Yesterday, our study on antibiotic cycling strategies in ICUs was published. Thanks to Joppe van Duijn, involved in all study phases, we could report that in 8 ICUs in 5 countries with 8,776 patients the unit-wide prevalence of antibiotic resistance was similar when cycling antibiotics every 6 weeks or when cycling antibiotics for every next patient treated (mixing). The study was motivated by prior mathematical models, of which most predicted that cycling would do better. So, now all can raise their voices: (1) “all models are wrong, but some are useful”; (2) “most studies are wrong, but some are useful”; or (3) “if model predictions are not confirmed, where did the study go wrong?”

Antibiotic cycling to control antibiotic resistance on a population level is perfectly suited for a theoretical approach. The first step is to simplify the assumed mode of action: patient 1 is treated with antibiotic A, which leads to carriage (through selection or mutation) with a bacterium resistant to antibiotic A. The same occurs in patients 2, 3 and 4 (of 20 patients in the ICU). Now, once we have more patients carrying the resistant bug, the likelihood of cross-transmission to other (yet uncolonized) patients increases, adding a “transmission force” to the “selection force”. Therefore, let us, from some time point on, treat patients with antibiotic B, that kills the bacteria resistant to antibiotic A. Problem gone, yet the same happens with bacteria becoming resistant to antibiotic B. Then we go to antibiotic C, and, thereafter, we can go back to antibiotic A. These periods between changing from A to B to C to A may be months (cycling) or may be after each patient that was treated (mixing).

As the world is more complex than this, we need to include that some patients are admitted with resistant bacteria (immediately changing the “transmission force”), and that some ICUs do better in preventing cross-transmission. And as we don’t know how good one antibiotic selects for resistance, compared to another antibiotic, we must make assumptions. And, damn, some of these bacteria build up resistance to more antibiotics. And at the end of the “road of uncertainty” we have physicians that prescribe antibiotics, and that are supposed to follow a protocol…… Very difficult to model.

Over the years, the mathematical models included more of these uncertainties and progressed from very “simple” (“simple models, for simple people”, quote Marc Lipsitch) to very complex. The most recent one, published in Molecular Biology and Evolution when we ware analyzing our study data, made a very serious attempt to include that complexity and then concluded: “neither cycling or mixing is a priori better than the other at mitigating selection for antibiotic resistance in the clinic”.

Is this the end of the story for antibiotic cycling? Probably yes, as with increasing antibiotic resistance the choices of antibiotics for cycling get exhausted in many places. Probably yes, as I’m afraid that our study illustrates  what is needed to provide a reliable answer from a clinical study (count on 6 years of work!, at least). And definitely not, as the theory predicts that cycling and mixing can be very beneficial. All needed is to find the circumstances that meet the conditions that were modelled. A potential area is the concept of collateral sensitivity, with cycling of antibiotics for the benefit of individual patients with chronic infections.