Sensitivity analyses
| Mortality increment (%) per degree below/above 18°C (95% CI) | |||
|---|---|---|---|
| Cold | Heat | AIC | |
| Main model | 2.3 (0.7 to 3.8) | 1.7 (−2.9 to 6.5) | 1943.7 |
| Alternative distributional assumptions (main=overdispersed Poisson) | |||
| Simple regression | 2.1 (0.5 to 3.7) | 1.8 (−2.5 to 6.4) | NA |
| Negative binomial | 2.5 (1.1 to 3.8) | 1.5 (−2.9 to 6.2) | NA |
| Alternative time spline (main=6) | |||
| 3-knot spline | 1.1 (−0.9 to 3.1) | −0.2 (−6.2 to 6.2) | 3043.8 |
| 9-knot spline | 2.3 (0.8 to 3.9) | 2.8 (−1.9 to 7.6) | 1824.0 |
| Additional step for smaller discontinuities (see web appendix A) | |||
| Years starting 57 and 58 | 2.6 (1.0 to 4.2) | 2.4 (−2.4 to 7.4) | 1903.8 |
| Year starting 64 | 2.4 (0.8 to 3.9) | 1.6 (−3.0 to 6.4) | 1909.5 |
| Years starting 75 and 76 | 2.4 (0.9 to 3.9) | 2.2 (−2.9 to 7.4) | 1812.6 |
| Alternative heat/cold threshold (main=18) | |||
| Threshold 15°C | 2.2 (0.5 to 4.0) | −0.2 (−2.8 to 2.4) | 1966.5 |
| Threshold 21°C | 2.1 (0.8 to 3.5) | 8.4 (−4.8 to 23.4) | 1938.0 |
| Influenza control | |||
| No adjustment for influenza | 3.0 (0.9 to 5.2) | 6.3 (−0.2 to 13.2) | 3152.1 |
AIC, Akaike's information criterion: (−2×ln(likelihood)+2(terms)); NA, not available.