We searched relevant studies in six electronic bibliographic databases including PubMed, Cochrane Library, Embase, Web of Science, China National Knowledge Infrastructure, and VIP Data. The search was undertaken by combining search terms for horticultural therapy and stress, with multiple synonymous terms, such as “gardening” and “pressure”. All databases were searched from inception to January 2023. Detailed search steps are presented in Appendix B.

Table 1 outlines the inclusion/exclusion criteria, according to the population, intervention, comparison, outcomes, and study design (PICOS).

Studies normally utilized physiological and psychological indicators to assess the outcomes of stress-reduction effects. Physiological indicators typically include blood pressure (systolic blood pressure and diastolic blood pressure), pulse pressure, saliva cortisol levels, salivary α-amylase (sAA), pulse rate (BPM), heart rate variability (HRV), electroencephalography (EEG), skin conductance (SC), skin temperature (SKT), facial thermal imaging, etc. Psychological indicators were mainly assessed by standardized tests including the Perceived Stress Scale (PSS), the Stress and Crisis Inventory (SCI-93), the Stress Response Scale (SRS-18), the Depression Anxiety Stress Scale (DASS21), the Labor Occupational Pressure Scale, the Geriatric Depression Scale (GDS-30), the Psychosocial wellbeing Index Short Form (PWI-SF), 4T-PROs-Stress, Rehabilitation Stress Scales, etc.

We imported all studies into EndNote X8. Two independent reviewers assessed the studies based on the inclusion and exclusion criteria after removing duplicate studies. A third reviewer would be brought in when two independent reviewers had divergent opinions.

We first read the title and abstract of each study, followed by a full-text screening work to decide if it should be included in the analysis. We extracted the following information from each study: (1) basic information, including the research title, first author, and publication year; (2) basic characteristics of the research subjects, including the sample size, age, and gender distribution of people included in each group; (3) details of intervention of horticultural therapy, including intervention activities, duration and settings; (4) critical elements of bias risk assessment; and (5) the outcome indicators.

We pooled the information of the individual studies in Revman5.4 software and R 4.0.3 (R Core Team, 2020) using the “meta” package. Researchers employed a random-effects model to account for study heterogeneity and effect sizes. We employed standardized mean differences (SMDs) because of the various indicators of the stress-relieving outcomes adopted in different studies. The data was compiled using 95 % confidence intervals (CIs). We employed standard I² tests to measure statistical heterogeneity, and we ran a sensitivity analysis to assess the reproducibility and stability of the results. Forest plots were used to visualize the results. Funnel plots were created to visually evaluate publication bias, while Egger's regression test was used to statistically evaluate publication bias.

We also used subgroup analysis to investigate the effects of differences in participants, environmental settings, and interventions of horticultural therapy, accounting for a total of 11 subgroups. As for the participant-related subgroups, we coded their stressors (from education vs. occupation vs. rehabilitation), age, gender, and nationality. The subgroup of environmental settings was coded as indoor, outdoor, combined, and virtual settings. We then categorized the outdoor settings into therapeutically and non-therapeutically designed environments based on the aims and intentions of the design, and we also divided the outdoor settings into farms, gardens, campus, and parks in which horticultural therapy was carried out, to further investigate which kind of outdoor environment could be more effective in stress reduction. We coded the intervention-related subgroups according to the types of activities, duration, frequency, and course. This facilitates researchers and practitioners in developing more effective activities for horticultural therapy.

Two independent reviewers critically assessed the quality of the eligible studies. To assess the risk of bias in the included studies with RCT designs, we utilized the RCT-specific bias risk assessment tool in the Cochrane handbook for systematic reviews of treatment (Higgins et al., 2011), which assesses randomization procedure biases, allocation concealment, and selective reporting. We used the Joanna Briggs Institute (JBI) critical appraisal tools to assess studies with quasi-experimental designs.

Figure 1 outlines the evaluation procedure. We originally yielded a total of 11,383 articles from PubMed (n = 269), Embase (n = 1,342), Cochrane Library (n = 8), Web of Science (n = 9,150), China National Knowledge Infrastructure (n = 441), and VIP Data (n = 173). Five hundred sixty-one articles were eliminated due to duplication, and 10,698 were removed after screening the titles and abstracts. Of the remaining 124 studies, 17 were removed because the full text was not available, 63 because they lacked comprehensive data, four because they were off-topic, and one because it was not in English or Chinese. Eight studies were further removed because the outcome indicators were irrelevant to stress reduction and detailed reasons are presented in Appendix C. There were 31 studies included in our final analysis (Kam and Siu, 2010; Gonzalez et al., 2011; Hawkins et al., 2011; Van Den Berg and Custers, 2011; Pálsdóttir et al., 2013; Chen et al., 2015; Lee et al., 2015, 2018a,b, 2022; Dewi et al., 2017; Huang et al., 2017; Park et al., 2017a,b; Han et al., 2018; Hassan et al., 2019; Shao et al., 2020; Siu et al., 2020; Tao et al., 2020, 2022; Tu et al., 2020; Wei et al., 2020; Chalmin-Pui et al., 2021; Gong and Chen, 2021; Kim et al., 2021; Meore et al., 2021; Szczepańska-Gieracha et al., 2021; Chan et al., 2022; Curzio et al., 2022; Du et al., 2022; Odeh et al., 2022).

Appendix D summarizes the characteristics of the studies included in our analysis, of which 21 were quasi-experimental studies and 10 were randomized controlled trials. The reported studies were published between 2010 and 2022, with slightly more articles published in 2020 (n = 5), 2021 (n = 5) and 2022 (n = 6). The sample size ranged from 8 to 113 (1,036 in total). Experimental and control group activities, detailed settings and performers are presented in Appendix E.

Allocation concealment and outcome assessment blinding were rated as unclear risks, whereas five studies did not describe in detail the method of random sequence generation and six studies had instances of participation withdrawal due to incomplete outcome data. The majority of studies were found to be of low risk of bias. We followed the JBI critical appraisal checklist to assess the quasi-experimental studies involved. Figure 2 shows the results of the risk evaluation.

Thirteen quasi-experimental studies and five studies with RCT designs were adopting physiological indicators to assess the stress-reduction effects, while ten quasi-experimental studies and six studies with RCT designs adopted psychological indicators. Therefore, we used SMDs to manage the differences in measurements, and the meta-analysis was estimated under a random-effects model.

Figure 3 shows the effects on the physiology indicators, with the outcomes slightly varied (SMD = −0.10, 95% CI [−0.24, 0.03], p = 0.13, I² = 83%) in terms of the influence of horticultural therapy on stress. We detected significant differences in the sensitivity analyses when removing (Tu et al., 2020) (SMD = −0.05, 95% CI [−0.15, 0.05], p = 0.33, I² = 73%).

In comparison, the psychological effectiveness was more significant (SMD = −0.73, 95% CI [−0.91, −0.54], p < 0.0001, I² = 44%), as shown in Figure 4. We removed all the studies included in this meta-analysis one by one. When the study of Meore et al. (2021) and Chan et al. (2022) was removed, the results showed that heterogeneity was reduced (SMD = −0.68, 95% CI [−0.86, −0.51], p < 0.0001, I² = 35%; SMD = −0.68, 95% CI [−0.86, −0.50], p < 0.0001, I² = 35%).

We used subgroup analysis to investigate the effects of differences in participants, environmental settings, and interventions of horticultural therapy. Figure 5 shows the subgroup analysis outcomes.

Funnel plots were created to visually evaluate publication bias. The funnel plot showed an approximate symmetrical distribution of study effect size, which suggests that there might not be any publication bias (Figure 6). Furthermore, Egger's regression test was used to statistically evaluate publication bias. The bias coefficient of Egger's test was < 0.0001, so there was a possibility of publication bias.

Stress is often linked to complicated stressors (Chauhan et al., 2015), such as individual factors, relationship characteristics, health, work and education, community, finances, and the environment (Brannen et al., 2009). There was a limited number of studies identifying participants' stressors. Future research with clearly defined stressors is needed to develop stress reduction strategies for specific stressors and to improve the practice of horticulture therapy. Gender, ethnicity, and age have an impact on people's stressors and stress levels.

People with educational stressors obtained better stress reduction benefits in horticultural therapy activities. These activities transferred students' focus from daily stressful situations to plants, allowing them to experience happy feelings (Oh et al., 2020).

Males obtained better stress-relieving effects than females. Our results were consistent with other empirical studies that the self-esteem levels and emotional state of males increased more significantly than females after green exercise (Barton and Pretty, 2010). Females consider stressors as more threatening (Ptacek et al., 1992) and adopt more emotion-focused responses compared to males (Matud, 2004), making it more difficult to benefit from the stress-relieving effects of horticultural therapy.

People of different ethnic groups also differed in their level of stress (Wei et al., 2011; Hamamura and Laird, 2014) as well as their stress management strategies (Lam and Zane, 2004; Sawaumi et al., 2015). This could explain the fact that better stress reduction on physiological indicators was achieved by Asian participants, while better stress reduction on psychological indicators was achieved by North American participants.

People over 60 years old obtained better stress-reduction benefits from horticultural therapy. Long-term stressors can be harmful to people's health, especially elderly people (Schneiderman et al., 2005; Hurst et al., 2013). A review found that horticulture therapy could improve the physical and psychological health of older persons, which is consistent with our findings (Lin et al., 2022). Gardening appears to activate many important protective mechanisms for active and healthy aging. Therefore, the elderly, particularly in nursing homes and retirement communities, could be provided with more opportunities for horticulture therapy.

The settings for horticultural therapy were essential, and they also had an important influence on the therapeutic benefits (Huxmann, 2016). Our results suggested that indoor and virtual environments were more effective in stress reduction than outdoor settings, which might be somewhat inconsistent with previous studies. This is possible because indoor and virtual environments had a relatively homogeneous and quiet atmosphere which were not likely to be affected by other distracting factors (e.g., other people, other animals, weather, temperature, sun exposure, noise, etc.,) (Guo et al., 2020). In other studies, for example, Brooks and colleagues argue that actual and virtual nature interactions were both beneficial to moods, though actual nature interactions yielded better outcomes (Brooks et al., 2017). Therefore, we encourage people to connect with “First Nature” and “Second Nature” as much as possible. From a practical standpoint, we recommend environments with both indoor and outdoor attributes, especially considering people with limited mobility and weather conditions that prevent outdoor activities.

We found that conducting horticultural therapy activities in gardens had greater effects on psychological indicators. Many studies found that gardens were more suitable environments for stress reduction (Kohlleppel et al., 2002; Coventry and White, 2018; Ulrich et al., 2020) than parks and green views in terms of psychological health (Marques et al., 2021). Meanwhile, the high biodiversity of gardens had a huge benefit in increasing the stress-relieving impact (Keniger et al., 2013; Oh et al., 2020).

The lack of direct comparisons between the various activities made it hard to verify whether one activity contributed to the reported effect (Murroni et al., 2021). This question has been answered in our subgroup analysis. Activities that activate the five senses, such as walking, meditation, transferring plants, and tasting and smelling were more effective. At the same time, it is important to consider the different intensities of activities for different groups of people when choosing the types of activity (Park et al., 2014; Lee et al., 2021), with a focus on low and medium-intensity activities.

It is also a key issue to determine the duration and frequency of horticultural therapy programs (Tu and Chiu, 2020). The 30–60 min session was more effective in physiology indicators and the < 30 min session was more effective in psychological indicators, which could achieve the stress-reduction goals and at the same time not make participants feel bored during the session. A total duration of 100–500 min could be more beneficial by maintaining the appeal and uniqueness while attracting people's attention and willingness to engage in the cyclical process of treatment.

Our findings supported the positive effect of horticultural therapy on stress reduction. Educational stressors achieved better results with horticultural therapy interventions. Seniors over 60 and males had a better stress reduction experience in horticultural therapy. Indoor and virtual areas were the most effective setting for horticultural therapy and we believed that a combination of outdoor and indoor areas was the optimal setting for horticultural therapy. At the same time, a total duration of 100–500 min provided better effects of stress reduction.

We developed a theoretical framework for horticulture therapy in terms of its stress-reduction effects on physiological and psychological indicators based on “Participants-Settings-Interventions” to provide a reference for future horticultural therapy activities (Table 2).

We also identified several limitations in this literature review. First, studies that were not published in English or Chinese were not included in this review and generalizability may be limited. Second, the lack of randomized controlled trials of high quality, though difficult to perform, also limited our outcomes. Only ten out of 31 reported studies were randomized controlled trials, let alone the participant withdrawal in several RCT studies. Finally, the number of articles in the study, the sample size of these articles, and the heterogeneity between studies would have affected the results of the subgroup analysis. Moreover, due to the lack of specific data in some of the included studies, we were unable to conduct a subgroup analysis of these studies.