A new approach to categorising continuous variables in prediction models: Proposal and validation

Abstract

When developing prediction models for application in clinical practice, health practitioners usually categorise clinical variables that are continuous in nature. Although categorisation is not regarded as advisable from a statistical point of view, due to loss of information and power, it is a common practice in medical research. Consequently, providing researchers with a useful and valid categorisation method could be a relevant issue when developing prediction models. Without recommending categorisation of continuous predictors, our aim is to propose a valid way to do it whenever it is considered necessary by clinical researchers. This paper focuses on categorising a continuous predictor within a logistic regression model, in such a way that the best discriminative ability is obtained in terms of the highest area under the receiver operating characteristic curve (AUC). The proposed methodology is validated when the optimal cut points' location is known in theory or in practice. In addition, the proposed method is applied to a real data-set of patients with an exacerbation of chronic obstructive pulmonary disease, in the context of the IRYSS-COPD study where a clinical prediction rule for severe evolution was being developed. The clinical variable PCO was categorised in a univariable and a multivariable setting.

Safety and treatment compliance of subcutaneous immunotherapy: A 30-year retrospective study

“…The binary outcome variable was the presence/absence of SEs (regardless of their actual number). In order to make the logistic regression estimates more easily interpretable, continuous variable such as age and year of inception were discretized into binary variables, using an optimal cut-point search algorithm [11]. The algorithm used determines the number and the location of the cut-points using the area under the curve (AUC) of the logistic model, suitably correcting the AUC obtained, which may be biased upward when the same data-set is used both to fit the logistic regression model (involved in the cut-point selection process) and compute the AUC [11].…”

Which urban land covers/uses are associated with residents’ mortality? A cross-sectional, ecological, pan-European study of 233 cities

“…This process was carried out separately for each macroregion. We categorised the data into quintiles rather than apply polynomial terms in order to aid interpretation and therefore help guide policymakers 22. Furthermore, this approach also allowed us to generate the marginal estimates, which were plotted.…”

Antibodies anti-Toxoplasma gondii and anti-Neospora caninum in backyard pigs from the state of Mato Grosso, Brazil

“…The continuous variable was categorized if the visual evaluation shown a nonlinear trend. We used the optimization methodology of the R package CatPredi, as described (BARRIO et al, 2017). Secondly, regarding the categorical variables, we used a chi-square or Fisher's exact test on the results, and then we included into the model the various predictor variables, and all the variables with a value of p < 0.20.…”