Abstract:
|
Selection bias can result in biased estimates of epidemiologic parameters (i.e. incidence, prevalence) in a survey study due to non-respondents. This paper focuses on the importance of collecting secondary data and appropriate method selection to reduce the bias. We use a population survey, along with secondary patients' information from Electronic Medical Record (EMR) as secondary data, from a health care system to estimate age-specific prevalence of urinary incontinence (UI) from six models: M1: direct estimate using early respondents M2: direct estimate using all respondents M3: application of an inverse probability weighting (IPW) scheme method 1 and secondary data M4: application of an IPW scheme to method 2 and secondary data M5: application of a multiple imputation (MI) scheme to method 1 and secondary data Method 6: application of a MI scheme to method 1 and secondary data M2, 3 and 4 results in similar age-specific prevalence estimate, while prevalence estimate from M1 is higher for older age groups. M5, 6 tend to overestimate. Inverse probability weighting method yields the most consistent estimate by incorporating the probability estimated from secondary data.
|