Rudolph et al. Respond to “Mathematization of Epidemiology”

This article is linked to "Optimally Choosing Medication Type for Patients With Opioid Use Disorder" and "Invited Commentary: On the Mathematization of Epidemiology as a Socially Engaged Quantitative Science" (https://doi.org/10.1093/aje/kwac217 and https://doi.org/10.1093/aje/kwad010). We appreciate the opportunity to respond to Cartus and Marshall’s (1) commentary on our paper (2). First, we would like to thank them for their kind feedback and their excellent discussion of numerous structural factors that impact the opioid epidemic and, in particular, the suboptimal and inequitable treatment of opioid use disorder (OUD). We wholeheartedly agree that upstream changes at the national, state, local, and clinical levels (like the recent repeal of the X-waiver requirement for buprenorphine prescribing) are sorely needed to improve access to and quality of medication for opioid use disorder (MOUD) treatments to all who could benefit. Cartus and Marshall raise an interesting set of issues regarding the potential trade-offs between investing in technical sophistication and public health considerations. We are sympathetic to the concern that the broader social context, policy environment, and general public health relevance may sometimes be considered secondary to the technical goals of a particular analysis. However, we do not see technical sophistication and public health relevance as a zero-sum game. In fact, we believe that technical sophistication can, in certain cases, even help clarify public health relevance. We can elaborate on the above with an example from our paper (2) where the authors argue that the “tension between methodological sophistication and practical considerations emerges clearly” (1, p. XXX): our decision to discard the subset of buprenorphine patients who received buprenorphine on a methadone schedule (i.e., attending clinic daily to receive medication instead of receiving a supply of medication every 1–6 weeks). First, we want to point out that community-based comparative effectiveness trials, such as those conducted by the National Institute on Drug Abuse Clinical Trials Network and included in our paper, do not compare medications in a vacuum but instead compare treatment regimens—of which medication is a key component but not the only component. In the set of trials we considered, there were 2 general types of buprenorphine regimens: buprenorphine treatment in a methadone setting and buprenorphine treatment in the typical office-based setting. (This issue can also be thought of as multiple versions of treatment.) Combining the subset of patients receiving buprenorphine in the methadone setting with the buprenorphine patients from the other 2 trials would have violated an assumption required for identifying our causal quantity of interest from the data, although we could have retained this group as a separate treatment regimen. Estimating a causal quantity (also called a causal estimand) that is not identified from the observed data has questionable utility, as the resulting quantity may be too bias-ridden to be meaningful. Lack of technical sophistication may miss this potential identifiability complication. Additionally, in this particular case, including a subset of people who received a regimen of buprenorphine treatment that does not commonly exist in the real world, and will never routinely exist given the patient-centered barrier of undue burden, also would not add practical value to the study. Thus, in this case, the causally principled decision based on identification assumptions and the practically principled decision based on patient care seem aligned. Causally principled methods do not inherently “obscure social context” or obviate a “public health question” (1, p. XXX). Instead, they can clarify what quantity can be identified and estimated and under what causal and statistical assumptions. It is critical, however, to consider the social, political, and historical contexts in which the data are situated and which produced the data, so that the resulting algorithms and/or findings based on that data can be imbued with appropriate interpretations. This challenge has been discussed in the algorithmic fairness/unfairness literature: (3–5) “Data is frequently imperfect in ways that allow these algorithms to inherit the prejudices of prior decision makers…and may reflect the widespread biases that persist in society at large” (5, p. 671). Bringing a formal discussion of algorithmic fairness to treatment rules learned for optimizing the treatment of OUD or other related substance use disorders is an important endeavor that could shape future work in this area. Finally, we agree with Cartus and Marshall on the need to focus on other OUD patient-centered outcomes and on the need for structural improvements to policies and clinical practice to improve the MOUD treatment landscape. As epidemiologists, our work can inform such improvements. However, work that is not rigorous, reproducible, or methodologically sound, or is divorced from the contexts giving rise to the data, may be more hindrance than help. For our work to be of value, it must answer practical, relevant questions and answer them well. Some research questions can be answered “well” without technical or methodological sophistication. Others may require it. Fortunately, instead of detracting from public health relevance, methodologic sophistication has the potential to clarify and improve it (4). Author affiliations: Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York, United States (Kara E. Rudolph, Nicholas T. Williams); Department of Population Health, New York University Grossman School of Medicine, New York, New York, United States (Iván Díaz); Department of Psychiatry, School of Medicine, Columbia University, New York, New York, United States (Sean X. Luo, Edward V. Nunes); New York State Psychiatric Institute, Columbia University, New York, New York, United States (Sean X. Luo, Edward V. Nunes); and Department of Psychiatry, New York University Grossman School of Medicine, New York, New York, United States (John Rotrosen). Conflict of interest: none declared.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.