The results in indicate that the subpopulation of women, aside from being ages 18-50 years, is estimated to be 66% white, 59% well-educated (more than a high school degree), well-represented in each of the income categories, more than 50% married, and 50% healthy in terms of periodontal health status; more than 90% have never had endometriosis, 73% have a low cotinine level, only 8% have never had a child, and 94% are not currently pregnant. In addition, the prevalence of diabetes is under 3%. We also note that the process of multiple imputation did not change the subpopulation estimates substantially, and provided complete data sets for the multivariable analyses.
Weighted estimates of frequency distributions for the analysis variables, for the subpopulation of interest (before and after multiple imputations of item-missing values).
Results from fitting the binary logistic regression model to the outcome broadly measuring periodontal health status (any periodontal disease vs. healthy) are presented in , before and after multiple imputations of the missing data. In the complete-case analyses, the following predictors were not found to have a significant relationship with periodontal disease in design-based Wald tests, and were dropped from the model: diabetes, age at first period, high serum cotinine, age, and the parity indicator. After dropping these predictors, all other covariates with the exception of endometriosis (education, ethnicity, income, and pregnancy) had a statistically significant (p < 0.05) association with the outcome in the complete case analysis. Due to missing data, only 2,052 observations (out of 4,136) were used to fit the initial model in the complete case analysis, and only 2,804 observations were used to fit the subsequent reduced model. The same reduced model was fitted when performing the multiple imputation analyses, and endometriosis was found to have a marginally significant association with the binary outcome.
Table 2 Weighted Estimates of Adjusted Odds Ratios (AORs) and design-based 95% confidence intervals (CIs), indicating the association of endometriosis with the binary periodontal health status outcome (before and after multiple imputations), in addition to all (more ...)
The results of the multivariable analyses presented in indicate that in general, white women in this subpopulation have the lowest odds of having poorer periodontal health (for example, Mexican-American women are estimated to have between 76% (multiple imputation analysis) and 84% (complete-case analysis) higher odds of having any periodontal disease compared to white women). In addition, higher income and higher education result in lower odds of having a poor outcome, while being pregnant increases the odds of having a poor outcome by roughly 40 to 50%. We note that endometriosis has a marginally significant association with the odds of having a poor outcome based on the multiple imputation analysis, where women who have been told that they have endometriosis have 31% higher odds of having a poor outcome (AOR = 1.31, 95% CI = 0.91, 1.88).
presents results from fitting the multinomial logistic regression model to the outcome measuring the four specific categories of dental health in the complete case analysis. Wald tests of the independent predictors in the complete case analysis revealed that the following predictors should be dropped from the model, due to lack of significance (p < 0.10): age at first period, diabetes, and the parity indicator. The design-based Wald test for the endometriosis indicator in the reduced model indicated that endometriosis had a statistically significant association with the four-category outcome (Wald test p = 0.0189) when controlling for the relationships of the other significant predictors with the outcome in the complete case analysis. Specifically, the results showed that women with endometriosis had significantly (57%) higher odds of having gingivitis and periodontitis relative to not having periodontal disease, compared to women without endometriosis (AOR: 1.57; 95% CI: 1.06, 2.33). Due to missing data, only 2,052 observations (out of 4,136) were used to fit the initial model, and only 2,664 observations were used to fit the reduced model.
Table 3 Weighted Estimates of Adjusted Odds Ratios (AORs) and design-based 95% confidence intervals (CIs), indicating the association of endometriosis with the four-category periodontal health status outcome (in the complete case analysis, before multiple imputations), (more ...)
The results from the multivariable analysis presented in suggest other meaningful associations based on the cases with complete data. In general, lower education levels increase the risk of having adverse outcomes; younger women have a higher odds of having gingivitis only relative to being healthy; white women again have a reduced odds of having adverse outcomes; higher income tends to result in a lower odds of having adverse outcomes; higher cotinine levels tend to increase the odds of having adverse outcomes; and currently being pregnant tends to increase the odds of having adverse outcomes as well.
The results based on the analysis of the complete data set (throwing out cases with missing data) were essentially replicated in the multiple imputation analyses, suggesting that the findings are robust. Specifically, the risk of having both adverse outcomes relative to being healthy was found to be increased by 44% after controlling for the relationships of the other predictors in with the four-category outcome (RRR = 1.44, 95% CI = 0.92, 2.25). Even after using a statistically valid technique to impute the missing values (14
), we have essentially the same findings, only using a complete data set rather than one with 50% of the cases lost due to missing data. This suggests the findings would remain stable in an even larger sample.