Inpatient morbidity and structural care burden in oropharyngeal carcinoma in Germany: a nationwide cross-institutional EHR analysis using an AI-enabled data network.

PURPOSE: This study aims to characterize hospitalization-level morbidity and care burden associated with oropharyngeal carcinoma (OPC) in Germany and assess the contribution of terminology-based extraction from narrative electronic health record (EHR) documentation beyond structured administrative coding. METHODS: This nationwide retrospective observational study analyzed inpatient data from 96 German tertiary care hospitals (2016-2022). OPC hospitalizations were identified using ICD-10-GM principal diagnosis codes. Treatment patterns, morbidity indicators, and resource utilization measures were derived from OPS procedural codes, structured clinical variables, and terminology-based extraction from predefined narrative EHR sections. Analyses were conducted at the hospitalization level. RESULTS: Among 13,925 OPC hospitalizations, major pharyngeal resections were documented in 9.0% of cases. The proportion reconstructed with microvascular free flaps more than doubled, while resections without reconstruction declined. Robotic-assisted surgery was documented in 0.1% of OPC hospitalizations. Mechanical ventilation was documented in 3.7% of cases, and 19.7% had documentation indicative of tracheostomy. Median hemoglobin decreased during hospitalization, and documented RBC transfusion use increased descriptively from 5.4% in 2016 to 7.6% in 2022. The in-hospital mortality rate was 2.8%. Terminology-based EHR extraction identified additional documentation of selected comorbidities and morbidity markers beyond structured coding alone. CONCLUSION: Hospitalization-level OPC care in German tertiary hospitals involves substantial inpatient care burden, including complex reconstruction, airway-related morbidity, and physiologic morbidity markers. Combining structured administrative data with terminology-based extraction from narrative EHR documentation provided additional routine-data markers beyond coding alone. These findings should be interpreted as complementary hospitalization-level information rather than as validated clinical endpoints or evidence of causal treatment effects.