Artificial intelligence in dentistry: Harnessing big data to predict oral cancer survival

Table 1 Number of oral cancer cases from various anatomical sites

ICD-O-3 codes	Sites	Number of cases
C000	External upper lip	413
C001	External lower lip	2444
C002	External lip, NOS	92
C003	Mucosa of upper lip	104
C004	Mucosa of lower lip	567
C005	Mucosa of lip, NOS	29
C006	Commissure of lip	85
C008	Overlapping lesion of lip	46
C009	Lip, NOS (excludes skin of lip C44.0)	153
C019	Base of tongue, NOS	10840
C020	Dorsal surface of tongue, NOS	652
C021	Border of tongue	2632
C022	Ventral surface of tongue, NOS	1688
C023	Anterior 2/3 of tongue, NOS	2807
C024	Lingual tonsil	170
C028	Overlapping lesion of tongue	581
C029	Tongue, NOS	3050
C030	Upper gum	821
C031	Lower gum	1680
C039	Gum, NOS	210
C040	Anterior floor of mouth	1362
C041	Lateral floor of mouth	352
C048	Overlapping lesion of floor of mouth	136
C049	Floor of mouth, NOS	2284
C050	Hard palate	1155
C051	Soft palate, NOS (excludes nasopharyngeal surface of soft palate C11.3)	1301
C052	Uvula	180
C058	Overlapping lesion of palate	206
C059	Palate, NOS	154
C060	Cheek mucosa	1787
C061	Vestibule of mouth	134
C062	Retromolar area	1413
C068	Overlapping lesion of other and unspecified parts of mouth	142
C069	Mouth, NOS	487
C079	Parotid gland	7111
C080	Submandibular gland	1149
C081	Sublingual gland	94
C088	Overlapping lesion of major salivary glands	6
C089	Major salivary gland, NOS (excludes minor salivary gland, NOS C06.9)	287
C090	Tonsillar fossa	1735
C091	Tonsillar pillar	888
C098	Overlapping lesion of tonsil	109
C099	Tonsil, NOS (excludes lingual tonsil C02.4 and pharyngeal tonsil C11.1)	9521
C100	Vallecula	282
C101	Anterior surface of epiglottis	88
C102	Lateral wall of oropharynx	184
C103	Posterior wall of oropharynx	246
C104	Branchial cleft (site of neoplasm)	37
C108	Overlapping lesion of oropharynx	277
C109	Oropharynx, NOS	940
C129	Pyriform sinus	1707
C130	Postcricoid region	78
C131	Hypopharyngeal aspect of aryepiglottic fold, NOS (excludes laryngeal aspect of aryepiglottic fold C32.1)	214
C132	Posterior wall of hypopharynx	250
C138	Overlapping lesion of hypopharynx	113
C139	Hypopharynx, NOS	816
C739	Thyroid gland	111425

ICD-O-3 codes: International Classification of Diseases for Oncology, 3rd ed; NOS: Not otherwise specified.

Table 2 List of all 10 variables included in the final machine learning model building and validation

Variables	Variable description
Age at diagnosis	This data item represents the age of the patient at diagnosis for this cancer. The code is three digits and represents the patient’s actual age in years
Year of diagnosis	The year of diagnosis is the year the tumor was first diagnosed by a recognized medical practitioner, whether clinically or microscopically confirmed
Month of diagnosis	The month of diagnosis is the month the tumor was first diagnosed by a recognized medical practitioner, whether clinically or microscopically confirmed
Primary site	This data item identifies the site in which the primary tumor originated. See the International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3)[18] for topography codes. The decimal point is eliminated
CS tumor size	Information on tumor size. Available for 2004-2015 diagnosis years. Earlier cases may be converted and new codes added which weren't available for use prior to the current version of CS. For more information, see http://seer.cancer.gov/seerstat/variables/seer/ajcc-stage[19]
CS extension	Information on extension of the tumor. Available for 2004-2015 diagnosis years. Earlier cases may be converted and new codes added which weren't available for use prior to the current version of CS. For more information, see http://seer.cancer.gov/seerstat/variables/seer/ajcc-stage[19]
CS lymph nodes eval	Available for 2004-2015, but not required for the entire timeframe. Will be blank in cases not collected. For more information, see http://seer.cancer.gov/seerstat/variables/seer/ajcc-stage[19]
Derived AJCC stage group	This is the AJCC “Stage Group” component that is derived from CS detailed site-specific codes, using the CS algorithm, effective with 2004-2015 diagnosis years. See the CS site-specific schema for details (http://seer.cancer.gov/seerstat/variables/seer/ajcc-stage)[19]
RX Summ-surg prim site	Surgery of primary site describes a surgical procedure that removes and/or destroys tissue of the primary site performed as part of the initial work-up or first course of therapy
Site recode ICD-O-3/WHO 2008	A recode based on primary site and ICD-O-3 Histology in order to make analyses of site/histology groups easier. For example, the lymphomas are excluded from stomach and Kaposi and mesothelioma are separate categories based on histology. For more information, see http://seer.cancer.gov/siterecode/icdo3_dwhoheme/index.html[20]

CS: Coding System; AJCC: American Joint Committee on Cancer; ICD-O-3: International Classification of Diseases for Oncology, 3rd ed; WHO: World Health Organization.

Table 3 Demographic characteristics of the sample (n = 177714)

Variable	Mean	SD	Median	n	%
Survival months/mo	60.35	40.98	54.00
Age at diagnosis/yr	54.62	16.10	55.00
Tumor size/(ID, cm)	22.56	21.74	19.00
Marital status
Single				35688	20.08
Married				110480	62.17
Separated				1746	0.98
Divorced				16401	9.23
Widowed				13055	7.35
Unmarried or domestic partner				344	0.19
Sex
Male				72179	40.62
Female				105535	59.38
Race
White				148556	83.60
Black				16051	9.03
Other				13107	7.38

SD: Standard deviation; ID: Diameter.

Table 4 Machine learning model performance

Performance indicators	Linear regression	Decision tree	Random forest	XGBoost
MSE	647.49	538.30	489.58	486.55
RMSE	25.45	23.20	22.13	22.06
MAE	18.21	14.45	13.63	13.55
R2 score	0.620	0.681	0.709	0.711
Adjusted R2 score	0.620	0.681	0.709	0.711

XGBoost: Extreme gradient boosting; MSE: Mean squared error; RMSE: Root mean squared error; MAE: Mean absolute error.