exKidneyBERT: a language model for kidney transplant pathology reports and the crucial role of extended vocabularies.

Background: Pathology reports contain key information about the patient's diagnosis as well as important gross and microscopic findings. These information-rich clinical reports offer an invaluable resource for clinical studies, but data extraction and analysis from such unstructured texts is often manual and tedious. While neural information retrieval systems (typically implemented as deep learning methods for natural language processing) are automatic and flexible, they typically require a large domain-specific text corpus for training, making them infeasible for many medical subdomains. Thus, an automated data extraction method for pathology reports that does not require a large training corpus would be of significant value and utility. Objective: To develop a language model-based neural information retrieval system that can be trained on small datasets and validate it by training it on renal transplant-pathology reports to extract relevant information for two predefined questions: (1) "What kind of rejection does the patient show?"; (2) "What is the grade of interstitial fibrosis and tubular atrophy (IFTA)?" Methods: Kidney BERT was developed by pre-training Clinical BERT on 3.4K renal transplant pathology reports and 1.5M words. Then, exKidneyBERT was developed by extending Clinical BERT's tokenizer with six technical keywords and repeating the pre-training procedure. This extended the model's vocabulary. All three models were fine-tuned with information retrieval heads. Results: The model with extended vocabulary, exKidneyBERT, outperformed Clinical BERT and Kidney BERT in both questions. For rejection, exKidneyBERT achieved an 83.3% overlap ratio for antibody-mediated rejection (ABMR) and 79.2% for T-cell mediated rejection (TCMR). For IFTA, exKidneyBERT had a 95.8% exact match rate. Conclusion: ExKidneyBERT is a high-performing model for extracting information from renal pathology reports. Additional pre-training of BERT language models on specialized small domains does not necessarily improve performance. Extending the BERT tokenizer's vocabulary library is essential for specialized domains to improve performance, especially when pre-training on small corpora.

Optimal 1-NN prototypes for pathological geometries.

Using prototype methods to reduce the size of training datasets can drastically reduce the computational cost of classification with instance-based learning algorithms like the k-Nearest Neighbour classifier. The number and distribution of prototypes required for the classifier to match its original performance is intimately related to the geometry of the training data. As a result, it is often difficult to find the optimal prototypes for a given dataset, and heuristic algorithms are used instead. However, we consider a particularly challenging setting where commonly used heuristic algorithms fail to find suitable prototypes and show that the optimal number of prototypes can instead be found analytically. We also propose an algorithm for finding nearly-optimal prototypes in this setting, and use it to empirically validate the theoretical results. Finally, we show that a parametric prototype generation method that normally cannot solve this pathological setting can actually find optimal prototypes when combined with the results of our theoretical analysis.

Support Vector Machines and logistic regression to predict temporal artery biopsy outcomes.

OBJECTIVE: Support vector machines (SVM) is a newer statistical method that has been reported to be advantageous to traditional logistic regression for clinical classification. We determine if SVM can better predict the results of temporal artery biopsy (TABx) for giant cell arteritis compared to logistic regression. METHOD: A database of 530 TABx patients with 10 covariates was used and randomly split into training and test sets. The area under the receiving operating curve (AUC), misclassification rate (MCR), and false negative rate (FN) were compared for SVM and logistic regression. AUC and MCR were used to tune the SVM. RESULTS: The SVM model with optimal AUC had gamma = 0.01267 and cost = 26.466, with 133 support vectors. The AUC/MCR/FN for logistic regression and SVM respectively were 0.827/0.184/0.524 and 0.825/0.168/0.571. CONCLUSION: In our dataset of 530 TABx subjects, SVM did not offer any distinct advantage over the logistic regression prediction model.

The k conditional nearest neighbor algorithm for classification and class probability estimation.

The k nearest neighbor (kNN) approach is a simple and effective nonparametric algorithm for classification. One of the drawbacks of kNN is that the method can only give coarse estimates of class probabilities, particularly for low values of k. To avoid this drawback, we propose a new nonparametric classification method based on nearest neighbors conditional on each class: the proposed approach calculates the distance between a new instance and the kth nearest neighbor from each class, estimates posterior probabilities of class memberships using the distances, and assigns the instance to the class with the largest posterior. We prove that the proposed approach converges to the Bayes classifier as the size of the training data increases. Further, we extend the proposed approach to an ensemble method. Experiments on benchmark data sets show that both the proposed approach and the ensemble version of the proposed approach on average outperform kNN, weighted kNN, probabilistic kNN and two similar algorithms (LMkNN and MLM-kHNN) in terms of the error rate. A simulation shows that kCNN may be useful for estimating posterior probabilities when the class distributions overlap.

Pay attention and you won't lose it: a deep learning approach to sequence imputation.

In most areas of machine learning, it is assumed that data quality is fairly consistent between training and inference. Unfortunately, in real systems, data are plagued by noise, loss, and various other quality reducing factors. While a number of deep learning algorithms solve end-stage problems of prediction and classification, very few aim to solve the intermediate problems of data pre-processing, cleaning, and restoration. Long Short-Term Memory (LSTM) networks have previously been proposed as a solution for data restoration, but they suffer from a major bottleneck: a large number of sequential operations. We propose using attention mechanisms to entirely replace the recurrent components of these data-restoration networks. We demonstrate that such an approach leads to reduced model sizes by as many as two orders of magnitude, a 2-fold to 4-fold reduction in training times, and 95% accuracy for automotive data restoration. We also show in a case study that this approach improves the performance of downstream algorithms reliant on clean data.

Nearest labelset using double distances for multi-label classification.

Multi-label classification is a type of supervised learning where an instance may belong to multiple labels simultaneously. Predicting each label independently has been criticized for not exploiting any correlation between labels. In this article we propose a novel approach, Nearest Labelset using Double Distances (NLDD), that predicts the labelset observed in the training data that minimizes a weighted sum of the distances in both the feature space and the label space to the new instance. The weights specify the relative tradeoff between the two distances. The weights are estimated from a binomial regression of the number of misclassified labels as a function of the two distances. Model parameters are estimated by maximum likelihood. NLDD only considers labelsets observed in the training data, thus implicitly taking into account label dependencies. Experiments on benchmark multi-label data sets show that the proposed method on average outperforms other well-known approaches in terms of 0/1 loss, and multi-label accuracy and ranks second on the F-measure (after a method called ECC) and on Hamming loss (after a method called RF-PCT).

Which literacy skills are associated with smoking?

BACKGROUND: Research has demonstrated associations between smoking and reading skills, but other literacy skills such as speaking, listening and numeracy are less studied despite our dependence on the use of numbers and the oral exchange to deliver information on the risks of smoking. METHODS: The authors used multivariable logistic regression to examine the effects of reading, numeracy, speaking and listening skills on: (1) becoming a regular smoker and (2) smoking cessation. Further, multivariable linear regression was used to examine the relation between literacy skills and amount smoked among current smokers. Models controlled for education, gender, age, race/ethnicity, income and, when relevant, age at which they became a regular smoker. RESULTS: For each grade equivalent increase in reading skills, the odds of quitting smoking increased by about 8% (OR=1.08, 95% CI 1.01 to 1.15). For every point increase in numeracy skills, the odds of quitting increased by about 24% (OR=1.24, 95% CI 1.06 to 1.46). No literacy skills were associated with becoming a regular smoker or current amount smoked. CONCLUSION: The ability to locate, understand and use information related to the risks of smoking may impact one's decision to quit. Messaging should be designed with the goal of being easily understood by all individuals regardless of literacy level.

Patients' literacy skills: more than just reading ability.

Limited literacy contributes to suboptimal care and outcomes for patients. The Institute of Medicine noted that future work in health literacy should consider multiple literacy skills. However, lacking empirical evidence of the relationship between different literacy skills, reading skills are often used as proxies of literacy in research and practice. Using a community-based sample of 618 individuals residing in Boston, Massachusetts, and Providence, Rhode Island, the authors conducted a principal component analysis on measures of four literacy skills--reading, numeracy, oral (speaking), and aural (listening)--to examine whether and to what extent literacy can, or should, be represented by a single measure. The first principal component represented overall literacy and could only explain 60% of the total variation in literacy skills among individuals. The second principal component differentiated between numeracy/reading and the oral/aural exchange. While reading and numeracy best represent overall literacy, patients' relative strengths may vary. Those with moderate reading ability may have high oral and aural language skills. Conversely, people who have difficulties speaking with or understanding a provider may read well. Effective communication with patients should rely on the oral exchange and written health information, and not rely on a single literacy skill.

Patient activation and advocacy: which literacy skills matter most?

Attention to the effect of a patient's literacy skills on health care interactions is relatively new. So, too, are studies of either structural or personal factors that inhibit or support a patient's ability to navigate health services and systems and to advocate for their own needs within a service delivery system. Contributions of the structural environment, of interpersonal dynamics, and of a variety of psychological and sociological factors in the relationship between patients and providers have long been under study. Less frequently examined is the advocacy role expected of patients. However, the complex nature of health care in the United States increasingly requires a proactive stance. This study examined whether four literacy skills (reading, numeracy, speaking, and listening) were associated with patient self-advocacy--a component of health literacy itself--when faced with a hypothetical barrier to scheduling a medical appointment. Although all literacy skills were significantly associated with advocacy when examined in isolation, greater speaking and listening skills remained significantly associated with better patient advocacy when all four skills were examined simultaneously. These findings suggest that speaking and listening skills and support for such skills may be important factors to consider when developing patient activation and advocacy skills.

Literacy skills and calculated 10-year risk of coronary heart disease.

BACKGROUND: Coronary heart disease (CHD) is a leading cause of morbidity and mortality. Reducing the disease burden requires an understanding of factors associated with the prevention and management of CHD. Literacy skills may be one such factor. OBJECTIVES: To examine the independent and interactive effects of four literacy skills: reading, numeracy, oral language (speaking) and aural language (listening) on calculated 10-year risk of CHD and to determine whether the relationships between literacy skills and CHD risk were similar for men and women. DESIGN: We used multivariable linear regression to assess the individual, combined, and interactive effects of the four literacy skills on risk of CHD, adjusting for education and race. PARTICIPANTS: Four hundred and nine English-speaking adults in Boston, MA and Providence, RI. MEASURES: Ten-year risk of coronary heart disease was calculated using the Framingham algorithm. Reading, oral language and aural language were measured using the Woodcock Johnson III Tests of Achievement. Numeracy was assessed through a modified version of the numeracy scale by Lipkus and colleagues. KEY RESULTS: When examined individually, reading (p = 0.007), numeracy (p = 0.001) and aural language (p = 0.004) skills were significantly associated with CHD risk among women; no literacy skills were associated with CHD risk in men. When examined together, there was some evidence for an interaction between numeracy and aural language among women suggesting that higher skills in one area (e.g., aural language) may compensate for difficulties in another resulting in an equally low risk of CHD. CONCLUSIONS: Results of this study not only provide important insight into the independent and interactive effects of literacy skills on risk of CHD, they also highlight the need for the development of easy-to use assessments of the oral exchange in the health care setting and the need to better understand which literacy skills are most important for a given health outcome.

Impact of small group size on neighbourhood influences in multilevel models.

BACKGROUND: Given the growing availability of multilevel data from national surveys, researchers interested in contextual effects may find themselves with a small number of individuals per group. Although there is a growing body of literature on sample size in multilevel modelling, few have explored the impact of group sizes of less than five. METHODS: In a simulated analysis of real data, the impact of a group size of less than five was examined on both a continuous and dichotomous outcome in a simple two-level multilevel model. Models with group sizes one to five were compared with models with complete data. Four different linear and logistic models were examined: empty models; models with a group-level covariate; models with an individual-level covariate and models with an aggregated group-level covariate. The study evaluated further whether the impact of small group size differed depending on the total number of groups. RESULTS: When the number of groups was large (N=459), neither fixed nor random components were affected by small group size, even when 90% of tracts had only one individual per tract and even when an aggregated group-level covariate was examined. As the number of groups decreased, the SE estimates of both fixed and random effects were inflated. Furthermore, group-level variance estimates were more affected than were fixed components. CONCLUSIONS: Datasets in which there is a small to moderate number of groups, with the majority of very small group size (n<5), size may fail to find or even consider a group-level effect when one may exist and also may be underpowered to detect fixed effects.

Children's mental health care following Hurricane Katrina: a field trial of trauma-focused psychotherapies.

New Orleans school children participated in an assessment and field trial of two interventions 15 months after Hurricane Katrina. Children (N = 195) reported on hurricane exposure, lifetime trauma exposure, peer and parent support, posttraumatic stress disorder (PTSD), and depressive symptoms. Teachers reported on behavior. At baseline, 60.5% screened positive for PTSD symptoms and were offered a group intervention at school or individual treatment at a mental health clinic. Uptake of the mental health care was uneven across intervention groups, with 98% beginning the school intervention, compared to 37% beginning at the clinic. Both treatments led to significant symptom reduction of PTSD symptoms, but many still had elevated PTSD symptoms at posttreatment. Implications for future postdisaster mental health work are discussed.

Disparities in the treatment of patients with IL-2 for metastatic renal cell carcinoma.

OBJECTIVES: The incidence of metastatic renal cell cancer (mRCC) is rising. To date, interleukin-2 (IL-2) is the only treatment offering a complete response rate for mRCC. We wish to test the hypothesis that the combination of restricted availability and expense associated with IL-2 administration results in differential access to the medication based on race and sex, despite similar clinical indications for its use. METHODS: We used data from the Surveillance, Epidemiology, and End Results program and the Centers for Medicare Services (CMS) to clinically characterize subjects with mRCC diagnosed from 1992 through 2002. We linked these subjects to claims identified in the CMS databases. We then assigned subjects to cohorts receiving radical nephrectomy, IL-2, both, or neither. A logistic model was created to identify factors that had significant independent effects on the receipt of IL-2. RESULTS: Three thousand seven hundred thirty individuals were identified with mRCC. After controlling for other variables, female subjects were less likely to receive IL-2 (O.R. 0.80). African American subjects were also less likely to receive IL-2 (O.R 0.55). Married individuals were much more likely to receive IL-2 (O.R 1.9). CONCLUSIONS: African Americans and women were much less likely to be treated with IL-2 after controlling for relevant clinical variables. These data document that the only therapy offering a complete response to patients with mRCC is less frequently given to those who are African American or female. It is possible that the racial and gender-based disparities in treatment with IL-2 will be replicated with newer, expensive treatment options for mRCC. Further prospective investigation into mitigating barriers to receipt of effective care for mRCC is urgently needed.

The neighborhood alcohol environment and alcohol-related morbidity.

AIMS: The aims of this study were (1) to examine the association between neighborhood alcohol outlet density and individual self-reported alcohol-related health outcomes in the last year-sexually transmitted infections (STI), motor vehicle accidents, injury, liver problems, hypertension and experienced violence; (2) to determine whether the relationship between morbidity and alcohol outlet density is mediated by individual alcohol consumption; and (3) to explore the role of alcohol outlet density in explaining any observed racial and ethnic differences in morbidity. METHOD: Hierarchical models from a random sample of Los Angeles, CA, and Louisiana residents (N = 2881) from 217 census tracts were utilized. The clustering of health and social outcomes according to neighborhood varied by health problem examined. RESULTS: There was substantial clustering of STI (intraclass correlation coefficient, ICC = 12.8%) and experienced violence (ICC = 13.0%); moderate clustering of liver problems (ICC = 3.5%) and hypertension (ICC = 3.9%); and low clustering of motor vehicle accident (ICC = 1.2%) and injury (ICC = 1.4%). Alcohol outlet density was significantly and positively associated with STI (crude OR = 1.80, 95% CI = 1.10-3.00), liver problems (crude OR = 1.33, 95% CI = 1.02-1.75) and experienced violence (crude OR = 1.31, 95% CI = 1.13-1.51) although not with other morbidity outcomes. Mediation analyses of morbidity outcomes revealed partial mediation of individual alcohol consumption in the relationship between alcohol density and STI and violence, and full mediation for liver problems. CONCLUSIONS: Findings support the concept that off-premise alcohol outlets in the neighborhood environment may impact health and social outcomes, either directly or indirectly, through individual alcohol consumption and these associations may be heterogeneous with respect to race and ethnicity.

Off-label prescribing to children in the United States outpatient setting.

OBJECTIVE: The aim of this study was to determine the frequency of off-label prescribing to children at United States outpatient visits and to determine how drug class, patient age, and physician specialty relate to off-label prescribing. METHODS: Data from the 2001 through 2004 National Ambulatory Medical Care Surveys (NAMCS) consisted of a sample of 7901 outpatient visits by children aged 0 through 17 years in which prescriptions were given, representative of an estimated 312 million visits. We compared FDA-approved age and indication to the child's age and diagnoses. We used multivariate logistic regression to determine adjusted differences in probabilities of off-label prescribing. RESULTS: Sixty-two percent of outpatient pediatric visits included off-label prescribing. Approximately 96% of cardiovascular-renal, 86% of pain, 80% of gastrointestinal, and 67% of pulmonary and dermatologic medication prescriptions were off label. Visits by children aged <6 years had a higher probability of off-label prescribing (P < .01), especially visits by children aged <1 year (74% adjusted probability). Visits to specialists also involved a significantly increased probability (68% vs 59% for general pediatricians, P < .01) of off-label prescribing. CONCLUSIONS: Despite recent studies and labeling changes of pediatric medications, the majority of pediatric outpatient visits involve off-label prescribing across all medication categories. Off-label prescribing is more frequent for younger children and those receiving care from specialist pediatricians. Increased dissemination of pediatric studies and label information may be helpful to guide clinical practice. Further research should be prioritized for the medications most commonly prescribed off label and to determine outcomes, causes, and appropriateness of off-label prescribing to children.

Social capital and the neighborhood alcohol environment.

We examine whether neighborhood alcohol outlet density is associated with reduced social capital and whether this relationship is mediated by perceived neighborhood safety. Hierarchical models from a random sample of Los Angeles, CA, and Louisiana residents (N=2,881) from 217 census tracts were utilized. Substantial proportions of the variance in collective efficacy (intraclass correlation coefficient, ICC=16.3%) and organizational participation (ICC=13.8%, median odds ratio=1.99) were attributable to differences between neighborhoods-suggesting that these factors may be influenced by neighborhood-level characteristics. Neighborhood alcohol outlet density was strongly associated with reduced indicators of social capital, and the relationship between collective efficacy and outlet density appears to be mediated by perceived neighborhood safety. Findings support the concept that off-premise alcohol outlets in the neighborhood environment may hinder the development of social capital, possibly through decreased positive social network expansion.

Surveying a nationally representative internet-based panel to obtain timely estimates of influenza vaccination rates.

We studied the feasibility of using an internet-based panel survey to obtain timely and accurate population-based data on influenza vaccination. We surveyed a nationally representative sample of US adults (n=3043) via the internet about use of influenza vaccination during the 2007-8 influenza vaccination season. We compared the internet-based rates to those from the 2004 and 2008 National Health Interview Surveys (NHIS). The internet-based rates were comparable to those from the NHIS and were obtained in less than six weeks following the end of influenza vaccination season. We conclude that an internet-based approach can yield accurate estimates of end-of-season influenza vaccination rates in time to support improved management of the subsequent season.

Support for Students Exposed to Trauma: A Pilot Study.

With high rates of trauma exposure among students, the need for intervention programs is clear. Delivery of such programs in the school setting eliminates key barriers to access, but there are few programs that demonstrate efficacy in this setting. Programs to date have been designed for delivery by clinicians, who are a scarce resource in many schools. This study describes preliminary feasibility and acceptability data from a pilot study of a new program, Support for Students Exposed to Trauma, adapted from the Cognitive Behavioral Intervention for Trauma in Schools (CBITS) program. Because of its "pilot" nature, all results from the study should be viewed as preliminary. Results show that the program can be implemented successfully by teachers and school counselors, with good satisfaction among students and parents. Pilot data show small reductions in symptoms among the students in the SSET program, suggesting that this program shows promise that warrants a full evaluation of effectiveness.

Alcohol outlet density and alcohol consumption in Los Angeles county and southern Louisiana.

The objective of this study was to assess the relationship between alcohol availability, as measured by the density of off-premise alcohol outlets, and alcohol consumption in Los Angeles county and southern Louisiana, USA. Consumption information was collected through a telephone survey of 2,881 households in Los Angeles county and pre-Katrina southern Louisiana, nested within 220 census tracts. Respondents' addresses were geo-coded and both neighbourhood (census tracts and buffers of varying sizes) and individual (network distance to the closest alcohol outlet) estimates of off-sale alcohol outlet density were computed. Alcohol outlet density was not associated with the percentage of people who were drinkers in either site. Alcohol outlet density was associated with the quantity of consumption among drinkers in Louisiana but not in Los Angeles. Outlet density within a one-mile buffer of the respondent's home was more strongly associated with alcohol consumption than outlet density in the respondent's census tract. The conclusion is that the relationship between neighbourhood alcohol outlet density and alcohol consumption is complex and may vary due to differences in neighbourhood design and travel patterns.