Study characteristics
The Search strategy (done in May, 2017) identified 1001 relevant citations. 28 additional studies were identified through asking clinicians with expertise about the topic area and reference mining. A total of 30 unique studies met the inclusion criteria enrolling 7381 patients (Fig. 2). Adults were enrolled in 14 studies [14,15,16,17,18,19,20,21,22,23,24,25,26,27], children in 12 studies [28,29,30,31,32,33,34,35,36,37,38,39], and both age groups in 4 studies [40,41,42,43]. The majority (17) were nonrandomized controlled studies [15, 17,18,19,20,21, 26, 27, 29,30,31,32,33, 35, 36, 38, 39]. There were 7 pre-post studies [14, 16, 23, 28, 37, 41, 42] and 6 RCTs [22, 24, 25, 34, 40, 43].
Description of the intervention
Cooking classes ranged in duration from 2 weeks to 104 weeks (2 years), with a mean duration of 21 weeks, whereas the number of sessions ranged from 1 session to 52 sessions, with a mean of 8 sessions. The class instructor was a chef in 9 studies [17, 18, 20, 22, 26, 29, 33, 35, 39], a dietitian in 10 studies [14, 15, 19, 21, 23,24,25, 28, 40, 42], an educator in 7 studies [27, 30,31,32, 34, 36, 38], and 4 studies did not report the class instructor. In 21 studies, the cooking classes were participatory [14, 15, 19,20,21,22, 24, 26, 28, 32,33,34,35,36,37,38,39,40,41,42,43], whereas 3 studies featured a cooking demonstration [16, 18, 30] and 6 studies were not clear in that aspect. Additional file 1: Table S1 lists the included studies’ characteristics.
The mean follow-up duration was 25 weeks. 27 studies reported a behavioral outcome [15,16,17,18,19,20, 22,23,24, 26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43], whereas 11 studies reported cardiometabolic outcomes and anthropometrics measures [16, 17, 19, 21, 22, 24, 25, 34, 38, 42, 43], and 2 studies reported on quality of life [15, 21]. Additional file 1: Table S2 lists the outcomes evaluated in each included study.
Risk of Bias
Six RCTs were assessed using the Cochrane Risk of Bias tool. All 6 RCTs have a high risk of bias, mainly because of inadequate procedures for randomization, concealment and blinding, or poor reporting of how these design features were achieved. The assessment of these studies is listed in Additional file 1: Table S3.
The remaining 24 studies were assessed using items from the Newcastle-Ottawa Scale. 17 studies were Nonrandomized controlled trials and 7 studies were Pre-Post studies. 6 studies had a high risk of bias, and 19 studies had a moderate risk of bias. Most studies were deficient in the component of comparability of the design or analysis (matching or confounder adjustment) and in the assessment of outcomes which relied on self-report in most cases and were subject to recall bias or interviewer bias (reporting bias). Additional file 1: Table S4 lists the risk of bias assessment for these 24 studies.
Effect on anthropometrics and cardiometabolic outcomes
Meta-analysis did not show a significant effect of culinary interventions on BMI, SBP, DBP or LDL-C.
Figure 3.a depicts the meta-analysis of BMI. Five studies [16, 17, 19, 22, 43] were analyzed for this outcome, with an overall population size of 1292 persons. The mean follow up duration was 30 weeks. The overall mean difference was − 0.07 kg/m2 (95% CI: -1.53, 1.40).
Figures 3.b and 3.c illustrate the meta-analysis of SBP and DBP, respectively. Four studies [19, 22, 25, 34] were analyzed for these outcomes, with an overall population size of 478 persons. The mean follow up duration was 15 weeks. The overall mean difference was − 5.31 mmHg (95% CI: -34.2, 23.58) for SBP and − 3.1 mmHg (95% CI: -23.82, 17.62) for DBP.
Figure 3.d represents the meta-analysis of LDL-C. Three studies [19, 22, 34] were analyzed for this outcome, with an overall population size of 410 persons. The mean follow up duration was 17 weeks. The overall mean difference was − 8.09 mg/dL (95% CI: -84.43, 68.25).
Only one study [34] evaluated the outcomes of glucose, insulin, and insulin resistance (HOMA-IR). None of these outcomes significantly changed (p values of 0.56, 0.88 and 0.85, respectively). Only one study evaluated the outcomes of HbA1c [22] which also did not change significantly (p = 0.58).
Attitudes, self-efficacy and dietary intake
Data on attitudes, self-efficacy and dietary intake were not reported in a way to allow quantitative analysis; therefore, these outcomes were reported narratively.
Participants’ attitudes (e.g. how likely are you to eat the following foods? [33] or eating healthy is important to me [20]) improved in adults and children (medium risk of bias).
Self-efficacy (e.g., do you believe you can eat correct portions? [22]) also improved in adults and children (medium risk of bias).
Both children and adults had improved healthy dietary intake after culinary interventions (medium risk of bias). For instance, Newman, 2005 [23] found a significant increase in total daily vegetable, fruit and fiber intake, as well as a significant decrease in fat intake after a cooking classes intervention. Quality of life in adults may have improved after the intervention in adults (2 small nonrandomized studies with medium risk of bias).
We conducted regression analysis to explore the effect of several possible effect modifiers on the success of the intervention. This analysis demonstrated no significant associations between the success of the intervention and sample size, cooking class provider, population (children vs. adults), whether the class was participatory (vs. demonstration), number of sessions, and the intervention duration. Results are depicted in Additional file 1: Table S5.
Interventions with additional components
Several studies employed multicomponent interventions. The additional components were gardening education, dietary education, physical activity recommendations, goal setting and grocery store tours. These studies are summarized in Additional file 1: Table S6. In general, these studies showed statistically significant improvements in participants’ self-efficacy, dietary intake and attitudes. In one study, Curtis et al. [40] performed a randomized cluster trial in which 169 families (with 589 individuals) were randomized to 3 groups: nutrition education, cooking classes, and cooking classes + nutrition education + goal setting. The authors analyzed fat, carbohydrate, protein, vitamin C, and iron intake as well as energy density. There was a substantial drop out rate (25% at 3 months and 60% at 18 months). Fat intake decreased and carbohydrate intake increased (this was the goal of the project) more in the combination group than in the education group. Cooking classes alone had an intermediate effect (not statistically different from either group). Differences were not sustained at 18 months. Energy density (calculated and expressed as KJ/g of food) was significantly lower in the combination group at 18 months, but not at 3 months.
Gatto et al. [34] performed a randomized cluster trial in which 4 elementary schools (319 students) were randomized to a control group or an intervention which consisted of 12 ninety minute sessions. 45 min were gardening/nutrition lessons and 45 min were cooking/nutrition lessons. There were significantly greater declines in BMI and waist circumference in the experimental group versus the control group. There was also a difference in fiber intake between the two groups (with a small increase in the experimental group and a large decline in the control group). Correction for multiple comparisons was not done because this was a preliminary study.
Evidence map
To summarize quantitative and qualitative results, the various outcomes of this systematic review are presented in an evidence map demonstrating the effect of culinary interventions, the risk of bias, and study design contributing to each outcome (Fig. 4).