Antibiotic resistance poses a growing threat, highlighting the need to optimize antibiotic prescribing in primary care. To establish optimal prescribing levels, understanding the relationship between prescribing decisions and adverse outcomes for common infections is crucial.
This study aims to work out whether variation in antibiotic prescribing is mainly driven by patient factors such as underlying chronic diseases, or due to doctors making different decisions for the same types of patients. Subsequently we will assess the health consequences of these different antibiotic prescribing decisions. Adverse outcomes, reflective of potential treatment failure or side-effects of the antibiotic, occurring within 3, 14, and 30 days of the initial GP visit will be evaluated.
Using various statistical techniques that adjust for differences between patients for which doctors make different antibiotic prescribing decisions, we will estimate the impact of those decisions on patient outcomes, including the need to get another antibiotic, the need to go to the doctor or hospital, or death shortly after the initial visit to the doctor. We expect that going to a high-prescribing practice offers patients no to little benefit in terms of health outcomes. We will also estimate health disparities in infection incidence, antibiotic treatment and outcomes and decompose the disparities.
Overall, this study seeks to inform current discussion around how to optimize antibiotic prescribing by examining variation and assessing relationships with important patient outcomes, and how to reduce disparities in infection incidence, antibiotic treatment, and outcomes.
Given the increasing threat of antibiotic resistance, there is an increasing recognition that there is a need to optimise antibiotic prescribing in primary care, where the majority of antibiotic prescriptions occur. To establish optimal prescribing levels, a better understanding of variation in antibiotic prescribing and the relationship between different prescribing decisions and the risk of adverse outcomes following common infections is required.
This study will estimate to what extent practice-level variation can be explained by case-mix, by practice-level preference, and by unexplained factors. Subsequently associations between practice-level antibiotic prescribing propensity and the risk of adverse outcomes following common infections will be assessed: bronchitis, cough, otitis media, rhinosinusitis, sore throat, asthma exacerbation, COPD exacerbation, gastroenteritis, impetigo, lower respiratory tract infection, upper respiratory tract infection, and urinary tract infection. We will also decompose patient-level variance in infection incidence, treatment, and outcomes between socio-economic groups.
The following adverse outcomes, potentially reflective of treatment failure, occurring within 3, 14, and 30 days of the first infection-related GP consultation will be evaluated: another GP consultation with(out) another antibiotic, any hospital admission, hospital admission with an infection-related diagnostic code, hospital outpatient visit, emergency department visit, any death, death with an infection-related diagnostic code.
We will use a grouped-treatment variables with individual covariates approach to overcome strong confounding by indication. Practice-level antibiotic prescribing propensity will be estimated for each common infection separately, accounting for age, sex, and comorbidities, where the prescribing propensity reflects the percentage of condition-specific consultations that resulted in an antibiotic prescription in each year.
We hypothesise that, after adjusting for case-mix, that the relationships are non-linear, leading to identification of levels of prescribing above which prescribing to more patients has no or negligible benefits. Where potentially valid instrumental variables can be identified, we will also perform instrumental variable analysis to address potential unmeasured confounding.
The following adverse outcomes, potentially reflective of treatment failure, occurring within 3, 14, and 30 days of the first infection-related GP consultation will be evaluated:
- Another GP consultation
- Another GP consultation with another antibiotic
- Any hospital admission
- Hospital admission with an infection-related diagnostic code
- Hospital outpatient visit
- Emergency department visit
- All-cause death (also evaluated at 90 days)
- Death with an infection-related diagnostic code (also evaluated at 90 days).
The primary analysis will focus on events occurring within 30 days for all outcomes, but to enable comparability with other studies the other time points will also be included.1
Koen Pouwels - Chief Investigator - University of Oxford
Koen Pouwels - Corresponding Applicant - University of Oxford
David Smith - Collaborator - University of Oxford
Mike Sharland - Collaborator - St George's, University of London
Murong Yang - Collaborator - University of Oxford
Vinh Nam Nguyen - Collaborator - University of Oxford
Yingfen Hsia - Collaborator - St George's, University of London
HES Accident and Emergency;HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;CPRD Aurum Ethnicity Record;CPRD Aurum Pregnancy Register