Table 3.  Sensitivity analysis: logistic regressions of whether or not the tooth lost 3 or more millimeters of attachment between the baseline and 48-month examinations

                                                Odds ratios                                                          

                                                                                Scenario #1:                                                                          Scenario #2:  

                                                                                all lost teeth                          Lost teeth treated as            all lost teeth

                                                                                ARE assumed                       missing data (same              are assumed NOT

Explanatory covariate                                          to have had ALI                   method as in Table 2)          to have had ALI

Tooth-specific characteristics

Worst attachment level on the tooth at  

baseline was 7 or more mm                3.17 * 2.69 * 1.65

Was less than 7 mm                                            -- -- --                                            

Tooth was a molar 1.83 * 1.81 * 1.80 *

Tooth was a premolar 1.13                                         1.33 1.34

Tooth was a canine                                             1.25                                         1.27 1.34

Tooth was an incisor                                           --                                             --                                             --                            

Person characteristics during follow-up

Person reported losing at least one tooth

due to a periodontal reason^† --                                             --                                             --            

Person lost at least one tooth, but did not     

                report that this was due to a

periodontal reason^† --                                             --                                             --                            

Person did not lose a tooth^† --                                             --                                             --            

Person did not report at least one dental

cleaning                                                 2.65 * 2.08 *                                     1.89 *

Person reported at least one dental cleaning --                                             --                                             --

Person characteristics at baseline

Does not floss                                                      1.42 * 1.86 *                                     1.86 *    

Flosses some                                                        --                                             --                                             --                            

Current cigar or cigarette smoker                      1.74 * 1.86 *                                     1.81 *

Never smoked or former smoker                        --                                             --                                             --                            

____________________________________________________________________________________________

Sample size

Number of persons                                              586                                          586                                          586                                         

Number of teeth                                                   11,557                                     10,702                                     11,557    

^† These variables were not included because only persons who lost a tooth were eligible to be asked these questions.  Therefore, including these variables would lead to artifactually high parameter estimates for their effects.

Statistically significant parameter estimates (p < 0.05) are highlighted in bold and italic font.

The outcome of interest (attachment loss of a specific tooth) was coded 1 if the tooth lost 3 or more millimeters of attachment between the baseline and 48-month examinations, and 0 if not.  The GENMOD procedure in SAS^®  was used with a logit link, binomial distribution, exchangeable correlation structure.  Although the tooth was the unit of analysis, clustering of teeth within persons was accounted for in the correlation structure, allowing for both tooth-specific and person-level characteristics to be tested.  Multicollinearity was measured using a procedure described by Belsley and colleagues,³⁵ which is available in the REG procedure of SAS.²⁷