![Sample our Behavioral Sciences journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/110801/TOC-Subject-Sample-2021-Behavioral-Sciences.jpg"})
Abstract
Objective: Differences between therapists in their average outcomes (i.e., therapist effects) have become a topic of increasing interest in psychotherapy research in the past decade. Relatively little work, however, has moved beyond identifying the presence of significant between-therapist variability in patient outcomes. The current study sought to examine the ways in which therapist effects emerge over the course of time in psychotherapy. Method: We used a large psychotherapy data set (n = 5828 patients seen by n = 158 therapists for 50,048 sessions of psychotherapy) and examined whether outcomes diverge for high-performing (HP) and low-performing (LP) therapists as treatment duration increases. Results: Therapists accounted for a small but significant proportion of variance in patient outcomes that was not explained by differences between therapists' caseload characteristics. The discrepancy in outcomes between HP and LP therapists increased as treatment duration increased (interaction coefficient = 0.071, p < .001). In addition, patients' trajectories of change were a function of their therapist's average outcome as well as the patient's duration of treatment (interaction coefficient = 0.060, p = .040). Conclusions: Indeed, patterns of change previously described ignoring between-therapist differences (e.g., dose-effect, good-enough level model) may vary systematically when disaggregated by therapist effect.
Resumo
Objetivo: Diferenças entre terapeutas em seus resultados médios (isto é, efeitos do terapeuta) tornaram-se um tópico de interesse em pesquisa em psicoterapia na última década. Relativamente poucos trabalhos, entretanto, foram além da identificação da presença de variabilidade significativa entre terapeutas nos resultados dos pacientes. O presente estudo procurou examinar as formas em que os efeitos do terapeuta emergem do longo do tempo na psicoterapia. Método: Utilizamos um grande bando de dados de psicoterapia (n=5828 pacientes atendidos por n=158 terapeutas para 50.048 sessões de psicoterapia) e examinamos se os resultados divergem para terapeutas de alto desempenho (AD) e de baixo desempenho (BD) à medida que a duração do tratamento aumenta. Resultados: Terapeutas explicaram uma pequena mas significativa proporção de variância nos resultados dos pacientes que não foi explicada pelas diferenças entre características dos casos dos terapeutas. A discrepância nos resultados entre os terapeutas AD e BD aumentou na medida em que a duração do tratamento aumentou (coeficiente de interação=0,071, p <0,001). Além disso, as trajetórias de mudança dos pacientes eram uma função da média de resultados do seu terapeuta, bem como a duração do tratamento do paciente (coeficiente de interação=0,060,p=0,040). Conclusões: De fato, padrões de mudança descritos anteriormente ignorando diferenças entre terapeutas (por exemplo, dose-efeito, modelo de nível suficientemente bom) podem variar sistematicamente quando desagregados pelo efeito do terapeuta.
Zusammenfassung
Ziel: Unterschiede zwischen Therapeuten bezüglich ihrer durchschnittlichen Ergebnisse (d.h. Therapeuteneffekte) haben im letzten Jahrzehnt in der Psychotherapieforschung zunehmend an Interesse gewonnen. Relativ wenig Forschung ist jedoch darüber hinausgegangen, das Vorhandensein signifikanter Unterschiede zwischen den Therapeuten bei den Patientenergebnissen festzustellen. In der vorliegenden Studie wurde untersucht, wie sich Therapeuteneffekte im Laufe der Zeit in der Psychotherapie herausbilden. Methode: Wir verwendeten einen großen psychotherapeutischen Datensatz (n=5828 Patienten gesehen von n=158 Therapeuten in 50.048 Psychotherapiesitzungen) und untersuchten, ob die Ergebnisse bei hochleistungsfähigen (HP) und leistungsschwachen (LP) Therapeuten divergieren, wenn die Behandlungsdauer zunimmt. Ergebnisse: Die Therapeuten konnten einen kleinen, aber signifikanten Anteil an der Varianz der Patientenergebnisse aufklären, der nicht durch Unterschiede in den Falldaten der Therapeuten erklärt werden konnte. Die Diskrepanz in den Ergebnissen zwischen HP- und LP- Therapeuten stieg mit zunehmender Behandlungsdauer (Interaktionskoeffizient=0,071, p <0,001). Darüber hinaus waren die Veränderungsverläufe der Patienten eine Funktion des durchschnittlichen Ergebnisses des Therapeuten sowie der Behandlungsdauer des Patienten (Interaktionskoeffizient=0,060, p=0,040). Schlussfolgerungen: In der Tat könnten Veränderungsmuster, die in vorherigen Beschreibungen Zwischen-Therapeuten-Unterschiede unbeachtet ließen (z.B. Dosis-Wirkung, Good-Enough-Level-Modell) systematisch variieren, wenn sie durch den Therapeuten-Effekt disaggregiert werden.
摘要
目的：過去十年，心理治療研究對治療師平均療效的差異主題(即治療師效果)越來越感興趣，但是從病人療效找出不同治療師之間的主要變異性之研究相對仍少，本研究乃探討隨著心理治療進程治療師效果如何展現。方法：使用一個大型心理治療資料庫(5828位病人，由158位治療師進行共50,048次心理治療)，並檢視隨著處遇的時間增加，高效能(HP)與低效能(LP)治療師的療效是否變得不同。結果：治療師因素在病人療效上可解釋的變異量雖小，但是是顯著的，此變異無法由治療師之間個案量等特徵的差異所解釋。HP與LP治療師在療效上的差異，隨著治療時間而增加(交互作用係數=0.071, p<.001)，並且，病人的變化軌跡隨治療師之平均療效與病人治療長度而異(交互作用係數=0.060, p=.040)。結論：過去所描述的變化型態的確忽略了治療師之間的差異(例如，次數—效果、恰到好處模式)，而在依治療師效果分別看時，其可能會有系統性地變動。
Obiettivo: le differenze tra terapeuti nei loro risultati medi (cioè gli effetti del terapeuta) sono diventate un argomento di crescente interesse nella ricerca in psicoterapia negli ultimi dieci anni. Relativamente pochi lavori, tuttavia, hanno tentato di andare oltre l'osservazione della presenza di una significativa variabilità tra terapeuti negli esiti dei pazienti. Questo studio ha tentato di esaminare i modi in cui emergono gli effetti del terapeuta nel corso del tempo in psicoterapia. Metodo: Abbiamo utilizzato un ampio set di dati di psicoterapie (n=5828 pazienti visti da n=158 terapeuti per 50.048 sedute di psicoterapia) ed esaminato se i risultati divergono tra terapeuti ad alte prestazioni (HP) e con basse prestazioni (LP) man mano che aumenta la durata del trattamento. Risultati: i terapeuti erano responsabili di una piccola ma significativa proporzione della varianza negli esiti dei pazienti che non era spiegata dalle differenze tra le caratteristiche del carico di lavoro dei terapeuti. La discrepanza nei risultati tra terapisti HP e LP aumentava con l'aumentare della durata del trattamento (coefficiente di interazione=0,071, p <0,001). Inoltre, le traiettorie di cambiamento dei pazienti erano una funzione del risultato medio del loro terapeuta e della durata del trattamento del paziente (coefficiente di interazione=0,060, p=0,040). Conclusioni: visti i risultati, i modelli di cambiamento precedentemente descritti che ignorano le differenze tra terapeuta (ad es. effetto dose, modello di livello sufficientemente buono) possono variare sistematicamente quando disaggregate per effetto del terapeuta.
Acknowledgements
A previous version of this paper was presented at the 46th International Meeting of the Society for Psychotherapy Research (SPR), June 2015, Philadelphia, PA, USA.
Notes
1 Of note, we refer here to dichotomous groups of HP and LP therapists and later to short versus long courses of treatment. Importantly, these groupings are referenced simply for rhetorical purposes. Both therapists' average effectiveness and treatment duration were treated continuously in all models.
2 Univariate descriptive statistics were computed on patient-level outcomes (within-patient d) in order to assess the presence of outliers. As a method for addressing outliers, we employed a typical three standard deviation cut-off. Although one would expect some percentage of cases to be this deviant from the mean, given the large sample, it seemed worthwhile employing a standard cut-off that was likely to exclude cases that may be artifacts of estimation procedures and data entry errors. In order to assess the impact of outliers, primary analyses were run with and without these patients excluded.
3 No differences were noted in results from primary analyses with or without outliers included.
4 The distribution of within-patient ds was relatively normally distributed with some evidence of a positive skew based on inspection of a QQ-normal plot.
5 The ICC was also computed in a model predicting post-test OQ scores controlling for pre-test OQ scores. The ICC was similar with this method (ICC = 0.011).
6 An additional model was fit with a random slope term that allowed the relationship between therapists' aggregate d and within-patient d to vary across therapists. A χ2 log-likelihood ratio test was used to assess improvement in model fit. The random slope model showed no indication of improved fit (χ2 < 0.01, p = .999) thus the random intercept model was used.
7 An alternative method was also used to assess the impact of treatment length depending on therapists' overall outcome. This was done due to concerns of over-fitting the two-level MLM by including a predictor variable (therapists' average outcome) that was statistically composed of the outcome (within-patient ds). Specifically, we examined the correlation between the random slope and the random intercept in a model that included treatment duration as the only predictor of within-patient ds. We were interested in interpreting the slope-intercept correlation. This correlation can be interpreted as reflecting the relationship between therapists' average outcome (i.e., the random intercepts) and the model's random slopes (which reflect the relationship between duration and patient outcome). In this model, treatment duration was centered within therapist (so it could be interpreted as reflecting the intercept for treatment duration for each therapist's average patient, rather than the intercept for a treatment duration of zero). The slope–intercept correlation in this model was quite large, r = 0.70. This correlation is consistent with the two-way interaction reported between treatment duration and therapists' average outcome, with HP therapists (i.e., with higher intercepts) showing larger reductions in symptom over time (i.e., higher slopes). As the “nlme” package does not report a confidence interval for the slope–intercept correlation, we ran a series of bootstrapped replications (n = 10,000) with replacement. The empirical 95% CI from this was [−0.01, 0.67] which was a marginally significant effect. Of note, this test may be underpowered, specifically due to the need to center the treatment length variable within therapists. This means that the intercept reflects the therapist's average effect, but not very exactly. For brief-treatment dyads, the intercept reflects a forecast of client status at a session not actually measured. For long-term dyads, the intercept reflects an intermediate outcome sometime during treatment. As the therapist variance increased with treatment length in this data set, this approach gives too little weight to longer term treatment outcomes, so likely attenuates the correlation between therapists' average outcome and treatment duration. Based on comments from an anonymous reviewer, an additional model was run that included a quadratic term for length of treatment. The interaction noted previously between therapists' aggregate d and length of treatment remained essentially unchanged (estimate = 0.070, p < .001) when the quadratic term was included. Further, in a model that added an interaction between this quadratic term and therapists' aggregate d found the interaction to be non-significant (estimate = 0.00033, p = .858).
8 The interaction remained significant when the covariates were entered individually as well.
9 It is also worth noting that a small ICC generally biases against the reported findings. A small amount of between-therapist variance relative to total variance reduces statistical power and may reflect a restricted range of between-therapist outcomes. Thus, one could interpret this small ICC as a conservative (rather than liberal) source of bias. The small ICC does not, however, necessarily bias against detecting the current findings. For example, an unusually low amount of between-therapist variance relative to total variance in the brief courses of treatment (e.g., three to four sessions) but not in the longer courses of treatment could influence the observed findings. Thus, it is important to replicate the current findings in another sample, ideally one with more typical therapist effects (e.g., Schiefele et al., Citation2016).