Children and Telehealth in Mental Healthcare: What We Have Learned From COVID‐19 and 40,000+ Sessions

The current study reports on internal review and research conducted at Achieve Behavioral Health (Achieve), a federally designated Certified Community Behavioral Health Center which delivers a broad spectrum of outpatient mental health services to the communities of Rockland and Orange Counties in NY. Achieve employs more than 130 clinical staff members in providing services to over 3200 clients annually. Clinic demographics reflect the clinic catchment area and are heavily influenced by the historically underserved Orthodox Jewish and Hasidic populations which make up a majority of the client population. Demographics for the current sample are listed in Table 1.

Data represent the total number of sessions of all clinical services rendered across multiple programs (standard outpatient, intensive outpatient, schools, medication management, etc.) as well as multiple departments and service types (psychotherapy, psychiatry, case management, etc.). The data were obtained from the naturally accruing clinical database generated by regular sessions entered in the electronic health records (EHRs).

Total number of sessions included in the analyzed data set is N=43,294, representing 2520 individual clients. All sessions occurred between the dates of January 1, 2020 and June 30, 2020, before, during, and after the peak of COVID‐19 spread in the NY catchment area (see Figure 1 for the number of positive tests in New York State by date).

In addition, two additional sources of data were used in some analyses, namely record of new calls‐in for service (N=729 recorded for this 6‐month period) and record of discharges (N=575). These records are recorded independently of clinic EHR but represent similar naturally accruing data collected over the course of regular clinic operations.

Data were captured and collated by the Achieve clinical data team and submitted to the research team as anonymized data. No additional data were collected for this study above and beyond that which are collected as part of regular clinic operations. Research was considered exempt from full review. A letter of exemption was received from the Western Copernicus Group Institutional Review Board.

Data reflect COVID‐19 as experienced in New York State during the first six months of 2020. The COVID‐19 lockdown began over the days of March 15–18, 2020, though, not formalized legally until the following week. The TH executive order in NYS was started retroactively to March 7, 2020. Lockdown restrictions began to be lifted during the week of May 15, 2020 (13). Although Achieve as an “essential service” was never fully closed to in‐person services, with minimal exception, all services were rendered remotely via TH from approximately March 17, 2020 through May 23, 2020. In line with falling infection rates, easing restrictions, and the desire of some clients and providers to resume in‐person services, a slow but steady return to FTF services began beginning May 24, 2020 and continued through the summer.

All analyses were performed using IBM SPSS (version 27). A variety of analytic approaches were necessary given the multiple lines of evidence encompassed in this study as well as the differing natures of data involved. For clarity, the following section is divided by hypothesis and specification of analytic technique is included after each hypothesis.

It was anticipated that during the period of COVID‐19 necessitated TH (January 1, 2020 to June 30, 2020) total clinical activity would demonstrate a relative drop in child clinical activity when compared to adult clinical activity. As it was further hypothesized that TH would be strongly implicated in this decrease and as TH was not evenly distributed over time, it was necessary to consider, separately and together, the main effects of date, TH and child status (child) on total sessions as well as the interaction of Date×TH×Child. For purposes of analysis, date was stratified across three levels of the COVID‐19 timeline (“1” – pre‐lockdown [January 1, 2020 to March 16, 2020]; “2” – lockdown [March 17, 2020 to May 23, 2020]; “3” – post‐lockdown [May 24, 2020 to June 30, 2020]). An outcome measure of number of sessions was then calculated as a count for each of these three periods with counts stratified by TH (“1” – face‐to‐face; “2” – telehealth). Due to the uneven time periods and the variable presence of individual subjects throughout the duration of the analysis period, an exposure variable (i.e., number of days eligible for sessions) was included as an offset in the analyses. A negative binomial regression analysis was run on count data (number of sessions) using a generalized estimating equation approach to handle repeated measures. Negative binomial regression was indicated over Poisson regression as mean of the outcome (3.11) was substantially smaller than its variance (37.54), indicating overdispersion. Due to the complexity of the above analysis and the inclusion of the three‐way interaction of Date×TH×Child alone includes 12 levels, effects shown to be significant are further reported in the form of descriptive differences.

It was further hypothesized that the differences between child and adult sessions would not be uniform across child age, but would indicate fewer sessions for younger children as compared to older children (for this analysis, child clients were divided into three age categories [ages 0–6; 7–11; 12–17 years] consistent with the clinical practice at Achieve). However, it was hypothesized that while different child age groups would be shown to be distinct from one another, it was also anticipated that each group would independently differ from the adult group, thereby lending conceptual validity to the child versus adult structure of analysis. The negative binomial regression was rerun with child replaced by age categories.

It was expected that records of intake and discharge would display a relative decrease in new requests for services (“calls‐in”) and a relative increase of discharges from service among child clients as compared to adult clients during the study time period. Chi‐square tests of independence were run to measure significance.

It was anticipated that while child sessions show a preference for FTF services this preference would not be consistent across service types. In particular, it was expected that psychotherapy sessions would yield a significant preference for FTF over TH, but that this preference would not be observed for child psychiatry sessions. Other categories of services might also display differences, so the factor of service type (“service”) was added to the analyses. In this variable, all sessions are stratified into one of four categories (“1 – psychotherapy; 2 – psychiatry; 3 – support services; 4 – other”). The category of “support services” includes the mental health services of psychiatric rehabilitation, targeted case management and peer support. The category of “other” includes group psychotherapy, crisis, intake, and nursing. Negative binomial regression analysis was repeated including the additional factor of service.

Running a fully saturated model that included the main effects of date, TH, and child as well as all interaction terms, it was found that all effects were significant in predicting number of sessions (see Table 2). Despite a greater number of child sessions as compared to adult sessions during the pre‐COVID‐19 period (see Figure 2), total sessions dropped to favor adults during the lockdown period as well as in the post‐lockdown period (see Figure 2). This preference was particularly apparent in TH versus FTF sessions, with adult clients demonstrating an overall preference for TH, exemplified in the significant shift away from TH and return to FTF post‐lockdown, which occurred at twice the rate in child sessions as adult sessions (see Figure 2). Refer Table 1 and Figure 3 for additional representation of the adult preference for TH.

When child status was replaced with the age‐stratification variable, the model and all effects remained significant. Notably, while the interaction of Date×TH×age was significant, consistent with the study hypotheses, the simple effects of age demonstrated significance at every level, with all three child ages demonstrating significant differences when compared to the adult group (“0–6:” χ² _{( 1)}=8.879, p=0.003; “7–11:” χ² _{( 1)}=6.631, p=0.01; “12–17:” χ² _{( 1)}=16.392.88, p<0.001). The model was rerun excluding adult sessions, and child age groups were compared against the reference group of 12–17. In this model, age was significant (Wald χ² _{( 2)}=2.542, p=0.02); however, parameter estimates of individual age groups were not significant (0–6: χ² _{( 1)}=2.542, p=0.11; 7–11: χ² _{( 1)}=3.700, p=0.054).

Monthly records of new calls‐in for service show a significant decline in the number of calls‐in for March and April (corresponding to the COVID‐19 lockdown and switch to TH) as compared to January and February (see Figure 4). New calls‐in began to increase again during May and June. Although the decline was seen in calls‐in for both child and adult clients, the drop in child calls‐in was greater with 114 and 106 calls‐in recorded in January and February falling to 46 and 20 in March and April as compared to adults (January, 78; February, 44; March, 36; April, 22). Furthermore, the increase in calls‐in during the post‐lockdown period was less for child clients relative to their pre‐lockdown numbers (May, 49; June, 80), as compared to adults (May, 61; June, 73). Overall, this difference was found to be significant (χ² _{(5, N=729)} = 21.745, p<0.01).

Monthly records of discharge from service show a similar, though directionally opposite pattern (see Figure 4). Significant increase in the number of discharges was recorded for the month of March (125) (compared to January [106] and February [78]). Though this increase leveled off in April (72) and May (84) increasing again in June (110), it is noteworthy that the initial increase reflected a significant number of discharges of child clients (March [74] compared to January [52] and February [30]). By contrast adult discharges were steady in March (51, as compared to January [54] and February [48]). This pattern of higher rates of discharge among child clients continued with April (38), May (55) and June (65) as compared to adults (April [34]; May [29]; June [45]). Overall, this difference was found to be significant (χ² _{(5, N=575)}=15.577, p<0.01).

The original model with Date×TH×Child was rerun with the inclusion of the variable service, and all main and interaction effects were shown to be significant (p<0.05) in predicting number of sessions, including the four‐way interaction Date×TH×Child×service.

As in the previous analysis, the general model of all TH services displayed significant relative decline among child TH services when compared to adult TH services (see Figure 3, as well as Figure 2 and Tables 1 and 2). Child TH services accounted for 16% fewer sessions than did child FTF, compared to adult sessions which showed a 13% increase in TH over FTF sessions. However, as hypothesized, this effect was not uniform across service type. Among psychotherapy sessions, a similar breakdown was demonstrated (child TH: −19%; adult TH: 14%). By contrast, psychiatry sessions increased with TH services and this increase was consistent among child and adult sessions (both demonstrating a 33% increase in TH psychiatry over FTF). Support services sessions also increased across both groups, though more so among adults (39% increase compared to 16% for child sessions). Other services demonstrated significant decline among both child and adult sessions, though here too the decrease was more pronounced among child sessions (adult: −24%; child: −43%).

Despite comprising a majority of both clients and billed sessions pre‐COVID‐19, children dropped to rates below those of adults with the advent of TH. The co‐occurring decrease in new calls‐in for service and increase in discharge during this same period together with rapid return to FTF services among child sessions not seen among adult sessions further illustrates a preference for FTF services among child clients and providers.

Differences in service type were notable when psychotherapy was compared with psychiatry sessions: whereas number of child psychotherapy sessions decreases significantly when delivered via TH, psychiatry does not, and even increases. A similar pattern holds true for support services. These more pragmatic services may be more useful at these times, but even more importantly, it supports the feasibility of increased access to child psychiatry, a crisis level issue well before the onset of COVID‐19 (14, 15).