Detectable and Lethal Prostate Cancer are Different Proteomic Diseases
Year: 2026
Type: article
Abstract: PSA testing detects prostate cancer but does not separate indolent from lethal disease. Whether the circulating proteins that precede a prostate cancer diagnosis are the same as those that precede death from prostate cancer has not been examined across the plasma proteome with time-to-event methods. Methods. In 24{,}090 men of the UK Biobank Olink proteomics subcohort without prostate cancer at baseline, we screened 2{,}920 proteins against incident prostate cancer (cancer registry, ICD-10 C61; 1{,}330 events) and death from prostate cancer (underlying cause C61; 126 deaths) over a median follow-up of 13.1 and 15.2 years. Each protein was residualized on age and ranked by Harrell's concordance ($C^*$)\footnote{$C^*$ denotes Harrell's concordance statistic for censored time-to-event data: the probability that, of two men whose order of events is known, the one with the earlier event has the higher score.}. Endpoint-specific Cox panels (Detection, Lethality), a stability-selected full-proteome model, a four-gate discovery against KLK3 and 24 clinical covariates, and a cross-cancer test against colorectal and lung cancer models were evaluated by nested cross-validation. Findings. KLK3 (PSA) ranked first for both endpoints ($C^* = 0.815$ for incidence, $0.798$ for mortality). Beyond KLK3 and GP2, the two top-10 lists shared no protein, and across all screened proteins the incidence and mortality rankings were uncorrelated (Spearman $\rho = 0.001$); eight of the ten leading mortality proteins were inversely associated with risk. In multivariable models the picture was different: age and KLK3 alone gave nested $C^* = 0.818$ for incidence and $0.822$ for mortality, and no panel improved on them. The Lethality panel, selected for mortality, predicted death less well than the Detection panel (nested $C^*$ 0.786 vs 0.825; difference $-0.039$, 95\% CI $-0.068$ to $-0.012$). The full-proteome mortality model retained KLK3 as its only stable protein in every training fold (nested $C^* = 0.821$). The colorectal and lung cancer models predicted prostate cancer incidence less well than the prostate model (nested $C^*$ 0.636 and 0.619 vs 0.810). Interpretation. At the level of single proteins, the circulating signatures preceding the diagnosis of prostate cancer and death from it are distinct. At the level of prediction, both endpoints are carried by age and KLK3, and with 126 deaths no lethality-specific protein added to them under nested validation. Distinguishing lethal disease will need cohorts with many more prostate cancer deaths and with grade and stage at diagnosis.
Author: Jacobs, Laurence A.
Language: English
Subfield: Pulmonary and Respiratory Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7225243414
FWCI:
Cites:
Cited by:
Related to: