Table 2.
Selection of the best available NTCP models per symptom category in head and neck cancer patients.
|
Author |
Patients, n, (OARs)/events |
Advanced RT techniques only |
Endpoint ≥ 12 m post-RT |
Model |
Variables included in final model |
Prospective toxicity collection |
Prediction probability |
Validation |
Others |
Overall quality |
| Xerostomia | ||||||||||
| Beetz 2012 [20] | 17/83 | Yes (IMRT) | No (6 m) | Logistic regression with bootstrapping | Xerostomia: Dmean PG contra, BL xerostomia Sticky saliva: Dmean SMG contra, Dmean SLG both, Dmean soft palate | Yes | AUC 0.68 (xerostomia), 0.70 (sticky saliva) | Yes (internal) | Good | |
| Lee 2014 [32] | 158/52 | Yes (IMRT) | Yes (3 m and 12 m) | LASSO logistic regression with bootstrapping | 12 m: Dmean PG ipsi/contra | Yes | AUC 12 m: 0.98 (HNSCC), 0.96 (NPC) | Yes (internal) | Good | |
| van Dijk 2017 [33] | 249/63 (sticky saliva)/100 (xerostomia) | No (IMRT, 3DCRT, VMAT) | Yes (1 y) | LASSO multivariate logistic regression | Xerostomia: Dmean PG contra, BL xerostomia, SRE GLRLM PG contra; Sticky saliva: Dmean SMG (both), BL sticky saliva, max HU both SMG | Yes | AUC xerostomia ± imaging biomarker 0.77 and 0.75, sticky saliva ± imaging biomarker 0.77 and 0.74 | Yes (internal) | Limited added value of imaging biomarkers | Good |
| Dysphagia | ||||||||||
| Christianen 2012 [19] | 354/NA | No (3DRT, IMRT) | No (6 m) | Logistic regression | Dmean SPC, Dmean supraglottic larynx | Yes | AUC 0.80 | Yes (external) | Good | |
| Feeding tube dependency | ||||||||||
| Wopken 2014 [31] | 355/38 | Yes (IMRT) | No (6 m), but clinically relevant | Logistic regression | T stage, weight loss, accelerated RT, chemo, Cetuximab, Dmean inf. PCM, Dmean PG contra, Dmean cricopharyngeal muscle | Yes | AUC 0,88 | Yes (internal) | Good | |
| Dysgeusia | ||||||||||
| Sapir 2016 [34] | 73/26 | Yes (IMRT) | No (3 m) | LKB | Dmean oral cavity | Yes | NA | No | Oral cavity as OAR | Fair |
| Esophageal stricture | ||||||||||
| Mavroidis 2003 [35] | 82/26 | No (3DCRT) | No 1–40 m (median 7 m) | LKB | Dmean esophagus | No | ROC = 0.84, X2 test = 0.95 | No | Poor | |
| Brain necrosis | ||||||||||
| Wang 2019 [36] | 749/38 | Yes (IMRT) | No (3.5–75 m, median 49 m) | Lasso binary regression | D0.5 cc and D10 selected for final model | No | AUC 0.68 (testing set) | Yes (internal) | Good | |
| Zeng 2015 [37] | 351/29 | Yes (IMRT) | No (6–100 m, median 76 m) | Logistic regression | D1cc | No | NA | No | Fair | |
| Nerve palsy | ||||||||||
| Chow 2019 [38] | 330 nerves/46 | Yes (IMRT) | No (min. FU 6 m, median 8.1 y) | Logistic regression | D1cc | No | AUC 0.83 | No | False high rate of palsies: min. FU all patients 6 m/healthy control 8 y | Fair |
| Trismus | ||||||||||
| Morimoto 2019 [39] | 132/30 | No (3DCRT, IMRT (percentage unclear) | No (6 m) | Logistic regression | Dmean TMJ contra, max. intercisial opening at BL | Yes | P = .182 (Hosmer and Lemeshow test) | No | Collinearity check; acc. fx in 95 patients | Good |
| Lindblom 2014 [40] | 121/50 | No (3DCRT, IMRT) | Yes for MID (21–127 m, median 66 m), unclear for QoL scores | Logistic regression | Different models with different variables studies; best fit for ipsi masseter for both endpoints | No (MID), yes (QoL) | 0.77 and 0.73 for model with endpoint MID and QoL, respectively and ipsi masseter as variable | Yes (internal) | 70 patients with acc. fx | Good |
| Hypothyroidism | ||||||||||
| Rønjom 2013 [41] | 203/35 | Yes (IMRT) | Yes (1 y, 2 y) | Logistic regression | Dmean thyroid gland, thyroid gland volume | No (but objective criterion) | NA | No | Good | |
| Hearing loss | ||||||||||
| Marzi 2015 [42] | 280 ears/73 | Yes (PT) | Yes (median FU 26 m) | LKB | Dmean inner ear | Yes | AUC 0.86 | No | Fair | |
| Tinnitus | ||||||||||
| Lee 2015 [43] | 422 ears/49 | Yes (IMRT) | Yes (51 m, range 36– 77 m) | LKB Logistic regression | Dmean cochlea ipsi | No | LKB: 0.76, Logistic: 0.76 | No | Influence of chemotherapy not accounted for | Fair |
| Dry eye | ||||||||||
| Bhandare 2012 [44] | 78/40 | No (EBRT 1996–2000) | No (mean 0.9 y) | Logistic regression with bootstrapping | Dmax lacrimal gland | No | NA | No | Fair | |
| Laryngeal edema | ||||||||||
| Rancati 2009 [45] | 48/25 | Yes (IMRT) | Yes (15 m) | Lyman Logit | Dmean larynx | No (but objective criterion) | NA | No | Fair | |
| Hypopituitarism | ||||||||||
| Marzi 2015 [42] | 103/45 | Yes (PT) | Yes (median 26 m) | LKB | Dmean pituitary gland | Yes | AUC 0.86 | No | Fair |
Abbreviations: NTCP, normal tissue complication probability; OAR, organ at risk; RT, radiotherapy; prosp., prospective; IMRT, intensity modulated radiation therapy; Dmean, mean dose; PG, parotid gland; Contra, contralateral; BL, baseline; SMG, submandibular gland; SLG, sublingual gland; AUC, area under curve; ipsi, ipsilateral; HNSCC, head neck squamous cell cancer; NPC, nasopharyngeal cancer; 3DCRT, 3-dimensional conformal radiotherapy; VMAT, volumetric modulated arc therapy; SRE GLRLM, short run emphasis gray level co-occurrence; HU, Hounsfield unit; NA, not assessed; PCM, Pharyngeal Constrictor Muscle; LKB, Lyman–Kutcher–Burman; ROC, receiver operator curve; min, minimum; FU, follow-up; TMJ, temporomandibular joint; acc. fx, accelerated fractionation; MID, maximal interincisal distance; QoL, quality of life; PT, proton therapy; EBRT, external beam radiotherapy.