Methods for Handling Left-Censored Data in Quantitative Microbial Risk Assessment.

Area of Science:

• Environmental microbiology
• Quantitative microbial risk assessment (QMRA)
• Statistical data analysis

Background:

• Left-censored data, common in environmental microbiology, presents challenges for accurate risk assessment.
• Decisions on handling data below the limit of detection (LOD) can significantly impact quantitative microbial risk assessment outcomes.
• Enterovirus data in water frequently exhibits censoring, necessitating robust analytical methods.

Purpose of the Study:

• To evaluate various statistical methods for handling left-censored enterovirus data in environmental samples.
• To compare the performance of five different methods in estimating microbial dose and infection risk under varying degrees of censoring.
• To identify the most reliable method for QMRA when dealing with data below the LOD.

Main Methods:

• Simulated enterovirus water data sets were generated with censoring levels from 10% to 97%.
• Five methods were applied: LOD/[Formula: see text] substitution, lognormal maximum likelihood estimation (MLE), Kaplan-Meier estimation (KM), and two imputation methods (MI method 1 and MI method 2).
• Root mean square error (RMSE) and bias were used to compare estimated and known doses and infection risks.

Main Results:

• MI method 1 demonstrated the lowest RMSE and bias for dose and infection risk across most censoring degrees.
• MI method 2 performed as the second-best method overall.
• For medium to severe censoring (35%-90%), MI method 1 generally yielded the least error.

Conclusions:

• Imputation methods, especially MI method 1, are effective for handling left-censored data in QMRA for enteroviruses.
• MI method 1 provides accurate risk estimations even under high censoring (97%).
• MI method 2 is a viable alternative, particularly when the underlying data distribution is uncertain, mitigating risks of misspecification.