Comparing In-Person, Telephonic, and Video-Based Treatment of Depression in Adult Primary Care During the COVID-19 Pandemic

Abstract

Introduction

The COVID-19 pandemic forced a rapid shift toward virtual modalities for the treatment of depression in primary care.

Methods

Participants were adults 18 years and older with a new episode of depression diagnosed in primary care between March 1, 2020, and May 21, 2021, and moderate-to-severe symptoms of depression at the time of diagnosis (N = 9619). Outcomes were 1) antidepressant medications prescribed and dispensed (referred to as received), as well as adherence to those medications; 2) referrals made to depression-related services and the receipt of those services; and 3) a follow-up visit completed with the diagnosing practitioner regardless of treatment actions.

Results

Patients were 42.4 ± 17.8 years old, and 77.6% had moderate-to-severe symptoms at diagnosis. Most patients were women (70.4%), 48.2% were Hispanic, and 8.4% were Black. Telephone visits were associated with 64% increased odds of having an antidepressant prescribed when compared to in-person visits. However, patients prescribed an antidepressant during a telephone visit were 52% less likely to receive this prescription when compared to patients who were prescribed an antidepressant during an in-person visit. Telephone and video visits were associated with 48% and 37% decreased odds, respectively, of having a follow-up visit with the prescribing practitioner when compared to an in-person visit.

Conclusion

Telemedicine for depression in adult primary care may result in greater antidepressant prescribing than in-person care, but these medications are less likely to be received. This study’s findings suggest that health systems should adjust electronic decision support tools (such as mail-order pharmacies) to ensure virtual care decisions are implemented.

Introduction

The COVID-19 pandemic forced a rapid shift toward virtual modalities for the treatment of depression in primary care. Before the pandemic, most visits to a practitioner for depression were in-person with primary care practitioners (PCPs), and most medications for depression were prescribed by PCPs. ^1,2 Since then these visits have largely transitioned to virtual formats. Between 2018 and 2020, the proportion of overall primary care visits performed virtually increased by 35%, and of those visits specifically for depression, > 50% were performed virtually, the highest proportion of any condition. ^3,4 Given that the pandemic has resulted in > 30% increase in depression symptoms over the past 4 years, ^5–7 understanding how virtual treatment modalities affect treatment actions and processes of care is critical. ⁸

The benefits of telehealth are well established, ^9,10 including decreased costs of care without affecting patient perceptions of care quality. ¹¹ Practitioners have also found virtual care to be an acceptable alternative to in-person care, ¹² and in specialty psychiatric settings, several studies have shown that telemedicine is an accessible, feasible, and safe modality for the treatment of depression, ^13–17 with treatment outcomes equivalent to those of in-person visits. ¹⁸ Much less is known about the virtual treatment of depression in primary care settings. One study found that telemedicine was associated with increased attendance for mental health appointments and fewer cancellations. ¹⁹ Virtual behavioral health programs that augment in-person primary care, such as mobile app platforms, lay person-delivered calls, and asynchronous telepsychiatry, have shown some effectiveness for depression care and symptom reduction when used adjunctively with primary care. ^20–22

Although such results are promising, knowledge remains limited about how the shift to virtual care during the pandemic has affected treatment actions, such as medication prescribing, adherence, and appropriate follow-up. ²³ In health conditions other than mental illness, medication adherence during the pandemic was mixed, facilitated by factors like increased health information delivered with technology but limited by other factors like medication shortages and challenges with transportation. ²⁴ Regarding psychotropic medications, increased prescribing and overall improved adherence were noted early in the pandemic, but this varied by age, with older patients having relatively lower rates of medication fills and adherence. ^25–27 This is consistent with other studies suggesting similar or improved adherence during the pandemic to medications for chronic disease, such as asthma. ^24,28,29 Referral behavior was also affected for psychiatric conditions during the pandemic, with decreased referrals early on and increased referrals over time. ³⁰

To date, the extent of the literature available examining the effect of virtual care on care processes, such as referral behavior and medication prescribing for the treatment of depression in primary care settings, is limited. The present study is a unique contribution to the literature because it studied the effect of telemedicine on processes of care for depression and how it varied by patient characteristics, such as gender, race, age, depression symptom severity, and by race, age, and gender concordance, with the PCP. Race and gender may lead to differences in treatment assignment ³¹ and outcomes, ³² but they also may affect the utilization of telehealth. ^33,34 Additionally, age is known to affect the care a patient receives, ³⁵ patient perceptions of care, ³⁶ and telehealth utilization. ³⁷ Depression symptom severity is generally used to guide treatment decisions and processes of care. ³⁸ Beyond patient characteristics alone, concordance of patient and practitioner characteristics, such as race, age, and gender, also affect treatment outcomes and the patients’ perceptions of their care. ^39,40

This study was designed to understand how virtual care during the COVID-19 pandemic compared to in-person care for the treatment of a new episode of depression in adult primary care settings. The primary outcomes were 1) any treatment received for depression, 2) medications prescribed, 3) medications dispensed (referred to as medication received), 4) referrals for depression-related services, 5) receipt of those services, and 6) having a follow-up visit with the diagnosing primary care practitioner regardless of any treatment actions (eg, prescribing medication or generating a referral). This study also examined whether the differences between in-person, telephonic, or video visits were associated with patient gender, race, age, baseline depression symptom acuity, and concordance for race, age, and gender with the diagnosing primary care practitioner.

Methods

Setting

This retrospective observational comparative effectiveness study was conducted in a large health system with 15 hospitals, 236 medical offices, and nearly 8000 practitioners serving 4.8 million members in all Southern California counties. Care for depression occurs in multiple settings, including primary care. ⁴¹ In general, ~ 50% of new depression episodes are diagnosed by PCPs who can prescribe medications and initiate referrals to psychiatry and collaborative depression care programs. ⁴¹ Regardless of a clinical care action, all patients are scheduled for a follow-up visit with the diagnosing practitioner 4–6 weeks following the new diagnosis/episode of depression. At the time of the study, if patients received a new diagnosis with a new prescription, they were not eligible for mail-order receipt of that medication. They were required to pick up the medication in person. All data were abstracted from electronic health records, and the health system’s institutional review board for human participants approved the study as minimal risk, granting a Health Insurance Portability and Accountability Act waiver of informed consent.

Participants

The selection of the sample for the study is shown in Figure 1. Eligible participants met the following criteria: 1) being > 18 years old; 2) having completed an in-person, telephone, or video visit within the departments of Family and Internal Medicine (adult primary care) during the period of March 1, 2020, to May 21, 2021; and 3) having a “new” depression diagnosis during a primary care visit in the study period. A new diagnosis of depression (referred to subsequently as a new episode) was defined as no diagnosis of depression and/or evidence of treatment for depression in the 12 months prior to the visit with the diagnosis. Criteria for determining a new episode of depression are provided in the Online Appendix and were based on the authors’ previous studies. ⁴² The term new episode is used herein instead of the term new diagnosis because it is difficult to ascertain an incident diagnosis with retrospective observational studies.

Figure 1:

Summary of selection of patients for the study. Depression symptoms were assessed during routine screening using the 9-item Patient Health Questionnaire (PHQ-9).

From these adult patients with a new episode of depression (N = 34,011), the authors selected those with moderate-to-severe symptoms of depression [Patient Health Questionnaire (PHQ-9) total score > 10] ³⁸ because these were the patients for whom the authors would expect to see treatment (N = 9619). The study period for determining eligibility criteria and treatment processes was between March 1, 2019, and June 30, 2022.

Outcomes

Outcomes were separated into 3 treatment actions that were standard of care for depression in the target health system: 1) antidepressant medications prescribed and dispensed (referred to as received), 2) referrals made to depression-related services and the receipt of those services (psychiatry visit and/or collaborative care visit), and 3) having a follow-up visit with the diagnosing practitioner. The authors also examined medication adherence if patients received a medication. A list of the medications used for analyses are provided in the Online Appendix.

Adherence to medication was calculated using definitions provided by the National Committee for Quality Assurance Healthcare Effectiveness Data and Information Set for the antidepressant medication acute (early adherence) and continuation (later adherence) phases of the metric. ⁴³ The specific calculation for both phases is provided in the Online Appendix. These adherence phases ensure that patients receive a 6-month course of medication for their depression. ⁴³

Exposure

The exposure was the type of outpatient primary care visit in which the diagnosis of the new episode of depression was made (in-person, telephone, or video). Although both telephone and video visits are considered virtual, there was reason to expect that video and in-person visits would have similar characteristics driving treatment decisions, including facial expressions, patient appearance, and eye contact, that would not be present in telephone visits without a visual component. ⁴⁴

Covariates

Covariates were chosen based upon their hypothesized or previously demonstrated impact on practitioner depression treatment decisions and outcomes and included age, gender, and race of the patient ⁴⁵ ; age, gender, and race concordance of the practitioner and patient ⁴⁶ ; and PHQ-9 total score at the time of diagnosis. The PHQ-9 is a self- or practitioner-administered questionnaire designed to screen for symptoms of depression, assist in diagnosis, and monitor symptom change over time. ³⁸ It contains 8 items that correspond to Diagnostic and Statistical Manual of Mental Illnesses, Fifth Editon, criteria for depression diagnosis and a ninth item that asks about suicidal ideation. Total scores range from 0 to 27 and are scored as 0–4 (no symptoms of depression), 5–9 (mild symptoms), 10–19 (moderate-to-severe symptoms), and 20–27 (severe symptoms). It is the most commonly used screener for depression in primary care settings in the US, ⁴⁷ and has been validated for many different cultures and is available in > 30 languages. ^48,49 The treatment actions, regardless of the type of visit, were also used as covariates appropriate to the outcome being analyzed (eg, depression service referral and receipt when looking at medication outcomes).

Analyses

Descriptive statistics for all outcomes and covariates were initially calculated by exposure (in-person, telephone, or video visit) using means and standard deviations (SDs) for continuous variables and frequencies and percentages for categorical variables (Table 1). Each outcome variable was dichotomized as yes or no, depending upon experiencing the outcome. Medications received were conditional on a prescription for that medication, and receipt of a depression-related service was conditional on a referral for a depression-related service. Having a follow-up visit with the diagnosing practitioner was not conditional on having a medication prescribed or a referral generated, as it was standard of care to have a follow-up visit with moderate-to-severe symptoms of depression regardless of the decision to treat.

Table 1:

Descriptive statistics for N = 9619 patients by type of visit in which a new episode of depression was diagnosed for all covariates and outcomes before and after inverse probability of treatment weighting (winsorized at the 95th percentile)

Measure at diagnosis	In-person (n = 5345) [mean ± SD or n (%)]	In-person weighted (n = 5345) [mean ± SD or %]	Telephone (n = 3630) [mean ± SD or n (%)]	Telephone weighted (n = 3630) [mean ± SD or %]	Video (n = 644) [mean ± SD or n (%)]	Video weighted (n = 644) [mean ± SD or %]
Patient age (y)	45.2 ± 19.1	45.2 ± 19.1	33.9 ± 15.2	44.4 ± 18.4	39.0 ± 15.8	39.6 ± 16.0
Patient gender (% woman)	3702 (69.3)	69.3	2624 (72.3)	69.7	448 (69.6)	69.4
Patient race or ethnicity
Hispanic	2741 (51.3)	51.3	1622 (44.7)	51.0	270 (41.9)	44.5
Black	437 (8.2)	8.2	318 (8.8)	8.2	53 (8.2)	8.3
Asian	380 (7.1)	7.1	229 (6.3)	7.2	43 (6.7)	6.8
White	1371 (25.7)	25.7	1135 (31.3)	25.8	199 (30.9)	30.3
Other	168 (3.1)	3.1	138 (3.8)	3.0	25 (3.9)	3.4
Not reported	248 (4.6)	4.6	188 (5.2)	4.7	54 (8.4)	6.8
PHQ-9 total score at diagnosis (range, 10–27)	15.8 ± 4.2	15.8 ± 4.2	16.3 ± 4.1	15.8 ± 4.2	16.1 ± 4.2	16.0 ± 4.2
Year of diagnosis (% 2021)	2639 (49.4)	49.4	1324 (36.5)	36.5	315 (48.9)	48.9
Age concordance (yes)	972 (18.2)	18.2	802 (22.1)	18.4	132 (20.5)	20.1
Gender concordance (yes)	3698 (69.2)	69.2	2398 (66.1)	69.1	453 (70.3)	70.8
Race concordance (yes)	1572 (29.4)	29.4	892 (24.6)	28.4	154 (23.9)	24.3
Prescribed medication (yes)	2254 (42.2)	42.2	2042 (56.3)	55.1	303 (47.0)	46.9
Received medication (yes)	1736 (77.0)	77.0	643 (31.5)	31.5	84 (27.7)	27.7
Depression-related services referral (yes)	1098 (20.5)	20.5	584 (16.1)	17.6	130 (20.2)	20.5
Depression-related services received (yes)	752 (68.5)	68.5	457 (78.3)	75.7	105 (80.8)	80.4
Diagnosing practitioner follow-up visit (yes)	3605 (67.4)	67.4	1881 (51.8)	52.9	342 (53.1)	53.7
No evidence of treatment (yes)	349 (6.5)	6.5	184 (5.1)	5.1	31 (4.8)	4.8

PHQ-9, Patient Health Questionnaire; SD, standard deviation.

To control for the nonrandom assignment of visit type, inverse probability of treatment weighting was performed with the twang package in R (version 2.6.1), ⁵⁰ which utilizes gradient boosted models to estimate propensity scores. Visit type groups were weighted using all covariates, with in-person visit as the reference group. Due to positive skewness, winsorized weights at the 95th percentile were used to mitigate the effects of extreme values. The association between visit type and each outcome was determined using multivariable logistic regression models. Each model was adjusted for the appropriate covariates and the inverse probability of treatment weights.

Two variables were added to each regression model to account for COVID-19 pandemic-related changes in practice over the study period: 1) year of diagnosis (2020 vs 2021), and 2) a year of diagnosis by visit type interaction term. These terms were added to the model based on the literature that showed referrals for psychiatric conditions during the pandemic decreased early on and then increased over time. ³⁰ This controlled for any bias that could be introduced by low rates of utilization when services were not available. All statistical analyses were performed using R Statistical Software (version 4.3.1). ⁵¹

Results

Participants

Patients (N = 9619) were 42.4 ± 17.8 years old and had an average PHQ-9 total score of 16.0 ± 4.2 [77.6% had moderate-to-severe symptoms (PHQ-9 total score 10–19), and 22.4% had severe symptoms (PHQ-9 total score 20–27)]. Most patients were women (70.4%), and 48.2% were Hispanic, 28.1% were White, 8.4% were Black, and 6.8% were Asian. Diagnosing primary care practitioners were 43.7 ± 8.8 years old, 64.9% were women, 49.1% identified as Asian, 28.1% White, 16.1% Hispanic, and 5.0% Black, and they had 9.9 ± 7.9 years of employment at the target health system. Only 20% of patients and practitioners were concordant for age and 27% were concordant for race. Concordance was greater for gender at 68%. Characteristics for patients and indicators of patient and practitioner concordance are shown in Table 1 for the unadjusted sample and the sample after inverse probability of treatment weighting.

Any Referral and/or Treatment for Depression

Most patients (94%) had either a referral for and/or received some treatment for their depression. Table 2 provides the adjusted results of factors related to referral and/or treatment for depression. Compared to patients whose new episode of depression was diagnosed at an in-person visit, those diagnosed during a telephone visit had a 35% [95% confidence interval (CI): 1.04–1.77; P = .03] and those diagnosed during a video visit had a 96% (95% CI: 1.07–3.59; P = .03) increased odds of having a referral for and/or receiving some treatment, respectively. Patients who were diagnosed in 2021 had a 26% (95% CI: 1.01–1.57; P = .04) increased odds of having a referral for and/or receiving some treatment compared to 2020, regardless of the type of visit in which they were diagnosed (interaction of visit type by year was not significant).

Table 2:

Adjusted model results for factors related to receiving any referral and/or treatment for depression (n = 9055; 94.1%) following the visit in which a new episode of depression was diagnosed

Measure	OR	95% LL	95% UL	P value
Type of visit for diagnosis
In-person (Ref)	—	—	—	—
Telephone	1.35	1.04	1.77	.03
Video	1.96	1.07	3.59	.03
Year of diagnosis
2020 (Ref)	—	—	—	—
2021	1.26	1.01	1.57	.04
Patient age (y)	1.01	1.00	1.01	.08
Patient gender
Woman (Ref)	—	—	—	—
Man	0.73	0.59	0.92	.006
Unknown	1.68	0.37	7.60	.50
Patient race or ethnicity
Hispanic (Ref)	—	—	—	—
Black	0.87	0.61	1.25	.46
White	1.40	1.07	1.84	.01
Asian	1.17	0.76	1.82	.47
Other	1.08	0.65	1.80	.76
Unknown	0.82	0.53	1.28	.38
PHQ-9 total score at diagnosis	1.08	1.05	1.11	< .001
Age concordance
No (Ref)	—	—	—	—
Yes	1.29	1.00	1.65	.048
Gender concordance
No (Ref)	—	—	—	—
Yes	1.26	1.01	1.57	.04
Race concordance
No (Ref)	—	—	—	—
Yes	0.78	0.61	1.00	.05
Type of visit * year of diagnosis
In-person * 2020 (Ref)	—	—	—	—
Telephone * 2021	0.88	0.58	1.33	.54
Video * 2021	0.50	0.23	1.10	.09

LL, lower limit; OR, odds ratio; PHQ, Patient Health Questionnaire; Ref, referent; UL, upper limit.

Patients who identified as White had a 40% (95% CI: 1.07–1.84; P =.01) increased odds of having a referral for and/or receiving some treatment when compared to patients who identified as Hispanic. Men had a 27% [odds ratio (OR) = 0.73; 95% CI: 0.59–0.92; P = .006] decreased odds of having a referral for and/or receiving some treatment when compared to women. Those with a higher PHQ-9 score at the time of diagnosis had an 8% (95% CI: 0.59–0.92; P = .006) increased odds of having a referral for and/or receiving some treatment. Finally, patients who had gender concordance with the diagnosing practitioner had a 26% (95% CI: 1.01–1.57; P = .004) increased odds of having a referral for and/or receiving some treatment compared to those who were not concordant.

Antidepressant Medications Prescribed and Received

Table 1 and Figure 2 provide unadjusted results for medication prescribing and receipt. Prescribing rates varied by visit type, from 56.3% in those diagnosed during a telephone visit to 42.2% during an in-person visit. Of those prescribed a medication, the rates of receiving that medication varied even more, with 77.0% of patients receiving medication after being prescribed when diagnosed during an in-person visit and 27.7% when diagnosed during a video visit. Table 3 provides the adjusted results of factors related to medications prescribed and received for depression. Compared to patients whose new episode of depression was diagnosed at an in-person visit, those diagnosed during a telephone visit had a 64% (95% CI: 1.45–1.86; P <.001) increased odds of having an antidepressant medication prescribed but a 52% decreased odds of receiving that medication after the visit (OR = 0.48; 95% CI: 0.41–0.56; P <.001). Although patients who were diagnosed during a video visit did not differ from those during a telephone visit in their odds of having a medication prescribed, they had a 65% (OR = 0.35; 95% CI: 0.25–0.49; P <.001) decreased odds of receiving that medication after the visit (OR = 0.35; 95% CI: 0.25–0.34; P <.001). These rates differed by year of diagnosis such that patients who were diagnosed during 2021 had a 26% (95% CI: 1.12–1.42; P < .001) and 21% (95% CI: 1.12–1.42; P < .001) increased odds of a prescription and receiving that medication, respectively, when compared to those diagnosed in 2020, regardless of the type of visit in which they were diagnosed (interaction of visit type by year was not significant).

Figure 2:

Graphic representation of unadjusted study outcomes across different primary care visit types. Study outcomes were 1) antidepressant medications prescribed and received, 2) referrals made to depression-related services and the receipt of those services (psychiatry visit and/or collaborative care visit), and 3) having a follow-up visit with the diagnosing practitioner.

Table 3:

Adjusted model results for factors related to a depression medication prescribed (n = 4599; 48% of the patient sample) at the visit in which a new episode of depression was diagnosed and received (n = 2463; 54% of those prescribed) after that visit

Measure	Medication prescribed (n = 4599)				Medication received (n = 2463)
	OR	95% LL	95% UL	P value	OR	95% LL	95% UL	P value
Type of visit for diagnosis
In person (Ref)	—	—	—	—	—	—	—	—
Telephone	1.64	1.45	1.86	< .001	0.48	0.41	0.56	< .001
Video	1.23	0.97	1.55	.09	0.35	0.25	0.49	< .001
Year of diagnosis
2020 (Ref)	—	—	—	—	—	—	—	—
2021	1.26	1.12	1.42	< .001	1.21	1.06	1.37	.003
Patient age (y)	1.00	1.00	1.00	.27	1.00	1.00	1.00	.20
Patient gender
Female (Ref)	—	—	—	—	—	—	—	—
Male	0.77	0.69	0.86	< .001	0.94	0.83	1.07	.36
Unknown	1.01	0.47	2.19	.98	0.87	0.32	2.41	.79
Patient race or ethnicity
Hispanic (Ref)	—	—	—	—	—	—	—	—
Black	0.66	0.55	0.80	< .001	0.63	0.51	0.78	< .001
White	1.39	1.23	1.57	< .001	1.31	1.15	1.49	< .001
Asian	1.02	0.82	1.25	.88	1.01	0.79	1.30	.97
Other	1.17	0.88	1.55	.27	1.05	0.77	1.42	.77
Unknown	1.38	1.08	1.75	.009	1.16	0.88	1.53	.28
PHQ-9 total score at diagnosis	1.02	1.01	1.03	< .001	1.05	1.03	1.06	< .001
Age concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.25	1.11	1.42	< .001	1.06	0.93	1.22	.39
Gender concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.00	0.89	1.11	.98	0.95	0.84	1.08	.43
Race concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.94	0.83	1.06	.30	0.99	0.87	1.13	.89
Depression service referral
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.73	0.58	0.92	.007	0.83	0.66	1.04	.10
Depression service received
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.82	0.63	1.06	.12	0.92	0.70	1.20	.53
Diagnosing practitioner follow-up visit
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.00	0.90	1.12	.95	0.98	0.87	1.11	.75
Type of visit * year of diagnosis
In person * 2020 (Ref)	—	—	—	—	—	—	—	—
Telephone * 2021	1.17	0.96	1.42	.11	0.79	0.62	1.00	.05
Video * 2021	0.94	0.68	1.32	.74	0.73	0.45	1.18	.19

LL, lower limit; OR, odds ratio; PHQ-9, Patient Health Questionnaire; Ref, referent; UL, upper limit.

Men had a 23% (OR = 0.77; 95% CI: 0.69–0.86; P < .001) decreased odds of having a medication prescribed than women; however, there were no gender differences in receiving that medication. When compared to patients who identified as Hispanic, those who identified as Black had a 33% (OR = 0.67; 95% CI: 0.56–0.82; P < .001) and 36% (OR = 0.64; 95% CI: 0.52–0.79; P < .001) decreased odds of having a medication prescribed and receiving that medication, respectively. Conversely, those who identified as White had a 39% (95% CI: 1.23–1.57; P < .001) and 31% (95% CI: 1.15–1.49; P < .001) increased odds of having a medication prescribed and receiving that medication, respectively, compared to patients who identified as Hispanic. As expected, patients with a higher PHQ-9 score at the time of diagnosis had a 2% (95% CI: 1.15–1.49; P < .001) and 3% (95% CI: 1.15–1.49; P < .001) increased odds of having a prescription and receiving that medication, respectively. Age concordance between practitioner and patient had a 25% (95% CI: 1.11–1.42; P < .001) increased odds of having a medication prescribed, but this did not affect receiving that medication. Finally, having a depression-related referral was associated with a 27% (OR = 0.73; 95% CI: 0.58–0.92; P =.007) decreased odds of having a medication prescribed, but it was not related to receiving that medication.

Depression-Related Services Referral and Receipt

Table 1 and Figure 2 provide unadjusted rates of depression-related referrals and receiving a service based on these referrals by visit type. Overall, 18.8% of patients newly diagnosed with depression received a depression-related referral, and 72.5% received a service based on that referral. Unlike the findings for antidepressant medications, there were no effects of visit type on either a referral for or receipt of depression-related services (Table 4). However, patients diagnosed in 2021 had a 15.68 (95% CI: 12.68–19.39; P < .001) and 14.41 (95% CI: 11.12–19.69; P < .001) increased odds of having a depression-related referral and receiving a service based on this referral, respectively, when compared to those diagnosed in 2020, which did not vary by visit type.

Table 4:

Adjusted model results for factors related to a depression-related service referral (n = 1812; 19% of the patient sample) at the visit in which a new episode of depression was diagnosed and a service received (n = 1314; 73% of those referred) after that visit

Measure	Depression-related service referral				Depression-related service received
	OR	95% LL	95% UL	P value	OR	95% LL	95% UL	P value
Type of visit for diagnosis
In person (Ref)	—	—	—	—	—	—	—	—
Telephone	1.09	0.76	1.54	.65	1.34	0.89	2.01	.16
Video	0.80	0.40	1.61	.53	0.80	0.34	1.87	.60
Year of diagnosis
2020 (Ref)	—	—	—	—	—	—	—	—
2021	15.70	12.70	19.40	< .001	14.40	11.10	18.70	< .001
Patient age (y)	1.00	1.00	1.00	.61	0.99	0.99	0.99	.05
Patient gender
Female (Ref)	—	—	—	—	—	—	—	—
Male	0.86	0.73	0.98	.07	0.85	0.71	1.01	.07
Unknown	1.66	7.87	3.36	< .001	2.16	7.16	4.36	< .001
Patient race or ethnicity
Hispanic (Ref)	—	—	—	—	—	—	—	—
Black	0.83	0.63	1.09	.18	0.92	0.69	1.23	.57
White	0.67	0.56	0.80	< .001	0.73	0.59	0.89	.002
Asian	1.01	0.76	1.36	.94	0.93	0.67	1.29	.66
Other	0.99	0.67	1.45	.95	1.17	0.77	1.78	.46
Unknown	0.68	0.42	0.81	.001	0.64	0.45	0.93	.02
PHQ-9 total score at diagnosis	1.03	1.00	1.00	.004	1.05	1.03	1.07	< .001
Age concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.93	0.78	1.11	.43	0.91	0.75	1.12	.38
Gender concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.19	1.02	1.40	.03	1.14	0.96	1.36	.13
Race concordance
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.82	0.69	0.97	.02	0.89	0.73	1.07	.22
Medication prescribed
No (Ref)	—	—	—	—	—	—	—	—
Yes	0.58	0.47	0.71	< .001	0.57	0.45	0.71	< .001
Medication received
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.14	0.91	1.43	.27	1.15	0.89	1.48	.30
Diagnosing practitioner follow-up visit
No (Ref)	—	—	—	—	—	—	—	—
Yes	1.35	1.16	1.58	< .001	1.39	1.18	1.65	< .001
Type of visit * year of diagnosis
In person * 2020 (Ref)	—	—	—	—	—	—	—	—
Telephone * 2021	1.28	0.88	1.87	.20	1.14	0.74	1.76	.54
Video * 2021	1.50	0.72	3.13	.28	1.83	0.75	4.46	.18

LL, lower limit; OR, odds ratio; PHQ-9, Patient Health Questionnaire; Ref, referent; UL, upper limit.

Unlike the findings for antidepressants, patients who identified as White had a 33% (OR = 0.67; 95% CI: 0.56–0.80; P < .001) and 27% (OR = 0.73; 95% CI: 0.59–0.89; P =.002) decreased odds of referral for and receiving a depression-related service, respectively, based on this referral when compared to patients who identified as Hispanic. As expected, patients with a higher PHQ-9 score at the time of diagnosis had a 3% (95% CI: 1.01–1.04; P =.004) and 5% (95% CI: 1.03–1.07; P < .001) increased odds of having a referral and receipt of a service, respectively. Patients who had race concordance between patient and practitioner had a 18% (OR = 0.82; 95% CI: 0.69–0.97; P =.02) decreased odds of having a depression-related referral, but there was no effect for receipt of a service based on that referral. Finally, patients who had a medication prescribed had a 42% (OR = 0.58; 95% CI: 0.47–0.71; P < .001) decreased odds and patients who had a PCP follow-up visit had a 35% (95% CI: 1.16–1.58; P < .001) increased odds of having a depression-related referral, with similar odds for receipt of a service based on that referral.

Diagnosing Practitioner Follow-Up Visit

Table 1 and Figure 2 provide unadjusted rates of PCP follow-up visit rates by visit type. Having a follow-up visit with the diagnosing practitioner varied by visit type, from 67.4% for in-person visits to 51.8% for telephone visits. Patients who had either a telephone or a video visit had a 48% (OR = 0.52; 95% CI: 0.46–0.59; P < .001) and a 37% (OR = 0.63; 95% CI: 0.50–0.80; P < .001) decreased odds, respectively, of having a follow-up visit with the diagnosing practitioner when compared to patients who had an in-person visit (Table 5). Those patients who had a diagnosis in 2021 had a 18% (OR = 0.82; 95% CI: 0.72–0.94; P = .003) decreased odds of having a follow-up visit with their diagnosing practitioner compared to those who had a visit in 2020, a finding that did not vary by visit type.

Table 5:

Adjusted model results for factors related to having a follow-up visit with the diagnosing practitioner after the visit in which a new episode of depression was diagnosed (n = 5696; 59% of the patient sample)

Measure	OR	95% LL	95% UL	P value
Type of visit for diagnosis
In person (Ref)	—	—	—	—
Telephone	0.52	0.46	0.59	< .001
Video	0.63	0.50	0.80	< .001
Year of diagnosis
2020 (Ref)	—	—	—	—
2021	0.82	0.72	0.94	.003
Patient age (y)	1.03	1.02	1.03	< .001
Patient gender
Woman (Ref)	—	—	—	—
Man	0.86	0.76	0.96	.009
Unknown	0.47	0.21	1.05	.07
Patient race or ethnicity
Hispanic (Ref)	—	—	—	—
Black	1.00	0.82	1.23	.95
White	0.94	0.82	1.07	.34
Asian	0.85	0.69	1.05	.14
Other	0.97	0.72	1.30	.84
Unknown	0.87	0.68	1.10	.24
PHQ-9 total score at diagnosis	0.97	0.96	0.98	< .001
Age concordance
No (Ref)	—	—	—	—
Yes	1.08	0.95	1.22	.27
Gender concordance
No (Ref)	—	—	—	—
Yes	1.42	1.27	1.59	< .001
Race concordance
No (Ref)	—	—	—	—
Yes	1.02	0.90	1.16	.71
Medication prescribed
No (Ref)	—	—	—	—
Yes	1.01	0.88	1.16	.89
Medication received
No (Ref)	—	—	—	—
Yes	0.97	0.83	1.14	.72
Depression service referral
No (Ref)	—	—	—	—
Yes	1.16	0.89	1.50	.28
Depression service received
No (Ref)	—	—	—	—
Yes	1.20	0.91	1.60	.20
Type of visit * year of diagnosis
In person * 2020 (Ref)	—	—	—	—
Telephone * 2021	0.99	0.81	1.20	.90
Video * 2021	0.94	0.67	1.33	.75

LL, lower limit; OR, odds ratio; PHQ-9, Patient Health Questionnaire; Ref, referent; UL, upper limit.

Older patients had a 3% (95% CI: 1.02–1.03; P < .001) increased odds of having a follow-up visit than younger patients, and men had a 14% (OR =0.86; 95% CI: 0.76–0.96; P =.009) decreased odds of having a follow-up visit compared to women. Patients who had practitioner–patient gender concordance had a 42% (OR = 1.42; 95% CI: 1.27–1.59; P < .001) increased odds of having a follow-up visit compared to those who did not have gender concordance. No other variables were significantly associated with the odds of having a follow-up visit with the diagnosing practitioner.

Antidepressant Medication Adherence

Findings for antidepressant medication adherence are presented in Table A1. Patients whose new episode of depression was diagnosed during a video visit had a 59% (OR = 0.41; 95% CI: 0.20–0.81; P =.01) decreased odds of adhering to the acute phase of antidepressant medication treatment as compared to patients diagnosed during an in-person visit. However, this effect may have only been for visits in 2020, as there was a significant interaction of type of visit and year of diagnosis for the continuation phase of medication treatment such that those patients who were diagnosed in 2021 during a video visit had a 292% (95% CI: 1.14–7.50; P =.03) increased odds of continuation adherence when compared to patients diagnosed in 2020 during an in-person visit.

In addition, patients who identified as White had a 30% (95% CI: 1.04–1.63; P =.02) increased odds of adhering to the continuation phase of medication treatment when compared to patients who identified as Hispanic.

Discussion

Patients whose new episode of depression was diagnosed during a telephone visit had a 64% increased odds of having an antidepressant prescribed but were 52% less likely to receive this prescription when compared to those diagnosed during in-person visits. Once the medication was received, a diagnosis made during a video visit in 2021 was associated with a 292% increased odds of continuation phase adherence to that medication when compared to patients whose diagnosis was made during in-person visits in 2020. In contrast to these findings, telephone and video visits were associated with 48% and 37% decreased odds, respectively, of having a follow-up visit with the diagnosing practitioner when compared to an in-person visit. In addition, there were no differences between visit types in which the diagnosis of depression was made and the odds of having a referral for or receiving a depression-related service.

These findings are difficult to compare to the existing literature in part because studies on how treatment modality affects treatment actions are limited. The findings of reduced follow-up with prescribing practitioners of antidepressants are novel but may align with a study showing decreased referrals early in the pandemic for psychiatric conditions, ³⁰ even though the present study did not show significant differences in referral behavior. Although antidepressant prescribing and use may have increased overall during the pandemic, ^27,52 there are no other studies to the authors’ knowledge that have demonstrated differences in rates of antidepressant receipt across visit type. In general, the authors found that video visits were associated with decreased odds of the acute phase of antidepressant medication but that in the latter part of the pandemic (2021) video visits were associated with a marked increased odds of continuing a full dose of antidepressant medication. It is not clear why point-of-decision care would affect the later adherence to antidepressants, and this finding merits future study.

The reasons for the differences reported between types of primary care visits are not clear. It is possible that prescribing antidepressants may happen more frequently during a telephonic visit because practitioners in virtual settings may not be able to use facial expression or body language to determine diagnosis and treatment that could prevent a more nuanced evaluation of the patient that would be possible during an in-person visit. ⁴⁴ In addition, it may not just be visual cues that are necessary for treatment decisions. Because the telephonic care group had more antidepressant prescribing than the video-based care group (56.3% vs 47.0%; Table 1), this suggests that some aspect of shared decision-making may be missing during virtual care even when a practitioner can see the patient.

Although the rates of prescribing were higher for telephone encounters, rates of follow-up with the diagnosing practitioner were lower than in-person encounters. This may be related to the lack of support services from medical staff (no longer colocated with the prescribing practitioner) during virtual care who review patient instructions immediately after visits, which might include scheduling a follow-up visit. This study’s findings of lower rates for follow-up are contrary to one other study in patients with a variety of health conditions, including depression, that found scheduled follow-up visits were more likely to be done during virtual care. ⁵³ The difference in findings could be because the authors of the present study used completed visits and not scheduled visits.

Additionally, the finding of decreased fulfillment rates of antidepressants once prescribed for both telephonic and video visits may also reflect the lack of colocated services available to a patient after their primary care encounter. The health system at the time in which this study took place required new prescriptions to be retrieved in-person, which created an additional trip for those patients having a virtual visit compared to those who were already at a facility with a colocated pharmacy. Even if a patient could pick up a prescription in a different pharmacy location that was more convenient to their home or work, they still had to travel to that pharmacy, and this may have added a barrier to receiving this medication.

This is one of the first studies to examine how patient and practitioner factors, such as demographics and concordance in those demographics, affect treatment decisions and receipt of those treatments during virtual and in-person visits. Generally speaking, regardless of the type of visit in which the diagnosis of depression was made, this study’s findings that patients who identified as Black were the least likely to be prescribed and to receive antidepressants when compared to patients who identified as Hispanic but that White patients were less likely to be referred to and receive depression-related services when compared to patients who identified as Hispanic have been supported by numerous studies in the literature. ^31,42,45

Although race concordance has received attention as a possible reason that patients do not adhere to their medications, ^54–56 this relationship was not found in this study. Race concordance has largely been shown to affect patient perceptions of care, ⁵⁴ but the extent of their effect on treatment actions is still unclear. For this study population, gender concordance was more likely to lead to differences in the treatment of depression, though this contrasts with some previous studies on gender concordance. ^57–59 Regardless, this study’s results are consistent with findings of racial and gender disparities in access to depression care, both in and out of primary care settings. ^60,61

Although this study’s findings were clear, the authors cannot be sure that the differences found in depression treatment actions between visit types were not a result of residual confounding. For example, patients who had higher technological literacy may have been more likely to have telephonic visits due to self-selection, and these patients may have been more likely to adhere to treatments, such as antidepressant medications. Several studies have found that socioeconomic status, race, rurality, and age may determine a patient’s use of telephonic or in-person care, ^44,62,63 though the literature is unclear regarding how this affects treatment outcomes. ^64–66 Additionally, clinic- and practitioner-level factors may be more important than patient demographics in determining variation in the provision of virtual care. ^44,67,68

To the extent possible, the authors’ analyses included patient and practitioner demographic factors, as well as race, age, and gender concordance between patients and practitioners, in both the matching of visit type and the regression models, but differences remained between visit types (Table 1). Unfortunately, the authors did not have measures in the electronic health record of patient socioeconomic status, access to reliable internet and phone service, or preferences for delivery of care. Future research doing retrospective observational comparative effectiveness studies for modes of depression care delivery should consider these variables in their statistical models.

Another limitation for the present study was its generalizability. The health system in which the study took place was not representative of patients with depression who are under- or uninsured. Future studies should be done in settings serving these groups of people. Despite these limitations, this study was one of the largest in the literature, with nearly 10,000 patients and a racially diverse sample (48.2% Hispanic and 8.4% Black). Related to generalizability, this study was conducted during the pandemic in which the diagnosis of depression may have been very different due to symptom presentation and severity, which in turn could explain any differences in treatment practices independent of the visit type. Further studies are needed to confirm that the present study’s findings are durable in the absence of a worldwide pandemic.

Conclusions

To the authors’ knowledge, no other published study has examined the factors contributing to differences in processes of depression care in primary care settings across in-person and virtual treatment modalities. This study’s findings highlight clear gaps in care that might result from a move to mostly virtual visits when diagnosing and treating patients in a primary care setting. Patients who had their depression diagnosed during a telephone or video visit were less likely to receive the medication that was prescribed and less likely to have a follow-up visit with the prescribing practitioner when compared to patients who had an in-person visit. As care shifts toward virtual modalities, all processes of care will need to be adjusted accordingly, including delivery of medications. Mail-order or virtual pharmacies will be critical for this. Future research will be needed to understand best practices in virtual care provision and symptom monitoring and whether these practices affect depression outcomes.

Supplementary Material

Online supplementary file 1: