Outcomes Of Dichotomizing A Continuous Variable In The Psychiatric Early Readmission Prediction Model



Objective: To illustrate the results of dichotomizing a continuous variable in a prediction model. Method: 202 patients who were discharged from the psychiatric ward, University Malaya Medical Centre (UMMC) from 27th August 2007 to 15th April 2008 were recruited. The general psychopathology was measured with Brief Psychiatric Rating Scale (BPRS-24). The information on age, gender, race, marital status, occupation, psychiatric diagnosis, first onset, electroconvulsive therapy, substance use, oral treatment and depot injection were collected. On follow up, the patients who had early readmission (<6 months) were identified. Univariate analysis of early readmission using independent t-test for continuous BPRS scale and Chi square test for dichotomized BPRS scale were conducted. Logistic regression model to determine early readmission based on all variables with BPRS score as continuous and dichotomized scale was compared.

Result: Both univariate tests showed that BPRS score was significantly associated with early readmission. The power of the test reduced from 0.97 to 0.65 after dichotomizing the BPRS score with alpha of 0.05. The performance of the logistic regression model decreased (AUC: 0.749 to 0.728; Nagelkerke R2: 0.244 to 0.211) and significance of hypothesis testing reduced (P value: 0.002 to 0.036) after dichotomization.

Conclusion: Dichotomizing a continuous variable in statistical analysis leads to underpowered of a test and reduced performance of a prediction model.

Keywords: Dichotomizing, prediction model, continuous variable


Dichotomizing, prediction model, continuous variable

Full Text: