The UK Biobank cohort is a sample of the British population containing health and lifestyle information for around 500,000 people between 40 and 69 years old at recruitment between 2006 and 2010. Possible participants were identified using the NHS patient registers living near an assessment center (~25 miles or ~40 Km) across England, Wales, and Scotland. The invitations to participate were sent via post. In total, 9.2 million invitations were sent, and 503,000 people agreed to participate (37). At the assessment centers, participants completed a touchscreen questionnaire answering questions related to socio-demographics, lifestyle, environmental, and health-related factors (37, 38). Participants were also verbally interviewed and gave urine, blood, and saliva samples. UK Biobank's protocol, and procedures are reported in detail elsewhere (39, 40). For this research, we only included participants who resided in England at the time they were recruited and had data collected via the touchscreen questionnaire, direct measurements at the assessment center, and had greenspace data available from UK Biobank project 4362, which assigned high-resolution Ordnance Survey MasterMap greenspace data in circular distance buffers around participant addresses (N = 204,478) (For the project's permits see the section “ethics approval and consent to participate” below).

The explanatory variables and health and wellbeing markers were selected a priori due to their previously assessed relationship with urban agriculture projects, and their relevance to public health (41–44).

The variable fruit and vegetable intake was created by adding the portions of fresh fruits, raw vegetables, and cooked vegetables obtained from three questions from the touchscreen questionnaire: (1) “On average, how many heaped tablespoons of COOKED vegetables would you eat per DAY? (Do not include potatoes; put ‘0' if you do not eat any),” (2) “On average, how many heaped tablespoons of SALAD or RAW vegetables would you eat per DAY? (Include lettuce and tomato in sandwiches; put ‘0' if you do not eat any),” and (3) “About how many pieces of FRESH fruit would you eat per DAY? (Count one apple, one banana, 10 grapes etc. as one piece; put ‘0' if you do not eat any).” The three questions have the same instruction for the participants (Please provide an average considering your intake over the last year. If you are unsure, please provide an estimate or select Do not know), and the same answer options (“Enter a number/Less than one/Do not know/Prefer not to answer”). If participants answered, “Less than one,” we replaced it for a value of 0.5 (45), and we excluded participants who answered “Do not know,” or “Prefer not to answer.” Based on the NHS values for fruit and vegetable portions, three heaped tablespoons of cooked vegetable and four heaped tablespoons equal one portion, consequently, before creating the variable fruit and vegetable intake, we divided cooked vegetable intake by three, and raw vegetable intake by four. Then we used the WHO recommendation of eating at least five portions of fruits and vegetables to categorize the variable; we labeled participants as “Low fruit and vegetable intake” for participants who ate less than five portions a day and “High fruit and vegetable intake” for those who ate five or more portions a day.

To assess physical activity, questions based on the short form of the International Physical Activity Questionnaire (IPAQ) were used (46), the protocol and procedures are reported elsewhere (47). Physical activity is reported in weekly Metabolic Equivalent Tasks (METs) (min/week), which relates to the energy expenditure doing moderate activities, vigorous activities, and walking, compared to sitting (46, 47). We used the UK Biobank variable “IPAQ activity group,” which categorizes people into three groups, “Low,” “Moderate,” and “High,” by using the classification processing rules published by the IPAQ (46).

Based on our hypothesis that people who practice urban agriculture are in constant proximity and interaction with nature, we wanted to assess the association of greenspace exposure and health and wellbeing markers in using a close distance. Hence, we assessed exposure to greenspace within 100 m from a participant's residential address using data from Ordnance Survey MasterMap Greenspace (UK Biobank project 4362). Methods to calculate greenspace percentage are reported elsewhere (48). The percentage of surrounding greenness or greenspace exposure was categorized using quartiles (Q1, Q2, Q3, and Q4).

Covariates were selected based on assumptions that they would confound the relationship between the exposure and outcome, and/or were important risk factors identified by other UK Biobank researchers for the health or wellbeing markers under investigation (49–56). We included sex (Female, Male), age at baseline assessment, ethnicity as a binary variable (two categories): (1) White, (2) Not white (original Biobank categories: “Mixed,” “Asian or Asian British,” “Black or Black British,” “Chinese,” and “Other ethnic group”), highest educational qualification (four categories): (1) College or University Degree; (2) Other professional qualification or education to age 18 (original Biobank categories: “A levels/AS levels or equivalent,” “NVQ or HND or HNC or equivalent,” and “Higher National Certificate (HNC) or equivalent, other professional qualifications”); (3) High School Qualifications (original Biobank categories: “CSEs or equivalent” and “O levels/GCSEs or equivalent”); (4) None of the above.

We defined neighborhood deprivation using the deciles of the index of multiple deprivation 2010 for England (57). We created the categories Low, Medium, and High Deprivation by collapsing the national deciles 1–3, 4–7, and 8–10, respectively. The index of multiple deprivation comprises assessments of seven different categories experienced by people: Income Deprivation, Employment Deprivation, Health Deprivation and Disability, Education Skills and Training Deprivation, Barriers to Housing and Services, Living Environment Deprivation, and Crime (58). We used the category “Smoking status” from the UK biobank that splits participants into current smokers, previous smokers, and participants who had never smoked.

We defined health markers BMI (three categories) by collapsing the original UK Biobank variable: (1) Healthy (≤ 24.9), (2) overweight (25–29.9), and obese (≥30) (59), blood pressure (binary): (1) Healthy (≤120 systolic; ≻ 80 diastolic) and (2) High (≥120 systolic; ≥80 diastolic) (60, 61), self-health assessment, and loneliness were considered as outcome variables to assess health and wellbeing. The questions “Do you often feel lonely?” (No = 0, Yes = 1) and “How often are you able to confide in someone close to you?” (0 = almost daily to once every few months; 1 = never, or almost never), participants were defined as lonely if they answered positively to both questions (score 2) (62, 63). Participants' self-health assessment was assessed using the question “In general how would you rate your overall health?” (Answer options: Poor/Fair/Good/Excellent/Do not know/Prefer not to answer). Participants who answered “Do not know” or “Prefer not to answer” were excluded as these responses are not clearly interpretable outcomes.

All the statistical analyses were carried out using the R software (64). Out of the initial 502,664 UK Biobank participants, we excluded participants with missing data in any of the explanatory variables and those who did not meet the inclusion criteria. To avoid losing data we classified NAs of confounders as “Missing” and used this as a category. To explore if healthy behaviors and exposures related to urban agriculture (fruit and vegetable intake, physical activity, and greenspace exposure) have a synergistic effect on health traits, we conducted Bootstrap General Linear Models (GLMs with bootstrap) with and without interactions (two-way and three-way) between the explanatory variables. We used logistic regression for blood pressure (reference = healthy) and loneliness (reference = no), and we used ordinal logistic regression to assess overall health rating (reference = poor) and BMI (reference = normal BMI), the variable overall health rating is the only variable that uses the unhealthy category (poor) as reference (62). All models were adjusted for age, sex, index of multiple deprivation, ethnicity, highest qualification, and smoking status. For each health and wellbeing marker we first fitted one single exposure model with one of the explanatory variables plus the covariates; secondly, we fitted a multi-exposure confounding model with the three explanatory variables—fruit and vegetable intake, physical activity, and greenspace exposure, plus the covariates without including interactions; finally, we fitted a multi-exposure model with two and three-levels interactions between the explanatory variables plus the covariates. To account for multiple testing, p values were corrected using the Benjamini & Hochberg method, setting the significance level at α= 0.05 and using two-tailed tests (65). To calculate the fit of the model and the 95% confidence intervals of the models' coefficients we bootstrapped all the models (1,000 generations) (66, 67). The tables in the main text show the outcomes of the exposure model with two and three-levels interactions; outcomes for single exposure and the multi-exposure confounding models are presented in Supplementary material 1. We used likelihood ratio test to compare the goodness of fit between the models with interactions and the models without interactions and statistical significance was defined by a two-side alpha level of p = 0.05. Models with interactions included have a better goodness of fit than those considering the health behaviors independently.

Single exp models show, as would be expected that both fruit and Vegetable intake and physical activity are positively associated with overall health rating; physical activity with decreased BMI; both physical activity and fruit and vegetable intake with decreased blood pressure, and physical activity and surrounding greenness with reduced loneliness. Contrary to expectations, however, fruit and Vegetable intake and surrounding greenness both were associated with increased BMI; the quartiles Q2 and Q3 of surrounding greenness with poor health, and with increased blood pressure.

After adjusting for age, sex, ethnicity, highest educational qualifications, smoking status, and Index of Multiple Deprivation with single exposure models show, as would be expected that both fruit and Vegetable intake and physical activity are positively associated with overall health rating; physical activity with decreased BMI; both physical activity and fruit and vegetable intake with decreased blood pressure, but the latter only for high levels of physical activity (OR = 0.894, 95% CI = 0.865, 0.924, p adjusted ≤ 0.001; Supplementary Table 5). Physical activity and surrounding greenness were associated with reduced loneliness, except for the 2nd quartile of greenspace exposure that was not statistically significant (OR = 0.979, 95% CI = 0.924, 1.038, p adjusted = 0.477; Supplementary Table 6). Contrary to expectations, however, fruit and vegetable intake and surrounding greenness were associated with increased BMI; the quartiles Q2 and Q3 of surrounding greenness with poor health, and with increased blood pressure.

We found largely similar associations between the explanatory variables and the health outcomes in the multiple-exposure models with no interactions, but all the association were slightly attenuated when compared to the single exposure models (Supplementary Table 5).

Multiple-exposure models with two- and three-way interactions show generally similar associations between the explanatory variables and the health outcomes, but with further attenuation. More closely aligned with expectations, Q2 (Q2: OR = 1.049, 95% CI = 0.984, 1.118, p adjusted = 0.249) and Q4 (Q4: OR = 1.049, 95% CI = 0.907, 1.030, p adjusted = 0.372) were no longer significantly associated with increased BMI, but Q3 in comparison with Q1 remained associated with worsened BMI (Table 1). Also more aligned with expectations, the quartiles Q2 and Q3 were no longer associated with poor health, and in reverse, as would be expected, Q4 became significantly associated with good health (OR = 0.896, 95% CI = 0.825, 0.974, p adjusted ≤ 0.032). More aligned with expectation the greenspace exposure quartiles Q2 and Q4 lost their positive associations with blood pressure, but Q3 remained significant (OR = 1.199, 95% CI = 1.097, 1.311, p adjusted ≤ 0.001; Table 1). Physical activity and greenspace exposure remained significant, showing a beneficial association with loneliness (Table 1).

Numerous epidemiological studies have shown associations between unhealthy lifestyle behaviors and life-threatening chronic diseases (68–70). Our current analysis shows that, after controlling for confounders, physical activity and fruit and vegetable intake show positive associations on most health and wellbeing markers. This is in-line with a large body of literature that demonstrates that being physically active reduces the chances of suffering from chronic diseases including various forms of cancer, high blood pressure problems, cholesterol, cardiorespiratory conditions, mental illnesses, bone and muscular problems, obesity, and sleep complications (71–74). Insufficient intake of fruit and vegetables has been shown to cause 31% of ischemic heart diseases, approximately 19% of gastrointestinal cancers, and 11% of stroke (75–77) whilst increased intake of vegetables on the other hand significantly reduces the risk of cardiovascular incidents (75). Contrary to expectations, however, meeting fruit and vegetable intake recommendations was shown to increase BMI in our analyses. A possible reason for this result is that for our sample, a high intake of fruit and vegetable might reflect a high intake of food in general, and research shows the size of portions is directly linked with high BMI (78).

In our single exposure models, although greenspace is shown to be associated with reduced loneliness, it was negative associated on health as measured by BMI, blood pressure, and overall health rating. Whilst the statistical significance of these associations is mostly removed in the multi-exposure and interaction models, this is still contrary to evidence from the literature. Greenspace exposure has been shown to be positively associated with mental health benefits (79–81), reduction in cardiovascular morbidity (82), type II diabetes (83, 84), and mortality (85). The reason for this observation in this study could be due to the 100 m distance buffer we used to measure greenspace exposure, whilst as most research identifies benefits for buffers of 300 m or more. For example, similar to our results, Zhang et al. (86) found a negative association between the percentage of green space and depression symptoms in adolescents at a 100 m buffer in adjusted models and no association once the models were corrected for confounders. Hartley et al. (87) found no significant association between the percentage of green space and several mental issues in kids and adolescents at a 200 m buffer. Still, they found a negative association between greenspace and anxiety for buffer distances at 400 and 800 m. Moreover, it has been shown that the benefits of natural spaces to people are not one-dimensional and depend on several factors, such as proximity to the green area, type of greenspace, how often the greenspace is visited, and the time spent in the greenspace (88–90). Coleman et al. (91) for instance demonstrated how air pollution may form part of the equation in the relationships between exercise, greenspace, and mortality. Therefore, further research that includes information on the type of green space and distance is needed. Another possible explanation for that outcome is the presence of a confusing variable not assessed here, such as working status and conditions, marital status and the quality of family relationships (92–95).

Returning to our hypothesis that interventions such as urban agriculture can bring more significant benefits to public health than interventions that only seek to address a single healthy lifestyle behavior: as stated in the introduction, research has shown the benefits of urban agriculture for health and wellbeing and the association between the practice of urban agriculture and healthy lifestyles (33). Our results continue to support the existing literature that multiple-lifestyle behaviors are better than single-focused approaches (111). However, more research is needed to understand the association that the different interactions between the urban environment, sociodemographic factors, and lifestyle behaviors have with health and wellbeing. Nevertheless, and more importantly, there is a need for health authorities to create strategies that motivate the population to engage in healthy behaviors (112). According to our results and the literature, urban agriculture could be a helpful tool due its multiple potential contributions to urban health (113).