A school-based non-randomized controlled interventional study with parallel group design was conducted in two co-educational English medium schools in a selected West Bengal municipal area. The data collection was done between March 2019 and January 2020.

Adolescents in seventh to tenth grades aged between 12 and 16 years, studying in the selected schools, whose parents gave consent and who provided assent for participation, were included in the study. Those who were absent at any phase of the study were excluded from the study. Multi-stage sampling was done. In the first stage, two schools were chosen based on enrolment and attendance (one for the intervention group and another for the control group) from the schools in the study area (Uttarpara-Kotrung Municipal area). In the selected schools, sections were chosen based on Probability Proportionate to Size (PPS) method in each grade. Then in the particular section, complete enumeration was done. The sample size was estimated using Fleiss' formula for difference in proportions for parallel group intervention study design (with both groups being equal in size) (24). For sample size calculation, a power of 80% and a confidence level of 95% were assumed. A design effect of 1.5 was used, and an attrition factor of 10% was considered. Based on a school-based study by Kebaili et al. (25), an increase in the proportion of intenders of healthy dietary habits (intention to eat vegetables regularly) among the control group taken as 5%, i.e., P_control = 0.05, and among intervention group 20%, i.e., P_intervention = 0.2. Therefore, using the Fleiss formula for difference in proportions, the minimum adequate sample size in each group was 76. With the design effect, the minimum sample size was 114 and using an attrition factor of 10%, the optimum sample size was 125 for each intervention and control group. Thus, the minimum sample size required in each group was 114; accounting for attrition, it was 125. Finally, after completing the post-intervention phase, 133 completed responses in the control group and 118 completed responses in the intervention group were obtained. Figure 1 depicts the selection of study participants.

Among the final 251 responses, a significant proportion of the intervention group (33.05%) belonged to the age group of 15 years, and the control group (37.60%) belonged to the age group of 14 years. Most participants in the intervention group (67.80%) were boys, whereas most participants (50.38%) were girls in the control group. The socio-demographic characteristics of the participants are given in Table 1.

Behavioral intention toward consumption of a healthy diet was determined utilizing the constructs from the TPB: (i) “Attitude” consisting of domains such as Behavioral belief and Evaluation of behavioral outcome, (ii) “Subjective norm” consisting of Normative beliefs and Motivation to comply, and (iii) “Perceived behavioral control” (PBC) consisting of Control beliefs and Perceived power. Items for intention measurement and identifying intenders were selected through elicitation interviews. These interviews were conducted based on the elicitation interview guide, which focused on the TPB's constructs. Elicitation interviews were done among 20 students of classes VII to X in a separate English medium co-educational school of the study area. The attitude was measured through the items: Sufficiency in consumption of vitamins and minerals, consumption of oily food, Sufficient protein intake, Obesity/overweight, Healthy life span, Taste of Food, and Cost of Food. For Subjective Norms, the items were: Mother, Father, Relatives/other family members, Friends/peers, Teachers, Contents of television, and Contents/discussions in social media. For the PBC construct, the items were as follows: choosing a healthy diet even when hungry, Not choosing unhealthy food even if they are tasteful, choosing to eat a healthy diet even when depressed or sad, opting to eat a healthy diet even when junk food is readily available, choosing to eat a healthy diet even during celebrations or parties, choosing to eat a healthy diet even while visiting any mall/theater, choosing to eat a healthy diet while traveling. Each item was measured on all three constructs through their respective domains using a dichotomous (Agree-Disagree) response. In this study, the intention was not directly computed; instead, the participants with higher intentions (the intenders) were identified in pre- and post-intervention phases through a combination of construct-wise measurements.

A pre-designed, pretested and validated questionnaire was used to assess behavioral intention. After initial preparation, the questionnaire was reviewed by a group of 5 experts who made necessary corrections. Pretesting of the questionnaire was done among 40 students. Cronbach's alpha measured internal consistency reliability construct-wise with values ranging between 0.79 and 0.86. The questionnaire consisted of socio-demographic characteristics and the three constructs of TPB that determined the intention to consume a healthy diet (refer to Supplementary material).

The intervention booklet and demonstration material were designed, developed, and pretested based on the construct-wise findings from elicitation interviews on healthy dietary practices. The conclusions of the pre-intervention survey were utilized in refining the content of the intervention tools. The intervention booklet contained information on different compositions of a healthy diet, accessible and affordable healthy food items, and the importance of developing and maintaining healthy dietary habits. During the interactive sessions, discussions were conducted on common myths and beliefs regarding healthy and unhealthy foods. Thrust was given to facilitating the enablers and overcoming the barriers to healthy dietary practices. Normative influencers and their role in motivating the participants to adopt healthy dietary habits were also discussed. The intervention tools were prepared with subject experts, following standard guidelines on healthy nutritional practices. The intervention materials were also pretested before implementation.

Baseline data were collected from the students of grades VII–X of both the intervention and the control groups with the help of the self-administered questionnaire after obtaining assent. Each participant was allotted an Identification (ID) code, which was further used to associate the data from the same participant upon follow-up. After baseline data collection, the intervention was provided to the intervention group. The intervention was imparted grade-wise, i.e., students from a particular grade were intervened at a time. Booklets were distributed among the participants, and then lectures were conducted based on the booklet's contents. Pre-designed computer-based slides were used as visual aids for pointing out the existing disconnects between their thoughts and action. In between lectures and demonstrations, interactive sessions and discussions were also conducted. After completion of the sessions, queries or doubts were clarified through a question-and-answer session. Follow-up data were collected 3 months after the conclusion of the intervention. Follow-up data were also collected from the control group using the same questionnaire but without intervention. At the end of follow-up data collection, health education interventions were conducted among the control group during the study period in the same manner as in the intervention group.

Pre-intervention and post-intervention data entered in a spreadsheet were linked through the unique codes. The cleaned dataset was used for statistical analysis using Statistical Package for Social Sciences (SPSS) software version 21 (IBM, Chicago, IL, USA). For computation purposes, items were coded 1 and 2, with 2 indicating a favorable response toward healthier habits. Negatively framed items were reverse-coded while maintaining the directionality of scoring, i.e., a favorable response was coded higher. Construct scores were calculated by multiplying the scores for the contributing domains (21), e.g., “Attitude” was calculated by multiplying the Behavioral belief score and evaluation of behavioral outcome score. A similar calculation was done for “Subjective Norm” and “Perceived Behavioral Control.” A linear combination of the responses of the respective item sets computed individual domain-specific scores. Intenders (i.e., those having the latent variable “Behavioral intention”) were identified by applying two-step cluster analysis with maximum likelihood estimation based on the calculated Attitude, Subjective norm, and Perceived behavioral control scores, separately for pre-intervention and post-intervention phases. Both intervention and control group measurements were taken for the cluster analysis. Those with a higher mean score in the constructs were considered intenders, and the remaining were non-intenders (refer to Table 2). Using the Silhouette measure, cluster analyses were found to be statistically adequate. The item-specific responses and proportions of intenders were compared between the pre-intervention and post-intervention phases in both the study groups.

The Chi-square test was employed to test for any statistically significant difference in each item in the two study groups before and after the intervention phase. The construct scores were compared (pre- vs. post-intervention) with the help of paired t-tests in each study group. The intervention effect was measured using Relative Risk (RR) for being in the higher intention cluster through Generalized Linear Model (GLM) with the log-linear link under Poisson distribution assumptions and robust standard errors. The model was adjusted for the effects of gender, the baseline (pre-intervention) intention cluster, and their interaction with the receipt of intervention (i.e., study groups). The model was statistically significant (P_χ2 <0.001). A two-tailed P-value of 0.05 or less was considered statistically significant in all the statistical techniques.

Approval for the study was obtained from the Institutional Ethics Committee of All India Institute of Hygiene and Public Health, Kolkata. Permission from the head of the schools was taken before data collection. Informed written assent was obtained from the participants, and informed written consent was obtained from the guardian.

Regarding the choice of food based on its cost, among intervention group participants, there was a significant difference post-intervention in their behavior belief and evaluation of behavioral outcome (Table 3). Relatives/other family members were the important normative influencers among the intervention group participants after the intervention (Table 4). The intervention group participants had greater control belief and perceived power for “opting to eat a healthy diet even when junk food is easily available” after the intervention (Table 5).

In both the intervention and control groups, the Subjective norm and Perceived behavioral control mean scores increased in the post-intervention phase with a broader range of scores (Figure 2). The RR of becoming an intender for healthy diet consumption in the Intervention group compared to the Control group was 2.07 (1.44–2.97). Table 6 represents the generalized linear model showing the effect of intervention in improving intention.

The change in intention status depends upon several external and environmental factors other than the constructs and the items in the TPB. Because of the focused use of the model framework, these exogenous variables were not accounted for in the study. As the participants' self-reported information on the constructs, there is a possibility of social desirability bias, despite the expressed confidentiality protocol.