Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Review
. 2020 Aug;146(8):2067-2075.
doi: 10.1007/s00432-020-03286-8. Epub 2020 Jun 17.

A clinician's guide for developing a prediction model: a case study using real-world data of patients with castration-resistant prostate cancer

Kevin M Veen #  1 , Isabel B de Angst #  2 , Mostafa M Mokhles  1 , Hans M Westgeest  3 , Malou Kuppen  4 , Carin A Uyl-de Groot  4 , Winald R Gerritsen  5 , Paul J M Kil  6 , Johanna J M Takkenberg  1
Affiliations
Review

A clinician's guide for developing a prediction model: a case study using real-world data of patients with castration-resistant prostate cancer

Kevin M Veen et al. J Cancer Res Clin Oncol. 2020 Aug.

Abstract

Purpose: With the increasing interest in treatment decision-making based on risk prediction models, it is essential for clinicians to understand the steps in developing and interpreting such models.

Methods: A retrospective registry of 20 Dutch hospitals with data on patients treated for castration-resistant prostate cancer was used to guide clinicians through the steps of developing a prediction model. The model of choice was the Cox proportional hazard model.

Results: Using the exemplary dataset several essential steps in prediction modelling are discussed including: coding of predictors, missing values, interaction, model specification and performance. An advanced method for appropriate selection of main effects, e.g. Least Absolute Shrinkage and Selection Operator (LASSO) regression, is described. Furthermore, the assumptions of Cox proportional hazard model are discussed, and how to handle violations of the proportional hazard assumption using time-varying coefficients.

Conclusion: This study provides a comprehensive detailed guide to bridge the gap between the statistician and clinician, based on a large dataset of real-world patients treated for castration-resistant prostate cancer.

Keywords: Castration-resistant prostate cancer; Cox proportional hazard model; Decision-making; Prediction modeling.

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no conflict of interest related to this paper.

Figures

Fig. 1
Fig. 1
Example of a continuous outcome (y-axis) and continuous predictor (x-axis). As is shown: with the assumption the relation is linear the model (red line) does not fit the observed data well (black dots)
Fig. 2
Fig. 2
Example of relaxation of the linear assumed association (red line) of a continuous outcome and predictor. This can be done either with natural splines (green line) or fractional polynomials (FP) (blue line). Using splines the data is divided in separate sections, and each section has its own estimate of the line. Using fractional polynomials the relationship is described as multiple polynomials, which can produce a very flexible line
Fig. 3
Fig. 3
a Example of a Schoenfeld residuals plot in order to check the proportional hazard assumption. When the hazard of WHO is assumed constant over time (blue line in part a), the assumption is violated, especially in the first ten months the blue line deviates from the red line. In part b we have two coefficients for WHO, one for the first ten months and one for more than ten months. Proportional hazards assumption is not violated anymore
See this image and copyright information in PMC

References

    1. Alba AC, Agoritsas T, Walsh M, Hanna S, Iorio A, Devereaux PJ et al (2017) Discrimination and calibration of clinical prediction models: users' guides to the medical literature. JAMA 318(14):1377–1384 - PubMed
    1. Babyak MA (2004) What you see may not be what you get: a brief, nontechnical introduction to overfitting in regression-type models. Psychosom Med 66(3):411–421 - PubMed
    1. Breheny P, Huang J (2015) Group descent algorithms for nonconvex penalized linear and logistic regression models with grouped predictors. Stat Comput 25(2):173–187 - PMC - PubMed
    1. Carroll KJ (2003) On the use and utility of the Weibull model in the analysis of survival data. Control Clin Trials 24(6):682–701 - PubMed
    1. Collins GS, Reitsma JB, Altman DG, Moons KG (2015) Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. J Clin Epidemiol 68(2):134–143 - PubMed

LinkOut - more resources