Premium
Implementing the use of nomograms by choosing threshold points in predictive models: 2012 updated P artin T ables vs a E uropean predictive nomogram for organ‐confined disease in prostate cancer
Author(s) -
Borque Ángel,
RubioBriones Jose,
Esteban Luis M.,
Sanz Gerardo,
DomínguezEscrig Jose,
RamírezBackhaus Miguel,
Calatrava Ana,
Solsona Eduardo
Publication year - 2014
Publication title -
bju international
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.773
H-Index - 148
eISSN - 1464-410X
pISSN - 1464-4096
DOI - 10.1111/bju.12532
Subject(s) - nomogram , receiver operating characteristic , confidence interval , medicine , cohort , predictive value , statistics , urology , mathematics
Objectives To implement the use of nomograms in clinical practice showing how to choose thresholds in nomograms’ predictions to select risk groups. To validate and compare the predictive ability and clinical utility of the H ospital U niversitario ‘ M iguel S ervet’ ( HUMS ) and the updated P artin T ables 2012 ( PT ‐2012) nomograms to predict organ‐confined disease ( OCD ) after radical prostatectomy ( RP ).Patients and Methods Cohort of 1285 patients with prostate cancer treated with RP at I nstituto V alenciano de O ncología ( IVO ) between 1986 and 2011. The predictive value of the nomograms was assessed by means of calibration curves, discrimination ability (area under the receiver operating characteristic ( ROC ) curve ( AUC ) and probability density functions). The clinical utility was evaluated through V ickers’ decision curves and thresholds were chosen through probability density functions.Results The calibration curves showed a minimal underestimation in low probabilities (<20%), a minimal overestimation in high probabilities (>50%) in the HUMS nomogram and a regular minimal overestimation in the PT ‐2012. Their AUC of 0.7285 (95% confidence interval [ CI ] 0.7010–0.7559) and 0.7288 (95% CI 0.7013–0.7562) respectively, show an adequate discrimination ability for both predictive models in the IVO cohort. The decision curves show similar net benefits for both models. In this study we advocate for a threshold of 53% for the identification of OCD .Conclusions The HUMS ‐nomogram and the PT ‐2012 predictions of OCD confirm their utility in a contemporary cohort of patients. Patients with a probability of OCD >53% should be classified as OCD , helping physicians to better counsel their patients. A selection of adequate thresholds, as presented in this paper, makes nomograms more accessible tools.