This study used a real-world, retrospective cohort design leveraging deidentified electronic health record data from Grow Therapy, a technology-enabled mental health platform, reported in accordance with STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidance for observational studies [25]. Adults aged 18 years or older with a first provider visit between January 1, 2022, and April 30, 2024, were eligible for inclusion. Patients were included in the intent-to-treat cohorts if they had clinically elevated symptoms at intake (PHQ-9 [Patient Health Questionnaire-9 item]>9 for the depressive symptom cohort; GAD-7 [Generalized Anxiety Disorder]>9 for the anxiety symptom cohort). Clinical outcome analyses used complete case cohorts requiring at least 2 completed PHQ-9 or GAD-7 assessments. Sample size was determined by the available population meeting these criteria, consistent with retrospective observational designs. Data included patient demographics, preferred provider characteristics selected during the matching process (specialty and identity attributes), number of completed visits, and routinely collected symptom assessments. All symptom data originated from standardized, validated electronic instruments integrated into the platform and administered at intake and approximately every third visit. No new data were collected for research purposes. Participants self-selected providers via a predefined interface, and all data used in this study were originally collected as part of routine clinical care.

This study was reviewed by Western Institutional Review Board Copernicus Group and deemed exempt under 45 Code of Federal Regulations §46.104(d)(4) (Confirmation ID 45689720), which covers secondary research using identifiable private information or biospecimens when data are recorded in a manner that no longer allows subjects to be identified. All individuals agreed to Grow Therapy’s privacy policy and informed consent procedures at intake, which permitted the use of deidentified data for research purposes. As this study involved secondary analyses of deidentified data, additional participant consent was not required. All data were extracted in deidentified form, with direct identifiers removed before analysis. Data were stored in HIPAA (Health Insurance Portability and Accountability Act)-compliant secure databases, and only aggregated results are reported. No compensation was provided to individuals whose clinical data were included in this secondary analysis. No images or materials containing individually identifiable patient information are included in this paper or supplementary materials. All analyses were conducted on deidentified data, which precluded the identification of individual participants.

Data on all adult patients (18 years or older) with a first Grow Therapy visit between January 1, 2022, and April 30, 2024, were included in this study’s sample, which was further broken down into 2 cohorts. The first cohort consisted of patients who had elevated levels of depressive symptoms, and the second cohort consisted of patients who had elevated levels of anxiety symptoms as described below. Data included preferred provider characteristics endorsed by the patient when seeking to find a new provider affiliated with Grow Therapy and deidentified clinical data collected as part of treatment engagement with a provider, as well as the total number of visits between the patient and provider. Data collected through April 30, 2025, were included in analyses to allow all patients to have at least 12 full months of follow-up.

Statistical analysis was performed in May 2025. Frequency counts and percentages are reported for all categorical data, and means and SDs are reported for continuous data. Bivariate (unadjusted) associations were examined via chi-square tests to determine differences between provider characteristics selected by patients and outcomes of interest, including number of visits dichotomized into 3 or more provider visits (yes or no) in the intent-to-treat populations and change in symptom of depression and anxiety scores on the PHQ-9 and the GAD-7 quantified as having an MCID (yes or no), respectively, in the complete case populations (subsets of the intent-to-treat populations with at least 2 completed PHQ-9 or GAD-7 assessments, respectively) using chi-square tests. Logistic regression models to test associations between selected provider characteristic and each outcome of interest were run adjusted for baseline demographic (age and gender) and clinical characteristics (baseline depression or anxiety symptom scores) based on theory, prior research findings, and significant variables identified in analyses that compared those in the intent-to-treat populations who were, vs those who were not, included in the complete case populations. The primary outcome of interest was having at least 3 provider visits in each intent-to-treat population (depressive symptom and anxiety symptom cohorts), and the secondary outcomes included achieving a depressive and anxiety symptom MCID in the depressive symptom complete case cohort and anxiety symptom complete case cohort, respectively. To address the potential for selection bias, sensitivity analyses were performed using the last observation carried forward method for missing clinical outcome data. Logistic regression models were expressed as odds ratios (ORs) and 95% CI of each OR. A 2-sided P≤.05 was considered indicative of statistical significance. All analyses were performed using R software (version 4.2; R Project for Statistical Computing).

The intent-to-treat depressive symptom and anxiety symptom cohorts used for engagement and sensitivity analyses consisted of 159,448 and 167,356 adult Grow Therapy patients aged 18 years or older, respectively, with elevated symptoms at baseline (PHQ-9>9 and GAD-7>9, respectively) who had a first provider visit between January 2022 and April 2024. There was a sizable overlap in the 2 intent-to-treat cohorts, with 123,284/159,448 (77.3%) of the clinically elevated depressive symptom intent-to-treat cohort also having clinically elevated anxiety symptoms, and 123,284/167,356 (73.7%) of the clinically elevated anxiety symptom intent-to-treat cohort also having clinically elevated depressive symptoms at baseline. The complete case samples used for clinical symptom analyses required at least 2 PHQ-9 or 2 GAD-7 assessments to enable the computation of a change score and included 72,008 patients in the depressive symptom cohort (72,008/159,448, 45.2% of the corresponding intent-to-treat sample) and 74,176 patients in the anxiety symptom cohort (74,176/167,356, 44.3% of the corresponding intent-to-treat sample). Again, there was a sizable overlap in the 2 intent-to-treat cohorts, with 54,039/72,008 (75%) of the clinically elevated depressive symptom complete case cohort also having clinically elevated anxiety symptoms and 54,039/74,176 (72.9%) of the clinically elevated anxiety symptom complete case cohort also having clinically elevated depressive symptoms at baseline (Figure 1).

In the intent-to-treat cohorts, females comprised the majority of patients (about 75% in each cohort), and patients had a mean age of 33.3 (SD 11.2) years and 33.0 (SD 10.8) years in the depressive and anxiety symptom cohorts, respectively. Provider specialty was selected by approximately 35% of both cohorts, and about 5% of both cohorts selected a provider identity before the first visit (Table 1). Those in the complete case sample were overrepresented by females, had a higher mean age, and had slightly more severe depressive symptoms but slightly less severe anxiety symptoms at baseline, although the differences were not clinically significant (depressive symptom cohort completers mean baseline PHQ-9 of 15.2, SD 4.1, vs noncompleters mean baseline PHQ-9 of 15.3, SD 4.2, and anxiety symptom cohort completers mean baseline GAD-7 of 14.9, SD 3.4, vs noncompleters mean baseline GAD-7 of 15.0, SD 3.4, P<.001 for each comparison).

The intent-to-treat samples included all adult Grow Therapy patients aged 18 years or older with a first provider visit in January 2022 through April 2024 who had clinically elevated depressive symptoms at baseline (PHQ-9 of 10-27; baseline depressive symptom cohort) or who had clinically elevated anxiety symptoms at baseline (GAD-7 of 10-21; baseline anxiety symptom cohort) as indicated.

A total of 123,284/159,448 (77.3%) of the clinically elevated depressive symptom intent-to-treat cohort also had clinically elevated anxiety symptoms and are included in both intent-to-treat cohorts.

A total of 123,284/167,356 (73.7%) of the clinically elevated anxiety symptom intent-to-treat cohort also had clinically elevated depressive symptoms at baseline and are included in both intent-to-treat cohorts.

The complete case samples included patients in each intent-to-treat sample who also had at least 2 completed PHQ-9 (baseline depressive symptom cohort) or GAD-7 assessments (baseline anxiety symptom cohort).

A total of 54,039/72,008 (75%) of the clinically elevated depressive symptom complete case cohort also had clinically elevated anxiety symptoms and are included in both complete case cohorts.

A total of 54,039/74,176 (72.9%) of the clinically elevated anxiety symptom complete case cohort also had clinically elevated depressive symptoms at baseline and are included in both complete case cohorts.

The mean number of provider visits in the intent-to-treat depressive and anxiety symptom cohorts was 11.4 (SD 16.0) and 11.3 (SD 15.7), respectively. In the depressive symptom cohort, 69.4% had at least 3 provider visits, and 28.7% had at least 12 provider visits. In the anxiety symptom cohort, 69.3% had at least 3 provider visits, and 28.6% had at least 12 provider visits. As expected, the complete case cohorts had higher levels of engagement metrics than the intent-to-treat cohorts (Table 2).

The intent-to-treat samples included all adult Grow Therapy patients aged 18 years or older with a first provider visit in January 2022 through April 2024 who had clinically elevated depressive symptoms at baseline (PHQ-9 of 10-27; baseline depressive symptom cohort) or who had clinically elevated anxiety symptoms at baseline (GAD-7 of 10-21; baseline anxiety symptom cohort) as indicated.

A total of 123,284/159,448 (77.3%) of the clinically elevated depressive symptom intent-to-treat cohort also had clinically elevated anxiety symptoms and are included in both intent-to-treat cohorts.

A total of 123,284/167,356 (73.7%) of the clinically elevated anxiety symptom intent-to-treat cohort also had clinically elevated depressive symptoms at baseline and are included in both intent-to-treat cohorts.

The complete case samples included patients in each intent-to-treat sample who also had at least 2 completed PHQ-9 (baseline depressive symptom cohort) or GAD-7 assessments (baseline anxiety symptom cohort).

A total of 54,039/72,008 (75%) of the clinically elevated depressive symptom complete case cohort also had clinically elevated anxiety symptoms and are included in both complete case cohorts.

A total of 54,039/74,176 (72.9%) of the clinically elevated anxiety symptom complete case cohort also had clinically elevated depressive symptoms at baseline and are included in both complete case cohorts.

In the complete case depressive symptom cohort, mean PHQ-9 declined from 15.2 (SD 4.1) to 9.1 (SD 6.1) from baseline to last assessment (average change of 6.1, SD 6.1, points; Table 2). In this cohort, 58.9% (95% CI 58.5%-59.2%) of patients had a PHQ-9 depressive symptom MCID. In the complete case anxiety symptom cohort, mean GAD-7 declined from 14.9 (SD 3.4) to 9.1 (SD 5.7) from baseline to last assessment (average change of 5.8, SD 5.7, points). In this cohort, 63% (95% CI 62.6%-63.3%) of patients had a GAD-7 anxiety symptom MCID. We did not use multiple imputation (MI) because greater than 50% of eligible adults had missing outcomes data, which also was not missing at random. Prior research indicates MI is unreliable when missingness exceeds 30% and the missing-at-random assumption is violated [32,33].

Unadjusted chi-square and adjusted logistic regression models indicated no significant associations between provider specialty selection and having at least 3 provider visits in both the depressive and anxiety symptom cohorts (Table 3). Patients in the depressive symptom cohort who did select a provider specialty, however, were about 8% more likely to have a depressive symptom MCID than those who did not (59.9% vs 58.3%, OR 1.08, 95% CI 1.04-1.11). Similarly, patients in the anxiety symptom cohort who selected a provider specialty were about 4% more likely to have an anxiety symptom MCID than those who did not (63.4% vs 62.7%, OR 1.04, 95% CI 1.01-1.07), although this association was only significant at P<.05 in the adjusted regression analysis (P<.001) and not the chi-square unadjusted analysis (P=.07).

The intent-to-treat samples used for the engagement models (models 1 and 3) included all adult Grow Therapy patients aged 18 years or older with a first provider visit in January 2022 through April 2024 who had clinically elevated depressive symptoms at baseline (PHQ-9 of 10-27; baseline depressive symptom cohort) or who had clinically elevated anxiety symptoms at baseline (GAD-7 of 10-21; baseline anxiety symptom cohort) as indicated.

A total of 123,284/159,448 (77.3%) of the clinically elevated depressive symptom intent-to-treat cohort also had clinically elevated anxiety symptoms and are included in both intent-to-treat cohorts.

A total of 123,284/167,356 (73.7%) of the clinically elevated anxiety symptom intent-to-treat cohort also had clinically elevated depressive symptoms at baseline and are included in both intent-to-treat cohorts.

The complete case samples used for the clinical symptom models (models 2, 3, 5, and 6) included patients in each intent-to-treat sample who also had at least 2 completed PHQ-9 (baseline depressive symptom cohort) or GAD-7 assessments (baseline anxiety symptom cohort).

A total of 54,039/72,008 (75%) of the clinically elevated depressive symptom complete case cohort also had clinically elevated anxiety symptoms and are included in both complete case cohorts.

A total of 54,039/74,176 (72.9%) of the clinically elevated anxiety symptom complete case cohort also had clinically elevated depressive symptoms at baseline and are included in both complete case cohorts.

Models adjusted for patient age (all models), gender (all models), baseline depression symptoms (all models), baseline anxiety symptoms (all models), and number of visits (models 2 and 4).

Unadjusted comparisons were done via chi-square tests.

Unadjusted chi-square and adjusted logistic regression models indicated that patients in both the depressive and anxiety symptom cohorts who selected a provider identity characteristic before the first visit were 14% and 11% more likely, respectively, to have at least 3 provider visits (72.1% vs 69.3%, OR 1.14, 95% CI 1.08-1.20, in the depressive symptom cohort and 71.6% vs 69.2%, OR 1.11, 95% CI 1.06-1.17) as compared to patients who did not select a provider identity characteristics. There were no significant unadjusted or adjusted associations, however, between the selection of a provider identity characteristic and having a depressive or anxiety symptom MCID in either cohort (Table 3).

Sensitivity analyses using the last observation carried forward for clinical outcomes in the intent-to-treat cohorts confirmed our main findings with a few exceptions. In unadjusted analyses, patients who selected both specialty and identity provider characteristics before enrollment were more likely to have a depressive symptom MCID in the depressive symptom cohort and an anxiety symptom MCID in the anxiety symptom cohort (P<.01 in each instance). These findings differed from the unadjusted main analyses which just found significant unadjusted associations between choosing the specialty of the provider and depressive symptom MCID in the depressive symptom cohort. The significance of the adjusted analyses, however, remained the same, with only those who selected a provider specialty, but not a provider identity, characteristic having a significantly greater likelihood of having a depressive symptom and an anxiety symptom MCID in the depressive symptom and anxiety symptom cohorts, respectively.

In this study, a little over a third of adult patients in each cohort selected a provider specialty, and about 5% of each cohort selected a provider identity characteristic. Secondary, exploratory analyses, however, indicated that a patient’s self-selection of each provider characteristic option was associated with engagement or clinical outcomes (see Table 4 for a summary of significant findings). These findings suggest the potential value of a patient self-selecting their desired provider characteristics in order to improve their therapeutic experience and subsequent outcomes. In this study, patients who selected a provider specialty were more likely to have a change in symptoms meeting the MCID criteria in both the depressive and anxiety symptom cohorts. These findings, while preliminary and exploratory, suggest that patients who knew their presenting issues and requested targeted help from providers might have been able to experience significantly better clinical outcomes by selecting providers based on self-awareness of their treatment needs. Furthermore, patients who selected a provider identity characteristic were more likely to have at least 3 provider visits than those who did not. Selecting a provider identity characteristic was not, however, associated with clinical outcomes in either the depressive or anxiety symptom cohort. Future studies with more rigorous designs that include an appropriate control group should seek to understand the impact of extended treatment beyond PHQ-9 and GAD-7 measures, including quantifying improvements in quality of life and the role of therapeutic alliance in the observed associations.

This is the first study of its kind to report engagement and clinical outcomes of a technology-enabled platform where patients have the choice and autonomy to select characteristics of the mental health provider treating them. Likewise, it is the first study to begin to explore the ability of patients to select provider characteristics as related to the levels of engagement and clinical outcomes found. Despite prior literature that has suggested the importance of patient choice in facilitating beneficial outcomes [18], little is known about this topic specifically related to technology-enabled platforms that provide this option.

Nonetheless, prior work examining digital mental health interventions has consistently noted low engagement rates and only modest levels of efficacy [6,7,10]. Levels of patient engagement in mental health care found in the current study, however, were quite high in comparison to even the prior recommendations from systematic reviews reported in the literature; for example, despite lower levels of actual engagement, 1 meta-analysis found that a typical internet-based CBT program recommends 5-9 visits per patient [34]. The present study’s findings are notable in this context as the proportion of patients engaging in Grow Therapy care with at least 3 provider visits (≈69%) and meeting depressive and anxiety symptom MCID criteria (≈59%-63%) is similar or higher than those reported in typical structured digital CBT programs that often have high rates of drop-out. In general, these proportions of those meeting criteria for an MCID of each outcome appeared to be similar or slightly higher than interventions previously described in the literature, especially supported digital mental health interventions delivered over the internet or via an app [23,35]. It is important to note, however, that the present analyses were conducted on a real-world sample with at least enough provider visits to capture at least 2 relevant outcome assessments, and no control group was included to enable group-level comparisons. Additional studies using more rigorous designs, such as a pragmatic RCT or using a real-world synthetic control group, that also adjust for other potential confounders of the explored associations (eg, personality characteristics, education, or other factors related to both the exposure and outcome of interest), are needed to confirm these preliminary findings.

Most previously published work has focused on naturally occurring concordance rather than patient selections of provider attributes, so our findings might not generalize to these instances. The current study examines behaviors somewhat unique to digital settings where patients can intentionally select provider characteristics unconstrained by in-person geographic location or appointment availability. This type of investigation is novel as compared to earlier concordance studies that often relied on clinic assignment, administrative matching, or geographic convenience instead of the ability to freely choose provider characteristics [11,12]. A growing body of research, however, has examined hybrid or digital care systems that have used performance-based or algorithmic matching; for example, Constantino et al [13] conducted an RCT and determined that matching patients to providers based on therapist strengths improved outcomes. The present study differs because patients could choose preferred provider characteristics, some of which they might have thought were beneficial to their care, rather than an algorithmic assignment. This distinction is important, as choice-based matching may promote early therapeutic alliance, autonomy, and expectancy effects, each of which has been linked to improved engagement and treatment success [26,36,37].

Our findings also extend recent work on digital personalization frameworks that have suggested that the inclusion of tailored content, therapist profiles, and culturally responsive design may bolster outcomes when integrated into digital mental health services [38]. While personalization in previous studies has typically involved system-driven tailoring, provider selection in this study reflects a patient-driven personalization strategy, representing an underexamined but potentially scalable approach.

Finally, few large-scale naturalistic studies have examined associations between provider-selection behaviors and clinical outcomes in digital care settings at a real-world scale. This study’s sample size exceeds most published digital mental health cohorts’ sizes and provides one of the strongest empirical demonstrations to date of how patient choice operates within commercially deployed teletherapy platforms. These findings thus contribute new evidence that scalable matching systems, when embedded within a flexible platform, may relate meaningfully to both engagement and symptom change.

Although this study’s design limits definitive conclusions, the exploratory findings provide preliminary evidence that a web-based platform that promotes patient selection of their provider could contribute to patient engagement and favorable clinical outcomes. Specifically, selecting a provider specialty was associated with significantly achieving MCID-level improvements in depressive and anxiety symptoms. Thus, the findings suggest that encouraging patients to select a provider specialty may positively impact clinical outcomes by ensuring a more targeted fit to the expertise of the provider. In addition, selecting a provider identity characteristic was associated with having 3 or more provider visits. This pattern aligns with theoretical models proposing that autonomy, personal relevance, and therapeutic fit enhance treatment continuation [26,36,38]. The significant association between selecting a provider identity and increased patient engagement may have resulted from patients feeling more comfortable from the onset of the patient-provider relationship due to the therapeutic alliance formed between the patient and provider, promoting cultural safety and trust [11,39]. It is somewhat surprising, however, that we did not find a significant association between the selection of provider identity and changes in clinical outcomes, potentially because of the small sample size of those selecting a provider identity or perhaps due to the current study’s design itself.

Additional studies that include control groups of patients who do not select provider characteristics and that adjust for more potential confounders would lead to a greater understanding of the preliminary signals found by these exploratory analyses. These findings suggest that scalable, technology-enabled patient-provider matching systems, particularly those that allow patients to filter by provider specialty or identity, may offer a feasible way to improve engagement and outcomes at scale. As platforms such as Grow Therapy expand, this approach could be further developed to support systems–level innovation in care delivery, particularly by personalizing access in ways that are difficult to replicate in traditional models of mental health referral and triage.

Although causality cannot be inferred, these findings underscore opportunities for designing more effective digital matching interfaces. For instance, platforms may consider highlighting evidence-based specialty tags, guiding patients toward providers with relevant expertise, or using A/B testing to evaluate which provider profiles maximize engagement. Similar approaches are increasingly used in digital triage and hybrid care systems [14,40].

In addition to the aforementioned need for more rigorous study designs to confirm these initial findings, future studies, such as pragmatic randomized or quasi-experimental studies, could also explore other potential reasons for the levels of engagement and clinical outcome success experienced by patients included in this study. For example, additional work is needed to explore potential reasons for the levels of engagement and clinical outcomes other than the selection of provider characteristics of specialty and identity. Further research is also needed to ensure that levels of outcomes are similar across different patient subgroups other than those defined by age, gender, and baseline symptomatology. These groups could be created based on social determinants of health or other vulnerable populations to ensure the equitable delivery, receipt, and experience of mental health care.

Additional work should also examine the mechanisms by which provider-selection behaviors and symptom changes are associated, perhaps via improved engagement or therapeutic alliance. Prior studies suggest that early engagement and therapeutic alliance can shape downstream clinical trajectories [28]. Research should also explore more granular subtypes of specialty and identity selection, such as having a culturally aligned provider, and whether this impacts findings. Greater attention to equity-centered matching is critical, especially as digital platforms expand access to diverse populations [23]. The current findings suggest the potential benefit of such alignment, but small sample sizes limited its evaluation. More robust investigations with larger sample sizes are needed to confirm these initial findings and to further explore more nuanced specialty and identity selections (eg, specific specialties or identity characteristics) in relation to longer-term engagement and outcomes.

This study has several limitations. First, our clinical outcome analyses relied on complete case data rather than using MI, which introduces potential selection bias. We deemed MI inappropriate because of substantial missing data (>50%) and nonrandom missingness after considering the evidence that MI can be invalid under these conditions [32,33]. Thus, clinical outcome findings may be generalizable only to patients who completed at least 2 PHQ-9 or GAD-7 assessments. Sensitivity analyses using last observation carried forward, however, yielded similar adjusted findings, increasing confidence in the robustness of observed associations. As reasons for missingness were not captured, we could not determine whether those lost to follow-up were dissatisfied with their treatment, realized early symptom improvement so did not “need” treatment anymore (“happy abandonment” [41]), or if some other, unobserved factors were in play.

Second, due to limited sample sizes for hundreds of possible provider specialty and identity combinations, this study examined only binary selections rather than specific match types (eg, trauma specialty; Black/African American identity). Future research with larger subgroup samples may explore more granular associations. Additional limitations include reliance on observational data without a control group, lack of adjustment for unmeasured confounders such as socioeconomic status or provider experience, and potential overlap between depressive and anxiety symptom cohorts. We also did not examine how engagement and outcomes relate to each other mechanistically; future studies should test whether engagement mediates clinical change. Additional studies to determine the amount and types of engagement needed to achieve beneficial outcomes, including whether engagement mediates the associations with clinical outcomes, will provide a better understanding of the findings of this study. Finally, because analyses did not adjust for multiple comparisons, findings should be interpreted as exploratory.

Third, a majority of the sample was in both elevated baseline symptoms of depression and anxiety cohorts, given the high prevalence of comorbid baseline symptoms, rendering the engagement analyses somewhat duplicative. The reported regression analyses, however, did report associations adjusted for both baseline depressive and anxiety symptoms. We did not, however, have access to other potentially confounding variables such as socioeconomic status, education, or provider experience to further adjust the models. These factors should be included in future research. Additionally, we did not adjust the analyses for multiple comparisons, given the exploratory nature of this study. Most reported significant associations, however, were at the P<.001 level of significance that would have remained significant even after adjustment.

Fourth, although this study examined the patient-driven selection of providers, the Grow Therapy platform’s interface, available provider pool, and algorithmic ordering of provider profiles may influence what options users see and select. These platform-level design factors were not examined but may partially explain patterns of selection and should be considered in future research [42].

Fifth, this study did not evaluate therapist characteristics beyond those made available to patients for selection (eg, provider experience, caseload, and therapeutic fidelity). Prior work suggests that therapist effects can meaningfully influence outcomes and unmeasured provider-level variability may confound associations [43].

Finally, measures of patient preference strength (eg, how strongly a patient valued a specific attribute) were not collected. Prior meta-analytic work suggests that preference strength, rather than preference presence alone, may predict outcomes [17].

More recent quality improvement initiatives targeting measurement-informed care, which promotes the completion and use of routine outcome monitoring as part of the treatment relationship between a provider and a patient [44], have improved clinical symptom measure completion rates at Grow Therapy since the sample’s inclusion of earlier data periods (2022-2024). Future outcome analyses that include these more robust samples will thus serve to enhance the generalizability of findings, as would the use of a study design that includes an appropriate control group.

In this large-scale real-world analysis, patients who selected provider characteristics before initiating care demonstrated meaningful differences in both engagement and clinical outcomes. While exploratory, these findings contribute new evidence that patient autonomy in selecting provider attributes, a feature uniquely enabled by digital mental health platforms, may play an important role in shaping therapeutic trajectories. This is likely the first real-world study of the associations between different types of provider selections and engagement and clinical symptom changes in a digital mental health platform for patients with elevated symptoms of depression or anxiety. Further research using more rigorous controlled designs is needed to confirm these preliminary findings and better understand how types of and more nuanced provider selection variables, such as specialty (eg, depression, anxiety, trauma, etc) and identity (Black/African American, Muslim, etc), are associated with engagement or clinical symptom outcomes. As the field continues to evolve toward more personalized models of digital care [19], understanding how patients navigate matching tools and how these behaviors relate to outcomes will be critical for developing scalable, equitable, and effective mental health services.