‐pharmacological interventions have failed. ¹² , ¹³ , ⁴⁵ Interestingly, risperidone appeared to worsen sleep disturbances (increase nighttime wakefulness) in individuals without such symptoms at baseline, consistent with findings from a previous open‐label trial. ⁴⁶ Antipsychotics can have both sedative and insomnia‐inducing effects, depending on their receptor activity and individual patient factors. ⁴⁷ Although the underlying mechanisms remain unclear, this observation reinforces the recommendation that risperidone be reserved for short‐term use, as treatment beyond 12 weeks is associated with an increased risk of adverse events. ¹² A previous analysis using a similar trial cohort also reported higher rates of cerebrovascular adverse events and mortality associated with risperidone use, ²² reinforcing the need to reserve pharmacological intervention as a last resort in this population.

In subgroup analyses, we observed that patients with normal BMI, those without active endocrine diseases at baseline, and those who are male and White may appear to benefit more from risperidone treatment, demonstrating statistically significant interactions with treatment effects, although the effect size is small. These findings suggest that certain patient characteristics, possibly reflecting underlying pharmacokinetic and pharmacodynamic factors, might influence how effectively risperidone can reduce psychosis, aggression, and anxiety/phobias. Risperidone primarily acts via dopamine D₂ and serotonin 5‐HT2A receptor antagonism, with moderate affinity for α1‐adrenergic and H₁ receptors. It is primarily metabolized by the hepatic enzyme cytochrome (CYP) 2D6 into its active metabolite, 9‐hydroxy‐risperidone, which shares similar pharmacological properties with risperidone itself. ⁴⁸ Both compounds are predominantly eliminated through renal excretion. ⁴⁸ The identified treatment modifiers—sex, ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² BMI, ⁵³ , ⁵⁴ , ⁵⁵ endocrine status, ⁵⁶ , ⁵⁷ and race/ethnicity ⁵⁸ , ⁵⁹ , ⁶⁰ —are consistent with previous evidence suggesting these factors may influence pharmacokinetics (e.g., altered drug distribution, ⁵⁴ , ⁵⁶ , ⁵⁷ CYP activity ⁴⁸ , ⁵⁰ , ⁵⁹ , ⁶¹ or plasma protein binding, ⁵³ , ⁵⁴ P‐glycoprotein function ⁴⁸ , ⁵⁴ ) or pharmacodynamics (e.g., variations in receptor D₂ and 5‐HT2A genes, ⁵⁵ , ⁶⁰ , ⁶² different receptor affinity ⁵¹ , ⁵² ) of risperidone. However, they only partially explain the observed variability as the precise relationship between risperidone plasma concentration and clinical outcomes remains uncertain ⁶¹ and may not follow a simple linear pattern. ⁶³ Additionally, risperidone‐induced weight gain ⁶⁴ may alter its metabolism and excretion, potentially influencing clinical outcomes. However, it is important to interpret these subgroup findings cautiously, as they were not prespecified. ³⁹ Thus, these findings should be viewed as hypothesis generating, requiring validation in future studies to confirm clinical implications. ³⁸

Among risperidone users, we found that achieving early therapeutic response or improvement at week 2 significantly predicted subsequent response at weeks 4 and 8 in both total and individual symptom scores. This finding aligns with a previous study on AD patients treated with antipsychotics, which reported that symptom improvements observed at week 2 were associated with treatment outcomes at week 8. ²¹ Clinically, evaluating response at week 2 could inform decisions about continuing antipsychotic treatment, helping to minimize associated risks. ¹⁴ However, BPSD may also be influenced by non‐pharmacological factors, such as addressing unmet needs or environmental triggers, ⁶⁵ which were not mentioned in the trial protocol. Higher baseline cognitive function (higher MMSE scores) was modestly linked to improvements in most BEHAVE‐AD subscales, but was paradoxically associated with slightly worse outcomes in psychosis and anxiety/phobia symptoms. These mixed findings reflect the complex and sometimes contradictory relationship between cognitive impairment and BPSD. Although some studies have reported psychosis was associated with worsening cognitive function, ⁶⁶ , ⁶⁷ others have shown minimal or even slight improvement. ⁴⁴ Given the modest associations observed, further research is needed to clarify whether baseline cognition can inform treatment decisions and optimize benefit–risk balance. The concomitant use of anxiolytics during the trials predicted higher overall symptom scores, possibly due to greater baseline symptom severity. However, this combination should be cautiously avoided due to an increased risk of adverse events, including stroke and mortality. ¹² Finally, higher symptom severity at baseline consistently predicted higher severity score compared to those without those conditions, suggesting the modest benefits observed with risperidone.

4.3. Strengths and limitations

This study analyzed data from multiple clinical trials conducted across different regions, enhancing the generalizability of the findings. The one‐stage (centered) IPD‐MA approach we used can minimize the presence of ecological bias across studies. ¹⁶ Additionally, using a random‐intercept, S‐learner strategy aligns with current recommendations for estimating individualized treatment effects in IPD‐MA analyses, which can provide a better performance. ¹⁹ However, several limitations should be acknowledged. First, not all existing risperidone trials were included in the analysis, as only those available through the YODA database were accessible. Although this may introduce selection bias and limit the exploration of a broader range of subgroups, our findings are consistent with those of a prior pooled meta‐analysis of 18 studies, supporting the robustness of our results. ⁴⁴ Second, important covariates, such as genetic factors influencing drug metabolism (e.g., CYP2D6 ⁴⁸ ) and dementia subtypes (e.g., AD, vascular dementia), were unavailable, as these data were not consistently collected across all trials. Third, as the study population was predominantly White, the findings may not generalize to Asian, African, and other race/ethnic groups, in which CYP2D6 pharmacogenetic variability can alter risperidone pharmacokinetics and response. Therefore, this study was not powered to detect potential interactions with different races/ethnicities. Fourth, detailed information about concomitant medications and specific comorbidities was not collected by the trials, potentially complicating the interpretation of our results. Additionally, reliance on the MAR assumption may bias pooled estimates. Although the inclusion of unpublished trials (BEL‐14 and INT‐83) can reduce publication bias, the absence of peer review may introduce some risk of bias. Finally, due to the limitations of the dataset, symptom assessment was restricted to the BEHAVE‐AD scale. The NPI, which is now more commonly used in both clinical and research settings, was not available in these trials. ⁶⁸

4.4. Future research and implications

This study adds to existing evidence supporting the effectiveness of risperidone in managing psychosis, aggression, and possibly anxiety in people with dementia. Nonetheless, developing a data‐driven prediction model to identify patients most likely to benefit or experience harm could substantially improve clinical decision making, support personalized BPSD management, and ultimately enhance patient outcomes. Previous research has demonstrated that well‐designed prediction models of antipsychotic use in people with schizophrenia can assist clinicians in personalizing treatment, optimizing therapeutic effectiveness, and reducing the likelihood of adverse effects. ⁶⁹ In addition, subgroups that may influence treatment response—possibly due to risperidone's pharmacokinetics and pharmacodynamics—were identified, and these hypotheses should be tested to enable clinically meaningful interpretation. Finally, early response at week 2 was strongly associated with achieving response at weeks 4 and 8, suggesting that early treatment evaluation could help guide more informed clinical decisions. ¹⁴

5. CONCLUSION

Risperidone demonstrates modest effectiveness for treating psychosis, aggression, and anxiety in people with dementia, while showing limited impact on other symptoms. Treatment response may vary depending on individual pharmacodynamic/pharmacokinetic‐related factors. Early improvement by week 2 appears to be a strong indicator of continued response at weeks 4 and 8. Future research should focus on developing tools to better balance the risks and benefits of antipsychotic use, ultimately supporting more personalized and effective treatment decisions.

CONFLICTS OF INTEREST STATEMENT

The authors report no competing interests. E.C.K.T. is supported by a Dementia Australia Research Foundation Mid‐Career Research Fellowship. Author disclosures are available in the supporting information.

CONSENT STATEMENT

Consent was not required for this study, which was approved by the University of Sydney Human Research Ethics Committee (Approval No. 2024/HE000171).

Supporting information

Supporting Information

Supporting Information

Le HT, Lau ECY, Lu CY, et al. Treatment modifiers and predictors of risperidone response in dementia: An individual participant data meta‐analysis of six randomized controlled trials. Alzheimer's Dement. 2025;21:e70665. 10.1002/alz.70665