In order to review the available body of knowledge concerning patient-specific predictors for treatment failure, a literature search in PubMed, PsycInfo (OVID interface, 1806 onwards), Embase (OVID interface, 1974 onwards), and Medline, and Pre-Medline (OVID interface, 1946 onwards) was conducted with help from a research librarian. This search was aimed at articles that combined the following search criteria and terms using subject headings (MeSH for Medline, Emtree for Embase, and APA thesaurus for PsycInfo), if available: predictors, treatment failure, depressive or anxiety disorders, and psychotherapy. Treatment failure was further defined as poor outcome, non-response, drop-out, deterioration, or not successful as the search criteria. The literature search was confined to English, German, and Dutch. A draft of the search strategy in Embase is included in Table 1.

The last search was conducted on October 2, 2025, earlier searches were conducted in March 2022 and November 2017. We included studies examining a form of psychotherapy among adult populations (older than 18 years) with anxiety and depressive disorders.

We found 933 articles within this search. Figure 1 shows the PRISMA flow chart. Out of all these articles, the first author (VP) made a further selection. First, all titles and abstracts were checked for exclusion criteria. They were excluded if there was no mentioning of psychotherapy, anxiety, or depressive disorders, or prediction of outcome. They were also excluded if they only reported on child and adolescent studies.

Next, the remaining 169 articles were screened by reading the abstracts and the conclusions for meeting exclusion criteria. Articles were excluded based on the above-described exclusion criteria, and when predictors were not amenable or were measured during therapy and not at the start of therapy. Articles were also excluded when there was no mentioning of treatment failure or outcome. The remaining 95 papers were inspected more precisely by reading them fully, in order to check for reported amenable predictors of treatment failure. If it was not clear what kind of predictors were used and how they were measured, the methods and results sections were read carefully to inspect the analyses in order to find out if the variables could be labeled as predictors. Articles were included when they reported on treatment failure, predictors being amenable before therapy, and patient-specific, and when studies were based on prediction analyses. To decrease the risk of bias, in case of doubt, the second and fourth co-author also read the articles before including or excluding them. This was the case in 35 articles. In total, 30 papers met the inclusion criteria and were included in this review. Table 2 summarizes these articles.

The main outcome of this review will be the predictors of treatment failure. A predictor was included if the variable was amenable to change and patient-specific, such as motivation or treatment attitude. As such, biological markers, demographic variables, comorbidity, or severity of complaints were excluded. We included only predictors of treatment failure, not predictors of follow-up or risk of relapse.

To improve readability, we categorized the predictors into categories based on both their definition and the instrument used to measure them in the studies included in the review (e.g., Motivation measured with the URICA). For instance, seeking social or emotional support, as mentioned by Grilo et al. (1998) and Marquett et al. (2013), was defined as a coping style and measured with a coping questionnaire. In contrast, perceived social support, mentioned by Marquett et al. (2013), Schilling et al. (2021), and Solomonov et al. (2021), was measured with instruments focusing on how patients perceive social support. This construct was therefore considered distinct from coping. Because of this conceptual distinction, we placed these seemingly similar constructs in different categories to ensure that each predictor reflected the specific theoretical framework and measurement approach used in the studies. Through this process, we specified seven predictor categories and one category “other” for predictors that did not fit elsewhere. The identified predictor categories are personality factors, coping strategies, motivational aspects, attributions, treatment attitude and expectancies, interpersonal relationships, and others.

Six studies examined personality factors in relation to treatment failure. Three of them used large samples of patients with depressive disorders (Blom et al., 2007; Marquett et al., 2013; Solomonov et al., 2021). Solomonov et al. (2021) studied the Big Five personality traits in problem-solving therapy and supportive therapy for late-life depression (age > 60) with executive dysfunction. Blom et al. (2007) used a sample of depressive adults following interpersonal psychotherapy. Both Blom et al. (2007) and Solomonov et al. (2021) found that only higher scores on neuroticism are a predictor for treatment failure. The other personality traits of the Big Five (i.e., extraversion, openness to experiences, conscientiousness, and agreeableness) were not found to be predictive of treatment failure (defined as drop-out and early non-response; Solomonov et al., 2021). Marquett et al. (2013) found that a lower score on openness to experience is a predictor of treatment failure (no significant clinical improvement) in a small sample of late-life depression (age > 60) treated with CBT.

Three studies focused on personality styles or indicators instead of traits (Grilo et al., 1998; Hoyer et al., 2016; Lutz et al., 2019). Only a more histrionic personality style and a less obsessive personality style predicted a higher probability of treatment failure (drop-out) in CBT for several diagnoses, with mostly affective, personality, and anxiety disorders (Lutz et al., 2019). Other predictors like harm-avoidance, self-esteem, shame, novelty seeking, dominance, and reward dependence were not found as predictors of treatment failure (treatment response and drop-out) in cognitive therapy for social phobia (Hoyer et al., 2016). Neither were personality styles, self-reported features from personality disorders from an internal perspective, such as avoidant or narcissistic personality styles, predictors of treatment failure in cognitive behavioral therapy, medication therapy, or combination treatment for panic disorder (Grilo et al., 1998).

Several coping strategies were studied as predictors of outcome. It is remarkable that all studies used different instruments to measure coping styles or strategies.

The use of seeking social or emotional support is the only coping strategy that was found to be predictive of treatment failure (drop-out and no clinically significant improvement) in more than one study (Grilo et al., 1998; Marquett et al., 2013). Samples used in these studies were a small sample of depressive older adults (age >60) following CBT (Marquett et al., 2013) and patients with uncomplicated panic disorder or with mild agoraphobia following CBT, pharmacotherapy, or combined treatment.

Other coping strategies were only studied as a predictor in one article and in different patient samples. Of those strategies, only higher self-consolation (with behavior like spending money, eating more, and drinking more alcohol) was associated with treatment failure (a high likelihood of not improving) after psychotherapy in depressed patients (Parker et al., 1986). Other coping strategies turned out to be not predictive of treatment failure. The Grilo et al. (1998) study revealed that confrontive coping, distancing, self-control, acceptance of responsibility, escape avoidance, problem solving, positive reappraisal, problem focusing, self-blame, wishful thinking and avoidance were not associated with treatment failure (drop-out) in the treatment of patients with a panic disorder. Finally, Lin and Farber (2021) found that self-concealment (the tendency to conceal distressing information aiming at self-protection and avoidance of stigma) measured at the baseline of psychotherapy, was not predictive of treatment failure (non-improvement) in depressive complaints.

Findings about motivation as a predictor of treatment failure are contradictory. For instance, Keijsers et al. (1994) found that poor motivation predicted treatment failure (no improvement) in obsessive compulsive disorder (OCD) treatment. In line with these findings Schilling et al. (2021) found that a sudden drop in therapy motivation was predictive for treatment failure (deterioration of complaints) in CBT for different diagnosis, although therapy motivation did not differ between on-track and not on-track patients. However, motivation was found not to be predictive of treatment failure (no improvement) of CBT worry treatment (Westra, 2011). All three studies used different questionnaires for measuring motivation.

Readiness for change, measured with the URICA, is seen as a different conceptualization of motivation. However, readiness for change was not predictive of treatment failure (no improvement and drop-out) in small samples of CBT for OCD and anxiety disorders (Marker et al., 2019; Steketee et al., 2011; Vogel et al., 2006).

Finally, change talk and counter change talk i.e., patient statements indicating support or opposition to change, an observational measure of motivation (Marker et al., 2019), did not predict treatment failure (drop-out) during early sessions of therapy in which psychoeducation and cognitive restructuring were offered. However, low change talk did predict treatment failure (drop-out) during exposure sessions later in CBT for anxiety disorders. Next to that, Westra (2011) found that low motivation to change and higher resistance to therapy, considered by Westra (2011) to be an indirect measure of motivation, were predictive of treatment failure (no improvement) in a small GAD-sample treated with CBT.

There are only two studies on attributions as predictor of treatment failure. They both study a different sort of attribution. Attributing their panic disorder to life stressors predicted treatment failure (drop-out) for patients receiving treatment for panic disorder (Grilo et al., 1998). Perception of health, defined as the way people rate their overall physical health, did not predict treatment failure (non-response) in a group of late-life depressive patients who received IPT (Miller et al., 1996).

Several studies found that treatment expectancy did not predict treatment failure (Chambless et al., 1997; LeBeau et al., 2013; Lutz et al., 2019; Renaud et al., 2014; Safren et al., 1997; Steketee et al., 2011). These outcomes are replicated in several studies with big sample sizes and different diagnosis.

On the other hand, outcome expectancy is found as a predictor of treatment failure in multiple studies (Keefe et al., 2021; Schindler et al., 2013; Solomonov et al., 2021). Patients with a more negative outcome expectancy were found to be more likely to have treatment failure (drop-out) for depression (Schindler et al., 2013; Solomonov et al., 2021). In a study comparing three treatments for panic disorder (CBT, focused psychodynamic therapy and applied relaxation), Keefe et al. (2021) found that outcome expectancies measured at session two were not predictive of treatment failure (drop-out) in the overall study. However, they did find that in the CBT group, lower outcome expectancies at session two were predictive of treatment failure (drop-out), which was not found for the other treatment conditions. Contrary to these findings, Vogel et al. (2006) did not find outcome expectancy as a predictor of treatment failure in a small sample of OCD patients with overt compulsions.

There are also predictors related to treatment expectancies that were only found in one study but not replicated in other studies. Patients with lower expectations of a strong alliance had a higher risk at treatment failure (drop-out) during supportive-expressive therapy than during medication or placebo therapy (Zilcha-Mano et al., 2016). In the early phase of panic disorder treatment, in which psychoeducation and cognitive restructuring were offered, higher process expectations (i.e., expectations on the therapy process and role expectations) were predictive of treatment failure (drop-out), but anxiety expectations (i.e., expectations on having panic-related symptoms) were not. In the behavioral phase of treatment, where exposure and anxiety-provoking exercises took place, anxiety and process expectations were not predictive of treatment failure (drop-out) (Bélanger et al., 2017).

Treatment attitude implies the way people think of certain therapies or views they have on psychotherapy in general. Treatment attitude is studied in depressive and panic disorder patient populations with large sample sizes (Arndt et al., 2020; Grilo et al., 1998; Moradveisi et al., 2014). Moradveisi et al. (2014) found that treatment attitudes are no predictor of treatment failure. Contrary to that, treatment failure (attrition, or the likelihood of drop-out) could be predicted by less favorable or negative treatment attitudes towards the offered treatment, found in an internet-based intervention for depression and panic disorder treatment (Arndt et al., 2020; Grilo et al., 1998).

Interpersonal relationships are a broad concept and include for example social support and attachment characteristics. Several studies found aspects of interpersonal relationships to be predictive of treatment failure. The only predictor within this category studied in multiple studies is perceived social support (Marquett et al., 2013; Schilling et al., 2021; Solomonov et al., 2021). Whereas Schilling et al. (2021) and Marquett et al. (2013) did not find perceived social support as a predictor of treatment failure in depression and other diagnosis, Solomonov et al. (2021) found low perceived social support to be predictive of treatment failure (early non-response) in psychotherapy for late-life depression.

Other interpersonal factors that were found as predictors for treatment failure are emotional overinvolvement from relatives and hostility coming from relatives (Chambless & Steketee, 1999) as well as higher impairment of interpersonal relationships (Lutz et al., 2019). Both were found to predict higher rates of drop-out in CBT for several diagnosis. Higher vindictive tendencies in interpersonal relationships were found to be predictive of treatment failure (drop-out) in pharmacotherapy for depressive disorder, but not in other treatment groups (Zilcha-Mano et al., 2016). Lastly, Solomonov et al. (2021) found that perceptions of the therapist as less accepting are predictive of treatment failure (early non-response) in psychotherapy for late-life depression.

However, not all studies found interpersonal relationships to be predictive of treatment failure. Although low end state functioning groups had higher levels of interpersonal hostility, Critchfield et al. (2007) found no predictors in interpersonal process behaviors in a small sample of patients with generalized anxiety disorder (GAD) receiving CBT. Further, dyadic adjustment, which focuses on consensus, satisfaction, cohesion, and expression of affection within a romantic relationship, did not predict treatment failure (dropout) in CBT for panic disorder (Bélanger et al., 2017). Perceived criticism was no predictor of treatment failure in treatment for patients with OCD or panic disorder (Chambless & Steketee, 1999). Hoyer et al. (2016) did not find interpersonal problems and attachment style as predictors of treatment failure in cognitive therapy for patients with social anxiety disorder. Finally, Strauss et al. (2017) found that partner-related attachment anxiety and avoidance in romantic relationships are no predictors of treatment failure (drop-out) in both CBT and psychodynamic therapy for patients with social anxiety disorder.

Several predictors that were studied in the included papers could not be classified under one of the categories of predictors discussed so far. Notably, none of those predictors were studied in multiple studies, whereas a number of the target groups in these studies were very specific, for example late life depression, or social anxiety disorder with a primary fear of public speaking.

First of all, patients who experience a high negative impact of stressful events are more likely to experience treatment failure (non-response) to CBT for late-life depression (Marquett et al., 2013). The same was true for an external locus of control. Interestingly, also patients who do not tend to assign blame for stressful events to others (decreased external blame) turned out to be more likely to experience treatment failure (non-response) (Marquett et al., 2013). Further, Arndt et al. (2020) studied different drop-out patterns in an internet-based depression intervention and found that low physical health increases the risk to belong to the drop-out group, whereas Johnson et al. (2014) found that stereotype confirmation concerns, defined as being afraid of confirming certain stereotypes leading to negative evaluations, are associated with higher risk of treatment failure (drop-out) in social anxiety disorder treatment. Moggia et al. (2020) measured self-ideal discrepancy, which is the difference between the rating of the ideal self and the self as it is now, self-others discrepancy, explained as how someone views him/herself in respect to others, and a dilemmatic construction of the self, which means that a person desires a change in one construct of the self that correlates with an undesirable change in another construct of the self. All of these measures appeared not to be predictive of treatment failure (non-improvement) in psychotherapy for depression. Finally, Schilling et al. (2021) found that emotion regulation was no predictor of treatment failure (deterioration of complaints) after CBT for different diagnosis.

In this paper, we reviewed studies on predictors of treatment failure (also defined as non-response, deterioration, and drop-out) for depressive and anxiety disorders. We specifically focused on amenable patient-specific predictors, such as motivation or coping styles. Knowledge of such factors might help us reduce treatment failure and thus improve treatment outcomes. Like Steketee and Chambless (1992) and Eilertsen and Eilertsen (2023), we encountered several problems within the existing literature. One important problem is the fact that both treatment failure and predictors are measured differently, which makes studies difficult to compare. Another problem is a lack of replication studies, and if predictors were replicated, this was done in studies using small or specific patient samples (e.g., elderly, social phobia with fear of speaking in public, or OCS with overt compulsions). Consequently, it is difficult to conclude the overall impact of these predictors.

However, a few predictors that stand out are treatment expectancy (replicated in three studies with a big sample size and different patient samples), neuroticism (replicated in two studies with the same instrument in different patient samples), the use of social support (replicated in two studies with different measuring instruments), readiness to change (replicated in three studies, in very small patient samples), outcome expectancy (two studies with different measuring instruments), and perceived social support (three studies, but only in late-life depression found as a predictor).

Focusing more on the different definitions of treatment failure, we see that about half of the included articles focus on drop-out and the other half on non-response. There are two articles that mention deterioration. Because drop-out, non-response, and deterioration are different constructs, we also looked at the difference in predictors per treatment failure definition. We see that almost none of the predictors are replicated within these treatment failure categories. The only predictors that remain are readiness to change and treatment expectancy. Readiness to change was studied in three different studies, two of which focused on drop-out, and both studies found that it was no predictor of drop-out. Treatment expectancy was studied in six different studies, all of which found that it was not a predictor of drop-out (three studies) or non-response (three studies). However, drop-out and non-response in all these studies were measured differently, which still makes it hard to draw firm conclusions about the predictive validity of these measures.

Targeting the identified predictors before or early in the treatment of anxiety and depressive disorders might help reduce treatment failure, which in turn might increase the cost-effectiveness of treatments. For example, in accordance with Constantino et al. (2018), assessing the outcome expectancy of patients at the start of treatment gives insight into the probability of a poor treatment outcome or even treatment failure. It offers an opportunity to assess whether interventions targeting the expectations of treatment should be integrated in the treatment in order to reduce the risk of treatment failure. Also, by assessing the use of social support early in treatment, interventions can be focused more on involving the social support system in therapy or enhancing coping strategies for asking for help, which might prevent treatment failure.

Unfortunately, as there are very few to no amenable factors that show strong evidence of being a predictor of treatment failure, clinicians need to be careful to use variables of which they assume they influence a poor therapy outcome. With the current knowledge, it is not possible to withhold patients from specific treatments or adapt treatments based on amenable predictors of treatment failure.

One important limitation is that for each predictor, only a limited number of studies have been carried out, often for only one specific treatment and/or disorder. For instance, CBT was the treatment of choice in five of the six studies into treatment expectancy as a predictor, whereas only two of them studied a sample of patients with the same disorder. As we cannot rule out the possibility that predictors of treatment failure differ for different disorders as well as for different treatments, predictors preferably should be studied in multiple treatment and patient groups to reach more definitive conclusions on overall predictors of treatment failure (Dalgleish et al., 2020). A further limitation is the absence of standardized measures for response rates (Loerinc et al., 2015), which makes it difficult to draw definitive conclusions about treatment outcome and -thus- about treatment failure. The reliable change index (RCI), as defined by Jacobson and Truax (1991), could address this problem (Loerinc et al., 2015), as by using this index treatment failure can be defined as meeting the criteria for no clinically significant change or deterioration at the end of treatment. The fact that the vast majority of studies reporting on predictors of treatment outcome only studied (or reported) predictors of treatment success or reported on poor(er) outcome is a third limitation. As we could not assume that poor(er) outcome is identical to treatment failure, or that results of treatment success can be inverted to treatment failure, most studies had to be excluded. Future studies would benefit from also studying and reporting on predictors of treatment failure instead of treatment success only. Another limitation is that the terms ‘association’ and ‘prediction’ were used interchangeably in both the results and discussion sections of papers excluded from this review. Hence, it was not always clear if the studied variable was in fact a predictor, or if the predictors were just correlated with the outcome. We only included the studies that were very clear that they measured predictors of treatment failure. The field would benefit from a more consistent and precise use of these terms in presenting results on predictors of treatment outcome. A final limitation is that this review is not pre-registered and is not completely written according to current scoping review guidelines (Tricco et al., 2018). This stems from the fact that the first search of this review was conducted in 2017, before the publication of these guidelines. We recommend a future scoping or systematic review on the same topic to replicate and renew these findings.

By knowing which factors are predictors of treatment failure, we can move towards studying how to intervene on these factors in order to decrease the odds of treatment failure. Based on the results of this review, there are no predictors that stand out as overall strong amenable predictors of treatment failure. Future studies are needed to determine whether already found predictors are replicable with the same instruments in different patient samples and with different treatments.