Table 2.
Marker | Unit for odds ratio | N | Odds ratio | (95% CI) | AUC for ROC | P value |
Predicting hypophysitis | ||||||
GNAL fold change | 0.10 | 20 | 5.13 | (1.95, infinity) | 1.00 | <0.001 |
GNAL pretreatment | 0.10 | 20 | 2.66 | (1.14, 7.29) | 0.79 | 0.02 |
GNAL posttreatment | 0.10 | 20 | 2.77 | (1.30, 9.03) | 0.92 | 0.004 |
ITM2B fold change | 0.10 | 20 | 3.93 | (1.59, infinity) | 1.00 | <0.001 |
ITM2B pretreatment | 0.10 | 20 | 0.99 | (0.20, 4.31) | 0.55 | >0.99 |
ITM2B posttreatment | 0.10 | 20 | 3.28 | (0.88, 15.99) | 0.75 | 0.08 |
Predicting pneumonitis | ||||||
CD74 fold change | 0.01 | 32 | 1.54 | (1.07, 2.21) | 1.00 | 0.02 |
CD74 pretreatment | 0.01 | 32 | 1.25 | (1.03, 1.52) | 0.76 | 0.03 |
CD74 posttreatment | 0.01 | 32 | 1.66 | (1.14, 2.42) | 0.95 | 0.01 |
Due to the separation of values between those with and without the adverse outcome, the odds ratios are very high, and the small sample sizes mean large confidence intervals. The units for the odds ratio show a continuous effect for every increment of that size over the narrow range of values present in our samples (Figs. 1 and 2). For example, a patient with a GNAL fold increase 0.10 higher than another patient is estimated to be 5.1 times more likely to have hypophysitis than the patient with the lower increase. Due to the very small sample size, the confidence interval for this estimate varies widely (1.95, infinity). Similarly, an increase of 0.25 (not shown) is associated with an odds ratio and confidence interval of 59.7 (5.3, infinity). The effect is so large for CD74 that the odds ratio unit is 0.01. An increase of 0.25 (not shown) is associated with an odds ratio and confidence interval of 50,294 (6, 422,160,000). P values set in boldface indicate statistical significance. ROC, receiver operating characteristic.