Table 3.
Logistic Regressions of Tooth Loss, Showing Effects as Odds Ratios with 95% Confidence Intervals and in Which the Tooth is the Unit of Analysis
| Characteristic | Model 1 (Race, SES Only) | Model 2 (Model 1, Plus Need Variables) | Model 3 (Model 2, but Only Includes Persons with a Dental Visit during Interval) |
|---|---|---|---|
| Race | |||
| African American | 2.18 (1.33, 3.57) | 1.39 (0.87, 2.23) | 1.75 (1.10, 2.79) |
| non-Hispanic white | 1.00 | 1.00 | 1.00 |
| Education | |||
| 8th grade or less | 1.96 (1.13, 3.39) | 1.31 (0.73, 2.33) | 2.04 (1.10, 3.78) |
| Some high school | 1.00 | 1.00 | 1.00 |
| Household Income | |||
| Below $20,000 | 1.60 (0.98, 2.61) | 1.38 (0.86, 2.22) | 1.71 (1.05, 2.77) |
| At or above | 1.00 | 1.00 | 1.00 |
| Need variables | — | (included)a | (included)a |
| Model Statistics | |||
| Number persons | 745 | 745 | 613 |
| Number tooth-intervals | 26,538 | 24,695 | 18,489 |
| Deviance | 7,018.9 | 4,961.8 | 4,323.2 |
Parameter estimates were converted to odds ratios with 95% confidence intervals. Statistically significant odds ratios (p < .05) appear in bold and italic font.
The outcome of interest (loss of a specific tooth) was coded 1 if the tooth was lost, and 0 if not. Therefore, this is a tooth-level analysis, not a person-level analysis (as appears in Table 2). Although the tooth is the unit of analysis, clustering of teeth within persons is accounted for in the correlation structure, allowing for both tooth-specific and person-level characteristics to be tested. The observation period was divided into two 24-month intervals, such that the need variables (clinical examination measures) from the baseline examination were used to predict tooth loss by the 24-month time point, and clinical variables from the 24-month exam were used to predict tooth loss between the 24-month and 48-month exams.
These need variables were included: tooth-specific variables determined by direct clinical examination (periodontal [gum] attachment level, upper/lower jaw location, tooth type, tooth severely loose, tooth had a fractured filling, root fragment, tooth surfaces with fillings), and two person-level signs/symptoms reported during the interval by interview (toothache and loose tooth). Consistent with the notion that clinical variables are important to understanding social disparities in tooth loss, each of these need variables were statistically significant with large effect magnitudes. The rationale for including these variables is found in Gilbert, Shelton et al. 2002. Because accounting for clinical detail is not the focus of this report, and for ease of presentation, results from the full model are not presented. However, they are available at http://nersp.nerdc.ufl.edu/~gilbert/supplemental.html.