Evaluation of Parametric Method Performance for Left-Censored Data and Recommendation of Using forCovid-19 Data Analysis

Abstract

Objectives: Left-censored data, which is commonly seen in clinical studies, are frequently encountered in the litera ture, especially in the fields of food, environment, microbiology, and biochemistry. In this study, the most appropriatedistribution between the negatively skewed distributions for left-censored data in Parametric Inverse Hazard Modelswas tried to be determined. Methods: Within the scope of the study, firstly, the data were produced uncensored according to different parametersof each distribution. Then, simulation studies were carried out in different censorship rates (15%, 25% and 35%) andvarious sample sizes (1000, 2000 and 3000) in order to determine the most appropriate distribution. AIC, AICC, HQIC,and CAIC information criteria were employed to compare the distribution performances. Since it was not possible tostudy simulations of all possible scenarios, scenarios similar to each other were generally preferred over others. Results: In the simulation results, the most appropriate distributions to be used for left-censored data in ParametricInverse Hazard Models were found as Generalized Inverse Weibull as well as Log-Logistic, Log-Normal, Inverse Normaland Gamma distributions. It was also detected that the Marshal-Olkin distribution revealed a superior performancecompared to the Modified Weibull, Generalized Gamma, Gamma, and Flexible Weibull distributions. Log logistics dis tribution gave the most appropriate result among the analyzed distributions in the examination made with real dataapplication. Conclusion: The use of censored data analysis in evaluations in terms of Covid-19 is quite additive, considering thatmore statistical evaluation will be needed in the next period of the epidemic. Improved estimates can be obtainedwith this approach, especially in Covid-19 data analysis.