The study evaluated a mobile phone–based application designed as an assessment and intervention tool to improve dietary behaviors and reduce weight. Designing the system began with formative research with overweight men and women to solicit feedback about dietary behaviors, current mobile phone and text and picture message habits, the type and frequency of text and picture messages helpful for weight loss, and nutrition-related topic areas that should be included in a weight loss program. Focus group participants also tested prototypes of the system by receiving and responding to sample text and picture messages. While men and women differed on the preferred number of messages per day, both groups agreed that messages related to motivation, progress with weight loss, tips, and hints would be welcomed.

The system was developed to be both personally tailored and interactive. Personal tailoring was accomplished by providing flexibility in the number and timing of receipt of messages each day. For example, users could choose different times during the day to receive a message—typically one in the morning and one in the evening, with one to three additional messages when the user thinks a reminder would be helpful. A database was developed of over 3000 SMS text and Multimedia Message Service (MMS; essentially small pictures) messages and 1500 rules that determined what message was sent based on the day of week, time of day, and the participant’s eating behaviors and previous replies, among other parameters. Approximately half of the messages requested a reply, with the balance providing tips, suggestions, and positive reinforcement or encouragement for improved behaviors. To minimize annoyance, the system was designed so if the user chose not to respond to a message, it reduced the number of messages requesting replies until a response was provided. Also, the user was given the opportunity to change the time and frequency of the messages after gaining experience with use of the system.

The intervention was organized by topics that changed each week as the participant proceeded through the intervention. The weekly topics included behavioral and dietary strategies known to positively influence weight control, including goal setting and self-monitoring, understanding calories, portion control, pedometers and physical activity, personal strategies for weight loss and overcoming barriers, volumetrics (consuming foods that are healthy and make one feel “full”), replacement and substitution, routine physical activity, organization and meal planning, strategies for eating out, strategies for creating healthy food and exercise environments, strength training, emotion eating, managing tough social situations, body image, and sticking with it. In addition to weekly topics, each week participants were instructed to weigh themselves, fill out worksheets outlining their food and exercise goals, and set goals for a daily 500 calorie reduction using the strategies provided in the corresponding weekly topic. For example, in week 3, participants learned about portion control. Also in week 3, participants were instructed to select portion control strategies that they would use in order to reduce their daily calories. To support self-monitoring, participants used the mobile phone to report their weight once a week. A graph of weight change was also sent to participants weekly.

To keep the text messaging novel and non-repetitive, the type and content of the messages sent throughout the week changed from day to day (Table 1), and participants never got the exact same message throughout the entire duration of the study. As an example, a person requesting two daily messages, one in the morning and the other in the evening, might receive the following:

For example, a participant who was identified as having difficulty eating fruits and vegetables might get the text message tip “In a rush? Buy pre-cut vegetables like carrots, celery, and mushrooms for a quick, easy, and low calorie snack!” or the question “Did you buy fresh raw vegetables at the grocery store for snacks this week? (y/n)”. If a participant responded to a text message that was a question, he or she would always get a response based on the answer provided. For the above question (“Did you buy fresh raw vegetables at the grocery store for snacks this week?”), if the participant responded “no,” he or she would see the response “Be sure to add your favorite raw veggies like carrots, red peppers, and mushrooms to your shopping list.”

The overall system consists of four components: (1) a Web-based application to enroll participants and set user preferences; (2) a database to store the participants’ records, rules, and messages sent and received; (3) an application to determine the appropriate timing and message to send and to process the received replies; and (4) a text message delivery/reception platform. The system also had tools that enabled continuous technical monitoring to recognize anomalies such as messages and rules missing from the database, logic mistakes, or unexpected responses from participants, indicating that an individual may be having difficulties with the system. These tools alerted the case manager, who could contact the participant by phone or email to prevent user frustration and increase adherence and satisfaction.

A baseline dietary assessment [20] for each participant was used to identify unique diet behavior challenges that may contribute to increased caloric intake (eg, snacking behaviors, pacing of consumption, and self-monitoring of food intake). The server processed these data to create goals to target based upon particular logic rules of the expert system. These goals were then presented to the user via text or MMS messages to serve as prompts for food selection and behavioral improvements. At intervention onset, participants were given a printed binder with nutrition topics and behavioral strategies to supplement the phone-based messaging and a food and exercise journal to support self-monitoring. They also received brief (5-15 minute) monthly phone calls from a trained health counselor to encourage continued participation in the program and to work through any technical issues they might be having with the intervention. During the counseling calls, the health counselor followed a script designed to assess progress and barriers to weight loss. In addition, the health counselor would inquire about any of the following: weekly topics (Which topics are helpful? What have you learned? Is there any information missing?); text messages (Are they helpful?); current physical activity and eating plan (What is your current physical activity plan? How are you cutting calories?); social support (Have you been getting social support from the people around you?); or environment (Have you made any changes to your environment?). The health counselor would then provide feedback on what was discussed during the phone call and offer solutions to overcoming barriers.

The usual care comparison group received the same baseline dietary assessment as the intervention group and was mailed one to two pages of print materials once a month for 4 months. The print materials differed from the binder received by the intervention group. Print materials included information and tips on nutrition and weight loss, walking basics, fruit and veggie basics, and how to make physical activity a habit. Some of the topics and content overlapped information provided in intervention group’s binder and text messages. The comparison group did not receive telephone calls or text/MMS messages.

Evaluation of the system was accomplished in a randomized controlled trial with participants randomized to either the intervention or usual care comparison group (Figure 1). Assessments were completed at baseline, 2 months, and 4 months. The study was approved by the Institutional Review Board at the University of California, San Diego, CA.

The primary outcome was weight in kilograms measured in the research offices using a calibrated scale.

Participants were recruited in 2007 from the community via newspaper ads, flyers, and announcements on craigslist. Interested individuals called the telephone number provided on the recruitment materials. Recruitment staff answered the telephone and screened potential participants for meeting the following inclusion criteria: 25 to 55 years old, overweight (BMI ≥ 25-39.9), not taking medications known to cause weight gain, and present use of a mobile phone for sending and receiving SMS messages or a willingness to learn. Individuals eligible and interested in the study were transferred to the study staff, who scheduled a baseline visit. During the scheduling process, individuals were assigned to either the comparison or intervention group by a computer-generated process using simple randomization. Participants were not informed of their group allocation status during this process. Following baseline measurements, neither study staff nor participants were blinded to participants’ allocation. As no studies using SMS messages for weight control had been previously published, we did not have guidance on sample size for powering the study, so we based enrollment numbers on our best estimate of likely effects.

Figure 1

Flow diagram of participants

Initial descriptive analyses examined sample characteristics, the randomization of participants, and distribution of variables. A mixed-model repeated-measures analysis compared the effect of the intervention group to the comparison group on weight status over the 4-month intervention period. The mixed-model analysis allowed for inclusion of all available data with missing values assumed missing at random. Model building methods outlined by Singer and Willett [21] were followed. An unconditional means model of baseline weight was evaluated first, followed systematically by models including a within-subject factor of time (0, baseline; 1, 2 months; 2, 4 months), a between-subjects factor of treatment group (1, intervention group; 0, comparison group), and the treatment by time interaction. Further models were planned to include sex, race (white versus non-white), ethnicity (Hispanic versus non-Hispanic), and age (mean centered). Neither race nor ethnicity independently influenced initial weight status or rate of weight change and were dropped from the models. A heterogeneous first-order autoregressive error covariance structure was specified for all models. Model fit was compared using the deviance statistic, the Akaike information criterion, and the Schwarz Bayesian criterion. Additionally, analysis of covariance (ANCOVA) models examined weight change between baseline and 4 months after adjusting for baseline weight, sex, and age. We handled missing for the ANCOVA analyses in two ways: (1) a completer analysis included participants with complete baseline and 4-month data only, and (2) an analysis using imputation (last observation carried forward [LOCF]) for participants missing 4-month data. All analyses were conducted using SPSS statistical software, version 12.0 (SPSS Inc, Chicago, IL, USA). All reported P values are for two-sided tests, with effects considered statistically significant at P < .05.