The lifetime prevalence of major depression in the United States ranges from 13% to 17% [1]. The Global Burden of Disease study places major depression within the top 3 sources of disability-adjusted life years in the United States and Europe [2]. There are a range of effective pharmacological and behavioral treatments for depression [3]. However, many patients and health care providers must work through multiple trials before finding the optimal regimen. The Sequenced Treatment Alternatives to Relieve Depression trial found that 37% of the patients responded to the first antidepressant prescribed, but up to 4 medication changes were needed to achieve a 67% response rate overall [3-5].

Patient adherence to depression medication treatment contributes significantly to these treatment difficulties. Approximately one-third of patients discontinue antidepressants within the first month of treatment and 44% discontinue them by the third month of treatment [6]. Kales et al [7] examined veterans recently prescribed antidepressants and found 29% nonadherence based on self-report and 53% nonadherence when based on the medication possession ratio using veterans Affairs (VA) pharmacy data. Samples and Mojtabai [8] conducted a national survey and found that the major reasons reported for discontinuation were side effect burden, patient experience that the medications are not working, and patients wanting to resolve depression without using medication. Although many patients attribute nonadherence to simple forgetting, Gadkari and McHorney [9] demonstrated that negative beliefs about medication predict unintentional nonadherence (eg, forgetting doses and running out of medication).

To improve antidepressant adherence, providers need effective tools to address reasons for self-discontinuation [8]. There have been many commercial mobile apps or texting technologies developed to promote medication adherence and management across a range of medical conditions [10]. Meta-analytic reviews of the use of mobile apps to promote psychiatric medication adherence in community settings indicate small to moderate effect sizes (ESs) [11-13]. One observed limitation of mobile app use in health care has been a significant drop off in use and patient engagement with the application over time [11]. Much of this prior work examined general apps that were not tailored to a specific medication (eg, antidepressants), and many were not integrated into the overall treatment setting, which may be the critical factor. Mohr et al [14] in an article entitled, “Three Problems with Current Digital Mental Health Research...and Three Things We Can Do About Them,” argue that mobile apps’ limited effectiveness is because of their being developed and tested outside of the clinic setting where they can have the most impact.

This study tested the acceptability, feasibility, and preliminary effectiveness of an intervention designed to improve adherence and depression outcomes among veterans initiating antidepressant medication. In the intervention, veterans used Annie, the Department of VA mobile app SMS, to support self-management of antidepressant medication use. Implementation and adoption of Annie have been explored in self-support for the hepatitis C virus. This pilot indicated low-to-moderate adoption and perceived value of the messaging service, but among adopters, there was enhanced treatment adherence and lower distress about failing hepatitis C virus treatment [15]. Shimada et al [16] reported that secure messaging adoption in VA was enhanced by leadership and technical support at the facility level, and early adopters of secure messaging in primary care demonstrated lower urgent care use.

For the program described herein, we purposely designed the Annie protocol to counter the key reasons for antidepressant self-discontinuation, such as side effect burden or prolonged delay (typically 6 wk) of treatment effect. To increase Veteran and provider satisfaction with Annie, study staff provided participants and providers visual representations of their Annie responses (ie, graphs and bar charts) between medical visits. Veterans also received training about My HealtheVet (MHV), VA’s patient portal, as a communication tool should any medication-related issues arise.

This pilot was a 3-armed randomized controlled trial of 53 veterans undergoing depression treatment at the Iowa City VA Health Care System. The proposed protocol was shared with the Iowa City VA Veteran Engagement Panel where veterans provided feedback and recommendations based on patient experiences. The panel members introduced the idea of presenting visual representations of patient-centered outcomes, such as side effect burden, in addition to the standard use of graphs to chart depressive symptom severity over time. The panel members also recommended that Annie’s responses include information about the Veteran Crisis Line and use more encouraging and patient-centered language for responses related to veterans experiencing medication side effects.

All aspects of the protocol were approved by the University of Iowa and Veteran’s Administration internal review board (IRB ID#201712732). It is registered on ClinicalTrials.gov (NCT03930849). The study coordinator provided informed consent to all participants, clearly describing the 3 arms of the study, the nature of the randomization, the amount of time involved, and all potential benefits and harms of participation. All study data were deidentified for the preparation of this manuscript. Participants in all treatment arms were reimbursed US $25 for completing all assessments for the study.

Veterans were eligible for the study if they were seeking care at either the Iowa City VA Health Care System Primary Care or Mental Health Service Line and were initiating new antidepressant treatment for depression OR they had a recent change in antidepressant medication. New antidepressant treatment was defined as the initiation of antidepressant treatment and no indication of any antidepressant use in the previous year. Veterans with medical record indications of comorbid major psychotic disorders, current active substance abuse, or cognitive disorders were excluded. Those with comorbid posttraumatic stress disorder were eligible. Veterans were not required to be experienced in using texting. However, they did need to own a cell phone (flip phone, smartphone, or Android) with texting functionality and a data plan that allowed for texting.

Veterans who met initial eligibility criteria were mailed an invitation letter that described the study and contained information on how to contact staff if they had questions or wanted to express whether they were interested in participating. Study staff then followed up with a phone call to veterans. Those who were interested and eligible were scheduled for an in-person visit with a study coordinator at a location convenient for them, either at their local VA clinic or at their home. At the visit, veterans went through consent procedures, were randomized to 1 of the 3 arms, as described in the Data Collection and Intervention Descriptions section, and then completed baseline assessments. Randomization was blocked by recruitment within the primary care or Mental Health Service Line at a local VA medical center. We recruited 8 veterans from primary care and 45 veterans from the Mental Health Service Line. The baseline assessments included the assessments described in the Data Collection and Intervention Descriptions section. The primary outcome was a change in self-reported medication adherence. The secondary outcomes were changes in depressive symptom burden and side effect burden.

The primary hypothesis was that Veteran participants in the SM+Coach and the SM-Only interventions would have better self-reported antidepressant treatment adherence than those in the AC condition as indicated by the total MARS score measured at baseline and 6 and 12 weeks after baseline. As this was a pilot study with a small sample size, we conducted a mixed effects model including a group (SM-Only, SM+Coach, and AC) by time (baseline, 6 wk, and 12 wk) interaction term to derive an F value and significance level for type 3 fixed effects. ESs were calculated separately for each intervention by comparing SM-Only versus AC and SM+Coach versus AC. We did so by determining an ES metric where the difference between the mean change scores for the 2 being compared was the numerator. The denominator for this metric was the SD of the combined mean change scores for the 2 intervention arms being compared [21]. ESs were categorized where 0.2−0.5=small ES, 0.5−0.8=moderate ES, and ≥0.8=large ES [22].

For this pilot study, 327 veterans were approached to participate. Of these, 120 (36.7%) were not interested, 72 (22%) were unable to contact, and 23 (7%) were ineligible. Of the 53 patients who provided consent, 3 dropped out before starting the study. Sample characteristics are presented in Table 2. The average age was 46 (SD 13.62) years, with 28% (15/53) women, 4% (2/53) African American, and 8% (4/53) Hispanic. Initial sample sizes immediately after randomization were 17 for the SM-Only arm, 18 for the SM+Coach arm, and 18 for the AC arm. Throughout the 12-week intervention, 4 discontinued SM-Only, 4 discontinued SM+Coach, and none discontinued AC. For each of the 3 arms, the sample was predominantly male, consistent with the Veteran enrollee population as a whole.

Table 2 presents descriptions of the Veteran participants in each of the 3 arms, as well as results on the 3 key outcomes of this study. Change from baseline to 12 weeks on the MARS (increase=poor adherence) declined in the 2 intervention arms, SM-Only (ES=0.09; P=.19) or SM+Coach (ES=0.85; P=.002), whereas poor adherence increased for those in the AC arm. These differences only reached statistical significance for the comparison between the SM+Coach intervention compared with the AC.

Depression scores on the 9-Item Patient Health Questionnaire decreased for all 3 arms, although the decline was slightly larger for the SM-Only (ES=0.32; P=.09) and the SM+Coach (ES=0.24; P=.28) arms than for the AC arm. None of these differences in depression changes between the groups reached statistical significance. The 2 intervention arms indicated a decrease in side effects on the Frequency, Intensity, and Burden of Side Effects Ratings, whereas the side effect assessment for the AC arm increased. These differences were not statistically significant but indicated moderate ES (SM-Only vs AC: ES=0.40; P=.07; SM+Coach: ES=0.54; P=.06)

In week 1, all veterans participants taking part in 1 of the 2 active treatment arms responded to at least one text. By week 6, 86% (30/35) had responded to at least one text, and in the final week, 80% (28/35) had responded to Annie at least once. The maximum number of days a participant could respond to a text was 5 days per week. In week 1, the average number of days responded across both treatment arms was 4.74 (SD 0.70). By week 6, this gradually declined to an average of 4.11 (SD 1.76) days. At week 12, Veteran participants responded to a text for an average of 3.37 (SD 2.02) days.

Table 3 presents Annie’s quotes and tips ordered from most helpful to least helpful. These scores were based on responses to the question, “How helpful did you find this?” (1 for “helpful,” 0 for “Not Helpful”). Although there was some variability in responses from veteran participants, most of the Annie’s quotes and tips were seen as helpful. The most helpful were those offering emotional support and promoting perseverance such as, “Every day may not be good...but there is something good in every day.” The lower-ranked quotes and tips tended toward concrete reminders about illness management such as “Do not adjust your medications on your own even if you are feeling better.”

This pilot study comparing 2 forms of a mobile health intervention to an AC condition demonstrated feasibility, acceptability, and small to moderate ES in improving antidepressant adherence and reducing side effect burden. It did not appear to reduce depressive symptom severity. Consistent with previous research, the intervention that included brief synchronic meetings with a coach appeared to have greater benefit than SMS alone. In a recent review of digital psychiatry, Torous et al [23] recommended that health care systems support digital navigators that assist both clinicians and patients in adopting mobile technologies to manage their health. Such navigators, akin to the coaches in our study, can promote ongoing engagement with the application so that the patient may eventually benefit from the technology.

Veteran participants who remained in the study engaged with the mobile app over 12 weeks in contrast to the prior studies. Moreover, 80% of the participants remained engaged with Annie at weeks 6 and 12 of the intervention. Prior estimates of patient engagement for mobile apps report lower retention on average [24] than what was observed in this study. Our retention was comparable with retention found for a mobile app targeting self-management of schizophrenia [11] and slightly higher than what was promoted in a study focusing specifically on people with depression, which found that app engagement dropped off significantly after 2 weeks [25]. The intervention designed herein combined secure messaging with some intermittent contact that may boost retention.

In addition to the coaching, our sustained retention may be because of the direct integration of Annie into the patient’s current treatment, and app results were shared with both Veteran participant and their provider. The actual content of the app was informed by the specific clinical goal at hand: encouraging veterans to remain on their antidepressant long enough to experience and evaluate its true effectiveness. The intervention also included tips and quotes intended to be supportive or inspirational and Veteran participants rated the degree to which these were helpful. It appears that those who provided general encouragement were rated as more helpful than those who recommended specific self-care behaviors. A similar patient engagement strategy using a chatbot with sophisticated branching logic was tested among individuals with eating disorders [26]. As observed in our study, these chatbot users rated the personalized messaging highly and remained engaged with the technology over time.

Simon et al [27] conducted a web messaging program to support depression follow-up care that combined algorithm-determined responses with real-time text communication with a care manager when required. The care manager then consulted with a physician about treatment options and responses to patient communications. Similar to our study, this program yielded improved antidepressant adherence but also achieved a greater reduction in depressive symptoms over 5 months.

The differential dropout between the 2 interventions and control condition suggests that a significant portion of these Veteran participants did not accept the Annie intervention. Consistent with prior research on mobile apps in mental health, dropout can be high and often occurs within the first month of an intervention. This limits our ability to estimate the true impact across the entire target population of veterans initiating antidepressants. Future research is needed to better understand these technology refusers and to determine whether there are strategies to enhance their experience and retention.

Annie was harnessed in the VA’s response to the COVID-19 pandemic [28]. The algorithm-derived messaging protocol assessed how veterans were feeling and their temperature and probed for difficulty breathing. Referrals to contact VA providers occurred based on Veteran responses. Program evaluation revealed that this likely reduced the overall cost of COVID-19 management as it supplanted direct contact with the VA. Moreover, 75% of the respondents to a postintervention survey indicated that Annie’s messages were helpful. In light of these successes, VA continues to explore how algorithm-based text messaging integrated into the care provided by a large health care system can improve population health in a cost-effective manner.

This study has several limitations. It was a pilot study designed to demonstrate feasibility and determine ES but was not powered to perform definitive tests of statistical significance between treatment arms. Moreover, there was a greater dropout in the active intervention arms than in the AC arm. It is not clear why this occurred, but it is possible that those with negative experiences around texting stopped study participation, thereby biasing estimates of positive impact and overall estimates of patient engagement over 12 weeks.

This study presents an SMS protocol developed specifically to address key barriers to antidepressant adherence. Consistent with the overarching recommendation to integrate technology into clinical contexts and practice, the program was developed to collect information demonstrated to undermine antidepressant adherence such as perceived effectiveness and side effect burden. SMS responses aimed to educate Veteran participants based on their responses to the probes. In addition, this information was presented in a graphic format and was shared with both the patients and their providers. Veterans engaged with Annie and there is preliminary support that this enhanced their self-management of depression, although differential impact on actual depressive symptom burden did not occur. However, there was differential dropout with significantly greater rates in the interventions that clouds the interpretation of results.