Discussion

Antibiotic prescribing is a complex process influenced by multiple factors such as physician’s behavior and experience, sex, patient expectations, organizational culture, etc. ^{Reference Wojcik, Ring, McCulloch, Willis, Williams and Kydonaki23} Although much has been written about those factors, there is a significant gap in the literature when it comes to documenting the impact of underlying health conditions on prescribing rates and the impact of social factors that inflate prescribing rates. In the current study, I considered antibiotic use from a mediation analysis perspective. There is a temporal relationship between the variables studied and antibiotic prescribing rates. This temporal relationship is a cornerstone of mediation analysis. Based on current knowledge, social determinants largely “create” chronic health conditions such as obesity, diabetes, and COPD. ^{Reference Singer, Bulled, Ostrach and Mendenhall24} In other words, chronic health conditions cluster in the same areas due to the social and environmental conditions, a phenomenon called “syndemics.” ^{Reference Singer, Bulled, Ostrach and Mendenhall24} For example, several researchers have documented that obesity is more prevalent among residents of poor communities. ^{Reference Kim, Landgraf and Colabianchi25} The same clustering has been observed for diabetes. ^{Reference Consolazio, Koster and Sarti26} Although COPD has been commonly associated with smoking, which is also more prevalent among underserved and poor populations, several researchers have documented a high prevalence of COPD among nonsmoking populations in rural and economically disadvantaged areas. ^{Reference Raju, Keet and Paulin27} Thus, the relationship between the poverty and mediators is well known. In fact, the chronic health conditions (eg, diabetes, obesity, COPD) examined in this study as mediators have been selected due to their high prevalence and syndemic association with the independent variable (poverty). The mediators in the current study have direct and linear relationship with outcome variables (eg, antibiotic prescribing rates) through direct or indirect mechanisms. For example, patients diagnosed with COPD may benefit from antibiotics and therefore, should receive antibiotics during episodes of exacerbation. ^{Reference Stevermer, Fisher and Lin11} On the other hand, patients diagnosed with diabetes and obesity may have a higher exposure rate to antibiotics because they are predisposed to bacterial infections, which may indirectly increase the need for antibiotic therapy. ^{Reference Abu-Ashour, Twells, Valcour and Gamble12,Reference Hussain, Curtain, Mirkazemi, Gadd, Peterson and Zaidi13} In current study, those relationships have been explored under steps 2 and 3 of mediation analysis. Each step demonstrated a strong, linear relationship with the outcome variable. In step 2, poverty was regressed against mediators, and in step 3 mediators were regressed against antibiotic prescribing rates. The relationship between prevalence of poverty and antibiotic prescribing rate was also statistically significant and showed a fully mediated process (96%), indicating that the entire effect of poverty on antibiotic prescribing was fully mediated by chronic health conditions (Table 3). This fact would allow the simplification of the conceptual model by omitting the effect of poverty on antibiotic prescribing rate because the direct effect of poverty prevalence on antibiotic prescribing rate was not significant (Table 3). The conceptual model can be presented as X (prevalence of poverty) → M (prevalence of COPD, obesity, and diabetes) → Y (antibiotic prescribing rates per 1,000 population).

The current study linear regression analysis for individual confounders (ie, physician density per 100,000 population and prevalence of aging population) did not show a statistically significant association with antibiotic prescribing rates, even though several studies have documented higher prescribing rates among elderly populations. ^{Reference Kabbani, Palms, Bartoces, Stone and Hicks19} One potential explanation is that the states with higher prescribing rates might have lower prevalence of elderly population due to lower life expectancy caused by poor socioeconomic status. No studies have explored the effect of physician density on antibiotic prescribing rates in communities. Therefore, it is hard to make valid conclusions based on the multiple regression analysis results presented in Table 2.

The current study had several strengths. These results clearly demonstrate strong, linear relationship between poverty and outpatient antibiotic prescribing rate with ∼39% of variability in prescribing explained by the prevalence of poverty in a state. I capitalized on the strengths of ecological studies and used community characteristics to explore relationships among the variables. Another strength of the study is that it has exposed the association of poverty with antibiotic prescribing through mediators. This association would have gone undetected if multiple regression analysis with prevalence of poverty had been used as one of the confounders because the effect of poverty becomes negligible due to the fully mediated process (Table 3). Thus, in the current study, I utilized the mediating effect of a factor, which would have been undetected in a regular regression analysis commonly used in public health research.

The scientific literature related to patient-specific factors leading to higher prescribing rates is limited. Only a handful of researchers have considered prescribing rates in relation to poverty, and all of them have noted that poor and underserved communities have higher rates of prescribing, although the causes underlying higher prescribing rates have not been fully investigated. ^{Reference Mölter, Belmonte and Palin28} Thus patient-specific factors, which are not currently on the agenda of antibiotic stewardship programs, must be investigated through the lenses of underlying health conditions and social determinants.

This study also had several limitations. Perhaps the most important limitation is the ecological design, which is subject to so-called ecological fallacy. Ecological fallacy prevents researchers from exploring individual-level data simply because, in ecological studies, the unit of analysis is a community (in this case a state). ^{Reference Szklo and Nieto29} Thus, ecological studies are subject to numerous confounders that are not always taken into account. ^{Reference Szklo and Nieto29} This study may also have been confounded by unknown confounders. Another limitation was the use of the prevalence of certain chronic medical conditions along with the prevalence of aging population and poverty using BRFSS data. The problem with BRFSS point-prevalence estimates is that quite often they overlap in 95% confidence intervals, which reduces the accuracy of prevalence estimates used in this analysis. ¹⁵ Furthermore, I was unable to separate diabetes type 1 from diabetes type 2 in data analysis due to the way BRFSS questionnaires are structured; the BRFSS questionnaire asks a question about ever being told by healthcare provider that he/she has diabetes. ¹⁵ Although variables in this study have been shown to have high validity and reliability, it is still possible that due to self-reported information collected by BRFSS, the rates used in this study under- or overestimated the true prevalence of medical conditions or household income. ¹⁵

For the outcome variable in this study, the IQVIA Xponent data presented on the CDC website do not include outpatient antibiotics dispensed from the federal facilities nor inpatient medications dispensed. Thus, these data may underestimate the true amount of antibiotics dispensed from community pharmacies. ¹⁴ Also, the rates presented in this study may have been influenced by the COVID-19 pandemic. And finally, IQVIA Xponent data used in the current study do not differentiate appropriately prescribed antibiotics from inappropriately prescribed ones. ¹⁴

Current discourse regarding antibiotic stewardship does not consider patient-specific factors and may in fact penalize physicians or organizations with higher prescribing rates. For example, peer-to-peer comparison has been shown to reduce antibiotic prescribing rates among top prescribers. ^{Reference Clegg, Bean, Ezzo, Hoth, Sheedy and Anderson30} Although reductions in prescribing rates are desirable, the drawback of this approach in stewardship programs may be the fact that those comparisons do not consider underlying health conditions of patients being treated by individual physicians. Also, physicians with higher prescribing rates may have patients with multiple underlying health conditions that in fact, require antibiotic therapy. Assuming that top prescribers are the ones who prescribe inappropriately and asking them to reduce their prescribing may not be quite desirable approach without considering patient-specific factors. The process of standardization, which would consider patient-specific characteristics, has yet to occur in the National Healthcare Safety Network (NHSN). ^{Reference van Santen, Edwards and Webb7} The standardized antibiotic administration ratio, usually used for facility comparisons proposed by the CDC, does not account for patient mix. ^{Reference van Santen, Edwards and Webb7} On the other hand, looking at antibiotic prescribing process without considering a broader context of social determinants is also detrimental to stewardship programs because without improving social determinants of populations under care, prescribing rates are unlikely to be reduced.