Validación del índice de salud prostática en un modelo predictivo de cáncer de próstata

espanolObjetivos Validar y analizar la utilidad clinica de un modelo predictivo de cancer de prostata que incorpora el biomarcador «[–2] proantigeno prostatico especifico» a traves del indice de salud prostatica (PHI) en la toma de decision para realizar una biopsia de prostata. Material y metodos Se aislo suero de 197 varones con indicacion de biopsia de prostata para la determinacion del antigeno prostatico especifico total (tPSA), fraccion libre de PSA (fPSA) y [-2] proPSA (p2PSA); el PHI se calculo como p2PSA/fPSA×√tPSA. Se crearon 2modelos predictivos que incorporaban variables clinicas junto a tPSA o a PHI. Se evaluo el rendimiento de PHI usando analisis de discriminacion mediante curvas ROC, calibracion interna y curvas de decision. Resultados Las areas bajo la curva para el modelo tPSA y el modelo PHI fueron de 0,71 y 0,85, respectivamente. PHI mostro mejor capacidad de discriminacion y mejor calibracion para predecir cancer de prostata, pero no para predecir un grado de Gleason en la biopsia ≥7. Las curvas de decision mostraron un beneficio neto superior del modelo PHI para el diagnostico de cancer de prostata cuando el umbral de probabilidad esta entre 15 y 35% y un mayor ahorro (20%) en el numero de biopsias. Conclusiones La incorporacion de p2PSA a traves de PHI a los modelos predictivos de cancer de prostata mejora la exactitud en la estratificacion del riesgo y ayuda en la toma de decision sobre realizar una biopsia de prostata. EnglishObjectives To validate and analyse the clinical usefulness of a predictive model of prostate cancer that incorporates the biomarker «[–2] pro prostate-specific antigen» using the prostate health index (PHI) in decision making for performing prostate biopsies. Material and methods We isolated serum from 197 men with an indication for prostate biopsy to determine the total prostate-specific antigen (tPSA), the free PSA fraction (fPSA) and the [-2] proPSA (p2PSA). The PHI was calculated as p2PSA/fPSA×√tPSA. We created 2 predictive models that incorporated clinical variables along with tPSA or PHI. The performance of PHI was assessed with a discriminant analysis using receiver operating characteristic curves, internal calibration and decision curves. Results The areas under the curve for the tPSA and PHI models were 0.71 and 0.85, respectively. The PHI model showed a better ability to discriminate and better calibration for predicting prostate cancer but not for predicting a Gleason score in the biopsy ≥7. The decision curves showed a greater net benefit with the PHI model for diagnosing prostate cancer when the probability threshold was 15-35% and greater savings (20%) in the number of biopsies. Conclusions The incorporation of p2PSA through PHI in predictive models of prostate cancer improves the accuracy of the risk stratification and helps in the decision-making process for performing prostate biopsies.