Table 2.
Estimates of serum prolactin thresholds to discriminate between micro- and macroadenomas in a cohort of N = 133 prolactinoma patients and associated performance metrics.
| Method | All patients (N = 133) | Female patients (N = 91) | Male patients (N = 42) | |||
|---|---|---|---|---|---|---|
| Bayesian logistic regression | Youden Index | Bayesian logistic regression | Youden Index | Bayesian logistic regression | Youden Index | |
| Prolactin threshold (µg/L) | 239.4 (44.0–451.2) | 230.0 (203.0–466.1) | 211.6 (29.0–426.2) | 203.0 (189.8–243.2) | 1,046.1 (582.2–2,325.9) | 1,179.0 (596.2–1,510.0) |
| AUROC | 0.91 (0.85–0.95) | 0.87 (0.78–0.94) | 0.93 (0.83–0.99) | |||
| Sensitivity | ||||||
| Global threshold | 0.79 (0.72–0.99) | 0.84 (0.70–0.94) | 0.61 (0.47–0.99) | 0.64 (0.42–0.75) | 0.97 (0.97–0.99) | 0.97 (0.97–0.97) |
| Gender threshold | – | 0.69 (0.47–0.99) | 0.75 (0.61–0.78) | 0.74 (0.54–0.97) | 0.74 (0.69–0.94) | |
| Specificity | ||||||
| Global threshold | 0.90 (0.10–0.97) | 0.91 (0.81–0.99) | 0.95 (0.09–0.99) | 0.95 (0.89–0.99) | 0.57 (0.14–0.71) | 0.57 (0.43–0.71) |
| Gender threshold | – | 0.91 (0.00–0.98) | 0.89 (0.82–0.95) | 0.99 (0.71–0.99) | 0.99 (0.71–0.99) | |
| Positive predictive value | ||||||
| Global threshold | 0.91 (0.56–0.96) | 0.92 (0.83–0.99) | 0.89 (0.42–0.99) | 0.88 (0.81–0.99) | 0.92 (0.85–0.94) | 0.92 (0.89–0.94) |
| Gender threshold | – | 0.84 (0.40–0.94) | 0.82 (0.74–0.88) | 0.99 (0.94–0.99) | 0.99 (0.94–0.99) | |
| Negative predictive value | ||||||
| Global threshold | 0.78 (0.74–0.94) | 0.83 (0.72–0.93) | 0.78 (0.74–0.93) | 0.80 (0.72–0.84) | 0.80 (0.50–0.99) | 0.80 (0.75–0.83) |
| Gender threshold | – | 0.82 (0.74–0.99) | 0.84 (0.79–0.85) | 0.43 (0.30–0.83) | 0.44 (0.39–0.71) | |
AUROC, area under the receiver operating characteristic.
The most likely estimates and 95% credible intervals are shown for thresholds derived with a multilevel Bayesian logistic regression framework (BLRM). Median values and bootstrapped 95% confidence intervals are shown for the threshold estimates derived with the Youden Index. Median and 95% credible intervals (for thresholds derived with the BLRM) and 95% confidence intervals (for thresholds derived with the Youden Index) are shown for the performance metrics. For the female and male patients, performance metrics are shown for two cases: First, when a global (gender-unspecific) threshold is used to compute the confusion matrix. Second, when a gender-specific threshold is used compute the confusion matrix.