Predictive models for preterm birth*
Model 1† | Model 2‡ | Model 3§ | |
OR (95% CI) | OR (95% CI) | OR (95% CI) | |
Ethnicity | |||
White/Caucasian (ref) | – | – | – |
Non-white | 1.50 (1.11 to 2.04) | 1.48 (1.11 to 1.96) | 1.49 (1.13 to 1.99) |
Parity | |||
Multiparous (ref) | – | – | – |
Primiparous | 1.49 (1.21 to 1.84) | 1.52 (1.19 to 1.93) | 1.53 (1.20 to 1.95) |
Body mass index | |||
Normal weight (ref) | – | – | – |
Underweight | 0.99 (0.46 to 2.10) | 1.01 (0.47 to 1.14) | 1.00 (0.35 to 2.83) |
Overweight | 1.18 (0.88 to 1.57) | 1.14 (0.76 to 1.68) | 1.13 (0.72 to 1.78) |
Obesity | 1.94 (1.41 to 2.65) | 1.95 (1.25 to 3.04) | 1.95 (1.16 to 3.30) |
Smoked before pregnancy | |||
No (ref) | – | – | – |
Yes | 1.20 (0.90 to 1.60) | 1.19 (0.78 to 1.79) | 1.19 (0.77 to 1.82) |
Depression during pregnancy | |||
No (ref) | – | – | – |
Yes | 1.10 (0.84 to 1.46) | 1.12 (0.76 to 1.66) | 1.13 (0.74 to 1.71) |
Household income | |||
≥$100 000 (ref) | – | – | – |
$70 000 to <$100 000 | 0.82 (0.61 to 1.12) | 0.82 (0.51 to 1.33) | 0.84 (0.55 to 1.28) |
$40 000 to <$70 000 | 0.75 (0.70 to 1.31) | 0.96 (0.57 to 1.62) | 0.99 (0.58 to 1.69) |
<$40 000 | 0.92 (0.71 to 1.66) | 1.05 (0.60 to 1.81) | 1.10 (0.63 to 1.88) |
Neighbourhood SES | – | ||
Q1 least deprived (ref) | – | – | |
Q2 | 0.86 (0.53 to 1.39) | 0.97 (0.64 to 1.49) | |
Q3 | 0.96 (0.58 to 1.59) | 0.87 (0.52 to 1.47) | |
Q4 | 0.99 (0.60 to 1.58) | 0.90 (0.51 to 1.59) | |
Q5 most deprived | 1.20 (0.63 to 1.85) | 1.01 (0.55 to 1.86) | |
Neighbourhood-level variance | – | 0.15 (0.03 to 0.89) | 0.14 (0.03 to 0.88) |
ICC (%)¶ | – | 4.45 (0.07 to 23.25) | 4.27 (0.06 to 23.59) |
MOR | – | 1.46 | 1.44 |
Proportion of neighbourhood level variance explained by neighborhoodneighbourhood SES (%) | – | 25.00 | 25.16 |
AUC | 0.60 (0.56 to 0.63) | 0.75 (0.73 to 0.78) | 0.75 (0.72 to 0.77) |
*Prediction models were developed in bootstrapped samples with 1000 replications.
†Conventional logistic regression model that includes individual level predictors.
‡Multilevel logistic regression model that includes random intercept at neighbourhood level, neighbourhood deprivation index and all the individual level predictors contained in the logistic regression model.
§Multi-level logistic regression model that includes random intercept at neighbourhood level, neighbourhood median personal income and all the individual level predictors contained in the logistic regression model.
¶ICC calculation follows standard logistic distribution with variance π2/3 for the level 1, where π denotes the mathematical constant 3.1416.
AUC, area under the receiver operating characteristic curve; ICC, intra-cluster correlation; MOR, median OR.