TABLE 2.
An Application of At-risk-measure Sampling in an Example Study Estimating the IRR of Bicycle Crashes Comparing Roadway Types in Atlanta, Georgia Between 1 October 2016 and 31 July 2018
| Measure | Scenario | Exposed (95% CI) | Unexposed (95% CI) | Exposure Ratio (95% CI) | IRR (95% CI) |
|---|---|---|---|---|---|
| Number of crashes | No bias adjustment | 104 (73, 140) | 25 (15, 38) | 4.2 (2.1, 8.2) | |
| Bicycle–distance (person–km) | No bias adjustment | 1.5 × 106 (1.2 × 106, 1.8 × 106) | 1.1 × 106 (0.9 × 106, 1.2 × 106) | 1.4 (1.1, 1.9) | 3.0 (1.4, 5.9) |
| Bicycle–distance (person–km) | IPSW only | 1.3 × 107 (0.5 × 107, 3.0 × 107) | 1.0 × 107 (0.3 × 107, 1.9 × 107) | 1.3 (1.1, 1.9) | 3.1 (1.4, 6.0) |
| Number of crashes | Standardization only | 3.4 (1.8, Inf) | |||
| Bicycle–distance (person–km) | Standardization only | 1.4 (1.1, 1.9) | |||
| Pseudo-IRa (per sampled person–km) | Standardization only | 5.6 × 10−5 (3.1 × 10−5, 7.9 × 10−5) | 2.3 × 10−5 (0.0, 3.5 × 10−5) | 2.4 (1.2, Inf) | |
| Number of crashes | IPSW, then standardization | 3.5 (1.8, Inf) | |||
| Bicycle–distance (person–km) | IPSW, then standardization | 1.4 (1.1, 2.0) | |||
| IR (per estimated person–km) | IPSW, then standardization | 6.4 × 10−6 (2.2 × 10−6, 19.5 × 10−6) | 2.5 × 10−6 (0.0, 8.0 × 10−6) | 2.6 (1.2, Inf) |
The 95% CI was estimated by the percentile method using nonparametric hierarchical bootstrapping (N, replications = 1,000). For scenarios including IPSW, the 95% CI also considers the uncertainty due to the parametric sampling-fraction model.
Please see text for exposure definition and additional detail.
a
The pseudo-incidence rate is the number of crashes divided by the sampled person–distance, following the terminology on p. 113 of Modern Epidemiology, 3rd Edition.21