Real-world evidence in localized and locally advanced prostate cancer: applying artificial intelligence to electronic health records.

Autores de IIS La Fe

• Antonio Conde Moreno

  Autor

Participantes ajenos a IIS La Fe

• Maroto JP
• Puente J
• Juárez Á
• Garcillán B
• Calderón JM
• Del Toro JM
• Valdivieso JL
• López-Menduiña M
• Sarró E
• Alcaraz A

Grupos

• Investigación Clínica y Traslacional en Cáncer

Abstract

PURPOSE: To provide real-world evidence of the clinical characteristics and outcomes of localized and locally advanced prostate cancer patients (LPC/LAPC). MATERIALS & METHODS: Observational and retrospective analysis using secondary data from electronic health records (EHR) of prostate cancer (PC) patients in eight Spanish hospitals (2014–2018). Data was extracted and analyzed using EHRead® technology, based on natural language processing and machine learning. LPC/LAPC patients were included and stratified by risk and by first treatment received. RESULTS: Twenty-two thousand one hundred sixty-six PC patients were identified,14,434 (65.1%) were classified as LPC/LAPC. Among them, 5,331 incident patients with sufficient data were selected for outcome analysis (real world overall survival [rwOS], metastasis and event free survival [MFS, EFS]) and were followed for a median time of 2.3 years. 36.5% were classified as LPC intermediate risk (IR), 26.0% LPC high risk (HR), 7.3% LPC low risk (LR), 5.9% LAPC, and 24.2% unknown risk. First treatment received was radiotherapy (RT) in 40.7%, radical prostatectomy (RP) in 37.1%, active surveillance (AS)/watchful waiting (WW) in 6.4%, brachytherapy (BT) in 4.2%, and androgen deprivation therapy monotherapy (ADT only) in 3.3%. rwOS and MFS worsened as risk increased. Patients treated with ADT only presented the worst baseline characteristics, showing limited clinical outcomes. The 36-month rwOS was 91% for LAPC patients, 93% for HR-LPC, 97% for IR-LPC, and 98% for LR-LPC. CONCLUSIONS: Despite using treatment with curative intent, patients experienced oncological events within a median of less than three years post-diagnosis. Our findings emphasize the need for risk stratification, and proactive strategies to improve clinical outcomes. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12885-025-14828-z.

Keywords

• Electronic health records; Localized and locally advanced prostate cancer; Machine learning; Natural language processing; Prostate cancer