Santaella-Tenorio et al. respond to: Re: Estimation of opioid misuse prevalence in New York State counties, 2007-2018. A Bayesian spatio-temporal abundance model approach

Le résumé fourni par la source

Bradley et al.1 discuss our article in which we estimate opioid misuse prevalence in New York State (NYS) using a spatiotemporal abundance model approach.2 We appreciate their thoughtful comments and take this opportunity to address the 5 main issues they posed. First, Bradley et al. correctly highlight that International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes 40–44, ×60–64, ×85, and Y10–14 should not be used in the emergency department visit data. We acknowledge this oversight in our data description. The emergency department data comprised only T40.0-T40.4 and T40.6 codes and excluded opioid “adverse effect” or “sequela” visits (T400X5 and T400X6, respectively). Second, Bradley et al., indicate that estimates from the National Survey on Drug Use and Health (NSDUH) are likely biased because NSDUH excludes individuals experiencing houselessness or incarceration and has a disproportionately high level of nonresponse among persons with substance use disorder. We agree with Bradley et al. and noted this limitation in our study: “Fifth, household survey estimates usually exclude populations of individuals who are incarcerated or experiencing homelessness, which underestimates the state prevalence of opioid misuse.”2 It is important to note that NSDUH does include homeless populations living in shelters and that NSDUH calibrates weights to account for nonresponse and poststratification, which may partly compensate for nonresponse. Our methodology integrates NSDUH data on state-level prevalence of people who misuse opioids to overcome challenges with identifiability, because person-level linkage is not possible across multiple data systems in NYS. As Bradley et al. indicate in their letter,1 this is a common issue in most states, including NYS. Although NSDUH data are not perfect, this is the only survey that provides annual estimates of opioid misuse for NYS. Also, relative risk estimates are generally robust to bias or misspecification of intercepts in abundance models, such as the one we used in our study.3 Third, Bradley et al. indicate that statewide opioid misuse prevalence estimates from the model are consistently lower than the sum of county-level estimates. The NSDUH estimates have error and are incorporated into the model with error, so the model would not necessarily center on the point estimates. The NSDUH estimates are multiyear estimates (blue bars in Figure 1 in our original article2), which smooths any temporal trend and leads to reduced uncertainty due to temporal averaging. If NSDUH yearly estimates were available, these would have had wider bounds, and our average model-based rates would likely have overlapped with the bounds from the NSDUH estimates. State-level estimates. Fourth, Bradley et al. are concerned that the summed county-level opioid misuse prevalence estimates included persons younger than 12 years. Our count estimates likely reflect primarily individuals aged 12 years or older. The NSDUH estimates used were for those aged 12 years or older. Also, the aggregated outcome data used likely were mostly from people in that age group. Methadone access is generally restricted to those aged 18 years or older, with some regulatory exceptions, and overdose deaths and emergency visits are rare in those younger than 12 years and unlikely to be due to opioid misuse. Finally, we fully agree with Bradley et al. that studies should be conducted in collaboration with state public health and substance use agency partners when possible. These partnerships can greatly improve the quality and impact of this research. We are committed to find opportunities to more closely work with the Office of Addiction Services and Supports and other state agency partners in future research. These collaborations can generate appropriate evidence based on granular data and information on populations that may be most at risk of opioid misuse and overdoses, which is crucial for advancing research on this subject effectively. None declared. The authors declare no conflicts of interest.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.