SpiroLLM Model Fields for COPD Diagnostic Report Generation

234,028 individuals from the UK Biobank underpin the SpiroLLM model for generating diagnostic reports from pulmonary function tests. The model extracts morphological features from respiratory curves and aligns them with PFT numerical values. It achieved a diagnostic AUROC of 0.8977 and maintained a 100% valid response rate in robustness tests.

Use Cases

• Fine-tune the SpiroLLM model using extracted morphological features from respiratory curves for COPD diagnosis.
• Align PFT numerical values with respiratory curve features in a unified latent space for multimodal analysis.
• Generate a comprehensive diagnostic report by leveraging the fused physiological signals and large language model capabilities.

Strengths

• Derived from a substantial cohort of 234,028 individuals from the UK Biobank.
• The underlying model achieved a high diagnostic AUROC of 0.8977 (95% CI: 0.88-0.91).
• Demonstrated robustness with a 100% valid response rate in tests with missing core data.

Limitations

• The dataset is very small at 9.5 KB, indicating it contains model parameters or extracted features, not raw patient data.
• Specific column definitions and data structure for the required fields are not provided.
• The temporal and geographic scope of the underlying UK Biobank cohort is not detailed in this context.

Provenance

Source

UK Biobank cohort, processed by author Shuhao Mei.

Collection Method

Model development leveraging respiratory spirogram time series and PFT numerical values.

Freshness

Last updated on 2026-03-24.

Geography

United Kingdom (inferred from UK Biobank).