Table 3.
Multiple logistic regression of ESCUW (vs no ESCUW) on performance on the separate examination components of each examination. Data are necessarily structurally missing as candidates who do not pass one component do not continue to take later components, so that missing values have been replaced using 100 multiple imputations, and there is also right and left truncation. The table shows the effect for each exam component, after partialling out the effects of all other components. The columns show b = loge(OR), OR (i.e. eb) and the OR for comparing the likelihood of ESCUW in candidates at the 2.5th percentile (i.e. 2 SDs below the population mean examination performance) and at the 97.5th percentile (i.e. 2 SDs above the population mean), performance of all other assessments being taken as at the mean. All b values for predictors are significant with p < 0.001
| Exam | Component | b = loge(OR) | OR = eb (95% CI) |
OR for ESCUW 2.5th percentile vs 95th percentile |
|---|---|---|---|---|
| MRCGP n = 27,561 |
Constant | − 4.482 | 0.011× | |
| AKT | − 0.307 | 0.736× (0.659, 0.821) | 3.33× | |
| CSA | − 0.569 | 0.566× (0.495, 0.647) | 9.30× | |
| MRCP(UK) n = 44,314 |
Constant | − 4.726 | 0.009× | |
| Part 1 | − 0.206 | 0.814× (0.703, 0.943) | 2.24× | |
| Part 2 | − 0.309 | 0.734× (0.617, 0.874) | 3.36× | |
| PACES | − 0.496 | 0.609× (0.538, 0.689) | 6.99× |