Unhealthy diet, physical inactivity, and social disconnection are important modifiable risk factors for non-communicable and other chronic diseases, which might be alleviated through nature-based community interventions. We tested whether a community gardening intervention could reduce these common health risks in an adult population that is diverse in terms of age, ethnicity, and socioeconomic status.

In this observer-blind, randomised, controlled trial, we recruited individuals who were on Denver Urban Garden waiting lists for community gardens in Denver and Aurora (CO, USA), aged 18 years or older, and had not gardened in the past 2 years. Participants were randomly assigned (1:1), using a randomised block design in block sizes of two, four, or six, to receive a community garden plot (intervention group) or remain on a waiting list and not garden (control group). Researchers were masked to group allocation. Primary outcomes were diet, physical activity, and anthropometry; secondary outcomes were perceived stress and anxiety. During spring (April to early June, before randomisation; timepoint 1 [T1]), autumn (late August to October; timepoint 2 [T2]), and winter (January to March, after the intervention; timepoint 3 [T3]), participants completed three diet recalls, 7-day accelerometry, surveys, and anthropometry. Analyses were done using the intention-to-treat principle (ie, including all participants randomly assigned to groups, and assessed as randomised). We used mixed models to test time-by-intervention hypotheses at an α level of 0·04, with T2 and T3 intervention effects at an α level of 0·005 (99·5% CI). Due to potential effects of the COVID-19 pandemic on outcomes, we excluded all participant data collected after Feb 1, 2020. This study is registered with ClinicalTrials.gov, NCT03089177, and data collection is now complete.

Between Jan 1, 2017, and June 15, 2019, 493 adults were screened and 291 completed baseline measures and were randomly assigned to the intervention (n=145) or control (n=146) groups. Mean age was 41·5 years (SD 13·5), 238 (82%) of 291 participants were female, 52 (18%) were male, 99 (34%) identified as Hispanic, and 191 (66%) identified as non-Hispanic. 237 (81%) completed measurements before the beginning of the COVID-19 pandemic. One (<1%) participant in the intervention group had an adverse allergic event in the garden. Significant time-by-intervention effects were observed for fibre intake (p=0·034), with mean between-group difference (intervention minus control) at T2 of 1·41 g per day (99·5% CI –2·09 to 4·92), and for moderate-to-vigorous physical activity (p=0·012), with mean between-group difference of 5·80 min per day (99·5% CI –4·44 to 16·05). We found no significant time-by-intervention interactions for combined fruit and vegetable intake, Healthy Eating Index (measured using Healthy Eating Index-2010), sedentary time, BMI, and waist circumference (all p>0·04). Difference score models showed greater reductions between T1 and T2 in perceived stress and anxiety among participants in the intervention group than among those in the control group.

Community gardening can provide a nature-based solution, accessible to a diverse population including new gardeners, to improve wellbeing and important behavioural risk factors for non-communicable and chronic diseases.

American Cancer Society, University of Colorado Cancer Centre, University of Colorado Boulder, National Institutes of Health, US Department of Agriculture National Institute of Food and Agriculture, Michigan AgBioResearch Hatch projects.

On March 1, 2022, we searched MEDLINE, PubMed, PsychINFO, and Advanced Google Search, with no date or language restrictions, using the terms “garden”, “gardening”, AND “randomised”. Our search identified mostly systematic reviews and non-randomised studies on the associations between garden participation and a range of health behaviours and outcomes including diet and activity, BMI, and health status, with most studies showing positive evidence for garden interventions as a health promotion strategy. Three studies were randomised controlled trials of gardening among adults in home settings, but none involved community gardens.

To our knowledge, the CAPS trial is the first randomised controlled trial to study the effect of community gardening on diet, physical activity, and psychosocial outcomes in urban adults who are diverse in terms of age, ethnicity, and socioeconomic status. We found that a multicomponent, community-based intervention increased fibre intake and moderate-to-vigorous physical activity and reduced perceived stress and anxiety, but did not significantly change anthropometric outcomes, healthy eating index, combined fruit and vegetable intake, or sedentary time.

Community gardens provide a theory-informed, non-medical, community-based, and nature-based opportunity to improve chronic disease risk factors including diet, physical activity, and stress and anxiety adults who are diverse in terms of age, ethnicity, and socioeconomic status. Evidence indicates that garden interventions are acceptable to diverse populations across different age groups and are feasible. Community garden interventions can be considered a part of community-based strategies, including nature-based social prescribing, to reduce risk factors for cancer and chronic disease, and more broadly, address health and wellbeing.

The Community Activation for Prevention Study (CAPS) was designed to study the effect of community gardens on health behaviours and psychosocial outcomes among adults who are diverse in terms of age, ethnicity, and socioeconomic status. Primary outcomes included diet, physical activity, and anthropometry, with secondary outcomes of perceived stress and anxiety.

The CAPS randomised controlled trial was run at 37 community gardens in Denver and Aurora in Colorado, USA, administered by Denver Urban Gardens (DUG). Garden waiting lists were used as the basis for recruitment of study participants. These waiting lists were pre-existing at each of the gardens included in our study. Before recruitment began, study staff canvassed the surrounding neighbourhoods and partnered with neighbourhood-based organisations to raise awareness about the community gardens and encourage residents to become involved by joining their local community garden waiting list. After 6 weeks of garden promotion activities, study staff, in coordination with DUG staff, invited individuals on the garden waiting lists to join the study. Participants were eligible if they were aged 18 years or older, able to give consent in English or Spanish, had not gardened in the past 2 years, and were willing to not garden during the study period. Only one person per household could participate. Once individuals agreed to participate, study staff obtained written informed consent and conducted the baseline assessments, and participants were randomly assigned within garden waiting lists.

The study protocol was approved by the University of Colorado Boulder Institutional Review Board (UCB IRB; protocol number 16-0644) and monitored by the University of Colorado Cancer Center Data Safety and Monitoring Committee. All study participants provided written informed consent before enrolment and randomisation. All seven primary outcomes were approved on Nov 1, 2016, by the UCB IRB Office of Human Subjects.

We used a pseudo-random-number generator (sample function, R studio) with a random seed to choose participant assignments. Participants were randomly assigned (1:1) independently within each community garden waiting list to either the community gardening group (intervention group) or to stay on the waiting list with no gardening (control group). The approach involved permuted block randomisation with varying block sizes. Given the size of each garden waiting list, the algorithm randomly selected blocks of size two, four, or six, so that the total number of participants in all blocks was equal to the number of people on the waiting list for the garden. Assignments were generated by study statisticians (DHG and KKH) who had no contact with participants. Randomised assignments were transmitted to the study coordinator in sealed envelopes. The study coordinator informed participants of their group allocation after baseline data collection. Participants randomly assigned to the control group were not eligible to be re-randomised in a subsequent year. Study participants were not masked to assignment, but study staff conducting assessments, investigators, and statisticians were masked to participant assignments until after data collection was completed and the planned analyses for primary outcomes were finished.

In May of each wave, after the typical last frost in Denver and Aurora, participants randomly assigned to the intervention gardening group were provided a standard community garden plot (average size of 10 m²), seeds and seedlings, and an introductory gardening course taught through DUG. Plot fees were covered by the trial. We worked with community garden leaders to secure two to six plots per garden for study participants. A garden might have been included in the study for one wave, two waves, or all three waves. If a garden was not included in subsequent waves, new gardens were recruited to the study. The community garden organisation staff, leaders, and members offered opportunities for social interaction, community building, and mentorship through events, workdays, and classes. After each wave of the trial, when a participant's trial participation was completed (ie, 1 year of participation), as recompense for agreeing to wait for a year to garden, control participants were offered a garden plot the next growing season, with plot fees, seeds and seedlings, and an introductory gardening course paid for by the trial.

Health surveys, including perceived measures of stress and anxiety, accelerometry, and dietary interviews, were administered to all participants at baseline before the gardening season and before random allocation (April to early June, timepoint 1 [T1]), during autumn harvest (ie, end of August to October; timepoint 2 [T2]), and during the winter (ie, January to March; after the intervention (timepoint 3 [T3]). A retention incentive (US$25 at T1, $50 at T2, and $75 at T3) was offered to all participants who attended each study assessment.

We had seven prespecified primary outcomes: fibre intake, combined fruit and vegetable intake, Healthy Eating Index, moderate-to-vigorous physical activity, sedentary time, BMI, and waist circumference.

For anthropometrics, participant height was measured with a portable stadiometer (Seca 213 Portable Stadiometer; Seca, Hangzhou, China), accurate to the nearest 0·1 cm. Bodyweight was measured with a digital platform scale (Seca 876 Digital Scale; Seca) to the nearest 0·23 kg (0·5 lb). Waist circumference was measured at the superior border of the iliac crest, using a cloth tape to the nearest 0·1 cm, while the participant was standing. BMI was calculated as kg/m².

We did all analyses under the intention-to-treat principle (ie, included all participants randomly assigned to groups), and analysed them as randomised. Because the COVID-19 pandemic affected data collection and might have had an effect on health behaviours, the study team agreed to exclude wave 3, T3 data from the analysis—ie, data collected after Feb 1, 2020. The decision was made a priori, while assessors and investigators were still masked to group assignments. We tested time-by-intervention hypotheses at an α level of 0·04 and intervention effects at T2 and T3 at an α level of 0·005 (99·5% CI), for a total type 1 error rate of (0·04 + 0·005 + 0·005 =) 0·05 per primary outcome. A significant time-by-intervention interaction indicated that the pattern of means over time for the intervention group differed from that of the control group.

Participants with and without missing data were compared overall and stratified by intervention assignment using Rao-Scott χ² tests (). We assessed the balance of demographic features across the two groups using the Rao-Scott test for categorical demographic variables and mixed models accounting for clustering within garden waiting lists for continuous demographic variables.

For dietary outcomes, we used general linear mixed models to assess time-by-intervention interaction. The outcome at each timepoint (T1, T2, and T3) was the mean of the response for the 24-h recalls. We modelled correlations using the Kronecker-product covariance matrix, with an unstructured model for the longitudinal repeated measures, and the best-fitting choice of one of two parameterisations to account for clustering: (1) a model with a waiting list random effect only, and (2) a model with a random term for assignment to gardening, nested within the waiting list. For all mixed models, we used jackknife studentised residuals to test modelling assumptions. We assessed significance using the Wald test and Kenward-Roger degrees of freedom. In preliminary modelling, we assessed wave effects and found no difference in outcome by wave. This indicated that data could be combined across the 3 years of recruitment, as we present in the Results. We did a sensitivity analysis to assess whether there was an interaction between social desirability and randomisation assignment, and then, if there were no interactions, whether the social desirability score was associated with dietary outcomes.

For physical activity outcomes, we used a similar approach as for the dietary outcomes, with the exception that the initial set of predictors also included an indicator variable for day type (ie, weekend vs weekday) and interactions between day type, measurement time, and randomisation assignment. After examining the use of wave as a predictor, we did a series of planned hypothesis tests to examine interactions with day type, and then tested the time-by-intervention interaction in the best fitting model.

We did modelling and hypothesis testing similarly for BMI and waist circumference as for the dietary and physical activity outcomes, and additionally adjusted for age and sex.

We had several preplanned secondary outcome analyses. We assessed secondary outcomes at an α level of 0·05. We calculated difference scores, T2 minus T1 and T3 minus T1, for the secondary outcomes of perceived stress and anxiety. We fit separate general linear mixed models for the T2 minus T1 and T3 minus T1 difference scores to test if the difference score differed by group assignment. Our models controlled for the baseline value and the interaction between baseline value and randomisation assignment. Under the assumption that participants were exchangeable within waiting lists, a random intercept for each garden waiting list produced a compound symmetric variance structure. We used a two degrees-of-freedom test of equality of both intercepts and slopes to test if there was a difference between intervention and control at the α level of 0·05.

We did not have an a priori hypothesis about differences at T3, but we present results for T3 for completeness.

The funders of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report.

Between Jan 1, 2017, and June 15, 2019, 493 individuals were screened, of whom 291 (59%) completed baseline measurements and were randomly assigned to either a garden plot (intervention group; n=145) or the waiting list (control group; n=146; ). 237 (81%) of the 291 participants attended a health visit with study staff and completed a health survey at all three timepoints for wave 1 (n=50; n=24 in the intervention group and n=26 in the control group) and wave 2 (n=97; n=47 in the intervention group and n=50 in the control group), and the first two timepoints in wave 3 (n=90; n=44 in the intervention group and n=46 in the control group). The number of participants who contributed data for each outcome at each timepoint (ie, T1 and T2 or T3, or both) in the intervention and control groups is shown in the figure. Differing numbers by timepoint, outcome measure, and randomisation assignment reflect loss to follow-up or participant refusal to complete a specific outcome measure at a specific timepoint. There was no differential missingness by group assignment, sex, age, or income (all p>0·05; ). Median time from enrolment to T2 was 154 days (IQR 139–180) and T3 was 312 days (290–348).

36 (25%) of 145 participants in the intervention group and 27 (18%) of 146 in the control group were lost to follow-up. During the study, one (1%) participant in the intervention group had an intervention-attributable adverse event: an allergic reaction in the garden. 13 (9%) of 145 participants assigned to the intervention decided not to garden after randomisation and before gardening began and were lost to follow-up after contributing some data at T1. Four (3%) of 146 participants in the control group refused to remain on the waiting list and began to garden; two (50%) remained in the study and contributed data during follow-up, and two (50%) left the study after T1 and did not contribute subsequent data. All participant data were analysed as randomised.

The time-by-intervention interaction was significant for fibre intake (p=0·034; ), with differences being non-significantly higher in the intervention group than in the control group at T2 (mean intake: intervention group 21·48 g per day [SE 0·91]; control group 20·07 g per day [0·87]; between-group difference 1·41 g per day [SE 1·24; 99·5% CI –2·09 to 4·92; p=0·25]). There were no significant time-by-intervention effects for the outcomes of combined fruit and vegetable intake and Healthy Eating Index (all p>0·04; ). For combined fruit and vegetable intake, the intervention group had a higher mean intake than the control group at T2 but the difference was not significant (mean intake: intervention group 4·96 servings per day [SE 0·26]; control group 4·49 servings per day [SE 0·24]; between-group difference 0·46 servings per day [SE 0·37; 99·5% CI –0·58 to 1·51; p=0·21]). In a sensitivity analysis, we found no time-by-intervention-by-social-desirability effect for any diet outcome (all p>0·05; data not shown).

We found a significant time-by-intervention effect for moderate-to-vigorous physical activity (p=0·012; ), but it was not higher in the intervention group than in the control group at T2 (mean activity: intervention group 54·92 min per day [SE 2·62]; control group 49·12 min per day [SE 2·58]; between-group difference 5·80 min per day [SE 3·62; 99·5% CI –4·44 to 16·05; p=0·11]). Moderate-to-vigorous physical activity did not differ between weekend and weekday (data not shown). We found no time-by-intervention effect on sedentary time (p=0·47; ). We found no significant time-by-intervention interaction effect on BMI (p=0·99) or waist circumference (p=0·31) with the gardening intervention versus the control group ()

The hypothesis is of no intervention effect on difference score, controlling for baseline values. Mean (SE) of the respective stress and anxiety scores at baseline are preported by percentile. GAD-7=General Anxiety Disorder 7. PSS-10=Perceived Stress Scale 10.

Our results suggest that community gardening, as an example of a multicomponent intervention, could be beneficial in changing some key risk factors for cancer and other chronic diseases, thus warranting further investigation. Data from our qualitative interviews and our process assessment will be examined to contextualise the quantitative results and understand how the trial affected study participants.

We included a broad range of outcomes in our study, which was necessary to measure the pleiotropic effects of community gardening. Multiple outcomes require multiple statistical tests, and multiple statistical tests increase the chance of an inflated type 1 error rate, and reduced replicability. In scientific research, researchers usually correct for multiple comparisons within each manuscript separately, and not over many papers based on data from a single trial. The rationale is that each report represents a separate experiment. A trial with multiple outcomes raises similar questions. The choice of whether or not to use a Bonferroni correction so that the entire trial has a total type 1 error rate of 0·05, or to use a total type 1 error rate of 0·05 per outcome is unclear. We chose the second approach, which controlled the type 1 error rate within each outcome, but did not correct for the multiple outcomes, because each outcome was a separate hypothesis.

Our study had several limitations. One limitation was the exclusion of data from wave 3 at T3 to account for the COVID-19 pandemic. The data were excluded because pandemic-related closures probably would have affected the primary and secondary outcomes and were excluded before unblinding and before any analysis had occurred. The effect of excluding the last timepoint of data resulted in a decreased sample size and thus might have diminished statistical power or attenuated results towards the null. By design, the trial did not have the power to assess intervention effects within subgroups stratified by race, ethnicity, sex, sexual orientation, and socioeconomic status. Diet, stress, and anxiety outcomes were self-reported, which might lead to information bias, although there was no evidence of social desirability bias for dietary reporting in a sensitivity analysis. Finally, a small number of participants refused their assignment or discontinued participation, and there might be other participants who did not report their status to the study staff. Crossover of participants between groups would only bias the results towards the null. Finally, this trial captured the community gardening experience only over 1 year. Whether the effects of intervention will be maintained beyond 1 year is unknown. Because of the time required to establish new activities such as gardening and maintain changes in health behaviours and health status, long-term follow-up would be useful to understand if and how gardeners maintain the changes they adopted in the first year of gardening.

This randomised controlled trial strengthens evidence for community gardening as a comprehensive multicomponent nature-based social intervention that can improve some health behaviours and reduce perceived stress and anxiety in a diverse urban population. Both are important for the prevention of chronic diseases and mental health disorders. Gardening is a nature-based solution that fits within the broader context of urban agriculture systems. A community garden is a setting that could be within reach for citizens across the world and can be tailored to meet the needs of people across different social and economic groups, cultures, geographies, and local environments. Land planners, health officials, and policy makers together can integrate gardens into the fabric of communities, recognise gardens as a primary and permanent natural space, similar to playgrounds, farmers’ markets, bicycle lanes, and public plazas, and invest in programming that supports gardeners across the lifespan.

JSL and KA conceived of the study. JSL, KA, and DHG generated the hypotheses. JSL, KA, DHG, KL, RFH, JAL, JRH, and TGH designed the study, interpreted the data, and reviewed the manuscript. JSL, KA, DHG, KKH, KL, and RFH wrote the first draft of the manuscript. JBC and KL cleaned and processed the physical activity data. DHG, KKH, and MP conducted the data cleaning, data analysis, and statistical analysis, with guidance from JSL, KA, RFH, KL, JRH, and TGH, and support from EC. AV, EC, and JBC supported aspects of the recruitment, retention, data collection, database management, and writing. AV managed all aspects of the study including recruitment, retention, data collection, participant engagement, and coordination of study consortium. JRH and TG managed the collection of dietary data. All authors contributed to the preparation of the manuscript through synthesis of the data, team discussions, and detailed reviews. All authors reviewed the final version of the manuscript, verified the underlying data reported in the manuscript, and accepted responsibility for submitting the article for publication. All authors had full access to all the data in the manuscript and approved the decision to submit for publication.

The de-identified participant data and data dictionary generated by the CAPS trial and the statistical code can be shared by request of investigators and after approval by the trial investigators and the Institutional Review Board of the University of Colorado Boulder. Data inquiries can be directed to Jill Litt, at [email protected]

JRH owns controlling interest in Connecting Health Innovations (CHI), a company that has licensed the right to his invention of the dietary inflammatory index (DII) and for which JRH and the University of South Carolina have secured a federally registered trademark for the DII. The CHI aims to develop computer and smartphone applications for patient counselling and dietary intervention in clinical settings. CHI owns all derivative products, including the energy-adjusted DII (E-DII). All CHI-related activity occurred outside the submitted work, which includes royalties paid to CHI. JRH confirms that the subject matter of this manuscript did not have any direct bearing on that work, nor has that activity exerted any influence on this project. All other authors declare no competing interests.

This study was funded by the Research Scholars Health Equity Grant (130091-RSG-16-169-01-CPPB) from the American Cancer Society. Additional support was provided by the University of Colorado Cancer Centre and the University of Colorado Boulder (Boulder, CO, USA). DHG received additional support from the National Institutes of Health (grant numbers R01GM121081, R25GM111901 and R25GM111901-04S1). KA received additional support from the US Department of Agriculture National Institute of Food and Agriculture Michigan AgBioResearch Hatch projects MICL02410 and MICL02711. We thank the participants whose commitment and dedication made this trial possible, the garden leaders who supported and believed in this study, our partner DUG, and, specifically, Michael Buchenau, former DUG Executive Director, who, for the past 15 years, has graciously collaborated with the University of Colorado to examine the health and social benefits of community gardening. We are grateful for the contributions of our CAPS advisory committee: Betsy Johnson, Melanie Morrison, Sarah Muntz, Lara Fahnestock, Michael Buchenau, Linda Appel Lipsius, and Laura Gerlick. Additionally, we are indebted to the dedication and perseverance of our study personnel, who assisted in all facets of participant recruitment, retention, garden support, and data collection: Hannah Buchenau, Erin Decker, Lara Fahnestock, Pallas Quist, Abby Bohannan, Alyssa Beavers, Kristin Lacy, Ashby Sachs, Heidi Kessler, and Catherine Erickson. We also thank Tessa Crume and Carol-Ann Mullin for their support in the design of the data management system.