This study was conducted in Shandong province in eastern China, which had a total population of 101, 699, 900 at the end of 2021 [20]. The method of purpose sampling was used in the study. Two counties S and Y in City L located in the western area of Shandong Province were selected as the target locations. The two counties have similar mid-level development indexes in terms of Gross Domestic Product (GDP), number of permanent residents, number of health technicians, and number of township hospitals. In addition, the pandemic prevention and control policies were highly synchronized during the COVID-19 pandemic in PHSs in Shandong, China. Therefore, the sample township hospitals are representative of the majority of township hospitals in rural Shandong. They can representatively display the changes of antibiotic consumption trend and the use pattern before and after the outbreak of the COVID-19 pandemic.

Township hospitals are primary healthcare centers in rural China. Their primary missions are to provide rural population with extensive primary outpatient services (i.e., establishment resident health records, common diseases treatment, non-communicable diseases control, vaccination and health education) and limited inpatient service (i.e., basic clinical examination, surgical treatment for a few diseases). Antibiotics are used by inpatients and outpatients, but most of them are consumed by outpatients of the township hospitals in Shandong, China. During the early period of the COVID-19 pandemic, the township hospitals changed their daily work and business hours. For example, they greatly shortened their opening hours, and most of the medical staff were deployed to assist in tracking down infected individuals and conduct epidemiological investigation. At the same time, the number of patients in township hospitals decreased sharply, for those, patients with fever were transferred to COVID-19 pandemic specific clinics. In addition, antibiotics in pharmacies were not available for patients without physicians’ prescription. Instead, antibiotics were mostly prescribed in PHSs in rural China.

Data were collected from all township hospitals (38 in total) in counties S and Y in City L in Shandong Province. In Shandong Province, PHSs are only allowed to purchase, store and distribute drugs listed in the Regulations on Drug Administration based on the zero price difference policy. At the same time, doctors in PHSs are only permitted to prescribe antibiotics listed in the List of Essential Drugs [21].

We conducted a 3-year natural, before and after, quasi-experimental study in the real world. A controlled interruption time series design was used to collect data at several time points before and after the COVID-19 pandemic outbreak, aiming to investigate the immediate and long-term impacts of the pandemic on antibiotic consumption. In this study, the effectiveness of data was estimated by controlling the baseline level and the trend [22], the longitudinal characteristics of the data ensured the robustness of results [23].

We retrospectively collected aggregated monthly antibiotic consumption data from 38 township hospitals in counties S and Y of City L in Shandong, China, from January 2019 to December 2021. The data were collected with the following method. Two researchers were responsible for preparing Excel forms with headers. These persons emailed feedback to the researchers. At the end of December 2019, and then it spread to Shandong Province due to population migration during Chinese New Year holidays [24]. Therefore, the antibiotic consumption data over 12 months before and 24 months after the outbreak of the COVID-19 pandemic was included. The antibiotic consumption data mainly included the following variables: names of township hospitals, manufacturer of antibiotics, unique chemical substance name, generic name, dosage form, unit strength, specification, unit (by box, bottle, or ampoule), price per unit, inventory at the beginning of the month, monthly purchase quantity, monthly ex-warehouse quantity, monthly inventory at the end of the month, etc.

The original data were managed in Microsoft Excel (version 2019). We strictly controlled the quality of the data, carefully sorted and summarized them after receiving the emails from the persons in charge of the township hospitals. The validity of the data was assessed by two researchers engaged in antibiotic resistance research. They mainly made logic proofreading and supplemented outliers or missing values. During the process of cleaning data, the data with problems were separately listed and sent back to the corresponding persons in charge via emails for supplementing or correction. For example, they should supplement monthly usage, specifications, and dosage forms of drug, and check whether the end-of-month inventory of the drugs as consistent with the beginning-of-month inventory of the next month.

We assessed antibiotic consumption data according to Anatomical Therapeutic Chemical (ATC) classification J01 (i.e., antibacterial for systemic use) [25]. Additional file 1: Table S1 provides a full list of antibiotics analyzed in this study. A total of 55 unique chemical substance names (55 ATC-5 codes) were identified for single or combined antibiotics; they were first classified into 19 ATC-4 classes and secondly into 6 ATC-3 groups. At the same time, antibiotics were classified according to the WHO AWaRe categories (version 2021), aiming to analyze the use patterns of antibiotics [26]. Only one type of antibiotic, Fosfomycin, was classified as Reserve. Therefore, the antibiotics of the Reserve category were not included in the analysis. Instead, only Access and Watch category antibiotics were included. The Access category antibiotics are recommended by the WHO as the first or second choice for empirical treatment, whereas the increased consumption of Watch category antibiotics will aggravate antibiotic resistance.

The study assessed the immediate and long-term impacts of the COVID-19 pandemic on the changes in antibiotic use. The primary outcome indicator of interest was the monthly antibiotic consumption. Antibiotic consumption was measures using Defined Daily Dose (DDD), an assumed average maintenance dose developed by WHO to compare drug consumptions [25]. In this study, the DDD value of each drug was determined according to the Guidelines for ATC classification and DDD assignment 2021 [26]. The DDD equivalence per package [DPP = (unit strength * package size/DDD)] of drugs was calculated according to ATC templates. The total consumption of each group of procured drugs (DDDs) was estimated as the summed DPPs of all-inclusive products [27].

$$DDD_ = \mathop \sum \limits_^ \left( \times N_ } \right)$$

where Ni represents the number of packages of a certain antibiotic product(i) used in the township hospitals.

This study used the descriptive statistical method to quantify the patterns and trends of antibiotic consumption. First, the relative changes in the overall antibiotic consumption among all PHSs in 2020 and 2021 compared to the corresponding periods in 2019 were described. Second, a monthly antibiotic consumption trend chart based on ATC classification and WHO AWaRe category was prepared to observe and describe changes in antibiotic consumption from January 2019 to December 2021, respectively.

By using interrupted time series analysis, the immediate and long-term impacts of the COVID-19 pandemic on the trends of antibiotic consumption in PHSs were assessed [23, 28]. The time unit in this study was month, for all the data were collected monthly at even intervals. With December 2019 taken as the intervention time point, the data over 12 months before the intervention and 24 months after the intervention was finally included in the study. A segmented regression model was established as follows:

$$Y_ = \beta_ + \beta_ \times Time_ + \beta_ \times Covid_ + \beta_ \times Longterm_ + \beta_ \times Cold + \varepsilon_$$

where \(Y_\) is the independent outcome variable of month t (antibiotic consumption DDDs); where \(Time_\) is a continuous time series variable (1,2,3… 36); Covid-19t represents the intervention time indicator before and after the outbreak of the COVID-19 pandemic. The intervention time nodes lie in the 12th month before the outbreak of COVID-19 pandemic (t = 0) and the 24th after the outbreak of COVID-19 pandemic (t = 1); where Long-termt represents the time since the outbreak of the COVID-19 pandemic; its value before the outbreak of COVID-19 pandemic is 0, while its value after the outbreak of COVID-19 pandemic is 1, 2, 3, …24, corresponding to January 2020 to December 2021 (long-term effect). In addition, a dummy variable Cold was set to control the extreme value of antibiotic consumption during the coldest period of Chinese Spring Festival, which was the wild data point of this study [28,29,30,31]. By referring to previous studies, we assigned 1 to the variable Cold in December and January of each year, and 0 for the rest of months of these the year [32].

In this model, β0 is used to estimate the level of antibiotic consumption at the beginning of the time series. β1 is used to estimate the change trends of antibiotic consumption prior to the outbreak of the COVID-19 pandemic. β2 is used to assess the changes immediately after the outbreak of the COVID-19 pandemic. β3 reflects the monthly changes in antibiotic consumption after the outbreak of the COVID-19 pandemic. If it is different from the trend before the outbreak of the COVID-19 pandemic, it suggests that the outbreak of the COVID-19 pandemic has a long-lasting impact on the antibiotic consumption. β4 is used to estimate the weather effect of the coldest period. \(\varepsilon_\) is an estimate of the random error at time t. In addition, Durbin–Watson test was performed to verify the existence of the first-order autocorrelation (if the value is about 2, it indicates that there is no autocorrelation) [33]. In case of an autocorrelation, the regression will be estimated using the Prais-Winsten method [32].

All statistical analyses were performed in STATA version 15.0 (STATA Crop LP, College Station, TX, USA), and P < 0.05 was considered statistically significant.