Benchmarking observational studies against randomised trials

<p dir="ltr">Randomised trials are regarded as the gold standard in evaluating the effectiveness and safety of medical interventions, but they are not always feasible due to cost, practical constraints, and ethical considerations. Observational studies can provide evidence either in the absence of randomised trials, or to complement findings from existing ones.</p><p dir="ltr">The target trial emulation framework provides a systematic approach for the design of observational studies, and enables common biases to be conceptualised and avoided. This thesis aimed to explore reasons why observational studies have differing results when benchmarked against randomised trials that ask similar questions.</p><p dir="ltr"><b>Paper I</b> assessed the practical implications of varying the definition of treatment assignment in observational studies, where one of two potential analogues is usually available - prescription or dispensation. A case study was set up in the Swedish Primary Care Cardiovascular Database using the target trial emulation framework to evaluate the effect of two separate antihypertensive drug classes - angiotensin converting enzyme inhibitors (ACEi)/ angiotensin receptor blockers (ARB) and calcium channel blockers - on cancer and ischaemic heart disease. Prescription data was first used to define treatment assignment, followed by dispensation data. Classes of confounder were sequentially adjusted for due to the varying confounding structure and uncertainty surrounding the relevant choice of confounders. Effect estimates did not vary with the definition of treatment assignment nor sequential adjustment for confounder class, but there were differences in the size and characteristics of the study populations.</p><p dir="ltr"><b>Paper II</b> compared the socioeconomic characteristics of individuals who were and were not enrolled in two registry-based randomised controlled trials, TASTE and VALIDATE-SWEDEHEART. Although differences between enrolled and non-enrolled individuals relating to clinical status, comorbidities, and demographics are well characterised, there is an absence of evidence assessing socioeconomic status. This study found that compared to non-enrolled individuals, those who were enrolled were more like to be born in Sweden, be employed, be in a relationship and have a higher income. However, the addition of socioeconomic characteristics to a predictive model for enrolment did not meaningfully improve its predictive capability, likely reflecting the close correlation between socioeconomic and clinical characteristics.</p><p dir="ltr"><b>Paper III</b> was a per-protocol analysis of the REDUCE-AMI trial, which had compared beta-blockers with no beta-blockers in the treatment of myocardial infarction with preserved left ventricular ejection fraction (≥50%). The intention-to-treat analysis of the trial had shown no benefit, which was not in alignment with a target trial emulation asking the same question. The target trial emulation had been published prior to the trial. However, the target trial emulation had only included individuals who were not on beta-blockers at baseline. This per-protocol analysis of Swedish participants in the REDUCE-AMI trial found no evidence of a large effect of beta-blockers in patients with myocardial infarction and preserved heart function. However, a benefit was observed among individuals who were not on beta-blockers at baseline, broadly in line with the target trial emulation.</p><p dir="ltr"><b>Paper IV</b> assessed the effect of ACEi/ARB following myocardial infarction with preserved left ventricular ejection fraction (≥50%). A target trial was emulated in Swedish registries which found no additional benefit of these medications compared to no ACEi/ARB. This observational study can be benchmarked in the future against randomised trials that address this question as no such trials currently exist.</p><p dir="ltr">Conclusion</p><p dir="ltr">Observational studies can provide evidence in the absence of randomised trials or complement the findings from existing ones. The process of benchmarking results from randomised trials against those from observational studies enables the systematic exploration of how the study designs differ. This thesis demonstrates that benchmarking may not be successful due differences in how study populations and treatment strategies are defined and implemented in observational studies.</p><h3 dir="ltr">List of scientific papers</h3><p dir="ltr">I. <b>Humphreys AB</b><b>C</b>, Matthews AA, Young JC, Berglund A, Lindahl B, Wettermark B, Dahabreh IJ, Kahan T, Hernan MA. The definition of treatment assignment in observational emulations of target trials - an empirical examination in the Swedish Primary Care Cardiovascular Database. Ann Epidemiol. 2025 Aug;108:56-62. <a href="https://doi.org/10.1016/j.annepidem.2025.06.003">https://doi.org/10.1016/j.annepidem.2025.06.003</a></p><p dir="ltr">II. <b>Humphreys AB</b><b>C</b>, Berglund A, Olarte Parra C, Louro J, Munier I, Lindahl B, James S, Fröbert O, Jernberg T, Hofmann R, Erlinge D, Härkönen J*, Matthews AA*. Differences in socioeconomic characteristics of individuals who are and are not enrolled in cardiovascular registry-based randomised controlled trials. (*co-senior authors) [Manuscript]</p><p dir="ltr">III. <b>Humphreys AB</b><b>C</b>*, Hofmann R*, Olarte Parra C, Berglund A, Lindahl B, Hernan MA, Matthews AA, Jernberg T. Long-term effect of beta-blockers after myocardial infarction with preserved left ventricular ejection fraction in a per-protocol analysis of the the REDUCE-AMI trial. (*co-first authors) [Manuscript]</p><p dir="ltr">IV. <b>Humphreys AB</b><b>C</b>, Lindahl B, Berglund A, Voelskow V, Fang S, Fröbert O, Hofmann R, Jernberg T, Hernán MA, Matthews AA. Long-term effectiveness of ACE inhibitors or angiotensin receptor blockers in myocardial infarction with preserved left ventricular ejection fraction. Eur Heart J Cardiovasc Pharmacother. 2025 Nov 4;11(7):600-609. <br><a href="https://doi.org/10.1093/ehjcvp/pvaf051">https://doi.org/10.1093/ehjcvp/pvaf051</a><br></p>