Your therapist may not notice when you're getting worse

If therapy is not helping you, or is quietly making things worse, both you and your therapist want to know early. Early is when there is still time to change course: to adjust the approach, to talk openly about what is not working, or to consider a different kind of help. The difficulty is that the research suggests your therapist may not notice on their own. In the studies that tested this directly, therapists identified very few of the clients who were getting worse.

I want to be clear from the start that this is not a post about bad therapists. It is about a blind spot that appears to be common across our profession, me included, and about practical ways of checking progress. The best-studied one is modest: a short questionnaire, filled in regularly and looked at together. It does not solve the problem, but it helps.

What the research found

One of the clearest tests comes from Hannan and colleagues (2005), at a university clinic. Forty-eight therapists were asked to predict which of their clients would end treatment worse off than when they started. Across 550 clients, the therapists predicted that 3 would deteriorate. In fact, 40 clients (7.3%) did. This happened even though the therapists had been told beforehand that the clinic's deterioration rate was about 8%. They knew that some clients get worse, yet they rarely recognised it in the clients they were seeing.

By contrast, a formal, algorithm-based method tested in the same study identified 100% of the deteriorated clients at termination, and 85% of them by the third session, according to the abstract. The 7.3% figure describes that one clinic, not a universal rate.

More recently, Østergård and colleagues (2024) found that 53 trainee therapists identified 1 of 8 clients who had deteriorated by the end of therapy, which fits the earlier pattern.

Why it may happen: we tend to rate ourselves highly

If therapists miss deterioration, the next question is why. Part of the answer may lie in how we see ourselves. In a survey of mental health professionals, Walfish and colleagues (2012) found that 25% rated their own skill at the 90th percentile or higher, and none rated themselves below average. The same clinicians also overestimated how much their clients improve and underestimated how often clients deteriorate, compared with published rates. A secondary summary of the study (therapymeetsnumbers.com) reports that the survey had 129 respondents and that the mean self-rating was at the 80th percentile. These two figures are not in the abstract, and I have not checked them against the full text, so please read them as second-hand.

Not everyone can be above average. This is a familiar human pattern, not a flaw peculiar to therapists, but it matters in our work because a clinician who believes clients rarely get worse has little reason to look for it.

Experience does not close the gap

It might seem that experience corrects this. The evidence suggests otherwise. Goldberg and colleagues (2016) followed 170 therapists and 6,591 clients in one clinical setting over time. On average, client outcomes did not improve as therapists gained experience. In fact, there was a very small but statistically significant decline.

That finding needs careful handling. The decline was very small, some therapists did improve, and early termination fell with experience. As such, the fair reading is not that experience makes therapists worse. It is that years in practice, on their own, do not reliably make us better at helping clients improve.

What helps, partly: routine progress feedback

If neither self-assessment nor experience closes the gap, it makes sense to gather information that does not depend on the therapist's impression. Routine progress feedback, also called routine outcome monitoring, refers to a client completing a brief, validated questionnaire on a regular basis, often every session, with the results reviewed over time so that both people can see whether things are changing.

The largest meta-analysis on this, by de Jong and colleagues (2021), pooled 58 studies and 21,699 patients. Feedback produced a small improvement in symptom reduction (d = 0.15) and a slightly larger one for clients who were not on track (d = 0.17). It also produced a small reduction in dropout (OR = 1.19). These are small standardised effects, not percentages. Importantly, the effect on deterioration was not significant: 5.4% of clients deteriorated without feedback and 4.6% with it (p = 0.07).

A second meta-analysis, by Lambert and colleagues (2018), looked at 24 studies of two well-known feedback systems, the OQ System and PCOMS. Two-thirds found that feedback-assisted therapy outperformed usual care by the same therapists, and the authors reported that feedback reduced deterioration. This review covers only two systems, and its lead author developed the OQ System.

Taken together, I would describe progress feedback as a partial fix. It appears to help, most clearly for clients who are off track, but the gains are small and the evidence on deterioration is mixed. It is not a cure, and I would not claim that it prevents deterioration.

Scores are one window, not the whole view

If a questionnaire is only a partial fix, what else belongs alongside it? In my view, progress measures help ground our sense of progress in actual data rather than impression alone. But I think the client's own self-report, and functional accounts from people who see the client outside the therapy room (a partner, family, friends, or someone at school or work), are just as valuable. Where we are unsure about a piece of information, it can be qualified rather than discarded: noted with its limits and weighed alongside the rest.

That is my clinical view, not a research finding, but research supports part of it. In a meta-analysis of adult studies, Achenbach and colleagues (2005) found that self-reports and reports from people who know the person agreed only moderately, with mean correlations ranging from about .30 to .68 depending on the problem area and the instruments used. A meta-analysis of 341 studies of children and adolescents by De Los Reyes and colleagues (2015) found low-to-moderate agreement between informants such as the young person, parents and teachers (mean r = .28 overall). The authors argue that much of this disagreement reflects real differences in how a young person behaves at home, at school and elsewhere.

Moderate agreement is the point. Because these views do not simply repeat each other, each adds information, and no single source (client, informant, therapist or questionnaire score) is the whole truth.

There is a second reason to look beyond symptom scores, which is what most progress questionnaires track. In a review of research on depression, McKnight and Kashdan (2009) found the relationship between symptoms and functioning to be weak and often bidirectional, and functional change often lagged behind symptom change. A symptom score can therefore miss how daily life, work and relationships are actually going.

In my own practice, I combine a scale or scaling questions, an objective questionnaire such as the PHQ-9 or GAD-7, the client's own self-reports, and descriptions and qualitative reports from people around the client outside the therapy room. I do not treat any one of these as the whole picture, for the reasons above. I only hear from people around a client with the client's informed consent, and the client decides who, if anyone, that is.

If you are in therapy

For clients, the practical point is that you do not have to rely on your therapist noticing everything, and your own account counts as real information. You can ask how progress will be checked, for example:

• "How will we know if this is working?"
• "Do you use any questionnaires to track progress?"
• "What happens if the scores show I'm not improving?"

A therapist who does not use a formal questionnaire is not necessarily doing poor work. These questions are not a test; they make progress something you look at together.

Feeling worse for a while can be part of therapy. A sustained slide is different, and it is worth raising directly. Saying "I don't think this is helping" is a normal and welcome part of therapy. It is not rude.

If you are a therapist

For clinicians, the first step is the hardest: assume you share the bias. I include myself in that.

From there, you may wish to consider using a brief, validated progress measure at every session. Examples include the OQ-45, the ORS and SRS used in PCOMS, the CORE-10, and the PHQ-9 and GAD-7 for specific symptoms. Look at the trend with the client rather than on your own, and read it alongside the client's own report and, where appropriate, accounts from people who see the client outside sessions. When a client is flagged as "not on track", treat that as a prompt to talk, not a verdict on the client or on you. Over time, it may also help to track your own aggregate outcomes rather than relying on years in practice. This is general information, not advice about any particular client or setting.

Limitations of this evidence

Each source has limits worth stating plainly.

• Walfish et al. (2012) was a self-report survey, compared with published rates rather than clinicians' own outcomes; the full text, with the authors' stated limitations, is paywalled. The 129 respondents and 80th percentile mean self-rating are second-hand (therapymeetsnumbers.com), not in the abstract, and unchecked against the full text of the original source, Walfish, S., McAlister, B., O'Donnell, P., & Lambert, M. J. (2012), Psychological Reports, 110(2), 639-644.
• Hannan et al. (2005) took place at one university clinic, with predictions gathered over only 3 weeks at different points in treatment (as noted by Østergård et al., 2024).
• Østergård et al. (2024) involved trainee therapists only, at one clinic.
• Goldberg et al. (2016) used naturalistic data from one setting. A 2023 correction (https://doi.org/10.1037/cou0000666) fixed reversed coding wording for early termination in the Method section only.
• de Jong et al. (2021) mixed randomised and non-randomised studies, its effects are small, and inconsistent reporting of moderators leaves it unclear when and why feedback works.
• Lambert et al. (2018) covers only two feedback systems, one-third of its studies did not find feedback superior, and the lead author developed the OQ System, a possible source of allegiance.
• Achenbach et al. (2005) rests on a thin evidence base: qualifying correlations were found in only 108 of about 51,000 articles screened (0.2%), from 52 journals over 10 years. The full text was not open access, so these limitations are taken from the abstract only.
• De Los Reyes et al. (2015) studied children and adolescents, so it applies to adults only by analogy. The authors note possible publication bias and that there are not yet established best practices for using multi-informant data in clinical work.
• McKnight and Kashdan (2009) focused on depression only. The evidence is correlational, and the authors flag measurement artifacts (such as floor and ceiling effects) that may distort the summary correlations.

Checking progress together

Taken as a whole, the evidence points to a human limit rather than a professional failing, and one that experience alone does not seem to correct. A short questionnaire, looked at together over time, appears to help, modestly, and in my view it works best as one window among several. I think that is worth saying openly, because it shifts the question from "Is my therapist good enough?" to "How are we checking whether this is working?" That second question is one that clients and clinicians can answer together.

This post is general information, not clinical advice. If you are in crisis in Canada, call or text 9-8-8. If you have concerns about your own therapy, I would encourage you to raise them with your therapist or another regulated professional.

References