CANCER MUSINGS PRESENTATION

Autor: Alberto Carmona-Bayonas, Paula Jiménez-Fonseca, Ramon Salazar Soler

03/09/2024
A New Blog Discussing Oncology Literature
“Science is one of the very few human activities – perhaps the only one – where errors are systematically criticized and fairly often, in time, corrected.”
—Karl Popper

Modern medical literature is a clear example of the clarity of the great philosopher Karl Popper (Vienna, July 28, 1902 - London, September 17, 1994) who stated that science advances thanks to discernible falsifiability; that is, that any proposed statement, to be considered scientific, must be capable of being independently tested by others. Any assertion that cannot be subjected to testing, whether due to its nature, metaphysics, or because the methods employed for its discovery are not reproducible, should not be considered. The core principle of science is that any claim must be testable and potentially falsifiable. This means that scientific knowledge is inherently provisional and open to change as new evidence emerges. This commitment to rigorous methodology, while crucial for ensuring the reliability of scientific findings, has occasionally been perceived as overly critical or challenging to established beliefs. At the time, establishment was embodied by the philosophers of the Vienna Circle who defended a more positivist empiricism, according to which an empirically generated statement was considered scientific truth. Two sides of the same coin that later generated two opposing approaches to science: Frequentism and Bayesianism.
The discussion between these two trends in statistical analysis in the social sciences and in medicine is legendary and will be part of the dialogue we want to promote in this blog. Both approaches are methodologically correct, and each one has its strengths and weaknesses; we would even dare to say that there are situations where each one has its optimal use. 

The frequentist approach is the most commonly used in pharmacological clinical research, which focuses on the development and optimization of medical treatments. A typical scenario involves a medical indication where two treatments exist, and researchers aim to determine if one is more effective or less toxic than the other (A > B or A+B > A) through a randomized clinical trial. To do this, researchers first define estimates of the potential benefit's magnitude based on preliminary clinical and preclinical studies. They then establish criteria for the maximum acceptable risk of error, both for false positives (type 1 error or alpha risk) and false negatives (type 2 error or beta risk). Based on these predefined error margins, a sample size is calculated, with larger sample sizes needed for smaller expected differences and lower acceptable error risks. Finally, a decision rule is established, and based on where the study results "fall," researchers decide whether to reject (or not) the null hypothesis of no difference. With the usual caveats of Popperian Falsifiability, the hypothesis of superiority of the treatment that emerges as the winner is accepted. This approach is most useful in clinical settings where there's a clear, albeit not necessarily predetermined, hypothesis of superiority. This typically applies to situations where a standard treatment already exists, which may be moderately or highly effective but has a suboptimal toxicity profile. In such cases, a new, experimental treatment may offer potential superiority in efficacy or reduced toxicity. This experimental treatment must have undergone rigorous prior evaluation, including a scientifically sound rationale for its mechanism of action, demonstrated efficacy and toxicology validated in preclinical models, and preliminary clinical studies

On other occasions or scenarios, a more heuristic or probabilistic approach is needed, which is less deterministic than the one produced by the frequentist approach. These may be related to less common diseases where there are simply not enough patients or unexpected situations arise, such as the recent COVID pandemic where the global health system collapsed, and we were forced to make decisions in resource-constrained conditions we had never faced before. In these situations, it is more feasible to apply a Bayesian approach based on prior knowledge, plausible certainties, or preconceptions, and the observation of prospective results, randomized or not. It is a mathematical system that provides probabilistic information called Likelihoods, which can be translated as degrees of certainty, and can be useful to support decision-making, although, unlike the frequentist method, it does so non-deterministically, without a decision rule, and with a focus on constantly revising its probabilistic percentages.

Regardless of the frequentist or Bayesian approach, in every experiment or trial, there is a high risk of error, noise, and bias that are difficult for researchers to control and are not always recognized in a first reading of scientific articles. If we focus only on preclinical studies in biomedicine, audits carried out a few years ago showed that between 50% and 89% of the results published in preclinical biomedical research were not reproducible by independent researchers.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4461318/; https://www.nature.com/articles/483531a
These results led to a positive reaction in the editorial and scientific community, resulting in a series of measures adopted by prestigious biomedical journals to ensure greater methodological rigor and control of the effects of noise and bias inherent in all biomedical experimentation. Clinical trials and their statistical methodology, whether Bayesian or frequentist, are not immune to methodological errors or the malicious effects of noise and bias. Multiple testing and its confounding variables, categorization of continuous variables, non-representativeness of the population or selection bias, and others such as evaluation, survivor, or immortal time biases are just a few examples of concepts that should be familiar to authors, reviewers, and readers of medical literature. However, unfortunately, on too many occasions, they go unnoticed and lead to erroneous interpretations, if not directly to erroneous results in articles published in peer-reviewed journals.
https://journals.plos.org/plosmedicine/article/info%3Adoi%2F10.1371%2Fjournal.pmed.0020124

The main objective of this blog is to provide a new, fresh, and independent approach to the interpretation of relevant articles or topics in the medical literature on cancer. We will write editorial comments in which we will provide an original vision and an in-depth analysis of the risk of errors, noise, and bias inherent in the biomedical sciences. When we comment on specific articles, we will reflect on the context and analyze the published study both in its design and in the interpretation of the results.
We will not be restricted by any editorial rules, beyond methodological correctness and a tone that is always respectful of the authors, as the work of medical scientific writing deserves our greatest admiration despite its usual imperfections, which we know and suffer from firsthand given our vocation and trajectory as authors of our own medical articles. In this small editorial group of 3 friends who have been united by a passion for oncology and epistemology, Bayesian and frequentist souls coexist. We have our contradictions and recognize the complexity and difficulties that lead to errors for all researchers. We will not pretend to be dogmatic or even to be right. Our intention is much humbler, as it is to provide a space for dialogue and scientific reflection, in which we certainly think we can contribute our grain of sand and help our readers to become familiar with relevant methodological concepts that are not always explained in sufficient detail and persistence in our medical schools and hospitals. Therefore, we will point out any imperfections we identify and express our discrepancies respectfully and objectively, always with reasoned arguments and accepting and acknowledging our own cognitive biases (another one to add to the list...). We will try to do it in a way that is entertaining and understandable for all clinicians, regardless of their level of knowledge of statistics and methodology. Although the topics or articles we comment on will be related to methodology and cancer in general, there will probably be a biased representation in favor of digestive cancer, due to our professional profiles in this area of oncology, but we are open and will be delighted to receiving publication proposals from physicians and researchers who are experts in other areas of oncology.

0 comentarios

Nombre:
Escribe un comentario:
Publicar