Achieving Rapid Blood Pressure Control With Digital Therapeutics: Retrospective Cohort and Machine Learning Study.

BACKGROUND: Behavioral therapies, such as electronic counseling and self-monitoring dispensed through mobile apps, have been shown to improve blood pressure, but the results vary and long-term engagement is a challenge. Machine learning is a rapidly advancing discipline that can be used to generate predictive and responsive models for the management and treatment of chronic conditions and shows potential for meaningfully improving outcomes. OBJECTIVE: The objectives of this retrospective analysis were to examine the effect of a novel digital therapeutic on blood pressure in adults with hypertension and to explore the ability of machine learning to predict participant completion of the intervention. METHODS: Participants with hypertension, who engaged with the digital intervention for at least 2 weeks and had paired blood pressure values, were identified from the intervention database. Participants were required to be ≥18 years old, reside in the United States, and own a smartphone. The digital intervention offers personalized behavior therapy, including goal setting, skill building, and self-monitoring. Participants reported blood pressure values at will, and changes were calculated using averages of baseline and final values for each participant. Machine learning was used to generate a model of participants who would complete the intervention. Random forest models were trained at days 1, 3, and 7 of the intervention, and the generalizability of the models was assessed using leave-one-out cross-validation. RESULTS: The primary cohort comprised 172 participants with hypertension, having paired blood pressure values, who were engaged with the intervention. Of the total, 86.1% participants were women, the mean age was 55.0 years (95% CI 53.7-56.2), baseline systolic blood pressure was 138.9 mmHg (95% CI 136.6-141.3), and diastolic was 86.2 mmHg (95% CI 84.8-87.7). Mean change was -11.5 mmHg for systolic blood pressure and -5.9 mmHg for diastolic blood pressure over a mean of 62.6 days (P<.001). Among participants with stage 2 hypertension, mean change was -17.6 mmHg for systolic blood pressure and -8.8 mmHg for diastolic blood pressure. Changes in blood pressure remained significant in a mixed-effects model accounting for the baseline systolic blood pressure, age, gender, and body mass index (P<.001). A total of 43% of the participants tracking their blood pressure at 12 weeks achieved the 2017 American College of Cardiology/American Heart Association definition of blood pressure control. The 7-day predictive model for intervention completion was trained on 427 participants, and the area under the receiver operating characteristic curve was .78. CONCLUSIONS: Reductions in blood pressure were observed in adults with hypertension who used the digital therapeutic. The degree of blood pressure reduction was clinically meaningful and achieved rapidly by a majority of the studied participants. Greater improvement was observed in participants with more severe hypertension at baseline. A successful proof of concept for using machine learning to predict intervention completion was presented.

Emergence of digital biomarkers to predict and modify treatment efficacy: machine learning study.

OBJECTIVES: Development of digital biomarkers to predict treatment response to a digital behavioural intervention. DESIGN: Machine learning using random forest classifiers on data generated through the use of a digital therapeutic which delivers behavioural therapy to treat cardiometabolic disease. Data from 13 explanatory variables (biometric and engagement in nature) generated in the first 28 days of a 12-week intervention were used to train models. Two levels of response to treatment were predicted: (1) systolic change ≥10 mm Hg (SC model), and (2) shift down to a blood pressure category of elevated or better (ER model). Models were validated using leave-one-out cross validation and evaluated using area under the curve receiver operating characteristics (AUROC) and specificity- sensitivity. Ability to predict treatment response with a subset of nine variables, including app use and baseline blood pressure, was also tested (models SC-APP and ER-APP). SETTING: Data generated through ad libitum use of a digital therapeutic in the USA. PARTICIPANTS: Deidentified data from 135 adults with a starting blood pressure ≥130/80, who tracked blood pressure for at least 7 weeks using the digital therapeutic. RESULTS: The SC model had an AUROC of 0.82 and a sensitivity of 58% at a specificity of 90%. The ER model had an AUROC of 0.69 and a sensitivity of 32% at a specificity at 91%. Dropping explanatory variables related to blood pressure resulted in an AUROC of 0.72 with a sensitivity of 42% at a specificity of 90% for the SC-APP model and an AUROC of 0.53 for the ER-APP model. CONCLUSIONS: Machine learning was used to transform data from a digital therapeutic into digital biomarkers that predicted treatment response in individual participants. Digital biomarkers have potential to improve treatment outcomes in a digital behavioural intervention.

Change in Glycemic Control With Use of a Digital Therapeutic in Adults With Type 2 Diabetes: Cohort Study.

BACKGROUND: Intensive lifestyle change can treat and even reverse type 2 diabetes. Digital therapeutics have the potential to deliver lifestyle as medicine for diabetes at scale. OBJECTIVE: This 12-week study investigates the effects of a novel digital therapeutic, FareWell, on hemoglobin A1c (HbA1c) and diabetes medication use. METHODS: Adults with type 2 diabetes and a mobile phone were recruited throughout the United States using Facebook advertisements. The intervention aim was to effect a sustainable shift to a plant-based dietary pattern and regular exercise by advancing culinary literacy and lifestyle skill acquisition. The intervention was delivered by an app paired with specialized human support, also delivered digitally. Health coaching was provided every 2 weeks by telephone, and a clinical team was available for participants requiring additional support. Participants self-reported current medications and HbA1c at the beginning and end of the 12-week program. Self-efficacy related to managing diabetes and maintaining dietary changes was assessed via survey. Engagement was recorded automatically through the app. RESULTS: We enrolled 118 participants with a baseline HbA1c >6.5%. Participants were 81.4% female (96/118) and resided in 38 US states with a mean age of 50.7 (SD 9.4) years, baseline body mass index of 38.1 (SD 8.8) kg/m2, and baseline HbA1c of 8.1% (SD 1.6). At 12 weeks, 86.2% (94/109) of participants were still using the app. Mean change in HbA1c was -0.8% (97/101, SD 1.3, P<.001) for those reporting end-study data. For participants with a baseline HbA1c >7.0% who did not change medications midstudy, HbA1c change was -1.1% (67/69, SD 1.4, P<.001). The proportion of participants with an end-study HbA1c <6.5% was 28% (22/97). After completion of the intervention, 17% (16/97) of participants reported a decrease in diabetic medication while 8% (8/97) reported an increase. A total of 57% (55/97) of participants achieved a composite outcome of reducing HbA1c, reducing diabetic medication use, or both; 92% (90/98) reported greater confidence in their ability to manage their diabetes compared to before the program, and 91% (89/98) reported greater confidence in their ability to maintain a healthy dietary pattern. Participants engaged with the app an average of 4.3 times per day. We observed a significantly greater decrease in HbA1c among participants in the highest tertile of app engagement compared to those in the lowest tertile of app engagement (P=.03). CONCLUSIONS: Clinically meaningful reductions in HbA1c were observed with use of the FareWell digital therapeutic. Greater glycemic control was observed with increasing app engagement. Engagement and retention were both high in this widely distributed sample.

FareWell and the How of Lifestyle Medicine.

The what of Lifestyle Medicine, including a whole foods dietary pattern, has been well established, but the how has remained elusive. How do we apply what we know in a cost-effective and widely accessible manner to prevent, treat, and even reverse chronic disease? Over the decade ahead, we believe the field of Lifestyle Medicine and the people who need it most will benefit from addressing the how. This article summarizes the founding and operational principles of FareWell Inc. - a digital therapeutics company targeting lifestyle-related cardiometabolic diseases. We outline our current use of mobile health technology and artificial intelligence and describe our early clinical experience, business model, and key anticipated challenges.