The predictive power of subgroups: An empirical approach to identify depressive symptom patterns that predict response to treatment

Abstract

BackgroundDepression research has been trying to improve the response rates to treatments by identifying a valid set of differential predictor variables. Potential candidates have been proposed, one of which were different subtypes of depression. However, the results on the predictive quality of subtypes on treatment are conflicting. MethodsThe analyzed data consisted of Hamilton Depression Rating Scales (HAM-D17) of 879 depressive inpatients, which were recruited in a naturalistic multicenter study. Mean length of stay was 9.9 weeks. In a first step, a Latent Class Analysis (LCA) was conducted to classify the patients into smaller groups. In a second step, the class variable was included in a Linear Mixed Effects model to predict the same patients’ response to treatment. ResultsFive classes were obtained from LCA, showing substantially different symptom profiles. One of the classes, with a symptom profile similar to melancholic depression, showed substantially slower response to treatment (i.e. estimated time to remission; 11.3 weeks) than the remaining classes in the study (6.6–8.6 weeks). LimitationsThe applied measurement instrument, the HAM-D17, did not include items for two additional, frequently found subtypes of depression: psychotic and atypical depression. Thus, these subtypes could not emerge in the LCA. Furthermore, there was no systematic variation of treatment in the data. Thus, a differential effect of the classes on treatment could not be measured. ConclusionsThe classification of patients according to their symptom profiles seems to be a potent predictor for treatment response. However, the obtained symptom patterns are not completely congruent with the theoretically proposed subgroups. Against the background of the results, dividing melancholic depression in a rather cognitive and vegetative subtype may be promising.