Cross-Scale Transfer Learning for Depression Severity Prediction: From PHQ-8 to HAMD-17 Across Languages and Clinical Paradigms

TL;DR

This paper proposes a sequential low-rank adaptation protocol for cross-scale transfer learning to predict depression severity scores from clinical interview transcripts. The protocol achieves state-of-the-art results on the HAMD-17 target dataset. It outperforms target-only training and non-LLM baselines.

Why It Matters

The study's findings have practical significance in addressing data scarcity in clinical datasets, enabling more accurate depression severity predictions. This can lead to better patient outcomes and more effective treatment plans. The approach can be applied to other clinical datasets and languages.

Key Findings

• Sequential low-rank adaptation improves depression severity prediction
• Native-language target input outperforms machine-translated input
• Correctly aligned source supervision yields the best results