Understanding and Predicting Symptom Trajectories in Psychological Care

Part one: RESPONSIVE REGULATION OF PSYCHOTHERAPY DURATION: A SYSTEMATIC REVIEW AND META-ANALYSIS OF THE “GOOD-ENOUGH LEVEL” LITERATURE Objectives This review aimed to examine the “Good-Enough Level” (GEL) concept that people respond differently to therapy, as well as examining the shape of change (Barkham et al., 1996). Methods Systematic searches took place using Medline, PsycINFO and Scopus databases. Key search terms were variants of: Good-enough level, dose-response, treatment duration, rate of change, treatment outcome, responsive regulation and psychotherapy. A key inclusion criterion was that cases must be stratified by treatment length to examine the GEL. A narrative synthesis was provided, with random effects meta-analysis where possible. Results Fifteen studies were synthesised (n=114,123), with five used in primary meta-analyses (n=46,921), and sub-group analyses performed on differential findings. High heterogeneity was observed making conclusions tentative, however there was no overall association between improvement and total sessions, which supports the GEL (r=-0.24 [95% CI -0.70, 0.36], p=0.2747). Increases in improvement associated with longer treatment may be an artefact of people terminating therapy early. Longer treatments were associated with higher baselines (r=0.15 [95% CI 0.08, 0.22], p<.001) and slower rates of change, in support of the GEL. Shapes of change varied, but emerging patterns suggested that longer treatments may see more linear trajectories, challenging the universal curvilinear trend suggested in the dose-response literature. Conclusions Support was found for the GEL: treatment length appears to be responsively regulated based on need, and there is heterogeneity in trajectories of change. However this may also occur within boundaries suggested by the dose-response literature. The models could co-exist within a concept of “boundaried responsive regulation”. Part two: THE DEVELOPMENT OF A DYNAMIC PROGRESS FEEDBACK SYSTEM TO GUIDE PSYCHOLOGICAL TREATMENT IN PRIMARY CARE Objectives This study aimed to develop dynamic progress feedback models, combining initial profile information from the Leeds Risk Index ([LRI] Delgadillo, Moreea, & Lutz, 2016) with weekly progress scores to provide personalised prognoses of recovery. Sub-aims were to assess generalisability, and to examine whether complex models outperformed parsimonious ones. Design A retrospective database analysis was used to construct the predictive models in one Improving Access to Psychological Therapies (IAPT) dataset, followed by cross-validation in a new IAPT dataset. Methods Models of increasing complexity were constructed, using backward elimination to retain significant predictors. Five predictors were used: baseline score, current session score, cumulative risk sums, cumulative individual standard deviations, and LRI profile group. Models were compared on how much variance they explained, and AUC, Kappa and Brier scores used for cross-validation. Results The models showed good predictive ability and cross-validated well in a new sample (e.g. explaining 39% of variance with an AUC of .775 by session four at low intensity). At high intensity treatment, the most complex model was superior at sessions one-to-six. At cross-validation, the more complex models saw higher scores, however differences between models were not statistically significant. Conclusions It was possible to build dynamic prediction systems that cross-validated in a new sample. Although complex models performed slightly better, this was not statistically significant at cross-validation. The question of whether to incorporate more complexity would therefore be a service-led decision. Further cross-validation and development into a clinical tool is intended.