Methods for estimating kidney disease stage transition probabilities using electronic medical records

Area of Science:

• Biostatistics
• Epidemiology
• Nephrology

Background:

• Chronic diseases, including chronic kidney disease (CKD), are staged by severity, influencing clinical decisions.
• Estimating transition rates between CKD stages is vital for patient management but challenging due to data variability.

Purpose of the Study:

• To develop and validate a method for estimating CKD stage transition rates using electronic health records (EHR).
• To address the challenges of irregular data collection and varying observation times in EHR data.

Main Methods:

• Utilized hidden Markov models (HMMs) to estimate transition rates, accounting for varying information levels and observation times.
• Employed a discretization method to transform daily data into manageable intervals (30, 90, 180 days) for computational efficiency.
• Validated the approach through simulation studies assessing accuracy, computation time, and the impact of informative observation times and missing data.

Main Results:

• The developed HMM approach demonstrated good performance, even with non-ignorable missing data.
• Applied to over 60,000 primary care patients with CKD (stage 2+), the method estimated transition rates between six disease states.
• Estimated transition rates were found to be similar between men and women.

Conclusions:

• The study presents a robust method for estimating CKD stage transition rates from EHR data, overcoming common analytical challenges.
• The findings provide valuable quantitative insights into the dynamics of CKD progression in a large primary care population.
• This methodology can inform clinical practice and public health strategies for managing chronic kidney disease.