Megan Hall
More Information
Thesis: Sampling Considerations in Intensive Longitudinal Data
Advisor: Matt Shotwell
BS, Biochemistry (minors in Mathematics and Applied Statistics), Western Kentucky University
At University of Kentucky since 2024. Currently Biomedical Data Science Assistant.
MS thesis abstract:
Ansa cervicalis stimulation (ACS) has been proposed as a treatment option for obstructive sleep apnea. Airflow, CPAP pressure, and ACS stimulation status were measured continuously during drug-induced sleep endoscopy. Data was divided into breaths and experimental variables were summarized across each breath. Airflow was summarized in three ways: Vimax, mean flow, and middle third mean flow. The treatment effects of ACS were ∆Pcrit and ∆Popen. ∆P is the difference in pressure between ACS and no stimulation required to reach set levels of airflow. Airflow is zero for ∆Pcrit and the average flow for non-flow limited breaths for ∆Popen. Breaths were divided into experimental conditions, which are a set of successive breaths a unique patient, nasal pressure and stimulation status. Current experimental protocol allows three breaths per experimental condition. ∆P estimates were compared via sampling first N breaths per experimental condition and a simulation of the same sampling method. ∆P estimates were also compared via bootstrap sampling the number of patients. For the breath sampling, ∆Pcrit sub-sample estimates approached the full-sample estimates at three breaths per experimental condition, while ∆Popen estimates did not. For the simulation, bias between the true ∆P and the simulated ∆P estimate decreased as number of breaths sampled increased. Some ∆P estimates (∆Popen for Vimax, ∆Pcrit for mean and middle third mean flow) do not plateau at the true ∆P, indicating bias in the ∆P calculation. Estimates tended to plateau at or before six breaths per condition. Standard error decreased as number of breaths sampled increased for the simulation, but standard error across flow measures stabilized by four breaths per condition for breath sampling. The number of patients needed to obtain stable estimates of ∆Pcrit and ∆Popen was small for Vimax and mean flow but greater for middle third mean flow. Because sampling three breaths per condition may yield biased estimates of ∆P, increasing the number of breaths per condition to six may be beneficial. Few patients were needed to obtain stable estimates of ∆P but increasing the number of patients decreased standard errors for all flow measures.