PREDICT

Prostate cancer · Radiotherapy

Pre-treatment risk of nodal or distant spread in prostate cancer, and whether to include the pelvis in radiotherapy.

See the study

Calculator

years

ng/mL

Biopsy Gleason
Primary pattern
Secondary pattern

Biopsy cores optional
positive
examined

Clinical T stage optional

T1 includes T1a to T1c. If T2 or T3 is not subclassified, choose T2a or T3a. N/A: not available.

Data are processed only in your browser: nothing is sent or stored.

The study

Development and multicohort external validation

The decision to include the pelvis still rests on the Roach formula, which uses only PSA and Gleason score and overestimates risk in contemporary patients. PREDICT uses the full biopsy report and estimates the risk that an adequate lymph node dissection would find.

73,203 SEER patients diagnosed between 2018 and 2022
0.793 AUC for nodal involvement in 17,164 operated patients, vs 0.749 for Roach
7.0 fewer pelvic fields per 100 patients than the POP-RT criterion, with the same coverage
1,923 operated patients in 4 external cohorts (106 node-positive)

Predicted risk matches observed risk

Proportion of patients with positive nodes in each band of PREDICT risk.

SEER, operated (17,164) External cohorts (1,923)

Bars: observed proportion; lines: 95% CI. With dissections of 15 or more nodes, the SEER values were 4.0%, 10.2%, 14.2% and 42.8%.

Fewer pelvic fields, same coverage

Per 100 operated SEER patients: who would receive pelvic irradiation under each criterion.

PREDICT
Roach formula

The PREDICT threshold that covers the same proportion of node-positive patients as the POP-RT criterion is 15.3%: 7.0 fewer pelvic fields per 100 patients (95% CI, 4.9 to 8.4).

Reference: pathologic nodal status in 17,164 operated patients without distant metastases. Cross-validated estimates.

Discrimination in development and in four external populations

AUC for pathologic nodal involvement (95% CI).

PREDICT Roach formula

PREDICT was applied without refitting or imputation, using the variables recorded in each cohort. Summary: random-effects meta-analysis (I², 0%).

01

Population

73,203 patients with histologically confirmed prostate adenocarcinoma diagnosed between 2018 and 2022 (SEER, 8 registries); 17,164 operated with lymph node dissection.

02

Predictors

Age, PSA, primary and secondary Gleason patterns, positive and examined cores, and clinical T stage. Missing variables are not imputed.

03

Model

Gradient boosting with monotonic constraints. A single value: the risk of distant metastasis plus the nodal risk — referenced to a 15-node dissection — in patients without them.

04

External validation

Rochester, Seoul, MSKCC and San Raffaele: 1,923 operated patients, 106 node-positive. Compared with the Roach formula at the RTOG 94-13 and POP-RT thresholds.

PREDICT provides a calibrated estimate of dissemination risk from routine variables, transports to external populations and selects patients for pelvic irradiation better than the Roach formula. We recommend its clinical use with a 15% threshold, adjustable to each population.

Authors

Agustin Rosich · Belem Cuervo · Joaquin Ferrer · Ester Sanchez · Pierina Debenedetti · Emiliano Rivero · Santiago Roldan · Federico Lorenzo

Unidad Académica de Radioterapia, Universidad de la República (UDELAR), Montevideo, Uruguay

Unidad Académica de Radioterapia Hospital de Clínicas Dr. Manuel Quintela Universidad de la República, Uruguay