Local Large Language Models for Complex Structured Tasks.

This paper introduces an approach that combines the language reasoning capabilities of large language models (LLMs) with the benefits of local training to tackle complex language tasks. The authors demonstrate their approach by extracting structured condition codes from pathology reports. The proposed approach utilizes local, fine-tuned LLMs to respond to specific generative instructions and provide structured outputs. Over 150k uncurated surgical pathology reports containing gross descriptions, final diagnoses, and condition codes were used. Different model architectures were trained and evaluated, including LLaMA, BERT, and LongFormer. The results show that the LLaMA-based models significantly outperform BERT-style models across all evaluated metrics. LLaMA models performed especially well with large datasets, demonstrating their ability to handle complex, multi-label tasks. Overall, this work presents an effective approach for utilizing LLMs to perform structured generative tasks on domain-specific language in the medical domain.

CLASSify: A Web-Based Tool for Machine Learning.

Machine learning classification problems are widespread in bioinformatics, but the technical knowledge required to perform model training, optimization, and inference can prevent researchers from utilizing this technology. This article presents an automated tool for machine learning classification problems to simplify the process of training models and producing results while providing informative visualizations and insights into the data. This tool supports both binary and multiclass classification problems, and it provides access to a variety of models and methods. Synthetic data can be generated within the interface to fill missing values, balance class labels, or generate entirely new datasets. It also provides support for feature evaluation and generates explainability scores to indicate which features influence the output the most. We present CLASSify, an open-source tool for simplifying the user experience of solving classification problems without the need for knowledge of machine learning.

Use of residential opioid use disorder treatment among Medicaid enrollees in nine states.

INTRODUCTION: Residential treatment is a key component of the opioid use disorder care continuum, but research has not measured well the differences in its use across states at the enrollee level. METHODS: This cross-sectional observational study used Medicaid claims data from nine states to document the prevalence of residential treatment for opioid use disorder and to describe the characteristics of patients receiving care. For each patient characteristic, chi-square and t-tests tested for differences in the distribution between individuals who did and did not receive residential care. RESULTS: Among 491,071 Medicaid enrollees with opioid use disorder, 7.5 % were treated in residential facilities in 2019, though this number ranged widely (0.3-14.6 %) across states. Residential patients were more likely to be younger, non-Hispanic White, male, and living in an urban area. Although residential patients were less likely than those without residential care to be eligible for Medicaid through disability, diagnoses for comorbid conditions were more frequently observed among residential patients. CONCLUSIONS: Results from this large, multi-state study add context to the ongoing national conversation around opioid use disorder treatment and policy, providing a baseline for future work.

Outpatient follow-up and use of medications for opioid use disorder after residential treatment among Medicaid enrollees in 10 states.

BACKGROUND: Follow-up after residential treatment is considered best practice in supporting patients with opioid use disorder (OUD) in their recovery. Yet, little is known about rates of follow-up after discharge. The objective of this analysis was to measure rates of follow-up and use of medications for OUD (MOUD) after residential treatment among Medicaid enrollees in 10 states, and to understand the enrollee and episode characteristics that are associated with both outcomes. METHODS: Using a distributed research network to analyze Medicaid claims data, we estimated the likelihood of 4 outcomes occurring within 7 and 30 days post-discharge from residential treatment for OUD using multinomial logit regression: no follow-up or MOUD, follow-up visit only, MOUD only, or both follow-up and MOUD. We used meta-analysis techniques to pool state-specific estimates into global estimates. RESULTS: We identified 90,639 episodes of residential treatment for OUD for 69,017 enrollees from 2018 to 2019. We found that 62.5% and 46.9% of episodes did not receive any follow-up or MOUD at 7 days and 30 days, respectively. In adjusted analyses, co-occurring mental health conditions, longer lengths of stay, prior receipt of MOUD or behavioral health counseling, and a recent ED visit for OUD were associated with a greater likelihood of receiving follow-up treatment including MOUD after discharge. CONCLUSIONS: Forty-seven percent of residential treatment episodes for Medicaid enrollees are not followed by an outpatient visit or MOUD, and thus are not following best practices.

Trends in Gabapentin Prescribing in a Commercially Insured U.S. Adult Population, 2009-2016.

BACKGROUND: Gabapentin is prescribed for a variety of conditions and is often used off label. It is important to understand the prevalence of gabapentin prescribing and the characteristics of individuals who are prescribed gabapentin, given increasing concern regarding its potential for misuse. OBJECTIVES: To (a) examine state- and region-level prevalence and trends in gabapentin prescribing from 2009 to 2016 and (b) characterize demographic and clinical characteristics of individuals prescribed gabapentin in a nationwide population of commercially insured adults. METHODS: This retrospective, longitudinal study examined trends in gabapentin prescribing from 2009 to 2016. The study population included individuals aged 18-64 years who were enrolled in a commercial insurance plan at any point from 2009 to 2016. Individuals who were prescribed gabapentin were defined as beneficiaries with at least 1 gabapentin prescription claim in a calendar year (CY). A cross-sectional descriptive analysis was performed to examine differences in demographic and clinical characteristics of individuals prescribed or not prescribed gabapentin in CY 2016. RESULTS: The prevalence of gabapentin prescribing nearly doubled from 2009 to 2016. During this time, gabapentin prescribing increased in every state (range: 44%-179%). State-specific prevalence rates in 2016 varied from 12.7 to 43.9 per 1,000 beneficiaries. Overall, 2.7% of beneficiaries filled ≥ 1 gabapentin prescription in 2016. Individuals prescribed gabapentin were more likely to fill opioid prescriptions (60.8% vs. 16.5%, P < 0.01); reside in the South (53.7% vs. 47%, P < 0.01); be female (62.5% vs. 52.3%, P < 0.01); and be aged 55-64 years (41.7% vs. 21.2%, P < 0.01) compared with the comparator. Individuals who were prescribed gabapentin also had significantly higher rates of seizure disorders, neuropathic pain, mental health disorders, substance use disorders, and diabetes. CONCLUSIONS: The prevalence of gabapentin prescribing among a U.S. privately insured population has increased steadily in recent years. Additional research should examine coprescribing of gabapentin in the context of the opioid epidemic. DISCLOSURES: The project described in this study was supported by the NIH National Center for Advancing Translational Sciences through grant number UL1TR001998. This study was also partially supported by grant number 2017-PM-BX-K026 (Data-Driven Responses to Prescription Drug Misuse in Kentucky) awarded by the Bureau of Justice Assistance. Viewpoints or opinions in this document are those of the authors and do not necessarily represent the official position or policies of the U.S. Department of Justice or the official views of the NIH. The authors do not have any conflicts of interest to report. Portions of this study have been previously presented in poster presentations at the 2019 Academy Health Annual Research Meeting; June 2-4, 2019; Washington, DC, and the 2019 University of Kentucky Substance Use Research Day; March 3, 2019; Lexington, KY.

Association of time to antibiotics and clinical outcomes in adult hematologic malignancy patients with febrile neutropenia.

Objective The objective of this study was to determine the clinical impact of time to antibiotic administration in adult inpatients who have hematologic malignancies and develop febrile neutropenia. Methods A retrospective chart review was conducted to screen for all febrile neutropenia events amongst adult hematologic malignancy patients between 1 January 2010 and 1 September 2014. All included patients were admitted to the hospital at the time of fever onset, having been admitted for a diagnosis other than febrile neutropenia. Descriptive statistics and logistic generalized estimated equations were used to analyze the data. Results Two hundred forty-four neutropenic fever events met inclusion criteria. Thirty-five events (14.34%) led to negative clinical outcomes (in-hospital mortality, intensive care unit transfer, or vasopressor requirement), with an in-house mortality rate of 7.4%. The time to antibiotics ranged from 10 min to 1495 min. The median time to antibiotics in the events that led to negative outcomes was 120 min compared to 102 min in the events that did not lead to the negative outcome ( p = 0.93). Conditional order sets were used to order empiric antibiotics in 176 events (72.1%) and significantly reduced time to antibiotics from 287 min to 143 min ( p = 0.0019). Conclusion Prolonged time to antibiotic administration in hematologic malignancy patients who develop neutropenic fever was not shown to be associated with negative clinical outcomes.

The influence of antidiabetic medications on the development and progression of prostate cancer.

BACKGROUND: The development of prostate tumors has been linked to co-morbid diabetes mellitus (DM) in several studies, potentially through the stimulation of insulin-like growth factor receptor (IGFR). This study evaluates the effect of anti-diabetic medication use on the development of high grade tumors and time to tumor progression compared to non-diabetics. METHODS: This retrospective, nested case control study identified patients with prostate cancer (PCa) from the Kentucky Medicaid Database. Cases were diagnosed with PCa and DM and using at least one of the following antidiabetic medications; sulfonylureas, insulin, metformin or TZDs. Cases were further stratified on their insulin exposure resulting from therapy. Controls were those with PCa without DM or any anti-diabetic medications. RESULTS: The use of metformin or TZDs trended toward decreased odds of high-grade tumors and decreased risk of progression, while sulfonylureas and high-dose insulin tended toward an increased odds of high-grade tumors and increase the risk of progression compared to non-diabetics. CONCLUSIONS: Future studies should be conducted to further evaluate the effects of anti-diabetic medications on tumor grade and time to prostate cancer progression.

Incidence and risk factors of clinically significant chemotherapy-induced thrombocytopenia in patients with solid tumors.

UNLABELLED: PURPOSE AND RELEVANCE: Chemotherapy-induced thrombocytopenia (CIT) can be a significant problem in patients with cancer, leading to numerous clinical complications. Understanding the types of patients at risk for these complications is essential to improve monitoring, counseling, and provide future targeted prophylaxis measures. Previous studies have limited prospective utility since they do not examine risk factors associated with complications from multi-agent regimens. This evaluation aims to identify the incidence and risk factors associated with clinical complications of CIT in patients receiving common chemotherapy regimens. METHODS: Retrospective evaluation of adult patients receiving first or second line regimens for the most common solid tumors associated with high rates (≥5%) of laboratory diagnosed thrombocytopenia. Patients were examined for clinically significant CIT (defined as platelet count <75,000 cells/µL as well as the presence of one of the following: bleeding, dose reduction/delay, platelet transfusion, or therapy cessation) and associated risk factors. RESULTS: About 254 patients receiving a total of 278 regimens were evaluated. The incidence of clinically significant CIT=10.1%; complications were most common in patients receiving cisplatin/gemcitabine for bladder cancer (57%), or carboplatin/gemcitabine (29%) or cisplatin/etoposide (18%) for lung cancer. Bladder cancer (OR=13.7 (2.89-64.7); p=0.001) and concurrent or recent infection (OR=3.8 (1.45-10.1); p=0.007) was found to increase the risk of clinical complications while smoking was found to have a protective effect (OR=0.17 (0.04-0.71)). CONCLUSIONS: The incidence of clinically significant CIT is most commonly seen in patients using cisplatin/gemcitabine for bladder cancer, or carboplatin/gemcitabine or cisplatin/etoposide for lung cancer. Further evaluation of these patients is warranted.

Community readiness for local smoke-free policy change.

PURPOSE: The purpose of this study was to use the Community Readiness Model to examine local smoke-free policy development. DESIGN, SETTING, AND SUBJECTS: A descriptive, cross-sectional design was used to assess 64 Kentucky communities. Dimensions of readiness included a community's knowledge of the problem and existing voluntary smoke-free policies; leadership for policy development; resources for policy development; climate surrounding policy development; existing voluntary policy efforts; and political climate for policy development. Dimension scores were summed to identify one of six overall readiness stages: (1) unawareness; (2) vague awareness; (3) preplanning; (4) preparation; (5) initiation; and (6) endorsement. ANALYSES: Correlations between dimensions and overall readiness scores were evaluated. One-way analysis of variance was used to evaluate regional trends, and multiple regression was used to assess the influence of sociodemographic/political variables on policy readiness. RESULTS: The knowledge dimension rated highest, and community climate rated lowest. Most communities were in the lower stages of readiness. No relationship was found between overall readiness and region (F [4,59] = 1.17; p > .05); nor were there regional differences among dimension scores. Smaller communities were less ready for local policy development than larger ones (adjusted R2 = .25; p = .003). CONCLUSIONS: The Community Readiness Model is appropriate for understanding local policy development, and it provides advocates with information that may prove helpful in advancing smoke-free policy.