Partially randomized preference trials in carceral health research: a methodology to advance equity and inclusion Authors: Justin Berk, Mike Gordon, Margaret A Handley, Helen E Jack Published: 2026 ABSTRACT AbstractBackgroundIncarcerated people experience disproportionately poor health outcomes yet often remain excluded from clinical research due to ethical concerns of limited autonomy and coercion. As a result, traditional randomized controlled trials face significant challenges in carceral settings, driven by mistrust of the medical system, randomization limiting choice, and logistical barriers with blinding and masking. These challenges necessitate alternative methodologies that balance scientific rigor while empowering participant choice.Main textThis commentary proposes partially randomized preference trials as a methodology to advance equity and inclusion in carceral health research. In these hybrid designs, participants with strong preferences choose their intervention, while those without strong preferences are randomized. This approach facilitates participant autonomy while maintaining causal inference capabilities. Further, the design allows researchers to quantify preference and selection effects, enhancing external validity (generalizability), and can increase patient recruitment and retention. We discuss practical considerations including sample size calculations and data analysis approaches that estimate treatment effects, selection bias, and preference bias.ConclusionsPartially randomized preference trials are a way to generate rigorous evidence in carceral settings while respecting participant autonomy. They may also be useful for engaging other communities with limited autonomy. This approach should be considered as a tool for advancing health equity and inclusion in research.Keywords: Partially randomized preference trials, RCTs, Carceral health, Health equity, Vulnerable populations, Incarcerated persons, Clinical trials Background Incarcerated individualsare one of society’s most marginalized and underserved groups [1]. With disproportionately high rates of chronic disease, mental illness, and substance use disorders, this population urgently needs improved care. Rigorous research to develop and evaluate interventions tailored to this population should contribute to improving care and outcomes. Yet carceral facilities remain largely excluded from clinical trials, national datasets, and health services research, creating a harmful paradox wherein those with the greatest health needs have the least access to evidence-based interventions [2, 3]. The exclusion of incarcerated people from research stems from multiple barriers: concerns about coercion among individuals with limited autonomy, complex regulatory requirements, institutional resistance, and reluctance from researchers wary of contributing to the long legacy of unethical exploitation of incarcerated populations [4–6]. This exclusion is particularly harmful because incarcerated populations embody significant intersectionality of disadvantage: racial and ethnic minorities are vastly overrepresented, most have experienced poverty, mental illness and substance use disorders are prevalent, and educational and health literacy levels are often low [1]. Arguably, the barriers to research have tipped the balance between protection and participation too far toward exclusion of a population that could benefit from research. Growing consensus supports expanded access to ethically conducted carceral health research [7, 8]. Randomization is ethically justified by the principle of clinical equipoise (i.e., the uncertainty, shared by the investigator and ideally by the participant, about which treatment is superior). In most settings, equipoise is assumed when participants enroll without strong treatment preferences. In carceral settings, however, this calculus is complicated: incarcerated individuals may suppress or express preferences not out of genuine uncertainty, but in response to coercion, power asymmetries, or fear of institutional consequences. Conversely, they may express strong preferences based on prior experience with particular treatments—experience that deserves to be honored in a historically exploitative setting with a vulnerable population, rather than overridden by randomization. This commentary proposes that partially randomized preference trials (PRPTs) in carceral health research offer a pragmatic methodology to advance inclusion when gathering critical data about evidence-based interventions for participants in this setting. PRPTs balance the scientific rigor needed for causal inference about interventions with the ethical imperative to honor participant autonomy—a particularly critical consideration for populations with limited freedom. PRPTs accommodate both possibilities: a randomized cohort consists of participants for whom genuine uncertainty holds and randomization is therefore ethically sound, while the preference cohort comprises those for whom equipoise (for whatever reason) does not apply. To be sure, we are not arguing that traditional RCT designs are unethical in carceral settings, but rather that PRPTs are an important ethical and scientific tool in this context. In particular, PRPTs offer an additional understanding of preferences which may be important data to gather to inform subsequent implementation in this setting, as carceral settings, where patient participation may not be captured in other ways (e.g., demand for non-formulary therapies). Research participation barriers in carceral settings Randomized controlled trials (RCTs) are considered the gold standard for establishing treatment efficacy, yet they have been described as a “Sisyphean” task in jails and prisons [9]. Typical limitations of RCTs can become amplified in carceral environments. For example, blinding and masking are especially difficult in shared living spaces where participants observe one another’s treatment conditions, complicating study integrity. Furthermore, patients with a preferred treatment may decline study enrollment and/or drop out immediately after randomization prior to receiving any intervention. The history of exploitative research during incarceration has appropriately led to stringent protections and reluctance on the part of researchers to pursue research in jails and prisons [4]. Researchers and regulators have concerns about randomization in two-arm studies (with an intervention and control group only), which could limit access to potentially beneficial interventions or create undue influence to participate in a potentially harmful intervention. Federal regulations on research in prisons and jails include several safeguards, including limiting use of control groups and conducting research that is no more than minimal risk [10]. Yet concerns about the risks of research have led some carceral facilities to ban research or trials completely. This paternalistic approach, while well-intentioned, denies incarcerated people the agency to make informed decisions about research participation. As a result of these barriers, most research in carceral settings relies on observational or quasi-experimental designs that, while valuable, provide less robust causal inference [11]. This methodological compromise limits our ability to establish treatment effectiveness and ultimately slows the translation of evidence into practice, contributing to a “prison implementation penalty” [12]. Main text Partially randomized preference trials: balancing rigor and autonomy Partially randomized preference trials (PRPTs) offer a methodological approach that addresses many of the ethical and practical challenges of conducting RCTs in carceral settings while maintaining scientific rigor. PRPTs represent a hybrid design between traditional RCTs and preference trials (i.e., prospective cohorts where participants choose their intervention). Like a standard RCT, they are relevant to situations in which there are two treatment options with no known superiority of either. PRPTs, however, are not typically used when comparing a treatment to a control. In a one-stage PRPT, participants are first asked at the time of enrollment, before randomization takes place, about their treatment preference. Those with a strong preference (often captured on a Likert scale) are allowed to choose their preferred intervention (preference cohort), while those without a strong preference are randomized between treatment options (randomized cohort) [13] (Fig. 1). Analytic framework for partially randomized preference trials The randomized cohort functions like a traditional randomized controlled trial, providing causal evidence about treatment effectiveness. However, PRPT data analysis goes beyond simple treatment comparisons; it allows researchers to examine how patient preferences influence treatment uptake and outcomes. Specifically, the design reveals whether receiving one’s preferred treatment independently improves outcomes, and whether people who prefer different treatments differ systematically in ways that affect their outcomes. In this way, PRPTs can offer richer, more interpretable data than an RCT while attaining greater ethical feasibility by preserving autonomy. Specifically, the PRPT offers several advantages: Real-world relevance. The preference cohort increases external validity in settings where participants have multiple treatment options and choice over their treatment assignment. Real-world relevance. The preference cohort increases external validity in settings where participants have multiple treatment options and choice over their treatment assignment. Enhanced participation. Allowing participants’ choice in their treatment arm has been shown to improve recruitment and retention rates compared to traditional RCTs, as individuals are less likely to drop out after receiving their preferred treatment [14]. Enhanced participation. Allowing participants’ choice in their treatment arm has been shown to improve recruitment and retention rates compared to traditional RCTs, as individuals are less likely to drop out after receiving their preferred treatment [14]. Improved understanding of preference bias. In a PRPT, preference arms allow researchers to compare outcomes between participants who chose a treatment and those who were randomized to it, making it possible to quantify how much preferences themselves influence engagement and outcome. Improved understanding of preference bias. In a PRPT, preference arms allow researchers to compare outcomes between participants who chose a treatment and those who were randomized to it, making it possible to quantify how much preferences themselves influence engagement and outcome. Improved understanding of selection bias. By comparing baseline characteristics of participants who self-select into different treatment arms, investigators can detect systematic differences between groups. These differences can be quantified and adjusted using methods unavailable in RCTs where preference-driven selection is unobservable. Improved understanding of selection bias. By comparing baseline characteristics of participants who self-select into different treatment arms, investigators can detect systematic differences between groups. These differences can be quantified and adjusted using methods unavailable in RCTs where preference-driven selection is unobservable. Reduced ethical concern about coercion. By allowing choice for those with preferences, they reduce concerns about coercion: participants are not forced into a treatment they do not want. Reduced ethical concern about coercion. By allowing choice for those with preferences, they reduce concerns about coercion: participants are not forced into a treatment they do not want. Reduced ethical concerns about just distribution of benefits. PRPTs mitigate concerns about unfair distribution of beneficial interventions, as those who have preference can access their preferred treatment while randomization only occurs among those who are genuinely undecided or ambivalent. Reduced ethical concerns about just distribution of benefits. PRPTs mitigate concerns about unfair distribution of beneficial interventions, as those who have preference can access their preferred treatment while randomization only occurs among those who are genuinely undecided or ambivalent. PRPTs have been successfully implemented across diverse clinical contexts including mental health treatment for PTSD, surgical versus medical management of Crohn's disease, and additional studies in gynecology, psychiatry, orthopedics, oncology, rheumatology, and weight loss programs [14–16]. Recently, PRPTs have been proposed or implemented in carceral settings internationally [17, 18]. As jails and prisons expand access to certain medications such as medications for opioid use disorder (MOUDs); it is more likely that individuals will be able to choose their preferred medications from the available options. Design considerations Implementing PRPTs requires careful planning around several methodological considerations that differ from standard RCT design [19]. Power calculations must account for the expected proportion choosing each treatment option, the magnitude of treatment effect, the potential size of preference and selection effects, outcome variance, and anticipated attrition rates. Prior research suggests that PRPTs may require approximately twice as many total participants compared to a standard RCT to maintain equivalent power for detecting treatment effects while also quantifying preference and selection effects [20]. The anticipated proportion of participants with strong preferences fundamentally shapes the trial design. If 100% of participants have strong preferences, the PRPT becomes a preference trial with no randomization; if 0% have preferences, it becomes a standard comparative effectiveness RCT. Larger randomized cohorts increase internal validity and causal inference strength, while larger preference cohorts enhance external validity by reflecting real-world treatment selection patterns (at least in settings where patients have access to multiple treatment options). Institutional constraints such as formulary restrictions can narrow that choice in carceral settings and therefore attenuate external validity. Design decisions must weigh the relative importance of isolating treatment effects versus understanding preference and selection effects based on the research question and clinical context. It also may be important to collect some pilot data to inform the feasibility of this design in a particular setting, by for instance, assessing preferences in a small sample. Approach to data analysis PRPT analysis typically employs regression models that include treatment assignment, cohort membership (preference versus randomized), strength of preference (often measured on a Likert scale), and relevant demographic variables. This analytic strategy allows researchers to separately estimate three key effects: the true treatment effect (the causal impact of the intervention), preference bias (outcome differences based on receiving a preferred versus non-preferred treatment), and selection bias (systematic differences between individuals who prefer different treatments). For example, researchers might compare outcomes across four groups: randomized participants receiving Treatment A, randomized participants receiving Treatment B, preference participants who chose Treatment A, and preference participants who chose Treatment B. (Fig. 1). By controlling for cohort membership and preference strength, the model can isolate whether outcomes differ due to the treatment itself or due to having choice and receiving one's preferred option. If outcomes are similar across preference and randomized groups receiving the same treatment, it suggests that offering choice does not substantively change treatment effectiveness. However, if preference cohort participants demonstrate better outcomes, interpretation requires caution: this could support the value of patient autonomy in clinical settings—particularly relevant for carceral populations where autonomy is otherwise limited—or it could reflect underlying differences between individuals who express preferences and those who do not (e.g., motivation, health literacy). The PRPT design allows researchers to disentangle these explanations by comparing baseline characteristics between cohorts and examining whether improved outcomes occur across both preference arms regardless of which treatment was chosen. Within the preference cohort, participants who choose one treatment may systematically differ from the other among important baseline characteristics (age, disease severity, motivation, etc.). Propensity score matching offers a statistical approach to address this selection bias. By calculating each participant’s likelihood of choosing a particular treatment based on their baseline covariates, researchers can match or weight participants to create more comparable groups. This allows for fairer comparisons between those who chose Treatment A versus Treatment B, helping to isolate whether observed outcome differences stem from the treatments themselves or from pre-existing differences between people who prefer different options. While the randomized cohort provides unbiased treatment effect estimates, propensity score methods strengthen causal inference from the preference cohort data [21]. Alternative PRPT designs The design described in this commentary is a one-stage PRPT, in which preference is assessed at enrollment. This approach preserves participant autonomy and enhances external validity but does not fully eliminate selection bias: the preference cohort consists of individuals who may differ systematically from those who are undecided. While preference can be a controlled variable in analysis, a more analytically rigorous alternative is the two-stage (doubly randomized) preference trial [22]. In this design, participants are first randomized to either a random arm or a preference arm; only those assigned to the preference arm are then permitted to choose their treatment. Because preferences do not influence the initial randomization, this structure removes the selection bias introduced by self-selection. Despite its methodological advantages, the two-stage design demands larger samples, greater complexity, and, like a traditional RCT, limits autonomy, which may raise ethical and practical challenges in prisons and jails. A broader taxonomy of preference-sensitive trial architectures exists. For example, in a “comprehensive cohort design” (CCD), all eligible participants are enrolled; those who consent to randomization are randomized, while those who decline receive their preferred treatment under the same protocol. Because refusal may reflect procedural anxiety or logistical barriers rather than genuine preference, the CCD does not formally measure preference strength and lacks the analytic framework to estimate preference and selection effects. Additional trial designs include “preference-informed complementary trials” where two complementary trials are offered within a study in which one trial excludes a treatment participants are strongly unwilling to receive (e.g., opioid therapy in pediatric patients) [23]. A participant may opt into whichever sub-trial excludes the treatment they are unwilling to receive and then is randomized within that trial. Patient preferences have also been proposed as a basis for adaptive allocation in platform trials [23]. Each design reflects a distinct trade-off between the rigor of preference effect estimation, internal validity, and implementation feasibility. Limitations and appropriate use cases PRPTs are not appropriate for all research questions and settings. PRPTs are not appropriate for placebo-controlled trials, blinded studies where treatment assignment must be concealed, or situations where nearly all participants would prefer the same option. When participant preferences are heavily imbalanced, the PRPT loses many of its benefits. The methodology is best suited when comparing two active treatments where non-uniform preferences might exist. Relevant examples in carceral settings include oral versus injectable medication formulations, pharmacotherapy versus psychotherapy for mental health or substance use disorders, treatments with different durations or side effect profiles (such as hepatitis C direct-acting antivirals with varying treatment lengths), or a behavioral intervention delivered by a peer versus by a social worker. Additionally, PRPTs’ reliance on larger sample sizes than standard RCTs may be prohibitive in resource-limited settings or for rare outcomes. Researchers must carefully consider whether the benefits of incorporating patient preference justify the increased complexity and sample size requirements. Conclusion Incarcerated individuals experience intersecting forms of marginalization that create both urgent health needs and substantial barriers to research participation. PRPTs offer a practical methodology to advance equity, diversity, and inclusion in carceral health research by balancing scientific rigor with respect for participant autonomy. Moreover, PRPT data can provide empirical insights on how and when autonomy itself can be therapeutic in this setting. The arguments for PRPTs in carceral settings similarly apply to other marginalized populations where autonomy, trust, and power imbalances present similar research barriers. As such, this methodology can apply to many situations where equity research is impeded due to such barriers. While not appropriate for all research questions, PRPTs represent an important addition to the methodological toolkit for inclusive research. By offering a design that accommodates participant autonomy without sacrificing scientific rigor, PRPTs represent a valuable complement to conventional RCTs in settings where preference and limited freedom are central ethical concerns, and may help generate the evidence needed to improve health outcomes for incarcerated individuals and other marginalized communities… Future work should focus on developing best practices for PRPT implementation and expanding their use in other populations with limited freedoms or restricted autonomy, such as people who use drugs, people experiencing homelessness, or others facing similar ethical and practical concerns. Authors’ contributions JB wrote initial draft and oversaw revisions. All authors contributed to revision and approved the final version of the manuscript. Funding Helen Jack DP2DA063100, R25DA037190. Justin Berk K23DA055695, R25DA037190. Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement No datasets were generated or analysed during the current study.