PROVIDENT: Development and Validation of a Machine Learning Model to Predict Neighborhood-level Overdose Risk in Rhode Island.

BACKGROUND: Drug overdose persists as a leading cause of death in the United States, but resources to address it remain limited. As a result, health authorities must consider where to allocate scarce resources within their jurisdictions. Machine learning offers a strategy to identify areas with increased future overdose risk to proactively allocate overdose prevention resources. This modeling study is embedded in a randomized trial to measure the effect of proactive resource allocation on statewide overdose rates in Rhode Island (RI). METHODS: We used statewide data from RI from 2016 to 2020 to develop an ensemble machine learning model predicting neighborhood-level fatal overdose risk. Our ensemble model integrated gradient boosting machine and super learner base models in a moving window framework to make predictions in 6-month intervals. Our performance target, developed a priori with the RI Department of Health, was to identify the 20% of RI neighborhoods containing at least 40% of statewide overdose deaths, including at least one neighborhood per municipality. The model was validated after trial launch. RESULTS: Our model selected priority neighborhoods capturing 40.2% of statewide overdose deaths during the test periods and 44.1% of statewide overdose deaths during validation periods. Our ensemble outperformed the base models during the test periods and performed comparably to the best-performing base model during the validation periods. CONCLUSIONS: We demonstrated the capacity for machine learning models to predict neighborhood-level fatal overdose risk to a degree of accuracy suitable for practitioners. Jurisdictions may consider predictive modeling as a tool to guide allocation of scarce resources.

Translating Predictive Analytics for Public Health Practice: A Case Study of Overdose Prevention in Rhode Island.

Prior applications of machine learning to population health have relied on conventional model assessment criteria, limiting the utility of models as decision support tools for public health practitioners. To facilitate practitioners' use of machine learning as a decision support tool for area-level intervention, we developed and applied 4 practice-based predictive model evaluation criteria (implementation capacity, preventive potential, health equity, and jurisdictional practicalities). We used a case study of overdose prevention in Rhode Island to illustrate how these criteria could inform public health practice and health equity promotion. We used Rhode Island overdose mortality records from January 2016-June 2020 (n = 1,408) and neighborhood-level US Census data. We employed 2 disparate machine learning models, Gaussian process and random forest, to illustrate the comparative utility of our criteria to guide interventions. Our models predicted 7.5%-36.4% of overdose deaths during the test period, illustrating the preventive potential of overdose interventions assuming 5%-20% statewide implementation capacities for neighborhood-level resource deployment. We describe the health equity implications of use of predictive modeling to guide interventions along the lines of urbanicity, racial/ethnic composition, and poverty. We then discuss considerations to complement predictive model evaluation criteria and inform the prevention and mitigation of spatially dynamic public health problems across the breadth of practice. This article is part of a Special Collection on Mental Health.

Neighborhood-Level Risk Factors for Severe Hyperglycemia among Emergency Department Patients without a Prior Diabetes Diagnosis.

A person's place of residence is a strong risk factor for important diagnosed chronic diseases such as diabetes. It is unclear whether neighborhood-level risk factors also predict the probability of undiagnosed disease. The objective of this study was to identify neighborhood-level variables associated with severe hyperglycemia among emergency department (ED) patients without a history of diabetes. We analyzed patients without previously diagnosed diabetes for whom a random serum glucose value was obtained in the ED. We defined random glucose values ≥ 200 mg/dL as severe hyperglycemia, indicating probable undiagnosed diabetes. Patient addresses were geocoded and matched with neighborhood-level socioeconomic measures from the American Community Survey and claims-based surveillance estimates of diabetes prevalence. Neighborhood-level exposure variables were standardized based on z-scores, and a series of logistic regression models were used to assess the association of selected exposures and hyperglycemia adjusting for biological and social individual-level risk factors for diabetes. Of 77,882 ED patients without a history of diabetes presenting in 2021, 1,715 (2.2%) had severe hyperglycemia. Many geospatial exposures were associated with uncontrolled hyperglycemia, even after controlling for individual-level risk factors. The most strongly associated neighborhood-level variables included lower markers of educational attainment, higher percentage of households where limited English is spoken, lower rates of white-collar employment, and higher rates of Medicaid insurance. Including these geospatial factors in risk assessment models may help identify important subgroups of patients with undiagnosed disease.

Identifying Predictors of Opioid Overdose Death at a Neighborhood Level With Machine Learning.

Predictors of opioid overdose death in neighborhoods are important to identify, both to understand characteristics of high-risk areas and to prioritize limited prevention and intervention resources. Machine learning methods could serve as a valuable tool for identifying neighborhood-level predictors. We examined statewide data on opioid overdose death from Rhode Island (log-transformed rates for 2016-2019) and 203 covariates from the American Community Survey for 742 US Census block groups. The analysis included a least absolute shrinkage and selection operator (LASSO) algorithm followed by variable importance rankings from a random forest algorithm. We employed double cross-validation, with 10 folds in the inner loop to train the model and 4 outer folds to assess predictive performance. The ranked variables included a range of dimensions of socioeconomic status, including education, income and wealth, residential stability, race/ethnicity, social isolation, and occupational status. The R2 value of the model on testing data was 0.17. While many predictors of overdose death were in established domains (education, income, occupation), we also identified novel domains (residential stability, racial/ethnic distribution, and social isolation). Predictive modeling with machine learning can identify new neighborhood-level predictors of overdose in the continually evolving opioid epidemic and anticipate the neighborhoods at high risk of overdose mortality.

Discovering anomalous patterns in large digital pathology images.

Advances in medical imaging technology have created opportunities for computer-aided diagnostic tools to assist human practitioners in identifying relevant patterns in massive, multiscale digital pathology slides. This work presents Hierarchical Linear Time Subset Scanning, a novel statistical method for pattern detection. Hierarchical Linear Time Subset Scanning exploits the hierarchical structure inherent in data produced through virtual microscopy in order to accurately and quickly identify regions of interest for pathologists to review. We take a digital image at various resolution levels, identify the most anomalous regions at a coarse level, and continue to analyze the data at increasingly granular resolutions until we accurately identify its most anomalous subregions. We demonstrate the performance of our novel method in identifying cancerous locations on digital slides of prostate biopsy samples and show that our methods detect regions of cancer in minutes with high accuracy, both as measured by the ROC curve (measuring ability to distinguish between benign and cancerous slides) and by the spatial precision-recall curve (measuring ability to pick out the malignant areas on a slide which contains cancer). Existing methods need small scale images (small areas of a slide preselected by the pathologist for analysis, eg, 32 × 32 pixels) and may not work effectively on large, raw digitized images of size 100K × 100K pixels. In this work, we provide a methodology to fill this significant gap by analyzing large digitized images and identifying regions of interest that may be indicative of cancer.

Housing-Sensitive Health Conditions Can Predict Poor-Quality Housing.

Improving housing quality may improve residents' health, but identifying buildings in poor repair is challenging. We developed a method to improve health-related building inspection targeting. Linking New York City Medicaid claims data to Landlord Watchlist data, we used machine learning to identify housing-sensitive health conditions correlated with a building's presence on the Watchlist. We identified twenty-three specific housing-sensitive health conditions in five broad categories consistent with the existing literature on housing and health. We used these results to generate a housing health index from building-level claims data that can be used to rank buildings by the likelihood that their poor quality is affecting residents' health. We found that buildings in the highest decile of the housing health index (controlling for building size, community district, and subsidization status) scored worse across a variety of housing quality indicators, validating our approach. We discuss how the housing health index could be used by local governments to target building inspections with a focus on improving health.

Fast subset scan for multivariate event detection.

We present new subset scan methods for multivariate event detection in massive space-time datasets. We extend the recently proposed 'fast subset scan' framework from univariate to multivariate data, enabling computationally efficient detection of irregular space-time clusters even when the numbers of spatial locations and data streams are large. For two variants of the multivariate subset scan, we demonstrate that the scan statistic can be efficiently optimized over proximity-constrained subsets of locations and over all subsets of the monitored data streams, enabling timely detection of emerging events and accurate characterization of the affected locations and streams. Using our new fast search algorithms, we perform an empirical comparison of the Subset Aggregation and Kulldorff multivariate subset scans on synthetic data and real-world disease surveillance tasks, demonstrating tradeoffs between the detection and characterization performance of the two methods.

Presyndromic surveillance for improved detection of emerging public health threats.

Existing public health surveillance systems that rely on predefined symptom categories, or syndromes, are effective at monitoring known illnesses, but there is a critical need for innovation in "presyndromic" surveillance that detects biothreats with rare or previously unseen symptomology. We introduce a data-driven, automated machine learning approach for presyndromic surveillance that learns newly emerging syndromes from free-text emergency department chief complaints, identifies localized case clusters among subpopulations, and incorporates practitioner feedback to automatically distinguish between relevant and irrelevant clusters, thus providing personalized, actionable decision support. Blinded evaluations by New York City's Department of Health and Mental Hygiene demonstrate that our approach identifies more events of public health interest and achieves a lower false-positive rate compared to a state-of-the-art baseline.

Preventing Overdose Using Information and Data from the Environment (PROVIDENT): protocol for a randomized, population-based, community intervention trial.

BACKGROUND AND AIMS: In light of the accelerating drug overdose epidemic in North America, new strategies are needed to identify communities most at risk to prioritize geographically the existing public health resources (e.g. street outreach, naloxone distribution efforts). We aimed to develop PROVIDENT (Preventing Overdose using Information and Data from the Environment), a machine learning-based forecasting tool to predict future overdose deaths at the census block group (i.e. neighbourhood) level. DESIGN: Randomized, population-based, community intervention trial. SETTING: Rhode Island, USA. PARTICIPANTS: All people who reside in Rhode Island during the study period may contribute data to either the model or the trial outcomes. INTERVENTION: Each of the state's 39 municipalities will be randomized to the intervention (PROVIDENT) or comparator condition. An interactive, web-based tool will be developed to visualize the PROVIDENT model predictions. Municipalities assigned to the treatment arm will receive neighbourhood risk predictions from the PROVIDENT model, and state agencies and community-based organizations will direct resources to neighbourhoods identified as high risk. Municipalities assigned to the control arm will continue to receive surveillance information and overdose prevention resources, but they will not receive neighbourhood risk predictions. MEASUREMENTS: The primary outcome is the municipal-level rate of fatal and non-fatal drug overdoses. Fatal overdoses will be defined as unintentional drug-related death; non-fatal overdoses will be defined as an emergency department visit for a suspected overdose reported through the state's syndromic surveillance system. Intervention efficacy will be assessed using Poisson or negative binomial regression to estimate incidence rate ratios comparing fatal and non-fatal overdose rates in treatment vs. control municipalities. COMMENTS: The findings will inform the utility of predictive modelling as a tool to improve public health decision-making and inform resource allocation to communities that should be prioritized for prevention, treatment, recovery and overdose rescue services.

Fast Bayesian scan statistics for multivariate event detection and visualization.

The multivariate Bayesian scan statistic (MBSS) is a recently proposed, general framework for event detection and characterization in multivariate space-time data. MBSS integrates prior information and observations from multiple data streams in a Bayesian framework, computing the posterior probability of each type of event in each space-time region. MBSS has been shown to have many advantages over previous event detection approaches, including improved timeliness and accuracy of detection, easy interpretation and visualization of results, and the ability to model and accurately differentiate between multiple event types. This work extends the MBSS framework to enable detection and visualization of irregularly shaped clusters in multivariate data, by defining a hierarchical prior over all subsets of locations. While a naive search over the exponentially many subsets would be computationally infeasible, we demonstrate that the total posterior probability that each location has been affected can be efficiently computed, enabling rapid detection and visualization of irregular clusters. We compare the run time and detection power of this 'Fast Subset Sums' method to our original MBSS approach (assuming a uniform prior over circular regions) on semi-synthetic outbreaks injected into real-world Emergency Department data from Allegheny County, Pennsylvania. We demonstrate substantial improvements in spatial accuracy and timeliness of detection, while maintaining the scalability and fast run time of the original MBSS method.

Understanding spatial patterns in rape reporting delays.

Under-reporting and delayed reporting of rape crime are severe issues that can complicate the prosecution of perpetrators and prevent rape survivors from receiving needed support. Building on a massive database of publicly available criminal reports from two US cities, we develop a machine learning framework to predict delayed reporting of rape to help tackle this issue. Motivated by large and unexplained spatial variation in reporting delays, we build predictive models to analyse spatial, temporal and socio-economic factors that might explain this variation. Our findings suggest that we can explain a substantial proportion of the variation in rape reporting delays using only openly available data. The insights from this study can be used to motivate targeted, data-driven policies to assist vulnerable communities. For example, we find that younger rape survivors and crimes committed during holiday seasons exhibit longer delays. Our insights can thus help organizations focused on supporting survivors of sexual violence to provide their services at the right place and time. Due to the non-confidential nature of the data used in our models, even community organizations lacking access to sensitive police data can use these findings to optimize their operations.

Big Data Analytics and the Struggle for Equity in Health Care: The Promise and Perils.

Big data is both a product and a function of technology and the ever-growing analytic and computational power. The potential impact of big data in health care innovation cannot be ignored. The technology-mediated transformative potential of big data is taking place within the context of historical inequities in health and health care. Although big data analytics, properly applied, hold great potential to target inequities and reduce disparities, we believe that the realization of this potential requires us to explicitly address concerns of fairness, equity, and transparency in the development of big data tools. To mitigate potential sources of bias and inequity in algorithmic decision-making, a multipronged and interdisciplinary approach is required, combining insights from data scientists and domain experts to design algorithmic decision-making approaches that explicitly account and correct for these issues.

Guggulsterone enhances head and neck cancer therapies via inhibition of signal transducer and activator of transcription-3.

Treatment of human head and neck squamous cell carcinoma (HNSCC) cell lines with guggulsterone, a widely available, well-tolerated nutraceutical, demonstrated dose-dependent decreases in cell viability with EC(50)s ranging from 5 to 8 microM. Guggulsterone induced apoptosis and cell cycle arrest, inhibited invasion and enhanced the efficacy of erlotinib, cetuximab and cisplatin in HNSCC cell lines. Guggulsterone induced decreased expression of both phosphotyrosine and total signal transducer and activator of transcription (STAT)-3, which contributed to guggulsterone's growth inhibitory effect. Hypoxia-inducible factor (HIF)-1alpha was also decreased in response to guggulsterone treatment. In a xenograft model of HNSCC, guggulsterone treatment resulted in increased apoptosis and decreased expression of STAT3. In vivo treatment with a guggulsterone-containing natural product, Guggulipid, resulted in decreased rates of tumor growth and enhancement of cetuximab's activity. Our results suggest that guggulsterone-mediated inhibition of STAT3 and HIF-1alpha provide a biologic rationale for further clinical investigation of this compound in the treatment of HNSCC.

An empirical comparison of spatial scan statistics for outbreak detection.

BACKGROUND: The spatial scan statistic is a widely used statistical method for the automatic detection of disease clusters from syndromic data. Recent work in the disease surveillance community has proposed many variants of Kulldorff's original spatial scan statistic, including expectation-based Poisson and Gaussian statistics, and incorporates a variety of time series analysis methods to obtain expected counts. We evaluate the detection performance of twelve variants of spatial scan, using synthetic outbreaks injected into four real-world public health datasets. RESULTS: The relative performance of methods varies substantially depending on the size of the injected outbreak, the average daily count of the background data, and whether seasonal and day-of-week trends are present. The expectation-based Poisson (EBP) method achieves high performance across a wide range of datasets and outbreak sizes, making it useful in typical detection scenarios where the outbreak characteristics are not known. Kulldorff's statistic outperforms EBP for small outbreaks in datasets with high average daily counts, but has extremely poor detection power for outbreaks affecting more than of the monitored locations. Randomization testing did not improve detection power for the four datasets considered, is computationally expensive, and can lead to high false positive rates. CONCLUSION: Our results suggest four main conclusions. First, spatial scan methods should be evaluated for a variety of different datasets and outbreak characteristics, since focusing only on a single scenario may give a misleading picture of which methods perform best. Second, we recommend the use of the expectation-based Poisson statistic rather than the traditional Kulldorff statistic when large outbreaks are of potential interest, or when average daily counts are low. Third, adjusting for seasonal and day-of-week trends can significantly improve performance in datasets where these trends are present. Finally, we recommend discontinuing the use of randomization testing in the spatial scan framework when sufficient historical data is available for empirical calibration of likelihood ratio scores.

Where did I get dengue? Detecting spatial clusters of infection risk with social network data.

Typical spatial disease surveillance systems associate a single address to each disease case reported, usually the residence address. Social network data offers a unique opportunity to obtain information on the spatial movements of individuals as well as their disease status as cases or controls. This provides information to identify visit locations with high risk of infection, even in regions where no one lives such as parks and entertainment zones. We develop two probability models to characterize the high-risk regions. We use a large Twitter dataset from Brazilian users to search for spatial clusters through analysis of the tweets' locations and textual content. We apply our models to both real-world and simulated data, demonstrating the advantage of our models as compared to the usual spatial scan statistic for this type of data.

Youth violence: What we know and what we need to know.

School shootings tear the fabric of society. In the wake of a school shooting, parents, pediatricians, policymakers, politicians, and the public search for "the" cause of the shooting. But there is no single cause. The causes of school shootings are extremely complex. After the Sandy Hook Elementary School rampage shooting in Newtown, Connecticut, we wrote a report for the National Science Foundation on what is known and not known about youth violence. This article summarizes and updates that report. After distinguishing violent behavior from aggressive behavior, we describe the prevalence of gun violence in the United States and age-related risks for violence. We delineate important differences between violence in the context of rare rampage school shootings, and much more common urban street violence. Acts of violence are influenced by multiple factors, often acting together. We summarize evidence on some major risk factors and protective factors for youth violence, highlighting individual and contextual factors, which often interact. We consider new quantitative "data mining" procedures that can be used to predict youth violence perpetrated by groups and individuals, recognizing critical issues of privacy and ethical concerns that arise in the prediction of violence. We also discuss implications of the current evidence for reducing youth violence, and we offer suggestions for future research. We conclude by arguing that the prevention of youth violence should be a national priority. (PsycINFO Database Record

Human Rights Event Detection from Heterogeneous Social Media Graphs.

Human rights organizations are increasingly monitoring social media for identification, verification, and documentation of human rights violations. Since manual extraction of events from the massive amount of online social network data is difficult and time-consuming, we propose an approach for automated, large-scale discovery and analysis of human rights-related events. We apply our recently developed Non-Parametric Heterogeneous Graph Scan (NPHGS), which models social media data such as Twitter as a heterogeneous network (with multiple different node types, features, and relationships) and detects emerging patterns in the network, to identify and characterize human rights events. NPHGS efficiently maximizes a nonparametric scan statistic (an aggregate measure of anomalousness) over connected subgraphs of the heterogeneous network to identify the most anomalous network clusters. It summarizes each event with information such as type of event, geographical locations, time, and participants, and provides documentation such as links to videos and news reports. Building on our previous work that demonstrates the utility of NPHGS for civil unrest prediction and rare disease outbreak detection, we present an analysis of human rights events detected by NPHGS using two years of Twitter data from Mexico. NPHGS was able to accurately detect relevant clusters of human rights-related tweets prior to international news sources, and in some cases, prior to local news reports. Analysis of social media using NPHGS could enhance the information-gathering missions of human rights organizations by pinpointing specific abuses, revealing events and details that may be blocked from traditional media sources, and providing evidence of emerging patterns of human rights violations. This could lead to more timely, targeted, and effective advocacy, as well as other potential interventions.

Revision arthroscopically assisted anterior cruciate ligament reconstruction with previously unharvested ipsilateral autografts.

BACKGROUND: Revision anterior cruciate ligament reconstruction requires flexibility and variability in treatment options. This study analyzed the functional outcomes and graft stability of 48 consecutive revision anterior cruciate ligament reconstructions using previously unharvested ipsilateral autografts. HYPOTHESIS: Using previously unharvested ipsilateral autografts will achieve similar outcomes to other graft choices in revising previously failed anterior cruciate ligament reconstructions. STUDY DESIGN: Prospective nonrandomized clinical trial. METHODS: Forty-eight patients (48 operations) were observed for 2 to 13 years (mean, 90 months). All agreed to have revision reconstruction with ipsilateral autografts. The details of the technique varied according to the original graft choice and the abnormality encountered. Concomitant procedures were necessary in 40 (84%) of 48 knees. Twenty-three patients (48%) had revision reconstruction with previously unharvested ipsilateral autogenous hamstring tendons. Ten (21%) were 2-stranded grafts, and 13 (27%) were 4-stranded (quadrupled) autografts. Twenty-five patients (52%) had revision reconstruction with previously unharvested ipsilateral patellar tendon autografts, 6 (12%) using the 2-incision rear-entry method and 19 (40%) using the single-incision technique. RESULTS: Results were evaluated with Lysholm and Gillquist scores and International Knee Documentation Committee ratings, including KT-2000 arthrometer examinations. Seventy-three percent of the patients had International Knee Documentation Committee normal (A) or nearly normal (B) knees (42% of the patients had A knees and 42% had B knees). Twelve percent of patients had C knees, and 4% had a D rating. Sixty-seven percent of the knees had a KT-2000 arthrometer side-to-side difference of 3 mm or less, and an additional 21% of the knees had a side-to-side difference of 3 to 5 mm; therefore, 94% of the grafts were functional or partially functional. Six percent of grafts had more than 5 mm of laxity and were considered failures. CONCLUSIONS: Previously unharvested ipsilateral autografts proved reliable in improving function and stability in revision anterior cruciate ligament reconstruction. However, outcomes were less favorable with revision reconstructions than with primary reconstructions.

Automatic detection of omissions in medication lists.

OBJECTIVE: Evidence suggests that the medication lists of patients are often incomplete and could negatively affect patient outcomes. In this article, the authors propose the application of collaborative filtering methods to the medication reconciliation task. Given a current medication list for a patient, the authors employ collaborative filtering approaches to predict drugs the patient could be taking but are missing from their observed list. DESIGN: The collaborative filtering approach presented in this paper emerges from the insight that an omission in a medication list is analogous to an item a consumer might purchase from a product list. Online retailers use collaborative filtering to recommend relevant products using retrospective purchase data. In this article, the authors argue that patient information in electronic medical records, combined with artificial intelligence methods, can enhance medication reconciliation. The authors formulate the detection of omissions in medication lists as a collaborative filtering problem. Detection of omissions is accomplished using several machine-learning approaches. The effectiveness of these approaches is evaluated using medication data from three long-term care centers. The authors also propose several decision-theoretic extensions to the methodology for incorporating medical knowledge into recommendations. RESULTS: Results show that collaborative filtering identifies the missing drug in the top-10 list about 40-50% of the time and the therapeutic class of the missing drug 50%-65% of the time at the three clinics in this study. CONCLUSION: Results suggest that collaborative filtering can be a valuable tool for reconciling medication lists, complementing currently recommended process-driven approaches. However, a one-size-fits-all approach is not optimal, and consideration should be given to context (eg, types of patients and drug regimens) and consequence (eg, the impact of omission on outcomes).