Association of patient, physician and visit characteristics with inappropriate antibiotic prescribing in Japanese primary care: a cross-sectional study

Abstract

Objective: To assess the prevalence and associated factors of inappropriate antibiotic prescribing in Japanese primary care.

Methods: This cross-sectional study analysed all antibiotic prescriptions written between 1 October 2022 and 30 September 2023, using an electronic health record database of primary care clinics across Japan. Using a previously developed classification algorithm, we determined whether diagnosis codes occurring on or during the 3 days before the antibiotic prescribing date 'always', 'sometimes' or 'never' justified antibiotic use. We classified antibiotic prescriptions into one of four mutually exclusive categories: 'appropriate' (associated with ≥1 'always' code), 'potentially appropriate' (associated with ≥1 'sometimes' code but no 'always' codes), 'inappropriate' (associated only with 'never' codes) and 'not associated with a recent diagnosis'. A linear probability model examined patient, physician and visit characteristics associated with inappropriate antibiotic prescribing among solo practice clinics.

Results: Analyses included 2 058 021 outpatient antibiotic prescriptions to 1 267 708 patients at 2809 clinics. Among these prescriptions, 176 181 (8.6%) were appropriate, 1 238 549 (60.2%) were potentially appropriate, 348 949 (17.0%) were inappropriate and 294 342 (14.3%) were not associated with a recent diagnosis. Among solo practice clinics, inappropriate prescribing was more likely to patients aged <18 versus ≥65 years (+2.6%; 95% CI +0.3% to +4.9%) or with Charlson Comorbidity Index score ≥2 vs 0 (+2.0%; 95% CI +0.6% to +3.4%), for physicians aged ≥65 versus <45 years (+7.3%; 95% CI +3.6% to +11.0%), for physicians in the highest tertile of antibiotic prescribing volume (+4.9%; 95% CI +3.0% to +6.8%), during telehealth visits (+3.9% vs office visits; 95% CI +0.02% to +7.7%) and during regular hours care versus after hours care (+2.1%; 95% CI +0.7% to +3.5%). These findings were qualitatively unchanged when including both solo and group practice clinics.

Conclusions: Targeting younger patients, patients with comorbidities, older physicians, physicians with high antibiotic prescribing, telehealth visits and regular hours care may further increase stewardship effectiveness.

Keywords: Communicable Disease Control; Epidemiological Monitoring; Public Health.

Figure 1. Association of patient, physician and visit characteristics with prevalence of inappropriate antibiotic prescriptions in solo practice clinics. To accurately attribute prescriptions to physician characteristics, this analysis focused on prescriptions written by physicians at solo practice clinics. We analysed 1 091 553 prescriptions from 1723 solo practice clinics, sequentially excluding 966 468 prescriptions from 86 clinics lacking the physician owner’s age and 1000 group practice clinics. Using this sample of antibiotic prescriptions, we examined the associations of patient, physician and visit characteristics with the prevalence of being categorised as ‘inappropriate’ using a multivariable linear regression model that adjusted for month indicators. We clustered standard errors at the physician level. We calculated adjusted prevalence of antibiotic prescriptions that were inappropriate using predictive margins, presented as bars with caps indicating 95% CIs. ^aWe classified patient volume at the physician level according to the tertiles of average number of visits per month, with ≤669 visits per month being ‘low’, 670–1215 visits per month being ‘middle’, and≥1216 visits per month being ‘high’. ^bWe classified antibiotic prescription volume at the physician level according to the tertiles of number of antibiotic prescriptions per 1000 visits, with ≤12.6 prescriptions being ‘low’, 12.7–40.1 prescriptions being ‘middle’, and ≥40.2 prescriptions being ‘high’. ^cWe tested whether the prevalence of inappropriate prescribing was equal in all regions (ie, whether at least one region was not the same as the others) using the F-test, with a P value of 0.74. CCI, Charlson Comorbidity Index.