Effect of gardening activities on domains of health: a systematic review and meta-analysis | BMC Public Health

Study identification

We adopted the PRISMA diagram to depict the flow of information throughout different phases of this systematic review. Figure 1 illustrates the study selection and screening process flow and maps the number of records identified, included and excluded and the reasons for exclusions. The initial search yielded 23,086 articles, and 95 records were identified, with no additional findings from reference list searches. After employing automation tools and eliminating irrelevant information from each database, 3862 articles were screened by title and abstract, and 3807 articles were excluded by title and abstract. Among the retrieved articles (n = 55), 14 duplicated articles were excluded. Twenty articles were excluded due to research design (n = 14), research status (n = 2), full-text unavailable (n = 2), and irrelevant outcome (n = 2), finally, 21 out of 41 articles were included in the systematic review and meta-analysis. The second-round search identified 173 records, after excluding studies that did not meet the inclusion criteria, two studies remained. Finally, 23 articles were included in the present systematic review and meta-analysis. The detailed reasons for exclusion are presented in Fig. 1.

Study selection process for the first round and second round selection. RCTs: randomized controlled trials

Study characteristics

A summary of the main characteristics of the included studies is provided in the Supplemental Material eTable 6. Among the included studies, thirteen studies were randomized controlled trials, and ten were quasi-experimental studies with intervention and control groups. The sample size ranged from 14 [22] to 381 [23], with a total of 4535 participants included. The participants had a mean age of 54.39 years, and the majority were females (63.25%). The majority of studies were conducted in the U.S. [22,23,24,25,26,27,28], four in Taiwan (China) [29,30,31,32], two in mainland China [33, 34], and two in Singapore [35, 36]. Single studies were selected from Denmark [37], Norway [38], France [39], the UK [40], the Netherlands [41], South Korea [42], the Dominican Republic [43] and Nicaragua [44].

Health was categorized into three domains: physical health, mental health and general health (summarized in Table 1). The most frequently reported outcomes included depression [30, 35, 36, 42], anxiety [24, 34,35,36, 42], emotional well-being [22, 26, 28, 31, 36, 40] and cognitive function [33, 35, 36, 39]. Fourteen studies were included in the meta-analysis of gardening activity and mental health [22,23,24, 29, 32,33,34,35,36, 38, 40, 42]; seven studies were included in the meta-analysis of gardening activity and general health [22, 23, 29, 31, 36, 40, 42]; and five studies were included in the meta-analysis of gardening activity and physical health [33, 35, 36, 39, 44].

Effects of gardening activity on mental-, physical and general health

The results of the meta-analysis of pooled mental health data from 14 trials investigating gardening interventions and control groups revealed high heterogeneity (100%). With 585 participants in the intervention group and 563 participants in the control group, the overall effect size was relatively low to medium (Z = 0.94, p = 0.35), without statistical significance (Fig. 2A). For general health, there were a total of 736 participants in gardening interventions, and for physical health, there were 776 participants. The heterogeneity for both general health (I² = 93%) and physical health (I² = 89%) was relatively high, with overall effect sizes of 1.35 (p = 0.18) and 1.20 (p = 0.23), respectively (Fig. 2B, C).

A, B, C Forest-plot of gardening and health domains

Given the high heterogeneity in each subcategory of health outcomes, a random effects model was applied. The meta-analysis showed that gardening activity had a moderate effect on mental health (SMD: −0.31; 95% CI: −0.97, 0.34), a medium-to-low effect on general health (SMD: −0.25; 95% CI: −0.62, 0.11), and a small effect on physical health (SMD: −0.08; 95% CI: −0.20, 0.05). Although the 95% confidence intervals for all these health domains included 0, the overall trend pointed toward a beneficial effect of gardening interventions.

Sensitivity analysis

We conducted sensitivity analyses of gardening interventions and health outcomes by excluding very low-quality trials. After removing trials rated as “high risk of bias” in at least one domain out of seven domains of the GRADE, the results indicated that gardening intervention is significantly effective for mental health (SMD: −0.82; 95% CI: −1.51, −0.13). Weak evidence was found for general health (SMD: −0.10; 95% CI: −1.16, 0.95) and physical health (SMD:0.06; 95% CI: −0.35, 0.46), with high heterogeneity (I² > 80%). The detailed results can be found in eFigure 1A, B and C.

Risk of bias and quality assessment

There was a minor degree of risk of bias observed in the included studies. The highest potential of bias was related to selection bias, as half of the trials did not adhere to the randomized allocation of participants (50%). Other potential biases included detection bias (22%) and performance bias (35%), which had relatively high percentages of “unclear risk of bias” and “high risk of bias”, respectively, compared with other types of biases (Fig. 3). There was one study obtained more than three “high risk of bias” scores, as their study was conducted in a low-income country where participants engaged in home gardening primarily to improve food security, increasing the risk of selection bias [44]. No other trials received more than three “high risk bias” bias scores for a domain (Supplemental Material eTable 5).

Risk of bias graph. Risk of bias items are presented as percentages across all included studies

The quality of evidence (GRADE) was rated for three health domains (i.e., mental health, general health, and physical health) (Table 2). The GRADE quality score was rated as moderate, low, or very low for mental health, general health or physical health, respectively, after evaluating inconsistency, indirectness, imprecision and publication bias. Inconsistency and imprecision were found for physical health, and no indirectness was detected for any of the three health outcomes. Detailed information regarding the quality assessment is shown in Table 2.