- A recent high profile report estimates that, by 2050, 10 million people will die every year due to antimicrobial resistance (AMR) unless a global response to the problem of AMR is mounted.
- There is undoubtedly a large clinical and public health burden associated with AMR, but it is challenging to quantify the associated excess morbidity and mortality.
- When estimates of the burden of AMR are provided, they should be accompanied by clear acknowledgment of the associated uncertainties regarding the incidence of infections, the prevalence of resistance, and the attributable mortality.
- Predictions always require assumptions, but modeling future scenarios using unreliable contemporary estimates is of questionable utility.
- Current global estimates of the burden of AMR are not very informative; we need detailed, reliable data to be able to improve AMR control measures, preferably based on comprehensive, population-based surveillance data from low-, middle-, and high-income countries.
In 2014, Lord Jim O’Neill and his team published a review commissioned by the United Kingdom government entitled, “Antimicrobial Resistance: Tackling a crisis for the health and wealth of nations” (the AMR Review) . The review estimated that antimicrobial resistance (AMR) could cause 10 million deaths a year by 2050. This estimate has become a familiar refrain; it has been quoted repeatedly by lay media, experts, and public health agencies. Frequently, only this specific, frightening conclusion is reproduced from the report, unaccompanied by caveats or confidence intervals. We acknowledge that there is a large clinical and public health burden associated with AMR, that this burden is likely to increase over time, and that urgent action is required [2,3]. However, we contend that unreliable global estimates like those provided in the AMR Review  potentially undermine, rather than support, the fight against a post-antibiotic era. In this essay, we will scrutinize the estimations of the burden of AMR provided by the AMR Review  and highlight the uncertainties behind these estimates. These uncertainties need to be addressed in order to produce more reliable, detailed, and actionable results.
The first international AMR burden estimates stem from the European Centre for Disease Prevention and Control (ECDC) report “The bacterial challenge: Time to react”  in 2009. Because the model estimates reported in the AMR Review  are partly based on the ECDC  methodology, we will discuss both reports, which estimated the burden of AMR in Europe and the world, respectively. The AMR Review  includes estimates produced by two different consultancy firms: RAND and KPMG. Because the most quoted phrase is derived from the KPMG estimates , the KPMG model will be represented whenever we refer to the AMR Review  throughout the remainder of this essay.
The Number of Bloodstream Infections in the World
Despite being published five years apart, the ECDC report  and AMR Review  are based on very similar assumptions. First, for both reports, data about the number of infections by selected bacterial pathogens (Escherichia coli, Klebsiella pneumoniae, and Staphylococcus aureus) for European countries are derived from the European Antimicrobial Resistance Surveillance network (EARS-Net). However, this is not a population-based surveillance network; EARS-Net only records invasive infections diagnosed in hospitals and for a variable proportion of the total number of hospitals in each country. Tertiary care hospitals are much more likely to participate in this program than smaller hospitals (Fig 1). In order to generate national estimates for number of infections, both reports extrapolate these EARS-Net estimates through use of the reported catchment population of EARS-Net. This means that the incidence rate of infections in predominantly tertiary care hospitals is applied to the whole country. Moreover, overlapping catchment areas between hospitals and uncertainty in catchment population estimates are ignored. For non-EARS-Net countries, included in the global perspective of the AMR Review , an even more crude approach was used; the number of infections was based on the EARS-Net average infection rates per 100,000 people multiplied by population size.