Premium
Identification of a five gene signature to predict time to biochemical recurrence after radical prostatectomy
Author(s) -
Corradi John P.,
Cumarasamy Christine White,
Staff Ilene,
Tortora Joseph,
Salner Andrew,
McLaughlin Tara,
Wagner Joseph
Publication year - 2021
Publication title -
the prostate
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.295
H-Index - 123
eISSN - 1097-0045
pISSN - 0270-4137
DOI - 10.1002/pros.24150
Subject(s) - prostatectomy , prostate cancer , biochemical recurrence , medicine , receiver operating characteristic , proportional hazards model , prostate , lasso (programming language) , prostate specific antigen , stage (stratigraphy) , urology , oncology , cancer , biology , computer science , paleontology , world wide web
Background Identification of novel biomarkers associated with high‐risk prostate cancer or biochemical recurrence can drive improvement in detection, prognosis, and treatment. However, studies can be limited by small sample sizes and sparse clinical follow‐up data. We utilized a large sample of prostate specimens to identify a predictive model of biochemical recurrence following radical prostatectomy and we validated this model in two external data sets. Methods We analyzed prostate specimens from patients undergoing radical prostatectomy at Hartford Hospital between 2008 and 2011. RNA isolated from formalin‐fixed paraffin‐embedded prostates was hybridized to a custom Affymetrix microarray. Regularized (least absolute shrinkage and selection operator [Lasso]) Cox regression was performed with cross‐validation to identify a model that incorporated gene expression and clinical factors to predict biochemical recurrence, defined as postoperative prostate‐specific antigen (PSA) > 0.2 ng/ml or receipt of triggered salvage treatment. Model performance was assessed using time‐dependent receiver operating curve (ROC) curves and survival plots. Results A total of 606 prostate specimens with gene expression and both pre‐ and postoperative PSA data were available for analysis. We identified a model that included Gleason grade and stage as well as five genes ( CNRIP1 , endoplasmic reticulum protein 44 [ ERP44 ], metaxin‐2 [ MTX2 ], Ras homolog family member U [ RHOU ], and OXR1 ). Using the Lasso method, we determined that the five gene model independently predicted biochemical recurrence better than a model that included Gleason grade and tumor stage alone. The time‐dependent ROCAUC for the five gene signature including Gleason grade and tumor stage was 0.868 compared to an AUC of 0.767 when Gleason grade and tumor stage were included alone. Low and high‐risk groups displayed significant differences in their recurrence‐free survival curves. The predictive model was subsequently validated on two independent data sets identified through the Gene Expression Omnibus. The model included genes ( RHOU , MTX2 , and ERP44 ) that have previously been implicated in prostate cancer biology. Conclusions Expression of a small number of genes is associated with an increased risk of biochemical recurrence independent of classical pathological hallmarks.