Desafiando el dogma de los cuidados intensivos: el papel del escepticismo clínico en la UCI

.css-9l3uo3{margin:0;font-family:»Roboto»,»Helvetica»,»Arial»,sans-serif;font-weight:400;font-size:1rem;line-height:1.5;letter-spacing:0.00938em;}“It is in the darkness of their eyes that men get lost.” (Black Elk)We continue to witness difficulties caused by the formation of critical care dogma without solid clinical evidence. Much of this arises from misinterpretation of medical literature. Every medical specialty wrestles with this problem, yet critical care faces a more severe and common version because of the urgency, complexity, and high stakes of ICU decision-making.During the last century, the work of Ronald Fisher in statistics and Karl Popper in philosophy shaped the concept of null hypothesis (H0) significance testing. The null hypothesis is usually summarized as “no difference” or “no effect,” and researchers calculate the P value to judge how compatible the dataset is with the H0. A very small P value provides evidence against the H0. One can never prove the H0; one can only evaluate its probability. In other words, a studied theory can never be ontologically proven. It can only fail to be falsified.Clinical skepticism, particularly in critical care, suggests an effort to find evidence for (ie, “to prove”) the H0. Consider a study claiming improved outcomes with penicillin therapy in the treatment of pneumococcal pneumonia. The H0 would imply that penicillin is not associated with better outcomes in the treatment of pneumococcal pneumonia. “Proving” the H0 would involve identifying factors that show no improvement with penicillin, which would make any outcome benefit statistically unlikely (ie, a high P value). This approach does not falsify the study’s intent, but it increases the likelihood of the H0 and decreases the likelihood that clinicians will adopt the study’s conclusions.Obvious Evidence & Impossible TrialsIn reality, the discovery and use of penicillin for streptococcal infections predate the era of randomized controlled trials. The majority of physicians accept the “disproval” of this H0 because of multiple converging forms of evidence: historically high streptococcal infection fatality rates before World War II, dramatic clinical improvement after penicillin administration, microbiologic eradication of bacteria, and consistent findings across time periods. Therefore, this H0 has become statistically very unlikely but not completely disproved; yet a properly conducted placebo-controlled randomized trial has never occurred and is ethically highly implausible.Frequentist (probabilistic) analysis interprets the long-term frequency of events as probabilities rather than subjective beliefs. Bayesian analysis begins with a set of prior beliefs and accounts for new knowledge based on Bayes’ theorem, suggesting that Bayesian analysis aligns more closely with the prior assumptions and new information used daily by intensivists and other clinicians during patient care.Why ICU Trials Struggle to Reveal TruthICU trials are uniquely complex. Outcomes are multifactorial, and mortality is a notoriously unreliable end point. Populations are heterogeneous; therefore, small effects may or may not reach prespecified thresholds for statistical significance. Depending on sample size, an intervention may produce an absolute benefit of only 1% to 2% with a P value less than 0.05, whereas an absolute benefit of 10% may be labeled nonsignificant because of a high P value. Smaller sample sizes may produce type II errors, which may be far more clinically meaningful than the statistically significant findings of larger trials.ICU trials are frequently underpowered because of difficulty with enrollment and small effect sizes, and type II errors may commonly occur. Many of these trials are deemed negative by the usual frequentist standards. Even negative findings may contain clinically relevant information when one examines confidence intervals and effect sizes. Significant ranges of truth may remain important.In addition, no evidence of effect does not imply evidence of no effect. Premature dismissal of negative trials may prove erroneous. Meta‑analyses may aid performance, but publication bias remains substantial. Trials with statistically plausible findings are more likely to be published, whereas those deemed negative may be discarded or ignored. This bias distorts the performance of meta‑analyses and professional guidelines.Bayesian Analysis at the BedsideAs noted, ICU physicians and other clinicians rarely focus on P values or confidence intervals while attending to a patient who is unstable. They focus on possible degrees of benefit and risk associated with interventions. Bayesian analysis mirrors this bedside reasoning, while frequentist analysis shapes publications and guidelines.Bayesian analysis aligns most closely with bedside clinical decision-making, while traditional frequentist analysis tends to shape publications more objectively.Frequentist analyses in the ICU often struggle because of high noise‑to‑signal ratios, multifactorial decision-making, and heterogeneous patient populations. The most relevant question is not “Is the H0 rejected?” but rather, “What does all available evidence imply about the possible benefits and risks for this particular critically ill patient?”Watch for my next column, which will feature further review of critical care research, examples, and recommendations.Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of Physician’s Weekly, their employees, and affiliates.ReferencesBlack Elk. Black Elk Speaks: Being the Life Story of a Holy Man of the Oglala Sioux. University of Nebraska Press; 1961.
Fisher, R. J R Stat Soc Ser B. 1955;17:69-78. https://doi.org/10.1111/j.2517-6161.1955.tb00180.x
Popper K. The Logic of Scientific Discovery. Hutchinson & Co; 1959.