PREDICTING PROSTATE CANCER INCIDENCE USING A LARGE, NATIONAL U.S. ELECTRONIC HEALTH RECORDS (EHR) DATABASE
Author(s)
Stephanie L. Wall, MPH, Reghan Lanning, MS, Maryam Ajose, MPH, anusorn thanataveerat, DrPH;
Veradigm, Raleigh, NC, USA
Veradigm, Raleigh, NC, USA
OBJECTIVES: This study aimed to evaluate predictive models for prostate cancer incidence within 24 months following prostate-specific antigen (PSA) assessments using U.S. real-world data, and to compare predictive model performance by incorporating PSA-specific information, including the index PSA value and PSA slope.
METHODS: The Veradigm Network EHR database was used to identify adults (≥18 years) with ≥1 PSA laboratory result between 01/01/2017-09/30/2023 (earliest test=index). Inclusion criteria included no evidence of prostate cancer anytime pre-index, EHR activity for ≥12 months prior to (baseline) and up to 24 months following the index date or until the first occurrence of a prostate cancer diagnosis within that timeframe (follow-up). All patients were required to have ≥1 additional PSA test post-index during follow-up.To address outcome imbalance, patients with a prostate cancer diagnosis were matched 1:4 to randomly sampled patients without prostate cancer. Multivariable logistic regression models were used to estimate the likelihood of developing prostate cancer during follow-up. Model 1 included core demographics and clinical characteristics; Model 2 added index PSA; Model 3 further incorporated PSA slope (categorized as “0.5-3.5 ng/mL/year” or “other”) to create nested models. Models’ performance was calculated using the area under the receiver operating characteristic curve (AUC).
RESULTS: Out of 1,807,823 eligible patients, 37,452 (2.1%) had evidence of prostate cancer during the follow-up period. The median time to prostate cancer diagnosis was 363 days, or nearly 1 year. Most patients were White (61.9%), Non-Hispanic/Unknown (94.5%), and resided in the Southern geographic region (42.3%). The AUC for Models 1-3 were 0.6751, 0.8650, and 0.8684, respectively.
CONCLUSIONS: The incorporation of PSA information into the models significantly enhanced its ability to predict the likelihood of incident prostate cancer within the 24-month follow-up period. Notably, PSA slope exhibited a strong association with the occurrence of incident prostate cancer.
METHODS: The Veradigm Network EHR database was used to identify adults (≥18 years) with ≥1 PSA laboratory result between 01/01/2017-09/30/2023 (earliest test=index). Inclusion criteria included no evidence of prostate cancer anytime pre-index, EHR activity for ≥12 months prior to (baseline) and up to 24 months following the index date or until the first occurrence of a prostate cancer diagnosis within that timeframe (follow-up). All patients were required to have ≥1 additional PSA test post-index during follow-up.To address outcome imbalance, patients with a prostate cancer diagnosis were matched 1:4 to randomly sampled patients without prostate cancer. Multivariable logistic regression models were used to estimate the likelihood of developing prostate cancer during follow-up. Model 1 included core demographics and clinical characteristics; Model 2 added index PSA; Model 3 further incorporated PSA slope (categorized as “0.5-3.5 ng/mL/year” or “other”) to create nested models. Models’ performance was calculated using the area under the receiver operating characteristic curve (AUC).
RESULTS: Out of 1,807,823 eligible patients, 37,452 (2.1%) had evidence of prostate cancer during the follow-up period. The median time to prostate cancer diagnosis was 363 days, or nearly 1 year. Most patients were White (61.9%), Non-Hispanic/Unknown (94.5%), and resided in the Southern geographic region (42.3%). The AUC for Models 1-3 were 0.6751, 0.8650, and 0.8684, respectively.
CONCLUSIONS: The incorporation of PSA information into the models significantly enhanced its ability to predict the likelihood of incident prostate cancer within the 24-month follow-up period. Notably, PSA slope exhibited a strong association with the occurrence of incident prostate cancer.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
RWD120
Topic
Real World Data & Information Systems
Disease
SDC: Oncology