Search

Acute kidney injury (AKI) is common after hematopoietic cell transplantation (HCT) and is associated with poorer outcomes. Risk factors for AKI after pediatric HCT are not fully understood. The study objective was to assess unique risk factors for AKI in the HCT population and evaluate post-HCT AKI patterns.We conducted a retrospective cohort study of patients 21 years of age who underwent HCT at Seattle Children's Hospital/Fred Hutchinson Cancer Center from September 2008 to July 2017 (n = 484). We defined AKI using KDIGO criteria. We collected demographics, baseline HCT characteristics, post-HCT complications, and mortality. Multinomial logistic regression was used to estimate association between AKI and potential risk factors. We used adjusted Cox proportional hazard ratios to evaluate differences in mortality.One hundred and eighty-six patients (38%) developed AKI. Seventy-nine (42%) had severe AKI and 27 (15%) required kidney replacement therapy. Fluid overload was common in all groups and 67% of those with severe AKI had 10% fluid overload. Nephrology was consulted in less than 50% of those with severe AKI. In multivariable analysis, risk of severe AKI was lower in those taking a calcineurin inhibitor (CNI). Risk of death was higher in severe AKI compared to no AKI (RR 4.6, 95% CI 2.6-8.1).AKI and fluid overload are common in pediatric patients after HCT. Severe AKI occurred less often with CNI use and was associated with higher mortality. Future interventions to reduce AKI and its associated complications such as fluid overload are approaches to reducing morbidity and mortality after HCT. A higher resolution version of the Graphical abstract is available as Supplementary information.

Introduction:Geroscience studies of low-dose rapamycin in laboratory species have identified numerous benefits, including reversing age-related cardiac dysfunction. Cardiovascular benefits have been observed in dogs with 10 weeks of treatment, raising questions about possible benefits and adverse effects of long-term use of low-dose rapamycin. The objectives of this study were to assess the impact of 6 months of low-dose rapamycin on echocardiographic indices of cardiac function in healthy dogs and to document the occurrence of adverse events.Methods:Seventeen client-owned dogs aged 6–10 years, weighing 18–36 kg, and without significant systemic disease were included in a prospective, randomized, placebo-controlled, masked clinical trial. Low-dose rapamycin (0.025 mg/kg) or placebo was administered three times per week for 6 months. Baseline, 6-month, and 12-month evaluation included physical examination, cardiology examination, and clinicopathology. Three-month evaluation included physical examination and clinicopathology. Owners completed online questionnaires every 2 weeks.Results:There were no statistically significant differences in echocardiographic parameters between rapamycin and placebo groups at 6 or 12 months. No clinically significant adverse events occurred. In 26.8% of the bi-weekly surveys owners whose dogs received rapamycin reported perceived positive changes in behavior or health, compared to 8.1% in the placebo group (p = 0.04).Discussion:While no clinically significant change in cardiac function was observed in dogs treated with low-dose rapamycin, the drug was well-tolerated with no significant adverse events.

BACKGROUND: With the use of artificial intelligence and machine learning techniques for biomedical informatics, security and privacy concerns over the data and subject identities have also become an important issue and essential research topic. Without intentional safeguards, machine learning models may find patterns and features to improve task performance that are associated with private personal information. OBJECTIVE: The privacy vulnerability of deep learning models for information extraction from medical textural contents needs to be quantified since the models are exposed to private health information and personally identifiable information. The objective of the study is to quantify the privacy vulnerability of the deep learning models for natural language processing and explore a proper way of securing patients’ information to mitigate confidentiality breaches. METHODS: The target model is the multitask convolutional neural network for information extraction from cancer pathology reports, where the data for training the model are from multiple state population-based cancer registries. This study proposes the following schemes to collect vocabularies from the cancer pathology reports; (a) words appearing in multiple registries, and (b) words that have higher mutual information. We performed membership inference attacks on the models in high-performance computing environments. RESULTS: The comparison outcomes suggest that the proposed vocabulary selection methods resulted in lower privacy vulnerability while maintaining the same level of clinical task performance.

Purpose: To determine if gene expression signature of invasive oral squamous cell carcinoma (OSCC) can subclassify OSCC based on survival.Experimental Design: We analyzed the expression of 131 genes in 119 OSCC, 35 normal, and 17 dysplastic mucosa to identify cluster-defined subgroups. Multivariate Cox regression was used to estimate the association between gene expression and survival. By stepwise Cox regression, the top predictive models of OSCC-specific survival were determined and compared by receiver operating characteristic analysis.Results: The 3-year overall mean ± SE survival for a cluster of 45 OSCC patients was 38.7 ± 0.09% compared with 69.1 ± 0.08% for the remaining patients. Multivariate analysis adjusted for age, sex, and stage showed that the 45 OSCC patient cluster had worse overall and OSCC-specific survival (hazard ratio, 3.31; 95% confidence interval, 1.66-6.58 and hazard ratio, 5.43; 95% confidence interval, 2.32-12.73, respectively). Stepwise Cox regression on the 131 probe sets revealed that a model with a term for LAMC2 (laminin γ2) gene expression best identified patients with worst OSCC-specific survival. We fit a Cox model with a term for a principal component analysis-derived risk score marker and two other models that combined stage with either LAMC2 or PCA. The area under the curve for models combining stage with either LAMC2 or PCA was 0.80 or 0.82, respectively, compared with 0.70 for stage alone (P = 0.013 and 0.008, respectively).Conclusions: Gene expression and stage combined predict survival of OSCC patients better than stage alone.

PDF file - 49K

Supplementary Tables 1-3 from A Sex-Specific Association between a 15q25 Variant and Upper Aerodigestive Tract Cancers

Background: The matrix metalloproteinases (MMP) cause degradation of the extracellular matrix and basement membranes, and thus may play a key role in cancer development.Methods: In our search for biomarkers for oral squamous cell carcinomas (OSCC), we compared primary OSCC, oral dysplasia and control subjects with respect to: (i) expression of MMP1, MMP3, MMP10, and MMP12 in oral epithelial tissue using Affymetrix U133 2.0 Plus GeneChip arrays, followed by quantitative reverse transcription-PCR (qRT-PCR) for MMP1, and (ii) determination of MMP1 and MMP3 concentrations in saliva.Results:MMP1 expression in primary OSCC (n = 119) was >200-fold higher (P = 7.16 × 10−40) compared with expression levels in nonneoplastic oral epithelium from controls (n = 35). qRT-PCR results on 30 cases and 22 controls confirmed this substantial differential expression. The exceptional discriminatory power to separate OSCC from controls was validated in two independent testing sets (AUC% = 100; 95% CI: 100–100 and AUC% = 98.4; 95% CI: 95.6–100). Salivary concentrations of MMP1 and MMP3 in OSCC patients (33 stage I/II, 26 stage III/IV) were 6.2 times (95% CI: 3.32–11.73) and 14.8 times (95% CI: 6.75–32.56) higher, respectively, than in controls, and displayed an increasing trend with higher stage disease.Conclusion: Tumor and salivary MMPs are robust diagnostic biomarkers of OSCC.Impact: The capacity of MMP gene expression to identify OSCC provides support for further investigation into MMPs as potential markers for OSCC development. Detection of MMP proteins in saliva in particular may provide a promising means to detect and monitor OSCC noninvasively. Cancer Epidemiol Biomarkers Prev; 20(12); 2628–36. ©2011 AACR.

Association of frequency of marijuana use, duration of marijuana, and joint years of marijuana use with oropharyngeal and oral tongue cancers

PDF file - 71K

PDF file - 51.3KB

Purpose: To determine the differential gene expression between oral squamous cell carcinoma (OSCC) with and without metastasis to cervical lymph nodes and to assess prediction of nodal metastasis by using molecular features.Experimental Design: We used Affymetrix U133 2.0 plus arrays to compare the tumor genome-wide gene expression of 73 node-positive OSCCs with 40 node-negative OSCCs (≥18 months). Multivariate linear regression was used to estimate the association between gene expression and nodal metastasis. Stepwise logistic regression and receiver operating characteristics (ROC) analysis were used to generate predictive models and to compare these with models by using tumor size alone.Results: We identified five genes differentially expressed between node-positive and node-negative OSCCs after adjusting for tumor size and human papillomavirus status: REEP1, RNF145, CTONG2002744, MYO5A, and FBXO32. Stepwise regression identified a four-gene model (MYO5A, RFN145, FBXO32, and CTONG2002744) as the most predictive of nodal metastasis. A leave-one-out ROC analysis revealed that our model had a higher area under the curve (AUC) for identifying occult nodal metastasis compared with that of a model by tumor size alone (respective AUC: 0.85 and 0.61; P = 0.011). A model combining tumor size and gene expression did not further improve the prediction of occult metastasis. Independent validation using 31 metastatic and 13 nonmetastatic cases revealed a significant underexpression of CTONG2002744 (P = 0.0004).Conclusions: These results suggest that our gene expression markers of OSCC metastasis hold promise for improving current clinical practice. Confirmation by others and functional studies of CTONG2002744 is warranted. Clin Cancer Res; 17(8); 2466–73. ©2011 AACR.

PDF file - 49K

Background: A history of diabetes is associated with an increased risk of several types of cancers. Whether diabetes is a risk factor for head and neck cancer (HNC) has received little attention.Methods: We pooled data from 12 case–control studies including 6,448 cases and 13,747 controls, and estimated OR and 95% CI for the associations between diabetes and HNC, adjusted for age, education level, sex, race/ethnicity, study center, cigarette smoking, alcohol use, and body mass index.Results: We observed a weak association between diabetes and the incidence of HNC overall (OR, 1.09; 95% CI: 0.95–1.24). However, we observed a modest association among never smokers (OR, 1.59; 95% CI: 1.22–2.07), and no association among ever smokers (OR, 0.96; 95% CI: 0.83–1.11); likelihood ratio test for interaction P = 0.001.Conclusion: A history of diabetes was weakly associated with HNC overall, but we observed evidence of effect modification by smoking status, with a positive association among those who never smoked cigarettes.Impact: This study suggests that glucose metabolism abnormalities may be a HNC risk factor in subgroups of the population. Prospective studies incorporating biomarkers are needed to improve our understanding of the relationship between diabetes and HNC risk, possibly providing new strategies in the prevention of HNC. Cancer Epidemiol Biomarkers Prev; 21(2); 294–304. ©2011 AACR.

Background: The incidence of oropharyngeal and oral tongue cancers has increased over the last 20 years which parallels increased use of marijuana among individuals born after 1950.Methods: A pooled analysis was conducted comprising individual-level data from nine case–control studies from the United States and Latin America in the INHANCE consortium. Self-reported information on marijuana smoking, demographic, and behavioral factors was obtained from 1,921 oropharyngeal cases, 356 oral tongue cases, and 7,639 controls.Results: Compared with never marijuana smokers, ever marijuana smokers had an elevated risk of oropharyngeal [adjusted OR (aOR), 1.24; 95% confidence interval (CI): 1.06–1.47] and a reduced risk of oral tongue cancer (aOR, 0.47; 95% CI, 0.29, 0.75). The risk of oropharyngeal cancer remained elevated among never tobacco and alcohol users. The risk of oral tongue cancer was reduced among never users of tobacco and alcohol. Sensitivity analysis adjusting for potential confounding by HPV exposure attenuated the association of marijuana use with oropharyngeal cancer (aOR, 0.99; 95% CI, 0.71–1.25), but had no effect on the oral tongue cancer association.Conclusions: These results suggest that the association of marijuana use with head and neck carcinoma may differ by tumor site.Impact: The associations of marijuana use with oropharyngeal and oral tongue cancer are consistent with both possible pro- and anticarcinogenic effects of cannabinoids. Additional work is needed to rule out various sources of bias, including residual confounding by HPV infection and misclassification of marijuana exposure. Cancer Epidemiol Biomarkers Prev; 23(1); 160–71. ©2013 AACR.

Description for sensitivity analysis to adjust for potential unmeasured confounding

Supplementary Figure S1; Supplementary Tables S1-S4.

Supplementary Data from A Genetic Expression Profile Associated with Oral Cancer Identifies a Group of Patients at High Risk of Poor Survival

PDF file - 50K

Background: Sequence variants located at 15q25 have been associated with lung cancer and propensity to smoke. We recently reported an association between rs16969968 and risk of upper aerodigestive tract (UADT) cancers (oral cavity, oropharynx, hypopharynx, larynx, and esophagus) in women (OR = 1.24, P = 0.003) with little effect in men (OR = 1.04, P = 0.35).Methods: In a coordinated genotyping study within the International Head and Neck Cancer Epidemiology (INHANCE) consortium, we have sought to replicate these findings in an additional 4,604 cases and 6,239 controls from 10 independent UADT cancer case–control studies.Results: rs16969968 was again associated with UADT cancers in women (OR = 1.21, 95% CI = 1.08–1.36, P = 0.001) and a similar lack of observed effect in men [OR = 1.02, 95% CI = 0.95–1.09, P = 0.66; P-heterogeneity (Phet) = 0.01]. In a pooled analysis of the original and current studies, totaling 8,572 UADT cancer cases and 11,558 controls, the association was observed among females (OR = 1.22, 95% CI = 1.12–1.34, P = 7 × 10−6) but not males (OR = 1.02, 95% CI = 0.97–1.08, P = 0.35; Phet = 6 × 10−4). There was little evidence for a sexdifference in the association between this variant and cigarettes smoked per day, with male and female rs16969968 variant carriers smoking approximately the same amount more in the 11,991 ever smokers in the pooled analysis of the 14 studies (Phet = 0.86).Conclusions: This study has confirmed a sex difference in the association between the 15q25 variant rs16969968 and UADT cancers.Impact: Further research is warranted to elucidate the mechanisms underlying these observations. Cancer Epidemiol Biomarkers Prev; 20(4); 658–64. ©2011 AACR.

Canine cognitive dysfunction (CCD) is a neurodegenerative disease in aging dogs. It has been described previously in relatively small cohorts of dogs using multiple different rating scales. This study aimed to use a minimally modified CCD rating scale developed by previous researchers to describe the prevalence of CCD more thoroughly in a large, nationwide cohort of companion dogs participating in the Dog Aging Project (DAP) (n = 15,019). Associations between various canine characteristics, predicted lifespan quartiles, and CCD were examined using univariable and multivariable logistic regression models and receiver operating curve (ROC) analysis. When controlling for all other characteristics, the odds of CCD increased 52% with each additional year of age. Among dogs of the same age, health status, breed type, and sterilization status, odds of CCD were 6.47 times higher in dogs who were not active compared to those who were very active. When controlling for age, breed type, activity level, and other comorbidities, dogs with a history of neurological, eye, or ear disorders had higher odds of CCD. Lifespan quartile analysis showed excellent discriminating ability between CCD positive and negative dogs. Weight-based lifespan quartile estimation could therefore serve as a tool to inform CCD screening by veterinarians.

We aim to reduce overfitting and model overconfidence by distilling the knowledge of an ensemble of deep learning models into a single model for the classification of cancer pathology reports.We consider the text classification problem that involves 5 individual tasks. The baseline model consists of a multitask convolutional neural network (MtCNN), and the implemented ensemble (teacher) consists of 1000 MtCNNs. We performed knowledge transfer by training a single model (student) with soft labels derived through the aggregation of ensemble predictions. We evaluate performance based on accuracy and abstention rates by using softmax thresholding.The student model outperforms the baseline MtCNN in terms of abstention rates and accuracy, thereby allowing the model to be used with a larger volume of documents when deployed. The highest boost was observed for subsite and histology, for which the student model classified an additional 1.81% reports for subsite and 3.33% reports for histology.Ensemble predictions provide a useful strategy for quantifying the uncertainty inherent in labeled data and thereby enable the construction of soft labels with estimated probabilities for multiple classes for a given document. Training models with the derived soft labels reduce model confidence in difficult-to-classify documents, thereby leading to a reduction in the number of highly confident wrong predictions.Ensemble model distillation is a simple tool to reduce model overconfidence in problems with extreme class imbalance and noisy datasets. These methods can facilitate the deployment of deep learning models in high-risk domains with low computational resources where minimizing inference time is required.

Objective:\ud We analyzed the pooled case-control data from the International Head and Neck Cancer Epidemiology (INHANCE) consortium to compare cigarette smoking and alcohol consumption risk factors for head and neck cancer between less developed and more developed countries.\ud \ud Subjects and Methods:\ud The location of each study was categorized as either a less developed or more developed country. We compared the risk of overall head and neck cancer and cancer of specific anatomic subsites associated with cigarette smoking and alcohol consumption. Additionally, age and sex distribution between categories was compared.\ud \ud Results:\ud The odds ratios for head and neck cancer sites associated with smoking duration differed between less developed and more developed countries. Smoking greater than 20 years conferred a higher risk for oral cavity and laryngeal cancer in more developed countries, whereas the risk was greater for oropharynx and hypopharynx cancer in less developed countries. Alcohol consumed for more than 20 years conferred a higher risk for oropharynx, hypopharynx, and larynx cancer in less developed countries. The proportion of cases that were young (

In the last decade, the widespread adoption of electronic health record documentation has created huge opportunities for information mining. Natural language processing (NLP) techniques using machine and deep learning are becoming increasingly widespread for information extraction tasks from unstructured clinical notes. Disparities in performance when deploying machine learning models in the real world have recently received considerable attention. In the clinical NLP domain, the robustness of convolutional neural networks (CNNs) for classifying cancer pathology reports under natural distribution shifts remains understudied. In this research, we aim to quantify and improve the performance of the CNN for text classification on out-of-distribution (OOD) datasets resulting from the natural evolution of clinical text in pathology reports. We identified class imbalance due to different prevalence of cancer types as one of the sources of performance drop and analyzed the impact of previous methods for addressing class imbalance when deploying models in real-world domains. Our results show that our novel class-specialized ensemble technique outperforms other methods for the classification of rare cancer types in terms of macro F1 scores. We also found that traditional ensemble methods perform better in top classes, leading to higher micro F1 scores. Based on our findings, we formulate a series of recommendations for other ML practitioners on how to build robust models with extremely imbalanced datasets in biomedical NLP applications.

High dietary glycaemic index (GI) and glycaemic load (GL) may increase cancer risk. However, limited information was available on GI and/or GL and head and neck cancer (HNC) risk. We conducted a pooled analysis on 8 case-control studies (4081 HNC cases; 7407 controls) from the International Head and Neck Cancer Epidemiology (INHANCE) consortium. We estimated the odds ratios (ORs) and 95% confidence intervals (CIs) of HNC, and its subsites, from fixed- or mixed-effects logistic models including centre-specific quartiles of GI or GL. GI, but not GL, had a weak positive association with HNC (ORQ4 vs. Q1 = 1.16; 95% CI = 1.02–1.31). In subsites, we found a positive association between GI and laryngeal cancer (ORQ4 vs. Q1 = 1.60; 95% CI = 1.30–1.96) and an inverse association between GL and oropharyngeal cancer (ORQ4 vs. Q1 = 0.78; 95% CI = 0.63–0.97). This pooled analysis indicates a modest positive association between GI and HNC, mainly driven by laryngeal cancer.

PURPOSE SEER registries do not report results of epidermal growth factor receptor ( EGFR) and anaplastic lymphoma kinase ( ALK) mutation tests. To facilitate population-based research in molecularly defined subgroups of non–small-cell lung cancer (NSCLC), we assessed the validity of natural language processing (NLP) for the ascertainment of EGFR and ALK testing from electronic pathology (e-path) reports of NSCLC cases included in two SEER registries: the Cancer Surveillance System (CSS) and the Kentucky Cancer Registry (KCR). METHODS We obtained 4,278 e-path reports from 1,634 patients who were diagnosed with stage IV nonsquamous NSCLC from September 1, 2011, to December 31, 2013, included in CSS. We used 855 CSS reports to train NLP systems for the ascertainment of EGFR and ALK test status (reported v not reported) and test results (positive v negative). We assessed sensitivity, specificity, and positive and negative predictive values in an internal validation sample of 3,423 CSS e-path reports and repeated the analysis in an external sample of 1,041 e-path reports from 565 KCR patients. Two oncologists manually reviewed all e-path reports to generate gold-standard data sets. RESULTS NLP systems yielded internal validity metrics that ranged from 0.95 to 1.00 for EGFR and ALK test status and results in CSS e-path reports. NLP showed high internal accuracy for the ascertainment of EGFR and ALK in CSS patients—F scores of 0.95 and 0.96, respectively. In the external validation analysis, NLP yielded metrics that ranged from 0.02 to 0.96 in KCR reports and F scores of 0.70 and 0.72, respectively, in KCR patients. CONCLUSION NLP is an internally valid method for the ascertainment of EGFR and ALK test information from e-path reports available in SEER registries, but future work is necessary to increase NLP external validity.

In a population-based case control study of testicular germ cell tumors (TGCT), we reported a strong positive association between serum levels of Wolff's Group 1 (potentially estrogenic) polychlorinated biphenyl (PCBs) and risk of TGCT, and the observed associations were similar for both seminoma and non-seminoma. While the observed specific associations between TGCT and Wolff's Group 1 PCBs cannot be easily explained by bias or confounding, a question can still be asked, that is, could the relationship between PCBs and TGCT differ by age at diagnosis? PCBs tend to bioaccumulate, with more heavily chlorinated PCB congeners tending to have longer half-lives. Half-lives of PCB congeners were reported ranging from 4.6 years for PCB-28 to 41.0 years for PCB-156. The half-life for the heavy PCB congeners (17.8 years) was found to be approximately twice that for the light PCBs (9.6 years) in early studies. Therefore, the same PCB concentration measured in a 20-year-old vs. a 55-year-old is unlikely to represent the same lifetime PCB exposure or type of PCB exposure. In this analysis, we stratified the data by median age of diagnosis of TGCT and further stratified by histologic type of TGCT (seminoma vs non-seminoma) to explore if the risk of TGCT associated with PCB exposures differs by age.

PURPOSE: Systemic therapy use in the last 30 days of life (DOL) for patients with advanced cancer is a low-value medical practice. We hypothesized that systemic therapy use in the last 30 DOL increased after approval of antiprogrammed cell death protein 1 immune checkpoint inhibitors (ICIs) and has contributed to increased health care utilization and spending. METHODS: We investigated the change in prevalence of any systemic therapy use in the last 30 DOL among patients with advanced solid tumors in the 4 years before and after antiprogrammed cell death protein 1 ICI approval in 2014. We used cases from the Western Washington Cancer Surveillance System linked to commercial and Medicare insurance. We calculated the difference in prevalence between the pre- and post-ICI periods. We also calculated the annual prevalence of any systemic therapy and ICI use in the last 30 DOL and measured health care utilization (emergency department visits and hospitalizations) and costs during the last 30 DOL. RESULTS: Eight thousand eight hundred seventy-one patients (median age 73 years) were included; 34% and 66% in the pre-and post-ICI period, respectively. Systemic therapy use in the last 30 DOL was lower in the post-ICI versus pre-ICI period (12.4% v 14.4%; difference −2.0% [95% CI, −3.5 to −0.5]). The annual prevalence of systemic therapy use in the last 30 DOL also declined, although ICI use rose. Patients treated with ICIs in last 30 DOL had more emergency department visits, hospitalizations, and higher costs. CONCLUSION: Systemic therapy use in the last 30 DOL was lower in the period after ICI approval. However, ICI use rose over time and had higher utilization and costs in the last 30 DOL. Systemic therapy use in the last 30 DOL warrants monitoring, especially as more ICI indications are approved.