Comparative Study Design

👑 Which Study Design Is the King of the Jungle? Ask different people and you’ll get different answers: 🧪 From the perspective of internal validity? Many will say: RCTs 📊 From the lens of external validity? You’ll hear: Large-scale complex surveys. 📚 Synthesizing evidence? Meta-analyses. Each has its moment in the sun depending on whose perspective you’re asking from. 🙅♂️ But what if we stopped asking humans altogether? 🔍 What if we looked at this from the point of view of the study designs themselves? Then the question becomes: 👉 Which design has shaped, inspired, or influenced the others the most—directly or indirectly? And the answer is clear: 🎯 Surveys are the true King. 📋 Surveys rooted in simple random sampling are the quiet backbone of epidemiology. Their influence permeates every major study design, serving as a benchmark or foundational element. Let’s explore why. 👇 1️⃣ Cluster Randomized Trials and Complex Surveys 🧩 In cluster randomized trials or complex surveys with cluster sampling, the simple random survey is the gold standard. 🧮 The simple random survey serves as a benchmark in cluster randomized trials or complex surveys, informing design effects or intracluster correlation to account for clustering. ✅ A simple random survey has a design effect of 1. 📈 Clustered designs inflate this, revealing the survey’s foundational role in quantifying efficiency and precision. 2️⃣ Case-Control Studies ⚖️ The survey’s influence extends to case-control studies, particularly in control selection. 🎯Population-based controls frequently use survey methodologies, such as sampling frames or random-digit dialing, to reflect the source population, though other approaches may apply. 📞 Even when a sampling frame is unavailable, methods like random-digit dialing or neighborhood controls still draw on survey principles. 🔄 Matching in case-control studies mirrors stratification—a core survey design principle. 📊 Stratification ensures representation, and matching is its mathematical cousin, underscoring the survey’s philosophical and practical imprint. 3️⃣ Randomized Controlled Trials 🧠 Even RCTs, often hailed as the pinnacle of causal inference, owe a debt to surveys. 🎲 Randomization itself is grounded in the principle of random selection, a concept intuitive to simple random sampling. 📊 When RCTs employ stratified randomization to ensure balance across key variables, they borrow directly from survey design’s playbook. 4️⃣ Cohort Studies 🔁 Surveys can serve as the foundation for other designs. 📋 A cross-sectional survey can evolve into a cohort study by adding follow-up waves for individuals meeting eligibility criteria (e.g., smokers). 💡 Think of the Longitudinal component of the Tobacco Use Supplement to the Current Population Survey (TUS-CPS). 🏁 Bottom line: 👑 Surveys are the quiet monarchs of study designs. They don’t always get the spotlight—but they set the stage. #Chisquares #StudyDesign

Today, I like to discuss unanchored indirect comparisons from a strategic point of view. There was a point in time at Novartis, that the #RWE CoE took the decision that all #NIS should be comparative (from a design perspective). I argued that for new assets launched in the market, having a single-arm design strategy is advisable, until we have a understanding on the properties of our products under real conditions. Today, I come across a different reality, that NIS are designed as single-arm and then comparative evidence is generated via the use of known methods, like the matched-adjusted indirect comparisons (#MAIC). Personally, I do not feel comfortable with this solution either, due to my previous experience in the indications of breast cancer (#BCa) and #glaucoma. In both cases, the matching between real data with published clinical data was not easy, since the overlap of the populations was very low (~1/3). That was resulting on 2/3 of the data to receive low weights, that it was like running the comparative study with a small sample size. Other limitations were related to the fact that data have being collected at different periods, outcomes were not consistently defined, unobserved confounders being present, etc. The attached paper on #psoriasis (2023) is a better example with MAIC being applied on clinical studies. The MAIC method was used to estimate comparative efficacy between #guselkumab and the #IL_17 inhibitors #secukinumab and #ixekizumab in the absence of head-to-head clinical trials. Advantages of MAIC included the ability to adjust for differences across trials when individual patient-level data were available for one treatment and only summary data for the comparator. This made it possible to provide comparative effectiveness estimates when randomized controlled trials (#RCTs) were not feasible. Disadvantages involved the risk of bias from unmeasured confounding, the need for very similar trial designs, and the requirement to adjust for all prognostic factors, which is not always guaranteed. MAIC was applied because, at the time, no direct RCT comparisons between the treatments existed, and decision-makers needed comparative efficacy data. The study later validated the MAIC results by comparing them with subsequent head-to-head RCTs (ECLIPSE and IXORA-R), finding consistent results across methods. At the end, the authors recommend using MAIC methods cautiously when RCTs are not available, especially when trials have similar designs and populations. They stress that MAICs can be valid but should be evaluated case-by-case and more empirical validations are necessary. To my experience MAIC should not be used to compare NIS vs RCTs and they might be of limited value in the comparison of evidence generated through NISs, due to the heterogeneity of study populations. Disclaimer: The opinions shared are solely my own and not express the views or opinions of any of my employers.

Choosing the Right Study Design The document outlines various study designs commonly used in clinical and epidemiological research. It begins by distinguishing between experimental and observational studies. In experimental studies, like Randomized Controlled Trials (RCTs), the investigator actively intervenes in patient care and records outcomes. RCTs are considered the gold standard for establishing causality due to randomization, which minimizes bias. However, they can be costly and not always feasible. In contrast, observational studies involve no intervention—researchers simply observe and record what happens. These include cohort studies, which follow individuals over time to assess incidence and are valuable for identifying potential causes but may be resource-intensive and prone to confounding. Case-control studies look backward in time, comparing patients with a disease to those without to identify prior exposures. They are efficient for rare diseases but limited in establishing temporal relationships. Cross-sectional studies assess data at a single point in time, often used to estimate disease prevalence or attitudes but cannot determine incidence or causality. Case series and case-note reviews offer descriptive insights, often used to describe new diseases or treatments but lack control groups and strong inferential power. The document stresses the importance of selecting a study design that fits the research question, especially when examining causality. More rigorous designs typically require more time and resources but provide stronger evidence. Poor sampling can introduce bias regardless of design. Link: https://lnkd.in/ecVTFwhG #statistics #studydesign

Studies using RWD for nonrandomised comparisons require important methodological considerations to minimize potential sources of bias and confounding, which need to be addressed through appropriate study designs and analytical methods. As an example, the DARWIN EU® catalogue includes the use of active comparator designs, which compare treatment alternatives commonly used for the same indication. This design mitigates confounding by indication and is restricted to new users whenever possible to minimize the potential for other biases. Self-controlled designs including self-controlled case series and self-controlled risk interval are also included for drug safety assessments. In such designs, comparisons are made by looking at different treatment periods within the same person, eliminating all time invariant confounding by design. Analytical strategies to assess potential bias due to measured or unmeasured confounding are also considered. Examples include the use of large-scale propensity scores as an adjustment approach to balance all measured covariates between treatments compared, and the use of negative control outcomes to inform the risk of systematic error and to enable the empirical calibration of estimates and p-values. Over its first three years, DARWIN EU® has played a pivotal role in advancing the EU regulators' vision to enable the use of RWE and establish its value for regulatory decision-making in Europe. Achieving this vision will improve the timeliness, accuracy and relevance of regulatory decisions, with the ultimate goal to better support the development and evaluation of medicines for patients. #DARWINEU #regulatory #realworldevidence #rwe #realworlddata #rwd #dataquality #dataanalytics #patientjourney #patientoutcomes #realworldoutcomes #EMA #clinicaleffectiveness #claims #ehr #clinicaltrials #datascience