In this episode, I’ll discuss possible design flaws in a recent study that suggested cefepime has a mortality benefit over piperacillin-tazobactam.
In a recent issue of JAMA Internal Medicine, a group of authors used instrumental variable (IV) analysis to conclude that the use of cefepime was associated with a significant decrease in mortality compared with piperacillin-tazobactam. These results contradict those of a randomized controlled trial, which showed no difference in mortality between these two antibiotics.
In the journal Clinical Infectious Diseases, a new group of authors has published an article describing limitations of the IV analysis study, and has offered an explanation of how the study design may have led to results that differ from the well-powered randomized controlled trial.
An IV analysis is supposed to be a way to take observational data and infer causal effects. Typically, one would expect this technique to be used when a prospective randomized trial is either not ethical or not practical to conduct.
The IV analysis by the JAMA authors used the piperacillin-tazobactam shortage as the instrumental variable to suggest that the use of piperacillin-tazobactam was associated with mortality.
However, the Clinical Infectious Diseases authors point out that the JAMA authors adjusted their results for the use of metronidazole. This is problematic because metronidazole use could be influenced by either the shortage of piperacillin-tazobactam OR disease severity (less severe disease receiving cefepime only and more severe receiving cefepime and metronidazole).
The Clinical Infectious Disease authors identify that this type of bias is called a ‘collider bias‘ and they highlight that it is not statistically appropriate to adjust for collider bias in an IV analysis. One reason not to make this adjustment is that it is akin to adjusting for a post-randomization variable in a randomized controlled trial – something that is accepted to be exploratory only and not reliable enough to draw a conclusion of causality. Another reason not to make this adjustment is that doing so essentially opens a backdoor pathway that links the initial antibiotic choice to an unmeasured factor (disease severity), which can result in spurious associations.
When the Clinical Infectious Disease authors performed an analysis without the metronidazole adjustment, they found no significant difference in mortality between piperacillin-tazobactam and cefepime-based antibiotic regimens.
The Clinical Infectious Disease authors concluded:
The recent IV analysis does not support a mortality benefit for cefepime; results appear dependent on incorrect analytical choices introducing bias. Clinicians should be aware of IV analysis complexities and assumptions when making causal inferences from observational data, especially when results contradict high-quality trials and antibiotic choice is the exposure
You may have heard of an example of collider bias before in something called “The Obesity Paradox“. This is a paradoxical finding where, within a group of patients with a chronic disease, obesity appears protective against mortality, even though in the overall population, obesity is associated with increased mortality. Take, for example, the observation that among patients with diabetes, those who are obese have a lower mortality than those who are non-obese. When the patient population is limited to those with diabetes, bias is introduced because non-obese patients in this population are more likely to have other risk factors for diabetes that can also affect mortality – the most obvious one being smoking. Therefore, it is likely that fewer obese diabetic patients smoke compared to non-obese diabetic patients, and the imbalance of obese vs non-obese smokers can distort mortality rates, making it appear (falsely) that obesity is protective of mortality.
In both examples, diabetes and metronidazole use are “colliders” that, when conditioned on, allow unmeasured confounders (like smoking or disease severity) to create a spurious association between the exposure (obesity or antibiotic choice) and mortality, leading to biased results.
The article in this episode is a selection from my Hospital Pharmacy Academy’s weekly literature digest. Have you ever felt like your physician colleagues are one step ahead of you with new literature developments? Every week, Academy members are provided a summary curated and explained by me of the top hospital pharmacy-related articles published that week from over 20 major journals and sources to save you time and keep you up to date with the literature. To get immediate access, go to pharmacyjoe.com/academy.
If you like this post, check out my book – A Pharmacist’s Guide to Inpatient Medical Emergencies: How to respond to code blue, rapid response calls, and other medical emergencies.
Leave a Reply