In partnership with Sanitas Medical Center, a medical group practice and healthcare facility with locations in the Houston and Dallas regions, research staff recruited patients from Sanitas clinics that met the study inclusion criteria. The inclusion criteria were being an adult (18 to 70 years) with a T2DM diagnosis and HbA1c > 7.0, receiving care at Sanitas Medical Center clinics, and fluent in English or Spanish. Additionally, we screened for eligibility following these criteria: (1) having access to reliable internet and a device, like a cell phone, tablet or laptop, (2) someone in the household that was able to go grocery shopping before each class.

The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of The UTHealth Houston (HSC-SPH-21-0555) on September 22nd, 2021. We started the rolling recruitment period from January 25th, 2022, to October 11th, 2022. Sanita’s care coordinators assessed individuals’ eligibility to participate and provided a printed flyer and sent a text message with a weblink to an online form for people who showed interest in the program. They completed the online interest form, sharing their contact information with UTHealth Houston. All recruitment materials were available in English and Spanish, and UTHealth Houston recruitment staff were fluent in both languages. After receiving their information, trained staff from UTHealth Houston contacted individuals via phone using an approved script for assessing eligibility and interest, and explained the program, including time commitment and potential advantages of participating in the program and minor risks while cooking at home (e.g., injuries with the use of knives, other cooking utensils, stoves, and ovens). After eligibility screening, participants provided written informed consent via REDcap (Research Electronic Data Capture), a secure, web-based application designed to support data capture for research studies (39). A bilingual staff member addressed any questions or concerns participants had regarding the program.

There are five main modifications that were made to adapt APHL for virtual delivery and NCCM implementation. First, the delivery format of the program was changed so that classes could be delivered through an online video conferencing platform (WebEx), with the main instructor hosting the virtual class live from the Nourish Teaching Kitchen at UTHealth Houston. One of the motivators for adapting APHL for virtual delivery was to reach a broader range of participants while COVID-19 restrictions (e.g., limitations on gathering in large groups) were still in place. Second, rather than providing food prescriptions, participants were provided electronic grocery cards via text message and email, to cover the cost of the ingredients and a shopping list for each class, which included tips on which items to choose to help participants with their shopping experience. These electronic grocery cards could be redeemed at any location of our chosen supermarket chain. Third, to further facilitate participant engagement and learning before and after class, we provided asynchronous virtual toolkit education content via text and email, which included links to access cooking skill videos, animated nutrition education videos, step by step recipes, shopping lists and kitchen set-up for each recipe, and nutrition and diabetes education handouts and worksheets. All of these materials were included in the study website; the development of this toolkit is described elsewhere (40). Participants used this toolkit to prepare for each culinary medicine session and as a reference post-session. Fourth, classes were offered during convenient times (usually after work hours, before standard dinner time) and the recipes were designed for a family of four. Fifth, cooking classes and toolkit learning materials were made more culturally inclusive (Table 1). To represent different ethnic groups, we specifically designed the intervention recipes to include a variety of cooking styles and flavors. For example, the stir fry was inspired by Asian cuisines, the turkey chili featured typical Hispanic/Latino flavors, and the pasta dish resembled more classic American cuisine. These recipes were pilot tested during the previously described APHL program (33). Additionally, the sessions were taught in two languages (English and Spanish).

This program consisted of five hands-on, 90-min virtual sessions. The sessions were facilitated by a native bilingual (English and Spanish) Registered Dietitian with culinary medicine training, with the aid of a co-teacher/assistant trainee. Kitchen setup was performed by the trainee, who additionally troubleshot technical issues (e.g., difficulties with accessing the video conferencing platform) and operated the kitchen studio cameras to alternate between a view of the instructor in the kitchen and an up-close, top view of the counter. This process allowed for lower staffing requirements for intervention delivery. The NCCM topics covered during the five sessions are presented in Table 1. Sessions were structured with a 15-min community connection and session introduction, 45–55 min of hands-on cooking, and 20–30 min of educational videos and facilitated group discussion. Participants were instructed on how to prepare their cooking station ahead of each class (i.e., have cutting board and chef knife ready, vegetables washed, and ingredients measured). Each session focused on two different cooking techniques widely shown to be the most important for improving healthy eating. Cooking videos explaining the different techniques were also provided ahead of time for participants to watch at their leisure and for reinforcement after sessions. During passive cooking time, 2–3 animated videos were played to facilitate discussion of the topics listed in Table 1. After each video, the instructor facilitated group discussion, with participants sharing their personal experiences and what they learned from or reactions to the videos. At the conclusion of each session, participants were encouraged to discuss likes, dislikes, recommendations for others, and how to personalize the recipes for their families – including adapting the recipe to fit cultural or individual preference. This conversation created additional ideas for creative ways to prepare food, driven by participants, to facilitate the incorporation of these recipes into their regular rotation. A printed toolkit with QR codes was mailed to participants’ homes that included all recipes and shopping lists along with handouts participants could fill out on their own time.

The study design was a single-arm prospective cohort intervention. The 6 cohorts (3 in English and 3 in Spanish) took place throughout 2022. The first three cohorts received the classes in the summer every other week. After implementing the program with these three cohorts, participants expressed an interest in having classes more frequently and we thought this might improve retention in the program. Thus, Cohorts 4–6, which were conducted during the fall, were conducted weekly, covering the same curriculum. We collected self-administered questionnaires at pre-test and post-test to evaluate participants’ dietary behaviors, nutrition knowledge, cooking skills and behaviors. Electronical medical record (EMR) data were collected at pre-test, post-test, and 6-month post-test (Figure 1). Additionally, we collected qualitative data from semi-structured interviews immediately post-intervention to evaluate patients’ experience during the intervention and satisfaction with the program overall.

The partnership with Sanitas Medical Center sites allowed the use of participants’ EMR data to evaluate the impact of the NCCM program on biometric outcomes. Data were abstracted and shared by clinic coordinators with UTHealth Houston at three timepoints: pre-test (within 90 days before starting the NCCM program), at program completion (within 90 days of NCCM program completion), and 6-month post-test (within 90 days of 6-months completion of NCCM). These time measurements were chosen based on similar studies (41–45). Biometric outcomes included glycosylated hemoglobin (HbA1c), Body Mass Index (BMI), systolic blood pressure (SBP), and diastolic blood pressure (DBP). Additionally, demographics (age, sex, race/ethnicity) were sourced from the patient EMR records.

Participants were asked to complete a 15-min electronic questionnaire via REDcap at two timepoints: a pre-test questionnaire before the first class and a post-test immediately after the last session. Participants received a $25 gift card as compensation for their time for completion of each questionnaire ($50 total for study questionnaire completion). The pre-test and post-test questionnaires collected information on dietary intake, cooking skills, cooking-related psychosocial constructs, and diabetes management (details in the Measures section). Additional demographics were collected at pre-test including education, employment status, preferred language spoken at home, transportation, and participation in government assistance programs.

After each session, participants received a link to a virtual comment card in which they were asked to share feedback on the specific recipe, cooking method, and overall class flow. Participants voluntarily responded to the comment cards, and completion varied by participant. Each participant identified the class they attended (sessions 1–5) and then were asked to answer questions on how helpful they found the cooking techniques, class activities, and animated videos. Participants selected ‘yes’ or ‘no’ if they would make the specific recipe again on their own, if ‘no’ was selected then they were asked to explain in an open response. An additional comment section at the end encouraged participants to leave further feedback for improvement. These comments and feedback were not anonymous since the comment cards were linked to the participants’ IDs.

Qualitative interviews were conducted virtually by the research team at the end of the intervention to gather more robust information on the participants’ experience, identify areas for additional improvement in both the intervention and study design, and explain quantitative outcomes. Participants received a $20 gift card as compensation for their time after completing the interview. The semi-structured interview guide was adapted from the interview questions from a similar program (33) and was focused on assessing key feasibility and acceptability outcomes through open-ended, non-leading, and progressive questions (43–45). The topics included enrollment, engagement, experience with classes and resources, and comments regarding the program overall (Table 2). At the conclusion of the intervention, all enrolled participants were invited via text and email to take part in the virtual interviews. Depending on the preference of the participant, the interviews were conducted in either Spanish or English and all interviews were recorded. For English interviews, a transcription was performed by a smart note transcription service (Otter.ai) and inspected by staff for accuracy. For Spanish interviews, a transcription and translation into English was performed by bilingual research staff.

Feasibility outcomes were recruitment, retention, acceptability, and satisfaction (39). The feasibility of recruitment was determined by the number of individuals who were invited to participate by Sanitas staff, screened for eligibility, and then enrolled in the study. Records were maintained for all stages of recruitment in REDCap. Retention was measured by the proportion of participants who remained in the pilot study at the end of the intervention and completed post-test data collection. This was measured through attendance records. As recommended for pilot studies, a priori targets for recruitment and retention were set using relevant literature and the authors’ prior experience: > 60% of eligible individuals enroll, and ≥ 70% of participants are retained in the study (33). Acceptability and satisfaction were measured informally with comment cards to continuously engage with participants during the program and formally through in-depth interviews with participants post-test to understand participants’ experience and learning journey throughout the NCCM program.

Short-term effectiveness outcomes (Table 3) were measured through self-administered questionnaires. Perceived health was measured by 1 item validated survey with response options from Excellent (6) to Very Poor (1), with higher scores indicating better perceived health (46). Average daily servings of fruits and vegetables were assessed with 2 items (1 for fruits and 1 for vegetables), with response options from None (1) to 4 or More servings (6). These items were adapted from a previously validated instrument to assess fruit and vegetable consumption (47). The two items were assessed separately, as well as summed, for a maximum score of 12, with higher numbers indicating a greater number of servings of fruits and vegetables (47). Frequency of healthy food consumption was measured by 7-item validated measure to assess typical food consumption behaviors, with response options from Not at all (1) to More than once a day (5); a sum was created, for a maximum score of 35, with higher numbers indicating greater frequency of healthy food consumption (48). Shopping, cooking, and eating behaviors were measured by 8 items, adapted from a previously validated survey, with response options from Never (1) to Always (5); items were summed for a maximum score of 40, with higher numbers indicating more ideal, healthy behaviors (48). Cooking self-efficacy was assessed by 10 items, not previously validated, at post-test only, five items that ask about self-efficacy before the program, and five ask about self-efficacy after the program, with response options of Not sure at all (1) to Extremely sure (5) (49). The scores were summed, and a higher score suggests greater self-efficacy for cooking. Diabetes self-management was measured by a 16-item validated scale, with response options from Does not apply to me (1) to Applies to me very much (4); a summed scale was created per standard protocol (50). Perceived barriers to healthy eating consisted of 9 items, not previously validated, with response options from Strongly Disagree (1) to Strongly Agree (5) for a maximum score of 45, with a lower number indicating fewer perceived barriers (48). And finally, nutrition knowledge was assessed with a single item, not previously validated, with response options providing different pictures of MyPlate; the correct answer was scored 1, and wrong answers were scored 0 (33).

Acceptability and satisfaction were assessed through analyses of in-depth interviews. Transcripts were uploaded in NVivo 12 (51), a qualitative data analysis computer software, and the analyses were conducted by two researchers independently. We used the framework approach (52, 53), involving three stages: data management, descriptive accounts, and explanatory accounts (53). During the data management stage, two research staff members familiarized themselves with the data, reading the interview transcripts, developing a coding index, and assigning codes for data analysis in NVivo 12. Twenty-seven percent of the transcripts were then double coded independently by another member of the research staff for quality control. Double coding resulted in 98.9% agreement. After transcript coding was complete, the researchers moved to the descriptive stage of analysis in which they identified associations between the codes and developed categories in which the data moved from anecdotal to more general themes. Finally, in the explanatory stage, researchers developed patterns within the themes, reflecting on the original data to ensure accurate representation of participant accounts.

We analyzed feasibility outcomes and demographic characteristics using descriptive statistics, including means and standard deviations or the observed number and percentage, as appropriate. Recruitment and retention outcomes were calculated as percent of eligible individuals enrolled and percent of participants completing questionnaires. We assessed changes from pre-test to post-test for questionnaire (perceived health, servings of fruits and vegetables, frequency of healthy food consumption, shopping, cooking, and eating behaviors, cooking self-efficacy, diabetes self-management, barriers to healthy eating and nutrition knowledge) and EMR data (HbA1c, BMI, SBP and DBP) using both unadjusted and adjusted multilevel mixed-effects models. We used linear, ordered, and logistic regression, as appropriate for each outcome. Models were adjusted for the individual as a random effect. Covariates were tested, and adjustments were performed when estimates changed by ≥10%.