Evaluation of Machine-Learning Algorithms for Predicting Opioid Overdose Risk Among Medicare Beneficiaries With Opioid Prescriptions

Developing and validating a machine-learning algorithm to predict opioid overdose in Medicaid beneficiaries in two US states: a prognostic modelling study

BACKGROUND

Little is known about whether machine-learning algorithms developed to predict opioid overdose using earlier years and from a single state will perform as well when applied to other populations. We aimed to develop a machine-learning algorithm to predict 3-month risk of opioid overdose using Pennsylvania Medicaid data and externally validated it in two data sources (ie, later years of Pennsylvania Medicaid data and data from a different state).

METHODS

This prognostic modelling study developed and validated a machine-learning algorithm to predict overdose in Medicaid beneficiaries with one or more opioid prescription in Pennsylvania and Arizona, USA. To predict risk of hospital or emergency department visits for overdose in the subsequent 3 months, we measured 284 potential predictors from pharmaceutical and health-care encounter claims data in 3-month periods, starting 3 months before the first opioid prescription and continuing until loss to follow-up or study end. We developed and internally validated a gradient-boosting machine algorithm to predict overdose using 2013-16 Pennsylvania Medicaid data (n=639 693). We externally validated the model using (1) 2017-18 Pennsylvania Medicaid data (n=318 585) and (2) 2015-17 Arizona Medicaid data (n=391 959). We reported several prediction performance metrics (eg, C-statistic, positive predictive value). Beneficiaries were stratified into risk-score subgroups to support clinical use.

FINDINGS

A total of 8641 (1·35%) 2013-16 Pennsylvania Medicaid beneficiaries, 2705 (0·85%) 2017-18 Pennsylvania Medicaid beneficiaries, and 2410 (0·61%) 2015-17 Arizona beneficiaries had one or more overdose during the study period. C-statistics for the algorithm predicting 3-month overdoses developed from the 2013-16 Pennsylvania training dataset and validated on the 2013-16 Pennsylvania internal validation dataset, 2017-18 Pennsylvania external validation dataset, and 2015-17 Arizona external validation dataset were 0·841 (95% CI 0·835-0·847), 0·828 (0·822-0·834), and 0·817 (0·807-0·826), respectively. In external validation datasets, 71 361 (22·4%) of 318 585 2017-18 Pennsylvania beneficiaries were in high-risk subgroups (positive predictive value of 0·38-4·08%; capturing 73% of overdoses in the subsequent 3 months) and 40 041 (10%) of 391 959 2015-17 Arizona beneficiaries were in high-risk subgroups (positive predictive value of 0·19-1·97%; capturing 55% of overdoses). Lower risk subgroups in both validation datasets had few individuals (≤0·2%) with an overdose.

INTERPRETATION

A machine-learning algorithm predicting opioid overdose derived from Pennsylvania Medicaid data performed well in external validation with more recent Pennsylvania data and with Arizona Medicaid data. The algorithm might be valuable for overdose risk prediction and stratification in Medicaid beneficiaries.

FUNDING

National Institute of Health, National Institute on Drug Abuse, National Institute on Aging.

Development and multimodal validation of a substance misuse algorithm for referral to treatment using artificial intelligence (SMART-AI): a retrospective deep learning study

Background

Substance misuse is a heterogeneous and complex set of behavioural conditions that are highly prevalent in hospital settings and frequently co-occur. Few hospital-wide solutions exist to comprehensively and reliably identify these conditions to prioritise care and guide treatment. The aim of this study was to apply natural language processing (NLP) to clinical notes collected in the electronic health record (EHR) to accurately screen for substance misuse.

Methods

The model was trained and developed on a reference dataset derived from a hospital-wide programme at Rush University Medical Center (RUMC), Chicago, IL, USA, that used structured diagnostic interviews to manually screen admitted patients over 27 months (between Oct 1, 2017, and Dec 31, 2019; n=54 915). The Alcohol Use Disorder Identification Test and Drug Abuse Screening Tool served as reference standards. The first 24 h of notes in the EHR were mapped to standardised medical vocabulary and fed into single-label, multilabel, and multilabel with auxillary-task neural network models. Temporal validation of the model was done using data from the subsequent 12 months on a subset of RUMC patients (n=16 917). External validation was done using data from Loyola University Medical Center, Chicago, IL, USA between Jan 1, 2007, and Sept 30, 2017 (n=1991 adult patients). The primary outcome was discrimination for alcohol misuse, opioid misuse, or non-opioid drug misuse. Discrimination was assessed by the area under the receiver operating characteristic curve (AUROC). Calibration slope and intercept were measured with the unreliability index. Bias assessments were performed across demographic subgroups.

Findings

The model was trained on a cohort that had 3·5% misuse (n=1 921) with any type of substance. 220 (11%) of 1921 patients with substance misuse had more than one type of misuse. The multilabel convolutional neural network classifier had a mean AUROC of 0·97 (95% CI 0·96–0·98) during temporal validation for all types of substance misuse. The model was well calibrated and showed good face validity with model features containing explicit mentions of aberrant drug-taking behaviour. A false-negative rate of 0·18–0·19 and a false-positive rate of 0·03 between non-Hispanic Black and non-Hispanic White groups occurred. In external validation, the AUROCs for alcohol and opioid misuse were 0·88 (95% CI 0·86–0·90) and 0·94 (0·92–0·95), respectively.

Interpretation

We developed a novel and accurate approach to leveraging the first 24 h of EHR notes for screening multiple types of substance misuse.

Funding

National Institute On Drug Abuse, National Institutes of Health.

Practical foundations of machine learning for addiction research. Part II. Workflow and use cases

In a continuum with applied statistics, machine learning offers a wide variety of tools to explore, analyze, and understand addiction data. These tools include algorithms that can leverage useful information from data to build models; these models can solve particular tasks to answer addiction scientific questions. In this second part of a two-part review on machine learning, we explain how to apply machine learning methods to addiction research. Like other analytical tools, machine learning methods require a careful implementation to carry out a reproducible and transparent research process with reliable results. This review describes a workflow to guide the application of machine learning in addiction research, detailing study design, data collection, data pre-processing, modeling, and results communication. How to train, validate, and test a model, detect and characterize overfitting, and determine an adequate sample size are some of the key issues when applying machine learning. We also illustrate the process and particular nuances with examples of how researchers in addiction have applied machine learning techniques with different goals, study designs, or data sources as well as explain the main limitations of machine learning approaches and how to best address them. A good use of machine learning enriches the addiction research toolkit.

Practical foundations of machine learning for addiction research. Part I. Methods and techniques

Machine learning assembles a broad set of methods and techniques to solve a wide range of problems, such as identifying individuals with substance use disorders (SUD), finding patterns in neuroimages, understanding SUD prognostic factors and their association, or determining addiction genetic underpinnings. However, the addiction research field underuses machine learning. This two-part narrative review focuses on machine learning tools and concepts, providing an introductory insight into their capabilities to facilitate their understanding and acquisition by addiction researchers. This first part presents supervised and unsupervised methods such as linear models, naive Bayes, support vector machines, artificial neural networks, and k-means. We illustrate each technique with examples of its use in current addiction research. We also present some open-source programming tools and methodological good practices that facilitate using these techniques. Throughout this work, we emphasize a continuum between applied statistics and machine learning, we show their commonalities, and provide sources for further reading to deepen the understanding of these methods. This two-part review is a primer for the next generation of addiction researchers incorporating machine learning in their projects. Researchers will find a bridge between applied statistics and machine learning, ways to expand their analytical toolkit, recommendations to incorporate well-established good practices in addiction data analysis (e.g., stating the rationale for using newer analytical tools, calculating sample size, improving reproducibility), and the vocabulary to enhance collaboration between researchers who do not conduct data analyses and those who do.

Risky alcohol use among patients dispensed opioid medications: A clinical community pharmacy study

BACKGROUND

Included among the significant risk factors for opioid overdose is concomitant use of other central nervous system depressants, particularly alcohol. Given the continued expansion of community pharmacy in the continuum of care, it is imperative to characterize alcohol use among pharmacy patients dispensed opioids in order to establish a foundation for identification and intervention in these settings.

METHODS

This secondary analysis utilized data from a one-time, cross-sectional health assessment conducted among patients dispensed opioid medications in 19 community pharmacies in Indiana and Ohio. Adult, English speaking, patients not receiving cancer care who were dispensed opioid medications were asked to self-report alcohol and substance use, behavioral and physical health, and demographic information. Descriptive and logistic regression analyses were employed to characterize alcohol use/risky alcohol use and patient characteristics associated therewith.

RESULTS

The analytical sample included 1494 individuals. Participants were on average 49 years of age (Standard Deviation=14.9)-with 6% being persons of color (n = 89). Weekly drinking was reported by 18.1% (n = 204) and daily drinking was reported by 6.8% (n = 77) of the study sample, with a total of 143 (9.6%) participants reporting moderate/high risk drinking. Males (Adjusted Odds Ratio [AOR]=1.94, 95% CI=1.3,2.9), those with higher pain interference (AOR=1.44, 95% CI=1.0,2.0), overdose history (AOR=1.93, 95% CI=1.1,3.5), sedative use (AOR=2.11, 95% CI=1.3,3.5), and tobacco use (AOR=2.41, 95% CI=1.6,3.7) had increased likelihood of moderate/high risk alcohol use (all p < 0.05).

CONCLUSIONS

Medication labeling and clinical guidelines clearly indicate that patients should abstain from concomitant use of opioids and alcohol. This study has identified rates and associated risk factors of risky alcohol use among a clinical sample of community pharmacy patients dispensed opioid medications. Continuing this line of research and potential clinical service development has the ability to improve patient safety through addressing a significant gap within the current opioid epidemic.

Machine Learning Improves Upon Clinicians' Prediction of End Stage Kidney Disease

Background and Objectives

Chronic kidney disease progression to ESKD is associated with a marked increase in mortality and morbidity. Its progression is highly variable and difficult to predict.

Methods

This is an observational, retrospective, single-centre study. The cohort was patients attending hospital and nephrology clinic at The Canberra Hospital from September 1996 to March 2018. Demographic data, vital signs, kidney function test, proteinuria, and serum glucose were extracted. The model was trained on the featurised time series data with XGBoost. Its performance was compared against six nephrologists and the Kidney Failure Risk Equation (KFRE).

Results

A total of 12,371 patients were included, with 2,388 were found to have an adequate density (three eGFR data points in the first 2 years) for subsequent analysis. Patients were divided into 80%/20% ratio for training and testing datasets.

ML model had superior performance than nephrologist in predicting ESKD within 2 years with 93.9% accuracy, 60% sensitivity, 97.7% specificity, 75% positive predictive value. The ML model was superior in all performance metrics to the KFRE 4- and 8-variable models.

eGFR and glucose were found to be highly contributing to the ESKD prediction performance.

Conclusions

The computational predictions had higher accuracy, specificity and positive predictive value, which indicates the potential integration into clinical workflows for decision support.

Development and validation of a risk-score model for opioid overdose using a national claims database

Opioid overdose can be serious adverse effects of opioid analgesics. Thus, several strategies to mitigate risk and reduce the harm of opioid overdose have been developed. However, despite a marked increase in opioid analgesic consumption in Korea, there have been no tools predicting the risk of opioid overdose in the Korean population. Using the national claims database of the Korean population, we identified patients who were incidentally prescribed non-injectable opioid analgesic (NIOA) at least once from 2017 to 2018 (N = 1,752,380). Among them, 866 cases of opioid overdose occurred, and per case, four controls were selected. Patients were randomly allocated to the development (80%) and validation (20%) cohort. Thirteen predictive variables were selected via logistic regression modelling, and a risk-score was assigned for each predictor. Our model showed good performance with c-statistics of 0.84 in the validation cohort. The developed risk score model is the first tool to identify high-risk patients for opioid overdose in Korea. It is expected to be applicable in the clinical setting and useful as a national level surveillance tool due to the easily calculable and identifiable predictors available from the claims database.

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.

Data needs and models for the opioid epidemic

The evolving nature of the opioid epidemic and continued increases in overdose deaths highlight a need for fundamental change in the collection and use of surveillance data to link them to implementation of effective service, treatment, and prevention approaches. Yet at present, the quality and timeliness of US surveillance data often limits data-driven approaches. We review current information needs, summarize limitations of existing data, propose complementary surveillance resources, and provide examples of promising approaches designed to meet the needs of data end-users. We conclude that there is a need for an approach that focuses on the needs of data end-users, such as public health systems leaders, policy makers, public, nonprofit and prepaid healthcare systems, and other systems, such as the justice system. Such an approach, which may require investments in new infrastructure, should prioritize improvements in data timeliness, sample representativeness, database linkage, and increased flexibility to adapt to shifts in the environment, while preserving the privacy of survey participants. Use of simulations, distributed research and data networks, alternative data sources, such as wastewater or digital data collection and use of blockchain technology, are some of promising avenues toward an improved and more user-centered surveillance system.

Using machine learning to classify patients on opioid use

Objectives

High-frequent opioid use tends to increase an individual's risk of opioid use disorder, overdose and death. Thus, it is important to predict an individuals' opioid use frequency to improve opioid prescription utilization outcomes.

Methods

Individuals receiving at least one opioid prescription from 2016 to 2018 in the national representative data, Medical Expenditure Panel Survey, were included. This study applied five machine learning (ML) techniques, including support vector machine, random forest, neural network, gradient boosting and XGBoost (extreme gradient boosting), to predict opioid use frequency. This study compared the performance of these ML models with penalized logistic regression. The study outcome was whether an individual lied in the upper 10% of the opioid prescription distribution. Predictors were selected based on Gelberg-Andersen's Behavioral Model of Health Services Utilization. The prediction performance was assessed using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC) in the test data. Patient characteristics as predictors for high-frequency use of opioids were ranked by the relative importance in prediction in the test data.

Key findings

Random forest and gradient boosting achieved the top values of both AUROC and AUPRC, outperforming logistic regression and three other ML methods. In the best performing model, the random forest, the following characteristics had high predictive power in the frequency of opioid use: age, number of chronic conditions, public insurance and self-perceived health status.

Conclusions

The results of this study demonstrate that ML techniques can be a promising and powerful technique in predicting the frequency of opioid use and health outcomes.

The potential for leveraging machine learning to filter medication alerts

OBJECTIVE

To evaluate the potential for machine learning to predict medication alerts that might be ignored by a user, and intelligently filter out those alerts from the user's view.

MATERIALS AND METHODS

We identified features (eg, patient and provider characteristics) proposed to modulate user responses to medication alerts through the literature; these features were then refined through expert review. Models were developed using rule-based and machine learning techniques (logistic regression, random forest, support vector machine, neural network, and LightGBM). We collected log data on alerts shown to users throughout 2019 at University of Utah Health. We sought to maximize precision while maintaining a false-negative rate <0.01, a threshold predefined through discussion with physicians and pharmacists. We developed models while maintaining a sensitivity of 0.99. Two null hypotheses were developed: H1-there is no difference in precision among prediction models; and H2-the removal of any feature category does not change precision.

RESULTS

A total of 3,481,634 medication alerts with 751 features were evaluated. With sensitivity fixed at 0.99, LightGBM achieved the highest precision of 0.192 and less than 0.01 for the pre-defined maximal false-negative rate by subject-matter experts (H1) (P < 0.001). This model could reduce alert volume by 54.1%. We removed different combinations of features (H2) and found that not all features significantly contributed to precision. Removing medication order features (eg, dosage) most significantly decreased precision (-0.147, P = 0.001).

CONCLUSIONS

Machine learning potentially enables the intelligent filtering of medication alerts.

Using data science to improve outcomes for persons with opioid use disorder

Medication treatment for opioid use disorder (MOUD) is an effective evidence-based therapy for decreasing opioid-related adverse outcomes. Effective strategies for retaining persons on MOUD, an essential step to improving outcomes, are needed as roughly half of all persons initiating MOUD discontinue within a year. Data science may be valuable and promising for improving MOUD retention by using "big data" (e.g., electronic health record data, claims data mobile/sensor data, social media data) and specific machine learning techniques (e.g., predictive modeling, natural language processing, reinforcement learning) to individualize patient care. Maximizing the utility of data science to improve MOUD retention requires a three-pronged approach: (1) increasing funding for data science research for OUD, (2) integrating data from multiple sources including treatment for OUD and general medical care as well as data not specific to medical care (e.g., mobile, sensor, and social media data), and (3) applying multiple data science approaches with integrated big data to provide insights and optimize advances in the OUD and overall addiction fields.

Applying Machine Learning in Distributed Data Networks for Pharmacoepidemiologic and Pharmacovigilance Studies: Opportunities, Challenges, and Considerations

Increasing availability of electronic health databases capturing real-world experiences with medical products has garnered much interest in their use for pharmacoepidemiologic and pharmacovigilance studies. The traditional practice of having numerous groups use single databases to accomplish similar tasks and address common questions about medical products can be made more efficient through well-coordinated multi-database studies, greatly facilitated through distributed data network (DDN) architectures. Access to larger amounts of electronic health data within DDNs has created a growing interest in using data-adaptive machine learning (ML) techniques that can automatically model complex associations in high-dimensional data with minimal human guidance. However, the siloed storage and diverse nature of the databases in DDNs create unique challenges for using ML. In this paper, we discuss opportunities, challenges, and considerations for applying ML in DDNs for pharmacoepidemiologic and pharmacovigilance studies. We first discuss major types of activities performed by DDNs and how ML may be used. Next, we discuss practical data-related factors influencing how DDNs work in practice. We then combine these discussions and jointly consider how opportunities for ML are affected by practical data-related factors for DDNs, leading to several challenges. We present different approaches for addressing these challenges and highlight efforts that real-world DDNs have taken or are currently taking to help mitigate them. Despite these challenges, the time is ripe for the emerging interest to use ML in DDNs, and the utility of these data-adaptive modeling techniques in pharmacoepidemiologic and pharmacovigilance studies will likely continue to increase in the coming years.

Ensemble learning to predict opioid-related overdose using statewide prescription drug monitoring program and hospital discharge data in the state of Tennessee

Objective

To develop and validate algorithms for predicting 30-day fatal and nonfatal opioid-related overdose using statewide data sources including prescription drug monitoring program data, Hospital Discharge Data System data, and Tennessee (TN) vital records. Current overdose prevention efforts in TN rely on descriptive and retrospective analyses without prognostication.

Materials and Methods

Study data included 3 041 668 TN patients with 71 479 191 controlled substance prescriptions from 2012 to 2017. Statewide data and socioeconomic indicators were used to train, ensemble, and calibrate 10 nonparametric “weak learner” models. Validation was performed using area under the receiver operating curve (AUROC), area under the precision recall curve, risk concentration, and Spiegelhalter z-test statistic.

Results

Within 30 days, 2574 fatal overdoses occurred after 4912 prescriptions (0.0069%) and 8455 nonfatal overdoses occurred after 19 460 prescriptions (0.027%). Discrimination and calibration improved after ensembling (AUROC: 0.79–0.83; Spiegelhalter P value: 0–.12). Risk concentration captured 47–52% of cases in the top quantiles of predicted probabilities.

Discussion

Partitioning and ensembling enabled all study data to be used given computational limits and helped mediate case imbalance. Predicting risk at the prescription level can aggregate risk to the patient, provider, pharmacy, county, and regional levels. Implementing these models into Tennessee Department of Health systems might enable more granular risk quantification. Prospective validation with more recent data is needed.

Conclusion

Predicting opioid-related overdose risk at statewide scales remains difficult and models like these, which required a partnership between an academic institution and state health agency to develop, may complement traditional epidemiological methods of risk identification and inform public health decisions.

Predicting 30-day readmissions in patients with heart failure using administrative data: a machine learning approach

AIMS

To develop machine-learning (ML) models trained on administrative data which predict risk of readmission in heart failure (HF) patients; evaluate and compare the ML model with the currently used LaCE score using clinically informative metrics.

METHODS AND RESULTS

This prognostic study was conducted in Alberta, Canada on 9,845 patients with confirmed HF admitted to hospital between 2012-2019. The outcome was unplanned all-cause hospital readmission within 30-days of discharge. 80% of the data was used for ML model development and 20% for independent validation. We reported, using the validation set, c-statistics (AUROCs)and performance metrics (likelihood ratio [LR], positive predictive values [PPV]) for the XGBoost model and a modified LaCE score within their respective predictive thresholds. Boosted tree-based classifiers had higher AUROCs (0.65 for XGBoost) compared to others (0.58 for Neural Network) and 0.57 for the modified LaCE. Within the predicted threshold range of the XGBoost classifier, the positive LR was 1.00 at the low end of predicted risk and 6.12 at the high end, resulting in a PPV (post-test probability) range of 21-62%; the pre-test probability of readmission was 20.9% using prevalence. The corresponding positive LRs and PPVs across LaCE score thresholds were 1.00-1.20 and 21-24%, respectively.

CONCLUSION

Despite predicting readmissions better than the LaCE, even the best ML model trained on administrative health data (XGBoost) did not provide substantially informative prediction performance as it only generated a moderate shift from pre to post-test probability. Health systems wishing to deploy such a tool should consider training ML models with additional data. Adding other techniques like Natural Language Processing, along with ML, to use other clinical information (like chart notes) might improve prediction performance.

Application of Machine Learning to Identify Clinically Meaningful Risk Group for Osteoporosis in Individuals Under the Recommended Age for Dual-Energy X-Ray Absorptiometry

Dual-energy X-ray absorptiometry (DXA) is the gold standard for diagnosing osteoporosis; it is generally recommended in men ≥ 70 and women ≥ 65 years old. Therefore, assessment of clinical risk factors for osteoporosis is very important in individuals under the recommended age for DXA. Here, we examine the diagnostic performance of machine learning-based prediction models for osteoporosis in individuals under the recommended age for DXA examination. Data of 2210 men aged 50-69 and 1099 women aged 50-64 obtained from the Korea National Health and Nutrition Examination Survey IV-V were analyzed. Extreme gradient boosting (XGBoost) was used to find relevant clinical features and applied to three machine learning models: XGBoost, logistic regression, and a multilayer perceptron. For the prediction of osteoporosis, the XGBoost model using the top 20 features extracted from XGBoost showed the most reliable performance with area under the receiver operating characteristic curve (AUROC) of 0.73 and 0.79 in men and women, respectively. We compared the diagnostic accuracy of the Shapley additive explanation values based on a risk-score model obtained from XGBoost and conventional osteoporosis risk assessment tools for prediction of osteoporosis using optimal cut-off values for each model. We observed that a cut-off risk score of ≥ 28 in men and ≥ 47 in women was optimal to classify a positive screening for osteoporosis (an AUROC of 0.86 in men and 0.91 in women). The XGBoost-based osteoporosis-prediction model outperformed conventional risk assessment tools. Therefore, machine learning-based prediction models are a more suitable option than conventional risk assessment methods for screening osteoporosis in individuals under the recommended age for DXA examination.

Outcomes associated with the use of medications for opioid use disorder during pregnancy

AIM

To test the effect of the duration of medication for opioid use disorder (MOUD) use during pregnancy on maternal, perinatal and neonatal outcomes.

DESIGN

Retrospective cohort analysis of claims, encounter and pharmacy data.

SETTING

Pennsylvania, USA.

PARTICIPANTS

We analyzed 13 320 pregnancies among 10 741 women with opioid use disorder aged 15-44 years enrolled in Pennsylvania Medicaid between 2009 and 2017.

MEASUREMENTS

We examined five outcomes during pregnancy and for 12 weeks postpartum: (1) overdose, (2) postpartum MOUD continuation, (3) preterm birth (< 37 weeks gestation), (4) term low birth weight (< 2500 g at ≥ 37 weeks) and (5) neonatal abstinence syndrome (NAS). Our primary exposure was the duration (count of weeks) of any MOUD use, including methadone or buprenorphine, during pregnancy.

FINDINGS

Among 13 320 pregnancies, 306 (2.3%) were complicated by an overdose, 1753 (13.2%) resulted in a preterm birth and 6787 (50.9%) continued MOUD postpartum. Among infants, 874 (7.6%) were low birth weight at term and 7706 (57.9%) were diagnosed with NAS. As the duration of MOUD use increased, we found a statistically significant decrease in the rate of overdose and preterm birth, a statistically significant increase in the rate of postpartum MOUD continuation and NAS and a decline in term low birth weight. Specifically, for each additional week of MOUD, the adjusted odds of overdose decreased by 2% [adjusted odds ratio (aOR) = 0.98; 95% confidence interval (CI) = 0.97, 0.99], preterm birth decreased by 1% (aOR = 0.99; 95% CI = 0.99, 1.00), postpartum MOUD continuation increased by 95% (aOR = 1.95; 95% CI = 1.87, 2.04) and NAS increased by 41% (aOR = 1.41; 95% CI = 1.35, 1.47). The odds of term low birth weight did not change (aOR = 1.00; 95% CI = 0.99, 1.00), although the rate declined with a longer duration of MOUD use during pregnancy.

CONCLUSIONS

Longer duration of medication for opioid use disorder use during pregnancy appears to be associated with improved maternal and perinatal outcomes.

The leading neighborhood-level predictors of drug overdose: A mixed machine learning and spatial approach

BACKGROUND

Drug overdose is a leading cause of unintentional death in the United States and has contributed significantly to a decline in life expectancy during recent years. To combat this health issue, this study aims to identify the leading neighborhood-level predictors of drug overdose and develop a model to predict areas at the highest risk of drug overdose using geographic information systems and machine learning (ML) techniques.

METHOD

Neighborhood-level (block group) predictors were grouped into three domains: socio-demographic factors, drug use variables, and protective resources. We explored different ML algorithms, accounting for spatial dependency, to identify leading predictors in each domain. Using geographically weighted regression and the best-performing ML algorithm, we combined the output prediction of three domains to produce a final ensemble model. The model performance was validated using classification evaluation metrics, spatial cross-validation, and spatial autocorrelation testing.

RESULTS

The variables contributing most to the predictive model included the proportion of households with food stamps, households with an annual income below $35,000, opioid prescription rate, smoking accessories expenditures, and accessibility to opioid treatment programs and hospitals. Compared to the error estimated from normal cross-validation, the generalized error of the model did not increase considerably in spatial cross-validation. The ensemble model using ML outperformed the GWR method.

CONCLUSION

This study identified strong neighborhood-level predictors that place a community at risk of experiencing drug overdoses, as well as protective factors. Our findings may shed light on several specific avenues for targeted intervention in neighborhoods at risk for high drug overdose burdens.

Key use cases for artificial intelligence to reduce the frequency of adverse drug events: a scoping review

Adverse drug events (ADEs) represent one of the most prevalent types of health-care-related harm, and there is substantial room for improvement in the way that they are currently predicted and detected. We conducted a scoping review to identify key use cases in which artificial intelligence (AI) could be leveraged to reduce the frequency of ADEs. We focused on modern machine learning techniques and natural language processing. 78 articles were included in the scoping review. Studies were heterogeneous and applied various AI techniques covering a wide range of medications and ADEs. We identified several key use cases in which AI could contribute to reducing the frequency and consequences of ADEs, through prediction to prevent ADEs and early detection to mitigate the effects. Most studies (73 [94%] of 78) assessed technical algorithm performance, and few studies evaluated the use of AI in clinical settings. Most articles (58 [74%] of 78) were published within the past 5 years, highlighting an emerging area of study. Availability of new types of data, such as genetic information, and access to unstructured clinical notes might further advance the field.

Advanced models for improved prediction of opioid-related overdose and suicide events among Veterans using administrative healthcare data

Veterans suffer disproportionate health impacts from the opioid epidemic, including overdose, suicide, and death. Prediction models based on electronic medical record data can be powerful tools for identifying patients at greatest risk of such outcomes. The Veterans Health Administration implemented the Stratification Tool for Opioid Risk Mitigation (STORM) in 2018. In this study we propose changes to the original STORM model and propose alternative models that improve risk prediction performance. The best of these proposed models uses a multivariate generalized linear mixed modeling (mGLMM) approach to produce separate predictions for overdose and suicide-related events (SRE) rather than a single prediction for combined outcomes. Further improvements include incorporation of additional data sources and new predictor variables in a longitudinal setting. Compared to a modified version of the STORM model with the same outcome, predictor and interaction terms, our proposed model has a significantly better prediction performance in terms of AUC (84% vs. 77%) and sensitivity (71% vs. 66%). The mGLMM performed particularly well in identifying patients at risk for SREs, where 72% of actual events were accurately predicted among patients with the 100,000 highest risk scores compared with 49.7% for the modified STORM model. The mGLMM’s strong performance in identifying true cases (sensitivity) among this highest risk group was the most important improvement given the model’s primary purpose for accurately identifying patients at most risk for adverse outcomes such that they are prioritized to receive risk mitigation interventions. Some predictors in the proposed model have markedly different associations with overdose and suicide risks, which will allow clinicians to better target interventions to the most relevant risks.

Using administrative data to predict cessation risk and identify novel predictors among new entrants to opioid agonist treatment

BACKGROUND

Longer retention in opioid agonist treatment (OAT) is associated with improved treatment outcomes but 12-month retention rates are often low. Innovative approaches are needed to strengthen retention in OAT. We develop and compare traditional and deep learning-extensions of Cox regression to examine the potential for predicting time in OAT at individuals' first episode entry.

METHODS

Retrospective cohort study in New South Wales, Australia including 16,576 people entering OAT for the first time between January 2006 and December 2017. We develop 12-month OAT cessation prediction models using traditional and deep learning-extensions of the Cox regression algorithm with predictors evaluated from linked administrative datasets. Proportion of explained variation, calibration, and discrimination are compared using 5 × 2 cross-validation.

RESULTS

Twelve-month cessation rate was 58.4%. The largest hazard ratios for earlier cessation from the deep learning model were observed for treatment factors, including private dosing points (HR=1.54, 95% CI=1.49-1.60) and buprenorphine medication (HR=1.43, 95% CI=1.39-1.46). Diagnostic codes for homelessness (HR=1.09, 95% CI=1.04-1.13), outpatient treatment for drug use disorders (HR=1.10, 95% CI=1.06-1.15), and occupant of vehicle accident (HR=1.04, 95% CI=1.01-1.07) from past-year health service presentations were identified as significant predictors of retention. We observed no improvement in performance of the deep learning model over traditional Cox regression.

CONCLUSIONS

Deep learning may be more useful in identifying novel risk factors of OAT retention from administrative data than evaluating individual-level risk. An increased focus on addressing structural issues at the population level and considering alternate models of care may be more effective at improving retention than delivering fully personalised OAT.

Accurate Medication Adherence Measurement Using Administrative Data for Frequently Hospitalized Patients

Background: Pharmacy administrative claims data remain an accessible and efficient source to measure medication adherence for frequently hospitalized patient populations that are systematically excluded from the landmark drug trials. Published pharmacotherapy studies use medication possession ratio (MPR) and proportion of days covered (PDC) to calculate medication adherence and usually fail to incorporate hospitalization and prescription overlap/gap from claims data. To make the cacophony of adherence measures clearer, this study created a refined hospital-adjusted algorithm to capture pharmacotherapy adherence among patients with end-stage renal disease (ESRD). Methods: The United States Renal Data System (USRDS) registry of ESRD was used to determine prescription-filling patterns of those receiving new prescriptions for oral P2Y12 inhibitors (P2Y12-I) between 2011 and 2015. P2Y12-I-naïve patients were followed until death, kidney transplantation, discontinuing medications, or loss to follow-up. After flagging/censoring key variables, the algorithm adjusted for hospital length of stay (LOS) and medication overlap. Hospital-adjusted medication adherence (HA-PDC) was calculated and compared with traditional MPR and PDC methods. Analyses were performed with SAS software. Results: Hospitalization occurred for 78% of the cohort (N = 46 514). The median LOS was 12 (interquartile range [IQR] = 2-34) days. MPR and PDC were 61% (IQR = 29%-94%) and 59% (IQR = 31%-93%), respectively. After applying adjustments for overlapping coverage days and hospital stays independently, HA-PDC adherence values changed in 41% and 52.7% of the cohort, respectively. When adjustments for overlap and hospital stay were made concurrently, HA-PDC adherence values changed in 68% of the cohort by 5.8% (HA-PDC median = 0.68, IQR = 0.31-0.93). HA-PDC declined over time (3M-6M-9M-12M). Nearly 48% of the cohort had a ≥30 days refill gap in the first 3 months, and this increased over time (P < .0001). Conclusions: Refill gaps should be investigated carefully to capture accurate pharmacotherapy adherence. HA-PDC measures increased adherence substantially when adjustments for hospital stay and medication refill overlaps are made. Furthermore, if hospitalizations were ignored for medications that are included in Medicare quality measures, such as Medicare STAR program, the apparent reduction in adherence might be associated with lower quality and health plan reimbursement.

Machine learning takes a village: Assessing neighbourhood-level vulnerability for an overdose and infectious disease outbreak

BACKGROUND

Multiple areas in the United States of America (USA) are experiencing high rates of overdose and outbreaks of bloodborne infections, including HIV and hepatitis C virus (HCV), due to non-sterile injection drug use. We aimed to identify neighbourhoods at increased vulnerability for overdose and infectious disease outbreaks in Rhode Island, USA. The primary aim was to pilot machine learning methods to identify which neighbourhood-level factors were important for creating "vulnerability assessment scores" across the state. The secondary aim was to engage stakeholders to pilot an interactive mapping tool and visualize the results.

METHODS

From September 2018 to November 2019, we conducted a neighbourhood-level vulnerability assessment and stakeholder engagement process named The VILLAGE Project (Vulnerability Investigation of underlying Local risk And Geographic Events). We developed a predictive analytics model using machine learning methods (LASSO, Elastic Net, and RIDGE) to identify areas with increased vulnerability to an outbreak of overdose, HIV and HCV, using census tract-level counts of overdose deaths as a proxy for injection drug use patterns and related health outcomes. Stakeholders reviewed mapping tools for face validity and community distribution.

RESULTS

Machine learning prediction models were suitable for estimating relative neighbourhood-level vulnerability to an outbreak. Variables of importance in the model included housing cost burden, prior overdose deaths, housing density, and education level. Eighty-nine census tracts (37%) with no prior overdose fatalities were identified as being vulnerable to such an outbreak, and nine of those were identified as having a vulnerability assessment score in the top 25%. Results were disseminated as a vulnerability stratification map and an online interactive mapping tool.

CONCLUSION

Machine learning methods are well suited to predict neighborhoods at higher vulnerability to an outbreak. These methods show promise as a tool to assess structural vulnerabilities and work to prevent outbreaks at the local level.

Machine learning risk prediction model for acute coronary syndrome and death from use of non-steroidal anti-inflammatory drugs in administrative data

Our aim was to investigate the usefulness of machine learning approaches on linked administrative health data at the population level in predicting older patients’ one-year risk of acute coronary syndrome and death following the use of non-steroidal anti-inflammatory drugs (NSAIDs). Patients from a Western Australian cardiovascular population who were supplied with NSAIDs between 1 Jan 2003 and 31 Dec 2004 were identified from Pharmaceutical Benefits Scheme data. Comorbidities from linked hospital admissions data and medication history were inputs. Admissions for acute coronary syndrome or death within one year from the first supply date were outputs. Machine learning classification methods were used to build models to predict ACS and death. Model performance was measured by the area under the receiver operating characteristic curve (AUC-ROC), sensitivity and specificity. There were 68,889 patients in the NSAIDs cohort with mean age 76 years and 54% were female. 1882 patients were admitted for acute coronary syndrome and 5405 patients died within one year after their first supply of NSAIDs. The multi-layer neural network, gradient boosting machine and support vector machine were applied to build various classification models. The gradient boosting machine achieved the best performance with an average AUC-ROC of 0.72 predicting ACS and 0.84 predicting death. Machine learning models applied to linked administrative data can potentially improve adverse outcome risk prediction. Further investigation of additional data and approaches are required to improve the performance for adverse outcome risk prediction.

Opioid Use Disorder and Overdose in Older Adults With Breast, Colorectal, or Prostate Cancer

BACKGROUND

Despite high rates of opioid therapy, evidence about the risk of preventable opioid harms among cancer survivors is underdeveloped. Our objective was to estimate the odds of opioid use disorder (OUD) and overdose following breast, colorectal, or prostate cancer diagnosis among Medicare beneficiaries.

METHODS

We conducted a retrospective cohort study using 2007-2014 Surveillance, Epidemiology, and End Results-Medicare data for cancer survivors with a first cancer diagnosis of stage 0-III breast, colorectal, or prostate cancer at age 66-89 years between 2008 and 2013. Cancer survivors were matched to up to 2 noncancer controls on age, sex, and Surveillance, Epidemiology, and End Results region. Using Firth logistic regression, we estimated adjusted 1-year odds of OUD or nonfatal opioid overdose associated with a cancer diagnosis. We also estimated adjusted odds of OUD and overdose separately and by cancer stage, prior opioid use, and follow-up time.

RESULTS

Among 69 889 cancer survivors and 125 007 controls, the unadjusted rates of OUD or nonfatal overdose were 25.2, 27.1, 38.9, and 12.4 events per 10 000 patients in the noncancer, breast, colorectal, and prostate samples, respectively. There was no association between cancer and OUD. Colorectal survivors had 2.3 times higher odds of opioid overdose compared with matched controls (adjusted odds ratio = 2.33, 95% confidence interval  = 1.49 to 3.67). Additionally, overdose risk was greater in those with more advanced disease, no prior opioid use, and preexisting mental health conditions.

CONCLUSIONS

Opioid overdose was a rare, but statistically significant, outcome following stage II-III colorectal cancer diagnosis, particularly among previously opioid-naïve patients. These patients may require heightened screening and intervention to prevent inadvertent adverse opioid harms.

Longitudinal K-means approaches to clustering and analyzing EHR opioid use trajectories for clinical subtypes

Identification of patient subtypes from retrospective Electronic Health Record (EHR) data is fraught with inherent modeling issues, such as missing data and variable length time intervals, and the results obtained are highly dependent on data pre-processing strategies. As we move towards personalized medicine, assessing accurate patient subtypes will be a key factor in creating patient specific treatment plans. Partitioning longitudinal trajectories from irregularly spaced and variable length time intervals is a well-established, but open problem. In this work, we present and compare k-means approaches for subtyping opioid use trajectories from EHR data. We then interpret the resulting subtypes using decision trees, examining how each subtype is influenced by opioid medication features and patient diagnoses, procedures, and demographics. Finally, we discuss how the subtypes can be incorporated in static machine learning models as features in predicting opioid overdose and adverse events. The proposed methods are general, and can be extended to other EHR prescription dosage trajectories.

Morphine Equivalent Total Dosage as Predictor of Adverse Outcomes in Opioid Prescribing

OBJECTIVE

The objective of this work was to develop a risk prediction model of opioid overdoses and opioid use disorder for patients at first opioid prescription and compare the predictive accuracy of using morphine equivalent total dosage with daily dosage as predictors.

DESIGN

Records from patients aged 18-79 years with opioid prescriptions between January 1, 2016 and June 30, 2019, no prior history of adverse outcomes, and no malignant cancer diagnoses were collected from the electronic health records system of a medium-sized central Ohio health care system (n = 219,276). A Cox proportional hazards model was developed to predict the adverse outcomes of opioid overdoses and opioid use disorder from patient sociodemographic, pharmacological, and clinical diagnoses factors.

RESULTS

573 patients experienced overdoses and 2,571 patients were diagnosed with OUD in the study time frame. Morphine equivalent total dosage of opioid prescriptions was identified as a stronger predictor of adverse outcomes (C = 0.797) than morphine equivalent daily dosage (C = 0.792), with best predictions from a model that includes both predictors (C = 0.803). In the model with both daily and total dosage predictors, patients receiving a high total/low daily dosage experienced a higher risk (HR = 2.17) than those receiving a low total/high daily dosage (HR = 2.02). Those receiving a high total/high daily dosage experienced the greatest risk of all (HR = 3.09).

CONCLUSIONS

These findings demonstrate the value of including morphine equivalent total dosage as a predictor of adverse opioid outcomes and suggest total dosage may be more strongly correlated with increased risk than daily dosage.

Assessing opioid overdose risk: a review of clinical prediction models utilizing patient-level data

Drug, and specifically opioid-related, overdoses remain a major public health problem in the United States. Multiple studies have examined individual risk factors associated with overdose risk, but research developing clinical risk prediction tools for overdose has only emerged in the last few years. We conducted a comprehensive review of the literature on patient-level factors associated with opioid-related overdose risk, with an emphasis on clinical risk prediction models for opioid-related overdose in the United States. Studies that developed and/or validated clinical prediction models were closely reviewed and evaluated to determine the state of the field. We identified 12 studies that reported risk prediction models for opioid-related overdose risk. Published models were developed from a variety of data sources, including Veterans Health Administration data, Medicare data, commercial insurance data, and statewide linked datasets. Studies reported model performance using measures of discrimination, usually at good-to-excellent levels, though they did not always assess calibration. C-statistics were better for models that included clinical predictors (c-statistics: 0.75-0.95) compared to models without them (c-statistics: 0.69-0.82). External validation of models was rare, and we found no studies evaluating implementation of models or risk prediction tools into clinical practice. A common feature of these models was a high rate of false positives, largely because opioid-related overdose is rare in the general population. Thus, efforts to implement prediction models into practice should take into account that published models overestimate overdose risk for many low-risk patients. Future prediction models assessing overdose risk should employ external validation and address model calibration. In order to translate findings from prediction models into clinical public health benefit, future studies should focus on developing clinical prediction tools based on prediction models, implementing these tools into clinical practice, and evaluating the impact of these models on treatment decisions, patient outcomes, and, ultimately, opioid overdose rates.

Predictors of attrition in a longitudinal population-based study of aging

BACKGROUND

Longitudinal studies predictably experience non-random attrition over time. Among older adults, risk factors for attrition may be similar to risk factors for outcomes such as cognitive decline and dementia, potentially biasing study results.

OBJECTIVE

To characterize participants lost to follow-up which can be useful in the study design and interpretation of results.

METHODS

In a longitudinal aging population study with 10 years of annual follow-up, we characterized the attrited participants (77%) compared to those who remained in the study. We used multivariable logistic regression models to identify attrition predictors. We then implemented four machine learning approaches to predict attrition status from one wave to the next and compared the results of all five approaches.

RESULTS

Multivariable logistic regression identified those more likely to drop out as older, male, not living with another study participant, having lower cognitive test scores and higher clinical dementia ratings, lower functional ability, fewer subjective memory complaints, no physical activity, reported hobbies, or engagement in social activities, worse self-rated health, and leaving the house less often. The four machine learning approaches using areas under the receiver operating characteristic curves produced similar discrimination results to the multivariable logistic regression model.

CONCLUSIONS

Attrition was most likely to occur in participants who were older, male, inactive, socially isolated, and cognitively impaired. Ignoring attrition would bias study results especially when the missing data might be related to the outcome (e.g. cognitive impairment or dementia). We discuss possible solutions including oversampling and other statistical modeling approaches.

Machine Learning in an Elderly Man with Heart Failure

Machine learning is a branch of artificial intelligence and can be used to predict important outcomes in a wide variety of medical conditions. With the widespread use of electronic medical records, the vast amount of data required for this process is now readily available. The following case demonstrates the application of machine learning to an elderly man with heart failure. The algorithms used, namely, decision tree and random forest, both correctly differentiated heart failure with preserved ejection fraction from heart failure with reduced ejection fraction. This has important treatment and prognostic ramifications and can be completed at the point of care while awaiting confirmation via echocardiogram. Viewing the machine learning process through a patient-centered lens, as in this case, highlights the key role we as physicians have in the implementation and supervision of machine learning.

Dementia risks identified by vocal features via telephone conversations: A novel machine learning prediction model

Due to difficulty in early diagnosis of Alzheimer's disease (AD) related to cost and differentiated capability, it is necessary to identify low-cost, accessible, and reliable tools for identifying AD risk in the preclinical stage. We hypothesized that cognitive ability, as expressed in the vocal features in daily conversation, is associated with AD progression. Thus, we have developed a novel machine learning prediction model to identify AD risk by using the rich voice data collected from daily conversations, and evaluated its predictive performance in comparison with a classification method based on the Japanese version of the Telephone Interview for Cognitive Status (TICS-J). We used 1,465 audio data files from 99 Healthy controls (HC) and 151 audio data files recorded from 24 AD patients derived from a dementia prevention program conducted by Hachioji City, Tokyo, between March and May 2020. After extracting vocal features from each audio file, we developed machine-learning models based on extreme gradient boosting (XGBoost), random forest (RF), and logistic regression (LR), using each audio file as one observation. We evaluated the predictive performance of the developed models by describing the receiver operating characteristic (ROC) curve, calculating the areas under the curve (AUCs), sensitivity, and specificity. Further, we conducted classifications by considering each participant as one observation, computing the average of their audio files' predictive value, and making comparisons with the predictive performance of the TICS-J based questionnaire. Of 1,616 audio files in total, 1,308 (81.0%) were randomly allocated to the training data and 308 (19.1%) to the validation data. For audio file-based prediction, the AUCs for XGboost, RF, and LR were 0.863 (95% confidence interval [CI]: 0.794-0.931), 0.882 (95% CI: 0.840-0.924), and 0.893 (95%CI: 0.832-0.954), respectively. For participant-based prediction, the AUC for XGboost, RF, LR, and TICS-J were 1.000 (95%CI: 1.000-1.000), 1.000 (95%CI: 1.000-1.000), 0.972 (95%CI: 0.918-1.000) and 0.917 (95%CI: 0.918-1.000), respectively. There was difference in predictive accuracy of XGBoost and TICS-J with almost approached significance (p = 0.065). Our novel prediction model using the vocal features of daily conversations demonstrated the potential to be useful for the AD risk assessment.

Endogenous opiates and behavior: 2019

This paper is the forty-second consecutive installment of the annual anthological review of research concerning the endogenous opioid system, summarizing articles published during 2019 that studied the behavioral effects of molecular, pharmacological and genetic manipulation of opioid peptides and receptors as well as effects of opioid/opiate agonists and antagonists. The review is subdivided into the following specific topics: molecular-biochemical effects and neurochemical localization studies of endogenous opioids and their receptors (1), the roles of these opioid peptides and receptors in pain and analgesia in animals (2) and humans (3), opioid-sensitive and opioid-insensitive effects of nonopioid analgesics (4), opioid peptide and receptor involvement in tolerance and dependence (5), stress and social status (6), learning and memory (7), eating and drinking (8), drug abuse and alcohol (9), sexual activity and hormones, pregnancy, development and endocrinology (10), mental illness and mood (11), seizures and neurologic disorders (12), electrical-related activity and neurophysiology (13), general activity and locomotion (14), gastrointestinal, renal and hepatic functions (15), cardiovascular responses (16), respiration and thermoregulation (17), and immunological responses (18).

A machine learning approach for the prediction of overall deceased donor organ yield

BACKGROUND

Optimizing organ yield (number of organs transplanted per donor) is a potentially modifiable way to increase the number of organs available for transplant. Models to predict the expected deceased donor organ yield have been developed based on ordinary least squares regression and logistic regression. However, alternative modeling methodologies incorporating machine learning may have superior performance compared with conventional approaches.

METHODS

We evaluated the predictive accuracy of 14 machine learning models for predicting overall organ yield in a cross-validation procedure. The models were parameterized using data from the Organ Procurement and Transplantation Network database from 2000 to 2018. The inclusion criteria for the study were adult deceased donors between 18 and 84 years of age that had at least 1 organ procured for transplantation.

RESULTS

A total of 89,520 donors met the inclusion criteria. Their mean (standard deviation) age was 44 (15) years, and approximately 58% were male. Our cross-validation analysis showed that a tree-based gradient boosting model outperformed the remaining 13 models. Compared with the currently used prediction models, the gradient boosting model improves prediction accuracy by reducing the mean absolute error between 3 and 11 organs per 100 donors.

CONCLUSION

Our analysis demonstrated that the gradient boosting methodology had the best performance in predicting overall deceased donor organ yield and can potentially serve as an aid to assess organ procurement organization performance.

Big data and predictive modelling for the opioid crisis: existing research and future potential

A need exists to accurately estimate overdose risk and improve understanding of how to deliver treatments and interventions in people with opioid use disorder in a way that reduces such risk. We consider opportunities for predictive analytics and routinely collected administrative data to evaluate how overdose could be reduced among people with opioid use disorder. Specifically, we summarise global trends in opioid use and overdoses; describe the use of big data in research into opioid overdose; consider the potential for predictive modelling, including machine learning, for prevention and monitoring of opioid overdoses; and outline the challenges and risks relating to the use of big data and machine learning in reducing harms that are related to opioid use. Future research for improving the coverage and provision of existing interventions, treatments, and resources for opioid use disorder requires collaboration of multiple agencies. Predictive modelling could transport the concept of stratified medicine to public health through novel methods, such as predictive modelling and emulated trials for evaluating diagnoses and prognoses of opioid use disorder, predicting treatment response, and providing targeted treatment recommendations.

Safe opioid prescribing: a prognostic machine learning approach to predicting 30-day risk after an opioid dispensation in Alberta, Canada

OBJECTIVE

To develop machine learning models employing administrative health data that can estimate risk of adverse outcomes within 30 days of an opioid dispensation for use by health departments or prescription monitoring programmes.

DESIGN, SETTING AND PARTICIPANTS

This prognostic study was conducted in Alberta, Canada between 2017 and 2018. Participants included all patients 18 years of age and older who received at least one opioid dispensation. Pregnant and cancer patients were excluded.

EXPOSURE

Each opioid dispensation served as an exposure.

MAIN OUTCOMES/MEASURES

Opioid-related adverse outcomes were identified from linked administrative health data. Machine learning algorithms were trained using 2017 data to predict risk of hospitalisation, emergency department visit and mortality within 30 days of an opioid dispensation. Two validation sets, using 2017 and 2018 data, were used to evaluate model performance. Model discrimination and calibration performance were assessed for all patients and those at higher risk. Machine learning discrimination was compared with current opioid guidelines.

RESULTS

Participants in the 2017 training set (n=275 150) and validation set (n=117 829) had similar baseline characteristics. In the 2017 validation set, c-statistics for the XGBoost, logistic regression and neural network classifiers were 0.87, 0.87 and 0.80, respectively. In the 2018 validation set (n=393 023), the corresponding c-statistics were 0.88, 0.88 and 0.82. C-statistics from the Canadian guidelines ranged from 0.54 to 0.69 while the US guidelines ranged from 0.50 to 0.62. The top five percentile of predicted risk for the XGBoost and logistic regression classifiers captured 42% of all events and translated into post-test probabilities of 13.38% and 13.45%, respectively, up from the pretest probability of 1.6%.

CONCLUSION

Machine learning classifiers, especially incorporating hospitalisation/physician claims data, have better predictive performance compared with guideline or prescription history only approaches when predicting 30-day risk of adverse outcomes. Prescription monitoring programmes and health departments with access to administrative data can use machine learning classifiers to effectively identify those at higher risk compared with current guideline-based approaches.

A Large-Scale Observational Study on the Temporal Trends and Risk Factors of Opioid Overdose: Real-World Evidence for Better Opioids

Background

The USA is in the midst of an opioid overdose epidemic. To address the epidemic, we conducted a large-scale population study on opioid overdose.

Objectives

The primary objective of this study was to evaluate the temporal trends and risk factors of inpatient opioid overdose. Based on its patterns, the secondary objective was to examine the innate properties of opioid analgesics underlying reduced overdose effects.

Methods

A retrospective cross-sectional study was conducted based on a large-scale inpatient electronic health records database, Cerner Health Facts^®, with (1) inclusion criteria for participants as patients admitted between 1 January, 2009 and 31 December, 2017 and (2) measurements as opioid overdose prevalence by year, demographics, and prescription opioid exposures.

Results

A total of 4,720,041 patients with 7,339,480 inpatient encounters were retrieved from Cerner Health Facts^®. Among them, 30.2% patients were aged 65+ years, 57.0% female, 70.1% Caucasian, 42.3% single, 32.0% from the South, and 80.8% in an urban area. From 2009 to 2017, annual opioid overdose prevalence per 1000 patients significantly increased from 3.7 to 11.9 with an adjusted odds ratio (aOR): 1.16, 95% confidence interval (CI) 1.15–1.16. Compared to the major demographic counterparts, being in (1) age group: 41–50 years (overall aOR 1.36, 95% CI 1.31–1.40) or 51–64 years (overall aOR 1.35, 95% CI 1.32–1.39), (2) marital status: divorced (overall aOR 1.19, 95% CI 1.15–1.23), and (3) census region: West (overall aOR 1.32, 95% CI 1.28–1.36) were significantly associated with a higher odds of opioid overdose. Prescription opioid exposures were also associated with an increased odds of opioid overdose, such as meperidine (overall aOR 1.09, 95% CI 1.06–1.13) and tramadol (overall aOR 2.20, 95% CI 2.14–2.27). Examination on the relationships between opioid analgesic properties and their association strengths, aORs, and opioid overdose showed that lower aOR values were significantly associated with (1) high molecular weight, (2) non-interaction with multi-drug resistance protein 1 or interaction with cytochrome P450 3A4, and (3) non-interaction with the delta opioid receptor or kappa opioid receptor.

Conclusions

The significant increasing trends of opioid overdose at the inpatient care setting from 2009 to 2017 suggested an ongoing need for efforts to combat the opioid overdose epidemic in the USA. Risk factors associated with opioid overdose included patient demographics and prescription opioid exposures. Moreover, there are physicochemical, pharmacokinetic, and pharmacodynamic properties underlying reduced overdose effects, which can be utilized to develop better opioids.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40801-021-00253-8.

Data Science Methods for Nursing-Relevant Patient Outcomes and Clinical Processes: The 2019 Literature Year in Review

Data science continues to be recognized and used within healthcare due to the increased availability of large data sets and advanced analytics. It can be challenging for nurse leaders to remain apprised of this rapidly changing landscape. In this article, we describe our findings from a scoping literature review of papers published in 2019 that use data science to explore, explain, and/or predict 15 phenomena of interest to nurses. Fourteen of the 15 phenomena were associated with at least one paper published in 2019. We identified the use of many contemporary data science methods (eg, natural language processing, neural networks) for many of the outcomes. We found many studies exploring Readmissions and Pressure Injuries. The topics of Artificial Intelligence/Machine Learning Acceptance, Burnout, Patient Safety, and Unit Culture were poorly represented. We hope that the studies described in this article help readers: (1) understand the breadth and depth of data science's ability to improve clinical processes and patient outcomes that are relevant to nurses and (2) identify gaps in the literature that are in need of exploration.

Prediction of Prolonged Opioid Use After Surgery in Adolescents: Insights From Machine Learning

BACKGROUND

Long-term opioid use has negative health care consequences. Patients who undergo surgery are at risk for prolonged opioid use after surgery (POUS). While risk factors have been previously identified, no methods currently exist to determine higher-risk patients. We assessed the ability of a variety of machine-learning algorithms to predict adolescents at risk of POUS and to identify factors associated with this risk.

METHODS

A retrospective cohort study was conducted using a national insurance claims database of adolescents aged 12-21 years who underwent 1 of 1297 surgeries, with general anesthesia, from January 1, 2011 to December 30, 2017. Logistic regression with an L2 penalty and with a logistic regression with an L1 lasso (Lasso) penalty, random forests, gradient boosting machines, and extreme gradient boosted models were trained using patient and provider characteristics to predict POUS (≥1 opioid prescription fill within 90-180 days after surgery) risk. Predictive capabilities were assessed using the area under the receiver-operating characteristic curve (AUC)/C-statistic, mean average precision (MAP); individual decision thresholds were compared using sensitivity, specificity, Youden Index, F1 score, and number needed to evaluate. The variables most strongly associated with POUS risk were identified using permutation importance.

RESULTS

Of 186,493 eligible patient surgical visits, 8410 (4.51%) had POUS. The top-performing algorithm achieved an overall AUC of 0.711 (95% confidence interval [CI], 0.699-0.723) and significantly higher AUCs for certain surgeries (eg, 0.823 for spinal fusion surgery and 0.812 for dental surgery). The variables with the strongest association with POUS were the days' supply of opioids and oral morphine milligram equivalents of opioids in the year before surgery.

CONCLUSIONS

Machine-learning models to predict POUS risk among adolescents show modest to strong results for different surgeries and reveal variables associated with higher risk. These results may inform health care system-specific identification of patients at higher risk for POUS and drive development of preventative measures.

Adolescent HIV-related behavioural prediction using machine learning: a foundation for precision HIV prevention

BACKGROUND

Precision prevention is increasingly important in HIV prevention research to move beyond universal interventions to those tailored for high-risk individuals. The current study was designed to develop machine learning algorithms for predicting adolescent HIV risk behaviours.

METHODS

Comprehensive longitudinal data on adolescent risk behaviours, perceptions, peer and family influence, and neighbourhood risk factors were collected from 2564 grade-10 students at baseline followed for 24 months over 2008-2012. Machine learning techniques [support vector machine (SVM) and random forests] were applied to innovatively leverage longitudinal data for robust HIV risk behaviour prediction. In this study, we focused on two adolescent risk behaviours: had ever had sex and had multiple sex partners. Twenty percent of the data were withheld for model testing.

RESULTS

The SVM model with cost-sensitive learning achieved the highest sensitivity, at 79.1%, specificity of 75.4% with AUC of 0.86 in predicting multiple sex partners on the training data (10-fold cross-validation), and sensitivity of 79.7%, specificity of 76.5% with AUC of 0.86 on the testing data. The random forest model obtained the best performance in predicting had ever had sex, yielding the sensitivity of 78.5%, specificity of 73.1% with AUC of 0.84 on the training data and sensitivity of 82.7%, specificity of 75.3% with AUC of 0.87 on the testing data.

CONCLUSION

Machine learning methods can be used to build effective prediction model(s) to identify adolescents who are likely to engage in HIV risk behaviours. This study builds a foundation for targeted intervention strategies and informs precision prevention efforts in school-setting.

Identifying risk of opioid use disorder for patients taking opioid medications with deep learning

OBJECTIVE

The United States is experiencing an opioid epidemic. In recent years, there were more than 10 million opioid misusers aged 12 years or older annually. Identifying patients at high risk of opioid use disorder (OUD) can help to make early clinical interventions to reduce the risk of OUD. Our goal is to develop and evaluate models to predict OUD for patients on opioid medications using electronic health records and deep learning methods. The resulting models help us to better understand OUD, providing new insights on the opioid epidemic. Further, these models provide a foundation for clinical tools to predict OUD before it occurs, permitting early interventions.

METHODS

Electronic health records of patients who have been prescribed with medications containing active opioid ingredients were extracted from Cerner's Health Facts database for encounters between January 1, 2008, and December 31, 2017. Long short-term memory models were applied to predict OUD risk based on five recent prior encounters before the target encounter and compared with logistic regression, random forest, decision tree, and dense neural network. Prediction performance was assessed using F1 score, precision, recall, and area under the receiver-operating characteristic curve.

RESULTS

The long short-term memory (LSTM) model provided promising prediction results which outperformed other methods, with an F1 score of 0.8023 (about 0.016 higher than dense neural network (DNN)) and an area under the receiver-operating characteristic curve (AUROC) of 0.9369 (about 0.145 higher than DNN).

CONCLUSIONS

LSTM-based sequential deep learning models can accurately predict OUD using a patient's history of electronic health records, with minimal prior domain knowledge. This tool has the potential to improve clinical decision support for early intervention and prevention to combat the opioid epidemic.

Enhancement in Performance of Septic Shock Prediction Using National Early Warning Score, Initial Triage Information, and Machine Learning Analysis

BACKGROUND

Several studies reported that the National Early Warning Score (NEWS) has shown superiority over other screening tools in discriminating emergency department (ED) patients who are likely to progress to septic shock.

OBJECTIVES

To improve the performance of the NEWS for septic shock prediction by adding variables collected during ED triage, and to implement a machine-learning algorithm.

METHODS

The study population comprised adult ED patients with suspected infection. To detect septic shock within 24 h after ED arrival, the Sepsis-3 clinical criteria and nine variables were used: NEWS, age, gender, systolic blood pressure, diastolic blood pressure, pulse rate, respiratory rate, body temperature, and oxygen saturation. The model was developed using logistic regression (LR), extreme gradient boosting (XGB), and artificial neural network (ANN) algorithms. The evaluations were performed using an area under the receiver operating characteristic curve (AUROC), Hosmer-Lemeshow test, and net reclassification index (NRI).

RESULTS

Overall, 41,687 patients were enrolled. The AUROC of the model with NEWS, age, gender, and the six vital signs (0.835-0.845) was better than that of the baseline model (0.804). The XGB model (AUROC 0.845) was the most accurate, compared with LR (0.844) and ANN (0.835). The LR and XGB models were well calibrated; however, the ANN showed poor calibration power. The LR and XGB models showed better reclassification than the baseline model with positive NRI.

CONCLUSION

The discrimination power of the model for screening septic shock using NEWS, age, gender, and the six vital signs collected at ED triage outperformed the baseline NEWS model.

Dual-trajectories of opioid and gabapentinoid use and risk of subsequent drug overdose among Medicare beneficiaries in the United States: a retrospective cohort study

BACKGROUND AND AIMS

Little is known about opioid and gabapentinoid (OPI-GABA) use duration and dose patterns' associations with adverse outcome risks. We examined associations between OPI-GABA dose and duration trajectories and subsequent drug overdose.

DESIGN

Retrospective cohort study.

SETTING

US Medicare.

PARTICIPANTS

Using a 5% sample (2011-16), we identified 71 005 fee-for-service Medicare beneficiaries with fibromyalgia, low back pain, neuropathy and/or osteoarthritis initiating OPIs and/or GABAs [mean age ± standard deviation (SD) = 65.5 ± 14.5 years, female = 68.1%, white = 76.8%].

MEASUREMENTS

Group-based multi-trajectory models identified distinct OPI-GABA use patterns during the year of OPI and/or GABA initiation, based on weekly average standardized daily dose (i.e. OPIs = morphine milligram equivalent, GABAs = minimum effective daily dose). We estimated models with three to 12 trajectories and selected the best model based on Bayesian information criterion (BIC) and Nagin's criteria. We estimated risk of time to first drug overdose diagnosis within 12 months following the index year, adjusting for socio-demographic and health factors using inverse probability of treatment weighted multivariable Cox proportional hazards models.

FINDINGS

We identified 10 distinct trajectories (BIC = -1 176 954; OPI-only = 3, GABA-only = 3, OPI-GABA = 4). Compared with OPI-only early discontinuers (40.6% of the cohort), 1-year drug overdose risk varied by trajectory group: consistent low-dose OPI-only users [16.6%; hazard ratio (HR) = 1.47, 95% confidence interval (CI) = 1.19-1.82], consistent high-dose OPI-only users (1.8%; HR = 4.57, 95% CI = 2.99-6.98), GABA-only early discontinuers (12.5%; HR = 1.39, 95% CI = 1.09-1.77), consistent low-dose GABA-only users (11.0%; HR = 1.44, 95% CI = 1.12-1.85), consistent high-dose GABA-only users (3.1%; HR = 1.43, 95% CI = 0.94-2.17), early discontinuation of OPIs and consistent low-dose GABA users (6.9%; HR = 1.24, 95% CI = 0.90-1.69), consistent low-dose OPI-GABA users (3.4%; HR = 2.49, 95% CI = 1.76-3.52), consistent low-dose OPI and high-dose GABA users (3.2%; HR = 2.46, 95% CI = 1.71-3.53) and consistent high-dose OPI and moderate-dose GABA users (0.9%; HR = 7.22, 95% CI = 4.46-11.69).

CONCLUSIONS

Risk of drug overdose varied substantially among US Medicare beneficiaries on different use trajectories of opioids and gabapentinoids. High-dose opioid-only users and all consistent opioid and gabapentinoid users (regardless of doses) had more than double the risk of subsequent drug overdose compared with opioid-only early discontinuers.

Predicting Mortality Risk After a Hospital or Emergency Department Visit for Nonfatal Opioid Overdose

BACKGROUND

Survivors of opioid overdose have substantially increased mortality risk, although this risk is not evenly distributed across individuals. No study has focused on predicting an individual's risk of death after a nonfatal opioid overdose.

OBJECTIVE

To predict risk of death after a nonfatal opioid overdose.

DESIGN AND PARTICIPANTS

This retrospective cohort study included 9686 Pennsylvania Medicaid beneficiaries with an emergency department or inpatient claim for nonfatal opioid overdose in 2014-2016. The index date was the first overdose claim during this period.

EXPOSURES, MAIN OUTCOME, AND MEASURES

Predictor candidates were measured in the 180 days before the index overdose. Primary outcome was 180-day all-cause mortality. Using a gradient boosting machine model, we classified beneficiaries into six subgroups according to their risk of mortality (< 25th percentile of the risk score, 25th to < 50th, 50th to < 75th, 75th to < 90th, 90th to < 98th, ≥ 98th). We then measured receipt of medication for opioid use disorder (OUD), risk mitigation interventions (e.g., prescriptions for naloxone), and prescription opioids filled in the 180 days after the index overdose, by risk subgroup.

KEY RESULTS

Of eligible beneficiaries, 347 (3.6%) died within 180 days after the index overdose. The C-statistic of the mortality prediction model was 0.71. In the highest risk subgroup, the observed 180-day mortality rate was 20.3%, while in the lowest risk subgroup, it was 1.5%. Medication for OUD and risk mitigation interventions after overdose were more commonly seen in lower risk groups, while opioid prescriptions were more likely to be used in higher risk groups (both p trends < .001).

CONCLUSIONS

A risk prediction model performed well for classifying mortality risk after a nonfatal opioid overdose. This prediction score can identify high-risk subgroups to target interventions to improve outcomes among overdose survivors.

Predicting postoperative opioid use with machine learning and insurance claims in opioid-naïve patients

BACKGROUND

The clinical impact of postoperative opioid use requires accurate prediction strategies to identify at-risk patients. We utilize preoperative claims data to predict postoperative opioid refill and new persistent use in opioid-naïve patients.

METHODS

A retrospective study was conducted on 112,898 opioid-naïve adult postoperative patients from Optum's de-identified Clinformatics® Data Mart database. Potential predictors included sociodemographic data, comorbidities, and prescriptions within one year prior to surgery.

RESULTS

Compared to linear models, non-linear models led to modest improvements in predicting refills - area under the receiver operating characteristics curve (AUROC) 0.68 vs. 0.67 (p < 0.05) - and performed identically in predicting new persistent use - AUROC = 0.66. Undergoing major surgery, opioid prescriptions within 30 days prior to surgery, and abdominal pain were useful in predicting refills; back/joint/head pain were the most important features in predicting new persistent use.

CONCLUSIONS

Preoperative patient attributes from insurance claims could potentially be useful in guiding prescription practices for opioid-naïve patients.

Integrating human services and criminal justice data with claims data to predict risk of opioid overdose among Medicaid beneficiaries: A machine-learning approach

Health system data incompletely capture the social risk factors for drug overdose. This study aimed to improve the accuracy of a machine-learning algorithm to predict opioid overdose risk by integrating human services and criminal justice data with health claims data to capture the social determinants of overdose risk. This prognostic study included Medicaid beneficiaries (n = 237,259) in Allegheny County, Pennsylvania enrolled between 2015 and 2018, randomly divided into training, testing, and validation samples. We measured 290 potential predictors (239 derived from Medicaid claims data) in 30-day periods, beginning with the first observed Medicaid enrollment date during the study period. Using a gradient boosting machine, we predicted a composite outcome (i.e., fatal or nonfatal opioid overdose constructed using medical examiner and claims data) in the subsequent month. We compared prediction performance between a Medicaid claims only model to one integrating human services and criminal justice data with Medicaid claims (i.e., integrated model) using several metrics (e.g., C-statistic, number needed to evaluate [NNE] to identify one overdose). Beneficiaries were stratified into risk-score decile subgroups. The samples (training = 79,087, testing = 79,086, validation = 79,086) had similar characteristics (age = 38±18 years, female = 56%, white = 48%, having at least one overdose = 1.7% during study period). Using the validation sample, the integrated model slightly improved on the Medicaid claims only model (C-statistic = 0.885; 95%CI = 0.877–0.892 vs. C-statistic = 0.871; 95%CI = 0.863–0.878), with small corresponding improvements in the NNE and positive predictive value. Nine of the top 30 most important predictors in the integrated model were human services and criminal justice variables. Using the integrated model, approximately 70% of individuals with overdoses were members of the top risk decile (overdose rates in the subsequent month = 47/10,000 beneficiaries). Few individuals in the bottom 9 deciles had overdose episodes (0-12/10,000). Machine-learning algorithms integrating claims and social service and criminal justice data modestly improved opioid overdose prediction among Medicaid beneficiaries for a large U.S. county heavily affected by the opioid crisis.

External validation of an opioid misuse machine learning classifier in hospitalized adult patients

BACKGROUND

Opioid misuse screening in hospitals is resource-intensive and rarely done. Many hospitalized patients are never offered opioid treatment. An automated approach leveraging routinely captured electronic health record (EHR) data may be easier for hospitals to institute. We previously derived and internally validated an opioid classifier in a separate hospital setting. The aim is to externally validate our previously published and open-source machine-learning classifier at a different hospital for identifying cases of opioid misuse.

METHODS

An observational cohort of 56,227 adult hospitalizations was examined between October 2017 and December 2019 during a hospital-wide substance use screening program with manual screening. Manually completed Drug Abuse Screening Test served as the reference standard to validate a convolutional neural network (CNN) classifier with coded word embedding features from the clinical notes of the EHR. The opioid classifier utilized all notes in the EHR and sensitivity analysis was also performed on the first 24 h of notes. Calibration was performed to account for the lower prevalence than in the original cohort.

RESULTS

Manual screening for substance misuse was completed in 67.8% (n = 56,227) with 1.1% (n = 628) identified with opioid misuse. The data for external validation included 2,482,900 notes with 67,969 unique clinical concept features. The opioid classifier had an AUC of 0.99 (95% CI 0.99-0.99) across the encounter and 0.98 (95% CI 0.98-0.99) using only the first 24 h of notes. In the calibrated classifier, the sensitivity and positive predictive value were 0.81 (95% CI 0.77-0.84) and 0.72 (95% CI 0.68-0.75). For the first 24 h, they were 0.75 (95% CI 0.71-0.78) and 0.61 (95% CI 0.57-0.64).

CONCLUSIONS

Our opioid misuse classifier had good discrimination during external validation. Our model may provide a comprehensive and automated approach to opioid misuse identification that augments current workflows and overcomes manual screening barriers.

Predicting opioid overdose risk of patients with opioid prescriptions using electronic health records based on temporal deep learning

The US is experiencing an opioid epidemic, and opioid overdose is causing more than 100 deaths per day. Early identification of patients at high risk of Opioid Overdose (OD) can help to make targeted preventative interventions. We aim to build a deep learning model that can predict the patients at high risk for opioid overdose and identify most relevant features. The study included the information of 5,231,614 patients from the Health Facts database with at least one opioid prescription between January 1, 2008 and December 31, 2017. Potential predictors (n = 1185) were extracted to build a feature matrix for prediction. Long Short-Term Memory (LSTM) based models were built to predict overdose risk in the next hospital visit. Prediction performance was compared with other machine learning methods assessed using machine learning metrics. Our sequential deep learning models built upon LSTM outperformed the other methods on opioid overdose prediction. LSTM with attention mechanism achieved the highest F-1 score (F-1 score: 0.7815, AUCROC: 0.8449). The model is also able to reveal top ranked predictive features by permutation important method, including medications and vital signs. This study demonstrates that a temporal deep learning based predictive model can achieve promising results on identifying risk of opioid overdose of patients using the history of electronic health records. It provides an alternative informatics-based approach to improving clinical decision support for possible early detection and intervention to reduce opioid overdose.

Using machine learning to identify diabetes patients with canagliflozin prescriptions at high-risk of lower extremity amputation using real-world data

AIMS

Canagliflozin, a sodium-glucose cotransporter 2 inhibitor indicated for lowering glucose, has been increasingly used in diabetes patients because of its beneficial effects on cardiovascular and renal outcomes. However, clinical trials have documented an increased risk of lower extremity amputations (LEA) associated with canagliflozin. We applied machine learning methods to predict LEA among diabetes patients treated with canagliflozin.

METHODS

Using claims data from a 5% random sample of Medicare beneficiaries, we identified 13 904 diabetes individuals initiating canagliflozin between April 2013 and December 2016. The samples were randomly and equally split into training and testing sets. We identified 41 predictor candidates using information from the year prior to canagliflozin initiation, and applied four machine learning approaches (elastic net, least absolute shrinkage and selection operator [LASSO], gradient boosting machine and random forests) to predict LEA risk after canagliflozin initiation.

RESULTS

The incidence rate of LEA was 0.57% over a median 1.5 years follow-up. LASSO produced the best prediction, yielding a C-statistic of 0.81 (95% CI: 0.76, 0.86). Among individuals categorized in the top 5% of the risk score, the actual incidence rate of LEA was 3.74%. Among the 16 factors selected by LASSO, history of LEA [adjusted odds ratio (aOR): 33.6 (13.8, 81.9)] and loop diuretic use [aOR: 3.6 (1.8,7.3)] had the strongest associations with LEA incidence.

CONCLUSIONS

Our machine learning model efficiently predicted the risk of LEA among diabetes patients undergoing canagliflozin treatment. The risk score may support optimized treatment decisions and thus improve health outcomes of diabetes patients.