This retrospective study analyzes secondary data from Rula Health, a DMHI that connects individuals with licensed therapists and psychiatric providers across the United States. Rula Health’s treatment model is grounded in MBC and personalized treatment to enhance clinical outcomes. Individuals were eligible for inclusion if they were adults aged 18 years or older and received therapy treatment through Rula Health between August 20, 2020, and March 31, 2025. Only therapy visits were included for analysis. Eligible participants were identified using existing clinical records collected in the course of routine care (ie, retrospective cohort sampling). Because the study was retrospective, no direct recruitment of participants occurred.

All patients at Rula Health complete an informed consent to care prior to treatment, which outlines the scope of services, confidentiality policies, and data usage information. This consent process is separate from and in addition to agreeing to Rula Health’s privacy policy, which outlines how their data may be used and protected. This study involves a retrospective analysis of deidentified secondary data, qualifying it for exemption from additional consent requirements under human participant research regulations. The Biomedical Research Alliance of New York (BRANY) Institutional Review Board (Study ID 25-035-2061) approved this study. Participants did not receive compensation for this study.

Patients learn about Rula Health through various referral sources, including primary care providers, insurance providers, employer-sponsored mental health benefits, health care partners, and direct online searches. The platform is designed to simplify access to mental health care while ensuring that patients have control over their provider selection. After completing a brief intake questionnaire, individuals are matched with a curated list of licensed therapists or psychiatric providers based on their mental health concerns, insurance coverage, and personal preferences, such as provider specialty, background, and availability. Rather than being assigned a provider, patients can choose from these matches, allowing them to select someone who best aligns with their needs and preferences.

Once matched, patients begin treatment through secure video sessions with their chosen provider. During the initial session, the provider collaborates with the patient to develop a personalized treatment plan that outlines therapy goals, intervention strategies, and expected progress. Rula Health’s care approach integrates MBC, meaning that patients complete standardized symptom assessments at regular intervals. These assessments help track changes in symptom severity and inform data-driven adjustments to treatment plans. Additionally, Rula Health’s model emphasizes provider quality, equipping clinicians with tools and insights that support evidence-based care while monitoring treatment effectiveness to ensure high standards of service.

Therapy sessions are scheduled based on individual patient needs, with some patients engaging in weekly sessions and others biweekly or a more flexible schedule, depending on symptom severity, treatment goals, and provider recommendations. Rula Health providers use a variety of evidence-based therapeutic approaches, including cognitive behavioral therapy, dialectical behavior therapy, acceptance and commitment therapy, and other empirically supported interventions tailored to each patient’s needs. The treatment process is collaborative, with therapists working closely with patients to establish personalized goals, track progress, and adjust interventions as needed. Patients complete assessments on their mental health symptoms at regular intervals, and Rula Health encourages providers to use these assessment scores to guide the patient’s treatment as part of MBC.

Upon intake, patients report demographic information including age, biological sex, race, and ethnicity. Providers report the patient’s primary diagnosis. Among patients with completed assessments, mental health symptom assessments were completed before each scheduled appointment or up to once per week starting August 20, 2020. Starting December 8, 2024, patients are asked to complete these assessments every 2 weeks. This analysis incorporates assessment data on depression and anxiety symptoms.

The Patient Health Questionnaire-9 (PHQ-9) is a 9-item tool that assesses the severity of depressive symptoms over the past 2 weeks. Patients rate each item on a 4-point Likert scale from 0 (“Not at all”) to 3 (“Nearly every day”). The total score ranges from 0 to 27. The PHQ-9 has demonstrated high internal consistency, with a Cronbach α of .89, and excellent test-retest reliability, with a correlation coefficient of 0.84 [28]. Higher scores indicate greater depression severity [29].

The Generalized Anxiety Disorder-7 (GAD-7) is a 7-item self-report questionnaire that evaluates the severity of anxiety symptoms over the past 2 weeks. Patients rate each item on a 4-point Likert scale, ranging from 0 (“Not at all”) to 3 (“Nearly every day”), with total scores spanning from 0 to 21. The GAD-7 is widely used and validated, demonstrating high internal consistency (Cronbach α=.89‐0.92), strong test-retest reliability (intraclass correlation=0.83) [30,31]. Higher scores indicate greater anxiety severity [32,33].

Data were limited to the first 12 visits because this is a common length of treatment in clinical trials [34], and thus estimates a relative episode of care at Rula Health. If outcomes data were missing for the 12th visit, or if the patient was not in treatment for a full 12 visits, data from the nearest visit prior to the 12th visit was used to calculate all change metrics. Because of the very large number of primary diagnoses given by providers, patient diagnoses were collapsed into 4 categories based on Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, categories: Anxiety Disorders, Depressive Disorders, Trauma and Stress-Related Disorders, and Other Disorders (the “Other” category captured any diagnosis not in the 3 categories). The primary diagnosis was reported for each visit; thus, a person-level variable representing each individual’s modal diagnostic category was calculated. The modal diagnostic category was used in all subsequent analyses. Demographic characteristics and descriptive statistics were calculated and reported for the sample. Cohen d was calculated to estimate the effect size of the change in depression and anxiety between baseline and up to 12 visits, with missing data handled using the last observation carried forward method. Due to high missingness in data at exactly 12 visits (~80%), sensitivity analyses were conducted to test for differences in PHQ-9 and GAD-7 change scores between patients with and without scores at 12 visits (see Multimedia Appendix 1). Given the difference observed between those with and without scores at 12 visits, multiple imputation was used to impute visit 12 PHQ-9 and GAD-7 scores to reduce bias and preserve statistical power in our findings. Because results were highly similar to the last observation carried forward analyses, these results are in the supplement.

Linear mixed models were used to model changes in depression and anxiety symptoms over time in treatment at Rula Health. Visual inspection of the raw data indicated that a non-linear function of time should be applied (see Figure 1). Visual fit has been cited as a way to determine knot placement in spline models [35-38]. As a result, we applied a linear spline model with a knot at visit 5, allowing us to model slopes between visits 1‐5 and 6‐12 separately. A base model with random intercepts was estimated for depression and anxiety outcomes separately, then demographic covariates including gender, age, race, ethnicity, and modal primary diagnosis were added. Interactions between demographic variables and the slopes between 1‐5 visits and 6‐12 visits were added to test whether treatment effects differed based on demographic characteristics. Separate models were estimated for each demographic variable to simplify the interpretation of the interaction terms. Satterthwaite’s method was used to approximate the degrees of freedom and statistical significance of the slopes in the models. Models with random slopes were tested but did not converge and thus were not used. Restricted maximum likelihood estimation was used in all models. All statistical analyses were conducted in R (version 4.3.1; R Foundation for Statistical Computing) [39].

A total of 365,741 adults and 2,685,103 therapy visits were included in the analysis. Patients had a mean age of 37.03 (SD 11.81) years and were mostly female (238,839/360,388, 66.27%), Caucasian/White (87,758/145,947, 60.13%), and non-Hispanic (115,033/168,633, 68.21%). Patients’ modal primary diagnoses were anxiety disorder (116,754/350,495, 33.31%), depressive disorder (82,833/350,495, 23.63%), and trauma and stress-related disorder (107,968/350,495, 30.80%) (Table 1). Patients attended an average of 7.34 (SD 4.33) sessions and were in treatment for an average of 80.36 (SD 95.14) days. On average, patients reported baseline depression and anxiety scores on the border between mild and moderate symptoms (mean baseline depression=9.41, SD 6.61; mean baseline anxiety=9.45, SD 5.65), and final scores within 12 visits decreased to an average of 6.37 (SD 5.83) for depression and 6.50 (SD 5.01) for anxiety symptoms (Table 2). Effect sizes (ie, Cohen d) for change from baseline to up to 12 visits were moderate for depression (−0.54, 95% CI −0.55 to −0.54) and anxiety (−0.58, 95% CI −0.58 to −0.57).

Raw mean PHQ-9 and GAD-7 score trajectories over 12 visits are presented in Figure 1. Linear mixed effects model results showed that depression symptoms decreased significantly over the first 5 visits (β=–0.74, P<.001), and to a lesser degree over visits 6‐12 (β=−0.032, P<.001). There was a moderate negative correlation between fixed effect intercept and visit 1‐5 slope (r=−0.56), suggesting that people who start with higher depression symptoms decrease more quickly over the first 5 visits. There was also a small positive correlation between fixed effect intercept and visit 6‐12 slope (r=0.19), indicating that people who start with higher depression decrease more slowly over visits 6‐12.

The main findings were consistent when main effects of demographic covariates were added to the model. Furthermore, male gender (β=−0.72, P<.001), greater age (β=−0.039, P<.001), African American/Black race (β=−0.62, P<.001), and non-Hispanic ethnicity (β=−0.09, P<.001) were associated with lower depression symptom scores. American Indian/Alaska Native (β=0.77, P<.001), Native Hawaiian/Pacific Islander (β=0.87, P<.001), and mixed-race (β=0.37, P<.001) patients had more severe depression symptoms. Primary diagnosis of an anxiety disorder (β=−2.81, P<.001), trauma and stress-related disorder (β=−2.97, P<.001), or other disorder (β=−2.12, P<.001) was associated with lower depression symptom scores than primary diagnosis of a depressive disorder (Table 3).

Linear mixed effects model results showed that anxiety symptom scores decreased significantly over the first 5 visits (β=−0.72, P<.001), and to a lesser degree over visits 6‐12 (β=−0.021, P<.001). There was a moderate negative correlation between fixed effect intercept and visit 1‐5 slope (r=−0.59), suggesting that people who start with higher anxiety decrease more quickly over the first 5 visits. There was also a small positive correlation between fixed effect intercept and visit 6‐12 slope (r=0.20), indicating that people who start with higher anxiety decrease less quickly over visits 6‐12, although this effect is smaller than the correlation for visits 1‐5.

Main findings generally remained consistent when main effects of demographic characteristics were added to the model. The slope for visits 6‐12 was much smaller after accounting for covariates, suggesting that much of the variance in anxiety scores over visits 6‐12 was explained by demographic and diagnosis-related factors. Furthermore, male gender (β=−0.68 P<.001), increased age (β=−0.030, P<.001), and Black/African American race (β=−0.78, P<.001) were associated with lower anxiety symptom scores. American Indian/Alaska Native race (β=0.51, P<.001), and Native Hawaiian/Pacific Islander race (β=0.71, P<.001) were associated with greater anxiety symptom scores. Modal primary diagnoses of an anxiety disorder (β=−0.57, P<.001), trauma and stress disorder (β=−1.41, P<.001), and other disorder (β=−1.02, P<.001) were associated with lower anxiety symptom scores (Table 4).

Results of the models, including interactions between demographic variables and the slopes between visits 1‐5 and visits 6‐12 for both depression and anxiety outcomes, are displayed in Table 5. Only interaction effects are displayed for simplicity. Results revealed that greater age was associated with slower improvement in both depression and anxiety over visits 1‐5. Black/African American race, American Indian/Alaska Native race, and Native Hawaiian/Pacific Islander race were associated with faster improvement in both depression and anxiety over visits 1‐5. Mixed race was associated with faster improvement in depression symptom scores over visits 1‐5, but not in the anxiety. Asian race was associated with slower improvement in anxiety symptoms over visits 1‐5. Non-Hispanic ethnicity was associated with slower improvement in both depression and anxiety over visits 1‐5. A primary diagnosis of an anxiety disorder was associated with slower improvement in depression but faster improvement in anxiety over visits 1‐5. A primary diagnosis of a trauma and stress-related disorder was associated with slower improvement in depression and anxiety, but to a lesser degree for anxiety symptoms. A primary diagnosis of other disorders was associated with slower improvement in both depression and anxiety over visits 1‐5.

Overall, effects for visits 6‐12 were much smaller than those for visits 1‐5. Male gender was associated with a greater decrease in both depression and anxiety symptoms over visits 6‐12. American Indian/Alaska Native race, Asian race, and mixed race were associated with faster improvements in both depression and anxiety over visits 6‐12. Non-Hispanic ethnicity was associated with slower improvements in both depression and anxiety over visits 6‐12. Primary diagnoses of anxiety disorder, trauma and stress disorder, and other disorders were associated with slower improvements in both depression and anxiety over visits 6‐12 compared to a diagnosis of a depressive disorder.

The purpose of this study was to examine the effects of participation in a commercial MBC DMHI, Rula Health, on changes in depression and anxiety over time in both subclinical and clinical patients. We aimed to (1) explore the trajectories of anxiety and depression symptoms and (2) examine the impact of demographics and primary diagnosis on depression and anxiety trajectories. We found that depression and anxiety scores significantly improved over time, with greater improvements in scores over the first 5 visits and slower but continued improvements over visits 6‐12. Greater symptom severity at baseline was associated with greater initial improvements, but this effect diminished after 5 visits. Demographic factors influenced symptom trajectories, with Asian individuals and non-Hispanic individuals experiencing slower improvement. In contrast, younger Black/African American, American Indian/Alaska Native, Native Hawaiian/Pacific Island, and mixed-race individuals experienced faster symptom improvements, especially in visits 1‐5. A primary diagnosis of anxiety or trauma-related disorders was related to slower improvements in depression symptoms but faster improvements in anxiety symptoms compared to those with a primary diagnosis of depression, whereas other diagnoses were associated with slower progress across both measures. Male gender and certain racial groups showed greater symptom reductions during visits 6‐12.

This study represents one of the few evaluations of a commercially available MBC DMHI demonstrating reductions in depression and anxiety symptoms using real-world data from both subclinical and clinical patients in a large sample. We examined over 360,000 patients who completed up to 12 sessions over an average of 81 days, resulting in over 2.5 million therapy visits. Other studies leveraging real-life patient data from digital or in-person practices have less than 1% of the sample size reported here, and none of these report findings in both subclinical and clinical patients [40,41]. Additionally, commercial DMHIs often restrict their samples to clinically significant cases to boost observed treatment effects or meet insurance reimbursement criteria [40,42,43], overlooking a large population experiencing early or mild symptoms who are at heightened risk of progressing to clinical disorders if left untreated. For example, individuals with subclinical anxiety and depression are at increased risk for developing clinical disorders by 2.5 and 3 times, respectively [3,4]. A meta-analysis shows that interventions targeting subclinical populations are among the most cost-effective strategies for reducing depression and anxiety [25]. This inclusive approach enhances the generalizability and clinical relevance of the findings and may have implications for the economic value of scalable DMHIs. More research is needed on the cost-effectiveness of MBC-based DMHIs. Overall, we found depression and anxiety decreased within 12 treatment visits across all patients and the full spectrum of severity. To our knowledge, no other commercial MBC DMHIs have reported comparable outcomes.

Symptoms of depression and anxiety decreased markedly over the first 5 visits of care at the MBC DMHI, and to a lesser degree over visits 6‐12. After experiencing a steep decline in symptoms early in care, Rula Health patients appear to enter a steady growth phase, where their initial progress stabilizes, and they continue improving at a more gradual pace while reducing the risk of setbacks [44,45]. Similar patterns of early symptom reductions have been observed in other studies of DMHIs [46-50], aligning with the idea of “sudden gains” in therapy, where symptoms improve quickly at first, then level out [26]. Notably, these patterns were observed in a sample that included both subclinical and clinical individuals, suggesting that DMHIs like Rula Health may be effective across a broad spectrum of symptom severity. While early improvements may reflect true therapeutic gains, they may also be partially influenced by regression to the mean—a statistical phenomenon in which individuals with extreme baseline scores naturally tend to show improvement over time, independent of treatment effects—which may have contributed to our findings [51]. Given that sudden gains are associated with better short- and long-term outcomes [52,53], future research should explore factors that drive early symptom improvements, such as provider quality and expertise [52,54].

Younger age was associated with more rapid improvements in depression symptoms compared to older individuals. Older adults may experience a slower response to psychotherapy due to factors such as longer illness duration, differences in cognitive flexibility, or variations in engagement with DMHIs [55]. Gender differences also emerged, with male patients showing greater symptom reductions later in treatment (visits 6‐12). Prior studies suggest that men often delay help-seeking behaviors and may require more time in treatment, as they are less likely to disclose emotional distress and engage with therapy [56]. In contrast, women tend to show earlier symptom reductions, likely due to higher baseline engagement, greater willingness to disclose distress, and a stronger tendency to seek formal psychological support [57]. These findings highlight the need to consider demographic factors in DMHIs, as age and gender influence symptom trajectories and treatment response.

At the start of care, patients using Rula Health’s platform are matched with providers based on stated preferences, including race and ethnicity. This approach reduces patient burden in navigating the provider search process and improves access to culturally responsive care. Observed racial and ethnic differences in symptom trajectories suggest that this MBC-based DMHI may support early symptom reduction among racial and ethnic minority populations, who have historically faced structural barriers to accessing mental health services [58-60]. Black/African American, American Indian/Alaska Native, Native Hawaiian/Pacific Island, and mixed-race individuals experienced faster symptom improvements, particularly within the first 5 visits. The rapid improvements observed in these groups reflect Rula Health’s commitment to providing accessible, high-quality care that meets the needs of diverse patients. Minority groups often enter treatment with greater distress due to systemic barriers to care, which may contribute to a steeper initial recovery trajectory once treatment begins [58-60]. We also found slower symptom reductions among Asian and non-Hispanic individuals, which may be influenced by cultural differences in symptom expression, help-seeking behaviors, or engagement with therapy, as some populations experience higher mental health stigma or prefer alternative coping strategies [61-63]. Future research should also explore how DMHIs, like Rula Health, can further tailor treatment approaches to ensure equitable outcomes across all racial and ethnic backgrounds.

Symptom improvements were most pronounced within their respective diagnostic categories, which aligns with the MBC model leveraged by this DMHI (Rula Health). Individuals with a primary depressive disorder experienced the fastest reduction in depression symptoms, while those with a primary anxiety disorder saw the quickest alleviation of anxiety symptoms. Unlike traditional therapy models, Rula Health’s MBC model ensures ongoing symptom tracking, supporting therapists to adapt treatment to shifting symptom trajectories. Most commercial DMHIs that integrate MBC typically track symptoms every 3 weeks or monthly—less frequently than Rula Health’s approach [42,64]. The MBC model used by Rula Health enables a responsive, data-driven approach that systematically tracks patient-reported outcomes to inform clinical decision-making. This framework facilitates early symptom detection and supports continued improvement across diagnostic categories over the course of up to 12 visits. Later in treatment, individuals with a primary diagnosis of depression improved faster in both depression and anxiety symptoms. Disorder-specific treatments may yield the fastest results in their target symptoms, while cross-diagnostic benefits, even in conditions that tend to be highly comorbid, such as anxiety, depressive, and trauma-related disorders, emerge more gradually. This slower cross-diagnostic improvement may reflect the additional cognitive and emotional processing required to generalize therapeutic gains across disorders [65,66]. Clinically, these results emphasize the importance of maintaining treatment beyond initial symptom relief to maximize broad-spectrum recovery [45,67], an approach that, to our knowledge, is more consistently implemented by Rula Health than by other commercial DMHIs.

This study has several strengths. First, it uses a large and diverse sample of patients receiving care through Rula Health’s MBC DMHI, allowing for insights that are generalizable across various demographic and diagnostic groups. Unlike studies relying on highly controlled clinical trials, this research examines real-world clinical data, making the findings more applicable to actual mental health care settings. The longitudinal tracking of symptom changes over 12 sessions provides a detailed picture of how depression and anxiety symptoms evolve over time, offering a nuanced understanding of both rapid early improvements and slower later-stage progress. By identifying key moderators of symptom improvement, such as baseline severity, demographics, and primary diagnoses, it contributes to a more nuanced understanding of variability in treatment response in DMHIs. This study adds to the body of research on DMHIs by investigating how depression and anxiety symptoms evolve over time in Rula Health, an MBC-integrated DMHI, emphasizing the influence of demographic and clinical factors on treatment response.

There are limitations that should be acknowledged. As an observational, non-randomized study, causal conclusions about treatment effectiveness are limited, and future research should incorporate randomized controlled trials to better assess treatment efficacy. Additionally, treatment engagement metrics (such as adherence to sessions, completion rates, and active participation) were not measured, which limited our ability to assess why symptoms improved over time. This study also lacks other variables that could have impacted treatment, such as social support, therapy modality (eg, cognitive behavioral therapy vs acceptance and commitment therapy), comorbid diagnoses, or life stressors. Future research should include these factors to better elucidate the mechanisms underlying treatment response and address these important knowledge gaps. Lastly, post-treatment follow-up data were not collected, preventing the evaluation of whether symptom improvements were maintained after care ended. Future studies should incorporate longitudinal follow-up to assess the durability of treatment effects over time.

This study provides valuable insights into how depression and anxiety symptoms evolve over time in patients receiving care through Rula Health, an MBC-integrated DMHI. We found that attending more therapy sessions was associated with greater symptom improvement. Symptom trajectories revealed that the most rapid improvements occurred within the first 5 visits, followed by a more gradual decline from visits 6‐12. Baseline severity, demographics, and primary diagnosis also influenced symptom trajectories, highlighting the various factors that shape treatment outcomes in mental health care. Our study highlights the impact of MBC DMHIs on clinical outcomes, providing evidence for the effectiveness of a digitally delivered MBC model in a large real-world patient sample.