Major depressive disorder represents a major public health problem in the Middle East and North Africa region (MENA). Antidepressant treatment resistance remains common and costly, affecting approximately 20–50% of patients depending on the definitions used. This study aims to identify socio-demographic and clinical factors associated with antidepressant resistance in Tunisian patients with major depressive disorder, and to develop predictive models of antidepressant treatment response. Prospective longitudinal study conducted among 67 adult patients (mean age: 42.2 ± 10.8 years) consulting at a psychiatry department in Tunis, between September 2020 and May 2021. Patients met DSM-5 criteria for a major depressive episode. Symptom severity was assessed using the Hamilton Depression Rating Scale (HAM-D) and psychiatric comorbidities using the Mini International Neuropsychiatric Interview (MINI). Resistance was defined as absence of response (< 50% reduction in HAM-D score) to two successive antidepressants administered at adequate dose and duration (6 weeks each). Three predictive models were developed and compared: logistic regression, CART decision tree, and random forest, validated by 5-fold stratified cross-validation. Of the 67 patients who completed the study, 48 (71.6%) responded to treatment and 19 (28.4%) were classified as resistant. The response rate to the first antidepressant was 42% (95% CI: 30.2–54.5%) and to the second was 50% (95% CI: 33.4–66.6%). Factors significantly associated with resistance included: initial episode severity (mean HAM-D score of 23 ± 4 in resistant vs. 18 ± 3 in responders, p < 0.001), duration of symptom evolution (11 ± 7 months vs. 7.4 ± 6 months, p = 0.026), presence of at least one psychiatric comorbidity (68.4% vs. 31.2%, p = 0.005), particularly posttraumatic stress disorder (PTSD) (26.3% vs. 4.2%, p = 0.007), and presence of disc herniation (21% vs. 6.2%, p = 0.007). Multivariate logistic regression analysis confirmed that psychiatric comorbidities (p = 0.005), PTSD (p = 0.008), disc herniation (p = 0.008), episode duration (p = 0.050), and initial severity (p < 0.001) were independent predictors of resistance. Overall remission rate was 32.8% (22/67), compared to response rate of 71.6%, indicating substantial residual symptomatology in partial responders. For predictive modeling, logistic regression showed the best performance with an AUC of 0.931 (95% CI: 0.892–0.970) and a Brier score of 0.095, outperforming random forest (AUC = 0.884) and CART tree (AUC = 0.848). However, this exceptional performance requires cautious interpretation given small sample size and potential overfitting; external validation is essential before clinical implementation. This study identifies clinical factors predictive of antidepressant treatment resistance in a Tunisian population. High initial severity, presence of psychiatric comorbidities (particularly PTSD), and prolonged duration of symptom evolution before treatment represent poor prognostic markers. The predictive models developed, while requiring external validation, offer proof of concept for developing clinical decision support tools in resource-limited settings. These results underscore the importance of early management, systematic assessment of comorbidities, and adaptation of therapeutic strategies in patients at high risk of resistance.
Othman et al. (Mon,) studied this question.