In total, 318 BetterHelp clients (of whom 254, 79.9%, were females), recruited from a larger pool of active BetterHelp users, participated in this study. BetterHelp users are individuals aged ≥18 years seeking to improve their quality of life. Users aged <18 years or under the care of a legal guardian were excluded from BetterHelp participation. Furthermore, individuals with thoughts of hurting themselves or others, those in urgent crisis or emergency situations, those diagnosed with a severe mental illness or advised to be in psychological supervision or psychiatric care, and those required to undergo therapy or counseling either by a court order or by any other authority were also excluded. BetterHelp users who met the eligibility criteria to participate in this study were invited to participate. Inclusion criteria for this study included a self-endorsed primary concern of feelings of overwhelming sadness, grief, or depression. Study participants were excluded from participation if preintervention levels of depression fell below mild clinical significance, that is, a score of <5 on the Patient Health Questionnaire (PHQ-9) [50], or if they had not engaged with BetterHelp for a minimum of 90 days. Users with preintervention depression in the mild to severe range, (ie, PHQ-9 scores ≥5 and ≤27) were included (Textbox 1). A total of 1148 BetterHelp users met all the eligibility criteria and were invited to participate. Our observed response rate of 27.70% (318/1148) is comparable to response rates seen in other Web-based survey studies [51]. The racial and ethnic makeup of this sample, as well as comorbid mental or physical health concerns, is unfortunately unknown as the BetterHelp platform does not currently collect this information. Ages of participants meeting the eligibility criteria ranged from 19 to 72 (mean 33.27 [SD 11.29]) years. At baseline, of 318, 119 (37.4%) participants met the criteria for mild depression, 91 (28.6%) for moderate depression, 75 (23.6%) for moderately severe depression, and 33 (10.4%) for severe depression [50,52,53] (Table 1).

This study was conducted in accordance with the Institutional Review Board and Office for the Protection of Human Subjects at the University of California, Berkeley. All subjects were informed of the risks and benefits of participating in the study and gave electronic informed consent to participate via the BetterHelp platform. Participants were provided with access to the BetterHelp platform and were not additionally compensated or incentivized to participate in any way. The research study was approved by the Institutional Review Board at the University of California, Berkeley.

Table 2 describes the sociodemographic characteristics (age, gender, health or financial status, and prior exposure to therapy) of the study sample. Overall, of 318 participants, 72 (22.6%) rated their financial status as poor, 167 (52.5%) as fair, and 49 (15.4%) as good. In addition, 24 (7.5%) participants rated their health status as poor, 165 (51.9%) as fair, and 118 (37.1%) as good; furthermore, 91 (28.6%) participants did not have prior exposure to counseling or therapy. Table 3 displays the modality usage in the BetterHelp user sample.

In this study, of 318 participants, 120 (37.8%) demonstrated a clinically significant improvement and 194 (62%) demonstrated a partial response (as defined by at least a 5-point score reduction on the PHQ-9 and a postintervention score ≤9, respectively [53]) after engaging with BetterHelp for 3 months; in addition, 63 (19.8%) participants qualified as being in remission (as defined by a postintervention score <5) by Time 2. Paired samples t test results revealed a significant decrease in symptom severity posttreatment with effect size in the medium range (pretreatment: mean 12.57 [SD 5.35]; posttreatment: mean 9.36 [SD 5.51]; t₃₁₇=10.80; P<.001, one-tailed; Cohen d=0.61; Figure 1). Mean PHQ-9 pretreatment scores reflected moderate levels of current depression, whereas mean posttreatment scores reflected mild levels of current depression, as determined by PHQ-9 clinical cutoffs [52]. As shown in Table 1, after using BetterHelp for 3 months, of 318 participants, 63 (19.8%) met the criteria for minimal depression, 141 (41.2%) for mild depression, 66 (20.8%) for moderate depression, 40 (12.6%) for moderately severe depression, and 18 (5.7%) for severe depression [52,53].

A one-way ANCOVA examining the differential effects of age and gender on therapy outcomes, covarying baseline symptom severity, revealed no significant differences. Outcomes did not significantly differ across age (F_45,242=0.98; P=.51) and gender (F_1,242=.092; P=.76).

A one-way ANCOVA examining effects of financial status, health status, and prior exposure to therapy on treatment outcomes, covarying baseline symptom severity, revealed a significant effect of prior exposure to therapy on treatment outcome (F_1,260=7.531; P=.006). Individuals with prior therapy exposure experienced significantly fewer gains after treatment compared with individuals without prior exposure (Figure 2). Treatment outcomes did not significantly differ across participant financial status (F_2,260=1.563; P=.21) or health status (F_2,260=1.575; P=.21).

Treatment dosage was measured as the sum of individual therapist-user interactions across modalities (text message, phone, video, and live chat). Response was measured as a binary variable indicating the presence or absence of clinical improvement, as defined by a PHQ-9 score change of ≥5 points. Mean number of interactions in this study was 125.3 (SD 392.5). Following the methodology laid out by Howard et al in the probit model [59], the dose corresponded to the log of the number of total interactions, to reduce skew. Because participants who did not interact with BetterHelp were not included in the sample, taking the log of 0 was not a possibility. Results revealed no significant dose-response effect (P=.14), which was maintained after adding baseline severity into the model.

To adjust PHQ-9 scores for regression to the mean, an ANCOVA in which the difference between baseline and posttreatment scores was regressed onto the mean-centered baseline score was used [58,60]. These additional analyses also revealed a significant mean difference (P<.001).

To assess the hypothesis that differences in treatment gains between individuals with and without prior therapy experience may be driven by differences in therapeutic alliance associated with prior therapy exposure, an independent samples t test was performed. Cronbach alpha for the 12 WAI-SR items was high (alpha=.946). The results revealed no significant effect of prior therapy experience on WAI-SR total scores (P=.55).

Our results indicate that multimodal digital psychotherapy may be an effective treatment for adult depression. Given our preliminary results demonstrating the effectiveness of this method of psychotherapy dissemination, we believe that multimodal digital psychotherapy may hold promise in overcoming some existing issues of psychotherapy access. Hence, continued research is needed. Users of a multimodal digital psychotherapy platform experienced significantly improved self-reported symptoms after engaging with BetterHelp, with 37.8% (120/318) users experiencing clinically significant improvement in depressive symptoms within 3 months. Moreover, no significant associations were found between changes in depressive symptoms and sociodemographic variables, including age, gender, and self-reported financial or physical health status. We suspect that increased accessibility and flexibility provided by a preference-based multimodal platform may be driving these latter findings [61,62]. Although this study cannot establish causality, our findings suggest that multimodal digital psychotherapy may be an effective solution to reduce existing barriers to accessible and preference-based digital psychotherapy.

Participants who had previously engaged in traditional face-to-face psychotherapy showed significantly less symptomatic improvement compared with those who had not. Among several potential explanations, we postulate that individuals with prior therapy experience may present with more chronic, complex, comorbid, or treatment-resistant forms of depression requiring a higher level of care [63-65]. Post-hoc analyses did not reveal any significant association between therapeutic relationship quality (as measured by the WAI) and prior therapy engagement, suggesting that this finding is not merely related to differences in therapist rapport and alliance linked to prior treatment exposure. Future research should seek to investigate differences in outcomes across diagnostic categories, taking into account commonly comorbid conditions known to influence response to treatment such as generalized anxiety disorder, attention deficit/hyperactivity disorder, and panic disorder. Baseline evaluation of the presence and severity of additional diagnoses may ensure that individuals with comorbid conditions are matched with an appropriately specialized therapist.

No general dose-response effect was detected in this study. Although this is not the first study to find nonsignificant dose-response effects of psychotherapy, we speculate that our existing measure of dosage may disguise existing dose-response effects, given that the quantity and significance of content shared in a single message may vary greatly by individual. Future research may seek to examine effects of word count as opposed to the message count, though due to user privacy concerns, researchers were not able to access these data at this time. Furthermore, the number of days spent interacting with the BetterHelp platform may prove to be a more valid measure of engagement than total interactions.

All study participants in this preference-based study utilized the SMS text message modality. A majority of participants utilized only one therapy modality over the course of the intervention, with only a single participant making use of all 4 modalities. We hypothesize that this pattern of use is driven by the increased convenience and flexibility provided by SMS text message-based therapy (ie, a previously scheduled appointment is not needed to utilize the SMS text message-based modality, whereas an appointment is needed to utilize live chat-, phone-, or video-based therapy). Future work will seek to test this hypothesis by obtaining qualitative data from participants regarding motivations behind modality choices, as well as to elucidate the added benefit of a multimodality psychotherapy platform (ie, using a randomized “text-only” vs “multimodal” design).

This exploratory investigation contains several limitations of note. First and foremost, we lacked a randomly assigned control group. Future investigations with appropriate controls, as well as a more comprehensive demographic and diagnostic screening, will enable a more valid approach. In addition, like many survey-based studies, the results of this work may be influenced by sample bias. It is possible that individuals who had notably positive or negative experiences with BetterHelp are those who chose to respond to our prompt to participate in research. We are further unable to investigate the effect of the type of psychotherapy provided (ie, cognitive behavioral therapy, acceptance and commitment therapy, etc) on outcomes. However, it is worth noting that the existing literature examining differential effects of different types of psychotherapies for depression does not suggest that psychotherapy type has a significant effect on outcomes in traditional psychotherapeutic settings [8]. In this study, financial and physical heath status were measured by self-report; such subjective indicators may not be fully valid. Future work may seek to utilize additional objective and sensitive measures of financial and physical health status to provide further insight. Finally, in this study, we lacked data on the potentially crucial moderator variables of race, ethnicity, and gender nonconformity.

Major depressive disorder is a pervasive and debilitating condition from which many individuals are unable to recover due to lack of accessible and appropriate treatment. The existing literature has demonstrated the use of digital technology as a feasible solution to this growing and widespread dilemma. This study examined the effect of a multimodal digital psychotherapy platform for the treatment of depression in adults. We proposed that a multimodal platform through which users can dynamically select from multiple modes of digital communication throughout therapy may be an effective method of delivering psychotherapy to adults with depression. Our study results demonstrate the initial effectiveness of such a model, with users experiencing significant symptom reduction after the intervention. This feasibility study’s preliminary demonstration of such a platform’s effectiveness provides an important first step in understanding the potential benefits of such a model of mental health care delivery. Given these results, subsequent research will aim to investigate the unique benefits of a preference-based multimodal platform compared with in-person, single- or bimodal psychotherapy dissemination.