Table 2.
Quantitative measures of the impact of quarantine on secondary cases in the community*.
| Analysis approach | Measure of association obtained from model | Variance estimation method | Estimated value of measure of association | p-value | 95% confidence interval | |
|---|---|---|---|---|---|---|
| Lower limit | Upper limit | |||||
| A) Generalized linear model with binomial error and identity link (additive logistic). Unit of analysis is community contacts, clustered within shared 204 index cases. (N = 8498). | Naïve Secondary Attack Rate Difference (incorporates information only for 204 index cases with one or more community contact; see text) | Asymptotic: Bootstrapped: | 0.00105 0.00105† |
0.786 0.881 |
-0.00665 -0.01269 |
0.00866 0.01479 |
| B) Canonical Poisson and negative binomial regression with log link functions. Unit of analysis is all index cases. (N = 332). | Beta coefficient | Poisson, asymptotic: Neg.Binomial asymptotic: Poisson, bootstrapped: |
-1.15209 -1.15209 -1.15209† |
0.026 0.057 0.589 |
-2.16906 -2.33953 -5.32984 |
-0.13511 0.03535 3.02567 |
| Secondary Case Count Ratio (SCCR) (SCCR = exponentiated Beta for quarantine) | Poisson, asymptotic: Neg.Binomial asymptotic: Poisson, bootstrapped: |
0.31598 0.31598 0.31598† |
0.026 0.057 0.573 |
0.11428 0.09637 0.00577 |
0.87362 1.03599 17.30158 |
|
| C) Generalized linear model with Poisson error term and identity link function, yielding an additive effect measure. | Secondary Case Count Difference (SCCD).(SCCD = Beta coefficient for quarantine) | Poisson, asymptotic: Neg.Binomial, asymptotic Poisson, bootstrapped: |
-0.13322 -0.13322 -0.13322† |
0.001 0.002 1.000 |
-0.21346 -0.21812 -1.18 × 1014 |
-0.05298 -0.04832 1.18 × 1014 |
| Unit of analysis is all index cases. (N = 332). | Number needed to quarantine (NNQ). (NNQ = 1/|SCCD|, above) | Poisson, asymptotic: Neg.Binomial, asymptotic: Poisson, bootstrapped: |
7.50647 7.50648 7.50647† |
0.001 0.002 1.000 |
4.68469 4.58462 -8.476 × 10-15 |
18.87661 20.69733 8.476 × 10-15 |
* Data from 332 index community-living probable or suspect Severe Acute Respiratory Syndrome (SARS) cases and 8,498 associated community contacts. Ontario, Canada, SARS outbreak, 2003.
† Bootstrapping used to obtain variance estimate; point estimate fixed from corresponding asymptotic model.