Substandard and falsified antibiotics are associated with antimicrobial resistance: a retrospective country-level analysis

Abstract

Objective

Substandard and falsified (SF) medicines pose a significant public health concern due to their elusive nature and potentially dire consequences, including the exacerbation of antimicrobial resistance (AMR). This study aims to quantitatively assess the relationship between the prevalence of SF antibiotics and AMR.

Methods

We combine the Infectious Diseases Data Observatory database (1962–2019), with Global Bacterial Antimicrobial Resistance Burden Estimates (2019) to investigate whether the prevalence of SF antibiotics is positively associated with AMR. Using a quasi-binomial regression, we analyse 257 unique observations across 63 countries and 7 antibiotic classes between 1992 and 2019. We adjust the model for socioeconomic, environmental, health-related, governance and livestock production factors.

Findings

We find that the prevalence of SF antibiotics is positively associated with AMR, after controlling for Gross Domestic Product per capita, population density, particulate matter 2.5, cardiovascular death rate, human antibiotic consumption, regulatory quality, livestock production. The prevalence of SF antibiotics is also positively associated with deaths per 100 000 people attributable to or associated with AMR. This is robust to adding more covariates and country fixed effects as well as dropping countries with a limited number of observations.

Conclusion

AMR ranks among the top global health threats, presenting a multifaceted challenge affecting humans, animals and environment. This study sheds light on the possible relationship between SF antibiotics and AMR. While the prevalence of SF antibiotics appears to be associated with AMR, further research and more representative data are needed to determine the extent to which this association could be explained by a direct causal relationship.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Substandard and falsified (SF) medicines, especially antibiotics, pose a significant public health concern. Existing research suggests that SF antibiotics may be associated with antimicrobial resistance (AMR), but the relationship between SF antibiotics and AMR prevalence remains underexplored.

WHAT THIS STUDY ADDS

• By utilising a comprehensive global dataset spanning multiple decades and controlling for socioeconomic, environmental, health-related, governance and livestock production factors, this study provides quantitative evidence of a positive association between the prevalence of SF antibiotics and AMR.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study sheds light on the possible relationship between SF antibiotics and AMR. Stakeholders may consider addressing SF medicines as part of broader efforts to mitigate AMR.

Introduction

Substandard and falsified (SF) medicines still do not receive enough attention in public health. Substandard medicines are authorised medical products that fail to meet either their quality standards or specifications, or both, due to within-factory errors or degradation in supply chains. Falsified medicines, instead, are defined as deliberately and fraudulently misrepresenting their identity, composition or source.¹ Due to the underground nature of this business, SF medicines are hard to detect. Yet, they are still a significant policy concern worldwide. In 2010, the UN Office on Drugs and Crime, for the first time, highlighted the trafficking of SF medical products as a substantial threat, alongside cocaine, maritime piracy and human trafficking. The illegal trafficking of SF medicines remains one of the most profitable sectors within the global trade of counterfeit goods. This is due to its low risk, large global market size, high profit margins and constant demand.^{2 3} In low and middle-income countries (LMICs), SF medicines are even more of a concern, as outright corruption and weak institutions may limit the government’s capacity to address inefficiencies along the drug supply chain, and individuals are more financially constrained to start with, limiting their choice of products when they go to purchase at a pharmacy.

Given the convenience sampling strategy commonly employed in surveys, obtaining representative figures on the prevalence of SF medicines is challenging.⁴ In 2017, the WHO estimated that in LMICs, 10.5% of medicines are of low quality.¹ In a similar meta-analysis Ozawa et al. (2018)⁵ estimated it at 13.6% (12.4% for antibiotics). Several other studies examined smaller samples, indicating that the prevalence of SF medicines varies significantly by country and type of medicine. Sub-Saharan Africa is noted for having one of the highest prevalences,⁶ and antimalarials and antibiotics have generally received more attention.^{7 8} Despite its clandestine nature, the pharmaceutical trade of SF medicines exhibits significant disparities, with a higher prevalence in LMICs compared with high-income ones. This phenomenon occurs despite the pharmaceutical market of the Middle East and African continents representing just about 3% of the global market share, compared with about 45% for North America.⁹

The consequences of SF medicines could be enormous in terms of health and socioeconomic impact.¹ SF medicines pose risks to patients, potentially resulting in adverse effects due to incorrect active ingredients (eg, toxicity), failure to cure diseases and causing prolonged illness and preventable deaths. From a health systems perspective, SF medicines may generate additional care requirements, resulting in increased out-of-pocket expenditures, burdening healthcare providers and eroding confidence in the healthcare system. Prolonged illness and a heightened disease burden caused by SF medicines can also bring income loss for patients and decreased productivity for businesses and the broader economy, thereby contributing to increased poverty.

More recently, researchers have argued that SF medicines could exacerbate the progression of antimicrobial resistance (AMR).10,12 AMR occurs when bacteria, viruses, fungi and parasites evolve over time and no longer respond to medicines, rendering infections more difficult to treat and increasing the risk of disease spread, severe illness and death.¹³ However, there is currently no quantitative evidence demonstrating a direct link between the prevalence of SF medicines and AMR. Despite the limitations of existing data, in this article, we endeavour to investigate whether the prevalence of SF medicines, specifically antibiotics, is associated with AMR, conditional on various socioeconomic, environmental, health-related, governance and livestock production factors.¹⁴

Methods

Data on prevalence of SF antibiotics

To quantify the prevalence of SF antibiotics, we used the Infectious Diseases Data Observatory (Medical Quality Scientific Literature Surveyor, IDDO) database,^{8 15} which compiled data on medication quality through systematic reviews of relevant papers or reports spanning 140 countries or locations from the years 1963 to 2021. As of February 2024, the original datasets contained 5788 observations, each representing a drug sample tested and report type. Among these, 4929 observations had no missing variables of interest, including country, drug name, year, samples tested and results indicating whether the medicine was SF. The categories encompassed antimalarials, antidiabetics, antiretrovirals, antibiotics, veterinary medicines and cardiovascular medicines. The original dataset comprised various report types and collection methodologies. Within the original dataset (n=3195) which varied by country and location, publication, sampling method, outlet, sample and drug types, 36.1% of the observations were categorised as SF, with specific rates as follows: 29.8% for antibiotics (n=1092), 34.1% for antidiabetics (n=105), 42.7% for antimalarials (n=851), 16.7% for antiretrovirals (n=391), 43.0% for cardiovascular drugs (n=456), and 55.4% for veterinary medicines (n=300).

In line with recent studies,^{8 12 16 17} for our analysis, we exclude recalls, warnings, alerts and case reports of patients not responding to medication as well as other study types not specifically designed to estimate the prevalence of individual sample quality or address epidemiological questions. The sample for analysis only includes convenience and/or random surveys for antibiotics, for a total of 433 observations. We measured the prevalence of SF antibiotics as proxied by the proportion of samples included in a convenience and/or random survey that failed at least one quality test.⁸ In order to combine this dataset with available data on AMR, we manually assigned antibiotic classes (Aminoglycosides, Aminopenicillin, Cephalosporins, Fluoroquinolones, Macrolides, Nitroimidazoles, Penicillin, Semisynthetic, Sulfone, Tetracycline, Trimethoprim-Sulfamethoxazole) for each drug and constructed the average prevalence of SF antibiotics by country, year and antibiotics class.⁸ This yields a dataset encompassing 316 observations for antibiotics tested across 11 antibiotic classes in 65 countries spanning the years 1992–2019.

In online supplemental appendix figure A1 panel A, the prevalence of SF antibiotics by country is depicted as the average across the years, ranging from 0% to 100%. The highest percentile exhibits varying rates from 26.8% to 100% across 15 countries: Bangladesh, Cambodia, Chad, Cyprus, Côte d'Ivoire, Ghana, Haiti, Kazakhstan, Mexico, Niger, Nigeria, Slovenia, Syrian Arab Republic, Ukraine and the USA. Conversely, the lowest percentile ranges from 0% to 5.7% across 16 countries: Afghanistan, Argentina, Azerbaijan, Brazil, Canada, Ecuador, Ethiopia, Germany, Iceland, Italy, Mongolia, New Zealand, Romania, South Africa, the United Kingdom and Uruguay.

Data on AMR prevalence

To quantify the burden of AMR, we used the Global Bacterial Antimicrobial Resistance Burden Estimates 2019, provided by the Institute for Health Metrics and Evaluation and the University of Oxford.^{18 19} This dataset comprises modelling estimates of deaths, disability-adjusted life years, years lived with disability and years of life lost attributed to AMR across 88 pathogen–drug combinations for 21 Global Burden of Disease Study (GBD) regions and seven super regions in the year 2019.

The original data, sourced from administrative records, laboratory results, surveillance systems and literature reviews, encompassed 471 million individual records and 7585 study-location-years observations to inform the models. These data were then combined with findings from GBD 2019 to estimate the prevalence of a given pathogen resistant to an antibiotic class using spatiotemporal modelling techniques.¹⁸

For our main analysis, we used the estimated prevalence of AMR (%) in each pathogen-drug combination, by country in the year 2019. Additionally, we presented similar results for deaths attributable (ie, deaths directly caused by drug-resistant infections, resulting from ineffective treatment) or associated (ie, deaths occurring from a drug-resistant infection) with AMR. The original dataset consisted of 31 416 observations (204 countries×7 antibiotic classes×22 pathogens). We reconstructed the overall AMR prevalence for each antibiotic class and country, resulting in 1428 unique observations (204 countries×seven antibiotic classes). The average AMR prevalence across all countries and antibiotic classes is 35.6%.

In online supplemental appendix figure A1 panel B, the estimated AMR prevalence by country in the year 2019 is depicted. The highest percentile of AMR prevalence ranges from 38.8% to 47.0% across 15 countries, including, Bolivia, Cameroon, China, Côte d'Ivoire, Ethiopia, Guatemala, Kenya, Madagascar, Mexico, Mongolia, Niger, Nigeria, Pakistan, Senegal and Tanzania. Conversely, the lowest percentile, varying from 4.5% to 28.4%, encompasses another 15 countries: Afghanistan, Canada, Germany, Haiti, Iceland, Italy, New Zealand, Papua New Guinea, Slovenia, South Africa, Tajikistan, Thailand, Uganda, the United Kingdom and the USA.

Empirical specification

To estimate the association between the prevalence of SF antibiotics, averaged across years and AMR prevalence, estimated in 2019, we use a quasi-binomial generalised linear model, with a logit link function. This model is well suited for proportion data that shows overdispersion. (When we explore alternative measures of the burden of AMR such as estimated deaths attributable to or associated with AMR per 100 000 people, we alternatively use a negative binomial generalised linear model, with a log link function, suited for count data with overdispersion.) We accomplished this by merging the two constructed datasets described earlier—on prevalence of SF medicines and AMR—by antibiotic class and country. The final dataset for analysis consists of 257 unique observations, drawing from 63 countries and 7 antibiotic classes. (The merged list of antibiotic classes contains aminoglycosides, aminopenicillin, cephalosporins, fluoroquinolones, macrolides, penicillin, trimethoprim-sulfamethoxazole. It is possible that observations did not merge because either countries or antibiotic classes were not present in both SF and AMR datasets. For example, the following antibiotic classes present in the SF dataset were not modelled in the AMR database (nitroimidazoles, semisynthetic sulfone and tetracycline).) In adjusting our basic empirical specification, we included covariates identified to be associated with AMR,¹⁴ encompassing socioeconomic, environmental, health-related, governance and livestock production factors. Additionally, we incorporated county fixed effects to control for unobserved heterogeneity at the county level.

Based on the findings from¹⁴ and publicly available data, we selected a subset of factors to include in our analysis: (1) socioeconomic factors: GDP per capita (in 1000 US$) in 2019²⁰ and population density in 2019²¹; (2) environmental indicators: air pollution measured as PM2.5 in 2019²²; (3) health-related indicators: cardiovascular death rate in 2019²³ and human antibiotic consumption in 2018²⁴; (4) governance indicators: regulatory quality in 2019 measured as the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development²⁵; and (5) livestock production in 2019.²⁶ We chose these covariates due to the existing evidence on being associated with AMR. Lower GDP per capita as well as population density were found to be statistically significantly associated with AMR likely due to the uncontrolled spread of resistant bacteria occurring in environments with inadequate sanitation services, limited access to healthcare and dense population.²⁷ This association could also be explained by culturing bias, since in lower income settings, cultures are often reserved for cases where patients fail to respond to initial antibiotic treatment, due to limited resources or laboratory access, making poorer countries appear to have higher AMR due to selective culturing, rather than actual differences in AMR prevalence.28,30 Described the association between PM2.5 and AMR globally suggesting that PM2.5 can not only carry antibiotic-resistant bacteria and genes, which can enter the human body through the respiratory system when inhaled, but also facilitate the transmission of resistance genes between bacteria. Cardiovascular death rate was found to be positively associated with AMR, as antimicrobials are frequently used in prevention and treatment of infection in patients with cardiovascular diseases.³¹ Evidence also exists on the associations between general human antibiotic consumption³² and livestock production^{33 34} and AMR. Finally, there exists a significant negative association between regulatory quality and AMR,³⁵ as poor governance may be associated with the misuse and overuse of antimicrobials for infection treatment, speeding up the spread of AMR.

An alternative empirical model included additional covariates: (1) socioeconomic factors: the Human Development Index in 2019,³⁶ the share of the population with unimproved drinking water in 2019,³⁷ the share of the population with unimproved sanitation services in 2019³⁷ and the proportion of the population living below the poverty line in 2019;³⁸ (2) five governance indicators that measure control of corruption, government effectiveness, political stability and absence of violence/terrorism, voice and accountability and rule of law in 2019^{25 39}; and (3) antibiotic policy and regulation factors measured as two categorical variables capturing the development stages of national action plans on AMR and national monitoring systems for antibiotic use in humans in 2019.⁴⁰ Please see online supplemental appendix table A1 for a list of all variables considered, their measurements and links to data sources.

Main results

Figure 1 illustrates the prevalence of SF antibiotics and AMR by country, using all available data. Colours representing continents suggest that Africa exhibits high AMR, yet with dispersed prevalence of SF antibiotics; notably, Ghana and Côte d'Ivoire record a prevalence of SF antibiotics higher than 50%. An exception is Sierra Leone, which shows both a low prevalence of SF antibiotics and AMR. Similarly, Asia demonstrates an AMR prevalence above 20%, with most countries having prevalence of SF antibiotics below 40%; however, Bangladesh and Syria exhibit a prevalence of SF antibiotics above 60%. Conversely, South America displays a low prevalence of SF antibiotics (below 20%) but a higher AMR prevalence (between 20% and 40%). Data from the two represented countries in Oceania show a prevalence of SF antibiotics and AMR below 20%. In contrast, there is no clear pattern for North and Central America: Canada shows a low prevalence of SF antibiotics and AMR, while the USA displays a prevalence of SF antibiotics of 30% and AMR prevalence close to 20%; Mexico fares worse on both dimensions, with a prevalence of SF antibiotics of 43% and an AMR prevalence of about 40%. Note that both the USA and some European countries show relatively high prevalence of SF antibiotics (30% in USA, 40% in Ukraine, 100% in Cyprus and Slovenia). While we argue that this could be influenced by data bias resulting from advanced testing technologies to detect SF medicines, better regulatory environments or the rise of online pharmacies that may also facilitate the distribution of SF medicines,⁴¹ we need to acknowledge that the number of observations in the data is limited (N=4 for USA, N=1 for Ukraine, N=1 for Cyprus, N=1 for Slovenia), thus the data may not be very representative for these countries. Similarly, a few Western European countries show extremely low prevalence of SF antibiotics (0% for Italy, Germany, United Kingdom and Iceland), which could also be more likely attributed to the limited available data (N=1 for Italy, N=2 for Germany, N=2 for United Kingdom, N=1 for Iceland). Taking these limitations into consideration, the patterns we found also remain consistent when considering only more recent data for prevalence of SF antibiotics (starting in 2000, online supplemental appendix figure A2) (We also explored results by continent. The main results are driven by African countries, but this is more likely due to a higher number of observations (n=133) than in other continents (n=89 for Asia, n=13 for North America, n=10 for Europe, n=8 for South America, n=4 for Oceania).).

Figure 1. Prevalence of substandard and falsified (SF) antibiotics and antimicrobial resistance (AMR), by country (1992–2019). The prevalence of SF antibiotics is measured across the years when data are available (1992–2019), while the AMR prevalence is only available in 2019. The two outliers are Cyprus (2005, 2007) and Slovenia (2005, 2007) where the prevalence of SF antibiotics is 100%. Data for Cyprus and Slovenia are from: (i) a survey focusing on the quality of generic clarithromycin products in 18 countries,⁵⁸ and (ii) a convenience survey studying active ingredients in 16 commercial formulations of ciprofloxacin tablets in different countries.⁵⁹.

Next, we delve into the association between the prevalence of SF antibiotics and AMR. Raw data reveal a positive association between the two variables. The coefficient of 1.006 (CI 1.002 to 1.010) indicates that for each unit increase in the prevalence of SF antibiotics, the odds of AMR prevalence increase by 1.006. This means that if the prevalence of SF drugs was to increase by one unit (eg, 1 percentage point increase), the odds of experiencing AMR would increase by approximately 0.6% (table 1, column 1). If, for example, SF prevalence would increase by 10 percentage points, the odds of AMR would increase by approximately 6.1%. Yet, note that specific estimates of AMR prevalence should be interpreted with caution as they depend on the baseline prevalence in the population.

Table 1. Association between the prevalence of substandard and falsified (SF) antibiotics and antimicrobial resistance (AMR).

	(1)	(2)	(3)	(4)	(5)
Prevalence of SF antibiotics	1.006	1.006	1.006	1.007	1.006
	(1.002–1.010)	(1.002–1.010)	(1.000–1.013)	(1.002–1.011)	(1.000–1.013)
GDP per capita (in 1000 $US)		0.979	0.998	0.984	0.993
		(0.969–0.989)	(0.989–1.008)	(0.972–0.996)	(0.986–1.000)
Population density (in 1000)		0.720	0.595	0.663	0.409
		(0.560–0.927)	(0.408–0.867)	(0.488–0.899)	(0.375–0.445)
PM 2.5		1.002	1.022	1.001	0.983
		(0.998–1.006)	(0.997–1.049)	(0.997–1.006)	(0.980–0.986)
Cardiovascular death rate		0.999	1.002	0.999	1.001
		(0.999–1.000)	(1.001–1.003)	(0.998–1.001)	(1.001–1.002)
Human antibiotic consumption		1.022	1.046	1.028	0.959
		(1.009–1.035)	(1.032–1.059)	(1.015–1.042)	(0.958–0.959)
Regulatory quality		1.004	0.590	0.835	0.286
		(0.858–1.175)	(0.511–0.680)	(0.593–1.177)	(0.260–0.315)
Livestock production (in 1000)		0.538	8.90e-07	0.713	555.3
		(0.0513–5.635)	(1.53e-08–5.18e-05)	(0.0601–8.464)	(270.9–1138)
Constant	0.482	0.439	0.189	0.682	159.4
	(0.416–0.558)	(0.270–0.714)	(0.0953–0.373)	(0.163–2.851)	(31.42–808.3)
Additional covariates				✓	✓
Country fixed effects			✓		✓
Observations	257	257	257	257	257
Mean dependent variable	0.356	0.356	0.356	0.356	0.356

AMR prevalence is estimated from the Global Bacterial Antimicrobial Resistance Burden Estimates (2019), provided by IHME and the University of Oxford. The prevalence of SF antibiotics is the proportion of antibiotics that were tested as substandard or falsified. Gross domestic product (GDP) per capita is in 1000 $US; population density is in 1000 people per sq. km of land area; particulate matter (PM) 2.5 is in micrograms per cubic metre; cardiovascular death rate refers to deaths due to cardiovascular disease per 100 000 people; human antibiotic consumption in doses for antibiotic drugs uses per 1000 population per day; regulation quality is a standardised index with higher score referring to better performance as gathered from 30+ data sources worldwide; livestock production is in 1000 tonnes including meat, milk and dairy products. The empirical model is a quasi-binomial generalised linear model with a logit link function. ORs and 95% CI are reported. See online supplemental appendix table A1 for a list of all variables considered, their measurements and links to data sources.

IHME, Institute for Health Metrics and Evaluation.

When we include a subsample of factors potentially associated with AMR (table 1, column 2), the coefficient has the same magnitude at 1.006 (CI 1.002 to 1.010). Among the covariates, GDP per capita and human antibiotic consumption show associations with AMR prevalence (table 1, columns 1 and 3). Constraining the variation to observations within countries by controlling for country-fixed effects (table 1, column 3), the results remain robust. The estimates on the prevalence of SF antibiotics persist when additional covariates are included, both without and with country-fixed effects (table 1, columns 4–5). (Please refer to online supplemental appendix table A2 (columns 1-2) for a display of the coefficients on the covariates in the alternative adjusted models (table 1, columns 4–5).) When additional covariates on drinking water and sanitation coverage, poverty rate and antibiotic policy and regulation (when available) are included (N=216) (online supplemental appendix table A2, column 3), the coefficient on prevalence of SF antibiotics remains robust (1.006, CI 1.001 to 1.012). In summary, these results suggest that, conditional on socioeconomic, environmental, health-related, governance indicators and livestock production factors, the prevalence of SF antibiotics is associated with AMR.

These estimates are also robust (1.004, CI 1.000 to 1.009) to the exclusion of countries with fewer than four observations during the study period, corresponding to the 25th percentile of the distribution (online supplemental appendix table A3). Additionally, we explore alternative measures of AMR burden in online supplemental appendix table A4, specifically estimated deaths attributable to or associated with AMR per 100 000 people. A positive association is observed between the prevalence of SF antibiotics and associated death rates related to AMR (online supplemental appendix table A4, column 4, 1.657, CI 1.104 to 2.486). When controlling for covariates and country-fixed effects, the coefficient magnitude remains larger for associated deaths (1.493, CI 0.967 to 2.305, online supplemental appendix table A4, column 6) than for attributable deaths (1.353, CI 0.671 to 2.728, online supplemental appendix table A4, column 3). We further examine heterogeneous associations by antibiotic class to assess whether this relationship is driven by specific classes of antibiotics. With the limitation of a small sample size in some subanalyses, we confirm that the positive association between SF antibiotic prevalence and AMR is primarily driven by penicillin (online supplemental appendix table A5, column 6; 1.008, CI 1.002 to 1.015). We also constructed the prevalence of SF non-antibiotics as the mean for all non-antibiotics by year and country (online supplemental appendix table A6). We test whether a similar association would exist between AMR and prevalence of SF products such as non-antibiotics, which would not be expected to be associated with AMR. We confirm that the prevalence of SF non-antibiotics is not associated with AMR (online supplemental appendix table A6, column 5, 1.000, CI 0.993 to 1.007), reinforcing our main results. Finally, we restrict the sample to random surveys within the SF antibiotics dataset. With a sample size of 61, this empirical specification does not reveal a positive association between SF antibiotics and AMR (online supplemental appendix table A7).

Discussion

AMR poses a significant challenge to global health.⁴² Various studies have examined the primary drivers of AMR, either through ecological^{14 43} or social science studies.^{14 44} Overall, the evidence consistently highlights (human or animal) antibiotic consumption as a key factor.¹⁴ This could stem from factors such as overprescription in human medicine⁴⁵ and antimicrobial use within the livestock sector for food production.^{46 47} Additional factors which have been linked to AMR may include socioeconomic factors like poverty,^{48 49} community access to water and sanitation service^{43 50} governance⁵¹ and out-of-pocket expenditures.⁴⁴ There is suggestive evidence that SF medicines with insufficient amounts of the active component are associated with the development of AMR.^{11 52} However, not many studies take this factor into consideration, likely due to the difficulty in detecting SF medicines. Despite these limitations, efforts have been made to combine available data into a comprehensive database of prevalence of SF medicines⁸ and for the burden of AMR.⁵³ This study uses the most comprehensive available databases on SF antibiotics1215,17 and the estimated burden of AMR^{18 19} to investigate the relationship between the prevalence of SF antibiotics and AMR, with a focus on antibiotics. Our study quantitatively demonstrates that the prevalence of SF antibiotics is associated with the estimated AMR prevalence as well as death rates attributable to or associated with AMR, when controlling for socioeconomic, demographic, environmental, health-related, governance and livestock production factors.

This research is not without limitations. First, we relied on the best available data to comprehensively gather information across countries and years on the prevalence of SF antibiotics as well as the estimated burden of AMR. Nevertheless, these data are limited for both key variables of interest, and further initiatives are required to better coordinate efforts to measure the prevalence of SF medicines and AMR worldwide. For example, the prevalence of SF medicines is defined as ‘the proportion of samples included in a prevalence survey that failed at least one quality test’,1215,17 but these data should not be interpreted as a proportion of the global supply of antibiotics is SF, since the data quality is still poor and hard to generalise. It is important to acknowledge that the IDDO database may have incomplete coverage, which can result in the under-reporting of instances of SF medicines, particularly in countries with weak surveillance systems. Consequently, the quality and reliability of the data may also vary across countries, affecting the overall accuracy of the prevalence estimates. Within countries, the data on SF medicines are usually from the sites of the studies and may not be representative of the country as a whole. There is further potential for bias, as the samples collected in surveys included may not be randomly selected, and surveys that do not find any SF medicines may not get published due to publication bias. Another concern relates to the limited number of studies available in the dataset that report the percentage of active pharmaceutical ingredients in the antibiotics, which would provide a more appropriate measure of the prevalence of SF medicines and could lead to an underestimation of the prevalence of substandard medicines. Similar concerns around data reliability may apply to the estimates of AMR prevalence.^{18 19} As mentioned before, poorer countries may appear to have higher AMR rates due to selective culturing practices rather than actual differences in AMR prevalence. Second, and related to the previous point, we are unable to fully explore the dynamics of the association between the prevalence of SF antibiotics and AMR. Moreover, we only have modelling estimates of the burden of AMR (%) or attributable or associated deaths for the year of 2019. Consequently, we cannot measure how the prevalence of SF antibiotics is associated with AMR over time. Although it is possible to measure the prevalence of SF antibiotics over time (1992–2019), the dataset on SF medicines has gaps in the time series data and may have temporal limitations, reflecting specific efforts by governments, international agencies or researchers changing over time.

Third, it is important to note that this analysis only explores associations and we are unable to draw causal conclusions that higher prevalence of SF antibiotics causes a higher (estimated) burden of AMR. Related to this, areas where there is high prevalence of SF antibiotics may also likely have other AMR drivers such as inappropriate prescriptions of antibiotics for which we lack global data.

We acknowledge that addressing this research question is inherently dynamic and complex. Our study seeks to assess the relationship between the prevalence of SF medicines and AMR by focusing on antibiotics and estimating a static empirical model. Given the current data availability, our best estimate for the prevalence of SF antibiotics is calculated as the average proportion of samples failing at least one quality test across years for each country. However, we recognise that the prevalence of SF antibiotics can vary significantly over time, between countries, and among antibiotic types, so producing an average across these factors may limit accuracy. To demonstrate the robustness of our estimates, we exclude countries with a limited number of observations over the study period, examine associations between SF antibiotics and alternative outcomes—such as estimated deaths attributable to or associated with AMR per 100 000 people—and explore heterogeneous associations by antibiotic class. More importantly, we attempt to control for observed factors confounding the relationship between SF antibiotics and AMR, but we note that our empirical model may not capture all possible confounders. Despite all these caveats, our analysis identifies SF antibiotics as a factor correlated with AMR across countries worldwide.

Conclusion

AMR poses a significant social and biological challenge, one that is exceedingly complex to tackle. The WHO identifies AMR as one of the top 10 threats to global health, warning that it has the potential to reverse decades of progress in reducing morbidity and mortality from infectious diseases. Bacterial AMR is linked to 4.95 million deaths, with 1.27 million directly attributable to it, and the burden is particularly pronounced in low-income settings, such as sub-Saharan Africa.¹⁸ Projections suggest that by 2050, annual deaths related to AMR could reach up to 10 million. It is estimated that AMR could result in a US$3.4 trillion reduction in GDP annually and push an additional 24 million people into extreme poverty within the next decade.⁵⁴

Understanding the factors associated with the burden of AMR is crucial for effectively addressing this global challenge. Our research indicates that substandard and falsified (SF) antibiotics are associated with the prevalence of AMR, even when accounting for other well-established factors correlated with AMR. Policymakers may consider prioritising addressing the issue of SF medicines and devise strategies to combat it. While there have been notable initiatives and successes in addressing SF medicines globally,^{55 56} significant gaps remain in pharmacovigilance programmes, data collection systems, human resources for detecting SF medicines and the development of simple and affordable testing methods (eg, Minilab) or new mobile authentication services (eg, Sproxil). AMR continues to escalate as a critical global issue affecting humans, animals and the environment. Embracing the One Health approach, prevention stands at the forefront of the necessary actions to mitigate the rise of AMR.⁵⁷ This study highlights pharmaceutical and chemical manufacturing as a factor associated with AMR. However, further research and more representative data are needed to determine the extent to which this association could be explained by a direct causal relationship. Stakeholders may consider addressing SF medicines as part of broader efforts to mitigate AMR.

Data availability statement

The data used for the analysis are publicly available.