Prevalence Estimation
Prevalence Estimation
One challenge in adequately responding to the opioid syndemic is the lack of data on the local prevalence of opioid misuse. National surveys are able to estimate prevalence of misuse at the state level but are generally unable to generate estimates for geographic areas smaller than states. However, we know that the prevalence of opioid misuse varies within states depending on many social and structural factors. To ensure resources and public health interventions are adequately and equitably distributed, local estimates of opioid misuse prevalence are needed. Due to the difficulty in quantifying opioid misuse in local areas, we are left with surveillance data sources that while easier to obtain, only provide indirect information about opioid misuse. These surveillance data are often collected by state and local public health agencies and are sometimes presented on publicly accessible dashboards.
The objective of our research is to take advantage of existing surveillance data sources to estimate the local prevalence of opioid misuse. Through integration of multiple data sources, we can synthesize the collection of indirect information and infer the underlying prevalence of opioid misuse. When individually linked data are available, we can apply techniques for capture-recapture estimation. More commonly, only aggregate data are available due to government restrictions on data sharing, which prevents application of capture-recapture methods. To overcome this limitation, we developed an integrated abundance model to estimate prevalence when only aggregate data are available.
Capture-recapture is an indirect estimation technique originally applied in ecology to estimate the population size of animal species. Fundamentally, the approach involves taking multiple samples, identifying individuals in each sample, and then determining which individuals were recaptured by multiple samples. When using administrative data, each sample is a list of individuals appearing in a data source. Each data source must have a common unique identifier so it can be determined if the same individual appears in multiple lists. Through special legislation, Massachusetts has created an individually linked Public Health Data Warehouse managed by the Massachusetts Department of Public Health. This data resource has facilitated the application and development of capture-recapture methods for the estimation of opioid misuse prevalence.
Wang J, Doogan N, Thompson K, Bernson D, Feaster D, Villani J, Chandler R, White LF, Kline D, Barocas JA. Massachusetts Prevalence of Opioid Use Disorder Estimation Revisited: Comparing a Bayesian Approach to Standard Capture–Recapture Methods. Epidemiology. 2023;34(6):841. doi:10.1097/EDE.0000000000001653
Wang J, Bernson D, Erdman EA, Villani J, Chandler R, Kline D, White LF, Barocas JA. Intersectional inequities and longitudinal prevalence estimates of opioid use disorder in Massachusetts 2014–2020: a multi-sample capture-recapture analysis. The Lancet Regional Health - Americas. 2024;32:100709. doi:10.1016/j.lana.2024.100709
Wang J, Kline DM, White LF. On the estimation of population size—A comparison of capture-recapture and multiplier-benchmark methods. Stat Methods Med Res. Published online September 30, 2024:09622802241275413. doi:10.1177/09622802241275413
Public health surveillance data are most commonly available in aggregate form. Since we cannot identify and link individuals, we cannot use capture-recapture methods. In ecology, abundance models were developed to estimate population size when only aggregate counts are available and rely on replication to identify model parameters.
We developed a Bayesian integrated abundance model to extend the ideas from ecology to estimate the prevalence of opioid misuse. Without replication, we rely on the integration of multiple data sources at multiple spatial scales to create "pseudo-replication" that enables identification of the model. Within this framework, each observed surveillance outcome is assumed to have imperfect detection. That is, each observed outcome is a proportion of the true unobserved population of interest. For example, overdose deaths reflect the number of people who misuse opioids who died, but many people did not die and so are not reflected in that outcome. If we think of each surveillance outcome like a diagnostic test for opioid misuse, then they all suffer from false negatives. Abundance models are designed to account for imperfect detection (i.e., false negatives) and enable estimation of prevalence relying on only aggregate data.
The ability to flexibly use aggregate data is an important innovation because it enables estimation in the vast majority of states that only have aggregate data available. Even more importantly, because aggregate data are available for sub-state areas (i.e., counties), we can use our integrated abundance model to generate local estimates of opioid misuse prevalence that can better inform the public health response to the opioid epidemic.
See an invited commentary in AJE on our work.
Hepler SA, Kline DM, Bonny A, McKnight E, Waller LA. An integrated abundance model for estimating county-level prevalence of opioid misuse in Ohio. Journal of the Royal Statistical Society Series A: Statistics in Society. 2023;186(1):43-60. doi:10.1093/jrsssa/qnac013
Santaella-Tenorio J, Hepler SA, Rivera-Aguirre A, Kline DM, Cerda M. Estimation of opioid misuse prevalence in New York State counties, 2007-2018. A Bayesian spatio-temporal abundance model approach. American Journal of Epidemiology. Published online March 6, 2024:kwae018. doi:10.1093/aje/kwae018
Kline DM, White BN, Lancaster KE, Egan KL, Murphy E, Miller WC, Hepler SA. Estimating prevalence of opioid misuse in North Carolina counties from 2016-2021: An integrated abundance model approach. Epidemiology. Published online 2025:10.1097/EDE.0000000000001838. doi:10.1097/EDE.0000000000001838
There are many challenging public health research questions related to the opioid syndemic. Many of these questions have been made more difficult because of the lack of local estimates of opioid misuse prevalence. Using estimates from our integrated abundance model, we can more directly address questions related to the allocation of resources, access to services, and impacts of policy.
McKnight ER, Dong Q, Brook DL, Hepler SA, Kline DM, Bonny AE. A Descriptive Study on Opioid Misuse Prevalence and Office-Based Buprenorphine Access in Ohio Prior to the Removal of the Drug Addiction Treatment Act of 2000 Waiver. Cureus. 2023;15(3):e36903. doi:10.7759/cureus.36903
Santaella-Tenorio J, Rivera-Aguirre A, Hepler S, Kline DM, Cantor J, DeYoreo M, Martins SS, Krawczyk N, Cerda M. Rates of receiving medication for opioid use disorder and opioid overdose deaths during the early synthetic opioid crisis: a county-level analysis. Epidemiology. Published online 2024:10.1097/EDE.0000000000001816. doi:10.1097/EDE.0000000000001816