Italian real world artificial intelligence (AI) based analysis of early-onset colorectal cancer: The ERCOLE study.

Abstract

e15539 Background: Early-onset colorectal cancer (EOCRC), defined as CRC diagnosed among individuals younger than 50 yrs, is becoming a public health issue, with rising incidence reported worldwide. In the US, EOCRC is the leading cause of cancer-related death among males aged 20-49 yrs and is second among female aged 40-49 yrs. Italy lacks a national cancer registry and available data regard mostly northern regions. Fondazione Policlinico Gemelli (FPG) Hospital is a national referral center for CRC (especially for middle and southern regions), ranking first for surgical procedures. Therefore, mining incidence rate at FPG Hospital contributes to provide a full picture of EOCRC incidence trend in Italy. Methods: The objective of this study was to assess the trends of EOCRC incidence at FPG along 15 yrs. Real world data collected from everyday clinical practice in the FPG Hospital Data Warehouse were extracted within the Gemelli GENERATOR framework. Study population was identified from records of pts from 2005 to 2019 matching two main inclusion criteria: pts hospitalized with a diagnosis of CRC (ICD-9 codes at discharge included in 153.* and 154. *, captured form structured data source); or pts with at least one pathology report of CRC (selected using clinically-validated text mining techniques from unstructured data source). Variables of interest, including demographics and surgical procedures, were extracted using SAS (SAS(R) Institute suite for ETL). Study population was stratified according to age (≤50yrs; > 50yrs), gender and tumor site. Primary endpoint was the annual percent change (APC). APC was estimated using the joint regression method with a maximum of 2 joint points allowed. Statistical analyses were conducted using R (v4.2.1). Results: 18606 consecutive single pts were included, of those 10.9% were EOCRC. From 2005 to 2019 a statistically significant APC increase for EOCRC was observed overall (9.2, CI 6.5-11.9, p < .001) and regardless of gender (male: 8.0, CI 4.7-11.3, p < .001; female: 10.1, CI 7.5-12.8, p < .001). Splitting the timeframe, a major trend was observed between 2010 and 2013 (APC: 30.5, 34.4 and 27.5, respectively overall, in male and in female pts). Primary tumor location was available for 10690 (of those 9.3% were EOCRC). Analyzing primary tumor location, a trend toward increased and decreased APC for EOCRC was observed for R-sided (3.6, p .105) and L-sided (-2.0, p .312) tumors, respectively, while a statistically significant increase was observed for rectum (4.5, p .02), specifically between 2012 and 2019 (13.1, p .02). Conclusions: This AI-based real world analysis confirms a rising incidence for EOCRC in Italy. Data has solid external consistency. Further analyses on familial history, life styles and MMR status are planned. The appropriate threshold age to start the screening needs to be addressed.