BERKELEY, CA (UroToday.com) - Some years ago Partin tables (PT) and risk classification methods appeared in the scientific arena of prostate cancer (PCa) prediction with increasing influence. Although those models are defined by their ease of use, our perception was that those models have not been common tools in clinical practice even after the development of web risk calculators. The lack of interpretability of predictive models could explain it. Furthermore, the advantages and disadvantages of applying those models remain unclear.
PT are focused on the prediction of pathological stage of PCa. These predictive models of pathological stage have remained in the back of analysis of PCa due to the emergence of predictive models of recurrence after primary treatment or metastasis. However, the prediction of organ-confined disease from clinic-pathological biopsy data has become crucial nowadays, perhaps more than ever before. The reason for that is that most of our patients come from population screening or opportunistic screening, so that we are offering them conservative approaches, i.e., active surveillance, focal therapy, or nerve-sparing radical prostatectomy. These therapies are proposed with the assumption that our patients have organ-confined PCa, but it is based on the experience of urologists and risk groups rather than on pathological certainty. Multivariate models predicting organ-confined PCa can help in this clinical decision-making as well in the counseling of our patients. On these grounds we decided to conduct an external validation in a cohort European Caucasian men of the last PT update in 2012 (PT-2012)[1] and a nomogram of organ-confined PCa previously published (Hospital Universitario Miguel Servet, HUMS-nomogram).[2, 3]
For this purpose, we performed an external validation beyond the traditional methodology. We used calibration curves to show how the predictions offered by these models correspond with the actual occurrence of organ-confined disease in this external validation cohort, the area under the ROC curve to assess the discriminatory ability of the predictive models, and Vickers’ decision curves to demonstrate the clinical utility of the model at different thresholds points. Moreover, in order to explore the predictive ability of the PT-2012 and HUMS-nomogram, we analyze the probability density functions (PDF) of organ-confined disease.
The PDF should provide a clear plot of the clinical benefit of applying these prediction models in our population. In this sense, we think that a prediction model needs a tool that informs the use of the predictions given by the nomograms. It is important to know what probability of organ-confined PCa is enough to consider it a true organ-confined PCa and, therefore, to offer a conservative therapy. It is advisable to provide a threshold point of probability in order to make decisions such as cutoffs of PSA, free PSA, PCA3, glucose, or hemoglobin, among others. A PSA above 20 ng/mL is considered a PCa and a prostate biopsy should be offered to patients. On the other hand, urologists tend to adopt a cutoff of 2.5 ng/mL to help them decide what a “normal” PSA is or not. Cutoffs points are essential in clinical practice.
However, researchers who develop nomograms, tables, and prediction models do not generally offer cutoffs to guide decision-making. A prediction probability of an organ-confined disease of 85% or 90% is a clear value to advise our patients. However, how to treat patients with these high estimated probabilities was already clear for urologists even before the application of predictive models because of their values of PSA, clinical stage, or Gleason-score that build the final prediction. Problems arise with probabilities distant from 85-90%, and this is a consequence of the failure of previous investigations to provide cut-off points of probabilities, which allow clear decisions. Vickers’ decision curves try to solve this problem showing the net benefit of choosing different cut-off points. In our study, we propose another complementary rule to choose them, the PDF.
Through these PDFs, we can easily draw the distribution of patients with and without the event (organ-confined PCa in our case) through their individual probabilities assigned by predictive models. We can also graphically identify how different cut-off points selected to the right or the left, identify patients with or without the event. This intuitive method should be used in our models to propose cut-off points and thus to show them to users to gain confidence in them.
In conclusion, the practical goals of our project are several. The study can help us to revitalize the predictive models for organ-confined PCa in contemporary conservative approaches, to validate the last PT-2012 update and the HUMS-nomogram in a cohort of European patients, and finally to offer both the scientific community and users a graphical and intuitive method to choose cut-off points in multivariate models, nomograms, and predictive tables.
References:
- Eifler JB, Feng Z, Lin BM, Partin MT, Humphreys EB, Han M, Epstein JI, Walsh PC, Trock BJ, Partin AW. An updated prostate cancer staging nomogram (Partin tables) based on cases from 2006 to 2011. BJU Int. 2013;111:22-9.
- Borque A, Sanz G, Allepuz C, Plaza L, Gil P, Rioja LA. The use of neural networks and logistic regression analysis for predicting pathological stage in men undergoing radical prostatectomy: a population based study. J Urol 2001; 166:1672-1678.
- Fernando AB, Escaño LM, Saiz GS, Sanz LA. Predictive models for biochemical recurrence of prostate cancer after local treatment. Nomograms. Arch Esp Urol 2012;65:39-50.
Written by:
Ángel Borque, MD, PhD as part of Beyond the Abstract on UroToday.com. This initiative offers a method of publishing for the professional urology community. Authors are given an opportunity to expand on the circumstances, limitations etc... of their research by referencing the published abstract.
Department of Urology, Hospital Universitario ‘Miguel Servet’, Zaragoza, Spain
More Information about Beyond the Abstract