Table 3 Characteristics of misclassifications (false-positives and false-negatives) between opioid classifier and manual screen (n = 1091)
From: External validation of an opioid misuse machine learning classifier in hospitalized adult patients
Patient characteristics | Likelihood of opioid misuse in patient chart review | p value | |||
|---|---|---|---|---|---|
Definitely | Highly Probable | Probable | Definitely Not | Â | |
n | 99 | 21 | 413 | 558 | Â |
Predicted probability from opioid classifier (mean ± SD) | 0.96 ± 0.07 | 0.93 ± 0.11 | 0.88 ± 0.12 | 0.81 ± 0.12 |  < 0.001 |
Age (mean ± SD) | 49.9 ± 13.3 | 52.4 ± 13.8 | 53.3 ± 12.3 | 45.3 ± 14.4 |  < 0.001 |
Chronic pain, n (%) | 26 (26.3%) | 6 (28.6%) | 127 (30.8%) | 107 (19.2%) | 0.001 |
Prior mention of opioid misuse, n (%) | 92 (92.9%) | 18 (85.7%) | 263 (63.7%) | 2 (0.4%) |  < 0.001 |
Trauma or intoxication evidence, n (%) | 12 (12.1%) | 0 (0%) | 5 (1.2%) | 2 (0.4%) |  < 0.001 |
Withdrawal or overdose symptoms, n (%) | 14 (14.1%) | 1 (4.8%) | 15 (3.6%) | 5 (0.9%) |  < 0.001 |
Naloxone administered, n (%) | 2 (2%) | 0 (0%) | 5 (1.2%) | 7 (1.3%) | 0.869 |
Positive drug screen, n (%) | Â | Â | Â | Â | Â |
 Opiates | 38 (38.4%) | 9 (42.9%) | 74 (17.9%) | 56 (10.0%) |  < 0.001 |
 Benzodiazepines | 11 (11.1%) | 3 (14.3%) | 37 (9.0%) | 104 (18.6%) |  < 0.001 |
 Cannabinoids | 11 (11.1%) | 3 (14.3%) | 28 (6.8%) | 78 (14.0%) | 0.005 |
 Cocaine metabolites | 26 (26.3%) | 2 (9.5%) | 45 (10.9%) | 73 (13.1%) | 0.001 |
 Other | 3 (3%) | 2 (9.5%) | 14 (3.4%) | 35 (6.3%) | 0.116 |
 All negative | 25 (25.3%) | 5 (23.8) | 70 (16.9%) | 162 (29.0%) |  < 0.001 |
 Not available | 38 (38.4%) | 10 (47.6%) | 254 (61.5%) | 164 (29.4%) |  < 0.001 |
Administered opioids during encounter, n (%) | 37 (37.4%) | 9 (42.9%) | 163 (39.5%) | 117 (21.0%) |  < 0.001 |
Administered benzodiazepines during encounter, n (%) | 17 (17.2%) | 3 (14.3%) | 37 (9.0%) | 210 (37.6%) |  < 0.001 |
Opioids prescribed upon discharge, n (%) | 15 (15.2%) | 3 (14.3%) | 83 (20.1%) | 58 (10.4%) |  < 0.001 |
Other substance misuse, n (%) | 56 (56.6%) | 13 (61.9%) | 239 (57.9%) | 373 (66.8%) | 0.021 |
 Alcohol | 21 | 5 | 84 | 256 |  < 0.001 |
 Cocaine metabolites | 39 | 6 | 166 | 132 |  < 0.001 |
 Cannabis | 17 | 5 | 78 | 103 | 0.91 |
 Other | 1 | 2 | 30 | 27 | 0.079 |
Diagnosed psychiatric disorders, n (%) | |||||
 Depression | 22 (22.2%) | 2 (9.5%) | 101 (24.5%) | 198 (35.5%) |  < 0.001 |
 Anxiety | 15 (15.2%) | 2 (9.5%) | 78 (18.9%) | 127 (22.8%) | 0.126 |
 Bipolar | 14 (14.1%) | 5 (23.8%) | 36 (8.7%) | 82 (14.7%) | 0.015 |
 Schizophrenia | 5 (5.1%) | 0 (0%) | 8 (1.9%) | 23 (4.1%) | 0.154 |
 Post traumatic disorder | 2 (2.0%) | 2 (9.5%) | 18 (4.4%) | 40 (7.2%) | 0.082 |
 Other | 9 (9.1%) | 1 (4.8%) | 22 (5.3%) | 45 (8.1%) | 0.315 |
  No psychiatric history | 55 (55.6%) | 14 (66.7%) | 244 (59.1%) | 240 (43.0%) |  < 0.001 |