Deep Neural Networks for Fine-Grained Surveillance of Overdose Mortality

Surveillance of drug overdose deaths relies on death certificates for identification of the substances that caused death. Drugs and drug classes can be identified through the International Classification of Diseases, Tenth Revision (ICD-10), codes present on death certificates. However, ICD-10 codes do not always provide high levels of specificity in drug identification. To achieve more fine-grained identification of substances on death certificate, the free-text cause-of-death section, completed by the medical certifier, must be analyzed. Current methods for analyzing free-text death certificates rely solely on lookup tables for identifying specific substances, which must be frequently updated and maintained. To improve identification of drugs on death certificates, a deep-learning named-entity recognition model was developed, utilizing data from the Kentucky Drug Overdose Fatality Surveillance System (2014-2019), which achieved an F1-score of 99.13%. This model can identify new drug misspellings and novel substances that are not present on current surveillance lookup tables, enhancing the surveillance of drug overdose deaths.

Changing Trends in Drug Overdose Mortality in Kentucky: An Examination of Race and Ethnicity, Age, and Contributing Drugs, 2016-2020

OBJECTIVES

Increased drug overdose mortality among non-Hispanic Black people in the United States in the past 5 years highlights the need for better tailored programs and services. We evaluated (1) changes in drug overdose mortality for various racial and ethnic groups and (2) drug involvement to inform drug overdose prevention efforts in Kentucky.

METHODS

We used Kentucky death certificates and postmortem toxicology reports from 2016-2020 (provisional data) to estimate changes in age-adjusted drug overdose death rates per 100 000 standard population.

RESULTS

The age-adjusted drug overdose death rate per 100 000 standard population among non-Hispanic Black residents doubled from 2016 (21.2) to 2020 (46.0), reaching the rate among non-Hispanic White residents in 2020 (48.7; P = .48). From 2016 to 2020, about 80% of these drug overdose deaths involved opioids; heroin involvement declined about 20 percentage points; fentanyl involvement increased about 30 percentage points. The number of psychostimulant-involved drug overdose deaths increased 513% among non-Hispanic Black residents and 191% among non-Hispanic White residents. Cocaine-involved drug overdose deaths increased among non-Hispanic Black residents but declined among non-Hispanic White residents. Drug overdose death rates were significantly lower among Hispanic residents than among non-Hispanic White residents.

CONCLUSIONS

Increased opioid-involved overdose deaths among non-Hispanic Black residents in Kentucky in combination with rapidly expanding concomitant psychostimulant involvement require increased understanding of the social, cultural, and illicit market circumstances driving these rapid trend changes. Our findings underscore the urgent need to expand treatment and harm reduction services to non-Hispanic Black residents with substance use disorder.

Driver injuries in heavy vs. light and medium truck local crashes, 2010-2019

OBJECTIVE

Multiple heavy truck driver injury studies exist, but there is a paucity of research on light and medium truck driver injuries. The objective of this study was to use first report of injury (FROI) data to: (a) compare demographic and injury characteristics; (b) assess workers' compensation (WC) claim disposition and lost work time status; and (c) describe injury scenarios by vehicle type for heavy truck and light/medium truck driver local crashes.

METHOD

Kentucky Department of Workers' Claims FROI quantitative and free text data were analyzed for years 2010-2019. Of 800 total FROIs, 451 involved heavy trucks and 349 involved light or medium trucks.

RESULTS

There was a higher light/medium truck driver crash FROI rate compared to the heavy truck driver crash FROI rate. There was a higher proportion of younger light/medium truck driver crash FROIs compared to younger heavy truck driver crash FROIs. The retail trade industry made up the largest percentage of light/medium truck local crash FROIs (47%); the transportation and warehousing industry was most frequently cited in heavy truck FROIs (46%). The heavy truck types most frequently identified in FROIs were semi-trucks (13%) and dump trucks (11%). The most common light/medium truck type identified was delivery trucks (30%). Most commonly, heavy truck crash FROIs involved rollovers, driving off/overcorrecting on narrow roadways, and driving downhill/unable to downshift. Light/medium truck crash FROIs most frequently involved being rear-ended, running red lights, and turning in front of other vehicles.

CONCLUSIONS

The utilization of WC FROI data highlighted top injury scenarios and specific vehicle types for targeting driver safety training among truck drivers, particularly light/medium truck drivers. Road safety policies regarding driver training, crash reviews, and in-vehicle monitoring systems are needed for truck drivers with previous crash injuries, especially for light and medium truck drivers.

PRACTICAL APPLICATIONS

Enhanced safety training on speeding on narrow roadways, on nearing intersections, and on downshifting on hills is needed for semi-truck, dump truck, and coal truck drivers with previous crash injuries. Rear-end crash prevention training (e.g., gradual stopping and checking mirrors) is needed for drivers of furniture, automotive parts and accessories, and groceries and soft drink delivery trucks with previous crash injuries.

Identification of work-related injury emergency department visits using International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes

INTRODUCTION

Emergency department (ED) visit discharge data are a less explored population-based data source used to identify work-related injuries. When using discharge data, work-relatedness is often determined by the expected payer of workers' compensation (WC). In October 2015, healthcare discharge data coding systems transitioned to the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). ICD-10-CM's structure offers potential new work-related codes to enhance work-related injury surveillance. This study identified work-related ED visits using relevant ICD-10-CM work-related injury codes. Cases identified using this method were compared with those identified using the WC expected payer approach.

METHODS

State ED visit discharge data (2016-2019) were analysed using the CDC's discharge data surveillance definition. Injuries were identified using a diagnosis code or an external cause-of-injury code in any field. Injuries were assessed by mechanism and expected payer. Literature searches and manual review of ICD-10-CM codes were conducted to identify possible work-related injury codes. Descriptive statistics were performed and assessed by expected payer.

RESULTS

WC was billed for 87 361 injury ED visits from 2016 to 2019. Falls were the most frequent injury mechanism. The 246 ICD-10-CM work-related codes identified 36% more work-related ED injury visits than using WC as the expected payer alone.

CONCLUSION

This study identified potential ICD-10-CM codes to expand occupational injury surveillance using discharge data beyond the traditional WC expected payer approach. Further studies are needed to validate the work-related injury codes and support the development of a work-related injury surveillance case definition.