Psychotherapists possess stable strengths and weaknesses in treating patients with different mental health problems; yet, such performance information (an effectiveness “report card”) is rarely harnessed in mental health care (MHC). We developed a multidimensional Match System for assigning patients to therapists who have empirically derived historical strengths in treating those patients' primary concerns.
Within a private community MHC consortium, our team (including a diverse stakeholder advisory board) compared the match with nonmeasurement-based case assignment as usual (CAU), hypothesizing a main effect of matching on the primary outcome of general symptomatic/functional impairment and 7 secondary outcomes (aim 1). Additionally, we evaluated moderators of matching (aim 2), hypothesizing that it would have an even greater effect for patient participants with more severe, complex problems. We also explored whether matching would be more beneficial for participants from racial and ethnic minority groups.
To prime the Match System before the trial, we used an established multistep method for assessing the performance of therapist participants (N = 48) across ≥ 15 historical cases based on these patients' pre- to posttreatment reporting on 12 symptomatic/functional domains measured by the Treatment Outcome Package (TOP), which had long been routinely administered in the MHC consortium: depression, quality of life, mania, panic/somatic anxiety, psychosis, substance use, social conflict, sexual functioning, sleep, suicidality, violence, and work functioning. For each of these outcome domains, prior research with a large outpatient reference sample used machine learning to establish normative, risk-adjusted change rates. Drawing on these reference data, every therapist in our sample was classified as “effective” (the therapist's average historical patient reliably exceeded the risk-adjusted threshold for expected change), “neutral” (the therapist's average historical patient met the threshold for expected change within a confidence interval), or “ineffective” (the therapist's average historical patient fell short of the threshold for expected change). Next, we recruited patient participants into the double-blind clinical trial. As the only exclusion criterion was patients not making their own MHC decisions, patients presented with varied mental health concerns. Of those screened at intake, 288 were randomly assigned to Match (n = 131) or CAU (n = 157). Matched patients were assigned to therapists at 1 of 5 match levels, ranging from effective on the patient's 3 most elevated TOP domains and not ineffective on any remaining TOP domain (highest) to not effective on the patient's single-most elevated domain but also not ineffective on any of the 12 TOP domains (lowest). CAU patients were assigned to therapists per usual pragmatic procedures. Therapists treated both Match and CAU patients, and treatment was delivered as usual. Patients completed study-related measurements at baseline and repeatedly through up to 16 weeks of treatment following baseline (ie, the trial's definition of posttreatment). For the primary outcome of general symptomatic/functional impairment, we averaged the scores (ie, SDs relative to the general population mean) across the 12 TOP subscales. Secondary outcomes were global psychological distress; domain-specific impairment (ie, most elevated/severe presenting problem); outcome consistency (ie, the degree to which matching reduced variability in patients' primary outcome); patient-therapist alliance quality; patient treatment outcome expectation (ie, belief that treatment can help); early treatment discontinuation (ie, attending <3 sessions); and overall patient-reported therapist satisfaction at termination. We conducted modified intent-to-treat analyses, excluding only those who withdrew consent, never began treatment, or provided no outcomes data beyond baseline. This approach resulted in an effective sample of 218 patients (Match, n = 99; CAU, n = 119). To account for data dependencies, we used multilevel models. For aim 1, we included a case-assignment arm as a predictor of between-patient (within-therapist) differences in posttreatment outcome levels and, for the outcomes assessed repeatedly, change over treatment. For aim 2, we included as predictors the relevant main effects and the arm × moderator interactions.
Most patients were female (67.4%) and White (88.5%), with a mean (SD) age of 33.9 (11.2) years. On average, patients demonstrated clinically significant impairment at baseline; for example, on their most elevated TOP domain, they scored almost 4 SDs above the mean of the general population. Match and CAU patients did not differ at baseline on any demographic or clinical variables. Patients provided a median (SD) of 12.14 (6.10) weeks of data, which did not differ between the arms. Compared with CAU, matching decreased weekly general impairment by a moderate to large degree (−0.03; 95% CI, −0.05 to −0.01; Cohen = 0.75). Compared with CAU, matching also decreased weekly global distress (−0.16; 95% CI, −0.30 to −0.02; Cohen = 0.50) and domain-specific impairment (−0.01; 95% CI, −0.01 to −0.006; Cohen = 0.63). Matching yielded moderately more consistent improvement in general impairment [χ(1) = 3.97; = .04; ϕ = 0.13] but did not outperform CAU on any other secondary outcome. The match effect on general impairment was more pronounced for patients with more severe problems, with greater problem complexity, and who identified as racial/ethnic minorities. No adverse events were reported.
Measurement-based matching to therapists' performance strengths can improve the outcomes of community-based, outpatient MHC delivered as usual.
The unmanipulated, as-usual treatments resulted in variable sessions/weeks attended and therefore some naturally occurring missing outcomes data. Additionally, the predominately White and moderately high-income sample as well as the lack of formal categorical diagnoses may limit generalizability.
