How best to study the influence of duration of psychotherapy

Observational and experimental studies indicate that talking therapies delivered in routine clinical practice usually last between six and ten appointments. Is this enough therapy and how would we know if it were? These questions are best addressed in an experimental design but randomised trials are carried out under controlled conditions wherein the duration of therapy in each arm is relatively fixed. That is not how it works in clinical practice and, although such trials are often rigorous, their results may lack external validity.

Generalisability is higher in observational research in which a wider variety of patients agree to undergo therapy. The main problem with clinical cohort studies is that they do not take account of the multitude of other factors that influence treatment outcomes. Randomisation is the only sure way to eliminate such bias. Furthermore, as pointed out some years ago, ^{Reference Howard, Kopta, Krause and Orlinsky2} both trials and observational studies provide answers in terms of mean responses to treatment in samples of patients. Neither design answers the key practitioner question, is this treatment working for this patient?

Shorter treatments, better outcome?

In this issue, Stiles and his colleagues ^{Reference Stiles, Barkham and Wheeler3} have conducted an interesting analysis of routine data collected from a cohort of 26 430 patients, who had received a range of talking therapies, undergone a planned end to that therapy and completed an outcome questionnaire at baseline and end of therapy. The therapy took place in primary and secondary care settings, universities, places of work and the voluntary sector in the UK. Patients attended a mean of 8.3 sessions in which almost half received more than one type of therapy. Their key finding is that duration of therapy had little relationship to recovery and this was the case in most therapy settings. In fact, shorter duration of treatment was positively associated with recovery. But as Stiles et al point out, given this was not an experimental study we cannot conclude that short treatments are best. As patients get better, they and their therapists agree to stop meeting. What Stiles et al call ‘responsive regulation of treatment duration’ is a sensible agreement that no more help is needed. Most professionals (given resource constraints on therapy in any health setting) work with patients until they are better without paying overly close attention to the duration of treatment.

Other approaches to cohort analysis

In what is now regarded as a classical systematic review and meta-analysis of randomised trials of psychotherapy conducted almost four decades ago, Smith and her colleagues ^{Reference Smith and Glass4} concluded that neither type nor duration of therapy had much impact on outcome. Successive trials and observational studies have largely confirmed that the duration of therapy is not closely associated with outcome, ^{Reference Molenaar, Boom, Peen, Schoevers, Van and Dekker5,Reference Knekt, Lindfors, Håkänen, Välikoski, Virtala and Laaksonen6} except in the more severely affected. However, meta-analyses of trials or large observational studies like that of Stiles et al, beg the question of which therapy fits which patient. ^{Reference Stiles, Barkham and Wheeler3} One would presume that some degree of matching of type and duration of therapy to the severity and nature of each patient's presentation will have occurred in most large observational cohorts. There is a suggestion of this in the data in that patients who had longer treatments were more likely to have higher Clinical Outcomes in Routine Evaluation – Outcome Measure (CORE-OM) scores at the outset. Confounding by indication is a technical way of saying that therapists make choices at the outset about how often and how long to see patients based on their clinical judgement and the degree of the patient's distress. However, little of this can be accounted for in this analysis. Furthermore, given that patients received a wide variety of therapies (in fact 41% of patients had more than one type), change may have been solely because of a non-specific therapist presence, regardless of the therapy delivered. ^{Reference Blow, Sprenkle and Davis7,Reference Martin, Garske and Davis8} An alternative approach using more of the data available might have been a multivariable analysis of the factors that predicted treatment duration, planned discharge or recovery, and that took into account clustering by therapist or service. In this way, such things as diagnosis, patients' characteristics, severity of symptoms at baseline and type of treatment provided might have given a more detailed answer to what determined treatment outcomes, however measured. An approach in which the four domains of the CORE-OM are treated as a multivariate (as opposed to composite) outcome may also have produced a finer grained picture of the result of therapy. Regardless of the analytical approach, however, treatment data of this type are also (by necessity) limited by measurement of outcome at the end of therapy, and thus we know nothing about how or whether improvements were sustained. It may be the case that a longer duration of therapy does not improve immediate outcomes, whereas consolidation of the therapy in more sessions leads to more secure long-term gains. ^{Reference Knekt, Lindfors, Håkänen, Välikoski, Virtala and Laaksonen6,Reference Sandell, Blomberg, Lazar, Carlsson, Broberg and Schubert9}

Experimental v. observational research

In an interesting comparison of experimental with observational research, Hansen et al first reviewed what they described as a ‘representative sample’ of randomised clinical studies. ^{Reference Hansen, Lambert and Forman10} They reported that the average number of treatment sessions received by patients was 12.7 and that between 57 and 67% of patients responded to treatment of this duration. ^{Reference Hansen, Lambert and Forman10} They then conducted similar research in observational data arising from routine practice where they found that the median number of treatment sessions ranged from three to eight and only 10% of patients recovered. They concluded that most patients entering clinical services were getting undertreatment. This is at odds with Stiles et al's findings where 60% of patients achieved reliable and clinically significant change. ^{Reference Stiles, Barkham and Wheeler3} It may be, however, that not only treatment setting but also choice of patients studied affects the kinds of recovery observed in routine clinical practice. Stiles et al's sample contained only patients who had undergone a planned ending to therapy; data on a further 4716 who did not have a planned ending, despite having completed a CORE-OM, were left out of the analysis. Thus, one might imagine that the patients included were the most likely to have recovered.

What does it all mean?

These real-world data show that outcomes from psychotherapy are messy and difficult to predict. They indicate that in talking therapies, in which patients and therapists plan the ending of treatment, most patients get well irrespective of the duration of that treatment.