Canine-assisted therapy in reducing stress and anxiety levels of university students: systematic review and meta-analysis of randomized controlled trials

Abstract

Background

Due to the high prevalence of mental health issues among university students worldwide, canine-assisted therapy (CAT) has emerged as a potential intervention to reduce student stress and anxiety. This study systematically reviews and meta-analyzes the effects of CAT on reducing stress and anxiety levels among university students.

Methods

Following the PRISMA 2020 guidelines, we conducted a systematic search across multiple databases (APA PsycINFO, PubMed, Duke Libraries, CNKI, Wanfang, and Google Scholar) for randomized controlled trials published in English and Chinese. Only studies incorporating professionally trained dogs and handlers were included. Two reviewers (SS and ZL) independently extracted data, and the risk of bias was assessed using the Cochrane Risk of Bias 2 tool. Effect sizes (Hedges’ g) were pooled using a random-effects meta-analysis to account for the anticipated clinical and methodological heterogeneity. Subgroup analyses were conducted to explore moderators such as intervention duration, baseline stress levels, and control condition types. Publication bias was evaluated using funnel plots, Egger’s test, and trim‑and‑fill analysis.

Results

Of 290 identified studies, 15 met the inclusion criteria and 14 were included in the meta-analysis. The meta-analysis yielded a statistically significant overall effect size of g = -0.67 (p <.001), indicating a moderate reduction in stress and anxiety among university students receiving CAT. Forest plots revealed effect sizes ranging from approximately − 1.34 to -0.13 across studies. Although substantial heterogeneity was observed, subgroup analyses showed that CAT was significantly more effective for students with high baseline stress and anxiety levels. Funnel-plot asymmetry suggested possible bias, but Egger’s test was non‑significant. Trim‑and‑fill analysis imputed two missing studies, adjusting the pooled effect to g = -0.59, indicating the findings remain robust.

Conclusion

CAT demonstrates promise in alleviating stress and anxiety among university students. Despite methodological variations and potential publication bias, the findings suggest that CAT may offer a feasible and accessible approach to enhancing mental well-being within university environments. Further studies are necessary to examine factors influencing methodological diversity and refine the integration of CAT within university settings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12906-025-04955-2.

Background

Discussions are increasingly pointing to a growing crisis in student mental health [1]. In 2021 alone, approximately 73% of students in the United States reported persistent psychological distress, a significant increase since 2013, reflecting a 50% rise in mental health challenges within this population [2]. The causes of these issues include academic workload pressure, loneliness, inadequate sleep, and being away from family [3]. Prolonged exposure to stress can contribute to the development of more severe mental health conditions, further increasing anxiety rates among university students [4]. Understanding different approaches to intervening in anxiety and stress is crucial, as these issues significantly affect students’ daily functioning, life satisfaction, and overall quality of life [5, 6]. When left unaddressed, stress and anxiety are associated with risk factors such as academic burnout, relationship difficulties, decreased productivity, and increased dropout rates [7]. In extreme cases, students may adopt maladaptive coping strategies, including excessive alcohol consumption, social withdrawal, avoidance behaviors and even suicidal tendencies [5].

In response to the impact of stress on both academic performance and overall well-being, universities have introduced various stress-reduction interventions such as mindfulness programs, peer support groups, and on-campus counseling services [8, 9]. However, the effectiveness of these interventions remains inconsistent due to underutilization and low student uptake. Peer support may deter students from disclosing personal issues to their peers, while mindfulness programs face operational challenges: universities must invest in staff training, program delivery, and materials, and students must dedicate considerable time and effort to learn and maintain mindfulness practices [10, 11]. On-campus counseling services are often strained under high demand, leading to long waiting times for students and increased counselor burnout and turnover [12]. Clinical interventions require qualified professionals and depend heavily on student motivation and engagement, which can further limit their accessibility and impact [13]. Given these considerations, there is a need for interventions that are more accessible and convenient for students, making canine-assisted therapy a promising option.

Canine-assisted therapy is a complementary intervention using professionally trained dogs to enhance individuals’ physical, social, emotional, or cognitive functioning [14, 15]. Therapy dogs and their handlers undergo specialized training to ensure that interactions are beneficial and secure [15]. Often referred to as humans’ best friends, therapy dogs have been successfully incorporated into various settings, including nursing homes, hospitals, and schools, tailored to diverse populations [16]. Even briefly interacting, such as hugging, petting, and sitting next to therapy dogs, can lower levels of stress hormones like epinephrine and norepinephrine, increase endorphin levels and oxytocin, leading to decreased stress, anxiety, and pain, improve mood, and better social bonding [15]. Recent studies have shown that canine-assisted therapy effectively reduces stress among nurses [17] and enhances the quality of life for Alzheimer’s patients [18].

Practically, canine-assisted therapy is relatively inexpensive, as dogs are easily transportable and can engage with a larger number of people during a single drop-in session, making it a cost-effective option for university-based interventions [19]. A report indicated that 62% of surveyed universities in the United States provide animal-assisted therapy (AAT) programs, the majority of which exclusively involved dogs [20]. This suggests that canine-assisted therapy is a viable intervention for addressing the escalating mental health concerns in universities, particularly given the high burnout rate among counselors and the financial implications of hiring more professional counselors [12].

Theoretically, canine-assisted therapy (CAT) is based on social support theory, offering a source of social connectedness and combating feelings of loneliness and isolation [21]. Social support, as defined by Thoits [22], refers to the social resources individuals can rely on when dealing with life challenges and stressors. According to this theory, social support has several dimensions: First, social support can be conceptualized as perceived support, reflecting an individual’s belief that support is available when needed [22]. Second, support can take various forms: instrumental, informational, or emotional. CAT primarily provides emotional support, which involves expressions of sympathy, caring, esteem, value, or encouragement [23]. Additionally, such social relationships extend beyond human interactions to encompass connections with animals. As McNicholas and Collis [24] noted, pets can complement existing human support networks and, in some cases, even serve as substitutes for lacking human connections. Pets provide support that releases individuals from relationship obligations, facilitates reorganization, and helps rebuild routines. In this context, CAT presents a unique type of social support that nurtures emotional connection in individuals. Furthermore, having a supportive social network has a direct impact on wellbeing, especially when it is large and is perceived to offer support, which can improve quality of life and reduce stress [25]. Consequently, the dogs and handlers involved in CAT play a crucial role in engaging individuals in social interactions and offering social support, thereby buffering them from stress and anxiety [21].

Recognizing that heightened stress and anxiety can decrease students’ quality of life and performance [26], CAT has emerged as a promising intervention across diverse populations. Previous research has linked CAT and animal-assisted interventions to mental health improvements such as reduced anxiety and pain-related symptoms in children [27, 28], as well as decreased PTSD symptoms in mixed-age samples [29, 30]. These findings support CAT’s broader potential as an intervention for promoting psychological well-being. Despite this, research specifically examining CAT’s impact within university settings remains limited. We identified nine relevant meta-analyses on animal-assisted therapy (Table 1), but only one focused on higher education [31]. While Huber et al. [31] explored a range of mental, cognitive, and physiological outcomes, with stress and anxiety as only part of the focus, their review also included general animal-assisted interventions (AAIs) involving various animals beyond dogs. Although that work provides helpful insights, our study takes a more focused approach to examine professionally facilitated CAT and its effects on stress and anxiety in university students. By doing so, we aim to build on existing literature and provide a clearer picture of how CAT may help support mental health in this population.

Table 1.

Past Meta-Analyses

Author, Year	Age Range	Outcomes Measured	Animals Involved	Intervention Type	Effect Size (CI)
Nimer & Lundahl, 2007 [32]	0–65+	Autism spectrum behaviors	Dogs, cats, rabbits, horses, aquatics, others	Animal-assisted therapy	Cohen’s d 0.72 (0.23, 1.22)
Virues-Ortega et al., 2012 [33]	40+	Anxiety	Dogs, cats, birds, rabbits, dolphins, others	Animal-assisted therapy	Hedge’s g -0.29 (-0.51, -0.07)
Charry-Sánchez et al., 2018 [34]	3–16	Gross motor function in cerebral palsy	Horses, dogs, elephants, dolphins	Animal-assisted therapy	Cohen’s d 1.61 (-2.00, 5.23)
Germain et al., 2018 [29]	8–62	Trauma-related symptoms	Horses, dogs, others	Animal-assisted psychotherapy	Hedge’s g 0.86 (0.53, 1.18)
Waite et al., 2018 [35]	2–88	Pain, anxiety, distress	Dogs	Animal-assisted intervention	Hedge’s g 1.65 (0.46, 2.83)
Hediger et al., 2021 [30]	4–86	Post-traumatic stress disorder symptoms	Dogs, horses, dolphins, sheep, farm animals, seals, others	Animal-assisted intervention	SMD -0.26 (-0.56, 0.04)
Feng et al., 2021 [27]	3–17	Anxiety	Dogs	Canine-assisted therapy	SMD -0.19 (-0.44, 0.06)
Zhang et al., 2021 [28]	3–18	Pain-related symptoms	Dogs	Animal-assisted therapy	MD -0.53(-0.77, -0.30)
Huber et al., 2024 [31]	18–22	Anxiety	Dogs, cats, fish, others	Animal-assisted intervention	Hedge’s g -0.57, (-1.45, 0.31)

Note: Only 1 meta-analysis was identified to be done in university settings [31]; 5 studies included other animals rather than just therapy dogs; 2 studies restricted the analyses to randomized controlled trials only [31, 34]; SMD = Standardized Mean Difference; MD = Mean Difference

Therefore, this paper aims to systematically review the effectiveness of canine-assisted therapy (CAT) in reducing stress and anxiety among university students. To address this gap, we conducted a meta-analysis focusing specifically on randomized controlled trials (RCTs) to establish a robust cause-and-effect relationship between CAT and mental health outcomes. By synthesizing evidence from these trials, we aim to provide a clearer understanding of CAT’s potential in supporting student mental wellness.

Methods

PROSPERO registration

Our systematic review and meta-analyses were developed following the PRISMA 2020 statement [36]. As of April 2025, our PROSPERO registration record has ID number CRD42024518138.

Study selection

We focused on studies involving undergraduate and graduate students of all genders, including those who prefer not to disclose their gender. Students clinically diagnosed with anxiety disorders were also eligible for inclusion, given previous research demonstrating the effectiveness of canine-assisted therapy (CAT) in treating anxiety [37].

This meta-analysis exclusively examined CAT. While conducting an overview of all similar treatments assisted by animals, we observed a standard confusion in the concepts and experimental designs of these interventions. For instance, it was often unclear whether the animals and caregivers involved were professionally trained, or whether handlers were present during sessions. Many animal-assisted interventions share overlapping features but also differ in important ways. However, inconsistent terminology and vague criteria make it hard to clearly distinguish between different types of animal-assisted therapy.

To address this, we used strict inclusion criteria to define CAT in our study. For an intervention to be considered CAT in our study, it must have two key criteria: (1) the incorporation of professionally trained dogs certified by relevant organizations; and (2) the dog handler must be present during the treatment [38]. The standardization of incorporating professionally trained, certified dogs and requiring the presence of a handler during CAT is crucial for the safety, therapeutic efficacy, and ethical treatment of the therapy dog [39]. Certified dogs are trained to remain calm, responsive, and adaptable in various therapeutic settings, reducing risks (e.g., agitation or biting) associated with patient inappropriate behaviors such as abrupt gestures, rough handling, and direct interactions [39, 40]. The handler plays a key role in supervising interactions, recognizing and mitigating signs of stress in the dog, and upholding ethical standards during sessions. Research shows that therapy dogs can experience stress or burnout if not properly managed, which may reduce the therapy’s effectiveness and negatively impact animal welfare [39–41]. Professional organizations like IAHAIO and AAII emphasize the importance of handler involvement to ensure safe, ethical, and effective CAT sessions. Without these elements, both the outcomes and integrity of the intervention could be compromised [39, 42]. For studies included in our analysis, if the presence of a handler was not explicitly stated, we assumed a handler was present since this is standard practice in CAT. With these established criteria, even though some papers might not name the intervention particularly as canine-assisted therapy, we still included them in our review as long as they involved trained therapy dogs and professional handlers.

In contrast to CAT, several related interventions, such as animal visitation programs (AVPs), animal-assisted psychotherapy (AAPT or AAP), and broader animal-assisted interventions (AAI), were excluded from this study. AVPs may involve informal interactions with animals that are not professionally trained, which does not meet this study’s requirement of incorporating certified therapy dogs [43]. AAPT, on the other hand, is a goal-directed, clinical intervention conducted by professional psychotherapists, where the animal is integrated as a key part of the therapeutic process aimed at reducing symptoms [44]. Therefore, it primarily focuses on patients with clinically diagnosed mental illnesses [45]. For example, an intervention led by a psychologist that teaches mindfulness by encouraging participants to pet a dog’s fur with the explicit goal of reducing anxiety would qualify as animal-assisted psychotherapy [44]. In contrast, CAT does not require clinical diagnoses or licensed therapists and is typically used in broader wellness or prevention contexts, such as providing stress relief for university students who may not have a diagnosed mental illness.

Broader animal-assisted interventions (AAI) were excluded for two main reasons. First, AAI is an umbrella term that includes a range of practices, such as animal-assisted activities (AAA), animal-assisted education (AAE), and animal-assisted therapy (AAT) [46]. AAT itself can involve a variety of animals, including dogs, cats, guinea pigs, birds, dolphins, and others [33]. CAT is a specific subtype of AAT that focuses exclusively on professionally trained dogs. Because this study centers on dog-based interventions, only those falling under CAT were included. Second, AAA and AAE may involve staff from various backgrounds, including educators, activity coordinators, or volunteers [47], while CAT requires trained dog handlers to ensure safety and structure. Therefore, AAA and AAE, which may not consistently involve trained animals and handlers, were excluded from this study.

The main comparator in this study was the control group of students who were not exposed to the canine-assisted therapy intervention. Control group designs may vary across studies, such as waitlist control, only handler control, watching videos about dogs’ control, and completing unrelated tasks. Our study also focused exclusively on randomized controlled trials to establish a causal relationship between the intervention and its effects on students’ stress and anxiety levels. We did not include cross-sectional or within-subjects pre-post designs, but only studies that consist of canine-assisted therapy as the treatment group and other forms of control groups.

Outcomes

The primary outcome was the decrease in reported stress and anxiety levels in the group exposed to canine-assisted therapy. Studies were included if they reported either or both constructs. Measurement of stress and anxiety levels was based on standardized and validated self-rating scales; we did not consider observers’ ratings. According to the American Psychological Association [48], anxiety is “an emotion characterized by feelings of tension, worried thoughts, and physical changes like increased blood pressure.” Since anxiety could be categorized into state and trait anxiety, we focused on state anxiety when studies reported anxiety levels separately. State anxiety is conceived as a situational increase in feelings of insecurity, while trait anxiety is a relatively stable personality characteristic [49]. With most canine-assisted therapy being implemented only occasionally in universities, examining its effect on state anxiety would be more beneficial for future recommendations, such as implementing it before stressful events like exams. Meanwhile, stress is defined as a state of worry or mental tension caused by a difficult situation [50]. Additionally, to explore potential moderators of the effectiveness of canine-assisted therapy in reducing stress and anxiety, we conducted subgroup analyses based on variations in intervention duration, control condition type, baseline stress/anxiety levels, and risk-of-bias ratings.

Sources and search strategy

We conducted a systematic literature search on April 10, 2025, across several primary databases: APA PsycInfo, PubMed, Duke Libraries, CNKI, and Wanfang. To maximize sensitivity, we did an all-field search in each database rather than limiting our query to titles and abstracts. In addition, we used Google Scholar as a supplementary source to identify any additional peer-reviewed publications and hand-searched the reference lists of relevant systematic reviews and included studies for potentially eligible publications. All references were imported into Covidence for deduplication and subsequent screening.

Our search strategy focused on three core concepts: the population (university students), the intervention (canine-assisted therapy and related animal-assisted interventions), and the outcomes (stress and anxiety). For example, we combined terms such as “college,” “university,” “undergraduate,” and “graduate” for the population; “canine-assisted therapy,” “dog-assisted therapy,” “animal-assisted therapy,” and similar variants for the intervention; and “stress,” “stress levels,” “anxiety,” and “anxiety levels” for the outcomes. This approach ensured that studies addressing all three concepts were retrieved. Detailed search strings for each database are provided in the supplementary materials. An example of the search string is as follows:

(“canine-assisted therapy” OR “canine therapy” OR CAT OR “dog-assisted therapy” OR “dog therapy” OR “canine-assisted intervention” OR “animal-assisted therapy” OR “animal therapy” OR “animal-assisted intervention” OR AAI) AND (college OR undergraduate OR graduate OR “higher education” OR “university students”) AND (stress OR “stress levels” OR anxiety OR “anxiety levels”).

Two reviewers (SS and ZL) independently screened titles and abstracts based on predefined inclusion criteria. Eligible studies were those mentioning canine-assisted therapy, targeting university students, and reporting measures of stress or anxiety. Articles that passed this initial screening underwent full-text review to verify study design, intervention implementation (including the use of trained therapy dogs and handlers), and outcome measurement. A checklist based on mutually agreed criteria was used during screening (see supplementary materials), and any disagreements were resolved through discussion. We restricted our search to randomized controlled trials published in English or Chinese. Only peer-reviewed, published studies were considered; gray literature was excluded. Our search included studies published up to April 2025, and any articles published after this were not considered for inclusion.

Data extraction

Data extraction was completed by both reviewers independently using an Excel sheet. We collected data on study designs, study participants, sample sizes, treatment conditions, control conditions, baseline measurements, reported post-measurements, duration of interventions, and the risk of bias. Risk of bias was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool [51]. Each study was evaluated for randomization process, deviations from the intended intervention, handling of missing outcome data, consistency of outcome measurement, and selection of reported results, with each domain rated as “low risk,” “some concerns,” or “high risk.”

To compute the effect sizes of each study, we primarily extracted post-means and standard deviations from treatment and control groups, following the recommendation by Lipsey and Wilson [52] for practical meta-analysis. If means and standard deviations were not reported, we extracted t-test values instead. For studies reporting results from ordinary least squares (OLS) regression models with a dummy-coded treatment variable (e.g., treatment = 1, control = 0), we extracted the unstandardized regression coefficient, its standard error, the standard deviation of the dependent variable, and group sample sizes to compute standardized effect sizes. These were included only when the regression compared treatment versus control within an RCT. If data such as sample sizes, intervention duration, or baseline values were missing, we contacted the authors to request the information. Missing data with no author response were imputed through mean substitution; in the case of subgroup analyses, such studies were excluded. Any disagreements during the extraction process were resolved through discussion, and when needed, we consulted the supervising author (CN).

Data analysis

We conducted a random-effects meta-analysis using Stata/SE 18.0 as we anticipated both clinical and methodological heterogeneity among canine-assisted therapy studies [31]. Given the considerable variability in intervention protocols, participant characteristics, and outcome measures in this field, the random-effects model is most appropriate because it assumes that each study estimates a different true effect [53], thus providing a more conservative and generalizable estimate of the overall intervention effect.

Effect sizes (Cohen’s d) were computed using the extracted post-intervention means and standard deviations, or t-test/regression statistics when necessary. The practical meta-analysis effect size calculator [54] was used to calculate effect sizes and standard errors for the meta-analysis. Subgroup analyses were conducted to examine the impact of intervention duration, differences based on risk-of-bias levels, and the effect of control condition type on intervention effectiveness. Control conditions were classified as passive if they involved no structured engagement (e.g., waitlist, routine academic activities) and as active if they involved structured or interactive components (e.g., interacting with handler-only sessions, watching animal-related videos). To assess the robustness of our findings for the intervention duration subgroup, we also conducted a sensitivity analysis by excluding multi-session studies and re-running the analysis.

For both meta- and subgroup analyses, we reported Hedges’ g, 95% confidence intervals, and p-values. The chi-square test of group differences was also reported for subgroup analyses to determine if effect sizes differed significantly across groups. We considered Hedges’ g values of 0.20, 0.50, and 0.80 as small, medium, and large overall intervention effect sizes [55], respectively, and a p-value less than 0.05 as significant. Heterogeneity across studies was assessed using the I² statistic. We interpreted I² values of 0–30% as insignificant, 30–60% as moderate, 60–75% as substantial, and 75–100% as considerable heterogeneity [56]. Small-study bias was assessed through Egger’s test, while potential publication bias was evaluated using trim-and-fill analysis and funnel plots. All tests and analyses were conducted using Stata/SE 18.0’s built-in functions; no external packages were used.

In addition, to account for differences in how baseline stress and anxiety were measured across studies, we standardized the baseline scores using z-scores. First, we calculated the pooled mean baseline score by weighting each study’s baseline score according to its sample size. We then estimated the pooled standard deviation by combining the variances from all studies while adjusting for the total number of studies. Finally, we computed a z-score for each study, representing how much its baseline stress and anxiety level deviated from the overall pooled mean. Based on these z-scores, we categorized studies into low (z < 0) and high (z > 0) baseline stress/anxiety groups in the later section to conduct a subgroup analysis.

Results

Study selection

After running our search strategy across the selected databases, we retrieved a total of 290 records (Fig. 1). Automatic deduplication in Covidence identified 34 duplicates, and an additional 2 duplicates were removed manually, leaving 254 articles for screening. Following the title and abstract screening, 224 records were excluded for not meeting the eligibility criteria (e.g., incorrect population, intervention, or study design), leaving 30 articles for full-text review. Of these, 15 were further excluded during full-text review for reasons such as not using a randomized controlled trial design, not measuring stress or anxiety outcomes (e.g., measured general mood only), involving mixed-species interventions (e.g., both cats and dogs), or using an intervention format outside the scope of our review (e.g., a therapy dog passively wandering in a lecture hall). Specific reasons for exclusion at the full-text stage are provided in the supplementary materials. No additional eligible studies were identified from the Chinese databases.

Fig. 1.

PRISMA flow chart. A total of 290 records were identified from databases, and 36 studies were removed before screening. Of 254 studies that were screened for titles and abstracts, 30 studies were eligible for full-text review. 15 studies were included in the review and 14 in the meta-analysis after finalization

In total, 15 individual studies [13, 21, 57–69] met al.l inclusion criteria. One study [61] met al.l inclusion criteria at the abstract and full-text screening stages; however, it was excluded from the meta-analysis due to insufficient statistical reporting. Specifically, the study did not provide post-intervention means, standard deviations, t-values, or enough information (e.g., standard error of regression coefficients) to compute an effect size, and our attempts to extract usable data did not meet our predefined extraction requirements. The remaining 14 studies were included in the meta-analysis.

Study characteristics

Most of the included studies were conducted in Western countries– 3 in Canada, 9 in the USA, and 3 in the UK– with a total sample of 1868 university students. Individual sample sizes ranged from 39 to 357 participants (M = 124, SD = 87). In most studies, the number of female participants exceeded that of males, with an overall male-to-female ratio of 1:3 (Male N = 424; Female N = 1322), and the average age of participants was 19.61 years (SD = 2.37). Most studies did not screen for prior anxiety or mental illness diagnoses; only one reported that 33% of participants had previously been diagnosed with an anxiety disorder [57]. All interventions involved live, in-person interactions with a therapy dog accompanied by a trained handler. Intervention session durations varied considerably, from as short as 1 min to as long as 120 min (M = 32, SD = 31.6). All studies allowed students to interact freely with therapy dogs; students could pet, sit with, or play with the dogs based on their preferences. Most studies implemented a single-session intervention; however, two studies administered the intervention over 6 and 7 weeks, respectively [60, 64]. Control conditions also varied. Examples included waitlist controls [64], watching videos of dogs [63], and completing the Family Life-Space Diagram [13]. The waitlist condition was the most common (N = 3), while a larger number of studies used active controls (e.g., engagement with handlers, interaction with other participants, or viewing non-animal-related presentations; N = 8) compared to passive controls (e.g., continuation of daily routines such as studying; N = 7).

Regarding outcome measures, 6 studies (40%) assessed anxiety using the State-Trait Anxiety Inventory (STAI), 5 studies (33%) evaluated stress via versions of the Perceived Stress Scale (PSS), and the remaining studies used alternative tools such as momentary stress ratings or diagram-based measures like the Family Life-Space Diagram. Although the majority of studies reported reductions in stress and anxiety levels in the intervention group, four of them did not yield statistically significant results [13, 58, 60, 68]. Additional details on study designs and individual study results are provided in the Summary Table of Papers (Table 2).

Table 2.

Summary table of papers (N = 15)

Paper	Sample Size	Intervention Duration	Control	Measured Outcome	Scale	Estimates of Cohen’s d	Estimates of Risk of Bias
Anderson & Brown, 2021 [57]	N = 89	35–45 min	Interaction with other participants in a separate room without dogs present	Anxiety	State-Trait Anxiety Inventory	-0.79	High
Barker et al., 2016 [13]	N = 57	15 min	Completion of the Family Life-Space Diagram (attention-control)	Perceived Stress	Perceived Stress Scale	-0.38	High
Barker et al., 2017 [58]	N = 74	15 min	Completion of the Family Life-Space Diagram (attention-control)	Perceived Stress	Family Life Space Diagram	-0.36	High
Binfet, 2017 [59]	N = 163	20 min	Engagement in a typical academic activity, such as studying course material	Perceived Stress	Perceived Stress Scale	-0.32	High
Binfet et al., 2021 [21]	N = 284	20 min	Interaction with a canine handler only	Momentary Stress	Momentary Stress (One item)	-0.99	Medium
Carr & Pendry, 2025 [60]	N = 145	7 sessions 2 h/session	Waitlist control	Anxiety	Beck Anxiety Inventory	-0.03	Medium
Crump & Derting, 2015 [61]	N = 61	30 min	Placement in another room	Perceived Stress	Perceived Stress Scale	NA	High
Grajfoner et al., 2017 [62]	N = 132	20 min	Interaction with a canine handler only	Anxiety	State-Trait Anxiety Inventory	-1.16	Medium
Haefelin et al., 2020 [63]	N = 165	1 min	Watching a video of playful dogs on a provided laptop	Anxiety	State-Trait Anxiety Inventory	-0.99	High
Kivlen et al., 2022 [64]	N = 104	6 sessions 35 min/session	Waitlist control	Anxiety	Patient-Reported Outcomes Measurement Information System	-0.40	Medium
Manville et al., 2022 [65]	N = 60	10 min	Watching a PowerPoint presentation with neutral, non-animal images	Anxiety	State-Trait Anxiety Inventory	-0.65	High
Spruin et al., 2020 [66]	N = 94	30 min	Standard treatment for student stress and anxiety: a one-on-one session with a student advisor	Anxiety	State-Trait Anxiety Inventory	-1.22	High
Trammell, 2019 [68]	N = 44	Not mentioned	Placement in a similar room, but without dogs present	Perceived Stress	Perceived Stress Scale	-1.34	Medium
Ward-Griffin et al., 2018 [69]	N = 357	90 min	Waitlist control	Perceived Stress	Perceived Stress Scale	-0.13	High
Williams et al., 2018 [67]	N = 39	12 min	Engagement in normal pre-exam routine, such as independent studying	Anxiety	State-Trait Anxiety Inventory	-0.96	High

Note. N = Analyzed Participants; All studies were conducted in Western countries (e.g., Canada, USA, and UK). All interventions involved students interacting with both dogs and handlers concurrently, while the control design varied across studies. Estimates of Cohen’s d were computed using post-intervention means and SDs of the intervention and control groups or t-test/regression statistics

Main effect

Results from random-effects meta-analysis revealed a statistically significant overall effect size of g = -0.672 (95% CI [-0.892, -0.452], z = -5.99, p <.001), indicating a moderate effect. However, four specific studies [13, 58, 60, 68] had no statistically significant effects. Among all studies, Trammell [67] demonstrated the strongest effect size of -1.342 (95% CI [-1.997, -0.687]), while Ward-Griffin et al. [68] reported the smallest significant effect size of -0.129 (95% CI [-0.379, 0.122]). Additionally, high heterogeneity was observed across the included studies (I² = 72.81, p <.001), suggesting substantial variability between intervention effects (Fig. 2).

Fig. 2.

Forest plot. Effect sizes were computed in Hedges’ g; CI = Confidence Interval; I² = Heterogeneity Value; Vertical solid line represents the null effect; Vertical dash line represents the weighted mean difference

Furthermore, the funnel plot (Fig. 3) raises concerns regarding potential publication bias, as 5 out of 14 studies fall outside the 95% confidence interval. For example, the largest effect size of -1.34 by Trammell [67] originated from the study with the second smallest sample size (N = 44), suggesting possible small-study effects. To formally assess this, we performed an Egger test, which yielded a non-significant result (z = -1.27, p =.21). Even though we did not find significant results for small study bias, it is still important to note that Egger’s test has low power to detect this bias when there are few studies. Thus, we complemented it with a trim-and-fill analysis for a more comprehensive assessment of potential publication bias.

Fig. 3.

Funnel Plot. Evidence of potential publication bias is suggested, with 5 out of 14 studies falling outside the 95% confidence interval

Trim and fill analysis

The trim and fill analysis imputed two potentially missing studies on the right side of the funnel plot, suggesting that smaller or non-significant studies may be underrepresented. The observed effect size was − 0.672 (95% CI [-0.892, -0.452]), while the adjusted effect size after imputation was − 0.592 (95% CI [-0.816, -0.368]). Although this adjusted effect size is slightly smaller in magnitude, it still reflects a moderate and statistically significant effect of canine-assisted therapy in reducing stress and anxiety among university students. The adjusted funnel plot (Fig. 4) showed a more symmetrical distribution, providing some evidence that the trim-and-fill method helped account for funnel plot asymmetry.

Fig. 4.

Adjusted Funnel Plot. A more symmetrical distribution is presented, with 2 missing studies imputed on the right side of the plot

However, as suggested by Sterne et al. [51], the trim-and-fill procedure assumes that funnel plot asymmetry is solely due to publication bias and may not perform well when substantial heterogeneity exists. Given the high heterogeneity observed in our meta-analysis, these results should be interpreted with caution. Other factors, such as clinical diversity or methodological variability across studies, may also contribute to the initial asymmetry.

Risk of bias

Across 5 risk of bias domains and 75 total outcomes, 17 of them were rated as low-risk, 43 were evaluated to have some concerns, and 12 were identified as high-risk. Overall, no studies were assessed as low-risk; 5 out of the 14 studies had some concerns, while the remaining were classified as high-risk. Similar to previous meta-analyses conducted on animal-assisted interventions, such as Huber et al. [31], we found that most studies could not implement blinding procedures since the presence of therapy canines during the intervention made it obvious to both researchers and possibly the participants that they were in the treatment condition. This, in turn, might have likely influenced the self-reported outcomes, thereby causing many studies to be rated as “some concerns” and high-risk. The overview of the risk of bias results can be referred to in Fig. 5, and specific details can be found in the supplementary materials.

Fig. 5.

Summary of the risk of bias. No studies were identified to have an overall low risk of bias

Subgroup analyses

Risk of bias

To address potential biases arising from variations in the risk of bias among studies included in our meta-analysis, we conducted a subgroup analysis comparing the effect sizes of studies categorized as “some concerns” and “high-risk”. The pooled effect size for studies with some concerns was − 0.780 (95% CI [-1.232, -0.328]), while for high-risk studies, it was − 0.607 (95% CI [-0.859, -0.456]). We further evaluated the differences in effect sizes between these groups using a chi-square test of group differences, yielding a non-significant result (Q_b = 0.43, p =.513). In general, our findings suggest that while studies with some concerns showed a larger effect size on average compared to high-risk studies, this difference was not significant.

Baseline differences

Another potential bias in our analysis is that the samples across studies were not homogeneous in their baseline stress and anxiety levels, which may have either overestimated or underestimated the effect of CAT. To account for this bias, we divided the studies’ baseline z-scores into two groups: a low baseline stress/anxiety group (z < 0) and a high baseline stress/anxiety group (z > 0). We then conducted a subgroup analysis to examine whether CAT’s effectiveness varied depending on students’ initial stress and anxiety levels.

Our analysis revealed that the pooled effect size for the low stress/anxiety group was − 0.494 (95% CI [-0.778, -0.209]), which is smaller in magnitude than the pooled effect size of -0.909 (95% CI [-1.225, -0.593]) for the high stress/anxiety group. The chi-square test of group differences indicated a statistically significant difference (Q_b = 3.67, p <.05) between the two groups. Furthermore, heterogeneity was more pronounced in the low stress/anxiety group (I² = 73.17, p <.001) compared to the high stress/anxiety group (I² = 59.65, p <.05). Overall, our subgroup analysis of baseline differences suggests that variations in students’ pre-intervention stress and anxiety levels likely contributed to the heterogeneity of the main outcome.

Passive and active controls

To examine whether the type of control condition influenced the effectiveness of canine-assisted therapy (CAT), we conducted a subgroup meta-analysis comparing studies using passive versus active control conditions. Interventions compared to passive controls (e.g., waitlist or typical study routines) yielded a pooled effect size of -0.503 (95% CI [-0.815, -0.191]), while those compared to active controls (e.g., interaction with a handler or engagement in attention-matched tasks) showed a stronger effect size of -0.834 (95% CI [-1.046, -0.622]). Although the test of subgroup differences did not reach statistical significance (Qb = 2.96, p =.085), the results suggest a potential trend toward greater CAT effectiveness when compared with more actively engaging control conditions.

Intervention duration

To explore potential differences in intervention effectiveness based on duration, we categorized sessions into three groups: 0–14 min (short), 15–29 min (medium), and ≥ 30 min (long). Currently, there are no standardized thresholds for session length in canine-assisted interventions. However, prior research suggests that both brief (5 min) and longer (20 min) sessions can significantly reduce stress [59], and one study found no significant differences in outcomes between sessions lasting 2-, 5-, and 10 min [70]. Given this variability and the absence of established standards, we selected our cutoffs to reflect common ranges reported in the literature. This approach may help clarify how session length relates to intervention effectiveness and improve replicability in future studies.

In the short-duration group (0–14 min), the pooled effect size was − 0.883 (95% CI [-1.170, -0.596]), with no observed heterogeneity (I² = 0.00, p =.564), suggesting a consistent and strong effect of brief sessions in reducing stress and anxiety. This consistency is notable given that one of the interventions [63] was as short as 1 min. The medium-duration group (15–29 min) also demonstrated a reduction in stress and anxiety, with a pooled effect size of -0.651 (95% CI [-1.001, -0.300]). However, substantial heterogeneity was observed (I² = 74.52, p <.01), indicating variability across studies. Similarly, the long-duration group (≥ 30 min) yielded an effect size of -0.497 (95% CI [-0.906, -0.088]) with high heterogeneity (I² = 79.66, p <.001), suggesting that longer sessions may be effective but vary in impact.

A chi-square test of group differences indicated that the differences between duration categories were not statistically significant (Qb = 2.31, p =.283). Therefore, while all durations show beneficial effects, no specific session length appeared as significantly more effective than others.

Two of the included studies implemented multi-session interventions lasting several weeks [60, 64]. To examine their influence, we also conducted a sensitivity analysis excluding these studies. The effect sizes and direction of results remained consistent, and no meaningful differences were observed across subgroups. This supports the robustness of our findings since the inclusion of multi-session studies did not substantially alter conclusions about intervention duration.

Discussion

Overall, our meta-analysis demonstrates that canine-assisted therapy (CAT) effectively reduces university students’ stress and anxiety, with a moderate pooled effect size. However, the substantial heterogeneity suggests that CAT’s effectiveness may vary depending on the study context. Although Egger’s test was not significant, the fact that one of the smallest trials produced the largest effect size points to possible small-study effects and calls for cautious interpretation.

Subgroup analyses help clarify some of this variability. Specifically, CAT produced a significantly larger effect among students with high baseline stress and anxiety compared to those with lower baseline levels. This finding suggests that CAT may function more effectively as a targeted preventive intervention, particularly for individuals already experiencing heightened psychological distress. Additionally, there were no significant differences in effect sizes between passive control groups (e.g., waitlist) and active control groups (e.g., attention-matched activities), suggesting that the observed benefits of CAT are not solely due to novelty or attention effects.

Although the shortest intervention duration (0–14 min) group showed the largest effect size, subgroup differences by session length were not statistically significant. This implies that, while intervention length might influence outcomes, no specific duration emerged as definitively superior. Instead, the consistent positive effects across all durations indicate CAT’s adaptability and its potential to be tailored flexibly to various program structures.

While potential publication bias cannot be entirely ruled out, given the funnel plot asymmetry and the imputation of two potentially missing studies via trim-and-fill analysis, the adjusted effect size remained statistically significant. This strengthens the overall conclusion that CAT is a promising intervention for improving student mental health.

From a theoretical standpoint, our findings align with social support theory. That is, interactions with therapy dogs and handlers may serve as a source of social support, resulting in reduced students’ anxiety and stress levels [22]. This unique form of social support provided by therapy dogs complements human support networks, as suggested by McNicholas and Collis [14, 24]. The non-judgmental, unconditional support facilitates emotional relief, helping individuals reduce their stress and anxiety levels. In particular, CAT facilitates emotionally safe, low-stakes social interactions, which may serve as an important buffer against stress and anxiety.

In summary, this study contributes to the growing body of literature supporting the effectiveness of CAT in reducing stress and anxiety, while also offering new insights into potential moderating factors such as baseline distress and intervention duration. It advances the field through rigorous inclusion of RCTs and by focusing specifically on university students, an underrepresented population in previous meta-analyses.

Limitations and future directions

Included evidence

Similar to the previous AAI meta-analysis by Huber et al. [31], a key limitation of the included studies is the predominance of female participants, with proportions ranging from 60 to 95%. This imbalance in participant demographics may limit the generalizability of the findings to the broader university student population, particularly given gender-based differences in stress perception and coping styles [50, 71]. The overrepresentation of female participants is likely due to recruitment practices in fields such as psychology and nursing, which tend to have higher proportions of female students. Future research should aim to address this limitation by recruiting more gender-diverse samples and by examining potential gender differences in CAT’s impact on stress and anxiety reduction among university students.

Another limitation is the high risk of bias present in many studies, largely due to factors inherent to CAT interventions. First, carers and individuals delivering the interventions were often aware of participants’ group assignments, as therapy dog interactions are conspicuous (participants would see and interact with dogs), making blinding infeasible. Second, outcome assessments may have been influenced by prior knowledge of CAT’s potential benefits, as many participants were psychology students who may have been familiar with the effects of human-animal interactions. Third, a substantial number of studies lacked pre-registration, meaning analyses were not always conducted according to a pre-specified plan finalized before unblinded outcome data were available, raising concerns about outcome reporting flexibility. Future research should adopt stronger trial designs, including pre-registration of outcomes and the use of validated, objective stress and anxiety biomarkers to enhance internal validity.

Review process

In addition, the studies included in this review exhibited high methodological heterogeneity, which may have influenced the interpretation of our results. We propose that this heterogeneity is partly attributable to the broad range of criteria for animal-assisted therapy (AAT), referred to in this analysis as canine-assisted therapy (CAT). Currently, there is no standardized theoretical framework for evaluating the effectiveness of AAT/CAT, leading to wide variations in intervention protocols, treatment goals, and outcome measures. For example, the duration and frequency of therapy dog interactions varied considerably, ranging from brief one-minute sessions to multi-week programs. Differences were also observed in the design of control conditions and outcome assessments, further contributing to methodological inconsistencies. Therefore, addressing these challenges by standardizing research procedures, such as establishing consensus definitions and treatment guidelines, would help improve replicability and intervention reliability in future CAT studies.

It is also worth noting that although this study aligns with social support theory, we did not directly measure social support factors such as emotional support, feelings of connectedness, or social isolation. Future studies should include explicit assessments of these constructs to better understand the mechanisms by which CAT influences mental health. By examining the role of therapy dog interactions in fostering a sense of connection and emotional relief, future studies can have deeper insight into the broader psychological benefits of CAT and clarify how social support mediates its effects on stress and anxiety reduction.

Implications

The implications of these findings are important for university administrators considering the implementation of CAT programs on their campuses. Despite the methodological limitations identified, CAT is still a promising solution for addressing the growing mental health challenges faced by university students, especially in the aftermath of the COVID-19 pandemic. Its cost-efficiency, scalability, and high-popularity, combined with demonstrated effectiveness in reducing stress and anxiety, make CAT a feasible option for institutions aiming to promote student well-being.

Universities may also consider integrating CAT into broader, multifaceted mental health strategies by pairing it with existing services such as counseling, mindfulness programs, or peer-led support groups. This integrated approach could encourage greater student engagement and enhance the overall effectiveness of mental health initiatives. Moreover, institutional investment in structured CAT programs (clear protocols for screening, supervision, and feedback) would contribute to the sustainability and standardization of these interventions, ensuring long-term benefits for student populations.

Conclusion

This meta-analysis provides strong evidence that canine-assisted therapy (CAT) is an effective intervention for reducing stress and anxiety among university students, with an overall moderate effect size and consistent benefits across session durations and control conditions. These findings suggest that CAT can be successfully implemented in higher education settings, especially for students experiencing high levels of pre-existing stress or anxiety. Importantly, this study contributes to existing literature by isolating the effects of CAT within university student populations. While methodological limitations remain, the results support the inclusion of CAT as part of a complementary suite of mental health interventions within universities.

Future research should focus on improving study design rigor, exploring long-term outcomes, and examining CAT’s differential effectiveness across diverse student subgroups. Establishing standardized intervention protocols and directly assessing mechanisms, such as social support and emotional bonding, will also be essential to optimizing CAT’s effectiveness. Ultimately, a deeper understanding of CAT’s therapeutic potential will help position this unique intervention as a valuable, accessible, and student-centered approach to mental health care in higher education.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

AAI

Animal-Assisted Interventions

AAA

Animal-Assisted Activities

AAE

Animal-Assisted Education

AAPT

Animal-Assisted Psychotherapy

AAT

Animal-Assisted Therapy

AVP

Animal Visitation Program

CAT

Canine-Assisted Therapy

RCT

Randomized Controlled Trial

Author contributions

SS analyzed and interpreted the data from the meta-analysis and subgroup analysis results. ZL was a major contributor to drafting the manuscript. Both SS and ZL contributed equally to study selection, screening, evaluating the levels of bias, and the overall design of the study. The second authors, ZW and SW, reviewed and edited the statistical results and manuscript. CN supervised the scientific quality of the entire meta-analytic process and revised the final manuscript.

Funding

The current study is not being funded.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and informed consent

Not applicable to secondary data.

Consent for publication

Not applicable.

Pre-Registration

PROSPERO registration record: CRD42024518138.

Competing interests

The authors declare no competing interests.