Interrupted time series analysis modelling adjusted monthly mortality risk with no, 3-month and 6-month lag periods
| All alerts n=172 | |||
| Risk ratio | 95% CI | Change (%) | |
| Modelling with no lag | |||
| Prealert trend | 1.05 | (1.04 to 1.05) | 5*** |
| Level change (over lag) | 0.62 | (0.58 to 0.66) | −38*** |
| Postlag trend | 0.99 | (0.98 to 1.00) | −1 |
| Modelling 3-month lag | |||
| Prealert trend | 1.05 | (1.04 to 1.05) | 5*** |
| Level change | 0.54 | (0.50 to 0.59) | −46*** |
| Postalert trend | 0.99 | (0.99 to 1.00) | −1 |
| Modelling 6-month lag | |||
| Prealert trend | 1.05 | (1.04 to 1.06) | 5*** |
| Level change | 0.48 | (0.44 to 0.53) | −52*** |
| Postalert trend | 0.99 | (0.98 to 1.00) | −1 |
Risk ratios are the model-estimated ratios of relative risk of death (observed number/expected number). Trend risk ratios are monthly increases/decreases. Our model measures the trend prior to an alert, the level change during varying lag periods and postlag trend in relative risk of death. Models are adjusted for autocorrelation. The model uses generalised estimating equations and the Wald test statistical significance was ***p<0.001. The 172 alerts were generated between January 2011 and November 2013 and sent to 93 acute National Health Service trusts in England.