Real World Evidence

RCT versus RWE: Good versus evil or yin and yang?
Almut Winterstein
University of Florida, USA

Clinicians, researchers and policy makers have been raised in a paradigm that places randomized clinical trials on top of a hierarchy of evidence or that dichotomizes study designs into randomized, which is equated to valid, and not randomized, which is equated to invalid or highly dubious. Major efforts to enhance drug safety research infrastructure have shifted our acceptance of observational designs, especially in instances where the adverse event is not anticipated and unrelated to a drug’s indication, resulting in limited confounding. Other instances where evidence from non-randomized studies is accepted include situations where randomization is not feasible. The most recent evolution of real-world evidence as main source of evidence for approval of new molecular entities or indications further challenges our historic understanding of the hierarchy of evidence and the scientific method.

Through randomization and blinding, comparison groups are largely balanced on both measured and unmeasured factors if the trial has sufficient sample size. Protocol-based outcomes ascertainment ensures unbiased, structured assessments regardless of exposure status or baseline characteristics. Used jointly, RCTs can mitigate both selection and measurement biases and support causal inferences. However, besides the escalating cost of RCTs and other feasibility issues, various problems arise that require supplemental methodological approaches to inform regulatory and clinical decision-making, including poor generalizability resulting in inductive fallacy; limited ability to explore effect modification; and significant delays in evidence generation.

Legislative action to address some of these shortcomings was formalized in the United States in the 21st Century Cures Act from 2016, which is designed to help accelerate medical product development. One central component is the concept of real-world evidence, i.e., evidence about the safety and effectiveness of medications derived from real-world data. Importantly, the Cures Act formalizes the concept that valid and actionable evidence can be derived from non-experimental settings using observational study designs and advanced analytic methods. In this presentation we aim to illustrate that dichotomous approaches that contrast RCTs and RWE are limited in their understanding of the full range of methodological challenges in making causal inferences and then generalizing such inferences for real-world decision-making. Those challenges are discussed across the spectrum of traditional RCTs, pragmatic RCTs that rely on RWD or hybrid designs, and observational studies that rely on RWD. The presentation will end with specific challenges for RWE research in the era of increasing data availability and artificial intelligence.

Diagnostic accuracy of claims data from 70 million people in the German statutory health insurance: Type 2 diabetes in men
Ralph Brinks1,2,3, Thaddaeus Toennies1, Annika Hoyer2
1Deutsches Diabetes-Zentrum, Germany; 2Department of Statistics, Ludwig-Maximilians-University Munich; 3Department of Rheumatology, University Hospital Duesseldorf

During estimation of excess mortality in people with type 2 diabetes in Germany based on aggregated claims data from about 70 million people in the statutory health insurance, we experienced and reported problems in the age groups below 60 years of age [1]. We hypothesized that diagnostic accuracy (sensitivity and specificity) might be the reason for those problems [1].

In the first part of this work, we ran a simulation study to assess the impact of the diagnostic accuracy on the estimation of excess mortality. It turns out that the specificity in the younger age groups has the greatest effect on the estimate in terms of bias of the excess mortality while the sensitivity has a much lower impact.

In the second part, we apply these findings to estimate the diagnostic accuracy of type 2 diabetes in men aged 20-90 based on the approach and data from [1]. We obtain that irrespective of the sensitivity, the false positive ratio (FPR) increases linearly from 0.5 to 2 per mil from age 20 to 50. At ages 50 to 70, the FPR is likely to drop to 0.5 per mil, followed by a steep linear increase to 5 per mil at age 90.

Our examination demonstrates the crucial impact of diagnostic accuracy on estimates based on secondary data. While for other epidemiological measures sensitivity might be more important, estimation of excess mortality crucially depends on the specificity of the data. We use this fact to estimate the age-specific FPR of diagnoses of type 2 diabetes in aggregated claims data.


[1] Brinks R, Tönnies T, Hoyer A (2020) DOI 10.1186/s13104-020-05046-w

Coronary artery calcification in the middle-aged and elderly population of Denmark
Oke Gerke1,2, Jes Sanddal Lindholt1,3, Barzan Haj Abdo1, Axel Cosmus Pyndt Diederichsen1,4
1Dept. of Clinical Research, University of Southern Denmark, DK; 2Dept. of Nuclear Medicine, Odense University Hospital, DK; 3Dept. of Cardiothoracic and Vascular Surgery, Odense University Hospital, DK; 4Dept. of Cardiology, Odense University Hospital, DK

Aims: Coronary artery calcification (CAC) measured on cardiac CT is an important risk marker for cardiovascular disease (CVD), and has been included in the prevention guidelines. The aim of this study was to describe CAC score reference values and to develop a free available CAC calculator in the middle-aged and elderly population. This work updates two previously published landmark studies on CAC score reference values, the American MESA study and the German HNR study [1,2]. Differences in curve-derivation compared to a recently published pooled analysis are discussed [3].

Methods: 17,252 participants from two population-based cardiac CT screening cohorts (DanRisk and DANCAVAS) were included [4,5]. The CAC score was measured as a part of s screening session. Positive CAC scores were log-transformed and nonparametrically regressed on age for each gender, and percentile curves were transposed according to proportions of zero CAC scores.

Results: Men had higher CAC scores than women, and the prevalence and extend of CAC increased steadily with age. An online CAC calculator was developed, After entering sex, age and CAC score, the CAC score percentile and the coronary age are depicted including a figure with the specific CAC score and 25%, 50%, 75% and 90% percentiles. The specific CAC score can be compared to the entire background population or only those without prior CVD.

Conclusion: This study provides modern population-based reference values of CAC scores in men and woman, and a freely accessible online CAC calculator. Physicians and patients are very familiar with blood pressure and lipids, but unfamiliar with CAC scores. Using the calculator makes it easy to see if a CAC value is low, moderate or high, when a physician in the future communicates and discusses a CAC score with a patient.


[1] Schmermund A et al. Population-based assessment of subclinical coronary atherosclerosis using electron-beam computed tomography. Atherosclerosis 2006;185(1):177-182.

[2] McClelland RL et al. Distribution of coronary artery calcium by race, gender, and age: results from the Multi-Ethnic Study of Atherosclerosis (MESA). Circulation 2006;113(1):30-37.

[3] de Ronde MWJ et al. A pooled-analysis of age and sex based coronary artery calcium scores percentiles. J Cardiovasc Comput Tomogr. 2020;14(5):414-420.

[4] Diederichsen AC et al. Discrepancy between coronary artery calcium score and HeartScore in middle-aged Danes: the DanRisk study. Eur J Prev Cardiol 2012;19(3):558-564.

[5] Diederichsen AC et al. The Danish Cardiovascular Screening Trial (DANCAVAS): study protocol for a randomized controlled trial. Trials 2015;16:554.