Persistence of resistance: a panel data analysis of the impact of antibiotic use on the spread of resistance

Antibiotic use data

Usage data for 26 European countries for 2008-2018 are obtained from the IQVIA MIDAS database. IQVIA reports the sales volume of antibiotic molecules used in human medicine, based on national surveys. Use the ATC/DDD Index 2020 to convert antibiotic sales to his defined daily dose (DDD). The World Health Organization defines DDD as her estimated average daily maintenance dose of drugs used for major indications in adults. Adjust the population using population estimates from the World Bank Databank to get his DDD per 1000 (DpTI).

Antibiotic molecules are active compounds that can be broadly classified into classes of antibiotics based on their mechanism of action against bacteria. Using the ATC/DDD index, we aggregate antibiotic DpTIs into 12 antibiotic classes, some of which contain only a single antibiotic (Supplementary Table S1). Therefore, sales are identified by antibiotic class, year, and country. We categorize the 26 European countries into Northern Europe, Southern Europe, Eastern Europe, and Western Europe, as detailed in the supplementary table. S2.

As shown in the Supplementary Figure, usage levels vary widely by class and country. S1–S3. In order to be able to compare the usage data, we convert the DpTI as follows: z-Score.a z-score is defined as the difference between the observed value and the sample mean divided by the standard deviation of the class for each country from 2008 to 2018.

Antibiotic resistance data

The European Antimicrobial Resistance Surveillance System (EARS–Net) collects data on antimicrobial resistance of eight different pathogens against 12 antibiotic classes, as reported in the Atlas of Surveillance of Infectious Diseases.Targeted bacteria and antibiotic combinations are detailed in the Supplementary Table S3According to the EARS–Net data document, all major geographic regions are covered and, on average, the data are considered representative of the national epidemiology. However, population coverage varies by reporting country and the population under surveillance changes over time.

The data document the percentage resistance of specific bacteria by specific antibiotic class, year and country.We classify eight different bacterial pathogens into Gram-positive and Gram-negative as detailed in the Supplementary Table S4Therefore, as shown in the Supplementary Figure, resistance data vary between 0 and 100% by country, combination of bacterium-antibiotic class, and year 2008-2018. S4–S7.

sample definition

Combine the two datasets using year, country, and antibiotic class to obtain resistance and usage data for 26 European countries, 11 years, and 26 bacterial class combinations. Create a panel on the universe of all units, with units given by country, year, and bacteria class. Some countries did not report resistance for all bacterial classes in all years, so the panel is unbalanced with 6586 observations.

Our main dependent variable is the prevalence of resistance given by the proportion of resistant isolates identified by EARS–Net.how to use z– score or change z-Score as the main explanatory variable, depending on the specifications adopted. For analysis, we select his six-year panel of resistance data covering 2012-2017. This selection helps set up an event study design with an event horizon from 1 year before usage to 4 years after usage so that usage data from 2008 to 2018 is required.

empirical model

We use three related empirical models to estimate the causal effect of use on resistance. – Study (E–S) model using binned endpoints. Binning refers to how the last lag (lead) is treated as an open interval that captures all known changes that have happened (or will occur) in the past (future).Our outcome variable is bacterial class resistance Iwithin the country cA year t, \(R_{i,{{{{{{\mathrm{c}}}}}}}},t}\). The explanatory variable of interest in model (1) is z– Usage score, Happening j Period for bacteria class Iwithin the country cA year t, \(U_{i,{{{{{{\mathrm{c}}}}}}}},t – j}\)The explanatory variables of interest in models (2) and (3) are z– Usage rate, Happening j Period for bacteria class Iwithin the country cA year t, \(\Delta U_{i,{{{{{{\mathrm{c}}}}}}}},t-j}\).

Models (1) and (2) are specified as

With fixed effect:

First difference:

The event window spans the period from one year before use (or change of use) to four years after. bacteria class (meters_I) and country (meters_c) the fixed effects of model (1). It accounts for unobserved bacteria, class, and country effects and is constant over time. In model (2), first-order differential resistance and use controls for these effects. For both models, the year fixed effect (I_t). The error term is given by e_{I, c, t}. Coefficient c_j and l_j It shows the marginal effects of use on resistance and measures the slope of these effects from one year to the next.estimated value of c_j and l_j Models (1) and (2) are only expected to differ significantly if the effect of use on resistance continues to evolve beyond the 4-year window. [25].

Model (3) is specified as

binned variable \(v_{{{{{{\mathrm{i}}}}}}}},c,t}^{\,j}\) Defined as follows:

Model (3) is a regression of binned levels of change.coefficient b_j is the therapeutic effect, j A period before or after use that dynamically expands over time and is expressed relative to a reference period. Base period coefficients are normalized to zero. Binning the upper and lower endpoints is equivalent to assuming \(\gamma_j = 0\) for all j> 4 and for all \(j\le – 2\)In other words, the model assumes that the effect of use on resistance is constant in all cases. j> 4 and for all \(j\le – 2\)Due to the nature and structure of binned variables, the ES model requires data two years before the event. These assumptions about effect windows admit that an infinite past and an infinite future cannot be used to estimate dynamic effects due to limited data availability and sample limitations. [25].

Model 3 provides easily interpretable coefficients b_j On the other hand, the coefficient c_j and l_j from models 1 and 2 should be linearly transformed to derive the dynamic effects. Statistical properties such as consistency and asymptotic normality are preserved during this linear transformation of D–L model estimates, and variances and covariances of estimated parameters are preserved. c_j and l_j can be used to recover the standard error of the estimate b_j Using the standard linear combination formula [25]These estimated dynamic treatment effects are unbiased under linear and additive assumptions.Event study and D–L models with binned endpoints are expected to yield similar treatment effects [25].

The reported dynamic treatment effects estimated using all three models are interpreted as effects on tolerance by 1 standard deviation increase in dosage. This effect can be observed in years before and after use. To identify dynamic treatment effects, no statistically significant effect on tolerance is required prior to use. In addition, identification is achieved within bacterial class and within the country over time.

It also estimates all models using segmented data. Specifically, we split the data into Gram-positive and Gram-negative data and assess whether their responses differ. Also, to assess whether the same effect is observed in all regions, we split the countries into four regions: West, North, East, and South.

Additionally, segment the data according to whether the usage change is positive or negative. Using our D–L spec, with slight modifications, we test the effect on resistance of positive and negative changes in usage separately. Our hypothesis is that a positive change in usage increases resistance, and a negative change in usage decreases resistance. An empirical model for evaluating this hypothesis is

where \(D_{i,{{{{{{\mathrm{c}}}}}}}},t – j}\) An indicator variable that indicates whether there was a positive or negative change above a certain threshold in antibiotic use to treat a bacterial class. I within the country c, jperiod away.