A Cluster Randomized Controlled Trial of Automated Internet Weight Loss Programs in Primary Care: Role of Automated Provider Feedback

Abstract

Objective

Despite the high prevalence of obesity and associated health risks in the US adult population, few primary care providers (PCPs) have time and training to provide weight management counseling to their patients. This study aims to compare referral to a comprehensive automated digital weight loss program, with or without provider email feedback, to usual care, on weight loss in patients with overweight or obesity.

Methods

Five hundred fifty adults (mean [SD] 51.4 [11.2] years, BMI 35.1 [5.5] kg/m², 72.0% female) were enrolled through their PCP (n=31). Providers were randomly assigned to refer their patients to a 12-month Internet weight loss intervention only (IWL), the intervention plus semi-automated feedback from the provider (IWL+PCP), or to usual care (EUC). Weight was measured at baseline, 3, 6, and 12 months.

Results

Weight changes (mean [SE]) at 12 months were −0.92 [0.46], −3.68 [0.46], and −3.58 [0.48] kg in the EUC, IWL, and IWL+PCP groups, respectively. Outcomes were significantly different in EUC vs. IWL and EUC vs. IWL+PCP (p<.001) but not in IWL vs. IWL+PCP.

Conclusions

Referral by a PCP to an automated weight loss program holds promise for patients with obesity. Future research should explore ways to further promote accountability and adherence.

INTRODUCTION

Though over 70% of US adults have overweight or obesity [1], weight counseling during visits with primary care providers (PCPs) is uncommon [2,3] and significantly declined from 2008 to 2013, despite an increase in national policies to improve obesity management [3]. Key barriers reported by PCPs include lack of training in obesity counseling [4] and time. Hence, there are no consistently effective and easily implemented interventions for primary care settings, making weight management referral options a target of opportunity for improving population health.

Though effective, in-person and telephone-based weight control programs have been difficult to disseminate in primary care [5]. Online weight control programs are increasingly effective [6] and may lend themselves to primary care settings because they overcome common implementation barriers and can expand the scale and reach of weight management programs, yet maintain consistent program quality [7–9].

While there are a number of effective online programs for obesity, more research is needed to determine 1) if automated online programs can be effective when integrated into primary care, and 2) whether outcomes and engagement are enhanced by provider involvement via email. Counseling and involvement by PCPs have been associated with weight loss [10] and research suggests patients would like providers to be more engaged in their weight management efforts [11]. Of the ways physicians can provide “supportive accountability” to their patients, “arranging follow-up” is an action uniquely suited to primary care settings and potentially effective [12].

The present study compared two online weight control programs versus a usual care control condition in primary care. Both active conditions included access to an automated online program, with features tested in other settings [13–15]. One condition included a novel method of primary care integration that efficiently expanded the role of the PCP by including a web feature to allow PCPs to follow the progress of their patients’ efforts. Both the website and PCP interface were informed by formative research with PCPs and patients [16].

The primary aim of this study was to test the efficacy of an Internet weight loss program (IWL), LoseNowPA, with referral from primary care, that provided automated feedback to patients compared to an enhanced usual care condition (EUC) on weight loss at 1 year, and further to test whether minimal PCP involvement (IWL + PCP) enhanced weight loss outcomes compared to the IWL program alone at 1 year. We hypothesized that a) patients in IWL would lose significantly more weight at 12 months than patients in EUC and, b) subjects in IWL+PCP would lose significantly more weight at 12 months than patients in EUC and IWL.

METHODS

Trial Design

The study was a 3-arm cluster randomized controlled trial with 31 PCPs and 550 patients conducted at two sites: a clinical site at Penn State College of Medicine (Hershey, PA) and intervention site at the University of North Carolina at Chapel Hill (Chapel Hill, NC). The study was approved by the Institutional Review Boards at both sites (PA IRB# 39237, NC IRB# 12–1661) and funded by the National Institute of Diabetes and Digestive and Kidney Diseases (R01-DK095078, Sciamanna and Tate, MPIs). A data and safety monitoring committee provided trial oversight.

Recruitment and Eligibility

Census data were used to send letters to PCPs within 100 miles of Penn State College of Medicine. Recruitment efforts first targeted cities with high minority rates. PCPs expressed interest via phone or web, and were screened by phone. Providers were included if they practiced at least 2 half days per week and agreed to avoid use of weight loss medications in participating patients. Providers were excluded if they served a specialty care population, completed a weight management fellowship, were unable to speak and read English, were pregnant or planning to become pregnant in the next 3 months, or planned to leave their practice location in next 12 months.

Patients were recruited mainly by letters mailed from their PCPs. Patients were between 21–70 years of age, with a body mass index (BMI) of 25–50 kg/m², and had Internet access at home or work. The BMI range was more inclusive than other primary care weight loss studies [17,18]. Other eligibility criteria required participants to have plans to see the PCP within the year, not lost 5% of their body weight in the past year, and no plans to move. Patients were excluded if increasing physical activity appeared unsafe, based on the Physical Activity Readiness Questionnaire (PAR-Q) [19], or if using insulin.

Interventions

Table 1 outlines the key features of each study arm. All PCPs participated in a single, 60 minute, orientation session with a physician investigator (CS), in which they completed informed consent, reviewed patient materials and goal-setting worksheet. The worksheet recommended a 10% weight loss as this amount leads to health improvements but allowed patients to self-select a goal [20]. It included a checklist of self-reported reasons for wanting to lose weight and discussion topics for a future clinical visit, based on the 5 A’s (Assess, Advise, Agree, Assist, Arrange) [12,21]. In the initial study protocol, the goal setting worksheet was planned as a collaborative activity between PCP and patient at a routine primary care visit during which the patient would be recruited. However, feasibility of recruiting the required number of patients at each practice within the recruitment period necessitated a protocol revision whereby recruitment occurred through direct mail. Instead, the worksheet was completed with study staff and copies were mailed to the patient and faxed to the PCP within 48 hours of the patient’s next non-acute visit.

Table 1:

Intervention Characteristics

	EUC	IWL	IWL + PCP
Single goal-setting phone call with research staff	X	X	X
Given Aim for a Healthy Weight booklet (NHLBI)	X	X	X
Comprehensive internet weight loss website:  • Self-monitoring diary for recording caloric intake with the option of using an external app/website (e.g., LoseIT) and enter summary information from the app into the study website each week  • Daily weight prompt encouraging participants to enter their weight as an initial online activity each day  • Computer-tailored feedback, once per week, based on self-reported progress and pre-programmed algorithms  • Instructional lessons: Eighteen “Core” behavioral weight loss lessons were included in an online library and available from the first day of the program.  • “Featured topics” were posted weekly from weeks 1–52 using a “blog style” conversational format and referred to core lesson topics and other standard content.  • Summary plan page and problem-solving tool, which allowed participants to make adjustments to their weight plan, method of self-monitoring, and diet/exercise goals  • Progress page where participants could view graphs of their weight, diet, and exercise, view progress toward their goals, and receive virtual badges for achieving weight loss milestones  • Message boards for connecting with other participants and study staff  • Resource page		X	X
4 automated text messages/week		X	X
Tailored monthly progress email		X	X
Optional Live Webinar (2 offered)		X	X
Optional Review of goal setting worksheet if regularly scheduled well-visit (faxed to practice 48 hours in advance of well-visit)	X	X	X
Review of IWL progress summary report with PCP if regularly scheduled well-visit			X
Biweekly emails from PCP based on weight loss progress and logins to website.			X

The EUC received a copy of the National Heart, Lung and Blood Institute’s Aim for a Healthy Weight booklet and completed the goal setting worksheet. This condition was designed to be an enhancement to usual care, as weight loss goal setting in primary care is not common [12,21].

Participants in the IWL and IWL + PCP conditions were given access to a comprehensive Internet behavioral weight loss program, adapted from the Diabetes Prevention Program, and proven effective in previous studies [13–15]. All participants received a “Getting Started” guide for weight loss, a user guide for the website, and a unique username and password. At baseline, the goal setting worksheet prompted intervention participants to set goals for how frequently they would use the IWL program (e.g., at least twice per week) based on prior work showing improved outcomes with more frequent use [13,22]. The IWL program recommended individualized standard caloric intake goals that promoted 1–2 pounds weight loss per week. Upon first login, participants chose an Eating Plan (e.g., how they would monitor calories) and an Activity Plan (e.g., progression based on current activity level). Participants were encouraged to weigh themselves frequently, at minimum, weekly.

Website features (Table 1) included a self-monitoring diary, computer-tailored feedback, instructional lessons, summary plan page and problem-solving tool, progress page, message board, and resource page. Each week, four automated text messages were sent to those IWL and IWL + PCP participants who opted into text messaging (91% of participants; not different between groups). The 4 messages served to: 1) introduce a new lesson topic, 2) provide general support and encourage motivation, 3) prompt use of a skill taught in the lesson, 4) provide feedback (with data point) on a behavior (e.g, steps, logins). Additionally, an automated tailored progress email was sent to participants monthly and an optional live webinar was offered to each group twice. (Examples in Supporting Information).

PCP Intervention (IWL + PCP Condition): During the orientation session, PCPs in the IWL+PCP condition received brief information about the role of Supportive Accountability [23] and how to promote patient accountability for weight loss through the intervention. They were given access to a Provider Portal website for tracking patient progress and managing patient messages. PCPs were encouraged, but not required, to view patients’ data (weight loss and logins). Every two weeks, PCPs received an email notification that new email messages were generated. The messages were tailored to a) weight loss progress, b) frequency of patient website logins, c) program week, and d) other factors, such as patient-reported motivation. In the portal, PCPs could view and edit the default messages for 6 days, after which messages were sent to the patient with or without provider edits (Sample messages in Supporting Information).

Outcomes

The primary outcome measure was change in body weight at 12 months (kgs) between the IWL and EUC conditions. Secondary outcomes included the weight change between the IWL and IWL+PCP conditions and the percentage of participants in each condition who lost at least 5% of their initial body weight, after 12 months. Two weight and height measurements were taken at the PCP offices by trained staff using calibrated equipment (Tanita BWB-800 scale, SECA 213 stadiometer). Chart review was used to record clinic weight measurements, if taken within 90 days of the 12-month follow-up, for participants with missing 12 month data. BMI was calculated by weight in kilograms, divided by height in meters squared. Blood pressure was measured at each study assessment using a digital monitor (Omron HEM 907XL) after the participant rested for five minutes in a seated position. Physical activity was measured with interview using the Paffenbarger Activity Questionnaire [24]. Dietary intake was assessed with Block Food Frequency, a semi-quantitative food frequency questionnaire [25]. Accountability was measured using one item, “It would bother me if my doctor thought less of me if I don’t control my weight, with a 7-point Likert scale in which higher scores indicated greater perceived accountability. Adherence was operationalized as patient logins to the study website, including total logins and percentage of patients who logged in at least once during each month.

Sample Size

The targeted sample size (N=27 PCPs and N=540 patients) was projected to provide 80% power to detect a 2.0 kg difference in Conditions A v. B, A v. C, and B v. C, consistent with a low to medium effect size of 0.30. Formulas developed by Cohen [26] were used to derive sample size calculations using a Type 1 error (alpha) of 5% (.05), 80% (.80) power, and a two-tailed test. As provider effects may induce positive correlation among patient outcomes, sample size calculations included an Intra-Provider Correlation Coefficient (ICC) of 0.1.

Randomization

The unit of randomization was the PCP. The study statistician (EL) prepared a computerized block randomization sequence with one study condition allocated to up to three providers at each practice site. Using the randomization sequence, the study coordinator prepared a sealed envelope containing the PCP condition assignment. After the PCP orientation and informed consent process, one of the physician investigators (CS) who was blinded to the condition during the study orientation, revealed the assigned condition. Participants were allocated to the study condition of their provider. Providers were chosen as the unit of randomization to balance feasibility (e.g., number of practices required) with risk of contamination. By design we allowed up to 3 physicians per practice (with one allocated to each study condition). It is possible that a patient saw a provider assigned to another condition during an acute-care (e.g., sick) visit. Risk for discussion about weight management during acute-care visits was deemed low because a) the intervention was via the Internet platform and email, b) the goal setting worksheet was only provided at one scheduled well-visit, and c) information on logins and weight loss for IWL+PCP participants was provided to their PCP’s unique login to the physician portal. Due to the nature of the intervention, blinding was not feasible.

Statistical Methods

All variables were summarized prior to analysis, and the distribution of continuous variables was assessed for normality. Baseline characteristics were compared between study groups using Chi-square tests or analysis of variance. Per our prespecified protocol, a linear mixed-effects model was utilized to make comparisons of the change in weight (kg) (primary outcome) using available data at all time points in the intention-to-treat population from baseline to 3, 6, and 12 months. Additional models were also run with adjustments for race and diabetes, which differed between conditions at baseline, however, results of these models were identical to the main models, thus unadjusted results are presented. A generalized estimating equations (GEE) model was also used to analyze the change in the proportion of subjects losing at least 5% weight over the course of the study. In addition to effects for study group, month, and interaction between those variables, models also included the PCP as a random effect. A Bonferroni adjustment applied to the three study group comparisons was used to maintain a family-wise Type I error rate of 0.05. The main effects analysis was first conducted using only the weights recorded by study staff. A sensitivity analysis was then conducted using available clinic weights to replace missing weight outcome data at 12 months. And finally, missing weights at 3, 6 and 12 months were imputed using multiple imputation models (n=100) that included baseline weight, gender, age, education, employment, marital status, hypertension status, medication usage, and diabetes status. The analysis for the primary outcome model was repeated using the imputed datasets.

Secondary analyses using the same linear mixed-effects model were carried out for systolic and diastolic blood pressure (mmHg), physical activity (cal/wk), food energy (kcal), and accountability score (1–7, higher=more accountability). Website utilization (logins) over the course of the study was compared between IWL and IWL+PCP study groups, by using approaches including Wilcoxon Rank Sum tests to compare the median number of logins between these two study groups during 3-month intervals (1–3, 4–6, 7–12), given that the number of logins was not normally distributed, and Chi-square tests to compare the percentage of participants with at least one login between these two study groups during each individual month. Analyses were performed using SAS software version 9.4 (SAS Institute, Cary, NC).

RESULTS

Study Participants

Letters were mailed to 1897 PCPs, of which 78 (4.1%) responded (for details, see Figure 1). Interested providers were screened by phone for eligibility and enrolled from May 2013 to October 2015. Thirty-one PCPs were randomized, and their patients were recruited. The PCP sample was 52% female: 90% self-identified as white, with average age 44.5 years. Most providers were MDs (71%) practicing in private solo/group practice (68%) (Table 2).

Figure 1:

CONSORT Diagram for PCP Participants

Abbreviations: PCP=Primary Care Provider, BMI= body mass index, EUC=Enhanced Usual Care, IWL=Internet Weight Loss program, IWL+PCP= Internet Weight Loss program plus feedback from primary care providers

Table 2:

Baseline Characteristics of LoseNowPA Primary Care Providers by Intervention Assignment

Characteristic	Total (N=31)	Group A (N=11)	Group B (N=9)	Group C (N=11)
	N (%)	N (%)	N (%)	N (%)
Age (years)	44.5 ± 10.2	41.6 ± 10.8	46.4 ± 8.7	45.8 ± 10.8
Gender  Male  Female	15 (48.4) 16 (51.6)	5 (45.5) 6 (54.5)	4 (44.4) 5 (55.6)	6 (54.5) 5 (45.5)
Ethnicity  Hispanic/Latino  Not Hispanic/Latino	0 (0.0) 31 (100.0)	0 (0.0) 11 (100.0)	0 (0.0) 9 (100.0)	0 (0.0) 11 (100.0)
Race  White  Non-White	28 (90.3) 3 (9.7)	11 (100.0) 0 (0.0)	8 (88.9) 1 (11.1)	9 (81.8) 2 (18.2)
BMI	23.6 ± 3.0	23.7 ± 3.3	24.0 ± 3.3	23.3 ± 2.6
BMI class  Normal  Overweight/Obese	21 (67.7) 10 (32.3)	6 (54.5) 5 (45.5)	6 (66.7) 3 (33.3)	9 (81.8) 2 (18.2)
Practice Specialty  Family Medicine  Internal Medicine	27 (87.1) 4 (12.9)	10 (90.9) 1 (9.1)	8 (88.9) 1 (11.1)	9 (81.8) 2 (18.2)
Provider Type  Physician  Advanced Practice Clinician	22 (71.0) 9 (29.0)	7 (63.6) 4 (36.4)	7 (77.8) 2 (22.2)	8 (72.7) 3 (27.3)
Practice Setting  Private Solo/Group Practice  Faculty Practice Plan	21 (67.7) 10 (32.3)	8 (72.7) 3 (27.3)	5 (55.6) 4 (44.4)	8 (72.7) 3 (27.3)

Letters were mailed to 12,000+ patients, of which 1,336 (11%) responded (Figure 2). Phone screenings were completed on 978 potential participants with 550 patients enrolled and randomized from August 2013 to November 2015. The patient sample was 72% female; 85% of participants self-identified as white, with average age 51.4 years and BMI 35.1 kg/m² (Table 3). Participants randomized to the IWL + PCP condition were more likely to be Black (p<.0001) and somewhat more likely to report a diagnosis of Type 2 diabetes at baseline (p=.08).

Figure 2:

CONSORT Diagram for Patient Participants

Abbreviations: PCP=Primary Care Provider, BMI= body mass index, EUC=Enhanced Usual Care, IWL=Internet Weight Loss program, IWL+PCP= Internet Weight Loss program plus feedback from primary care providers

Table 3:

Characteristics of LoseNowPA Patient Participants at Baseline by Intervention Assignment

Characteristic	Total (N=550)	EUC (N=187)	IWL (N=181)	IWL+PCP (N=182)	P-value*
	N (%)	N (%)	N (%)	N (%)
Age (years)	51.4 ± 11.2	50.4 ± 11.7	51.9 ± 11.0	51.8 ± 10.8	0.364
Gender  Female	396 (72.0)	128 (68.5)	130 (71.8)	138 (75.8)	0.288
Ethnicity1  Hispanic/Latino	24 (4.6)	9 (4.9)	10 (5.9)	5 (2.9)	0.375
Race  White  Black  Other	470 (85.4) 51 (9.3) 29 (5.3)	172 (92.0) 5 (2.7) 10 (5.3)	152 (84.0) 17 (9.4) 12 (6.6)	146 (80.2) 29 (15.9) 7 (3.9)	<0.001
Education (N=527)  High School or less  College 1–3 years  College 4+ years	133 (25.2) 177 (33.6) 217 (41.2)	44 (24.0) 60 (32.8) 79 (43.2)	45 (26.6) 54 (32.0) 70 (41.4)	44 (25.1) 63 (36.0) 68 (38.9)	0.892
Employed (N=524)  Yes	397 (75.8)	137 (74.9)	131 (77.1)	129 (75.4)	0.884
Diabetes (N=547)  Yes	61 (11.2)	16 (8.7)	17 (9.4)	28 (15.4)	0.083
Smoking (N=528)  Yes	44 (8.3)	18 (9.8)	10 (5.9)	16 (9.1)	0.374
Weight (kg)	97.7 ± 18.7	97.0 ± 18.7	99.2 ± 19.6	96.8 ± 17.8	0.404
Body Mass Index (kg/m2)	35.1 ± 5.5	34.7 ± 5.6	35.8 ± 5.7	35.0 ± 5.2	0.146
Body Mass Index (kg/m2)  Overweight (25.0 – 29.9)  Obese Class I (30.0 – 34.9)  Obese Class II (35.0 – 39.9)  Obese Class III (40.0 – 50.0)	105 (19.1) 193 (35.1) 141 (25.6) 111 (20.2)	44 (23.5) 61 (32.6) 48 (25.7) 34 (18.2)	30 (16.6) 63 (34.8) 45 (24.9) 43 (23.7)	31 (17.0) 69 (37.9) 48 (26.4) 34 (18.7)	0.484
Systolic Blood Pressure (mmHg)	125.2 ± 15.4	127.0 ± 15.2	124.0 ± 14.7	124.4 ± 16.0	0.123
Diastolic Blood Pressure (mmHg)	79.4 ± 10.4	79.9 ± 10.6	79.6 ± 9.8	78.8 ± 10.7	0.625
Taking BP Medication (N=549)	159 (29.0)	57 (30.5)	57 (31.7)	45 (24.7)	0.296
Accountability (1–7) (N=518)	4.2 ± 2.0	4.2 ± 2.0	4.2 ± 2.0	4.2 ± 2.0	0.963
Food Energy (kcal) (N=481)	1733.2 ± 816.2	1725.8 ± 709.8	1667.3 ± 764.8	1803.7 ± 957.7	0.337
Food Energy from fat (%kcal) (N=481)	38.8 ± 6.6	40.0 ± 6.5	38.1 ± 6.6	38.2 ± 6.7	0.010
Physical activity (cal/wk) (N=537)	761.5 ± 760.1	762.6 ± 801.2	808.7 ± 727.9	712.6 ± 749.8	0.491

^*ANOVA or Chi-square test, exact tests used as needed, N (%) or mean ± SD

Follow-up weights at 12 months were obtained on 82.9% of the baseline sample, with no differences in follow-up between conditions (83.4% EUC, 86.9% IWL and 79.1% IWL+PCP, p=0.197). Clinic weights were used in the sensitivity analysis for a follow up rate of 87.5% (86.3% EUC, 90.6% IWL, 85.2% IWL+PCP, p=0.269). Predictors of 12-month attrition were examined including age, race, gender, education, income, baseline BMI, diagnosis of diabetes, and BP medication usage. Patients lost to follow-up were heavier at baseline (BMI 37 kg/m²) compared to those who returned for 12-month follow-up (BMI 34 kg/m²; p=.010).

Weight Loss

At 6 months, mean (SE) weight losses in EUC, IWL, and IWL+PCP were −1.55 (.38), −4.49 (.38), and −4.05 (.40) kg and by 12 months the weight losses were −.92 (.46), −3.68 (.46), and −3.58 (.48) kg, respectively (Table 4). On the primary outcome at 12 months, the IWL lost significantly more weight vs. EUC (Bonferroni adjusted p-value <.001). The IWL+PCP also showed greater weight loss at 12 months vs. EUC (Bonferroni adjusted p-value <.001); the difference between IWL and IWL+PCP was not significant (Bonferroni adjusted p-value= 1.0). Results of two different sensitivity analyses on the primary outcome at 12 months, 1) using clinic weights available from the medical chart and 2) using multiple imputation analyses, showed an identical pattern of results with the IWL and IWL+PCP having significantly greater weight losses compared to EUC; IWL and IWL+PCP did not differ from each other (Table 4).

Table 4:

Main Outcomes: Mean Weight Loss in the Intention-to-Treat Population ^*

Variable	EUC (N=187)	IWL (N=181)	IWL + PCP (N=182)	P-value (adjusted for 3 comparisons)
				IWL vs. EUC	IWL + PCP vs. EUC	IWL + PCP vs. IWL
Model 1: Main Effects Model using Study Weights, only (unadjusted) **
Change in weight (kg)
At month 3	−1.43±0.27	−3.41±0.27	−2.95±0.29	<0.001	<0.001	0.705
At month 6	−1.55±0.38	−4.49±0.38	−4.05±0.40	<0.001	<0.001	1.0
At month 12	−0.92±0.46	−3.68±0.46	−3.58±0.48	<0.001	<0.001	1.0
Model 2: Main Effects Model using Study and Clinic Weights (unadjusted) **
Change in weight (kg)
At month 3	−1.43±0.27	−3.37±0.27	−2.88±0.28	<0.001	<0.001	0.642
At month 6	−1.54±0.38	−4.32±0.38	−3.91±0.39	<0.001	<0.001	1.0
At month 12	−0.84±0.45	−3.35±0.45	−3.35±0.46	<0.001	<0.001	1.0
Model 3: Main Effects Model using Multiple Imputation ***
Change in weight (kg)
At month 3	−1.65±0.35	−3.25±0.33	−2.89±0.36	<0.001	<0.001	0.464
At month 6	−1.86±0.46	−4.46±0.48	−4.12±0.48	<0.001	<0.001	0.606
At month 12	−1.02±0.54	−3.25±0.55	−3.19±0.55	0.004	0.005	0.939

^*Plus–minus values are means ± standard error (SE).

^**Unadjusted model

^***Adjusted for race and diabetes; Imputation models (n=100) included weight at baseline, month, gender, age, education, employment, marital status, hypertension medication, diabetic status. 100 iterations.

Secondary Outcomes

A greater proportion of participants randomized to the two interventions (IWL and IWL+PCP) achieved 5% or more weight loss (12 months; 17% of EUC, 38% of IWL, and 34% of IWL+PCP) (Bonferroni-adjusted p=.003 (IWL vs. EUC), and p=.017 (IWL+PCP vs. EUC).

There were no between group differences on any other secondary outcomes including systolic and diastolic BP, self-reported calorie intake, self-reported calories expended through physical activity, and perceived accountability (Table 5).

Table 5.

Secondary Outcome changes between groups in the Intention-to-Treat Population^*

Variable	EUC (N=184)	IWL (N=181)	IWL + PCP (N=182)	P-value (adjusted for 3 comparisons)
				Web vs. Usual	Web+ vs. Usual	Web+ vs. Web
Change in weight (%)
At month 3	−1.39±0.47	−3.34±0.48	−3.02±0.50	0.012	0.054	1.0
At month 6	−1.59±0.47	−4.45±0.48	−4.24±0.50	<0.001	<0.001	1.0
At month 12	−0.87±0.47	−3.63±0.48	−3.64±0.49	<0.001	<0.001	1.0
Change in BMI
At month 3	−0.50±0.10	−1.20±0.10	−1.08±0.10	<0.001	<0.001	1.0
At month 6	−0.55±0.13	−1.59±0.14	−1.49±0.14	<0.001	<0.001	1.0
At month 12	−0.33±0.16	−1.31±0.16	−1.31±0.17	<0.001	<0.001	1.0
Lost at least 5%
At month 3	10.97%	31.41%	31.62%	<0.001	<0.001	1.0
At month 6	15.23%	42.75%	37.50%	<0.001	0.001	1.0
At month 12	17.31%	37.82%	34.03%	0.003	0.017	1.0
Change in systolic BP (mm)
At month 3	−0.90±0.97	−1.36±0.98	0.01±1.03	1.0	1.0	1.0
At month 6	−1.23±1.23	−0.35±1.26	1.64±1.31	1.0	0.330	0.821
At month 12	−0.65±1.37	−1.47±1.38	0.61±1.42	1.0	1.0	0.883
Change in diastolic BP (mm)
At month 3	−0.24±0.64	−1.26±0.64	−1.31±0.68	0.780	0.750	1.0
At month 6	−0.95±0.82	−1.51±0.84	−0.51±0.86	1.0	1.0	1.0
At month 12	−0.64±0.91	−1.58±0.92	−0.73±0.95	1.0	1.0	1.0
Change in physical activity (kcal/wk)1
At month 3	119.10±64.31	256.28±64.31	304.27±68.10	0.395	0.145	1.0
At month 6	103.68±78.24	278.97±80.96	146.82±82.82	0.359	1.0	0.762
At month 12	180.42±84.19	174.34±85.13	272.17±86.86	1.0	1.0	1.000
Change in dietary intake (kcal)
At month 3	n/a	n/a	n/a	n/a	n/a	n/a
At month 6	−280.2±49.86	−328.4±53.92	−430.7±56.56	1.0	0.139	0.573
At month 12	−226.0±64.52	−277.7±68.90	−345.4±71.77	1.0	0.650	1.0
Change in accountability2
At month 3	n/a	n/a	n/a	n/a	n/a	n/a
At month 6	n/a	n/a	n/a	n/a	n/a	n/a
At month 12	−0.56±0.20	−0.23±0.21	−0.52±0.22	0.776	1.0	1.0

^*Plus–minus values are means ± standard error (SE).

¹Change is physical activity was measured with the Paffenbarger Activity Questionnaire.

²For the measure of accountability, higher scores indicate greater perceived accountability.

Engagement

Provider Use of the Portal: Providers logged into the portal on 51.4% of weeks (range 3.8%−82.8%). Logging into the portal allowed review of patient progress and messages at a glance. Editing a message required additional steps. The average number of messages sent per provider was 446 and varied based on the number of patients recruited in their panel (range 367–520). Of all provider messages sent, 1.2 % were edited or personalized (range 0%−5.5%).

Participant-Provider IWL Discussion: While patient progress summaries were only provided to IWL+PCPs in advance of the non-acute visit, patient-reported discussions of the program with the provider at a non-acute visit were similar between IWL and IWL+PCP groups. In IWL, 31% of those randomized reported a discussion of the program at a non-acute visit with the provider (53% of participants (n=96) reported at least one scheduled, non-acute care visit, 59% (n=57) of those said that their PCP discussed the program with them). In the IWL+PCP group, 33% of those randomized reported a discussion of the program at a non-acute visit (49% of participants (n=89) reported at least one scheduled, non-acute care visit, 67% (n=60) of those said that their PCP discussed the program with them).

Participant Use of IWL: The majority (89–90%) of participants in IWL and IWL+PCP groups logged into the website at least once during the first month. Participation rates did not differ between the two interventions on measures of average usage (median) nor the proportion of users logging in at a minimum frequency each month (1x per month) over time (Table 6). Participation was highest during the first 3 months of the program with mean logins of 55 and 43 (IWL and IWL+PCP, respectively), translating to approximately 3 and 5 logins/week. In contrast, during the last 6 months, mean logins were approximately 1 per week. Data on percent of participants still logging in at least once per month show at least 50% of intervention participants were participating through month 7; however, by month 12 about 40% were still logging in to the website.

Table 6.

Website Utilization between groups in the Intention-to-Treat Population

Variable	IWL (N=181)	IWL + PCP (N=182)	P-value
Login total (median)
Month 1–3	19.0 (6.0, 71.0)	21.0 (5.0, 50.0)	0.408
Month 3–6	5.0 (0.0, 28.0)	7.0 (0.0, 23.0)	0.943
Month 6–12	2.0 (0.0, 25.0)	4.0 (0.0, 26.0)	0.767
Login total (mean±SD)
Month 1–3	54.9±81.9	43.1±65.5
Month 3–6	28.4±61.0	24.4±48.1
Month 6–12	33.6±91.1	27.5±56.7
Users who logged in at least once in the month (%)
Month 1	89.0%	90.1%	0.732
Month 2	77.4%	77.4%	1.0
Month 3	73.5%	68.5	0.297
Month 4	59.7%	60.2%	0.915
Month 5	57.5%	60.2%	0.593
Month 6	56.4%	57.5%	0.832
Month 7	51.4%	50.8%	0.916
Month 8	45.3%	48.6%	0.527
Month 9	45.9%	50.8%	0.344
Month 10	41.4%	43.7%	0.671
Month 11	37.0%	40.3%	0.517
Month 12	39.2%	40.9%	0.748

DISCUSSION

The primary aim of this study was to determine the effect of offering an IWL program with tailored automated weekly feedback in primary care settings compared with enhanced usual care. A secondary aim was to determine whether minimal PCP involvement through a patient progress portal with bi-weekly, semi-automated email contacts from the PCP would increase weight loss over the IWL alone. Results showed that both the IWL and IWL+PCP conditions produced significantly greater weight loss compared with EUC over 12 months but that the addition of semi-automated provider emails did not enhance these effects.

Since this study utilized an automated Internet approach, these findings are quite important as PCPs lack the training and time to meaningfully engage in weight loss efforts with their patients. Padwal and colleagues observed that a clinic-based, fully automated intervention did not lead to significant weight losses, though the intervention was not interactive [27]. One of the longer and larger studies (n=840) to date, by Baer and colleagues, evaluated an online program compared to usual care and to an online program combined with population health management supported by clinical support staff [7]. The website used in their intervention arms provided lessons, videos, meal plans, and tracking tools. At 12 months, there were significant weight loss differences by group (–1.2 kg usual care, –1.9 kg online only, –3.1 kg in combined). Other studies have had larger weight losses. Thomas et al, conducted a study with 154 patients recruited from primary care with weight losses of −5.4 kgs at 6 months. Their IWL used weekly videos and automated feedback [9].

Our study’s results can also be examined in context of reviews of primary care weight loss programs without technology in the US [28] and UK [29], with losses ranging from 0.1 kg to 7.7 kg across trials using in person or phone-based care. In the present study, without any ongoing face-to-face, telephone or consistent weight management counseling, weight losses were 3.5–4.5% on average at 6 or 12 months, and the proportion of the population who achieved a clinically meaningful weight loss over 1 year was doubled compared to enhanced usual care. While weight losses were of a magnitude to potentially impact other clinical markers, we did not observe significant differences on secondary clinical and behavioral outcomes. Other primary care weight management studies have also demonstrated clinically meaningful weight losses without significant changes or differences on outcomes like blood pressure [17,18,30]. It is also not uncommon to observe between-group differences on weight but not on self-reported behavioral measures [31–33].

The IWL+PCP condition was meant to enhance aspects of the 5A’s [34]and encourage patient accountability to their PCP for weight loss using a two-tiered, sustainable approach in which patients could manage their weight using the website and providers could track patient progress via an online portal that would automatically send biweekly, tailored emails to promote this supportive accountability. We hypothesized that PCP involvement would increase weight losses over IWL alone, mainly through greater accountability to the provider and website utilization, though that did not occur. In fact, weight losses were nearly identical among participants in IWL and IWL+PCP.

This study was not designed to understand the mechanism of this effect, but it could be due to a lower level of engagement with the messages by providers. Participants received 24 messages from their providers and, since few were modified, it is possible that patients realized that they were not written by the provider. We offered providers a templated email, pre-populated with patient data to encourage continued study participation. The message portal—including the feature that allowed physicians to view and edit messages—was designed based on physician feedback in formative focus groups [16]. We allowed providers to choose statements for introductions and endings of messages that reflected their “voice,” however, the messages were pre-written and focused on a limited set of variables, thus, may have become repetitive to patients. Provider review of system generated messages in the portal was highly variable (Range: 4–82% of weeks; average 51% of weeks). Given that providers were following 11–24 patients, with bi-weekly messages over 1 year, the average total number of messages available for review was 446. Despite logging in to review messages about half the time, providers edited only 1.2% of messages. This was perhaps not surprising given heavy patient load and increasing demands on most PCPs [35]. However, it is unclear if the lack of provider message edits suggests that the messages were to the providers’ liking, if they were too busy to edit messages, not interested in weight management, or something else.

It is also possible that the effect of this condition was limited by not including a baseline visit between patients and providers. Our study conceptualization included an initial visit to allow collaborative goal setting with providers offering study participation. However, study feasibility required a shift to expedite recruitment resulting in an approach similar to nurse case management interventions [36], where the communications are delivered by a third party. About half of the sample reported a non-acute visit during the study period and a discussion of the program occurred in about 60% of these visits, thus only a third of randomized patients had a conversation with their doctor about the IWL at some point during the year and this did not differ between IWL and IWL+PCP. It is possible that if providers initially met with patients to establish goals, then modified the emails to reflect their priorities, the effect might have been greater. However, it is unclear if this is realistic to expect of providers or if it is necessary for a population level resource given the comparable weight losses in the IWL condition.

Weight losses and engagement with the IWL are encouraging since there were no weekly counselor contacts or kick-off meetings. Indeed, 90% of patients referred to the Internet program logged in initially. During months 1–3 they averaged 3–5 logins per week, a frequency that suggests ongoing utilization of self-monitoring logs and other features. Over time, this frequency declined to about once weekly. The proportion of the population logging in at least monthly suggests that most patients were still engaged through month 7; engagement remained at about 40% from months 8–12. This compares to engagement using this same criterion in an internet program for 2-year college students over 2 years [37]. As early as month 3, only 30–40% of students logged in at all, suggesting that primary care may be a setting in which users feel more accountability or motivation to use a digital program.

This study is not without limitations: lack of cost-effectiveness, limited assessment of accountability, and limited diversity. While our sample was over 25% male, 60% without a college degree, included patients with diabetes and hypertension, and a range of BMI categories, the sample lacked racial and ethnic diversity. We prioritized providers with offices in diverse areas, however our resultant sample was predominantly non-Hispanic white and may generalize to those living with overweight and obesity who identify with diverse racial and ethnic groups. It is not possible to separate the effects of the multi-component intervention on patient outcomes. Patient engagement declined over time; a well-documented barrier in digital interventions [6]. Finally, our response rate to direct mail recruitment letters to patients and providers, while consistent with other direct mail recruitment, was low. This may reflect a disinterest in research, management of obesity, digital health programs or something else. Implementation science and dissemination studies that focus on uptake with different enrollment methods and to programs with different offerings are needed before widescale dissemination of the approach is considered.

This study also has some important strengths. It meets the call for more research on the optimal use of technology in primary care and models for PCP involvement in weight loss [6,38]. It showed that an automated weight control program can be efficacious in clinical populations with minimal PCP interaction. The intervention was built upon behavior change techniques associated with improved weight outcomes [39], as well as features PCPs desired for digital programs. Other strengths include the cluster-randomized, 3 arm design, low attrition, and external validity by recruiting participants in a real-world setting.

Interventions that are effective in clinical populations but do not qualify for reimbursement have historically been difficult to disseminate. Having just experienced a time during COVID-19 when most non-acute health care services required remote delivery, there may be greater opportunity to disseminate automated remote interventions.

STUDY IMPORTANCE QUESTIONS.

What is already known about this subject?

• Internet and mobile weight loss resources are widely available and RCTs have established general efficacy of comprehensive digital programs.

• Digital weight loss resources may be particularly helpful to primary care weight loss efforts as doctors can refer patients who may benefit at lower cost than in person programs.

• Little is known about the efficacy of automated internet programs in a primary care setting or the role of semi-automated email follow-up from the provider.

What are the new findings in your manuscript?

• Weight losses achieved with referral from the primary care provider to an automated internet program were significantly better than usual care at 6 and 12 months; however, adding semi-automated provider emails in an effort to increase accountability did not improve outcomes.

How might your results change the direction of research or the focus of clinical practice?

• Evidence-based digital programs that provide ongoing feedback via tailored automated messages may be a lower cost alternative for some primary care patient referrals and deserve greater study.