Miguel Hernan

Background: Lithium has long been considered to reduce suicidal behaviour in patients with affective disorders, but evidence from large real-world populations remains limited. Objective: To estimate the effects of lithium initiation and continuation on suicide deaths and non-lethal suicidal behaviours among adults with bipolar disorder or major depressive disorder. Methods: We emulated two target trials using electronic health records and administrative claims from US veterans between January 2010 and December 2022. The first trial benchmarked results against the CSP-590 randomised trial by estimating the 1-year risk of suicide-related events (non-fatal suicide attempts, hospitalisations to prevent suicide or suicide deaths) among patients with a recent suicide attempt initiating lithium versus not initiating lithium. The second trial extended follow-up to 10 years to estimate risks of suicide deaths and non-lethal suicidal behaviours separately, including subgroup analyses by age and diagnosis. An additional analysis removed the requirement of a prior suicide attempt. Findings: In the benchmarking analysis, the 1-year risk ratio for suicide-related events comparing lithium initiation with no initiation was 1.06 (95% CI 1.01 to 1.12), consistent with findings from the CSP-590 trial. In the extended analysis, the 10-year per-protocol risk ratio for suicide death was 1.00 (95% CI 0.86 to 1.15) and for non-lethal suicidal behaviours was 0.96 (95% CI 0.90 to 1.02). Results were similar among individuals with and without a prior suicide attempt. Conclusion: When added to ongoing pharmacological treatment for patients with affective disorders, lithium may not substantially reduce suicide risk highlighting the need for additional suicide prevention strategies in this population. Limitations include lack of information on adherence, dosage or blood levels of lithium which may have obscured existing differences. Clinical implications: Our findings suggest that adding lithium therapy to the ongoing treatment regimens of patients with affective disorders for the sole purpose of reducing suicide may not be well-tolerated or meaningfully reduce suicide risk.

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have reshaped the clinical approach to managing obesity and type 2 diabetes. As approved indications have expanded, use of GLP-1RAs has increased rapidly in the United States. Although randomized trials demonstrate strong efficacy, many questions remain about their optimal use in clinical practice. Real-world data (RWD) from electronic health records, registries, insurance claims, and other sources offer a promising avenue to address these questions. However, concerns about data quality, selection bias, and incomplete ascertainment of medication use and outcomes pose significant challenges to the validity of the resulting evidence. In May 2025, the National Institute of Diabetes and Digestive and Kidney Diseases convened experts from regulatory agencies, payer organizations, and academia to explore these challenges. This second of 2 synopsis articles on the workshop summarizes the discussion around the strengths and limitations of various RWD sources and methodological approaches to strengthen causal inference and generalizability. Presenters highlighted pragmatic clinical trials and target trial emulation as strategies to generate stronger real-world evidence (RWE) that is relevant to both clinical practice and policy. The workshop underscored that careful attention to study design, data limitations, and analytic approach is essential to yield RWE that informs clinicians, patients, payers, and policymakers.

This article analyzes the results of the 2024 ADIA Lab Causal Discovery Challenge, a large-scale competition designed to evaluate whether machine learning methods can infer causal roles from observational data when ground-truth causal structures are known. Participants classified variables into eight causal categories—including confounders, mediators, colliders, and causal antecedents or consequences of treatment and outcome variables—using 47,000 synthetic datasets generated from known causal directed acyclic graphs. The Challenge attracted 1,904 registered participants and 3,343 submitted solutions worldwide. Our results show that supervised learning approaches substantially outperformed traditional constraint-based causal discovery methods in this controlled setting. Top-performing solutions achieved multiclass balanced accuracies as high as 77%, compared with approximately 40% for baseline methods. The most successful approaches combined rich feature engineering, edge-centric representations, ensemble learning, and graph-based architectures inspired by graph neural networks. Performance differences across graph structures revealed that some causal environments are systematically more difficult for discovery algorithms than others. The findings suggest that, when large libraries of labeled dataset–graph pairs are available, causal-role recognition can be effectively approached as a supervised learning problem, with potential applications in factor research, macro-financial forecasting, stress testing, and portfolio construction.

Randomized trials provide evidence for regulatory and clinical decision making in oncology. However, trials cannot answer every important clinical question. Evidence gaps often persist in under-represented or small patient populations and for questions about head-to-head comparisons of active treatments, complex treatment strategies (including treatment sequencing and other dynamic treatment strategies), and long-term or rare outcomes. To address these questions, clinicians and researchers increasingly turn to observational (real-world) data. The target trial framework provides a systematic approach for designing observational analyses that attempt to emulate a hypothetical pragmatic trial. This process requires that investigators specify the key components of the causal question as elements of the target trial protocol: eligibility criteria, treatment strategies, assignment procedures, follow-up, outcomes, and causal contrasts. Explicitly emulating the target trial helps investigators avoid common study design flaws that can lead to immortal time and selection bias. Although the target trial framework helps to avoid such design-related biases, it cannot eliminate biases due to inherent data limitations, such as confounding and measurement error. Here we review target trial emulation to supplement evidence from randomized trials and inform regulatory and clinical decision making in oncology. Understanding the strengths and limitations of the target trial framework improves the integration of real-world evidence into modern cancer care.

Descriptions of the target trial framework often appear to assume that investigators knew the protocol of the target trial from the start of the investigation. In practical applications, however, the protocol of the target trial is constrained by the observational data that investigators have access to. When using preexisting observational databases, the target trial protocol typically needs to be iteratively developed as investigators learn about the data. Here we examine the process by which investigators adapt their original causal question, operationalized into the protocol of a hypothetical trial (the index trial), to the available data. This adaptive process results in a final causal question, operationalized into the protocol of another hypothetical trial (the target trial), that depends both on the original causal question and on the available data. As a result, prespecification of an emulatable target trial protocol is not generally possible because adaptations are expected after inspecting the data. The adaptive nature of the specification of the target trial protocol raises important questions about the types of data examinations that are permissible to guide the adaptations, the procedures for transparent reporting of each adaptation and its rationale, and the possibility of prespecifying the rules that will govern the investigators’ decisions to adapt the protocol to the data.

The Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack—Lipid-Lowering Trial (ALLHAT-LLT) was a pragmatic randomized trial conducted in 513 clinical centers across North America from 1994 to 2002. ALLHAT-LLT, which used standard of care (usual care) as the control treatment, did not find an effect of assignment to pravastatin on all-cause death. We aimed to use the trial data to estimate the effect of adhering to the assigned treatment, that is, the per-protocol effect. Moderately hypercholesterolemic (low-density lipoprotein cholesterol 100-189 mg/dL), hypertensive participants ≥55 years old were included and randomized to pravastatin (40 mg/day unless contraindicated) or usual care (lipid-lowering treatment discouraged unless clinically indicated). Of 10 355 individuals randomized, we included 9741 with complete baseline data. One-third of individuals in the pravastatin arm were non-adherent to the assigned treatment strategy at year 5. After adjusting for factors related to adherence, the 5 year risk difference comparing pravastatin to usual care was −3.0% (−4.9%, −1.3%), and the risk ratio was 0.79 (0.68, 0.91) for death. Findings were consistent across different definitions of treatment strategies. Our results suggest that the use of pravastatin under full adherence was protective against death, even if assignment to pravastatin was not clearly protective.

Implementation of sequential trial emulation for the analysis of observational databases. The 'SEQTaRget' software accommodates time-varying treatments and confounders, as well as binary and failure time outcomes. 'SEQTaRget' allows to compare both static and dynamic strategies, can be used to estimate observational analogs of intention-to-treat and per-protocol effects, and can adjust for potential selection bias induced by losses-to-follow-up. (Paper to come).

Importance: When randomized trials are unavailable or not feasible, observational studies can be used to answer causal questions about the comparative effects of interventions by attempting to emulate a hypothetical pragmatic randomized trial (target trial). Published guidance to aid reporting of these studies is not available.Objective: To develop consensus-based guidance for reporting observational studies performed to estimate causal effects by explicitly emulating a target trial.Design, Setting, and Participants: The Transparent Reporting of Observational Studies Emulating a Target Trial (TARGET) guideline was developed using the Enhancing the Quality and Transparency of Health Research (EQUATOR) framework. The development included (1) a systematic review of reporting practices in published studies that had explicitly aimed to emulate a target trial; (2) a 2-round online survey (August 2023 to March 2024; 18 expert participants from 6 countries) to assess the importance of candidate items selected from previous research and to identify additional items; (3) a 3-day expert consensus meeting (June 2024; 18 panelists) to refine the scope of the guideline and draft the checklist; and (4) pilot of the draft checklist with stakeholders (n = 108; September 2024 to February 2025). The checklist was further refined based on feedback on successive drafts.Findings: The 21-item TARGET checklist is organized into 6 sections (abstract, introduction, methods, results, discussion, other information). TARGET provides guidance for reporting observational studies of interventions explicitly emulating a parallel group, individually randomized target trial, with adjustment for baseline confounders. Key recommendations are to (1) identify the study as an observational emulation of a target trial; (2) summarize the causal question and reason for emulating a target trial, (3) clearly specify the target trial protocol (ie, the causal estimand, identifying assumptions, data analysis plan) and how it was mapped to the observational data, and (4) report the estimate obtained for each causal estimand, its precision, and findings from additional analyses to assess the sensitivity of the estimates to assumptions, and design and analysis choices.Conclusions and Relevance: Application of the TARGET guideline recommendations aims to improve reporting transparency and peer review and help researchers, clinicians, and other readers interpret and apply the results.