Timely Overdose Death Reporting Is Challenging but We Must Do Better

Cause of death determinations for overdoses can require months or, in some cases, more than a year.¹ Such delays impede timely surveillance and action. Rossen and colleagues^1,2 have proposed using historical data to determine degrees of underreporting and then applying multiplication factors to upweight known deaths to estimated final counts. In this issue of AJPH, Friedman and Akre (p. 1284) used these provisional overdose death estimates to report monthly overdose deaths in 2020. They found that the greatest number of overdose deaths ever recorded in the United States occurred in mid-2020 during the COVID-19 pandemic.

We have long worked with provisional death records in North Carolina, with a focus on overdoses. Here we offer a context on the challenges of timely overdose death counting, corroborate the findings of Friedman and Akre, and share strategies to address data delays.

Readers may not know how a death record comes to be “provisional.” Details differ by state; we offer the following summary for North Carolina. Both before and long after a death, data are generated and collected via medicolegal, public health, and community response systems and are used to assist in death categorization. The death certificate, once completed by the declaring physician, is submitted to centralized vital records. There it is coded into an electronic system housed at the North Carolina State Center for Health Statistics. Literal text fields are standardized to International Classification of Diseases, 10th Revision (ICD-10) codes, both manually and with software.³ Public health personnel use these data to understand disease burdens and to prevent future deaths.

Complex causes of death such as overdoses may require special investigations, autopsies, or substance and specimen testing. These deaths may be temporarily assigned an ICD-10 code of R99 (ill-defined and unknown cause of mortality) to indicate that data are pending. It is these pending-data deaths, many later classified as overdoses, that delay timely death reporting.

LIMITATIONS

We agree with Friedman and Akre that the central limitation is that 2020 overdose death estimations are based on historical, prepandemic overdose death patterns. Reporting lags may have differed during the pandemic, leading to underestimated results. Although state data completeness was trending up nationwide,¹ pandemic challenges may have led to more incomplete data in 2020. It also is possible that the proportion of data-pending deaths anticipated to become overdose deaths changed during the pandemic. We do not yet have data on whether some deaths with R99 codes were newly related to COVID-19.

In addition, death information flows through multiple separated teams and data system silos, any of which may be down, upgrading, or changing. For example, the North Carolina State Center for Health Statistics is transitioning to an electronic death reporting system, and the state’s Office of the Chief Medical Examiner is simultaneously upgrading its centralized data system. Medicolegal and public health staff, essential workers already overburdened and underresourced as a result of the overdose epidemic, have been doubly stretched by increases in COVID-19 deaths. These changes and challenges have caused delays in death record processing.

Even with these limitations, we have confidence in Friedman and Akre’s conclusions. In line with the confirmatory emergency medical service trends they share, we have seen similar spikes in overdoses in syndromic surveillance emergency department data⁴ that support their findings. Our team reviewing R99-coded deaths found similar spikes and longer lag times. Friedman and Akre suggest that this may indicate their record-breaking counts are still underestimates.

TAKING ACTION FOR BETTER DATA

Timely counting of overdose deaths is critical. What can be done about these delays? We echo Friedman and Akre’s call for both staff and infrastructure funding, and offer several approaches, as outlined subsequently.

Partnerships

Data partnerships with internal and external organizations have facilitated funding, research, timely reporting, action, and change across many sectors. Successful partnerships promote communication, transparency, understanding of others’ scope of practice, and assistance in timely, conscious use of provisional data.

Faster Sources of Death Data

Other data can be used to circumvent delays. For example, states can add suspected overdose checkboxes to death certificates,⁵ and medical examiners can flag suspected overdoses at first contact (i.e., within days of death). Literal text from timely emergency medical service and law enforcement reports, patient electronic medical records, and medical examiner investigation reports can provide additional details on the substances involved.

Faster Sources of Overdose Data

Overdose-related indicators exist across dozens of data sources, including substance abuse treatment visits, calls requesting treatment resources, prescription and dispensing data, overdose reversal reports, and social service data. Data lags vary, but near-real-time emergency department syndromic surveillance systems can identify accelerating trends in overdoses within days. Data dashboards such as that in North Carolina⁶ can aggregate indicators and local programs, providing more timely awareness than offered through death counts alone.

Flow Diagrams

Building Unified Modeling Language activity diagrams⁷ can help map the interrelated flow of data and teamwork after an overdose death. The process of building an accurate activity diagram invariably requires collaboration and interviews to identify delay points caused by overwork, underfunding, or antiquated systems. This collaboration and mapping catalyzes change.

Infrastructure and Informatics Approaches

We echo Friedman and Akre’s call for investments in public health infrastructure. Public health has called for a modernization of our many antiquated systems with the “Data: Elemental to Health” campaign.⁸ Infrastructure improvements are not constrained to back-end databases and front-end visualization tools, but also include robust connections between systems and “bench science” lab machines to prepare and test samples quickly. Machine learning approaches can play a role; we have found that incorporating demographic data helps predict the percentage of pending-data deaths that will become overdoses.

Interoperability and Accessibility

We agree with Friedman and Akre that granular data to detect sharp overdose spikes are needed. Public Health 3.0⁹ calls for interoperable access to timely, granular, and actionable data, including novel data sources such as modeled estimates. The Centers for Disease Control and Prevention could provide these modeled monthly and rolling estimates on its Web site, where many data sources are already provided in easily accessible formats. However, we have also experienced hesitancy in integrating and communicating modeled data. Those already challenged with differentiating quality, lag, and biases among data sources may be further challenged in combining “true” and modeled data. However, the line between modeled and unmodeled data is often gray; integrating modeled estimates may promote increased data literacy.

CONCLUSIONS

We share Friedman and Akre’s call to increase the response to the ongoing overdose epidemic in the United States, even during the COVID-19 pandemic. They aptly endorse both downstream harm reduction and upstream structural responses; we concur.

Moreover, the overdose epidemic in the United States and the COVID-19 pandemic are not simply unrelated, simultaneous events. They may be “syndemics,”¹⁰ co-occurring at higher frequencies in underresourced communities with mutually deleterious effects. They share respiratory system suppression as a mechanism of death, and COVID-19 survivors may experience chronic pain that can lead to overdose. Increases in unemployment, financial burdens, overworked front-line workers, and long-term COVID-19-associated illnesses have likely had a negative impact on the overdose epidemic, compounding decreased access to mental, social, and medical services.

Retrospective funding later in an epidemic helps, but this approach leaves public health perpetually playing catch up. Public health funding, staffing, and informatics infrastructure must be sufficiently and proactively supported to attain and maintain the data timeliness needed for response as epidemics develop.