Table 2 List of recommendations for future studies estimating the cost of antibiotic resistance (ABR) and related interventions

Recommendations for primary data collection

 • Capture all economic costs related to ABR in hospital patients, not just the directly observed outcomes such as increased length of stay

 • Explore use of g-methods to correct for both time-dependent biases and time-dependent confounders in studies evaluating time-varying exposures in hospital-based studies

 • Exhaustively investigate potential confounders that need to be collected and investigated in ABR cost studies, ideally using formal causal inference methods such as causal diagrams

 • Collect data on lost earnings and out-of-pocket expenses of patients and caregivers so that the wider household and societal costs of prolonged hospital stay and premature mortality can be captured; this is especially important in settings with high out-of-pocket medical expenses

 • Consider reporting measures of the psychosocial burden of suffering to patients and caregivers associated with illness, either by monetising the value of avoided suffering or by reporting this separately in units such as quality- or disability-adjusted life years

 • Consider both quality (internal validity) and broader representativeness (external validity) of data collected before extrapolating from study sites to wider regions such as the national or international level; if possible, data from multiple sites should be synthesised using meta-analysis or meta-regression (including geospatial variables, if appropriate)

 • Implications of ABR outside the hospital setting should be considered unless they are known to be negligible

Recommendations for further methodological development

 • Investigate how levels of ABR may lead to increased costs for everyone, including patients with susceptible infections, those receiving antibiotics prophylactically and patients who are unable to access hospital beds because they are occupied by patients whose hospital stay has been extended by having a resistant infection

 • Explore the use of longitudinal data from prospective cohorts or large linked patient databases to understand the relationships between antibiotic use, ABR and costs of ABR

 • Ecological methods, such as regression, may allow extrapolation of site- or region-specific costs to a national or global level, adjusting for levels of ABR as well as other variables; however, further research is needed to investigate the implications of model simplifications, such as assuming linear and static relationships between antibiotic use and ABR, and the use of alternative modelling methods

 • Insights from transmission dynamic models of bacterial ecology and from economic models of antibiotic market dynamics need to be combined in order to inform optimal policies

 • Explore ways that long-term projections and macro-economic modelling can be incorporated into economic evaluations of ABR-related interventions