First-Cycle PSA Math Model Predicts Who Benefits Most From Adaptive Prostate Cancer Therapy

Why This Study Matters

Adaptive therapy is an evolution-based treatment strategy that intentionally controls tumor burden rather than eliminating it outright, using planned treatment breaks to delay the emergence of drug resistance. It has shown promise in prostate cancer, but patient responses vary widely, and clinicians currently have no reliable way to predict — up front — which patients stand to benefit most from an adaptive schedule versus continuous treatment.

That gap matters because a single time-point PSA measurement, the standard tool oncologists use to monitor prostate cancer, cannot by itself characterize a tumor's underlying growth dynamics or predict how a given patient will respond to treatment breaks. The authors set out to build a better predictive tool: a mathematical biomarker derived from the pattern of PSA change during just the first treatment cycle.

Study Design

Following the TRIPOD reporting guideline for prediction-model studies, the authors built a two-population differential equation model describing tumor growth as competing dynamics between drug-sensitive and drug-resistant cell populations. From this model they derived three first-cycle metrics: an adaptive therapy score, an expected time to progression (TTP), and an expected mean daily dose. The analysis drew on 53 patients across 2 independent, previously published clinical cohorts, spanning a combined enrollment period from 1996 to 2022, using just 1 cycle of PSA data per patient. Because all data came from previously published sources, no new institutional review board approval was required, and the model and biomarker derivations were fully prespecified before the external validation cohort was analyzed. As a retrospective modeling study rather than a prospective randomized trial, conventional statistics such as number of enrolling sites and randomization ratio are not reported in the source.

Patient Population

Enrollment into the Bruchovsky et al cohort required histologically confirmed adenocarcinoma with a rising serum PSA level after radiotherapy and no evidence of distant metastasis, with PSA and testosterone monitored every 4 weeks. Patients in the Zhang et al cohort had already progressed to metastatic castrate-resistant disease under first-line androgen deprivation therapy before enrollment.

Primary Endpoint Results

The primary outcome was time to progression (TTP); mean daily dose and overall survival were also assessed. Using only first-cycle PSA dynamics, the adaptive therapy score outperformed every standard PSA-based metric tested in both cohorts:

In the Zhang et al cohort, the adaptive therapy score also showed a strong rank correlation with clinical TTP (Spearman rho=0.76, P=.002). On simulated data from 5 virtual patient twins, expected TTP predicted the magnitude of TTP with a mean absolute error of 4.8% (SD 3.2%), and correlated with clinical outcomes (R-squared=0.65, P=.10 for expected TTP; R-squared=0.94, P=.01 for expected mean daily dose). As a separate internal consistency check against full patient-history treatment simulations — rather than against real clinical outcomes — the adaptive therapy score, expected mean daily dose, and expected TTP each showed strong agreement with their simulated benchmarks (R-squared=0.98, P=.004; R-squared=0.97, P=.006; and R-squared=0.98, P=.001, respectively). Extended follow-up of the Zhang et al cohort showed that both the adaptive therapy score and expected TTP were significantly associated with prolonged overall survival, while standard empirical PSA metrics showed no association with OS.

Subgroup Analyses

Not specified in source. This modeling study did not stratify its results by demographic or biomarker-defined patient subgroups (for example, age, race, or PD-L1/biomarker status). Its central comparison — summarized in the Primary Endpoint Results table above — was the adaptive therapy score against standard PSA-based metrics within each of the two independent cohorts, and no additional subgroup-level statistics beyond that head-to-head comparison were reported in the full text.

Safety Profile

Not specified in source. This is a retrospective computational modeling and biomarker-validation study built on previously collected PSA and outcome data — it is not a prospective treatment or safety trial, and the full text reports no adverse event or toxicity data. Readers should not infer safety conclusions about intermittent androgen deprivation therapy or adaptive abiraterone acetate from this article.

Interpretation and Broader Context

The authors frame their contribution plainly, arguing that mechanistically informed biomarkers built from first-cycle PSA dynamics can identify, before treatment even begins, which patients are most likely to benefit from an adaptive therapy schedule.

In survival analyses on 2 distinct clinical cohorts, these mechanistically informed biomarkers exhibit significantly stronger prognostic associations with TTP than traditional phenomenological PSA metrics.
— Discussion, JAMA Oncology

The paper also proposes an illustrative clinical workflow: patients could complete one cycle of standard adaptive therapy, have their adaptive therapy score and expected TTP computed from that cycle's PSA kinetics, and then be stratified into continuous therapy, standard adaptive therapy, or an enhanced adaptive therapy protocol. The authors are explicit that the 40% adaptive therapy score threshold used in this illustrative pipeline is not a validated clinical cutoff — it demonstrates feasibility only, and real-world implementation would require data-driven calibration against larger, prospectively collected cohorts. They note their framework is well suited to support contemporary de-escalation phase 3 trials in prostate cancer, while also flagging cautionary results from earlier landmark studies that found intermittent therapy could produce inferior outcomes in certain patient populations (those with nonmetastatic disease) — underscoring the need for exactly the kind of patient-selection biomarker this study proposes.

Limitations: the authors identify small cohort size (53 patients total across both cohorts) as the study's main limitation, and call for validation in larger, multicycle cohorts such as the ongoing phase 2 ANZADAPT trial (NCT05393791). They also note that patient-specific model parameters may drift over multiple treatment cycles as tumors continue to evolve, and propose that future work adopt a hierarchical Bayesian framework to continually update biomarker estimates as new patient data accumulates.