We are all modellers. We all have our own internal models of how the world works, from the interaction between the time of day and how long the coffee queue will be, to whether a certain antibiotic will work to treat an infection. These internal models rely on data as well as past experience – did the antibiotic work to treat any infections last week? But also, was anyone actually given this antibiotic last week? And if so, how likely is this antibiotic to work for a patient’s infection today?

For antibiotic resistance, the data to support our models are hampered in many ways – from biased collection to lack of collection. A new report from the Mapping AMR and AMU Partnership (MAAP) reveals the vast gaps in our collection, and hence knowledge, of the prevalence of AMR in Africa.

Based on data from 2016-2019 across 205 laboratories in 14 African countries, the report found that only five out of the 15 pathogen-drug resistance combinations prioritised by the World Health Organization (WHO) were consistently tested. Moreover, antibiotic consumption data from these 14 countries showed that amoxicillin, doxycycline, sulfamethoxazole/trimethoprim and ciprofloxacin compromised more than two-thirds of all antibiotics used in healthcare settings. A worrying selection force as well as a worrying gap in access to needed antibiotics. 

The report also highlighted the disconnect between clinics and laboratories. The data a clinician needs to inform their internal models is whether an antibiotic will treat an infection. If empiric prescribing of antibiotic is working, then they have the data they need. If not, then they need to send a sample from the infection to the lab. The latter becomes the routinely collected surveillance data currently available and biased towards overestimating resistance that is used in the models of public health researchers and policymakers. However this is only from patients in whom empiric antibiotics are failing rather than those in which they are working. What ecological models of AMR need are the additional data: the empiric antibiotic prescribing routines, who is being given them, and the measurement of the antibiotics that are working.

Understanding how and why data is collected is vital to inform the uncertainty in any model of how the world is working and what it really means. We need to reduce this uncertainty by improving data collection, which will also help to support and fulfil the recommendations from the MAAP report - which include bringing microbiology laboratories to places that need it across Africa to improve AMR and antibiotic consumption data; supporting analysis for public health action; adapting national AMR control strategies and national treatment guidelines, as well as defining continental AMR research priorities. We also need to do better at presenting this uncertainty – how should it inform the models our clinicians are using, as well as the models of local and global decision makers. This will drive improved sampling that all of our AMR models need to get better at both treating patients and preventing future AMR selection.