The healthcare costs of antimicrobial resistance in Lebanon: a multi-centre prospective cohort study from the payer perspective

Antibiotics are known to generate a positive externality on society by preventing the spread of bacterial infections [1‐4]. Since their discovery in the past century, the misuse and abuse of antibiotics [5‐9] has led to the selection of resistant bacteria and generated extra costs that yielded to a negative societal externality by increasing the burden of bacterial illnesses and eroding the effectiveness of antibiotics for future cases [4‐10]. In recent years, antibiotic resistance became a public health priority and triggered global actions. One Health initiatives [11‐13] tried to address current related market failures and support measures to mitigate its detrimental worldwide impact [14, 15]. With the increased limitations in healthcare resources, expenditures on programs that reduce resistance must efficiently be allocated [2]. Accurate estimates of the economic burden of antimicrobial resistance (AMR) provide a baseline tool to study the cost-effectiveness of interventions to mitigate this global crisis and gives policymakers the incentives needed to invest in research and funding programs that track and prevent the spread of resistance [10‐19]. Literature review in the field shows that health economics failed so far to demonstrate the real cost of resistance [10‐20]. Published studies quantifying the economic burden of AMR reported a wide variability of cost estimates and showed heterogeneity of methodological assumptions [10, 18, 21‐27]. The majority of data findings originate from high-income countries with unique Health Systems that limit the comparability of results with Low- and Middle-income countries (LMICs) and highlight the gap in research in these regions [28]. Lebanon is a middle-income country facing multiple challenges ranging from scarce financial resources, migrants and asylum seekers challenges, loose health system rules and regulations, economic and ecosystem issues, in addition to the fragmented health system and the diversity of health policy. In a country struggling for cost containment, the estimates of the economic and healthcare burden of AMR needed from multiple perspectives.

Research in the field has mainly focused on the epidemiology of selected pathogens and their virulence [29, 30] and the health impact of resistance [29, 31‐50]. Current data on AMR originate from academic research and the society of infectious diseases that both collaborate with different public and private hospitals [29, 30, 46]. The economic impact of AMR in Lebanon may be high compared with other middle-income countries due to the availability of antibiotics over-the-counter [29], the high incidence of AMR [29], the misuse and abuse of antibiotics in human health, animal health and the agricultural sector [51, 52]. Quantifying the burden of AMR in the healthcare settings is the only starting point to estimate the costs of resistance and provide high-quality estimates based on available recommendations from different systematic reviews, published literature, and experts in the field [10, 18, 20, 22, 27]. Here, we aimed to quantify the excess healthcare costs, length of stay, and in-hospital mortality associated with antimicrobial-susceptible versus -resistant infections from the payer perspective using a conventional method for studying the costs of community-associated-infections (CAIs) and two different methodologies of analysis previously adapted in the literature [53, 54] for estimating the costs of healthcare-associated- infections (HAIs) and compare attributable results.

The republic of Lebanon is a democratic, parliamentary state, classified by the World Bank (2020) as an upper-middle income country [55]. It is located in the near east, at the crossroads of the three continents: Europe, Asia, and Africa. The long history of conflicts contributed to the economic downturn and fragile health system and to the development of non-governmental organizations (NGOs) and private institutions that became the major provider of health services as well as the main contractor to the Ministry of Public Health (MOPH) for the provision of healthcare [56]. The total number of hospitals in Lebanon is 165, among which the private sector accounts for the vast majority of 137 institutions mostly located in the capital region. Hospitals in Lebanon cover acute as well as long-term stays and offer all types of medical and surgical services. The vast majority of public institutions have less than 100 beds and autonomous governance while the total number of hospital beds is 2550 [56]. The private sector is the primary provider of healthcare services in Lebanon. The number of beds varies between 80 and 400 beds. This sector is well equipped with highly trained staff and variable financial resources, offering superior multidisciplinary services. The private sector makes up a total of 12,648 hospital beds.

The range of bacterial isolates associated with the community- and healthcare-infections are listed as high alerts by the World Health Organization (WHO) and the Center for Diseases Prevention and Control (CDC). The isolated bacteria highly encountered in cases of HAIs compared with CAIs are Acinetobacter baumannii carbapenem-resistant, Pseudomonas aeruginosa carbapenem-resistant, multidrug resistant (MDR) Pseudomonas aeruginosa, and Methicillin-resistant Staphylococcus aureus (MRSA). Enterobacteriaceae either ESBL-producing or carbapenem-resistant are nearly equally associated with CAIs and HAIs. (Table 1).

The baseline demographic characteristics of enrolled patient’s show that multiple factors are associated with resistant cases either arising at baseline or during hospitalization. Prior exposure to antibiotics was the only common risk factor for developing either community- or -healthcare-associated infections due to resistant bacteria. A higher Charlson index score was a significant risk for developing CAIs with resistant bacteria. During hospitalization, clinical factors significantly associated with infections due to resistant cases included transfer to the ICU, insertion of a urinary catheter, mechanical ventilation, days of mechanical ventilation, and inappropriateness of antibiotic therapy. Not all listed clinical parameters exhibited significant results in the matched cohort group. CAIs and HAIs associated with resistant bacteria were at a higher risk of all-cause mortality and an increase in the hospital charges estimates. The median LOS post-infection did not significantly differ in matched patients with resistant compared with susceptible cohorts in both groups (CAIs and post-HAIs). (Table 2).

The estimates of the excess length of stay and hospital charges are incurred post-infection per patient case either community-or healthcare-associated infections due to antibiotic-resistant compared with -susceptible bacteria. The linear regression shows that CAIs with resistant bacteria extend the length of hospital stay by a statistically significant estimate of (2.2 [95% CI, 1.2–3.3]; p < 0.001) days and result in an additional ($889 [95% CI, 378–1400]; p = 0.001) excess hospital charges compared with susceptible cohorts. HAIs due to resistant bacteria showed a statistically significant higher excess hospital charges ($1297 [95% CI, 627–1966]; p < 0.001) and longer excess LOS (2.1 days [95% CI, 0.9–3.2]; p < 0.001). Regression results in the matched cohort group showed more conservative estimates compared with the Post-infection HAIs group by a statistically significant lower difference of 31% in excess LOS and 14% excess hospital charges. Previous exposure to antibiotics and stay in the ICU were the independent variables associated with excess LOS in the post-HAI and matched cohort analysis. Results show that higher comorbidity index scores were associated with increased LOS in the CAIs and post-HAIs analysis compared with the matched cohort analysis. Independent variables in the model associated with excess hospital charges were advanced age, and the uses of antibiotics before admission in the post-HAI analysis, the high Charlson comorbidity index score, and ICU stay in the matched cohort analysis model. (Table 3).

The results of logistic regression that identifies factors associated with in-hospital mortality. The independent effect of infections with resistant bacteria on in-hospital mortality was significant in cohorts with HAIs (odds ratio, 0.517 [95% CI, 0.327–0.820]; p = 0.05). A higher Charlson comorbidity index score increased the odds of in-hospital mortality in all cohort groups. An increase in time to infection was an additional independent factor associated with in-hospital mortality due to HAIs in both the Post-HAIs and matched groups. Cox regression models explore the daily hazard of reaching the endpoint of discharge alive or in-hospital mortality in individuals with CAIs and HAIs. The hazard ratio of infection with resistant bacteria did not appear to be significantly associated with an increased hazard of in-hospital mortality. (Table 4).

Our study is the first in Lebanon to quantify nationwide the economic burden of hospitalized cases with CAIs or HAIs from the payer perspective. We compared infections due to antimicrobial-resistant with -susceptible bacteria rather than with uninfected as a control group to avoid overestimation of the economic burden of resistance [23, 24, 72‐75]. Although the payer perspective is considered narrow in scope, it allows to demonstrate the differences in outcomes estimates between resistant versus susceptible cases and offer baseline data for policymakers to evaluate the cost-effectiveness of current infection control and prevention practices to prevent AMR, in an era where the health system is struggling with scarce resources [10, 18, 27, 76].

The estimation of costs followed the bottom-up microcosting, considered a reliable method for identifying and valuing relevant cost components in the healthcare setting [17, 77]. The time and date stamp allowed to adjust for multiple confounders arising before the onset of infection and to avoid the allocation of intermediate mediators in the causal pathway [23, 27, 53, 78‐80].We only considered the severity of illness represented by the APACHE score in the descriptive analysis because the timing of the score estimation was unknown. Accounting for the severity of illness score is thought to accurately predict the clinical outcomes associated with resistance if the optimal time to estimate the severity of illness is 24 h before the first culture-positive specimen sample [23, 79‐82]. We adjusted for the time-dependent bias by matching cohorts based on the day of infection [27, 67, 69].

We used the matched cohort with HAIs methodology at the study analysis stage according to age, gender, comorbidity, and timing of infection [23, 27, 53, 79, 83, 84]. The propensity score matching potentially reduces residual confounding compared with paired cohorts that control for a limited number of variables [10, 85].A published study by Nelson and colleagues (2015) [53] compared three methodologies of analysis: the conventional method, matched cohort analysis that accounted for the timing of infection, and the post-HAI method. Results showed more conservative estimates from the matched cohort analysis compared with the methodology based on post-infection data [53]. Our study showed a 26% reduction in excess hospital charges and a 13% decrease in excess LOS in the matched cohort analysis compared with the post-infection analysis.

Similar to a study by Neidell et al. [54], the comparison of LOS and costs due to resistant versus susceptible infections in healthcare settings showed significantly higher estimates associated with HAIs compared with CAIs. Healthcare- and community-associated infections due to susceptible versus resistant bacteria were associated with significantly higher median excess LOS (2.69 days versus 2.2 days) and mean excess hospital charges ($1807 versus $889). We used the survival analysis to estimate the hazard of in-hospital mortality attributable to resistant cases of infection. The results showed no significant increase in hazard ratio according to the three methods of analysis. The limited sample size associated with death cases during hospitalization and the lack of patient follow-up for at least 30 days post-discharge may potentially be contributory factors.

Systematic reviews in the field [10, 18, 23‐27] showed a wide variability of estimates and heterogeneity of the methodology quantifying the burden of AMR. Failure to account for the HAIs as a time-varying exposure and to adjust for multiple confounders, competing risks, and for the time-varying confounders generate erroneous results [10, 18, 23, 27]. The multistate model is a recommended methodology of healthcare costs analysis associated with AMR [18, 27]. This method accounts for time-dependent bias and competing risks but do not adjust for time-varying confounding [18, 27, 86]. The g-formulae may offer more reliable results by accounting for the time-varying confounders in addition to time-dependent bias and competing risks [10, 87, 88]. The g-formula estimates joint causal effects in the presence of a time-varying exposure and confounding. The g-formulae relies on assumptions that should be carefully determined to allow the robustness of the results [89]. Matching is a proposed method to avoid the time-dependent bias if the timing of infection was considered [21, 25]. Multiple studies used this method [55] and matched for patient characteristics like age, and gender or used diagnostic code, the severity of illness, and or the LOS before infection [10]. This reliability of these characteristics as an indicator of AMR infections is arguable not adequately tested [10, 85]. Each methodology of analysis and study design influences the quality of the studies despite that each has pitfalls and limitations [10, 18, 27, 87, 90] Jit et al. 2019 [10] proposed a conceptual framework for quantifying the economic burden of AMR and postulated that obtaining accurate estimates requires rigor and innovation in using the existing methodologies.

A standardized method for study design and analysis is highly needed. This method must take into consideration cross-borders differences and limitations to allow comparability and generalizability of results. Quantifying the burden of resistance may influence the fight against AMR if it provides evidence-based estimates [10, 18, 23, 27]. The global economic downturn has a profound impact on population health and the sustainability of healthcare systems leading to multiple cuts on the budget in both the public and private sectors [91]. Re-allocation of resources and health policy rely on quality data to orientate health expenditures. Data findings may allow us to examine the cost-effectiveness of current practices undertaken to mitigate resistance and orientate for the efficient prioritization of resource allocation to enhance global health [10]. The healthcare costs represent part of the burden of AMR to society. The opportunity cost associated with resistance shows what is missing due to resistance to enhance the welfare of the community if the payer is a public entity or to improve coverage of other diseases if the payer is a private entity. In Lebanon, implemented health policy to tackle AMR needs to strengthen the population-based surveillance on AMR consider the pre-requisite to study the economic burden of resistance [21] and expand to cover not only the human health but also the animal health and agriculture sector interfaces.

The limitations of the study are the factors that may affect the accuracy of the data. Our study included a wide range of infections due to different gram-positive and gram-negative bacteria. Targeting either CAI or HAI type due to an antimicrobial-resistant compared with the same -susceptible species can generate more reliable estimates attributable to the associated outcomes. The same remark applied when targeting one specific mechanism of acquired resistance like carbapenemase production. The methodology of antimicrobial susceptibility testing is another concern due to the lack of standardization throughout the enrolled hospitals or at least no homogeneous methods and breakpoints taken into account to define the resistance. For cost estimation, we followed the bottom-up microcosting from the payer perspective in terms of hospital charges that may vary depending on the health system coverage and third-party contracts. For further accuracy, studies taking into account these limitations are recommended to confirm our findings: the standardization of study design and analysis methodology would allow the comparability and generalizability of results for a more informed national and global action to fight AMR.

Our study is the first nationwide study that quantifies the economic burden of antimicrobial resistance (AMR) in Lebanese hospitals, in comparison with sensitive bacteria. Matched cohort analysis showed more conservative estimates in cases of HAIs. The differences in estimates highlight the need for a unified methodology to estimate the burden of AMR and accurately advise health policymakers and prioritize resource expenditure on infection prevention and control programs.