記住我
Prior to the COVID-19 pandemic, the lifetime prevalence of major depressive disorder (MDD) was over 20% for adults in the United States [], and the majority (71%) of cases were untreated []. Compared with the prepandemic period, depressive symptoms became over three times more prevalent [] during the pandemic, with up to 48% of citizens of developed nations reporting clinically significant depression []. At the same time, pandemic constraints critically challenged the provision of mental health services. Cost-effective, scalable, affordable, and accessible interventions were urgently needed, and the use of virtual care expanded quickly []. However, the efficacy of modern virtual interventions had not been systematically examined. Thus, the aim of this systematic review and meta-analysis was to fill this gap to inform clinical, administrative, and policy decision-making.
Prior systematic reviews and meta-analyses examined the evidence supporting the efficacy of computerized or virtual cognitive behavioral therapy (CBT) for MDD or depressive symptoms compared with no treatment or treatment as usual (TAU) (ie, referring participants to primary care providers or other health clinics in their local community to manage their depressive symptoms). Moreover, meta-analyses [] and umbrella summaries across meta-analyses [] have suggested that virtual treatment works at least as well as in-person treatment for those with depressive symptoms. Prior meta-analyses of virtual treatments for adults included studies conducted before 2016, and many included adults with depressive symptoms or various depression and anxiety diagnoses [,-]. Since 2016, individual studies of virtual interventions have proliferated, expanding beyond CBT [,], and increased in rigor. As a result, a comparison of modern virtual interventions with not only waitlist or TAU but also with face-to-face interventions [,] for adults with MDD is feasible and warranted given that face-to-face psychotherapy had become impractical and, in certain settings, impossible.
Information evaluating whether virtual care is an efficacious alternative to individual, face-to-face intervention with a therapist is needed for clinicians, health systems, payers, and policymakers. In addition, data to guide decisions about which existing virtual interventions are most efficacious for treating depressive disorder are lacking. In the absence of such data, common assumptions about the superiority of in-person treatment have guided clinical decisions and policies regarding depression treatment.
The purpose of this systematic review was to answer three clinically relevant key questions (KQs) for depressive disorders (ie, MDD, persistent depressive disorder, or dysthymia) based on studies conducted in the last 10 years.
KQ1: Does virtual intervention provide better clinical outcomes than no treatment, TAU, or attention control, defined as a rigorous control condition that simulates active treatment without the active ingredient (ie, does it work)?
KQ2: Does in-person intervention provide better outcomes than virtual intervention (ie, is in-person intervention better)?
KQ3: Does one type of virtual intervention provide better outcomes than another type of virtual intervention (ie, what works best)?
The KQs were structured based on Agency for Healthcare and Research Quality (AHRQ) guidance for decision-making related to best practices in treatment [].
We used the Cochrane Handbook for Systematic Reviews of Interventions methods [] and AHRQ guidance for grading the strength of evidence []. The protocol for this systematic review and meta-analysis is published in the Open Science Framework []. Reporting conforms to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [].
Search Strategy and Selection CriteriaFor this systematic review and meta-analysis, we searched the PubMed, EMBASE, and PsycINFO databases for trials published from January 1, 2010, to October 30, 2021, for MeSH (Medical Subject Headings) and major headings listed in Table A1 of . Relevant systematic reviews and meta-analyses were used to identify additional existing literature, and ClinicalTrials.gov was searched to identify unpublished trials.
The study criteria were selected to inform clinical decision-making and policy in the United States. Eligible studies were randomized controlled trials (RCTs) of adults with a clinical diagnosis of MDD, dysthymic disorder, or persistent depressive disorder that tested a virtual psychological intervention for depression in at least one study arm, reported an outcome using a validated depression measure (see Table A2 of ), and were conducted in countries with a very high human development index (see Table 3 of for a list of eligible countries, []). To ensure generalizability of the results to individuals with major depression with access to current technology in the United States, we included studies conducted in similarly highly developed nations. To ensure comparability across studies, we included studies with standard, validated measures of depression, both self-reported and clinician-rated. Because evidence-based treatments for depression differ for children and adults, we excluded studies of children from this review.
References identified through searches were imported into Covidence Systematic Review software (Veritas Health Innovation, Melbourne, Australia). Two reviewers independently screened the titles and abstracts of all references according to the inclusion and exclusion criteria. Studies included by either reviewer were retrieved for full-text screening by two independent reviewers for eligibility. Discrepancies between reviewers were resolved through discussions and consensus.
Data ExtractionOne author extracted summary data from the included trials into standardized forms, and a second senior author (BNG, CES, or LL) checked the data for accuracy. Two authors independently rated the risk of bias across nine domains (see Table A4 in ) using the Cochrane Risk of Bias tool for RCTs [] modified for psychotherapy outcome research []. Disagreements were resolved by discussion and consensus. Trials with a high risk of bias were excluded, although sensitivity analyses were performed to determine the impact on the results (see Figure A1 in ).
Data Synthesis and AnalysisPrimary outcomes were rates of remission (ie, no longer meeting the clinical cutoff for depression), rates of response (ie, a clinically significant reduction in depressive symptoms), and depression severity at posttreatment. We calculated odds ratios (ORs) with 95% CIs for remission and response, and calculated standardized mean differences (SMDs, Cohen d) with 95% CIs for differences in symptom severity between groups. Forest plots were generated for all outcomes with sufficient data.
To determine the appropriateness of quantitative analyses, the senior authors (BNG, CES, LL, AEB) assessed the clinical and methodological heterogeneity of studies under consideration []. We performed meta-analyses using the meta package (v 4.19-2) in R version 3.6.1 [] when two or more trials reported data on outcomes of interest with low levels of heterogeneity. Effect sizes were weighted by their inverse variance. To account for variability in the different study populations, we used random-effects models to estimate pooled or comparative effects with three or more studies. However, because the effect estimates from smaller studies (which are generally more prone to bias) are more influential in random-effects models, we used fixed-effects models in analyses with fewer than three studies to ensure that a small study would not overinfluence the estimates [].
Statistical heterogeneity in effects between studies included in each meta-analysis was assessed by calculating the χ2 statistic (Q) and the I2 statistic, assessing the proportion of variation in study estimates due to heterogeneity rather than sampling error []. In instances of high heterogeneity, we performed sensitivity analyses to determine the extent to which excluding dissimilar studies changed the overall effect estimates. Most studies had only two study arms (ie, intervention and control); however, two studies had two intervention arms in addition to a waitlist control arm [,]. A two-level model was used if there was no significant difference between the three-level model and a two-level model based on a likelihood ratio test []. Using AHRQ guidelines [], we assessed the overall strength of evidence (SOE) considering four factors: directness, consistency, precision, and bias. SOE was assessed by one author (LL) and checked for consensus with two other authors (BNG and CES). We began each SOE assessment with a rating of high and downgraded for factors that reduced the level of confidence. The resulting definitions of high, moderate, low, and insufficient SOE grades are summarized in Table A5 of .
Database and manual searching yielded 3797 citations for consideration (), 24 of which met the criteria for inclusion in this review. The characteristics of participants included in each study are summarized in . Participants in all of the included trials were diagnosed with MDD, and those with severe psychiatric comorbidities such as any psychotic disorder or active substance use disorder, bipolar disorder, or acute risk of suicidality were excluded.
Figure 1. Study selection. *See Table A2 in . View this figureTable 1.
Participant characteristics in each included trial.ReferenceaMDDb diagnostic measurecAverage MDD severity at baselineIntervention conditionComparison conditionAge (years), mean (SD)Women, n (%)Some colleged, n (%)aEach row represents an intervention arm. Some references are listed more than once because they provided data from multiple intervention arms.
bMDD: major depressive disorder.
cParticipants of all trials were diagnosed with MDD.
dSome college means any self-reported level of educational attainment greater than high school or equivalent.
eKQ: key question.
fMINI: Mini International Neuropsychiatric Interview.
gSCID: Structured Clinical Interview for Axis-I Disorders.
hNR: not reported.
iCIDI: Composite International Diagnostic Interview.
jTAU: treatment as usual.
kAll participants were diagnosed with MDD with perinatal onset.
lICD-10: International Classification of Diseases, 10th revision.
mIncluded in systematic review but excluded from meta-analysis due to differences in methods from other studies.
nHAMD: Hamilton Rating Scale for Depression.
The characteristics of each trial, including the length of intervention, treatment modality and mode, provider type, and comparison condition, are summarized in .
Risk of bias assessments across the nine individual domains and an overall summary is presented for each study in Table A4 of ; detailed information on intervention outcomes is presented in -; and SOE ratings alongside a summary of results are presented in .
Table 2. Trial characteristics.ReferenceaLength of intervention (weeks)Intervention conditionComparison conditionaEach row represents an intervention arm. Some references are listed more than once because they provided data from multiple intervention arms.
bKQ: key question.
cCBT: cognitive behavioral therapy.
dNA: not applicable.
eACT: acceptance and commitment therapy.
fBA: behavioral activation.
gTAU: treatment as usual.
hIPT: interpersonal therapy.
iOBGYN: obstetrician/gynecologist.
jPCP: primary care provider.
kVA: Veteran’s Administration.
lPsychoed: psychoeducation.
mIncluded in systematic review but excluded from meta-analysis due to differences in methods from other studies.
Figure 2. Forest plots of virtual intervention compared with waitlist control clinical outcomes. ΔBDI: Change in Beck Depression Inventory Score; ΔPHQ: Change in Patient Health Questionnaire-9 Score; BDI: Beck Depression Inventory; CES-D: Center for Epidemiologic Studies Depression Scale; MADRS-SR: Montgomery–Åsberg Depression Rating Scale – Self-Report Questionnaire; MDD: Major Depressive Disorder; PHQ-9: Patient Health Questionnaire-9. View this figure
Figure 3. Forest plots of virtual intervention compared with treatment as usual (TAU) clinical outcomes. ΔMADRS: Change in Montgomery–Åsberg Depression Rating Scale Score; ΔMADRS-SR: Change in Montgomery–Åsberg Depression Rating Scale – Self-Report Questionnaire Score; BDI: Beck Depression Inventory; EPDS: Edinburgh Postnatal Depression Scale; MDD: Major Depressive Disorder; MADRS: Montgomery–Åsberg Depression Rating Scale Interview; MADRS-SR: Montgomery–Åsberg Depression Rating Scale – Self-Report Questionnaire; PHQ-9: Patient Health Questionnaire-9. View this figure
Figure 4. Forest plots for virtual intervention (internet-based cognitive behavioral therapy [iCBT]) compared with virtual sham intervention clinical outcomes. ΔBDI: Change in Beck Depression Inventory Score; BDI: Beck Depression Inventory; HAMD: Hamilton Depression Rating Scale; IDS-SR: Inventory for Depressive Symptomatology – Self-Report; MDD: Major Depressive Disorder; SCID: Semi-Structured Clinical Interview for DSM Disorders. View this figure
Figure 5. Forest plots for in-person intervention compared with virtual intervention clinical outcomes (key question 2). ΔBDI: Change in Beck Depression Inventory Score; ΔHAMD: Change in Hamilton Depression Rating Scale Score; ΔMADRS: Change in Montgomery–Åsberg Depression Rating Scale Score; BDI: Beck Depression Inventory; HAMD: Hamilton Depression Rating Scale; MDD: Major Depressive Disorder; SCID: Semi-Structured Clinical Interview for DSM Disorders. View this figureTable 3.
Strength of evidence for each outcome organized by key question (KQ).OutcomeStudy design, duration, sample size (N), eventsa (n)Effect size (95% CI)Factors that affect the strength of evidenceOverall evidence strength and direction of effectFindingsKQ1a (virtual vs waitlist)aBased on risk of bias, precision of estimate, directness of comparison, and consistency.
bRCT: randomized controlled trial.
cOR: odds ratio.
dROB: risk of bias.
eImprecision is based on the number of events <300 events, or n=400 for continuous events or very wide confidence intervals; precision was the primary variable that influenced strength of evidence ratings given that most trials had low risk of bias and were direct and consistent.
fSOE: strength of evidence.
gInconsistent was based on I2>75%.
hSMD: standardized mean difference.
iTAU: treatment as usual.
jCBT: cognitive behavioral therapy.
Efficacy of Virtual Intervention Versus Waitlist Control (KQ1a)The efficacy of virtual interventions compared with waitlist was assessed in seven double-blinded RCTs [,-] (). Most trials compared virtual CBT with waitlist [,,,,]: four trials examined virtual CBT guided by mental health providers [,,,], two examined unguided virtual CBT [,], and one examined CBT provided via email []. Two studies examined virtual adaptations of evidence-based therapies other than CBT (ie, combined acceptance and commitment therapy and behavioral activation [BA] [] and problem-solving therapy []).
Remission was evaluated in four trials [,-]. Two of the trials included comparisons of two different intervention arms against waitlist control groups: Berger et al [] examined both guided and unguided virtual CBT compared with waitlist, and Vernmark et al [] examined both guided virtual CBT and CBT provided via email compared with waitlist. Meta-analysis including a total of five comparisons across three studies showed that the odds of remission were 10 times higher (95% CI 5.70-18.60; N=619; high SOE) with virtual intervention compared with waitlist (). Response was measured in three separate comparisons across two studies [,]. The odds of response did not substantially differ between virtual interventions and waitlist (OR 3.57, 95% CI 0.86-14.87; N=221; low SOE). Depression severity at posttreatment was assessed in seven trials [11,22–27]. Virtual interventions resulted in lower depression severity at posttreatment compared with waitlist (SMD 0.81, 95% CI 0.52-1.10; N=1071; moderate SOE).
Efficacy of Virtual Intervention Versus TAU (KQ1b)Efficacy of virtual interventions compared with TAU was evaluated in eight double-blinded RCTs [,-] (). Three of the trials focused on interventions for specific populations: those with perinatal-onset MDD [,,], a majority-male veteran cohort [], and a Latinx Spanish-speaking population [] (). Most virtual interventions involved guided virtual CBT [-,-]; however, two trials provided synchronous telehealth interventions, including interpersonal therapy delivered by nurses via telephone [] and supportive therapy plus pharmacotherapy provided by a psychiatrist via video visits []. Most TAU study arms consisted of primary care appointments, scheduled on an as-needed basis delivered by physicians [-,].
Remission was evaluated in four trials [,,,]. The odds of remission were two times higher with virtual intervention compared with TAU (OR 2.27, 95% CI 1.54-3.35; N=512; moderate SOE) (). Response was evaluated in three trials [,,]. The odds of response were nearly three times higher with virtual intervention compared with TAU (OR 2.95, 95% CI 1.51-5.75; N=450; moderate SOE). Depression severity was evaluated in seven trials [,-]. Virtual intervention resulted in a lower depression severity at posttreatment compared with TAU (SMD 0.59, 95% CI 0.13-1.05; N=1533; moderate SOE).
Efficacy of Virtual Therapy Versus Attention Control (KQ1c)Five trials compared a virtual adaptation of an evidence-based intervention (eg, CBT [,,], BA [], or psychodynamic therapy []) with a virtual control (ie, mindfulness []) or sham condition (). Of these, three studies compared virtual CBT with attention control conditions, which included online psychoeducation [,] and progressive muscle relaxation []. These three studies were included in one set of meta-analyses based on the consistency in interventions (virtual CBT) and attention control conditions ().
Remission and response were assessed in two trials [,], both of which compared virtual CBT to virtual psychoeducation and favored virtual CBT in terms of both remission and response (). The odds of remission were higher with virtual CBT compared with virtual psychoeducation (OR 1.92, 95% CI 1.10-3.35; N=226; low SOE), whereas there was no statistically significant difference in response rates between virtual CBT and virtual psychoeducation (OR 1.68, 95% CI 0.96-2.96; N=226; low SOE). Depression severity at posttreatment was evaluated in three trials [,,]. All three studies favored virtual CBT compared with an attention control condition. Virtual CBT resulted in lower depression severity at posttreatment compared with virtual psychoeducation (SMD 0.25, 95% CI 0.09-0.42; N=573; high SOE).
Efficacy of In-Person Versus Virtual Intervention (KQ2)Efficacy of in-person compared with virtual delivery of behavioral therapy, either CBT or BA, was evaluated in four RCTs [,-] (), three of which were noninferiority trials [-] powered to evaluate whether virtual therapy provides at least the same benefit to the patient as in-person therapy. Two of the virtual interventions were guided virtual CBT [,] and two were synchronous telehealth interventions [,]. One trial compared in-person group CBT with virtual CBT [].
Remission was evaluated in four trials [,,]. In no study did remission rates for the in-person intervention exceed those seen in virtual interventions (). Indeed, one trial reported significantly lower remission rates in the in-person CBT groups compared with virtual CBT (19% vs 52%; P<.005) []. The odds of remission with the in-person intervention were not higher than those with the virtual intervention (OR 0.84, 95% CI 0.51-1.37; N=789; moderate SOE). Sensitivity analysis excludi
留言 (0)