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.

Interrupted time series analysis of drug overdose fatalities in service-related industries versus non-service-related industries during the COVID-19 pandemic, 2018-2021.

BACKGROUND: Variation among industries in the association between COVID-19-related closing or reopening orders and drug overdose deaths is unknown. The objectives of this study were to compare drug overdose decedent demographics, annual drug overdose fatality rates and monthly drug overdose fatality rates by specific industry within the service-related industry sector, and to perform an interrupted time series analysis comparing weekly drug overdose mortality counts in service-related and non-service-related industries, examining the COVID-19 pre-pandemic and pandemic phases by Kentucky closing and reopening orders. METHODS: Kentucky drug overdose death certificate and toxicology testing data for years 2018-2021 were analysed using Χ2 and interrupted time series methods. RESULTS: Before the pandemic, annual drug overdose fatality rates in service-related industries were higher than in non-service-related industries. However, these trends reversed during the pandemic. Both service-related and non-service-related industry groups experienced increased fatal drug overdoses at change points associated with the gubernatorial business closure orders, although the magnitude of the increase differed between the two groups. Young, female and black workers in service-related industries had higher frequencies of drug overdose deaths compared with decedents in the non-service-related industries. CONCLUSION: Spikes in drug overdose mortality in both service-related and non-service-related industries during the pandemic highlight the need to consider and include industries and occupations, as well as worker populations vulnerable to infectious diseases, as integral stakeholder groups when developing and implementing drug overdose prevention interventions, and implementing infectious disease surveillance systems.

Trends in ICD-10-CM-Coded Administrative Datasets for Injury Surveillance and Research.

OBJECTIVES: Accurate injury surveillance depends on data quality in administrative datasets created for billing and reimbursement. Significant effort has been devoted to testing the ability of candidate injury case definitions to identify injury cases accurately in these datasets. We used interviews with experienced coders, informed by a review of the current literature, to identify three clinical coding trends that may affect the consistency of surveillance data: "clinical documentation improvement or clinical documentation integrity" (CDI), coding by treating clinicians, and certain electronic health record features. METHODS: An extensive literature review informed interviews with coding experts to identify potential issues in coding practice. To determine whether physician coding was associated with information loss, we analyzed data from two hospitals serving the same geographic area. One hospital had used physician coding of emergency department data for the past decade; the other used professional coders. We compared the proportion of emergency department records missing external cause of injury codes and assessed the variation for statistical significance. RESULTS: CDI audits review patient records to ensure that billing information includes every relevant International Classification of Diseases, Tenth Revision, Clinical Modification code. This approach has increased payment rates awarded to Medicare Advantage plans because additional codes increase the patient acuity level and case mix index. The impact of CDI audits on injury data needs further investigation. The pilot analysis addressing information loss with physician coding found a higher level of external cause coding with clinician self-coding, possibly because of the coding software. Finally, widespread "copy and paste" in patient electronic health records has the potential to increase reported injuries. CONCLUSIONS: Injury surveillance relies on billing and reimbursement records. Financial motivations may interfere with the consistency of surveillance findings and mislead injury epidemiologists. Further investigation is essential to ensure the integrity of surveillance findings.