Robust estimation of dementia prevalence from two-phase surveys with non-responders via propensity score stratification

BMC Medical Research Methodology

Research Article

Ke Deng¹  Chong Shen¹  Luning Wang²  Jiping Tan²  Yiming Zhao³  Minyue Pei³  Xiaoxiao Wang³  Nan Li³ 

[1] Center for Statistical Science, Department of Industrial Engineering, Tsinghua University, No. 30, Shuangqing Road, Haidian District, 100084, Beijing, People’s Republic of China;Geriatric Neurology Department of The Second Medical Center & National Clinical Research Center for Geriatric Diseases, Chinese PLA General Hospital, No. 28, Fuxing Road, Haidian District, 100039, Beijing, People’s Republic of China;Research Center of Clinical Epidemiology, Peking University Third Hospital, No. 49, Huayuan North Road, Haidian District, 100191, Beijing, People’s Republic of China;Key Laboratory of Epidemiology of Major Diseases (Peking University), Ministry of Education, Beijing, People’s Republic of China;

关键词: Prevalence estimation;    Missing data;    Propensity score;   

DOI  :  10.1186/s12874-023-01954-0

 received in 2022-09-20, accepted in 2023-05-16,  发布年份 2023

来源: Springer

【 摘 要 】

BackgroundMissing diagnoses are common in cross-sectional studies of dementia, and this missingness is usually related to whether the respondent has dementia or not. Failure to properly address this issue can lead to underestimation of prevalence. To obtain accurate prevalence estimates, we propose different estimation methods within the framework of propensity score stratification (PSS), which can significantly reduce the negative impact of non-response on prevalence estimates.MethodsTo obtain accurate estimates of dementia prevalence, we calculated the propensity score (PS) of each participant to be a non-responder using logistic regression with demographic information, cognitive tests and physical function variables as covariates. We then divided all participants into five equal-sized strata based on their PS. The stratum-specific prevalence of dementia was estimated using simple estimation (SE), regression estimation (RE), and regression estimation with multiple imputation (REMI). These stratum-specific estimates were integrated to obtain an overall estimate of dementia prevalence.ResultsThe estimated prevalence of dementia using SE, RE, and REMI with PSS was 12.24%, 12.28%, and 12.20%, respectively. These estimates showed higher consistency than the estimates obtained without PSS, which were 11.64%, 12.33%, and 11.98%, respectively. Furthermore, considering only the observed diagnoses, the prevalence in the same group was found to be 9.95%, which is significantly lower than the prevalence estimated by our proposed method. This suggested that prevalence estimates obtained without properly accounting for missing data might underestimate the true prevalence.ConclusionEstimating the prevalence of dementia using the PSS provides a more robust and less biased estimate.

【 授权许可】

CC BY   
© The Author(s) 2023

【 图 表 】

40517_2023_258_Article_IEq119.gif

Fig. 1

40517_2023_258_Article_IEq115.gif

40517_2023_258_Article_IEq114.gif

40517_2023_258_Article_IEq113.gif

Fig. 1

Fig. 1

40517_2023_256_Article_IEq14.gif

Fig. 8

Fig. 6

Fig. 7

40517_2023_256_Article_IEq11.gif

Fig. 5

40517_2023_256_Article_IEq5.gif

40517_2023_256_Article_IEq4.gif

41116_2023_36_Article_IEq117.gif

41116_2023_36_Article_IEq116.gif

41116_2023_36_Article_IEq786.gif

41116_2023_36_Article_IEq785.gif

41116_2023_36_Article_IEq783.gif

41116_2023_36_Article_IEq781.gif

41116_2023_36_Article_IEq778.gif

41116_2023_36_Article_IEq777.gif

41116_2023_36_Article_IEq774.gif

Fig. 2

41116_2023_36_Article_IEq765.gif

41116_2023_36_Article_IEq724.gif

40517_2023_258_Article_IEq51.gif