Open AccessCorrecting Misreported Multinomial Outcome Data Based on Logistic Regression Model with Application to Stroke Mortality in Thailand
Author(s) -
Arinda Ma-a-lee,
Nattakit Pipatjaturon,
Phattrawan Tongkumchum
Publication year - 2017
Publication title -
walailak journal of science and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.146
H-Index - 15
eISSN - 2228-835X
pISSN - 1686-3933
DOI - 10.48048/wjst.2018.2817
Subject(s) - multinomial logistic regression , confidence interval , medicine , logistic regression , statistics , stroke (engine) , confounding , demography , regression analysis , mathematics , mechanical engineering , sociology , engineering
Causes of death in Thailand are misreported; about 40 % of deaths have been recorded as “ill-defined”. This study aims to describe statistical methods to correct misreported multinomial outcome by using verbal autopsy (VA) data. Since the outcome is a nominal variable, with 21 levels, the appropriate model for systematic analysis of death by ICD-10 code is multinomial regression. Moreover, it is simpler and more informative to separately fit logistic regression models to the 21 outcome cause groups, and then rescale the results to ensure that the total number of estimated deaths for each group match those reported in the corresponding populations. This method also gives confidence intervals for percentages of deaths in cause groups for levels of each risk factor, adjusted for other risk factors. These confidence intervals are compared with bar charts of sample percentages, to assess evidence of confounding bias. The methods were illustrated using stroke deaths. Area plots are used to show results by gender, age group, and year. The most misclassified stroke deaths were ill-defined, other cardio vascular disease, mental and nerve (outside-hospital), septicemia, and respiratory disease (in-hospital).