Objective: The accuracy of 1997 Partin Tables' lymph node invasion (LNI) predictions exhibits important variability in different testing populations. We explored the LNI predictive accuracy in radical prostatectomy (RP) patients from Montreal, Canada. Moreover, we assessed the extent of change in predictive accuracy related to a modification of PSA coding from categorical to continuous.
Methods: We used pretreatment serum PSA, clinical stage, and biopsy Gleason sum from 537 men treated with RP to compare predicted and observed rates of LNI. Accuracy was quantified with receiver-operating characteristics curves.
Results: Accuracy was 0.760 in 369 evaluable patients, when categorically coded pretreatment PSA (0-4, 4.1-10, 10.1-20, 20.1+) was combined with clinical stage and biopsy Gleason sum. A 2.7% accuracy increase was noted when categorically coded PSA was replaced with continuously coded values.
Conclusion: Partin Tables' LNI predictions showed comparable accuracy to a community-based sample from the United States (0.766), and to a recent, multi-institutional sample (0.740). However, accuracy was lower than reported in internal (0.818), and external (0.837) academic, validation cohorts. Accuracy of LNI predictions was appreciably higher, when continuously coded PSA was used.