Background. The timing of cancer onset and progression is central to early detection strategies. Classical screening theory relies on statistical quantities such as mean sojourn time, inferred indirectly from population data. In contrast, the biological age of an individual tumor—from clonal origin to diagnosis—remains unmeasured, despite its importance for understanding tumor growth dynamics and screening windows.

Objectives. To develop and validate a DNA methylation–based molecular clock that estimates patient-specific tumor age. Unlike conventional epigenetic clocks, which track age-related methylation drift, this clock measures tumor mitotic age (cell divisions since clonal origin) and corresponding calendar age.

Methods. We developed a molecular clock based on ensembles of neutrally fluctuating CpG sites (fCpGs) whose stochastic methylation dynamics are measured using standard DNA methylation arrays. Tumor ages were derived from the epigenetic entropy of fCpGs using statistical inference and mathematical modeling. Cancer site-specific clocks were applied to breast cancer and neuroblastoma. In breast cancer, we trained the clock using methylation array data from 634 primary tumors and validated it in independent primary and paired primary–metastasis cohorts. We estimated tumor ages by molecular subtype and examined associations with prognostic markers, gene expression pathways, and immune infiltration. In neuroblastoma, we developed and validated a clock using array data from 213 tumors in an unscreened cohort and 105 tumors in an independent cohort, and reconstructed tumor initiation relative to historical infant screening windows.

Results. In breast cancer, mitotically younger tumors were more aggressive, with higher proliferation and genomic instability, whereas older tumors showed increased immune infiltration, consistent with stronger immune-mediated growth suppression. Estimated calendar ages ranged from months to decades and differed markedly by subtype, with a median of 1.0 year in basal cancers versus 6.5 years in luminal A cancers. In neuroblastoma, aggressive stage 4 tumors predominantly initiated after the first year of life and had short preclinical growth periods, whereas indolent tumors often began in utero or early infancy. This indicates that neuroblastoma screening failed due to a mismatch between the screening window and the onset of lethal disease rather than insufficient test sensitivity.

Conclusions. We introduce a molecular approach for measuring individual tumor age. As a measure of time since clonal origin, it provides an upper bound on tumor-specific sojourn time while remaining independent of screening modality, a key advantage over classical screening parameters. Direct measurement of heterogeneity in tumor growth histories offers a new empirical basis for refining screening windows and intervals, and for anticipating screening benefit and overdiagnosis.