Table 4.
Drug characteristics with poorly discriminative index keywords but achieved an overall good predictive performance
| Category | Characteristic | Position of IK(s) | n(pos) | Best AUC* | Top-20 predictive tokens/words (AUC†) |
|---|---|---|---|---|---|
| AE | Cystitis | 10.7 pct | 13 | 1.000 | nsaids, cyclooxygenase, nimesulide, meloxicam (0.999); nsaid, diclofenac, naproxen, antiinflammatory, non-steroidal (0.998); ibuprofen, anti-inflammatory, nonsteroidal (0.997); ketoprofen, antipyretic (0.996); indomethacin (0.993); osteoarthritis (0.991); pge2 (0.991), prostanoid (0.988), thromboxane (0.986), prostaglandin (0.985) |
| AE | Dyslipidaemia | 14.0 pct | 13 | 0.900 | Aldosterone (0.95); acetazolamide, mineralocorticoid (0.94); deoxycorticosterone (0.93), pge2, indomethacin, hearing (0.88); spironolactone, mineralocorticoids, hyponatremia, renin, adh, ace (0.87); furosemide (0.86); insipidus, asthmatic, prostaglandins, fev1, pra, phenylbutazone (0.85) |
| AE | Migraine | 76.0 pct | 14 | 0.906 | angiotensin (0.89), plasminogen, dbp, insulin (0.85); infarction, low-density, losartan (0.84); hormonal, brachial, run-in, fixed-dose (0.83); lipoprotein, valsartan, endothelium-dependent, renin (0.82); pravastatin, hba1c (0.81); angiotensinogen, chd, smoking (0.80) |
| AE | Oesophagitis | 10.0 pct | 19 | 0.919 | Metastases (0.92); marrow (0.90); weekly (0.88); metastatic, antitumor (0.87); cancer (0.86); 3-year, toxicity, regimen, nadir, breast, metastasis (0.85); myeloma, survival, cancers, prostate (0.84); regimens, remission, cytotoxic, melphalan (0.83) |
| AE | Paralytic ileus | 59.4 pct | 12 | 0.961 | amitriptyline (0.96); tricyclic (0.95); antidepressants, anticholinergic, antidepressant (0.94); neuroleptics, chlorpromazine, depressive, overdose (0.93); tca, serotonin, diazepam (0.92); intoxication (0.91); thioridazine, clonidine (0.90); psychotropic, affective, psychological, antinociceptive (0.89); constipation (0.87) |
| AE | Thrombocytopenic purpura | 10.6 pct | 11 | 1 | infarction, ejection (0.95); intra-arterial (0.94); echocardiography (0.93); st-segment (0.93); echocardiographic (0.90); beta-blocking (0.89); beta-blocker (0.87); cardiology, diacetolol, bopindolol, bucindolol, beta-adrenoceptor-blocking, adp, beta-ars, beta1-selective, beta-adrenoblockers, cardioselectivity, atenolol, non-fatal (0.86) |
| MC | 5HT3 antagonists | 10.0 pct | 4 | 1 | 5-ht3ra, 5-ht3ra/dexamethasone, granisetron, ondansetron, 1966-september, 5-ht3, tropisetron, 5-hydroxytryptamine3, anti-emetic (1); dolasetron, emetic, ramosetron, 5-ht3-receptor, cinv, 5ht3, emetogenic, type-3, emesis, setrons, pov (0.999) |
| MC | Antibacterials (ear) | 14.7 pct | 4 | 0.98 | enrofloxacin, chloramphenicol, gentamycin (0.99); oxytetracycline, kanamycin, polymyxin, colistin, gentamicin, bacitracin, neomycin, povidone-iodine, fusidic, streptomycin, bacterial, septicemia, swabs (0.98); anaerobic, peru, tetracycline, aminoglycoside (0.97) |
*) Refers to the best area under the ROC curve achieved by the four machine learning algorithms as evaluated by stratified CVs. †) Refers to the AUCs used during the feature selection process. Abbreviations: AE: Adverse events; IK: index keywords; MC: minor drug classes; pct percentile. n(pos): number of positive examples in each drug characteristic dataset.