Cross-Scale Transfer Learning for Depression Severity Prediction: From PHQ-8 to HAMD-17 Across Languages and Clinical Paradigms
Sequential LoRA transfer from English PHQ-8 interviews to Chinese HAMD-17 consultations improves depression-score prediction with Qwen3.
The paper proposes sequential LoRA transfer for depression-severity prediction when labeled clinical interviews are scarce. A Qwen3 model is first fine-tuned on 189 English DAIC-WOZ sessions scored with PHQ-8, then the adapter initializes training on 100 Chinese PDCH consultations scored with HAMD-17. On the target task, Qwen3-0.6B reached MAE 4.96, RMSE 6.59, and macro-F1 0.36, while Qwen3-1.7B reached 4.38, 5.62, and 0.46, outperforming target-only and non-LLM baselines. The authors describe a single-site exploratory evaluation and do not claim screening or diagnostic utility.
- Sequential LoRA fine-tunes Qwen3 on DAIC-WOZ PHQ-8, then PDCH HAMD-17.
- Qwen3-1.7B reached MAE 4.38, RMSE 5.62, and macro-F1 0.46.
- Native Chinese input outperformed machine-translated English on the target task.
- Authors report a single-site study without claimed diagnostic utility.
Full article195 words · extracted from arxiv.org · click to collapse
This work addresses continuous depression-severity score prediction from clinical interview transcripts under data scarcity. We propose a sequential low-rank adaptation (LoRA) protocol for cross-scale transfer: a Qwen3 backbone with a bounded regression head is first fine-tuned on the English DAIC-WOZ dataset (189 avatar-mediated sessions, PHQ-8), and the adapter then initializes fine-tuning on the Chinese PDCH dataset (100 real clinical consultations, HAMD-17), where a reinitialised, scale-specific head predicts the clinician-assigned score. All configurations use patient-level stratified 5-fold, 2-repeat cross-validation. On the data-scarce HAMD-17 target, the sequential protocol attains the best point-estimate MAE , RMSE, and macro-$F_1$ on both 0.6B and 1.7B backbones, outperforming target-only training and non-LLM baselines---4.96/6.59/0.36 with Qwen3-0.6B and 4.38/5.62/0.46 with Qwen3-1.7B. Ablations suggest that correctly aligned source supervision gives the best point estimates (unsupervised exposure and shuffled-label controls also show partial gains), that native-Chinese target input outperforms machine-translated English input, and that the reversed order yields no clear gain within run-to-run variance. The study is an exploratory, single-site internal evaluation: it does not establish screening or diagnostic utility, nor separately identify the contribution of the scale, language, or paradigm shifts. To our knowledge, no prior study evaluates this specific DAIC-WOZ-to-PDCH sequential transfer setting.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.28430