Building the adult growth hormone deficiency data mart: a Real-World model of AI-driven clinical data extraction in a single Italian center.
An AI-driven approach successfully created a Data Mart for AGHD, potentially enhancing real-world data analysis but requires broader validation.
Where it sits
this study against the rest of the hgh (somatropin) corpusSummary and findings
The study developed an AI-driven methodology to create a Data Mart for Adult Growth Hormone Deficiency (AGHD) using clinical data from a single Italian center. Among 210 identified patients, 188 were validated as having AGHD. The Data Mart includes diagnostic modalities, etiology, biochemical data, comorbidities, and growth hormone replacement therapy information.
Abstract
<h4>Purpose</h4>Adult Growth Hormone Deficiency (AGHD) is a complex and under-recognized condition, mainly managed in outpatient settings and characterized by fragmented and heterogeneous clinical data. Population registries provide high-quality evidence but require long timeframes and substantial resources. The aim of this study was to design and validate an artificial intelligence (AI)-driven methodology for the construction of a disease-specific AGHD Data Mart from routine clinical data within a single high-volume center, to support real-world evidence generation and future advanced analytics.<h4>Methods</h4>A standardized Data Science framework, based on automated extraction from hospital data warehouses and electronic medical records, integrating structured data and unstructured clinical narratives through natural language processing, was implemented. AGHD patients were identified using a combination of ICD-9 codes and text-mining applied to outpatient reports. A multidisciplinary workflow ensured clinical validation of extracted data. The Data Mart described patient identification from first hospital access (T0) and included diagnostic modality, etiology, biochemical data, comorbidities, and growth hormone replacement therapy.<h4>Results</h4>Among 210 identified patients, 188 were validated as AGHD after expert review. Diagnoses were based on dynamic testing (28.2%), panhypopituitarism with low IGF-1 (54.3%), or AI-assisted identification (17.6%). Etiology was retrieved in 87.8% of cases, with post-surgical causes being the most frequent. 37.8% of patients were receiving rhGH therapy. Specific trends for IGF-1 values for single patients were described.<h4>Conclusion</h4>This study represents the first AI-driven AGHD Data Mart and demonstrates the feasibility of constructing the Data Mart from routine clinical data. This approach offers a complementary framework for structured RWD extraction and may support future longitudinal analyses and AI-based clinical support in AGHD.
Background
Adult Growth Hormone Deficiency (AGHD) is a condition that is often under-recognized and managed in outpatient settings, leading to fragmented clinical data. Traditional population registries provide valuable evidence but are resource-intensive and slow to develop. This study addresses the need for a more efficient method of data collection by utilizing AI-driven techniques to establish a comprehensive AGHD Data Mart, which could facilitate real-world evidence generation and advanced analytics.
Methods
The study employed a standardized Data Science framework to extract data from hospital data warehouses and electronic medical records. This included both structured data and unstructured clinical narratives processed through natural language processing. AGHD patients were identified using ICD-9 codes and text-mining of outpatient reports. A multidisciplinary team validated the clinical data, which encompassed diagnostic modalities, etiology, biochemical data, comorbidities, and growth hormone replacement therapy.
Results
Out of 210 patients identified, 188 were confirmed to have AGHD after expert review. Diagnoses were made based on dynamic testing (28.2%), panhypopituitarism with low IGF-1 (54.3%), and AI-assisted identification (17.6%). The etiology was determined in 87.8% of cases, with post-surgical causes being the most common. Additionally, 37.8% of patients were undergoing recombinant human growth hormone (rhGH) therapy.
Interpretation
This study demonstrates the feasibility of using AI-driven methods to construct a Data Mart from routine clinical data, offering a new framework for structured real-world data extraction. While the study provides a promising model, the single-center design limits its generalizability. The findings suggest potential for future longitudinal analyses and AI-based clinical support in AGHD, but further validation in diverse clinical settings is necessary.
Key findings
- 210 patients identified, 188 validated as AGHD.
- 28.2% diagnosed via dynamic testing.
- 54.3% diagnosed with panhypopituitarism and low IGF-1.
- 17.6% identified through AI-assisted methods.
- 37.8% of patients receiving rhGH therapy.
Limitations
- Single-center study
- AI-driven methods need further validation
- Potential biases in data extraction
- Limited generalizability
- Retrospective design