Personalized Decision Making for Biopsies in Prostate Cancer Active Surveillance Programs
Pith reviewed 2026-05-24 22:22 UTC · model grok-4.3
The pith
Joint models generate visit-specific progression risks that trigger biopsies only when needed, cutting the median number from ten to four with nearly identical detection delays.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Joint models for longitudinal PSA, DRE and time-to-progression data produce a patient- and visit-specific cumulative risk; a biopsy is performed at the current visit only when this risk exceeds a pre-specified threshold, resulting in a median of four biopsies (IQR 2-5) versus ten (IQR 3-10) under annual scheduling and median detection delays of 0.7 years (IQR 0.3-1.0) versus 0.5 years (IQR 0.3-0.8).
What carries the argument
Joint model that combines longitudinal trajectories of PSA and DRE with a survival sub-model for time to progression, yielding an updated cumulative risk at each follow-up visit that is compared directly against a decision threshold.
If this is right
- The number of biopsies required per detected progression drops substantially while the timing of detection stays comparable.
- Patients experience fewer invasive procedures without a meaningful increase in the window during which progression could remain undetected.
- The same risk-threshold rule can be recalibrated for different acceptable delay tolerances by changing the threshold value.
Where Pith is reading between the lines
- The approach could be extended to incorporate patient-specific utilities for biopsy burden versus delay, turning the fixed threshold into a personalized utility-weighted one.
- If the joint model is updated with new data streams such as MRI results, the risk estimates and resulting biopsy savings could increase further.
- Similar risk-based scheduling logic might apply to surveillance protocols in other low-risk cancers where repeated invasive tests are currently used on a fixed calendar.
Load-bearing premise
The joint model fitted to historical PRIAS measurements produces accurate patient-specific cumulative risks of progression at future visits that can be directly thresholded for biopsy decisions.
What would settle it
A prospective trial in which the personalized arm shows a median increase in detection delay of more than one year or a median biopsy reduction of fewer than two procedures compared with annual scheduling.
read the original abstract
Background: Low-risk prostate cancer patients enrolled in active surveillance programs commonly undergo biopsies for examination of cancer progression. Biopsies are conducted as per a fixed and frequent schedule (e.g., annual biopsies). Since biopsies are burdensome, patients do not always comply with the schedule, which increases the risk of delayed detection of cancer progression. Objective: Our aim is to better balance the number of biopsies (burden) and the delay in detection of cancer progression (less is beneficial), by personalizing the decision of conducting biopsies. Data Sources: We use patient data of the world's largest active surveillance program (PRIAS). It enrolled 5270 patients, had 866 cancer progressions, and an average of nine prostate-specific antigen (PSA) and five digital rectal examination (DRE) measurements per patient. Methods: Using joint models for time-to-event and longitudinal data, we model the historical DRE and PSA measurements, and biopsy results of a patient at each follow-up visit. This results in a visit and patient-specific cumulative risk of cancer progression. If this risk is above a certain threshold, we schedule a biopsy. We compare this personalized approach with the currently practiced biopsy schedules via an extensive and realistic simulation study, based on a replica of the patients from the PRIAS program. Results: The personalized approach saved a median of six biopsies (median: 4, IQR: 2-5), compared to the annual schedule (median: 10, IQR: 3-10). However, the delay in detection of progression (years) is similar for the personalized (median: 0.7, IQR: 0.3-1.0) and the annual schedule (median: 0.5, IQR: 0.3-0.8). Conclusions: We conclude that personalized schedules provide substantially better balance in the number of biopsies per detected progression for men with low-risk prostate cancer.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes personalizing biopsy decisions in prostate cancer active surveillance by fitting joint models to longitudinal PSA/DRE measurements and time-to-progression outcomes from the PRIAS cohort (5270 patients). At each visit, a patient-specific cumulative risk of progression is computed and a biopsy is scheduled if the risk exceeds a threshold. This strategy is compared to fixed schedules (e.g., annual) in a simulation study that replicates PRIAS patients, with the central claim being a median reduction of six biopsies (4 vs. 10) while maintaining comparable detection delay (0.7 vs. 0.5 years).
Significance. If the performance metrics hold under external validation, the work could reduce biopsy burden in low-risk prostate cancer management without compromising timely detection of progression, addressing a practical compliance issue in active surveillance programs. The scale of the PRIAS data and the use of joint models for dynamic risk prediction are strengths that would support clinical translation if the simulation bias concerns are resolved.
major comments (3)
- [Methods (simulation study)] Methods, simulation study description: The evaluation generates trajectories and event times from the same fitted joint model used to produce the risk predictions (a 'replica of the patients from the PRIAS program'), creating an in-sample assessment. This setup risks optimistic bias in the reported median biopsy savings and delay metrics, as model misspecification (e.g., in the association between PSA trajectory and progression hazard) would not be detected; no held-out real patients or external cohorts are mentioned.
- [Methods (joint model and decision rule)] Methods, risk threshold and joint model fitting: The threshold for scheduling biopsies is invoked without stating whether it was pre-specified, cross-validated, or tuned on the same PRIAS data that supplies the simulation patients. Absence of sensitivity analysis to threshold choice directly affects the robustness of the central claim that the personalized schedule achieves a median of six fewer biopsies.
- [Results / Abstract] Results and Abstract: The reported performance metrics (biopsy counts and delays) are generated from quantities defined by the fitted model itself, with no reported external validation, missing-data handling details, or calibration checks for the cumulative risk predictions; this circularity undermines the conclusion that personalized schedules provide 'substantially better balance'.
minor comments (2)
- [Abstract] Abstract: Provide the numerical value of the risk threshold used and note any sensitivity analyses, as these are essential for interpreting the simulation results.
- [Methods] Notation: Clarify whether the cumulative risk is the dynamic prediction from the joint model at each visit or a fixed quantity, to avoid ambiguity in the decision rule.
Simulated Author's Rebuttal
We thank the referee for the constructive comments on our manuscript. We address each of the major comments below and have made revisions to the manuscript accordingly where appropriate.
read point-by-point responses
-
Referee: Methods, simulation study description: The evaluation generates trajectories and event times from the same fitted joint model used to produce the risk predictions (a 'replica of the patients from the PRIAS program'), creating an in-sample assessment. This setup risks optimistic bias in the reported median biopsy savings and delay metrics, as model misspecification (e.g., in the association between PSA trajectory and progression hazard) would not be detected; no held-out real patients or external cohorts are mentioned.
Authors: The simulation study is designed to evaluate the decision-making strategy under the data-generating process defined by the joint model fitted to the PRIAS data. This approach is common in methodological papers on dynamic prediction to allow access to the true event times for calculating detection delays. We agree that this may lead to optimistic performance estimates if the model is misspecified, and we have added a paragraph in the Discussion section acknowledging this limitation and the need for future external validation studies. revision: yes
-
Referee: Methods, risk threshold and joint model fitting: The threshold for scheduling biopsies is invoked without stating whether it was pre-specified, cross-validated, or tuned on the same PRIAS data that supplies the simulation patients. Absence of sensitivity analysis to threshold choice directly affects the robustness of the central claim that the personalized schedule achieves a median of six fewer biopsies.
Authors: The risk threshold was selected to achieve a clinically acceptable trade-off between biopsy reduction and detection delay, based on exploratory analyses. To address the concern, we have performed a sensitivity analysis by varying the threshold and included the results in the revised manuscript, showing that the biopsy savings remain substantial across a range of thresholds. revision: yes
-
Referee: Results / Abstract: The reported performance metrics (biopsy counts and delays) are generated from quantities defined by the fitted model itself, with no reported external validation, missing-data handling details, or calibration checks for the cumulative risk predictions; this circularity undermines the conclusion that personalized schedules provide 'substantially better balance'.
Authors: We have clarified in the Methods that the joint model accounts for irregular visit times and missing measurements via the shared random effects. Additionally, we have added calibration assessments for the risk predictions in the supplementary materials. We have revised the Abstract and Conclusions to emphasize that the results are based on simulation under the fitted model, and moderated the claim of 'substantially better balance' to reflect this. revision: partial
- External validation using held-out or independent patient cohorts, as no such data were available for this analysis.
Circularity Check
Simulation study evaluates personalized biopsy policy on trajectories generated from the same fitted joint model
specific steps
-
fitted input called prediction
[Methods (joint model + simulation study)]
"Using joint models for time-to-event and longitudinal data, we model the historical DRE and PSA measurements, and biopsy results of a patient at each follow-up visit. This results in a visit and patient-specific cumulative risk of cancer progression. If this risk is above a certain threshold, we schedule a biopsy. We compare this personalized approach with the currently practiced biopsy schedules via an extensive and realistic simulation study, based on a replica of the patients from the PRIAS program."
The joint model is estimated on the PRIAS data; the simulation then creates replica patients whose PSA/DRE trajectories and progression times are drawn from that same fitted model. Applying the risk-threshold rule (also derived from the model) to these simulated trajectories therefore evaluates the policy on data whose generative process is identical to the model, rendering the biopsy-savings and delay figures statistically dependent on the model being correctly specified.
full rationale
The paper fits a joint model to the full PRIAS cohort to obtain patient-specific cumulative progression risks, then thresholds those risks to decide biopsies. The reported performance (median 6 fewer biopsies, comparable 0.7-year delay) is obtained from a simulation study that replicates PRIAS patients. Because the simulation generates new longitudinal trajectories and event times from the fitted joint model itself, the risk predictions and resulting biopsy counts are computed on data whose distribution is defined by the model parameters; any misspecification is invisible and the metrics reduce to in-sample behavior of the fitted model. This matches the fitted-input-called-prediction pattern and forces the central claim.
Axiom & Free-Parameter Ledger
free parameters (1)
- risk threshold for biopsy
axioms (1)
- domain assumption Joint model assumptions (random effects structure, proportional hazards, correct specification of longitudinal trajectories for PSA and DRE)
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.