Empirically derived decision trees for the treatment of late-life depression

Abstract

Objective: Several predictors of treatment response in late-life depression have been reported in the literature. The aim of this analysis was to develop a clinically useful algorithm that would allow clinicians to predict which patients will likely respond to treatment and thereby guide clinical decision making.

Method: A total of 461 patients with late-life depression were treated under structured conditions for up to 12 weeks and assessed weekly with the 17-item Hamilton Rating Scale for Depression (HAM-D-17). The authors developed a hierarchy of predictors of treatment response using signal-detection theory. The authors developed two models, one minimizing false predictions of future response and one minimizing false predictions of future nonresponse, to offer clinicians two clinically useful treatment algorithms.

Results: In the first model, early symptom improvement (defined by the relative change in HAM-D-17 total score from baseline to week 4), lower baseline anxiety, and an older age of onset predict response at 12 weeks. In the second model, early symptom improvement represents the principal guide in tailoring treatment, followed by baseline anxiety level, baseline sleep disturbance, and--for a minority of patients--the adequacy of previous antidepressant treatment.

Conclusions: Our two models, developed to help clinicians in different clinical circumstances, illustrate the possibility of tailoring the treatment of late-life depression based on clinical characteristics and confirm the importance of early observed changes in clinical status.

FIGURE 1. Hierarchy of Predictors of Treatment Response With an Aggressive Treatment Approach (i.e., a cutoff point for sensitivity of 0.30)^a

^a Main outcome=full response status at week 12. Proportion of responders at 12 weeks=60%. A change in Hamilton Rating Scale for Depression score at week 4 of less than 45% from baseline predicts a less than half (43%) chance of responding at week 12. For the patients in this group, higher baseline anxiety predicts a 39% chance of responding at week 12. For patients with higher baseline anxiety, a younger age of onset predicts a 32% chance of responding at week 12.

FIGURE 2. Hierarchy of Predictors of Treatment Response With a Conservative Treatment Approach (i.e., a cutoff point for sensitivity of 0.70)^a

^a Main outcome=full response status at week 12. Proportion of responders at 12 weeks=60%. A change in Hamilton Rating Scale for Depression score at week 4 is the main predictor of treatment response in this model. A change of less than 39% from baseline predicts a 39% chance of responding at week 12, followed by a change of less than 18% from baseline, predicting a 25% chance of responding at week 12. For the subjects with an 18% or higher change in Hamilton Rating Scale for Depression score at week 4, higher baseline sleep disturbance predicts a 19% chance of responding at week 12.

FIGURE 3. Hierarchy of Predictors of Treatment Response Using a Cutoff Point for Sensitivity of 0.7 (conservative treatment approach) and Including the Antidepressant Treatment History Form Score as One of the Predictors of Treatment Response^a

^a Main outcome=full response status at week 12. Proportion of responders at 12 weeks=64%. For the antidepressant treatment history form, a score of ≥3=probably adequate antidepressant treatment history (trial of more than 4 weeks of an antidepressant at an adequate dose). For the antidepressant treatment history form a score of <3=inadequate antidepressant history (trial of less than 4 weeks or of more than 4 weeks but with an inadequate dose). High anxiety=at least moderate anxiety symptoms. Low anxiety=mild or no anxiety symptoms. Change in Hamilton Rating Scale for Depression (HAM-D) at week 4 of less than 30% from baseline predicts a 36% chance of responding at week 12. For the subjects with a lower than 30% change in HAM-D score at week 4, a history of at least one adequate antidepressant trial predicts a 14% chance of responding at week 12. For the subjects with a 30% or higher change in HAM-D at week 4, the next predictor of treatment response is baseline anxiety. A higher baseline anxiety score predicts a lower chance of responding at 12 weeks (41%), whereas a lower baseline anxiety score predicts a 79% chance of responding at week 12.