Psychotherapy Disruption Before and After the Transition to Virtual Mental Health Care Induced by the COVID-19 Pandemic

This study was conducted within the Mental Health Research Network (MHRN). The three health systems participating in this study were HealthPartners (Minnesota), Henry Ford Health (Michigan), and the Washington State region of Kaiser Permanente. Collectively, these systems serve over 3 million patients per year across five U.S. states. The institutional review boards at each participating system approved data use for this study.

This study included 110,089 unique patients who had a psychiatric diagnosis, attended at least two psychotherapy visits during the study period (before or after onset of the COVID-19 pandemic, defined as March 14, 2020), and were members of one of the participating health systems’ affiliated health plans for at least 10 months per year in 2019 and 2020. A total of 555,620 visits were captured in the 9-month period before COVID-19 (starting June 14, 2019) and in the 9-month period after COVID-19 onset (ending December 15, 2020).

Data were captured from the MHRN virtual data warehouse (VDW) at each participating health system, which is used to facilitate data consistency and sharing across participating sites (19). The VDW is a federated data model that combines electronic health record and insurance claims data by using the same data definitions and variables across sites. It consists of data on patients’ sociodemographic characteristics, diagnoses, procedures, visit types, and medications and of other administrative data recorded during clinical visits. Psychiatric diagnoses were defined by using standard ICD-10 diagnosis codes (F01–F99) (20). These diagnoses were subdivided into the following types: attention-deficit hyperactivity disorder (ADHD), anxiety disorder, autism spectrum disorder, bipolar spectrum disorder, conduct or disruptive disorder, dementia, depressive disorder, personality disorder, schizophrenia spectrum disorder, other psychosis, and substance use disorder. These categories of mental health conditions were not mutually exclusive, so individuals could be represented in multiple categories. Patient demographic characteristics, including age, race-ethnicity, and gender, were recorded during clinical visits. In addition, for each patient, census block data were used to create the geocoded variables of estimated income (“low income” was indicated when individuals resided in a census block where >20% of residents had an income below $40,000 per year), estimated education (“low education” was indicated when the average education in a neighborhood was less than a 4-year college degree), and geographic location (urban or suburban vs. rural).

Psychotherapy visits were defined as any visit greater than or equal to 30 minutes with a specialty mental health provider and with any CPT procedure code of 90785–90862, indicating either initial psychotherapy evaluation or individual psychotherapy (21). Virtual care was identified by using CPT codes with a GQ modifier for asynchronous telehealth video visits or with a GT modifier for synchronous and interactive telehealth video visits, or was otherwise designated as a telehealth video or telephone visit via local health system records. Disruptions in receipt of psychotherapy were measured via a gap of >45 days between visits for each individual—a definition that has been used in prior research (22, 23).

Statistical analysis began with calculating descriptive statistics for the demographic characteristics and psychiatric diagnoses of patients in the sample. Unique psychotherapy visits were stratified by period (before and after onset of COVID-19) and summarized in terms of psychotherapy disruption (yes or no: whether a gap of >45 days occurred between visits) and encounter type (in person vs. virtual). Mean and median lengths of time between visits were computed overall and within periods. To assess whether disruption disproportionately affected specific subgroups of patients, adjusted logistic mixed-effects models were used to examine patient- and visit-level factors associated with the dichotomous outcome of psychotherapy disruption during the observation period (June 2019–December 2020). Model covariates included period (before and after onset of COVID-19), number of previous psychotherapy visits, demographic characteristics (age, race-ethnicity, gender, geographic location, estimated income, and estimated education), and psychiatric diagnoses. The correlation of repeated visits with prior visits by one individual was accounted for with a patient-specific random intercept. For logistic mixed-effects models, beta coefficients and standard errors were calculated, along with adjusted odds ratios, 95% confidence intervals, and p values. Finally, interaction terms were added to the models to examine how the association of period with psychotherapy disruption varied by sociodemographic characteristics and psychiatric diagnoses. Each interaction effect was tested in a separate model, where each interaction term was added to the base model with visit factors and demographic characteristics. The marginal effect of each demographic characteristic and psychiatric diagnosis was estimated by using partial derivatives from the interaction terms to describe the effect of COVID-19 on the length between visits. Statistical significance was defined as p<0.05. Analyses were conducted with SAS, version 9.4.

The overall population sample included 110,089 unique individuals with at least one psychiatric diagnosis who made 555,620 psychotherapy visits at three health systems during the study period (9 months before and 9 months after COVID-19 onset). More total visits occurred during the preonset period (N=284,826) compared with the postonset period (N=270,794). The mean age of the sample was 46.6 years (median=46 years). The sample was 68.9% female, and 67.9% of patients identified as White. Nearly 8% of the sample lived in rural areas. Approximately 5% of the sample lived in neighborhoods with an estimated per-resident annual income of less than $40,000 per year, and 25.5% lived in neighborhoods whose residents had an estimated average education lower than a 4-year college degree. The most frequently diagnosed mental health conditions within the sample were anxiety (72.7%) and depressive (65.3%) disorders. Conduct or disruptive disorders were least common in the sample (0.6%) (Table 1).

As shown in Table 2, among the total number of visits occurring during the 18-month observation period, most occurred before the onset of COVID-19. Approximately one-third (35.4%) of all psychotherapy visits in the initial 9-month period were followed by a disruption in psychotherapy receipt of >45 days, whereas only 17.9% of all visits in the 9-month period after COVID-19 onset were followed by such a disruption. Only 3.1% of all visits recorded during the preonset period were virtual visits, whereas 51.8% of visits were virtual in the postonset period. Furthermore, the median time between visits in the preonset period was 27 days, whereas in the postonset period it was 14 days. The differences in the composition of virtual visits and length between visits were both statistically significant (p<0.001).

During the entire observation period (Table 3), male patients and those identifying as Asian, Hispanic, or another race or ethnicity were less likely to have a disruption between psychotherapy visits. Older individuals and those living in rural areas were more likely to have a disruption. Individuals with depressive, anxiety, or bipolar disorders were less likely to have a disruption, whereas those with schizophrenia, ADHD, autism, conduct or disruptive disorders, dementia, or personality disorders were more likely to have a disruption. After adjustment for visit, demographic, and mental health factors, individuals were less likely to have disruptions in psychotherapy receipt after (vs. before) the onset of COVID-19 and the widespread switch to virtual care (adjusted OR=0.45).

Because the sociodemographic main effects were statistically significant, we then added interaction terms to assess for the interaction of each sociodemographic factor with each period. Although the estimates of the interaction terms were not significant at the level of α=0.05, the marginal effects of period on every assessed categorical sociodemographic group were calculated (adjusted OR=0.64–0.70) (Table 4). In addition, the interaction between each psychiatric diagnosis and each period was assessed separately. Varying magnitudes of disruption were observed across different mental health conditions when comparing the periods before and after the onset of the COVID-19 pandemic. Patients with any mental health condition were less likely to have disruptions in psychotherapy receipt after COVID-19 onset and the accompanying rapid shift to virtual care, compared with before COVID-19 onset (Figure 1 and the online supplement to this article).

This study represents the largest population-level examination of psychotherapy treatment disruption in the United States and compared disruptions during the 9 months before the onset of COVID-19 with those occurring after COVID-19 onset, when a rapid shift from nearly all in-person visits to mostly virtual visits was observed. Consistent with other large U.S. health systems, the health systems participating in this study demonstrated sizable and rapid shifts from in-person to virtual mental health care immediately after the onset of the COVID-19 pandemic (24). Overall, in this study, the rapid transition to virtual care resulted in better continuity of psychotherapy visit attendance, as defined by fewer disruptions in care of >45 days. In addition, individuals had significantly fewer days between appointments during the postonset period. Thus, in this large population sample, individuals were more likely to return for additional psychotherapy visits after the widespread shift to virtual care. These population-level findings are consistent with other recent studies with smaller samples that showed better adherence to virtual mental health care compared with in-person visits (13, 25). In the present study, about 48% of psychotherapy visits during the postonset period were still conducted in person, which reflects variation in the degree to which these systems shifted to virtual care in response to local COVID-19 severity. Despite this finding, it is possible that improved engagement in psychotherapy during this period was reflective of a greater need for mental health services during COVID-19 rather than of the shift to virtual care. Future research should examine this relationship further.

The overall findings in this study support prior research showing that older individuals and those living in rural areas may experience more disruption in mental health care (26, 27). Contrary to prior research (28, 29), however, this study did not find disparities in psychotherapy adherence among patients belonging to historically minoritized racial-ethnic groups. Whereas most other studies have focused on access to all psychotherapy visits, the present study focused on disruptions in care after an initial visit. Prior research demonstrates that initial access may be the biggest barrier to receipt of psychotherapy (23, 30).

Despite concerns that there may be inequities in use of and access to virtual mental health care among certain groups of individuals (31, 32), we did not find sociodemographic subgroup differences in psychotherapy disruption after more widespread availability of virtual care. In addition, individuals with any of the mental health conditions included in this study were less likely to have disruptions in psychotherapy after the proliferation of virtual care. This finding is consistent with prior studies demonstrating that virtual care may be particularly beneficial in supporting ongoing care with providers (33). Psychotherapy may be particularly well suited for virtual care. Despite the narrowed gap in psychotherapy engagement through the shift to virtual care, there remain significant gaps in access to mental health services among those with psychiatric diagnoses. More research is also needed to better understand access to and use of other types of virtual mental health care, such as virtual medication management visits with psychiatrists and use of integrated care models in non–behavioral health settings.

Moving forward, virtual mental health care appears likely to continue and to represent a large portion of all psychotherapy services provided. Virtual care may have many benefits compared with in-person care, including greater ease of access, fewer barriers related to transportation, less need to take off time from work, and less stigma. Individuals in rural areas may benefit substantially from ongoing use of virtual care. In addition, therapists may have the ability to better observe living environments. Nonetheless, successful virtual mental health care relies on strong, reliable Internet connections as well as good and up-to-date computer equipment. Furthermore, privacy concerns about virtual care remain. More research is also needed on therapist-patient communication and rapport as well as on the ability of therapists to assess nonverbal cues, such as affect and facial and body expressions, during virtual sessions. Finally, it is important to evaluate patient preferences when determining whether virtual or in-person psychotherapy is most appropriate (34–36).

This study must be evaluated in the context of its limitations. We did not evaluate use of a patient’s first psychotherapy visit, so the findings may not represent initial use of services. Differences in access to psychotherapy should be examined in future research on specific populations. Similarly, this study was not able to examine reasons for disengagement from care. This study was conducted in three large health systems that possessed virtual mental health care infrastructure. Some systems may not have developed the infrastructure to provide these services. This study did not examine use of virtual care for medication management or for types of care other than psychotherapy, which may present different challenges. In addition, some patients in this study had missing data, including on race-ethnicity, which may have affected the results. We also were not able to differentiate between various modalities of psychotherapy that were delivered. Severity of illness and motivation to seek care were not assessed in this study. Even though differences in psychotherapy disruption were observed, this study did not assess whether differences in disease outcomes were present. The possibility remains that differences in population-level severity of illness before versus after the onset of COVID-19 may have affected the results. Similarly, patients seeking care in the postonset period may have perceived greater need for services, had more flexible schedules, or been more motivated to attend visits compared with individuals receiving care prior to the pandemic. Finally, this study was unable to assess the reason for the difference in total visits before versus after the onset of the pandemic. This difference may be due to unmeasured variance among the populations seeking care during both periods or to limitations in appointment availability—either throughout the postonset period or during the immediate weeks after the onset of the pandemic, when these systems were transitioning to virtual care.

In a sample of >110,000 individuals with mental health conditions, the population of patients experienced fewer disruptions in psychotherapy receipt after the onset of the COVID-19 pandemic and the rapid shift to virtual care. These findings support continued use of virtual psychotherapy as an option for care when appropriate infrastructure is in place. In addition, these findings support the continuation of policies that provide access to and coverage for virtual psychotherapy.