Healthcare Resource Utilization and Costs Among Overweight and Obese Users of a Digital Weight Loss Intervention Compared to Non-Users: A Retrospective Analysis

Background: The Noom Weight program is a smartphone-based weight management program that utilizes cognitive-behavioral therapy techniques to motivate users to achieve weight loss through a comprehensive lifestyle intervention. Objective: This retrospective database analysis aimed to evaluate the impact of Noom Weight use on healthcare resource utilization (HRU) and healthcare costs among overweight and obese patients. Methods: Electronic health records (EHR) data, claims data, and Noom program data were used to conduct the analysis. The study included 43,047 Noom Weight users and 14,555 non-Noom users aged 18-80 with a body mass index (BMI) ?25 kg/m² and residing in the U.S. The index date was defined as the first day of a 3-month treatment window during which Noom Weight was used at least once per week, on average. Inverse probability treatment weighting (IPTW) was used to balance sociodemographic covariates between the two cohorts. HRU and costs for inpatient visits, outpatient visits, telehealth visits, surgeries, and prescriptions were analyzed. Results: Within 12 months post-index, Noom Weight users had, on average, $20.10 less inpatient costs (95% CI: -$30.08, -$10.12), $124.33 less outpatient costs (95% CI: -$159.76, -$88.89), $313.82 less overall prescription costs (95% CI: -$565.42, -$62.21), and $450.39 less overall healthcare costs (95% CI: -$706.28, -$


Healthcare Resource Utilization and Costs Among Overweight and Obese Users of a Digital Weight Loss Intervention Compared to Non-Users: A Retrospective Analysis
No, I do not wish to publish my submitted manuscript as a preprint. 2) If accepted for publication in a JMIR journal, would you like the PDF to be visible to the public?
Yes, please make my accepted manuscript PDF available to anyone at any time (Recommended).
Yes, but please make my accepted manuscript PDF available only to logged-in users; I understand that the title and abstract will remain v Yes, but only make the title and abstract visible (see Important note, above). I understand that if I later pay to participate in <a href="http

Healthcare Resource Utilization and Costs Among Overweight and Obese Users of a Digital Weight Loss Intervention Compared to Non-Users: A Retrospective Analysis Introduction
Rising rates of obesity globally [1] have led to substantial increases in related healthcare expenditures. From 2000-2018, the age-adjusted rate of obesity in the United States (U.S.) increased from 30.5% to 42.4%, with 76.5% of the adult population classified as either overweight or obese in 2018 [2]. Between 1998 and 2016, obesity-related healthcare spending in the U.S. increased from $111.7-$170.3 billion [3] to $182.0-$288.0 billion (2021 USD) [4][5]. Obesity is associated with an increase in direct annual medical costs ranging from $1,961 [5] to $3,423 [4] per individual (2021 USD). Furthermore, obesity-related comorbidities are among the greatest contributors to total U.S. annual medical expenditures, including $126.0 billion for diabetes, $101.2 billion for ischemic heart disease, $89.5 billion for hypertension, and $29.9 billion for hyperlipidemia (2021 USD) [6].
There is a pressing need for effective strategies to address these rising costs and the growing prevalence of obesity [7]. Standard dietary interventions that maintain an energy deficit typically produce an average maximal weight loss of 4 to 12 kg after 6 months, with smaller sustained losses of 4 to 10 kg after 1 year, and only 3 to 4 kg after 2 years [8]. Lifestyle change is typically required to sustain weight losses, and as such, lifestyle interventions are among the most effective approaches [9][10][11].
In-person interventions, though effective, can be time consuming, expensive, and thus unappealing to many potential participants [12][13]. In addition to these barriers, limited program availability, and potential lack of reimbursement [14][15] further limit widespread participation. Remote interventions using mobile health (mHealth) technologies, such as telephone calls, text messages, and smartphone apps, have been effective in the treatment of obesity [16][17][18][19], and can address many of the limitations associated with in-person treatment [20]. By maintaining regular interaction with healthcare providers, directly delivering educational content, support, and motivation in a widely accessible, convenient, and affordable format, engagement and adherence are improved [17,21]. Though the evidence for clinical effectiveness continues to grow, the literature lacks sufficient data on healthcare cost savings from mHealth programs [22].
Noom (Noom, New York, NY) is a mHealth program that delivers a comprehensive lifestyle intervention through educational articles, coaching, support groups, diet and exercise tracking, and techniques based on principles of cognitive behavioral therapy. Noom has two health programs: Noom Weight for weight management and Noom Mood for stress management. Previous work has shown that 56% of Noom Weight starters achieve weight loss of ≥5% of initial body weight after 6 months [23], a threshold shown to produce clinically meaningful improvements in health by improving lipid profiles and reducing the risks of developing diabetes and hypertension [8,24]. A retrospective analysis of more than 11,000 users who opened the program at least once after week 8 showed that the majority of users lost ≥5% weight loss at 32 weeks (79%) and 52 weeks (82%), with the proportion of users losing 10% body weight or more increasing from 30% to 40% over the same period [25]. The degree of weight loss achieved has also been demonstrated to be strongly associated with user engagement levels [16,[25][26][27][28][29]. Although the clinical benefit of Noom Weight has been established, its economic impact, such as on health care resource utilization (HRU) and associated costs, has not been thoroughly evaluated or reported in the literature to date.
We conducted a retrospective study using real-world data from the Noom Weight user database, electronic health records (EHR), and a commercial claims database to assess the impact of Noom Weight on HRU and healthcare costs for Noom Weight users in the U.S.. Using propensity score analyses, HRU and costs in overweight and obese Noom Weight users were compared to demographically similar non-Noom users at 12 and 24 months of follow-up. We hypothesized that Noom Weight users would demonstrate lower HRU and healthcare costs compared to individuals who did not use Noom Weight. A secondary aim was to explore the potential correlation of observed impacts on HRU and costs with changes in obesity-related clinical outcomes.

Study Design
This retrospective, longitudinal cohort study used a dataset based on new Noom Weight registrants between July 31 st , 2018, and July 31 st , 2020, including self-reported demographic data recorded at the time of registration (e.g., age, sex, height, weight) and activity data (e.g., body weight measurements, food intake, physical activity) recorded longitudinally thereafter. This dataset was linked with a cohort of patients across EVERSANA EHR and open claims data, which included anonymized patient identifiers, vitals, and healthcare provider visits with associated diagnoses and procedures for US patients, including those with commercial insurance, Medicare, and Medicaid coverage. EVERSANA's EHR Dataset is an aggregation and standardization of electronic health record (EHR) data into a common data model for more than 120 million U.S. patients. The data are derived from more than 2,000 outpatient/ambulatory health centers, more than 500 hospitals, more than 30 health systems (including academic medical centers), and more than 50 unique EMR platform providers across all 50 states in the U.S. All database records are statistically de-identified and certified to be fully compliant with US patient confidentiality requirements set forth in the Health Insurance Portability and Accountability Act of 1996. Because this study used only de-identified patient records and did not involve the collection, use, or transmittal of individually identifiable data, Institutional Review Board approval to conduct this study was not necessary. HRU, healthcare costs, and obesityrelated clinical outcomes were compared between Noom Weight users and demographically similar non-Noom Weight users with overweight or obesity. Figure 1 provides a detailed summary of the study design.

Cohorts Noom Weight Users
Noom Weight users were required to have an initial treatment window of continuous Noom use lasting at least three months. Continuous use was defined as opening the Noom program at least once per week, on average. Each user's unique index date was defined as the first day of Noom Weight use in the treatment window. If users recorded multiple eligible 3-month treatment windows, the earliest eligible window was used. Users were required to be U.S. residents aged 18-80 and have a baseline body mass index (BMI) of ≥25 kg/m 2 . A minimum of 12 months of medical records pre-and post-index, and a minimum documented claims activity of at least one claim in the 12-month preindex period and at least one claim in the 12-month post-index period were required. Additionally, users included in the 24-month post-index analysis were required to have a second claim in the 12-to 24-month window.
Users were excluded if they had a history of medical conditions that would significantly affect body weight or the ability to fully engage in a comprehensive lifestyle intervention during the study period, including acquired immunodeficiency syndrome (AIDS), cancer (all types), end-stage organ failure, hemiplegia, paraplegia, uncontrolled human immunodeficiency virus (HIV), pregnancy, or wasting syndrome. Patients were also excluded if they had surgeries or acute-onset conditions affecting body weight, including bariatric surgery and cerebrovascular disease, prior to the end of the initial 3-month treatment window. Comorbidities were identified in EHR using International Classification of Diseases with Clinical Modification 9 th (ICD-9-CM) and 10 th (ICD-10-CM) revision codes.

Non-Noom Weight Users
A control cohort of non-Noom Weight users otherwise meeting the inclusion/exclusion criteria defined above for Noom Weight users was also identified using EHR and claims data. Index dates for non-Noom Weight users were defined as the date of the first qualifying BMI entry (≥25 kg/m 2 ) recorded in EHR between July 31 st , 2018, and July 31 st , 2020.

Baseline Covariates
Baseline covariates including BMI, sex, age, and U.S. census region were derived from Noom data for Noom Weight users, and from EHR for non-Noom Weight users. Type of insurance coverage was derived from claims data for both cohorts. Covariates were balanced between cohorts using inverse probability of treatment weighting (IPTW) prior to analyses.

Healthcare Resource Utilization
HRU was determined from all submitted insurance claims for any service. Claims were categorized based on the recorded place of service and type of claim, including inpatient visits, length of inpatient stay (in days), outpatient visits (including numbers of clinic, office, and outpatient hospital visits), telehealth visits, other/unknown visits, surgeries, total prescriptions, and obesity-specific prescriptions. Unique visits were counted as single events regardless of the extent of services rendered during the visit; total prescriptions included the total count of all prescribed medications. For each service type, the number of usages per patient, as well as the number of usages per patient among only those patients with ≥1 usage, were determined at 12-and 24-months post-index.

Healthcare Costs
Healthcare costs were determined based on remitted insurance claims and included all unique entries with valid Current Procedural Terminology (CPT) codes, Healthcare Common Procedure Coding Systems (HCPCS) codes, or National Drug Codes (NDC). In cases where remitted amounts were not available, costs were imputed using the median remitted amount for similarly coded claims, aggregated on the claimant's insurance type, age group, sex, and state of residence. Prescription costs included only paid claims; submitted claims that were not reimbursed were excluded. Obesityspecific prescription costs included all medications approved for short-term or chronic weight management, or those commonly prescribed off-label. Costs per patient were calculated at 12 and 24 months for each service type among all patients, as well as among only those patients with ≥1 use of each service type. All costs were reported in U.S. dollars (USD) and adjusted for inflation to 2021 dollars using the medical consumer price index (CPI) inflation factors from the Federal Reserve Economic Data [32].

Statistical Analysis
Propensity score matching was conducted with inverse probability of treatment weighting to balance the Noom Weight and non-Noom cohorts with respect to age, sex, geographic region, insurance plan, and BMI. Stabilized weights for reweighting were generated with the average treatment effect as the estimand. Summary statistics were expressed as mean and standard deviation (SD) for continuous variables and frequency and percentages for categorical variables. Standardized mean differences (SMDs) were used to confirm covariate balance, with absolute SMDs <0.10 indicating potential balance. Mean differences between cohorts at 12 and 24 months were reported for HRU and costs. Generalized linear models were used to report incidence rate ratios for each HRU service (using a Poisson distribution with a log link) and cost ratios for the overall costs (using a gamma distribution with a log link). All analyses were conducted using R 3.6.1.

Subgroup Analysis
Subgroup analyses were conducted by stratifying cohorts according to diagnosis of type 2 diabetes (T2D) (yes vs. no), diagnosis of hypertension (yes vs. no), index BMI (≥35 vs. <35), Noom Weight use duration (≥6 months vs. <6 months), and Noom Weight engagement level (high vs. low). Engagement was classified as "high" if Noom Weight program was opened ≥6 days per week on average and classified as "low" if opened <6 days per week during the initial 3-month treatment period.

Patient Demographics
A total of 114,691 Noom Weight users were represented in all three linked data sources, and 78,375 of these had valid index dates. After exclusions for comorbidities and inclusion criteria for index BMI, index age, Noom Weight use, and claims activity were applied, 43,047 Noom Weight users were included for the 12 month analyses and 14,141 for the 24 month analyses. A total of 107,519 non-Noom Weight users were identified in both EHR and claims data, with 95,005 having valid index dates. All inclusion/exclusion criteria were met by non-Noom Weight users for the 12-month (N=14,587) and 24-month (N=6,487) analyses.
Baseline demographics are shown in Table 1 before and after IPTW. Before IPTW, unweighted mean (SD) ages at baseline were 51.6 (12.0) years for Noom Weight users and 52.7 (14.3) years for non-Noom Weight users (SMD: -0.077), and 82.8% of Noom Weight users and 54.7% of non-Noom Weight users were female (SMD: -0.635). After IPTW, mean ages were equivalent between cohorts (51.9 years, SMD: 0.001) and proportions of females were identical at 75.6% for both Noom Weight users and non-Noom Weight users (SMD: 0.000). All other covariates were also well balanced following IPTW, with the proportion of balanced covariates (absolute SMDs <0.10) increasing from 23% to 100%. Relevant comorbid conditions before weighting are presented in Table 1.

Healthcare Resource Utilization
Noom Weight users had statistically significantly lower HRU compared to non-Noom Weight users in the majority of places of services in both the 12-month (Table 2) and 24-month (Table 3)  , although significant differences did not persist at 24 months post-index. Additional analyses limited to patients with at least 1 encounter of each service type showed lower outpatient service use at 12 months post-index, and fewer prescriptions at 12-and 24-months post-index for Noom Weight users compared to non-Noom Weight users (Table 2). In subgroups without T2D or hypertension and with BMI <35, Noom Weight users had lower usage of more service types compared to non-Noom Weight users than did subgroups with T2D or hypertension and with BMI ≥35, respectively (Supplemental Tables 3a-3c). For example, while fewer outpatient visits were recorded among Noom Weight users compared to non-Noom Weight users in both subgroups with and without T2D at 12 months post-index, significant differences (reductions) were also observed for Noom Weight users compared to non-Noom Weight users in inpatient visits, inpatient days, surgeries, prescriptions, and obesity-specific prescriptions only in the subgroup without T2D (Table 3a). Similarly, relatively fewer significant differences between Noom Weight and non-Noom Weight users were observed for the subgroup with hypertension (Table 3b) and with BMI ≥35 (Table 3c) in the respective subgroup analyses. The differences between subgroups were more pronounced at 24 months post-index.
More than three-quarters of Noom Weight users were categorized as high-engaged (76.1%), with the remaining 23.9% categorized as low-engaged. High-engaged Noom Weight users had significantly fewer prescriptions (overall and obesity-specific) than low-engaged users at 12 months (overall MD:  (Table 3d). These differences remained significant at 24 months post-index for high-engaged Noom Weight users; for low-engaged Noom Weight users, only the differences in inpatient visits and outpatient visits remained significant at 24 months post-index, and increased obesity-specific prescriptions in lowengaged Noom Weight users compared to non-Noom Weight users were also noted at 12 and 24 months post-index.  (Table 3e). The pattern of significant differences for Noom Weight users of both durations was similar to that for Noom Weight engagement level at 12 months post-index, with fewer inpatient visits, inpatient days, outpatient visits, surgeries, and prescriptions than for non-Noom Weight users. This pattern of significant differences was unchanged at 24 months post-index for Noom Weight users with ≥6 months of use; for Noom Weight users with <6 months of use, only inpatient visits, inpatient days, a subset of outpatient visits (outpatient-hospital visits only), and prescriptions were significantly lower than those for non-Noom Weight users.  (Table 5). Expenditures for inpatient services, outpatient services and overall prescriptions were lower for Noom Weight users compared to non-Noom Weight users at 12 months, while telehealth expenditures were slightly higher. Of these, the reduction in outpatient expenditures, overall prescriptions, and overall costs remained statistically significant through 24 months. The additional analysis limited to patients with at least 1 encounter of each service type (Table 4) showed significantly lower overall and obesity-specific prescription costs at both time points, as well as significantly lower outpatient costs at 12 months for Noom Weight users compared to non-Noom Weight users.  Results of the subgroup analyses for costs showed trends similar to those for HRU. Overall, significantly lower costs were seen for Noom Weight users compared to non-Noom Weight users in more service types among cases without T2D (vs. cases with T2D), without hypertension (vs. cases without hypertension), and with BMI <35 (vs. cases with BMI ≥35) (Supplemental Tables 5a-5c). Despite lower HRU among high-engaged versus low-engaged Noom Weight users and Noom Weight users with longer versus shorter duration of use, no significant corresponding differences in costs were observed between these groups (Supplemental Tables 5d, 5e). Compared to non-Noom Weight users, highly-engaged Noom Weight users had significantly lower costs at 12 months post-index for inpatient visits, outpatient visits, and overall costs, and significantly lower prescription costs and overall costs at 24 months post-index. Low-engaged Noom Weight users had fewer differences in costs compared to non-Noom Weight users, with significantly lower costs for inpatient visits and outpatient, and overall costs visits at 12 months post-index, and no significant differences at 24 months post-index. The pattern of significant cost differences for Noom use ≥6 months and Noom use <6 months compared to non-Noom Weight users was similar to those for high-engaged and lowengaged Noom Weight users, respectively, at 12-and 24-months post-index.

Principal Results
We showed that healthcare resource utilization (HRU) is lower for Noom Weight users compared with non-Noom Weight users at 12-and 24-months post-index. Per user, 0.03 fewer inpatient visits, 0.83 fewer outpatient visits, 0.01 fewer surgeries, and 1.39 fewer prescriptions were recorded among Noom Weight users compared to non-Noom Weight users at 12 months post-index. At 24 months post-index, 0.04 fewer inpatient visits, 0.58 fewer outpatient visits, 0.01 fewer surgeries, and 3.13 fewer prescriptions were recorded among Noom Weight users compared to non-Noom Weight users. Noom Weight users had higher usage of telehealth services at 12 months post-index (0.02 per user), perhaps because of increased connectivity to digital health services due to their use of Noom Weight or because of increased health responsibility as a result of the program [30]. There was also a greater number of obesity-specific prescriptions for Noom Weight users compared to non-Noom Weight users at 12 months (0.08 per user), which may be related to more health-conscious behavior [30] among newly registered Noom Weight users, potentially leading to higher rates of prescriptions. A statistically significant difference did not persist at 24 months.
Results also showed significantly lower healthcare costs for Noom Weight users compared to non-Noom Weight users at both 12 months and 24 months post-index. Overall costs for Noom Weight users were $450 lower per person at 12 months and $1,219 lower per person at 24 months compared to individuals who did not use Noom Weight. Further, extending similar findings at 12 months, outpatient services costs ($80 per person) and prescription costs ($1,139 per person) were lower for Noom Weight users than non-Noom Weight users at 24 months post-index.
Overall, our findings demonstrate significantly lower HRU and costs at 12 and 24 months for Noom Weight users compared to demographically similar non-Noom Weight users, with greater impact on HRU and costs observed for Noom Weight users without T2D, without hypertension, with BMI <35, with higher Noom engagement, and with longer duration of Noom use.

Limitations
This was an observational study, which therefore does not permit causal associations to be drawn between Noom Weight use and HRU and cost outcomes. Another important limitation was the restricted sample size due to linking of three separate databases. Users were required to be present in all three data sources for inclusion, which sharply reduced the size of the available population. This also adds a risk of bias, as underlying, systematic exclusions due to missing data that may have affected patients in any one database would have been projected across all three databases, including those not previously affected by them. The requirement for Noom Weight users to use the program for 3 months may have biased this cohort to include more health-conscious and motivated users, though it should be noted that this engagement criteria is similar to previously studied Noom Weight populations and this study's inclusion requirement to use the program at least ten times total during that time period is relatively low [25-27, 29, 31-32]. This study also included only U.S. subjects, which may limit its generalizability. However, previous work has shown comparable effectiveness of Noom Weight use for weight loss across different regions and income levels [16,28]. This may suggest similar cross-national effects on HRU and costs, which would be more affected by access to healthcare and existing healthcare utilization patterns in each country than by differential impact of Noom Weight use. In addition, the study cohorts described here were mostly female and were aged <80, further limiting the generalizability of the results.
Some potential imbalances between cohorts may not have been accounted for in our IPTW analyses. Potential racial imbalances could not be accounted for as a large proportion (>40%) of both the Noom and non-Noom Weight user cohorts had either non-specific or missing information for race, which prevented reweighting on this variable. Pre-existing comorbidities were also not included in reweighting in order to permit subgroup analyses based on comorbid conditions. However, these were nevertheless reasonably well balanced in the reweighted cohorts. There may also be other confounding variables impacting HRU and costs that were not identified or accounted for in our analyses (e.g., education level, income bracket). The potential impact of other common weight loss interventions used concurrently, such as weight loss programs and anti-obesity medications, on study results also requires further investigation.
Open claims data were used, which allowed the assessment of direct medical costs in all care settings (e.g., inpatient, outpatient) and provided large sample sizes covering patients with diverse backgrounds and medical needs. However, there are limitations associated with the use of open claims data. Open claims databases effectively capture patient activity longitudinally, but do not necessarily capture all patient claims activity within a given time period. HRU involving service providers not included in the database will not be captured, giving a potentially incomplete picture of HRU and costs, biasing results if certain types of HRU are less well represented, and potentially excluding otherwise eligible patients for claims inactivity if they have unobserved claims. We applied a minimum claim activity criterion of one claim per 12-month period during the study period to mimic a continuous enrollment criterion that would be applied to a closed claims dataset. Although this was a low threshold that preserved sample size, it may have introduced some bias toward patients more likely to file claims, and therefore potentially sicker patients. As not all submitted claims in open claims databases are remitted, missing values were imputed to estimate costs. Imputation may potentially over-or underestimate true costs, and systematically bias any subcategory of HRU that is particularly affected by missing data. Finally, since open claims databases are based on a large convenience sample that is not random, there may be potential biases or issues with generalizability.
Discrete surgical visits could be readily determined from claims data for HRU analyses. However, individual CPT codes for activities within each surgery were frequently not available and were aggregated under a master code for the entire procedure. This prevented meaningful cost assessments for surgeries, which would require enumeration of the specific line items, and surgical costs were therefore only captured as a subset of overall costs.

Comparison with Prior Work
Noom Weight has previously been shown to be an effective treatment for obesity, frequently producing weight loss exceeding 5% of initial body weight [23,27,[32][33] in as little as 8 weeks [34] and persisting for up to 52 weeks [25]. However, Noom's impact, and the impact of mHealth technologies generally, on HRU and costs in overweight and obese users compared to a demographically similar control group has not been previously reported. Therefore, this study contributes to a substantial gap in the literature in which there is limited data on healthcare costs and utilization associated with digital programs. Below, we compare these findings to the few publications reporting on the impact of nonsurgical (i.e., behavioral, not including bariatric surgery) weight loss on HRU and costs.
One study compared healthcare costs over three years for 4,790 users of an employer-sponsored, digital weight loss program with a propensity-matched control group (N=4790) who did not use the program. Overall costs for those who used the program were $771 lower per person over three years compared to non-users. year alone, and these were not statistically significant compared to those who lost no weight. In comparison, statistically significant overall cost reductions were incurred by Noom Weight users over non-Noom Weight users that increased from year 1 through year 2 in our study, suggesting that the cost impact of Noom Weight use may be longer lasting than that observed with nonsurgical weight loss interventions in general. This is consistent with the typical trend of long-term weight regain (potentially correlating with increased costs) among those with nonsurgical weight loss in the absence of intensive lifestyle interventions, such as Noom Weight [8].
The degree of weight loss among Noom Weight users is closely tied to the level of user engagement, with greater loss among patients who more frequently read articles, log data, and interact with coaches [25,27]. Similar results have also been reported with other weight loss programs [36]. In our study, higher Noom Weight engagement was also associated with lower HRU in terms of the number of prescriptions claimed (Table 3). High-engaged Noom Weight users claimed 0.95 (almost 1 unit) fewer prescriptions than low-engaged Noom Weight users through 12 months, increasing to 2.79 fewer prescriptions through 24 months. This was also true for obesity-specific prescriptions, which were fewer for high-engaged compared to low-engaged Noom Weight users at 12 months (-0.16) and 24 months (-0.52). While this did not translate into statistically significantly lower costs for highengaged compared to low-engaged Noom Weight users, high Noom engagement was associated with statistically significantly lower overall costs of -$462 per user (95% CI: -775.62, -148.39) compared to non-Noom Weight use at 12 months, as well as -$1,446 lower overall costs (95% CI: -2,469, -422) and -$1,366 lower prescription costs (95% CI: -2,377, -355) compared to non-Noom Weight use at 24 months (Table 5d). In comparison, costs for these service types were not statistically significantly different for low-engaged Noom Weight users compared to non-Noom Weight users at 12 or 24 months.
Cost-effectiveness data for mHealth interventions for weight loss are limited. A review of 39 studies in upper and upper-middle income countries showed generally good and growing evidence of mHealth cost-effectiveness, although only two of the studies focused on interventions specifically for obesity [37]. One of the studies demonstrated cost-effectiveness of the mHealth intervention [38], while the other did not [39]. A more recent systematic review of 44 studies on mHealth interventions for older adults found evidence in favor of text-based interactions, but only limited evidence for complex smartphone interventions; however, none of the included studies focused on interventions specifically for obesity [40]. One cost-effectiveness modeling study from a New Zealand health system perspective suggested that the incremental cost of one quality-adjusted life year (QALY) for a national promotion campaign of smartphone programs for weight loss compared to no promotion campaign ranged between $20,100 and $53,600 (2011 USD), depending on program adoption rates. The national promotion campaign was considered borderline cost-effective against a threshold of $30,000 USD [41]. While our study did not specifically address the cost-effectiveness of Noom Weight, the results demonstrating lower HRU and costs for Noom Weight users compared to non-Noom Weight users can support future economic evaluations.

Conclusions
To our knowledge, this is the first study using real-world data to show the economic impact of mHealth use for the treatment of overweight and obesity compared to a control cohort not using the mHealth intervention. We show lower healthcare resource utilization (HRU) and costs for users of the Noom Weight mHealth program compared to non-Noom Weight users over a 2-year follow-up period. Comprehensively examining all service types, we found that inpatient visits, outpatient visits, surgical visits, and prescriptions were lower for Noom Weight users compared to non-Noom Weight users for up to 24 months after initiating Noom Weight. Costs per Noom Weight user were statistically significantly lower by $80 for outpatient services, $1,139 for prescriptions, and $1,219 overall at 24 months, which could correspond to an average of $609 per person per year during that period. These cost estimates compare favorably to previously studied programs. By linking Noom Weight data, EHR data, and claims data, we were able to conduct several subgroup analyses for HRU and costs, including analyses based on T2D or hypertension diagnosis, duration of Noom Weight use, user engagement level, and index BMI. Further research is required to establish the relationship between changes in weight and BMI, as well as in comorbidities, with changes in HRU and costs, including the impact of differential levels of weight loss. In addition, as this study focused on direct healthcare costs only, future research should investigate the impact of mHealth interventions on indirect costs (e.g., productivity costs) as well.

Figures
STaRT-RWE diagram of study design. AIDS: acquired immunodeficiency syndrome; BMI: body-mass index; COV: covariate assessment window; EXCL: exclusion assessment window; HIV: human immunodeficiency virus; INCL: inclusion assessment window.