Extrapolating data from clinical trials as we treat patients in real life

Randomised clinical trials are the gold standard for assessing the efficacy of an intervention. However, their results are not always generalisable to real-world populations. Prof. Sharon Giordano (MD Anderson, Houston, USA) sketched how real-world data sometimes can fill this gap.

Although being the gold standard, randomised clinical trials have a couple of shortcomings that challenge the application of the outcomes of the trials when treating real-world patients, Giordano said. First of all, clinical trials don’t include (all) representative populations. In general, patients who participate in clinical trials are younger than the average real-world patient, have less comorbidities, have a better performance status, have a higher social-economic status, and often more late-stage cancer. As a consequence, it is not clear whether or not results of the clinical trial are also applicable to patient populations who where excluded from the trial, in particular patients who are older and/or have more advanced disease than the trial participants, patients who have comorbidities, or female patients who are pregnant (or even males with breast cancer!). Effectiveness also depends on many additional issues like comorbidities, adherence, and side effects. In addition, Giordano said, randomised studies are not always feasible, such as for rare tumors, are sometimes considered unethical, and sometimes fail because doctors and/or patients are unwilling to support randomisation. Finally, randomised clinical trials often do not study patient-centred outcomes, do not follow patients for sufficient duration to assess the late effects of therapy, and have insufficient numbers to study rare outcomes.

In these situations, data from observational studies may be used to help fill in the gaps in knowledge, according to Prof. Giordano. For example, results from one observational study showed a much higher rate of hospitalisation during chemotherapy in patients older than 65 years than was reported in clinical trials [1]. However, the major threat to the validity of observational studies are several types of bias: selection bias, performance bias, detection bias, attrition bias, selective outcome-reporting bias. For example, selection bias can result in patients with a poorer prognosis getting the treatment under investigation, resulting in a worse survival among treated patients. Or the other way around: patients with better underlying health are selected to receive the treatment under investigation, resulting in better survival. Therefore, it is of utmost importance to assess and address the risks of bias in observational studies, in order to estimate whether the observational study is likely to produce valid results, Giordano emphasised. There are statistical tools to address these biases, like multivariate regression models and propensity score analysis, but these techniques will never be able to completely take the biases away. As illustration, Giordano showed the results of an observational study that suggested that treatment for prostate cancer decreases the chance of dying of heart disease or diabetes [2].

Giordano concluded her lecture by briefly addressing two other issues that make it challenging to apply data from a clinical trial to real-world patients. Sometimes in a clinical trial, an intervention shows a statistically significant, but (very) small, benefit. For example, in the APHINITY trial, the absolute benefit after 3 years was only 0.9% and the number needed to treat (with adjuvant pertuzumab) was 112 [3]. Likewise, the ExteNET trial reported an absolute benefit after 2 years of adjuvant therapy with neratinib of 2.3% (number needed to treat of 44) [4]. In these cases, it is difficult to balance the pros and cons of applying treatment to an individual patient, Giordano remarked. Finally, applying outcomes from clinical studies to daily practice can be difficult when the standard of care has been changed by the time the results of the trial are presented. For example, the KATHERINE trial showed clinical benefit from adjuvant treatment with TDM-1 compared to adjuvant treatment with trastuzumab in patients with Her2-positive breast cancer and residual disease after neoadjuvant therapy [5]. However, less than 20% of the patients in KATHERINE were treated with dual anti-Her2 neoadjuvant therapy, which is the standard now. So, this makes it difficult to estimate the benefit of TDM-1 adjuvant therapy for patients who have residual disease after dual anti-Her2 neoadjuvant therapy. In these areas of uncertainty, shared decision-making becomes critical, Giordano said.

1. Barcenas CH, et al. J Clin Oncol 2014; 32: 2010-2017.
2. Giordano SH, et al. Cancer 2008; 112: 2456- 2466.
3. Von Minckwitz G, et al. N Engl J Med 2017; 377:122-131.
4. Holmes MM, et al. Lancet Oncol. 2017; 18: 1688-1700.
5. Von Minckwitz G, et al. N Engl J Med 2019; 380: 617-628.

Posted on 21 May 201927 March 2024