Comparison of Cox Proportional Hazards Regression and Multiple Logistic Regression Models for Identifying Factors...

Retinoblastoma is the most common intraocular rare disease in children arising in the retinal cells. This study aimed to evaluate the consistency of the findings of factors associated with mortality in retinoblastoma children between the Cox Proportional Hazards Regression Model (Cox PHR) and the Multiple Logistic Regression Model (MLR).

A retrospective cohort study was conducted from January 2004 to April 2023 on 402 children diagnosed with retinoblastoma who were admitted to hospitals with paediatric ophthalmology services in Malaysia. The statistical analyses used to model the factors associated with mortality were Cox PHR and the MLR. The five key elements evaluated for parameter estimates encompass direction, estimation, precision, significance, and magnitude of risk. The performance of the models was assessed using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC).

Among 402 children, 22 (5.5%) died, 27 (6.7%) were lost to follow-up, and 353 (87.8%) remained alive. Prolonged delay in diagnosis and non-compliance with follow-up were significant factors associated with mortality in both models. In the Cox PHR model, prolonged delay in diagnosis (AHR: 4.45, 95% CI: 1.86-10.62, p = 0.001) and non-compliance with follow-up (AHR: 5.21, 95% CI: 2.24-12.10, p < 0.001) increased mortality risk. Similar findings were observed in the MLR model, with prolonged delay in diagnosis (AOR: 4.73, 95% CI: 1.84-12.15, p = 0.001) and non-compliance with follow-up (AOR: 9.51, 95% CI: 3.75-24.11, p < 0.001) significantly increasing the odds of mortality. Both regression models showed consistent direction, and significant results.

Both Cox PHR and MLR models consistently identified prolonged delay in diagnosis and non-compliance with follow-up as factors associated with mortality among children with retinoblastoma.