Fruit and vegetable consumption and risk of depression: accumulative evidence from an updated systematic review and meta-analysis of epidemiological studies | British Journal of Nutrition | Cambridge Core (original) (raw)

Findings from systematic review

The characteristics of twenty-seven studies included in this systematic review are presented in Tables 1 and 2. These studies were published between 2001 and 2017. Among included studies, sixteen had a cross-sectional design( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Allgöwer, Wardle and Steptoe 23 , Reference Niu, Guo and Kakizaki 25 – Reference Bishwajit, O’Leary and Ghosh 28 , Reference Shahar, Hassan and Sundar 43 , Reference Whitaker, Sharpe and Wilcox 45 – Reference Mikolajczyk, El Ansari and Maxwell 50 ), nine studies had a cohort design( Reference Akbaraly, Sabia and Shipley 10 , Reference Chi, Wang and Tsai 17 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Rienks, Dobson and Mishra 29 , Reference Woo, Ho and Yu 30 , Reference Liu, Xie and Chou 42 , Reference Kingsbury, Dupuis and Jacka 51 ) and the remaining two studies were case–control studies( Reference Michalak, Zhang and Jacobi 40 , Reference Payne, Steck and George 52 ). In all, ten publications were reported from European countries( Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Allgöwer, Wardle and Steptoe 23 , Reference Verger, Lions and Ventelou 26 , Reference Michalak, Zhang and Jacobi 40 , Reference El Ansari, Adetunji and Oskrochi 46 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Kingsbury, Dupuis and Jacka 51 ), five studies were reported from American countries( Reference McMartin, Jacka and Colman 5 , Reference Gangwisch, Hale and Garcia 22 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Castellanos, Connell and Lee 49 , Reference Payne, Steck and George 52 ) and eight studies were reported from Asia( Reference Chi, Wang and Tsai 17 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Aihara, Minai and Aoyama 27 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Woo, Ho and Yu 30 , Reference Liu, Xie and Chou 42 , Reference Shahar, Hassan and Sundar 43 ), and the remaining four studies were reported from Australia( Reference Mihrshahi, Dobson and Mishra 19 , Reference Rienks, Dobson and Mishra 29 , Reference Crichton, Bryan and Hodgson 47 , Reference Forsyth, Williams and Deane 48 ). All published studies were conducted in adult populations. Four studies were conducted on women( Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Whitaker, Sharpe and Wilcox 45 ), one on men only( Reference Castellanos, Connell and Lee 49 ) and twenty-two studies on both genders. Sample sizes ranged from seventy-one people in cross-sectional studies to 125 428 in cohort studies. In total, 289 018 participants were studied. The number of cases varied from 51 to 9739. Most included studies had used FFQ to assess dietary intakes, except for ten studies that had applied 24-h recalls( Reference Shahar, Hassan and Sundar 43 , Reference Whitaker, Sharpe and Wilcox 45 ), dietary history( Reference Niu, Guo and Kakizaki 25 , Reference Forsyth, Williams and Deane 48 ) and food questionnaire( Reference Woo, Lynn and Lau 21 , Reference Allgöwer, Wardle and Steptoe 23 , Reference Aihara, Minai and Aoyama 27 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Castellanos, Connell and Lee 49 , Reference Kingsbury, Dupuis and Jacka 51 ). A total of sixteen studies had considered fruit intake( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Hintikka, Tolmunen and Honkalampi 16 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Lynn and Lau 21 – Reference Allgöwer, Wardle and Steptoe 23 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Rienks, Dobson and Mishra 29 , Reference Liu, Xie and Chou 42 , Reference El Ansari, Adetunji and Oskrochi 46 , Reference Forsyth, Williams and Deane 48 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Payne, Steck and George 52 ), nineteen studies had assessed consumption of vegetables( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Hintikka, Tolmunen and Honkalampi 16 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Lynn and Lau 21 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 – Reference Woo, Ho and Yu 30 , Reference Michalak, Zhang and Jacobi 40 , Reference Shahar, Hassan and Sundar 43 , Reference El Ansari, Adetunji and Oskrochi 46 – Reference Forsyth, Williams and Deane 48 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Payne, Steck and George 52 ) and in seven other studies total consumption of fruit and vegetables was examined( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Verger, Lions and Ventelou 26 , Reference Aihara, Minai and Aoyama 27 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Castellanos, Connell and Lee 49 , Reference Kingsbury, Dupuis and Jacka 51 ). One study had considered plant foods including vegetables, legumes, fruits and nuts as the exposure variableReference Crichton, Bryan and Hodgson 47 ). For assessment of mental disorders, nine studies used Center for Epidemiological Studies-Depression Scale (CES-D)( Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Chi, Wang and Tsai 17 , Reference Mihrshahi, Dobson and Mishra 19 , Reference Verger, Lions and Ventelou 26 , Reference Rienks, Dobson and Mishra 29 , Reference Liu, Xie and Chou 42 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Crichton, Bryan and Hodgson 47 , Reference Castellanos, Connell and Lee 49 ), five studies used Geriatric Depression Scale( Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Aihara, Minai and Aoyama 27 , Reference Woo, Ho and Yu 30 , Reference Shahar, Hassan and Sundar 43 ), four others had applied Beck’s Depression Inventory (BDI)( Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Allgöwer, Wardle and Steptoe 23 , Reference El Ansari, Adetunji and Oskrochi 46 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Abdi, Salehi and Ghodsi 53 ) and the remaining eight studies had used other questionnaires( Reference McMartin, Jacka and Colman 5 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Michalak, Zhang and Jacobi 40 , Reference Forsyth, Williams and Deane 48 , Reference Kingsbury, Dupuis and Jacka 51 , Reference Payne, Steck and George 52 ).

Table 1 Characteristics of studies that reported the relationship between fruit and vegetable intake and depression (Odds ratios, relative risks (RR), hazard ratios (HR) and 95 % confidence intervals; mean values and _β_-coefficients with their standard errors)

Table 2 Characteristics of studies that reported the relationship between fruit and vegetable intake and anxiety (Relative risks (RR), _β_-coefficients and 95 % confidence intervals)

With regard to outcome, only four studies had considered anxiety( Reference McMartin, Jacka and Colman 5 , Reference Michalak, Zhang and Jacobi 40 , Reference Crichton, Bryan and Hodgson 47 , Reference Forsyth, Williams and Deane 48 ). One study had reported OR( Reference Michalak, Zhang and Jacobi 40 ), one had reported correlation coefficients( Reference Forsyth, Williams and Deane 48 ) and two studies had reported _β_-coefficients( Reference McMartin, Jacka and Colman 5 , Reference Crichton, Bryan and Hodgson 47 ). Owing to the heterogenous form of reporting of the findings, we were unable to perform meta-analysis about anxiety. All studies had considered depression. Four studies reported beta coefficients( Reference El Ansari, Adetunji and Oskrochi 46 , Reference Crichton, Bryan and Hodgson 47 , Reference Castellanos, Connell and Lee 49 – Reference Payne, Steck and George 52 ), one had reported standardised β ( Reference Whitaker, Sharpe and Wilcox 45 ), one reported correlation coefficient( Reference Forsyth, Williams and Deane 48 ) and others had reported RR, HR or OR. OR for the association of fruit intake and depression ranged between 0·61 and 1·10 in different studies. The corresponding figures for vegetable intake varied from 0·63 to 2·75. A total of fourteen studies were of high quality( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Chi, Wang and Tsai 17 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Niu, Guo and Kakizaki 25 , Reference Aihara, Minai and Aoyama 27 – Reference Woo, Ho and Yu 30 , Reference Crichton, Bryan and Hodgson 47 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Kingsbury, Dupuis and Jacka 51 ). Most studies had adjusted for age( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 – Reference Bishwajit, O’Leary and Ghosh 28 , Reference Woo, Ho and Yu 30 , Reference Michalak, Zhang and Jacobi 40 , Reference Shahar, Hassan and Sundar 43 , Reference Whitaker, Sharpe and Wilcox 45 – Reference Crichton, Bryan and Hodgson 47 , Reference Payne, Steck and George 52 ), sex( Reference McMartin, Jacka and Colman 5 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Verger, Lions and Ventelou 26 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Liu, Xie and Chou 42 , Reference Shahar, Hassan and Sundar 43 , Reference El Ansari, Adetunji and Oskrochi 46 , Reference Crichton, Bryan and Hodgson 47 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Payne, Steck and George 52 ), education( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Rienks, Dobson and Mishra 29 , Reference Michalak, Zhang and Jacobi 40 , Reference Liu, Xie and Chou 42 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Crichton, Bryan and Hodgson 47 , Reference Castellanos, Connell and Lee 49 , Reference Payne, Steck and George 52 ), physical activity( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Verger, Lions and Ventelou 26 , Reference Rienks, Dobson and Mishra 29 , Reference Crichton, Bryan and Hodgson 47 ), energy intake( Reference Akbaraly, Sabia and Shipley 10 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Niu, Guo and Kakizaki 25 , Reference Verger, Lions and Ventelou 26 , Reference Rienks, Dobson and Mishra 29 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Crichton, Bryan and Hodgson 47 ), BMI( Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Verger, Lions and Ventelou 26 , Reference Rienks, Dobson and Mishra 29 , Reference Crichton, Bryan and Hodgson 47 , Reference Payne, Steck and George 52 ) and smoking( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Hintikka, Tolmunen and Honkalampi 16 – Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Rienks, Dobson and Mishra 29 , Reference Liu, Xie and Chou 42 , Reference Crichton, Bryan and Hodgson 47 ). Some had also controlled for income( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Woo, Lynn and Lau 21 , Reference Gangwisch, Hale and Garcia 22 , Reference Verger, Lions and Ventelou 26 , Reference Rienks, Dobson and Mishra 29 , Reference Castellanos, Connell and Lee 49 ), marital status( Reference Akbaraly, Sabia and Shipley 10 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Rienks, Dobson and Mishra 29 , Reference Michalak, Zhang and Jacobi 40 , Reference Whitaker, Sharpe and Wilcox 45 ) and chronic diseases( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Chi, Wang and Tsai 17 , Reference Woo, Lynn and Lau 21 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Aihara, Minai and Aoyama 27 ).

Cross-sectional design( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Allgöwer, Wardle and Steptoe 23 , Reference Niu, Guo and Kakizaki 25 , Reference Aihara, Minai and Aoyama 27 , Reference Liu, Xie and Chou 42 , Reference Whitaker, Sharpe and Wilcox 45 – Reference Crichton, Bryan and Hodgson 47 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Payne, Steck and George 52 ), invalid exposure or outcome assessment tools( Reference McMartin, Jacka and Colman 5 , Reference Akbaraly, Sabia and Shipley 10 , Reference Chi, Wang and Tsai 17 , Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Liu, Xie and Chou 42 , Reference Shahar, Hassan and Sundar 43 , Reference Castellanos, Connell and Lee 49 – Reference Kingsbury, Dupuis and Jacka 51 ), lack of controlling for potential confounders( Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Gangwisch, Hale and Garcia 22 , Reference Niu, Guo and Kakizaki 25 , Reference Woo, Ho and Yu 30 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Kingsbury, Dupuis and Jacka 51 ), using self-reported questionnaire( Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Chi, Wang and Tsai 17 , Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Woo, Ho and Yu 30 , Reference Mikolajczyk, El Ansari and Maxwell 50 , Reference Payne, Steck and George 52 ), low respondent rate( Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Ho and Yu 30 , Reference Kingsbury, Dupuis and Jacka 51 ) and misclassification of participants based on assessing dietary intake by FFQ( Reference Sanchez-Villegas, Delgado-Rodriguez and Alonso 18 , Reference Gangwisch, Hale and Garcia 22 ) were the most common limitations of earlier studies. However, large sample size( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Lynn and Lau 21 , Reference Rienks, Dobson and Mishra 29 , Reference Liu, Xie and Chou 42 , Reference Kingsbury, Dupuis and Jacka 51 ), adjustment for most potential confounders( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Rienks, Dobson and Mishra 29 ), valid exposure or outcome assessment tools( Reference Mihrshahi, Dobson and Mishra 19 , Reference Woo, Lynn and Lau 21 , Reference Rienks, Dobson and Mishra 29 , Reference Liu, Xie and Chou 42 , Reference Whitaker, Sharpe and Wilcox 45 , Reference Kingsbury, Dupuis and Jacka 51 ) and the first study in a special population( Reference Bishwajit, O’Leary and Ghosh 28 , Reference Whitaker, Sharpe and Wilcox 45 ) were mostly reported as strengths of these publications.

Findings from the meta-analysis of fruit intake and risk of depression

Combining six effect sizes from six cohort studies( Reference Akbaraly, Sabia and Shipley 10 , Reference Chi, Wang and Tsai 17 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Rienks, Dobson and Mishra 29 ), we found that the highest v. the lowest intake of fruit was associated with a 17 % significant reduction in the risk of depression (Fig. 2) (overall RR=0·83; 95 % CI 0·71, 0·98). However, heterogeneity was significant (I 2=84·5 %, P<0·001). To investigate the source of heterogeneity, subgroup analyses were performed on the basis of location, sex, outcome assessment tools and study quality (Table 3). Sex (for female: overall RR=0·92; 95 % CI 0·78, 0·98, I 2=87·3 %, P<0·001, and for both: overall RR=0·72; 95 % CI 0·61, 0·84, I 2=7 %, _P_=0·34) and location (for Asian countries: overall RR=0·87; 95 % CI 0·78, 0·96, I 2squared=0 %, _P_=0·62, and for non-Asian countries: overall RR=0·81; 95 % CI 0·62, 1·05, I 2=89·1 %, P<0·001) were the sources of heterogeneity. Combining nine effect sizes from six cross-sectional studies( Reference McMartin, Jacka and Colman 5 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Allgöwer, Wardle and Steptoe 23 , Reference Liu, Yan and Li 24 , Reference Bishwajit, O’Leary and Ghosh 28 ) indicated that the highest v. the lowest intake of fruit was associated with a 24 % reduction in the risk of depression (overall RR=0·76; 95 % CI 0·63, 0·92, I 2=82·7 %, P<0·001) (Fig. 3). To investigate the source of heterogeneity, subgroup analysis was done on the basis of location, sex, dietary assessment tools, outcome assessment tools, energy adjustment and study quality (Table 3). Location (for Asian countries: overall RR=0·62; 95 % CI 0·41, 0·95, I 2=88·6 %, P<0·001, and for non-Asian countries: overall RR=0·93; 95 % CI 0·86, 1·01, I 2=0 %, _P_=0·62) and outcome assessment tools (for BDI: overall RR=0·89; 95 % CI 0·78, 1·01, I 2=0 %, _P_=0·72), for self-reported questionnaire: overall RR=0·58; 95 % CI 0·24, 1·38, , I 2=93·2 %, P<0·001, and for other outcome assessment tools: overall RR=0·76; 95 % CI 0·54, 1·07, I 2=87·5 %, P<0·001) were the sources of heterogeneity.

Fig. 2 Forest plots of the association between fruit consumption and risk of depression in cohort studies. RR, relative risk.

Fig. 3 Forest plots of the association between fruit consumption and depression in cross-sectional studies. RR, relative risk.

Table 3 Results of subgroup analysis for fruit intake and risk of depression based on study design (Odds ratios and 95 % confidence intervals)

For non-linear dose–response meta-analysis on fruit consumption, there was only one cohort study that had provided relevant information. Therefore, we excluded this cohort study from the dose–response analysis owing to insufficient number of cohort studies. Finally, the non-linear dose–response analysis on fruit consumption was confined to two cross-sectional studies( Reference Woo, Lynn and Lau 21 , Reference Bishwajit, O’Leary and Ghosh 28 ), which had provided four effect sizes. In this non-linear dose–response analysis, we found that increased intake of fruit was not associated with reduced odds of depression (P non-linearity=0·12) (Fig. 4).

Fig. 4 Dose–response association between fruit consumption and risk of depression in cross-sectional studies. , Linear model; , spline model.

On the basis of meta-regression on three effect sizes from three cohort studies( Reference Chi, Wang and Tsai 17 , Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 ), we found an inverse linear association between fruit intake and risk of depression, such that every 100-g increase in intake of fruit was associated with a 3 % reduction in the risk of depression (RR=0·97; 95 % CI 0·95, 0·99). With regard to cross-sectional studies( Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Bishwajit, O’Leary and Ghosh 28 ), fruit intake was not linearly associated with the risk of depression (RR=1·00; 95 % CI 0·99, 1·01). Sensitivity analysis showed that the overall effect did not vary substantially with the exclusion of any study. The Begg’s test (_P_=0·46) and Egger’s test (_P_=0·07) had shown no publication bias.

Findings from the meta-analysis of vegetable intake and risk of depression

Summary effect from seven effect sizes provided by seven cohort studies( Reference Akbaraly, Sabia and Shipley 10 , Reference Chi, Wang and Tsai 17 – Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Rienks, Dobson and Mishra 29 , Reference Woo, Ho and Yu 30 ) showed that high intake of vegetables was associated with a 14 % significant reduction in the risk of depression (overall RR=0·86; 95 % CI 0·75, 0·98, I 2=66·1 %, _P_=0·004) (Fig. 5). To find the source of heterogeneity, subgroup analysis was conducted on the basis of location, sex, outcome assessment tools and study quality. Sex (for female: overall RR=0·93; 95 % CI 0·86, 1·00, I 2=27·4, _P_=0·25, and for both: overall RR=0·81; 95 % CI 0·64, 1·02, I 2=53·8 %, _P_=0·09), location (Asian countries: overall RR=0·88; 95 % CI 0·79, 0·98, I 2=0 %, _P_=0·38, and for non-Asian countries: overall RR=0·84; 95 % CI 0·68, 1·04, I 2=81·9 %, _P_=0·001) and outcome assessment tools (for CES-D: overall RR=0·81; 95 % CI 0·65, 1·02, I 2=82·5 %, _P_=0·001, and for other outcome assessment tools: overall RR=0·89; 95 % CI 0·80, 0·99, I 2=0 %, _P_=0·45) were the sources of heterogeneity (Table 4). Combining eight effect sizes from six cross-sectional studies( Reference McMartin, Jacka and Colman 5 , Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 , Reference Shahar, Hassan and Sundar 43 ) indicated that higher intake of vegetables was associated with a 25 % reduced risk of depression compared with lower intake of vegetables (overall RR=0·75; 95 % CI 0·62, 0·91, I 2=56·8 %, _P_=0·023) (Fig. 6). Subgroup analysis on the basis of location, dietary assessment tools, outcome assessment tools, energy adjustment and quality score were performed to investigate the source of heterogeneity (Table 4). None of these variables explained heterogeneity.

Fig. 5 Forest plots of the association between consumption of vegetables and risk of depression in cohort studies. RR, relative risk.

Fig. 6 Forest plots of the association between consumption of vegetables and depression in cross-sectional studies. RR, relative risk.

Table 4 Results of subgroup analysis for vegetable intake and risk of depression based on study design (Odds ratios and 95 % confidence intervals)

A non-linear dose–response meta-analysis of cohort studies on vegetable consumption( Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Woo, Ho and Yu 30 ) revealed that increased intake of vegetables (at the level of 100–400 g/d) was associated with a reduced risk of depression; however, vegetable consumption in excess of 400 g/d was associated with an increased risk of depression (Fig. 7(a)). However, these findings were not statistically significant (P non-linearty=0·97). The non-linear dose–response meta-analysis of cross-sectional studies( Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 ) indicated that increased intake of vegetables was associated with a reduced risk of depression (P non-linearty<0·001) (Fig. 7(b)).

Fig. 7 Dose–response association between consumption of vegetables and risk of depression in cohort (a) and cross-sectional (b) studies. , Linear model; , spline model.

We performed linear meta-regression analysis because of a non-significant non-linear association between vegetable consumption and risk of depression. Four effect sizes from four cohort studies( Reference Chi, Wang and Tsai 17 , Reference Mihrshahi, Dobson and Mishra 19 , Reference Gangwisch, Hale and Garcia 22 , Reference Woo, Ho and Yu 30 ) were included in linear meta-regression. We found that every 100-g increase in consumption of vegetables was associated with a 3 % reduced risk of depression (RR=0·97, 95 % CI 0·95, 0·98). With regard to cross-sectional studies, combining six effect sizes from four studies( Reference Hintikka, Tolmunen and Honkalampi 16 , Reference Woo, Lynn and Lau 21 , Reference Niu, Guo and Kakizaki 25 , Reference Bishwajit, O’Leary and Ghosh 28 ), we found that every 100-g increased intake of vegetables was associated with a 5 % reduced odds of depression (RR=0·95; 95 % CI 0·91, 0·98). Sensitivity analysis showed that none of the studies had a significant effect on the overall effect size. No evidence of publication bias was seen (Egger’s test: _P_=0·20, Begg’s test: _P_=0·56).

Findings from the meta-analysis on total fruit and vegetable intake and risk of depression

Overall effect from six effect sizes, obtained from four cross-sectional studies( Reference McMartin, Jacka and Colman 5 , Reference Konttinen, Mannisto and Sarlio-Lahteenkorva 14 , Reference Verger, Lions and Ventelou 26 , Reference Aihara, Minai and Aoyama 27 ), revealed that high intake of fruit and vegetables was associated with a 20 % reduced risk of depression (Fig. 8) (overall RR=0·80; 95 % CI 0·65, 0·98). However, significant heterogeneity was observed between studies (I 2=71·1 %, _P_=0·004). We performed subgroup analysis on the basis of sex, location, dietary assessment tools, outcome assessment tools and quality score to investigate the source of heterogeneity (Table 5). Location (for Asian countries: overall RR=1·53; 95 % CI 0·96, 2·41, I 2=0 %, _P_=0·99, and for non-Asian countries: overall RR=0·72; 95 % CI 0·60, 0·87, I 2=69·1 %, _P_=0·02) and dietary assessment tools (for FFQ: overall RR=0·72; 95 % CI 0·60, 0·87, I 2=69·1 %, _P_=0·02, and for questionnaire: overall RR=1·53; 95 % CI 0·96, 2·41, I 2=0 %, _P_=0·99) were the sources of heterogeneity. Sensitivity analysis revealed that none of the studies had significantly influenced the overall effect. No evidence of publication bias was found (Egger’s test: _P_=0·82, Begg’s test: _P_=0·57).

Fig. 8 Forest plots of the association between total intake of fruits and vegetables and risk of depression. RR, relative risk.

Table 5 Results of subgroup analysis for total intake of fruit and vegetables and risk of depression (Odds ratios and 95 % confidence intervals)