Screening and diagnosis when within-individual observations are Markov-dependent

Area of Science:

• Biostatistics
• Medical Screening
• Time Series Analysis

Background:

• Current statistical models for screening studies assume independent repeat observations within subjects.
• This assumption impacts the accuracy of regression to the mean calculations and misclassification probabilities.
• Existing models may not adequately capture the temporal dependencies in repeated measurements.

Purpose of the Study:

• To extend the statistical model for regression to the mean in screening studies to incorporate Markov-dependent repeat observations.
• To provide new expressions for regression to the mean considering time delays and averaged measurements.
• To introduce a diagnostic tool for the conditional distribution of an individual's long-term mean and estimate autocorrelation in short time series.

Main Methods:

• Developed a Markov-dependent statistical model for repeat observations within subjects.
• Derived new formulas for regression to the mean incorporating time intervals and measurement averaging.
• Utilized the extended model to describe the conditional distribution of long-term means.
• Presented a method for estimating autocorrelation coefficients from short time series data.
• Applied the method to estimate autocorrelation in diastolic blood pressure measurements.

Main Results:

• The extended model provides more accurate regression to the mean estimates when observations are dependent.
• New expressions for regression to the mean are dependent on the time delay and averaging strategy.
• The conditional distribution of the long-term mean serves as a useful diagnostic tool.
• A practical method for estimating autocorrelation in short time series was demonstrated.
• Autocorrelation in diastolic blood pressure was successfully estimated from daily repeat observations.

Conclusions:

• The assumption of independent observations in screening studies can lead to inaccurate regression to the mean estimates.
• Incorporating Markov dependence in statistical models improves the analysis of repeated measurements.
• The developed methods offer enhanced tools for analyzing screening data and understanding individual health trajectories.
• This research has implications for designing more accurate screening protocols and interpreting health data over time.