Unsupervised Hierarchical Clustering of Head and Neck Cancer Patients by Pre-Treatment Plasma Metabolomics Creates Prognostic Metabolic Subtypes

There is growing evidence that the metabolism is deeply intertwined with head and neck squamous cell carcinoma (HNSCC) progression and survival but little is known about circulating metabolite patterns and their clinical potential. We performed unsupervised hierarchical clustering of 209 HNSCC patients via pre-treatment plasma metabolomics to identify metabolic subtypes. We annotated the subtypes via pathway enrichment analysis and investigated their association with overall and progression-free survival. We stratified the survival analyses by smoking history. High-resolution metabolomics extracted 186 laboratory-confirmed metabolites. The optimal model created two patient clusters, of subtypes A and B, corresponding to 41% and 59% of the study population, respectively. Fatty acid biosynthesis, acetyl-CoA transport, arginine and proline, as well as the galactose metabolism pathways differentiated the subtypes. Relative to subtype B, subtype A patients experienced significantly worse overall and progression-free survival but only among ever-smokers. The estimated three-year overall survival was 61% for subtype A and 86% for subtype B; log-rank p = 0.001. The association with survival was independent of HPV status and other HNSCC risk factors (adjusted hazard ratio = 3.58, 95% CI: 1.46, 8.78). Our findings suggest that a non-invasive metabolomic biomarker would add crucial information to clinical risk stratification and raise translational research questions about testing such a biomarker in clinical trials.

Outcomes Stratification of Head and Neck Cancer Using Pre- and Post-treatment DNA Methylation From Peripheral Blood

PURPOSE

Established prognostic factors for head and neck squamous cell carcinoma (HNSCC) mostly consist of clinical and tumor features assessed before treatment. We report a novel application of DNA methylation in peripheral blood before and after radiation therapy to further improve outcomes stratification.

METHODS AND MATERIALS

Peripheral blood samples from patients with nonmetastatic HNSCC were obtained for methylation analysis 1 week before and 1 month after radiation therapy. Patients were randomized 1:1 to a Discovery Cohort or a Validation Cohort. In the Discovery Cohort, associations between genome-wide methylation change (posttreatment minus pretreatment) and recurrence-free survival (RFS) as well as overall survival (OS) were evaluated using Cox regression. A methylation risk score (MRS) was then constructed from methylation levels at the top associated sites, filtered for residing within the regulatory regions of genes expressed in cells of hematopoietic lineage. The prognostic value of MRS was separately assessed in the Discovery and Validation Cohorts.

RESULTS

Between December 2013 and September 2018, 115 patients participated in this study. Human papilloma virus negative status, oral cavity cancer, gastrostomy tube insertion, and higher neutrophil count before radiation therapy were associated with shorter RFS and OS (P < .05). Genes downstream of the methylation sites comprising MRS are HIF1A, SF1, LGALS9, and FUT5, involved in hypoxia response, blood cell maturation, and immune modulation. High MRS (in the top third) was significantly associated with worse RFS (hazard ratio [HR], 7.1; 95% confidence interval [CI], 1.4-35.5; P = .016) and OS (HR, 15.9; 95% CI, 1.6-153.6; P = .017) in the Discovery Cohort, independent of the aforementioned risk factors. These findings were replicated in the Validation Cohort, for which high MRS also independently predicted worse RFS (HR, 10.2; 95%, CI 2.4-43.4; P = .002) and OS (HR, 3.7; 95% CI, 1.3-10.4; P = .015).

CONCLUSIONS

We successfully trained and validated a signature of DNA methylation in peripheral blood before and after radiation therapy that stratified outcomes among patients with HNSCC, implicating the potential for genomics-tailored surveillance and consolidation treatment.

Metabolic Pathways Associated With Psychoneurological Symptoms in Children With Cancer Receiving Chemotherapy

CONTEXT

Children with cancer undergoing chemotherapy experience a cluster of psychoneurological symptoms (PNS), including pain, fatigue, anxiety, and depressive symptoms. Metabolomics is promising to differentiate metabolic pathways associated with the PNS cluster.

OBJECTIVES

Identify metabolic pathways associated with the PNS cluster in children with cancer before and after chemotherapy.

METHODS

Pain, fatigue, anxiety, and depressive symptoms were assessed using the Pediatric PROMIS scales. T-scores were computed and divided dichotomously by a cutoff point of 50; the PNS cluster was a sum of the four symptoms ranging from 0 (all T-scores <50) to 4 (all T-scores ≥50). Serum metabolites were processed using liquid chromatography mass-spectrometry untargeted metabolomics approach. Linear regression models examined metabolites associated with the PNS cluster. Metabolic pathway enrichment analysis was performed.

RESULTS

Participant demographics (n = 40) were 55% female, 60% white, 62.5% aged 13-19 years, and 62.5% diagnoses of Hodgkin's lymphoma and B-cell acute lymphocytic leukemia. Among 9276 unique metabolic features, 454 were associated with pain, 281 with fatigue, 596 with anxiety, 551 with depressive symptoms, and 300 with the PNS cluster across one chemotherapy cycle. Fatty acids pathways were associated with pain: de novo fatty acid biosynthesis (p < .001), fatty acid metabolism (p = .001), fatty acid activation (p = .004), and omega-3 fatty acid metabolism (p = .009). Tryptophan amino acid pathway was associated with fatigue (p < .001), anxiety (p = .015), and the PNS cluster (p = .037). Carnitine shuttle was associated with the PNS cluster (p = .015).

CONCLUSION

Fatty acids and amino acids pathways were associated with PNS in children undergoing chemotherapy. These findings require further investigation in a larger sample.

Multiomics Analysis of Structural Magnetic Resonance Imaging of the Brain and Cerebrospinal Fluid Metabolomics in Cognitively Normal and Impaired Adults

INTRODUCTION

Integrating brain imaging with large scale omics data may identify novel mechanisms of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). We integrated and analyzed brain magnetic resonance imaging (MRI) with cerebrospinal fluid (CSF) metabolomics to elucidate metabolic mechanisms and create a "metabolic map" of the brain in prodromal AD.

METHODS

In 145 subjects (85 cognitively normal controls and 60 with MCI), we derived voxel-wise gray matter volume via whole-brain structural MRI and conducted high-resolution untargeted metabolomics on CSF. Using a data-driven approach consisting of partial least squares discriminant analysis, a multiomics network clustering algorithm, and metabolic pathway analysis, we described dysregulated metabolic pathways in CSF mapped to brain regions associated with MCI in our cohort.

RESULTS

The multiomics network algorithm clustered metabolites with contiguous imaging voxels into seven distinct communities corresponding to the following brain regions: hippocampus/parahippocampal gyrus (three distinct clusters), thalamus, posterior thalamus, parietal cortex, and occipital lobe. Metabolic pathway analysis indicated dysregulated metabolic activity in the urea cycle, and many amino acids (arginine, histidine, lysine, glycine, tryptophan, methionine, valine, glutamate, beta-alanine, and purine) was significantly associated with those regions (P < 0.05).

CONCLUSION

By integrating CSF metabolomics data with structural MRI data, we linked specific AD-susceptible brain regions to disrupted metabolic pathways involving nitrogen excretion and amino acid metabolism critical for cognitive function. Our findings and analytical approach may extend drug and biomarker research toward more multiomics approaches.

Plasma metabolic phenotypes of HPV-associated vs smoking-associated head and neck cancer and patient survival

BACKGROUND

Metabolic differences between human papillomavirus (HPV)-associated head and neck squamous cell carcinoma (HNSCC) and smoking-associated HNSCC may partially explain differences in prognosis. The former relies on mitochondrial oxidative phosphorylation (OXPHOS) while the latter relies on glycolysis. These differences have not been studied in blood.

METHODS

We extracted metabolites using untargeted liquid chromatography high-resolution mass spectrometry from pretreatment plasma in a cohort of 55 HPV-associated and 82 smoking-associated HNSCC subjects. Metabolic pathway enrichment analysis of differentially expressed metabolites produced pathway-based signatures. Significant pathways (P<0.05) were reduced via principal components analysis and assessed with overall survival via Cox models. We classified each subject as glycolytic or OXPHOS phenotype and assessed it with survival.

RESULTS

Of 2,410 analyzed metabolites, 191 were differentially expressed. Relative to smoking-associated HNSCC, bile acid biosynthesis (P<0.0001) and octadecatrienoic acid beta-oxidation (P=0.01), were upregulated in HPV-associated HNSCC, while galactose metabolism (P=0.001) and vitamin B6 metabolism (P=0.01) were downregulated; the first two suggest an OXPHOS phenotype while the latter two suggest glycolytic. First principal components of bile acid biosynthesis (HR=0.52 per standard deviation, 95% CI:0.38-0.72, P<0.001) and octadecatrienoic acid beta-oxidation (HR=0.54 per sd, 95% CI:0.38-0.78, P<0.001) were significantly associated with overall survival independent of HPV and smoking. The glycolytic vs OXPHOS phenotype was also independently associated with survival (HR=3.17, 95% CI:1.07-9.35; P=0.04).

CONCLUSIONS

Plasma metabolites related to glycolysis and mitochondrial OXPHOS may be biomarkers of HNSCC patient prognosis independent of HPV or smoking. Future investigations should determine if they predict treatment efficacy.

IMPACT

Blood metabolomics may be a useful marker to aid HNSCC patient prognosis.

Grady Health System's Mobile Integrated Health Program: A Statistical Analysis of Low-Acuity 911 Calls

Grady's Mobile Integrated Health (MIH) program works to manage outpatient health concerns that otherwise burden EDs, improve quality of care, and connect patients to the appropriate level of care and resources. This prospective study collected data from 09/01/2019-03/31/2020 to analyze Grady's MIH response to low-acuity 911 calls compared to a traditional EMS (ACLS/BLS) response. A total of 2,759 EMS calls were reviewed. These calls comprised the four most common emergency medical dispatch codes for Grady's MIH response: i) "sick person other pain," ii) "diabetic alert behaving normally," iii) "back pain," and iv) "falls." Descriptive statistics and multivariable logistic regressions (MLR) were performed to compare disposition differences between MIH and traditional EMS services in whether calls were mitigated on-scene or transported. For MIH responses (n=300), 66.1% were mitigated on-scene. Comparatively, for traditional EMS responses (n=263), 11.4% were mitigated on-scene. The MLR model found the odds that a patient was mitigated on-scene for an MIH response were 24 times that for an ACLS/BLS response (OR=24.19, p<0.001) after adjusting for patient sex, ethnicity, age, blood pressure, heart rate, pain response, glucose, time of day, and EMD code. The magnitude of the odds ratio significantly differed based on the dispatch code. The results of this study indicate that utilizing Grady's current MIH model is an effective way to mitigate low-acuity 911 concerns and decrease unnecessary ED utilization, while potentially reducing hospital readmissions and healthcare costs.

Association of Epigenetic Age Acceleration With Risk Factors, Survival, and Quality of Life in Patients With Head and Neck Cancer

PURPOSE

Epigenetic age acceleration (EAA) is robustly linked with mortality and morbidity. This study examined risk factors of EAA and its association with overall survival (OS), progression-free survival (PFS), and quality of life (QOL) in patients with head and neck cancer (HNC) receiving radiation therapy.

METHODS AND MATERIALS

Patients without distant metastasis were enrolled and followed before and at the end of radiation therapy and at 6 and 12 months after radiation therapy. EAA was calculated with DNAmPhenoAge at all 4 time points. Risk factors included demographic characteristics, lifestyle, clinical characteristics, treatment-related symptoms, and blood biomarkers. Survival data were collected until August 2020, and QOL was measured using Functional Assessment of Cancer Therapy-HNC.

RESULTS

Increased comorbidity, symptoms unrelated to human papilloma virus, and more severe treatment-related symptoms were associated with higher EAA (P = .03 to P < .001). A nonlinear association (quadratic) between body mass index (BMI) and EAA was observed: decreased BMI (<35 kg/m²; P = .04) and increased BMI (≥35 kg/m²; P = .01) were linked to higher EAA. Increased EAA (per year) was associated with worse OS (hazard ratio [HR], 1.11 [95% confidence interval {CI}, 1.03-1.18; P = .004]; HR, 1.10 [95% CI, 1.01-1.19; P = .02] for EAA at 6 and 12 months after treatment, respectively) and PFS (HR, 1.10 [95% CI, 1.02-1.19; P = .02]; HR, 1.14 [95% CI, 1.06-1.23; P < .001]; and HR, 1.08 [95% CI, 1.02-1.14; P = .01]) for EAA before, immediately after, and 6 months after radiation therapy, respectively) and QOL over time (β = -0.61; P = .001). An average of 3.25 to 3.33 years of age acceleration across time, which was responsible for 33% to 44% higher HRs of OS and PFS, was observed in those who died or developed recurrence compared with those who did not (all P < .001).

CONCLUSIONS

Compared with demographic and lifestyle factors, clinical characteristics were more likely to contribute to faster biological aging in patients with HNC. Acceleration in epigenetic age resulted in more aggressive adverse events, including OS and PFS. EAA could be considered as a marker for cancer outcomes, and decelerating aging could improve survival and QOL.

Abstract PO-033: Metabolic differences between human papillomavirus-related and smoking-related squamous cell carcinoma of the head and neck independently predict overall patient survival

Background: Human papillomavirus (HPV) status and tobacco use are the strongest prognostic factors of survival for squamous cell carcinoma of the head and neck (SCCHN). Patients with HPV-related SCCHN live more than twice as long as patients with tobacco-related SCCHN. The underlying metabolic dysregulation characterizing these two clinically distinct groups may pose exciting translational targets that are yet unknown. Using a metabolome-wide approach, we investigated the metabolic differences of HPV vs smoking-related SCCHN in a large patient population. We then linked those metabolic differences to patient survival. Methods: We recruited 137 patients from Emory University hospitals and clinics: 55 HPV-positive never smokers (i.e., HPV-related SCCHN) and 82 HPV-negative current or former smokers (i.e., smoking-related SCCHN). We performed untargeted high-resolution metabolomics using liquid chromatography-mass spectrometry on pre-treatment patient plasma at the Emory Clinical Biomarkers Laboratory. Our statistical approach has three steps. 1) Metabolite feature selection: We aggregated the results from two selection algorithms (5-fold cross-validated partial least squares discriminant analysis and support vector machine) to determine the metabolites that differentiated HPV vs smoking-related SCCHN. 2) Metabolic pathway analysis (Mummichog): We identified statistically significant (P&lt;0.05) metabolic pathways stemming from the step 1 metabolites. 3) Metabolic profile and survival analysis: We created a metabolite-based linear principal component using the significant pathway metabolites in step 2 and estimated its association with overall survival using adjusted Cox models. Results: Of the 9,640 extracted metabolites, the two models found 229 that differentiated HPV vs smoking-related SCCHN. Metabolic pathway analysis found fatty acid and steroid metabolism was significantly upregulated in HPV patients: bile acid biosynthesis (P&lt;0.0001), 21-carbon steroid hormone metabolism (P=&lt;0.0001), fatty acid beta-oxidation (P=0.04). Pathways related to sugar metabolism were significantly upregulated in smoking patients: galactose metabolism (P=0.003), starch and sucrose metabolism (P=0.007), hexose phosphorylation (P=0.03), sialic acid metabolism (P=0.03). A principal component comprised of the 16 pathway metabolites was significantly associated with overall survival (HR=1.27 per standard deviation, P=0.01, 95% CI: 1.06-1.51) independent of HPV, smoking, tumor site, age, sex, race, and body mass index. Conclusions: Metabolites corresponding to fatty acid, steroid, and sugar metabolism differentiated HPV-related SCCHN from smoking-related SCCHN. They may be clinically informative as prognostic markers of survival. Our novel findings in patient plasma parallel those from HPV positive vs negative SCCHN cell lines, suggesting that metabolites in blood may be putative biomarkers of tumor metabolic activity and thus potentially useful for drug discovery.

Citation Format: Ronald C. Eldridge, Karan Uppal, Jonathan Beitler, Kristin Higgins, Evanthia C. Wommack, Dong M. Shin, Nabil F. Saba, Andrew Miller, Deborah W. Bruner, Canhua Xiao. Metabolic differences between human papillomavirus-related and smoking-related squamous cell carcinoma of the head and neck independently predict overall patient survival [abstract]. In: Abstracts: AACR Special Virtual Conference on Epigenetics and Metabolism; October 15-16, 2020; 2020 Oct 15-16. Philadelphia (PA): AACR; Cancer Res 2020;80(23 Suppl):Abstract nr PO-033.

Lipidome signatures of metastasis in a transgenic mouse model of sonic hedgehog medulloblastoma

Medulloblastoma (MB), the most common malignant pediatric brain tumor, has high propensity to metastasize. Currently, the standard treatment for MB patients includes radiation therapy administered to the entire brain and spine for the purpose of treating or preventing against metastasis. Due to this aggressive treatment, the majority of long-term survivors will be left with permanent and debilitating neurocognitive impairment, for the 30-40% patients that fail to respond to treatment, all will relapse with terminal metastatic disease. An understanding of the underlying biology that drives MB metastasis is lacking, and is critically needed in order to develop targeted therapeutics for its prevention. To examine the metastatic biology of sonic hedgehog (SHH) MB, the human MB subgroup with the worst clinical outcome in children, we first generated a robust SmoA1-Math-GFP mouse model that reliably reproduces human SHH MB whereby metastases can be visualized under fluorescence microscopy. Lipidome alterations associated with metastasis were then investigated by applying ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) under positive ionization mode to primary tumor samples collected from mice without (n = 18) and with (n = 7) metastasis. Thirty-four discriminant lipids associated with SHH MB metastasis were successfully annotated, including ceramides (Cers), sphingomyelins (SMs), triacylglycerols (TGs), diacylglycerols (DGs), phosphatidylcholines (PCs), and phosphatidic acids (PAs). This study provides deeper insights into dysregulations of lipid metabolism associated with SHH MB metastatic progression, and thus serves as a guide toward novel targeted therapies.

Endogenous estradiol and inflammation biomarkers: potential interacting mechanisms of obesity-related disease

PURPOSE

Disentangling the effects of endogenous estrogens and inflammation on obesity-related diseases requires a clearer understanding of how the two biological mechanisms relate to each other.

METHODS

We studied 155 healthy postmenopausal women not taking menopausal hormone therapy enrolled in the Prostate Lung Colorectal and Ovarian (PLCO) screening cancer trial. From a baseline blood draw, we measured endogenous estradiol and 69 inflammation biomarkers: cytokines, chemokines, adipokines, angiogenic factors, growth factors, acute phase proteins, and soluble receptors. We evaluated the estradiol-inflammation relationship by assessing associations across different models (linear, ordinal logistic, and binary logistic) using a variety of estradiol classifications. We additionally investigated the estradiol-inflammation relationship stratified by baseline obesity status (BMI < 30 stratum and BMI > 30 stratum).

RESULTS

Associations of estradiol with 7 inflammation biomarkers met p < 0.05 statistical significance in linear and ordinal models: C-reactive protein (CRP), adiponectin, chemokine (C-X-C motif) ligand-6, thymus activation-regulated chemokine, eosinophil chemotactic protein, plasminogen activator inhibitor-1, and serum amyloid A. The positive association between estradiol and CRP was robust to model changes. Each standard deviation increase in endogenous estradiol doubled a woman's odds of having CRP levels higher than the study median (odds ratio 2.29; 95% confidence interval 1.28, 4.09). Estradiol was consistently inversely associated with adiponectin. Other estradiol-inflammation biomarker associations were not robust to model changes.

CONCLUSIONS

Endogenous estradiol appears to be associated with CRP and adiponectin; the evidence is limited for other inflammation biomarkers.

Changing functional status within 6 months posttreatment is prognostic of overall survival in patients with head and neck cancer: NRG Oncology Study

BACKGROUND

Is posttreatment functional status prognostic of overall survival in patients with head and neck cancer (HNC).

METHODS

In an HNC clinical trial, 495 patients had two posttreatment functional assessments measuring diet, public eating, and speech within 6 months. Patients were grouped by impairment (highly, moderately, modestly, or not impaired) and determined if they improved, declined, or did not change from the first assessment to the second. Multivariable Cox models estimated overall mortality.

RESULTS

Across all three scales, the change in posttreatment patient function strongly predicted overall survival. In diet, patients who declined to highly impaired had three times the mortality of patients who were not impaired at both assessments (hazard ratio [HR] = 3.60; 95% confidence interval, 2.02-6.42). For patients improving from highly impaired, mortality was statistically similar to patients with no impairment (HR = 1.38; 95% CI, 0.82-2.31).

CONCLUSIONS

Posttreatment functional status is a strong prognostic marker of survival in patients with HNC.

Abstract 2418: Are we underestimating cancer mortality? A mediation model shows a larger impact from smoking

Purpose: Cancer mortality is estimated using underlying cause of death. For instance, to estimate the risk of head and neck cancer (HNC) mortality attributed to smoking, ‘death by HNC’ is the outcome (ie, smoking → HNC death). This conventional approach has notable flaws. It relies on accurate reporting on cause of death, a potential source of misclassification, and it excludes any mortality effects in cancer patients who die from other causes, a potential source of selection bias. Regarding HNC, 80% of all patients are excluded using this approach. These potential biases may result in underestimating cancer mortality from important public health factors such as smoking, obesity and physical activity. To overcome these biases, we are proposing a new approach to estimate cancer mortality. The method uses a causal mediation model and we demonstrate it in a smoking and HNC example.

Methods: The model estimates cancer mortality through a mediated pathway. In our example, smoking is the exposure, incident HNC is the mediator, and all-cause mortality is the outcome (smoking → HNC incidence → all deaths). The mediated effect is the risk of mortality attributed to HNC, which in turn, is attributed to smoking. We compare the results in the mediation model to the conventional approach using Aalen’s additive hazards, which estimates the absolute increase in deaths in smokers compared to non-smokers; the mediation model also uses logistic regression for the HNC mediator. Both models are adjusted for age, sex, race, body mass index, diet, alcohol use, physical activity, marital status, education, and reported health. We used the NIH-AARP Diet and Health cohort, which has smoking information at baseline, and cancer incidence and mortality during follow-up on 518,100 U.S. subjects.

Results: In the study, 2,279 subjects developed HNC during follow-up, among whom, 508 (22%) died from HNC; 125,442 subjects died from any cause. Under the conventional method, 11.0 more deaths (per 100,000 person-years) are attributed to HNC from current smokers relative to never smokers (95% CI: 8.0, 13.9); 1.8 deaths in former smokers (95% CI: 0.4, 3.2). Under the mediation method, 13.4 more deaths are attributed to HNC from current smokers relative to never smokers (95% CI: 9.3, 16.8); 2.2 deaths in former smokers (95% CI: 1.2, 3.4). Comparing the two estimates, the conventional approach underestimates HNC mortality attributed to smoking by 18%. Noteworthy, the underestimate for current smokers manifests primarily in men, while the underestimate for former smokers manifests primarily in women.

Conclusions: The mediation method overcomes limitations in prior approaches of cancer mortality. In our example, the method estimated a stronger effect on HNC mortality by smoking. Further study is warranted as published estimates of cancer mortality by notable public health risk factors (eg, smoking, obesity, physical activity) may be underestimating their impact.

Citation Format: Ronald C. Eldridge, W. Dana Flanders, Colin Adler, Deborah W. Bruner, Canhua Xiao. Are we underestimating cancer mortality? A mediation model shows a larger impact from smoking [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 2418.

Abstract 4257: Predicting fatigue levels of head and neck cancer patients with gene expression using machine learning

Introduction: Cancer-related fatigue is physical, emotional, or cognitive exhaustion which can affect treatment adherence and quality of life, and predict survival. Head and neck cancer (HNC) patients experience high levels of fatigue due to the degree of radiotherapy (RT), but that alone does not fully explain it. Recent studies suggest that fatigue may be related to a patient's personalized metabolic and inflammatory response, suggesting a patient's gene expression (GE) may be used to predict and monitor fatigue - a precursor to managing it. In this analysis, we examined whether GE is predictive of pre-RT patient reported fatigue using cross-validated penalized Lasso regression - a machine learning approach.

Study population: From Emory University Clinics, 44 HNC patients donated blood samples before undergoing RT. GE was assessed using an Affymetrix Clariom S Human microarray which measures gene transcripts for roughly 24,000 genes; each probe was log-transformed, normalized, and standardized. The validated 20-item self-report multidimensional fatigue inventory questionnaire measured each patient's continuous fatigue score.

Methods: To predict fatigue, we used leave-one-out cross validation (CV). This means we built a Lasso regression model using GE probes from 43 of 44 subjects and used that model to predict fatigue for the remaining subject; this process was repeated 44 times until all subjects had a predicted fatigue score. We chose penalized Lasso regression because the ‘penalty' performs variable selection and mitigates collinearity between the GE probes; the penalty was chosen by an extra layer of CV not described. We compared the predicted fatigue scores to the corresponding patient reported fatigue scores using R2 (higher values mean stronger correlation and better prediction).

Results: To test the approach, we allowed the Lasso regression to build a model based on all subjects and predict fatigue on all subjects. This prediction was expected to be high, and it was (R2=0.98). However, the approach was less successful predicting fatigue during the leave-one-out CV prediction (R2=0.15). The probes most influential predicting fatigue are linked to genes involved in tryptophan metabolism (precursor to the neurotransmitter serotonin), double strand break DNA repair, and transforming growth factor beta receptor signaling (inflammation cytokine).

Conclusions: Gene expression may predict fatigue in head and neck cancer patients, but there is room for improvement. The model suggests that a patient's personalized DNA repair process, metabolic and inflammatory response may play key roles in patient fatigue. Integrating more data focused on these biological processes, for instance patient metabolomics, may improve the model performance. Future efforts include collecting a larger sample, trying alternative machine learning methods, and testing the robustness of the model over time.

Citation Format: Ronald C. Eldridge, Andrew H. Miller, Deborah W. Bruner, Jonathan J. Beitler, Kristin A. Higgins, Evanthia C. Wommack, Linh Kha Huynh, Nabil F. Saba, Dong M. Shin, Canhua Xiao. Predicting fatigue levels of head and neck cancer patients with gene expression using machine learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 4257.

A Novel Application of Structural Equation Modeling Estimates the Association between Oxidative Stress and Colorectal Adenoma

In vitro evidence implicates oxidative stress in many adverse health conditions, including colorectal neoplasia. In human studies, however, oxidative stress is measured by imperfect biomarkers, which are inconsistently associated with health outcomes. Structural equation modeling (SEM) offers one possible solution by modeling a latent (unobserved) construct from multiple biomarkers. Our goal was to investigate the association of a latent oxidative stress variable with colorectal adenoma. Using SEM, we analyzed pooled data from two cross-sectional studies of colorectal adenoma (n = 526) that measured five plasma biomarkers of oxidative stress and inflammation that comprised the latent oxidative stress variable: F₂-isoprostanes (FIP), fluorescent oxidation products (FOP), mitochondrial DNA (MtDNA) copy number, γ-tocopherol (Gtoc), and C-reactive protein (CRP). Higher levels of oxidative stress were associated with colorectal adenoma [OR = 3.23 per SD increase in oxidative stress; 95% confidence interval (CI), 1.28-8.18]. The latent variable estimate was considerably stronger than the associations of adenoma with the individual biomarkers, which were modest and mostly nonsignificant. Risk factors were associated with adenoma via the oxidative stress pathway, particularly overweight and obesity with an OR = 1.50; 95% CI, 1.10-2.81; and OR = 2.95; 95% CI, 1.28-12.45, respectively. Oxidative stress may be positively associated with colorectal adenoma, and important risk factors may act through this mechanism, but the cross-sectional design of the current study precludes observing the directionality of associations. The presence of an adenoma could affect levels of the circulating biomarkers; thus, we should be cautious of strong conclusions until the findings are replicated in a follow-up study. Cancer Prev Res; 11(1); 52-58. ©2017 AACR.

Smoking and subsequent human papillomavirus infection: a mediation analysis

PURPOSE

Smoking is an established risk factor for a human papillomavirus (HPV) infection advancing to cervical precancer and cancer, but its role earlier in the natural history is less clear. Smoking is inversely associated with possessing HPV antibodies from a past infection suggesting that smoking may influence acquiring subsequent infections.

METHODS

In a cohort of 1976 U.S. women, we evaluate whether reduced antibodies to HPV-16 is a mechanism for smoking's role on acquiring a subsequent HPV-16 infection, through the analytic technique of causal mediation analysis. We posit a causal model and estimate two counterfactually defined effects: a smoking impaired antibody-mediated indirect effect and a nonmediated direct effect representing all other potential mechanisms of smoking.

RESULTS

Compared to never smokers, current smokers had increased odds of HPV-16 infection by the antibody-mediated indirect effect (odds ratio [OR] = 1.29; 95% confidence interval [CI]: 1.11, 1.73); the estimated direct effect was very imprecise (OR = 0.57; 95% CI, 0.26-1.13). We observed a stronger estimated indirect effect among women who smoked at least half a pack of cigarettes daily (OR = 1.61, 95% CI, 1.27-2.15) than among women who smoked less than that threshold (OR = 1.09; 95% CI, 0.94-1.44).

CONCLUSIONS

This is the first study to directly test the mechanism underlying smoking as an HPV cofactor. The results support current smoking as a risk factor earlier in the natural history of HPV and are consistent with the hypothesis that smoking increases the risk of a subsequent infection by reducing immunity.

Using multiple biomarkers and determinants to obtain a better measurement of oxidative stress: a latent variable structural equation model approach

PURPOSE

Since oxidative stress involves a variety of cellular changes, no single biomarker can serve as a complete measure of this complex biological process. The analytic technique of structural equation modeling (SEM) provides a possible solution to this problem by modelling a latent (unobserved) variable constructed from the covariance of multiple biomarkers.

METHODS

Using three pooled datasets, we modelled a latent oxidative stress variable from five biomarkers related to oxidative stress: F₂-isoprostanes (FIP), fluorescent oxidation products, mitochondrial DNA copy number, γ-tocopherol (Gtoc) and C-reactive protein (CRP, an inflammation marker closely linked to oxidative stress). We validated the latent variable by assessing its relation to pro- and anti-oxidant exposures.

RESULTS

FIP, Gtoc and CRP characterized the latent oxidative stress variable. Obesity, smoking, aspirin use and β-carotene were statistically significantly associated with oxidative stress in the theorized directions; the same exposures were weakly and inconsistently associated with the individual biomarkers.

CONCLUSIONS

Our results suggest that using SEM with latent variables decreases the biomarker-specific variability, and may produce a better measure of oxidative stress than do single variables. This methodology can be applied to similar areas of research in which a single biomarker is not sufficient to fully describe a complex biological phenomenon.

Prediagnostic circulating inflammation markers and endometrial cancer risk in the prostate, lung, colorectal and ovarian cancer (PLCO) screening trial

Inflammation is proposed to increase risk of developing endometrial cancer, but few prospective epidemiologic studies have investigated the relationship between circulating inflammation markers and endometrial cancer risk. In a nested case-control study within the PLCO Screening Trial we measured serum levels of 64 inflammation-related biomarkers in 284 incident endometrial cancer cases and 284 matched controls. Using multivariable logistic regression inflammation markers were evaluated individually and combined into a cross-validated inflammation score. Of 64 markers, 22 were associated with endometrial cancer risk at p < 0.05 and 17 of 22 markers remained associated after multiple testing corrections. After adjusting for BMI and estradiol, SERPINE1 [quartile(Q)4 vs. Q1 odds ratio (OR) (95% confidence interval (CI)), p trend = 2.43 (0.94-6.29), 0.03] and VEGFA [2.56 (1.52-4.30), 0.0002] were positively associated with endometrial cancer risk, while CCL3 [0.46 (0.27-0.77), 0.01], IL13 [0.55 (0.33-0.93), 0.01], IL21 [0.52 (0.31-0.87), 0.01], IL1B [0.51 (0.30-0.86), 0.01] and IL23 [0.60 (0.35-1.03), 0.02] were inversely associated with risk. We observed large differences in ORs across BMI-inflammation score categories. Endometrial cancer risk was most pronounced among obese women with the highest inflammation score tertile (T) [10.25 (3.56-29.55) vs. normal BMI/T1]. Several inflammation markers were prospectively associated with endometrial cancer, including adipokines, pro- and anti-inflammatory cytokines, angiogenic factors and acute phase proteins. Inverse associations with anti-inflammatory markers (IL13, IL21), other inflammation markers/mediators (CCL3, IL1B, IL23), and a robust positive association between VEGFA and endometrial cancer risk were independent of BMI and estradiol, suggesting that these factors may influence risk through other mechanisms.

Abstract 852: Smoking and HPV antibodies, a mediation analysis of HPV re-infection

Background: Naturally acquired antibodies against human papillomavirus (HPV) infection can protect against recurring infection. Smoking can adversely affect a person's immune response, both adaptive and innate systems; and it is also associated with increased risk of HPV-related precancers. But, it is unclear whether smoking results in higher HPV incidence because of an impaired immune response.

Objective: We investigated whether lowered HPV-16 antibody titers mediates the effect of smoking on incident HPV-16 infection.

Methods: We used a path analytic model where smoking lowers a person's antibody titer, thereby resulting in a higher risk of acquiring an incident re-infection. The model estimated the indirect effect of smoking to infection, mediated by antibodies; it also estimated the direct effect, which is not mediated. The path model includes two logistic equations with effects and confidence intervals estimated by parametric bootstrapping. Both equations were adjusted for age, age at sexual initiation, lifetime number of sexual partners, and ever diagnosis of a sexually transmitted infection other than HPV.

Study Population: A subset of women in the Atypical Squamous Cells of Undetermined Significance/Low-Grade Squamous Intraepithelial Lesion Triage Study provided blood samples at baseline. The samples were assessed for HPV antibodies using a multiplex Luminex assay. Smoking and sexual habits were assessed by a baseline questionnaire. During two years of follow-up, bi-annual cervical specimens were genotyped for HPV by PCR assays. The analytic population (n = 1,978) were women without a cervical intraepithelial lesion grade 2 or more severe and HPV-16 DNA negative at baseline. Among those women, 131 tested HPV-16 DNA positive during follow-up.

Results: Compared to never smokers, current smokers had an increased risk of incident HPV-16 infection by the antibody mediated pathway (OR = 1.24, 95% CI: 1.08, 1.60); the direct effect was non-significant. For former smokers, neither direct nor indirect effects were significant. The indirect effect of current smoking increased for women who smoked at least half a pack of cigarettes daily (OR = 1.47, 95% CI: 1.20, 2.02).

Conclusions: This is the first analytic model to suggest that current smoking increases the risk of incident HPV-16 infection through a reduction of HPV-16 antibody titers. The association appears dose-dependent, increasing for number of cigarettes smoked per day. Because our follow-up time was limited, we could not measure, and thus, extend these results to incident precancer outcomes.

Citation Format: Ronald C. Eldridge, Michael Pawlita, Lauren Wilson, Philip E. Castle, Tim Waterboer, Patti E. Gravitt, Mark Schiffman, Nicolas Wentzensen. Smoking and HPV antibodies, a mediation analysis of HPV re-infection. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 852. doi:10.1158/1538-7445.AM2015-852

Invited Commentary: Clinical Utility of Prediction Models for Rare Outcomes--The Example of Pancreatic Cancer

Translating relative risk estimates into absolute risks is important in evaluating the potential clinical and public health relevance of etiologic discoveries. Predicting high absolute risk is challenging, particularly for rare endpoints such as pancreatic cancer. Recent efforts to develop risk prediction models for pancreatic cancer have found moderate risk levels for very small parts of the population. A new approach in which clinical symptoms and medication use are evaluated in addition to information on risk factors is presented by Risch et al. in this issue of the Journal (Am J Epidemiol. 2015;182(1):26-34). The authors estimated absolute risks based on the relative risks obtained from their case-control study. Their absolute risk estimates were higher than those from previous approaches but remained restricted to a very small proportion of the general population. In the present commentary, we address issues of absolute risk stratification (particularly for rare diseases), specific analytic methods, and how actionable information will differ based on the disease and possible intervention. We suggest that moving from cancer-specific models to broader models used to predict risk for multiple outcomes can make risk prediction for rare diseases more effective. When considering translational goals, it is important to estimate absolute risk at the early stages of etiologic research. The results can be sobering but allow focusing on the most promising goals.

Uncontrolled confounding in studies of screening effectiveness: an example of colonoscopy

OBJECTIVE

To estimate the expected magnitude of error produced by uncontrolled confounding from health behaviours in observational medical record-based studies evaluating effectiveness of screening colonoscopy.

METHODS

We used data from the prospective National Institutes of Health American Association of Retired Persons (NIH-AARP) Diet and Health Study to assess the impact of health behaviour related factors (lifestyle, education, and use of non-steroidal anti-inflammatory drugs [NSAID]) on the association between colonoscopy and colorectal cancer (CRC) mortality. We first examined the difference between adjusted and unadjusted results within the cohort data, and then estimated a broader range of likely confounding errors based on the Breslow-Day approach that uses prevalence of confounders among persons with and without exposure, and the rate ratio reflecting the association between these confounders and the outcome of interest. As dietary factors and habits are often inter-correlated, we combined these variables (physical activity, body mass index, waist-to-hip ratio, alcohol consumption, and intakes of red meat, processed meat, fibre, milk, and calcium) into a "healthy lifestyle score" (HLS).

RESULTS

The estimated error (a ratio of biased-to-true result) attributable to confounding by HLS was 0.959-0.997, indicating less than 5% departure from the true effect of colonoscopy on CRC mortality. The corresponding errors ranged from 0.970 to 0.996 for NSAID, and from 0.974 to 1.006 for education (all ≤3% difference). The results for other CRC screening tests were similar.

CONCLUSION

Health behaviour-related confounders, either alone or in combination, seem unlikely to strongly affect the association between colonoscopy and CRC mortality in observational studies of CRC screening.

Identification of a BRCA2-specific modifier locus at 6p24 related to breast cancer risk

Common genetic variants contribute to the observed variation in breast cancer risk for BRCA2 mutation carriers; those known to date have all been found through population-based genome-wide association studies (GWAS). To comprehensively identify breast cancer risk modifying loci for BRCA2 mutation carriers, we conducted a deep replication of an ongoing GWAS discovery study. Using the ranked P-values of the breast cancer associations with the imputed genotype of 1.4 M SNPs, 19,029 SNPs were selected and designed for inclusion on a custom Illumina array that included a total of 211,155 SNPs as part of a multi-consortial project. DNA samples from 3,881 breast cancer affected and 4,330 unaffected BRCA2 mutation carriers from 47 studies belonging to the Consortium of Investigators of Modifiers of BRCA1/2 were genotyped and available for analysis. We replicated previously reported breast cancer susceptibility alleles in these BRCA2 mutation carriers and for several regions (including FGFR2, MAP3K1, CDKN2A/B, and PTHLH) identified SNPs that have stronger evidence of association than those previously published. We also identified a novel susceptibility allele at 6p24 that was inversely associated with risk in BRCA2 mutation carriers (rs9348512; per allele HR = 0.85, 95% CI 0.80-0.90, P = 3.9 × 10(-8)). This SNP was not associated with breast cancer risk either in the general population or in BRCA1 mutation carriers. The locus lies within a region containing TFAP2A, which encodes a transcriptional activation protein that interacts with several tumor suppressor genes. This report identifies the first breast cancer risk locus specific to a BRCA2 mutation background. This comprehensive update of novel and previously reported breast cancer susceptibility loci contributes to the establishment of a panel of SNPs that modify breast cancer risk in BRCA2 mutation carriers. This panel may have clinical utility for women with BRCA2 mutations weighing options for medical prevention of breast cancer.

Jewish ethnicity and pancreatic cancer mortality in a large U.S. cohort

BACKGROUND

An association between Jewish ethnicity and pancreatic cancer risk was suggested by analyses comparing pancreatic cancer mortality rates between Jews and non-Jews in New York in the 1950s. These analyses lacked information on potential confounding factors and the association between Jewish ethnicity and pancreatic cancer has not been examined in any contemporary U.S. population or in any cohort study.

METHODS

We examined the association between Jewish ethnicity and pancreatic cancer mortality among approximately 1 million participants in the Cancer Prevention Study II cohort. Participants completed a questionnaire at enrollment in 1982 which included information on religion, smoking, obesity, and diabetes. During follow-up through 2006, there were 6,727 pancreatic cancer deaths, including 480 among Jewish participants. Proportional hazards modeling was used to calculate multivariable rate ratios (RR).

RESULTS

After adjusting for age, sex, smoking, body mass index, and diabetes, pancreatic cancer mortality was higher among Jewish participants than among non-Jewish whites (RR = 1.43; 95% CI, 1.30-1.57). In analyses by birthplace, RRs were 1.59 (95% CI, 1.31-1.93) for North American-born Jews with North American-born parents, 1.43 (95% CI, 1.27-1.61) for North American-born Jews with 1 or more parents born outside North America, and 1.03 (0.73, 1.44) for Jews born outside North America (P(heterogeneity) = 0.07).

CONCLUSIONS

These results support a higher risk of developing pancreatic cancer among U.S. Jews that is not explained by established risk factors.

IMPACT

Future studies may clarify the role of specific environmental or genetic factors responsible for higher risk among U.S. Jews.