Fighting La Résistance

Can narrowing beta lactam antibiotics reduce gram negative resistance?

Antibiotics are undoubtedly one of the greatest medical innovations. Unfortunately, with their increasing usage has come a problem - antimicrobial resistance (AMR). Through selective pressure, more and more bacteria are developing resistance to the antibiotics we use every day.

Penicillin was discovered in 1928, but by 1940 there had already been discovered a strain of E coli which produced enzymes to destroy it, rendering it resistant. This issue has grown extremely serious over the past decades as bacteria have outpaced the development of new antibiotics, and we have become increasingly reliant on antibiotics of last resort. AMR is recognised as one of the most important global health threats of our time.

As such, mechanisms to reduce AMR gather a lot of interest. Antimicrobial stewardship (AMS) is the practice of optimising the use of antibiotics, with one of it’s main goals being to help reduce AMR. AMS includes decreasing use of antibiotics when not needed, and ensuring the best antibiotics are used for any given indication. One common goal in AMS is trying to narrow the spectrum of antibiotics being used- trying to use the most targeted antibiotic to the known, or likely cause of infection whilst reducing collateral damage. However, it is unknown how effective this practice is at actually reducing rates of resistance. That is where our new study comes in.

What’s it about?

The authors of the study, “Preventing New Gram-negative Resistance Through Beta-lactam De-escalation in Hospitalized Patients With Sepsis: A Retrospective Cohort Study”, look back through medical records to identify patients started on one of a family of antibiotics called beta-lactams, which are the most commonly used and usually most effective types of antibiotics. These include the group of penicillins (e.g. amoxicillin), cephalosporins (e.g. ceftriaxone) and carbapenems (e.g. meropenem). Some beta lactams are very broad - meropenem is arguably the broadest spectrum antibiotic in routine use. Some are very narrow - Flucloxacillin is the drug of choice to treat Staphylococcus aureus but is ineffective for most other bacteria. The authors seek to answer the question as to whether narrowing the spectrum of antibiotic used for a given patient reduces their chances of later developing resistant bacteria (in particular, gram negative bacteria).

How was it done?

The authors scored different beta lactams based on how broad spectrum they were, then categorised patients into three groups: Those whose beta lactam antibiotics were de-escalated, those whose beta lactam stayed the same, and those whose antibiotics were escalated. They then performed a “survival” comparison to see which group accrued the most new gram negative resistant bacteria over the follow up period of 60 days. As many of these patients are high risk of dying due to their infections, they used a specific type of survival analysis called a Fine-Gray proportional hazard - instead of censoring anyone who died (removing them from the analysis), death was treated as a competing hazard. This meant patients were still included in the analysis, but their contribution was re-weighted to reflect the fact they had died before they were able to get new resistant organisms (we will come back to this later). They included several other variables in the model to adjust for them, such as disease severity, demographics etc.

What did they find?

The study is notable as it is the first to suggest antibiotic de-escalation can reduce the burden of gram negative resistance at a patient level. Patients who had their beta-lactam antibiotics de-escalated were less likely to develop a new gram negative resistant organism than those whose antibiotics remained the same, with a Hazard Ratio of 0.59 (95% confidence interval 0.48 - 0.73). This makes this study a powerful addition to the AMS armoury, demonstrating it is not just better for the population at large to improve narrow targeting of antibiotics, but it even reduces the risk for the individual patient you are treating.

Are there any caveats?

Alas, as is often the cause with observational studies trying to determine causal associations, there are a smorgasbord of caveats to this study. The first is that the Fine-Gray model may not have been the most appropriate to use for the analysis. If you do not censor patients who died of their infection and include them in the analysis, then if patients whose antibiotics are narrowed simply die more often, it will look as though they have less gram negative resistance simply because more of them died and never had an opportunity to develop resistance. This kind of bias has been demonstrated in studies such as this one, where beta blockers appeared to protect against death from prostate cancer simply because people on beta blockers were more likely to die of cardiovascular disease (hence why they were on beta blockers in the first place) instead of the cancer.

The authors did a sensitivity analysis only including patients who survived (which would ultimately be more similar to a traditional Cox proportional hazard analysis), and fortunately the result looked the same.

Perhaps the biggest caveat to the study however is something called confounding by indication. The patients were not allocated to one of these three groups by random. A doctor had to make a decision whether to de-escalate the patients antibiotics or not, and hopefully they did this for a reason. If the reason to de-escalate was that they deemed the patient low risk for having or developing gram negative resistance, then of course we would expect those patients to subsequently develop less gram negative resistance - that was the whole reason they were de-escalated in the first place.

It is very difficult to account for this type of bias. There are various techniques to try and get around it such as propensity score matching, where you try to identify the factors which predict which group the patients will end up in and statistically adjust for them, but there is always a high risk of residual confounding since, ultimately, the groups are inherently different.

The only definitive answer to this would be a randomised trial. This would be very difficult to do for a research question like this (although maybe not impossible), and for effect sizes which may be much smaller than those observed here, would require very large sample sizes.

Summary

This study makes a welcome and overdue contribution to the question as to whether de-escalating antibiotics can reduce AMR at a patient level, and would suggest the answer to be , “Yes”. However, as with most studies of this nature, there is a very significant risk of bias, especially through confounding by indication. For now, we will continue to try and optimise the use of antimicrobials through AMS and eagerly await future research examining its impacts on patient and population levels of AMR.