Abstract:
|
One of the most challenging aspects of health care is patient adherence to medication. Adherence to medication and symptom levels are often measured as continuous outcomes but currently analyzed after dichotomization, often yielding misleading results. Improved analytic methods that utilize the whole continuum of adherence and symptom levels, taking into consideration the complex nature of patients' adherence behavior are needed. We develop a set of analytic methods that incorporates adherence and symptom levels as continuous variables and thereby reveal relationships between non-adherence and symptom return not detected by traditional methods. We evaluate the effect of non-adherence by measuring the lag between dose level and symptom level time series, by adapting and extending two statistical methods recently developed for neuroimaging and physics applications: 1) temporal kernel canonical correlation analysis (tkcca) and visibility graph analysis (vga). We compare the proposed methods among themselves and with traditional methods in terms of sensitivity to detect the effect of non-adherence, sample size requirements and handling of missing data, using simulations and real data.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.