A text mining on clinical transcripts of good and poor outcome psychotherapies

The text mining of clinical transcripts is broadly used in psychotherapy research, but is limited to top-down approaches, with a-priori vocabularies that code the transcripts according to a theoretical predetermined framework. Nevertheless, the semantic level that a word or clinical intervention can assume depends on the relational field in which the discourse is produced. Thus, bottom-up approaches seem to be particularly meaningful in addressing such a relevant issue. With the aim of investigating possible similarities and differences between good outcome and poor outcome psychotherapies, we applied a multivariate analysis on the transcripts of eight single cases of brief experiential psychotherapy (four good outcome vs four poor outcome cases), in order to identify the general core themes, and their difference according to therapy outcome. The results showed a significant difference in the number of context units classified in two of the six core themes (clusters) between good and poor outcome cases (χ2, df=5, p<0,01). These findings show how the bottom-up technique of text analysis on clinical transcripts turned out to be an enlightening tool to let their latent dimensions emerge, setting the clinical process and outcome, and therefore, providing a very useful tool for clinical purposes.