Efficient risk-based collection of biospecimens in cohort studies: Designing a prospective study of diagnostic performance for multicancer detection tests.

In cohort studies, it can be infeasible to collect specimens on an entire cohort. For example, to estimate sensitivity of multiple Multi-Cancer Detection (MCD) assays, we desire an extra 80mL of cell-free DNA (cfDNA) blood, but this much extra blood is too expensive for us to collect on everyone. We propose a novel epidemiologic study design that efficiently oversamples those at highest baseline disease risk from whom to collect specimens, to increase the number of future cases with cfDNA blood collection. The variance reduction ratio from our risk-based subsample versus a simple random (sub)sample (SRS) depends primarily on the ratio of risk model sensitivity to the fraction of the cohort selected for specimen collection subject to constraining the risk model specificity. In a simulation where we chose 34% of Prostate, Lung, Colorectal, and Ovarian Screening Trial cohort at highest risk of lung cancer for cfDNA blood collection, we could enrich the number of lung cancers 2.42-fold and the standard deviation of lung-cancer MCD sensitivity was 31-33% reduced versus SRS. Risk-based collection of specimens on a subsample of the cohort could be a feasible and efficient approach to collecting extra specimens for molecular epidemiology.

The risk and benefit profiles of US eligible lung cancer screening attendees vs. non-attendees.

BACKGROUND: The United States Preventive Services Task Force (USPSTF) recommend lung-cancer screening for individuals aged 50-80 with ≥20 pack-years and ≤15 quit-years, but uptake is low. The risk and benefit profiles of screening attendees are unknown; consequently, the impact and lost opportunity of ongoing lung-cancer screening in the US remains unclear. METHODS: We estimated lung-cancer death risk (using the Lung Cancer Death Risk Assessment Tool) and life gained from screening (using the LYFS-CT model) for individuals 50-79 who ever-smoked in the US-representative 2022 Behavioral Risk Factor Surveillance System. We compared lung-cancer death risk and life-gained among USPSTF-eligible individuals by screening status (self-reported screened vs not screened in past year), and estimated the number of lung-cancer deaths averted and life-years gained under current screening levels and if everyone eligible was screened. RESULTS: USPSTF-eligibility was 33.7% (95%CI:33.1-34.4%), of whom 17.9% (95%CI : 17.0-18.8%) self-reported screening. Screening uptake increased with increasing lung-cancer death risk quintile (Q1 = 5.2% (95%CI : 3.0%-8.8%); Q5 = 21.8% (95%CI : 20.3%-23.3%)) and life-gain from screening quintile (Q1 = 6.2% (95%CI : 3.8%-9.9%); Q5 = 20.8% (95%CI : 19.5%-22.2%)). Screened individuals had higher lung-cancer death risk (Risk Ratio [RR]=1.35, 95%CI : 1.26-1.46) and life-years gained (RR = 1.19, 95%CI : 1.12-1.25) than unscreened individuals. Currently screening averts 19,306 lung-cancer deaths and gains 237,564 life-years; screening everyone eligible would additionally avert 56,956 lung-cancer deaths and gain 751,850 life-years. Two-thirds of USPSTF-lung-eligible women were up-to-date with breast-cancer screening, but only 17.3% attended lung screening in the past year. CONCLUSIONS: Eligible screening attendees had higher lung-cancer death risk and benefit from screening. Higher rates of screening could substantially increase the number of lung-cancer deaths prevented.

Cleaning and Harmonizing Medical Image Data for Reliable AI: Lessons Learned from Longitudinal Oral Cancer Natural History Study Data.

For deep learning-based machine learning, not only are large and sufficiently diverse data crucial but their good qualities are equally important. However, in real-world applications, it is very common that raw source data may contain incorrect, noisy, inconsistent, improperly formatted and sometimes missing elements, particularly, when the datasets are large and sourced from many sites. In this paper, we present our work towards preparing and making image data ready for the development of AI-driven approaches for studying various aspects of the natural history of oral cancer. Specifically, we focus on two aspects: 1) cleaning the image data; and 2) extracting the annotation information. Data cleaning includes removing duplicates, identifying missing data, correcting errors, standardizing data sets, and removing personal sensitive information, toward combining data sourced from different study sites. These steps are often collectively referred to as data harmonization. Annotation information extraction includes identifying crucial or valuable texts that are manually entered by clinical providers related to the image paths/names and standardizing of the texts of labels. Both are important for the successful deep learning algorithm development and data analyses. Specifically, we provide details on the data under consideration, describe the challenges and issues we observed that motivated our work, present specific approaches and methods that we used to clean and standardize the image data and extract labelling information. Further, we discuss the ways to increase efficiency of the process and the lessons learned. Research ideas on automating the process with ML-driven techniques are also presented and discussed. Our intent in reporting and discussing such work in detail is to help provide insights in automating or, minimally, increasing the efficiency of these critical yet often under-reported processes.

Population-level incidence of human papillomavirus-positive oropharyngeal, cervical, and anal cancers, by smoking status.

We estimated the population-level incidence of human papillomavirus (HPV)-positive oropharyngeal, cervical, and anal cancers by smoking status. We combined HPV DNA genotyping data from the Centers for Disease Control and Prevention's Cancer Registry Sentinel Surveillance System with data from the Kentucky Cancer Registry and Behavioral Risk Factor Surveillance System across smoking status. During 2004-2005 and 2014-2015 in Kentucky, most cases of oropharyngeal (63.3%), anal (59.7%), and cervical (54.9%) cancer were among individuals who ever smoked. The population-level incidence rate was higher among individuals who ever smoked than among those who never smoked for HPV-positive oropharyngeal (7.8 vs 2.1; adjusted incidence rate ratio = 2.6), cervical (13.7 vs 6.8; adjusted incidence rate ratio = 2.0), and anal (3.9 vs 1.6; adjusted incidence rate ratio = 2.5) cancers. These findings indicate that smoking is associated with increased risk of HPV-positive oropharyngeal, cervical, and anal cancers, and the population-level burden of these cancers is higher among individuals who ever smoked.

Absolute lung cancer risk increases among individuals with >15 quit-years: Analyses to inform the update of the American Cancer Society lung cancer screening guidelines.

BACKGROUND: This report quantifies counteracting effects of quit-years and concomitant aging on lung cancer risk, especially on exceeding 15 quit-years, when the US Preventive Services Task Force (USPSTF) recommends curtailing lung-cancer screening. METHODS: Cox models were fitted to estimate absolute lung cancer risk among Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO) and National Lung Screening Trial (NLST) participants who ever smoked. Absolute lung cancer risk and gainable years of life from screening for individuals aged 50 to 80 in the US-representative National Health Interview Survey (NHIS) 2015-2018 who ever smoked were projected. Relaxing USPSTF recommendations to 20/25/30 quit-years versus augmenting USPSTF criteria with individuals whose estimated gain in life expectancy from screening exceeded 16.2 days according to the Life Years From Screening-CT (LYFS-CT) prediction model was compared. RESULTS: Absolute lung cancer risk increased by 8.7%/year (95% CI, 7.7%-9.7%; p < .001) as individuals aged beyond 15 quit-years in the PLCO, with similar results in NHIS and NLST. For example, mean 5-year lung cancer risk for those aged 65 years with 15 quit-years = 1.47% (95% CI, 1.35%-1.59%) versus 1.76% (95% CI, 1.62%-1.90%) for those aged 70 years with 20 quit-years in the PLCO. Removing the quit-year criterion would make 4.9 million more people eligible and increase the proportion of preventable lung cancer deaths prevented (sensitivity) from 63.7% to 74.2%. Alternatively, augmentation using LYFS-CT would make 1.7 million more people eligible while increasing the lung cancer death sensitivity to 74.0%. CONCLUSIONS: Because of aging, absolute lung cancer risk increases beyond 15 quit-years, which does not support exemption from screening or curtailing screening once it has been initiated. Compared with relaxing the USPSTF quit-year criterion, augmentation using LYFS-CT could prevent most of the deaths at substantially superior efficiency, while also preventing deaths among individuals who currently smoke with low intensity or long duration.

Inflammatory Tongue Conditions and Risk of Oral Tongue Cancer Among the US Elderly Individuals.

PURPOSE: The incidence of oral tongue cancers has increased since the 1980s among US men and women for unknown reasons. We investigated associations of inflammatory tongue conditions with risk of cancers of the oral tongue, other oral cavity, and oropharynx among the US elderly individuals (age 65 years or older). METHODS: We conducted a case-control study (2,534 oral tongue cancers, 6,832 other oral cavity cancers, 9,373 oropharyngeal cancers, and 200,000 controls) within the SEER-Medicare data set (1992-2013). Medicare records were used to identify patients with clinically diagnosed inflammatory tongue conditions (glossitis, benign migratory glossitis, median rhomboid glossitis, atrophic glossitis, glossodynia, other specified conditions [eg, atrophy and hypertrophy], and other unspecified conditions) and oral precancer (leukoplakia/erythroplakia). Only conditions preceding cancer/control selection by >12 months were included. RESULTS: The prevalence of inflammatory tongue conditions was significantly higher in patients with tongue cancer than controls (6.0% v 0.6%; odds ratios [ORs], adjusted for age, sex, race, Medicare utilization, and precancer, 5.8 [95% CI, 4.7 to 7.2]). This overall association primarily arose from glossitis, 5.6 (95% CI, 4.4 to 7.2); other specified conditions, 9.1 (95% CI, 5.5 to 15.2); and other unspecified conditions, 13.7 (95% CI, 8.0 to 23.7). These associations remained strongly elevated >5 years preceding tongue cancer (arguing against reverse causation), for conditions diagnosed by a specialist (arguing against misclassification), and among patients who received an oral biopsy (arguing against missed cancer). During 2013, an estimated 1 in 11 patients with oral tongue cancer had a preceding diagnosis of inflammatory tongue conditions. Associations of inflammatory tongue conditions were relatively weak for other oral cavity cancers (ORs, 1.8 [95% CI, 1.5 to 2.3]) and oropharyngeal cancer (OR, 1.3 [95% CI, 1.0 to 1.6]) and were observed only closest to cancer diagnosis. CONCLUSION: Inflammatory tongue conditions were associated with strongly increased risks of oral tongue cancers and preceded cancer diagnosis by several years, underscoring the need for increased clinical surveillance among patients with such apparently benign diagnoses.

Methods for Using Race and Ethnicity in Prediction Models for Lung Cancer Screening Eligibility.

Importance: Using race and ethnicity in clinical prediction models can reduce or inadvertently increase racial and ethnic disparities in medical decisions. Objective: To compare eligibility for lung cancer screening in a contemporary representative US population by refitting the life-years gained from screening-computed tomography (LYFS-CT) model to exclude race and ethnicity vs a counterfactual eligibility approach that recalculates life expectancy for racial and ethnic minority individuals using the same covariates but substitutes White race and uses the higher predicted life expectancy, ensuring that historically underserved groups are not penalized. Design, Setting, and Participants: The 2 submodels composing LYFS-CT NoRace were refit and externally validated without race and ethnicity: the lung cancer death submodel in participants of a large clinical trial (recruited 1993-2001; followed up until December 31, 2009) who ever smoked (n = 39 180) and the all-cause mortality submodel in the National Health Interview Survey (NHIS) 1997-2001 participants aged 40 to 80 years who ever smoked (n = 74 842, followed up until December 31, 2006). Screening eligibility was examined in NHIS 2015-2018 participants aged 50 to 80 years who ever smoked. Data were analyzed from June 2021 to September 2022. Exposure: Including and removing race and ethnicity (African American, Asian American, Hispanic American, White) in each LYFS-CT submodel. Main Outcomes and Measures: By race and ethnicity: calibration of the LYFS-CT NoRace model and the counterfactual approach (ratio of expected to observed [E/O] outcomes), US individuals eligible for screening, predicted days of life gained from screening by LYFS-CT. Results: The NHIS 2015-2018 included 25 601 individuals aged 50 to 80 years who ever smoked (2769 African American, 649 Asian American, 1855 Hispanic American, and 20 328 White individuals). Removing race and ethnicity from the submodels underestimated lung cancer death risk (expected/observed [E/O], 0.72; 95% CI, 0.52-1.00) and all-cause mortality (E/O, 0.90; 95% CI, 0.86-0.94) in African American individuals. It also overestimated mortality in Hispanic American (E/O, 1.08, 95% CI, 1.00-1.16) and Asian American individuals (E/O, 1.14, 95% CI, 1.01-1.30). Consequently, the LYFS-CT NoRace model increased Hispanic American and Asian American eligibility by 108% and 73%, respectively, while reducing African American eligibility by 39%. Using LYFS-CT with the counterfactual all-cause mortality model better maintained calibration across groups and increased African American eligibility by 13% without reducing eligibility for Hispanic American and Asian American individuals. Conclusions and Relevance: In this study, removing race and ethnicity miscalibrated LYFS-CT submodels and substantially reduced African American eligibility for lung cancer screening. Under counterfactual eligibility, no one became ineligible, and African American eligibility increased, demonstrating the potential for maintaining model accuracy while reducing disparities.

Global epidemiologic patterns of oropharyngeal cancer incidence trends.

BACKGROUND: The emergence of human papillomavirus (HPV)-positive oropharyngeal cancer and evolving tobacco use patterns have changed the landscape of head and neck cancer epidemiology internationally. We investigated updated trends in oropharyngeal cancer incidence worldwide. METHODS: We analyzed cancer incidence data between 1993 and 2012 from 42 countries using the Cancer Incidence in Five Continents database volumes V through XI. Trends in oropharyngeal cancer incidence were compared with oral cavity cancers and lung squamous cell carcinomas using log-linear regression and age period-cohort modeling. RESULTS: In total, 156 567 oropharyngeal cancer, 146 693 oral cavity cancer, and 621 947 lung squamous cell carcinoma patients were included. Oropharyngeal cancer incidence increased (P < .05) in 19 and 23 countries in men and women, respectively. In countries with increasing male oropharyngeal cancer incidence, all but 1 had statistically significant decreases in lung squamous cell carcinoma incidence, and all but 2 had decreasing or nonsignificant net drifts for oral cavity cancer. Increased oropharyngeal cancer incidence was observed both in middle-aged (40-59 years) and older (≥60 years) male cohorts, with strong nonlinear birth cohort effects. In 20 countries where oropharyngeal cancer incidence increased for women and age period-cohort analysis was possible, 13 had negative or nonsignificant lung squamous cell carcinoma net drifts, including 4 countries with higher oropharyngeal cancer net drifts vs both lung squamous cell carcinoma and oral cavity cancer (P < .05 for all comparisons). CONCLUSIONS: Increasing oropharyngeal cancer incidence is seen among an expanding array of countries worldwide. In men, increased oropharyngeal cancer is extending to older age groups, likely driven by human papillomavirus-related birth cohort effects. In women, more diverse patterns were observed, suggesting a complex interplay of risks factors varying by country, including several countries where female oropharyngeal cancer increases may be driven by HPV.

Human Papillomavirus and Rising Oropharyngeal Cancer Incidence in the United States.

PURPOSE: Recent increases in incidence and survival of oropharyngeal cancers in the United States have been attributed to human papillomavirus (HPV) infection, but empirical evidence is lacking. PATIENTS AND METHODS: HPV status was determined for all 271 oropharyngeal cancers (1984-2004) collected by the three population-based cancer registries in the Surveillance, Epidemiology, and End Results (SEER) Residual Tissue Repositories Program by using polymerase chain reaction and genotyping (Inno-LiPA), HPV16 viral load, and HPV16 mRNA expression. Trends in HPV prevalence across four calendar periods were estimated by using logistic regression. Observed HPV prevalence was reweighted to all oropharyngeal cancers within the cancer registries to account for nonrandom selection and to calculate incidence trends. Survival of HPV-positive and HPV-negative patients was compared by using Kaplan-Meier and multivariable Cox regression analyses. RESULTS: HPV prevalence in oropharyngeal cancers significantly increased over calendar time regardless of HPV detection assay (P trend < .05). For example, HPV prevalence by Inno-LiPA increased from 16.3% during 1984 to 1989 to 71.7% during 2000 to 2004. Median survival was significantly longer for HPV-positive than for HPV-negative patients (131 v 20 months; log-rank P < .001; adjusted hazard ratio, 0.31; 95% CI, 0.21 to 0.46). Survival significantly increased across calendar periods for HPV-positive (P = .003) but not for HPV-negative patients (P = .18). Population-level incidence of HPV-positive oropharyngeal cancers increased by 225% (95% CI, 208% to 242%) from 1988 to 2004 (from 0.8 per 100,000 to 2.6 per 100,000), and incidence for HPV-negative cancers declined by 50% (95% CI, 47% to 53%; from 2.0 per 100,000 to 1.0 per 100,000). If recent incidence trends continue, the annual number of HPV-positive oropharyngeal cancers is expected to surpass the annual number of cervical cancers by the year 2020. CONCLUSION: Increases in the population-level incidence and survival of oropharyngeal cancers in the United States since 1984 are caused by HPV infection.

Upper age limits for US male human papillomavirus vaccination for oropharyngeal cancer prevention: a microsimulation-based modeling study.

BACKGROUND: Human papillomavirus (HVP)-positive oropharyngeal cancer is the most common HPV-associated cancer in the United States. The age at acquisition of oral HPV infections that cause oropharyngeal cancer (causal infections) is unknown; consequently, the benefit of vaccination of US men aged 27-45 years remains uncertain. METHODS: We developed a microsimulation-based, individual-level, state-transition model of oral HPV16 and HPV16-positive oropharyngeal cancer among heterosexual US men aged 15-84 years, calibrated to population-level data. We estimated the benefit of vaccination of men aged 27-45 years for prevention of oropharyngeal cancer, accounting for direct- and indirect effects (ie, herd effects) of male and female vaccination. RESULTS: In the absence of vaccination, most (70%) causal oral HPV16 infections are acquired by age 26 years, and 29% are acquired between ages 27 and 45 years. Among men aged 15-45 years in 2021 (1976-2006 birth cohorts), status quo vaccination of men through age 26 years is estimated to prevent 95% of 153 450 vaccine-preventable cancers. Assuming 100% vaccination in 2021, extending the upper age limit to 30, 35, 40, or 45 years for men aged 27-45 years (1976-1994 cohorts) is estimated to yield small benefits (3.0%, 4.2%, 5.1%, and 5.6% additional cancers prevented, respectively). Importantly, status quo vaccination of men through age 26 years is predicted to result in notable declines in HPV16-positive oropharyngeal cancer incidence in young men by 2035 (51% and 24% declines at ages 40-44 years and 45-49 years, respectively) and noticeable declines (12%) overall by 2045. CONCLUSION: Most causal oral HPV16 infections in US men are acquired by age 26 years, underscoring limited benefit from vaccination of men aged 27-45 years for prevention of HPV16-positive oropharyngeal cancers.

Exclusive use of different types of tobacco products, exposure to secondhand tobacco smoke and risk of subtypes of head and neck cancer among Indian males.

Tobacco products are used in vary many forms in India. Although the risk of tobacco uses in developing head and neck cancer (HNC) is known, risk by exclusive use of different tobacco products on HNC and its subtypes is poorly understood. A case-control study was conducted at a tertiary cancer hospital, which receives cases from different geographical regions of India with use of different types of tobacco products. The study included 824 oral cavity (OC), 149 oropharynx (OPX) 104 hypopharyngeal (HPX) and 81 larynx (LX) cancer cases and 1206 visitor controls. Information on 11 different types of tobacco products and exposure to secondhand smoke was collected through structured questionnaires. Odds ratios (OR) and 95% confidence intervals (CI), for the association of various HNC subtypes with exclusive use of each tobacco product compared to nonusers of tobacco were estimated using logistic regression models, after adjusting for potential confounders. Exclusive use of any type of smokeless tobacco product was strongly associated with all subtypes of HNC. Gutka chewing (only) had highest risk (OR = 33.67; 95% CI = 19.8-57.0) while exclusive users of betel quid with tobacco (BQ + T), tobacco quid, Khaini, Mawa and Mishri users had a OR of 14.77, 24.20, 5.33, 2.96 and 3.32, respectively, for development of OC. Bidi smoking and secondhand smoke was independently associated with increased risk of HNC. Our study indicates that tobacco control policies should focus on product specific awareness messaging that switching between tobacco product types is not a safe alternative to complete cessation.

Variation in Human Papillomavirus Vaccination Effectiveness in the US by Age at Vaccination.

This survey study assesses the status and timing of HPV vaccination as self-reported by female participants in the National Health and Nutrition Examination Survey from 2011 to 2018.

The Current Burden of Oropharyngeal Cancer: A Global Assessment Based on GLOBOCAN 2020.

BACKGROUND: Oropharyngeal cancer (OPC) is a complex disease whose etiologies, either related to risk factors such as smoking or alcohol, or linked to HPV infection, are believed to be responsible for wide gender and geographical variability. This study depicts the current burden of OPC worldwide. METHODS: Estimated OPC new cases, deaths, age-standardized rates (ASR) for both incidence and mortality in 2020 were obtained from the GLOBOCAN database for each country and across 20 UN-defined world regions by sex. The incidence-to-mortality ratio (IMR) was also estimated from ASR. RESULTS: Worldwide, 98,400 new cases and 48,100 OPC deaths were estimated in 2020, with ASR of 1.1 and 0.51 per 100,000 for incidence and mortality, respectively. ASR for both incidence and mortality were approximately four times higher in men and varied greatly across geographical regions and countries within the same region. Higher incidence was estimated in Europe, North-America, Australia, and New Zealand. Mortality was the highest in Central-East Europe, Western Europe, Melanesia, South-Central Asia, and the Caribbean. South-Central Asia, most African areas, and Central America exhibited the lowest IMR values, whereas North-America, Australia, New Zealand, and North-Europe had the highest. CONCLUSIONS: The marked geographical and gender variability in OPC incidence and mortality is likely to reflect the distribution of risk factors and the diverse prevalence of HPV-negative and HPV-positive cases. IMPACT: Findings are likely to drive future research, support the development of targeted strategies to counteract disease burden, establish priorities for prevention and treatment programs, and address inequality in access to services.

Automatic Detection of Oral Lesion Measurement Ruler Toward Computer-Aided Image-Based Oral Cancer Screening.

Intelligent computer-aided algorithms analyzing photographs of various mouth regions can help in reducing the high subjectivity in human assessment of oral lesions. Very often, in the images, a ruler is placed near a suspected lesion to indicate its location and as a physical size reference. In this paper, we compared two deep-learning networks: ResNeSt and ViT, to automatically identify ruler images. Even though the ImageN et 1K dataset contains a "ruler" class label, the pre-trained models showed low sensitivity. After fine-tuning with our data, the two networks achieved high performance on our test set as well as a hold-out test set from a different provider. Heatmaps generated using three saliency methods: GradCam and XRAI for ResNeSt model, and Attention Rollout for ViT model, demonstrate the effectiveness of our technique. Clinical Relevance- This is a pre-processing step in automated visual evaluation for oral cancer screening.

Sex disparities in the incidence of 21 cancer types: Quantification of the contribution of risk factors.

BACKGROUND: Cancer incidence is higher in men than in women at most shared anatomic sites for currently unknown reasons. The authors quantified the extent to which behaviors (smoking and alcohol use), anthropometrics (body mass index and height), lifestyles (physical activity, diet, medications), and medical history collectively explain the male predominance of risk at 21 shared cancer sites. METHODS: Prospective cohort analyses (n = 171,274 male and n = 122,826 female participants; age range, 50-71 years) in the National Institutes of Health-AARP Diet and Health Study (1995-2011). Cancer-specific Cox regression models were used to estimate male-to-female hazard ratios (HRs). The degree to which risk factors explained the observed male-female risk disparity was quantified using the Peters-Belson method. RESULTS: There were 26,693 incident cancers (17,951 in men and 8742 in women). Incidence was significantly lower in men than in women only for thyroid and gallbladder cancers. At most other anatomic sites, the risks were higher in men than in women (adjusted HR range, 1.3-10.8), with the strongest increases for bladder cancer (HR, 3.33; 95% confidence interval [CI], 2.93-3.79), gastric cardia cancer (HR, 3.49; 95% CI, 2.26-5.37), larynx cancer (HR, 3.53; 95% CI, 2.46-5.06), and esophageal adenocarcinoma (HR, 10.80; 95% CI, 7.33-15.90). Risk factors explained a statistically significant (nonzero) proportion of the observed male excess for esophageal adenocarcinoma and cancers of liver, other biliary tract, bladder, skin, colon, rectum, and lung. However, only a modest proportion of the male excess was explained by risk factors (ranging from 50% for lung cancer to 11% for esophageal adenocarcinoma). CONCLUSIONS: Men have a higher risk of cancer than women at most shared anatomic sites. Such male predominance is largely unexplained by risk factors, underscoring a role for sex-related biologic factors.

The Evolving Epidemiology of Oral Cavity and Oropharyngeal Cancers.

In 1988, Blot and colleagues reported results from a U.S. case-control study of oral cavity or pharyngeal (oropharyngeal and hypopharyngeal) cancers, with results showing independent associations of smoking and alcohol with increased risk, multiplicative interaction effects between smoking and alcohol, and that nearly three-quarters of these cancers are attributable to smoking and alcohol. The report by Blot and colleagues represents a landmark in oropharyngeal cancer epidemiology. This study, the largest at the time, introduced several novel concepts in oropharyngeal cancer epidemiology that remain relevant today-etiologic heterogeneity, statistical interaction effects, adjusted attributable fractions, and disparities by sex and race/ethnicity. Perhaps the most significant recognition in the field since 1988 is the etiologic association of human papillomavirus (HPV, primarily HPV16) with cancers arising in the oropharynx. Today, more than 80% of oropharyngeal cancers in the United States are caused by HPV while only approximately 3% of oral cavity cancers are caused by HPV. This etiologic heterogeneity across head and cancer subsites revealed by HPV is manifest at the genetic/genomic, epidemiologic, and clinical levels. Tobacco and alcohol remain the major etiologic factors for oral cavity cancers while HPV is the major cause of oropharyngeal cancers. Thus, tobacco and alcohol control and prophylactic HPV vaccination remain the most promising prevention tools for oral cavity and oropharyngeal cancers at this time. Importantly, the ever-emerging alternative tobacco products, such as smokeless tobacco/snus, hookah and water pipes, e-cigarettes, flavored cigars and cigarillos, and oral dissolvable products, represent a key public health concern and the carcinogenic effects of these products remains an active area of investigation. See related article by Blot and colleagues, Cancer Res 1988;48:3282-7.

Oral Cavity Anatomical Site Image Classification and Analysis.

Oral cavity cancer is a common cancer that can result in breathing, swallowing, drinking, eating problems as well as speech impairment, and there is high mortality for the advanced stage. Its diagnosis is confirmed through histopathology. It is of critical importance to determine the need for biopsy and identify the correct location. Deep learning has demonstrated great promise/success in several image-based medical screening/diagnostic applications. However, automated visual evaluation of oral cavity lesions has received limited attention in the literature. Since the disease can occur in different parts of the oral cavity, a first step is to identify the images of different anatomical sites. We automatically generate labels for six sites which will help in lesion detection in a subsequent analytical module. We apply a recently proposed network called ResNeSt that incorporates channel-wise attention with multi-path representation and demonstrate high performance on the test set. The average F1-score for all classes and accuracy are both 0.96. Moreover, we provide a detailed discussion on class activation maps obtained from both correct and incorrect predictions to analyze algorithm behavior. The highlighted regions in the class activation maps generally correlate considerably well with the region of interest perceived and expected by expert human observers. The insights and knowledge gained from the analysis are helpful in not only algorithm improvement, but also aiding the development of the other key components in the process of computer assisted oral cancer screening.