Variable | IRR (95% CI) | % change | p Value |

Practice deprivation score | 1.016 (1.01 to 1.02) | 1.6% | <0.0001 |

Distance from hospital (miles) | 0.99 (0.98 to 0.996) | −1.0% | 0.0008 |

Size of practice (no of patients) | 0.998 (0.997 to 0.999) | −0.2% | 0.0002 |

Age (% of practice patients aged 65+ years) | 1.03 (1.02 to 1.04) | 3.0% | <0.0001 |

% of Practice white ethnicity | 1.005 (1.003 to 1.008) | 0.5% | <0.0001 |

% of Practice male | 0.98 (0.96 to 0.99) | −2.0% | 0.04 |

% Able to book with a specific GP | 0.993 (0.990 to 0.996) | −0.7% | 0.0001 |

Backward stepwise evictions | p Value when removed from model |

% Response rate to access survey | 0.95 |

% Able to book 2 days ahead | 0.92 |

Total QOF clinical points | 0.39 |

Total QOF organisational points | 0.48 |

% Satisfied with opening hours | 0.32 |

% Satisfied with phone access | 0.31 |

% Able to get appointment in 48 h | 0.09 |

N=145 for all variables.

Final multivariable model after backward selection process has been implemented for level 2 variables. Figures in the second column are incident rate ratios (IRR). Subtracting 1 from the IRR and then multiplying by 100 gives the percentage change in the expected admissions count for a 1 unit increase in the predictor. So for practice deprivation score, for every extra deprivation point, the expected admissions count increases by 1.6%. IRR less than 1.0 represent decreases and IRR greater than 1.0 represent increases in the count. Statistical model: negative binomial regression, using log of list size as the offset.

GP, general practitioner; QOF, Quality and Outcomes Framework.