Table 3. Logistic regression model^a showing factors associated with transmission unconfirmed by DNA fingerprinting in tuberculosis case pairs identified in contact investigations in seven sites, 1996–2000.

Factor	Adjusted odds ratio (95% confidence intervals)	p value
Smear-negative source case	2.0 (1.2 to 3.1)	0.001
Foreign-born secondary case	3.4 (2.0 to 6.0)	<0.0001
Secondary case <15 yrs	0.3 (0.2 to 0.8)	0.01
Secondary case >65 yrs	1.7 (0.8 to 3.6)	0.2
Workplace setting	2.7 (0.9 to 5.7)	0.08

^aThe model also included in the multiple logistic regression analysis all levels of the following factors that were not significantly associated with unconfirmed transmission: cavitary disease in source case, race/ethnicity of case pair, and HIV status of secondary case. In the goodness-of-fit test, full model’s p value was 0.885; and the p value for a reduced model containing only the variables with significant associations was 0.725. Thus, both models fit adequately. For the full model, –2 x log likelihood was 564.044; for the reduced model, –2 x log likelihood was 571.469. The difference was distributed as a chi-square variable with 1 degree of freedom. The full model gave a better fit than the reduced model and was retained (p=0.0064, chi square >7.425).