Antibacterial use by birth year and birth season in children 0-2 years in Norway

antibacterials in children follows seasonal cycles, and time to first treatment depends on birth season. The aim of this study was to describe dispensing rate, one-year periodic prevalence, and age at first prescription in children aged 0-2 years in Norway. Methods: We used data from the Norwegian prescription database and included all dispensed prescriptions on systemic antibacterials in 2008-2017 during the first three years of life to children born 2005-2014. We calculated age by subtracting birth month and birth year from date of collection of prescription. We used multiple linear regression to investigate the effect of birth season on age at first dispensed prescription. Results: We included 714 262 prescriptions to 281 888 individuals (53.1% boys). In 2016, one-year-old boys had the highest periodic prevalence (35.6%) and the highest dispense rate (545/1000 individuals), followed by one-year-old girls (32.6%, 478/1000 individuals). The lowest prevalence and dispense rate in all age groups was found towards the end of the period. Winter months had the highest proportion of dispensed prescriptions, and children born in autumn were significantly younger when collecting their first prescription. On average, boys collected their first prescription 26 days younger compared to girls. Conclusion:


INTRODUCTION
In 2017, there were > 800 000 consultations for children aged 0-5 years at general practitioners or emergency rooms in Norway. Among these, 34% were diagnosed with a respiratory tract infection (RTI) (1). In 2017, phenoxymethylpenicillin, Penicillin V (PcV), a common RTI antibacterial, was the most frequently dispensed drug in the age group 0-17 years in Norway (2). On the other hand, less than 30% of upper RTIs and radiological verified pneumonia in children < 5 years presumably have bacterial etiology (3,4).
In Norway in 2017, sex-and age specific one-year prevalence on dispensing of antibacterials was highest for children of 2 years, where 24% of girls and 27% of boys received at least one prescription (2). In Denmark in 2012, there were similar patterns with the highest incidence rate among those < 2 years, and the highest periodic prevalence among those 0-1 year, followed by the age group 2-4 years (5).
A recent Danish study investigated the association between birth season and age at first dispense of antibacterials and overall dispensing of antibacterials. The study included children born in 2004-2012 and showed a seasonally varying risk for receiving an antibacterial drug conditional on age. For example, children born in fall had at six months, a 4.9 percent point higher risk having received an antibacterial compared to children born in spring (6).
The high use of antibacterials in young children is of concern due to emerging antimicrobial resistance, and changes in the gut microbiota early in life, possibly associated with increasing risk for chronic diseases later in life (7)(8)(9)(10). The national strategy to curb development of antibacterial resistance seek to reduce the overall consumption of antibacterial agents, with a special focus on broad-spectrum agents (11). The strategy aims to reduce antibacterial use by 30% from 2012 to the end of 2020, targeting unnecessary or irrational prescribing to children receiving antibacterials for RTIs of viral etiology.
The aim of this study is to describe the dispensing rate and one-year periodic prevalence of systemic antibacterials to children aged 0-2 years in Norway in the period 2008-2017, and to explore if there is an association between birth season and age at first redeemed prescription for antibacterials.

MATERIAL AND METHODS
Data were from the Norwegian prescription database (NorPD). Details on NorPD is described elsewhere (12).
We used the following variables from NorPD: sex, year of birth, month of birth, unique personal identifier (ID), county of residence, date for redeemed prescription from pharmacy and ATC-code on fifth level.
Our data material consists of all prescriptions on ATC code J01 with a valid ID and month of birth from 2008-2017 to children born in 2005-2014. We excluded records for individuals > 1095 days (3 years), and those who died during the observation period. In addition, we aggregated prescriptions if an individual had received two or more prescriptions on the same antibacterial drug (ATC level 5) on the same date, as we considered them as the same treatment episode. Population statistics was obtained from Statistics Norway (SSB).

Analysis
Our data material contains information on birth month, but not date of birth. Consequently, we assigned a fictitious birth date of the 15th of their birth month, and age was calculated as date of dispense minus this date.
We calculated the annual sex specific dispense rate for use of antibacterials from 2008-2017. We also calculated the one-year periodic prevalence of children having received at least one prescription in each calendar year, and used population statistics on January 1 st each year as the denominator.
Furthermore, to investigate the effect of birth season on age at firstly dispensed prescription, we used a multiple linear regression model including children born in 2008-2014. We excluded prescription records to those born in 2005-2007 as we could not fully capture the outcome variable (age at first dispensed prescription) during this period. We further excluded those receiving more than 10 prescriptions during their first three years of life and those without a valid county (Fig.  1). Variables included in the regression analysis were based on a literature review and available data. Directed acyclic graphs (DAGs) were used to visualize the causal structure of the model. In addition to birth season, we included sex, birth year and county of residence. A geographical dummy variable identified the three counties with the highest dispensing rate, and the three with the lowest in 2014. These were respectively Nord-Trøndelag, Akershus and Vestfold, and Finnmark, Troms and Telemark. We used the remaining 13 counties as reference.
The other reference categories in the regression model were fall as birth season, birth year 2008, and female sex. All variables were included as categorical variables as they did not comply with the requirement of linearity between outcome and independent variables.

Data management and ethics
All calculations were performed using SPSS version 27 and STATA version 15.
The project was approved by Regional Committees for Medical and Health Research Ethics (REK) and was considered to be within the purpose of the NorPD, and was approved according to registry legislation (13).

RESULTS
The data contains dispensed systemic antibacterials from Norwegian pharmacies in the period 2008 to 2017, for children born in 2005 to 2014. In the period 714 262 prescriptions were dispensed of which 53.7% were to boys. The records include dispensed prescriptions to 281 888 individuals of which 53.1% were boys (Fig. 1). Table 1  The dispense rate was higher in boys compared with girls for all age groups. 1-year-olds had the highest dispense rate, followed by 2-year-olds. The overall highest dispense rate was in 2010 with 502 per 1000. In the same year, the rate was 763 and 665 per 1000 respectively, for 1-year old boys and girls (Fig. 2). There has been a fall in total dispense rate since 2012, apart from a small rise in 2014 and 2016. 2016 does not  The one-year periodic prevalence was highest for 1year-olds for both boys and girls (Fig. 3). The one-year prevalence was over 40% in 1-year-old boys in 2008 and 2010-2012 but decreased until 2015. For girls, the highest periodic prevalence was approximately 38% in 2010 and 2012. The overall periodic prevalence decreased from 28.9% in 2008 to 24.4% in 2017.
In our data, the most dispensed antibacterial substances were Penicillin V (PcV), amoxicillin (AMX) and erythromycin (ERY) ( Table 1). Over the period, the proportion of redeemed PcV prescriptions increased from 41% in 2008 to 48% in 2017, the total proportion of AMX and ERY decreased from 41% to 31%.
Most prescriptions were dispensed in winter months (December, January, and February), and was lowest in summer months (June, July and partly August). We further visualized the monthly proportion of children redeeming an antibacterial prescription in the first three years of life, stratified for birth season (Fig. 4). The snowflake shows when the respective birth seasons experience winter (defined as January). The figure  shows a cyclic order for all birth seasons, with a peak when the birth season experiences winter. For example, children born in spring (March-May) will experience winter in the age 8-10 months, 20-22 months, and 32-34 months, coinciding with the peak of proportion of redeemed prescriptions in all three periods. The same is observed for the other birth seasons.

Multiple linear regression
Our multiple linear regression analysis included 523 267 records to 227 577 individuals (53% boys). We excluded 143 557 prescription records to those born 2005-2007, 46 861 records to those receiving more than 10 prescriptions and 577 records with missing county (Fig. 1). Table 2 shows the results of the multiple regression analysis, with age (in days) at first dispensed prescription as the outcome variable. The intersection in the regression analysis was 534 days for first dispensed prescription, representing the reference category, a girl born in fall of 2008 in one of the reference counties. The birth seasons spring, summer and winter, birth year 2009-2012, gender and dummy variables for counties were all significant covariates. Children born in autumn were significantly younger when redeeming their first prescription of an antibacterial agent, followed by summer, being 4 days older on average, winter (16 days older) and spring (22 days older). Boys had their first dispensed prescription 26 days younger compared to girls.

DISCUSSION
In this nationwide drug utilization study, we describe dispense rate, periodic prevalence, and age at first dispensed prescription in children aged 0-2 years in Norway. Overall, we found a reduction in dispense rate for the years 2008-2017 for both girls and boys. The reduction was most pronounced in 2-year-olds. This trend is positive considering the goal of the national strategy of a 30% reduction from 2012 to the end of 2020 (11). The one-year periodic prevalence has also fallen over the period, but the decrease in dispense rate is more pronounced than the change in one-year prevalence, suggesting that the fall in proportion of users has changed less than the number of prescriptions per child. Almost half the children received only one prescription during their first three years of life, suggesting a restrictive use in this age group. We found a higher dispense rate of antibacterials for boys in all age groups, although the sex difference decreased for 2year-olds. Some studies suggest that this may be due to sex differences such as higher incidence of acute otitis media in boys (14,15).
Since 2008 we have seen several events that could affect the dispense rate. For example, updates of national guidelines for use of antibacterials in primary health care (16)(17)(18), the KUPP campaign (Norwegian Prescription Peer Academic Detailing intervention) (19,20), ENORM (Educational intervention in NORwegian Municipalities for antibiotic treatment in line with guidelines) (21), and the publication of the national strategy against antibiotic resistance (11). Another Norwegian study showed a reduced relative risk for both acute otitis media and lower respiratory tract infections after the introduction of the 13-valent pneumococcal conjugate vaccine in the childhood vaccination program (22,23).
Our study also shows that around 80% of all antibacterials dispensed were PcV, AMX and ERY (Table  1), all typical RTI antibacterials. This corresponds with RTIs being the single most common reason for children visiting a physician (1). We also observed an incline in use of PcV in line with quality indicators from the Norwegian Directorate of Health and current guidelines (18,24). However, the proportion of PcV is still lower than for Swedish children aged 0-6 years in 2017 (25).
The dispense rate has fallen since 2012, except for a small rise in 2014 for both boys and girls, 1-and 2-yearolds. This is in accordance with another Norwegian study using similar records from NorPD (26). The dispense rates in our analysis are low compared to similar studies from five European countries. A study from 2014 included data for 2008 from Denmark, Italy, Germany, Netherlands, and United Kingdom. In 2008 the dispense rate for Italian children 0-4 years was 1393 per 1000 person years, compared to the Netherlands having the lowest rate at 523 per 1000 (27). The rate in Norway were comparable to those in the Netherlands in 2008, where children 0-2 years redeemed 481 prescriptions per 1000 individuals.
Germany had a noticeably higher dispense rate compared to Denmark, Netherlands and Norway in 2008, but has since showed a reduction, particularly among children aged 0-5 years (28). Dispense rate was reduced from 2010 to 2018 in 0-1-year-olds from 630 to 320 per 1000, almost a 50% decrease. We observed around 34% decrease for 2-year-old boys and girls from 2008 to 2017, and a 25% decrease for 1-year-old boys and 21% decrease for 1-year-old girls from 2008 to 2016. The starting level in 2010 for German children were noticeably higher than Norwegian dispense rates, possibly allowing a greater relative reduction in rates. For example, in Germany children aged 0-1 year redeemed 630 prescriptions per 1000 persons, and 1212 per 1000 persons for children aged 2-5 years. The total number for dispensed prescriptions for children 0-2 years in Norway was in the same period 502 per 1000 individuals.
A more recent study from Hungary, Norway and Portugal identified antibacterial use as packages/child/ year in 2014 with children age 0-19, stratified by age groups. This study identifies the youngest children (0-4 years) as the most frequent users of antibacterials, with respectively 2.2, 0.4 and 1.0 packages/child/year in Hungary, Norway and Portugal (29). The results from Norway are on par with our results, even though we are using different age stratification and proxy-measure for antibacterial use. This study misses the large variation for the age group 0-4 years that we have identified in our study.
Between 2008 and 2010 there were large fluctuations in dispensing rate (Fig. 2). There is a drop in 2009, and it has been speculated if this is associated to the H1N1swine flu pandemic affecting Norway in the fall 2009 and improved hygiene measures during this period. This coincides with reduced contact rate for RTIs at the GPs and emergency rooms during the pandemic (30). A causal connection is still not clear.
We calculated age with a monthly fictitious birth date (set at the 15 th ). Several other studies on antibacterial use in children do not report method of age calculation. For example, three Norwegian studies have used data from NorPD to calculate age specific periodic prevalence (2,31,32). All identified the highest one-year periodic prevalence in 2-year-olds, and higher in boys than girls. These findings differ from ours, identifying 1-year-olds with the highest dispense rate and prevalence. Only one study gives information on definition of age, where year of dispensed prescription minus birth year is used. This has a margin of error of almost a year, whilst our method has a margin of error of maximum 16 days. Danish researchers used similar register data for the period 2000-2012, where age was calculated like our method. The results coincided with our results, identifying 1-year-olds with the highest dispense rate, followed by 2-year-olds, and the same for periodic prevalence (5). Calculating age with a low margin of error adds information value and indicates that aggregating age in larger age groups leads to substantial loss of information. For small children it can be of value to define age more specific than using year of birth, given that dispense rate and prevalence changes rapidly over a short time.
Our results show that a significantly higher number of prescriptions are dispensed in the winter months. Large variations in seasonality for use of antibiotics increasingly occur amongst those with high consumption (33), and possibly suggests suboptimal treatment. Fig. 4 also shows this trend. Winter season seems to be a predictor for if and when a child receives antibacterial treatment. All birth seasons show cyclic fluctuations with a peak when the birth season experience winter.
In the multiple linear regression analysis, we found an association between birth season and average age in days at first dispensed prescription. Our hypothesis was that children born in the fall would on average be younger when receiving their first dispensed prescription. These assumptions were congruent with the estimates from our analysis. However, the difference was moderate, but smallest for children born in summer, only being on average 3-4 days older at first dispensed prescription. There was a larger difference in average age for sex, girls being 26 days older on average on the first dispensed prescription. The results match that boys have a higher dispense rate and one-year prevalence generally at 0-2 years old (Figs. 2,3).
Kinlaw et al. used a Kaplan-Meier analysis to estimate the risk for receiving an antibacterial agent in the first year of life in Danish children (6). The authors reported a risk dependent partly on age. These results are comparable to our results (Fig. 4). Due to some differences in definition, we saw the highest proportion in dispensed prescriptions amongst children born in summer at 6 months old, but the peak for dispensed antibacterials was found to be February in Kinlaw's study, and January in ours.
It is difficult to decide whether differences in age at first dispensed prescription is of clinical relevance, and this should be explored with other study designs with access to clinical data and microbiome data. Even though all included covariates were significant in predicting the outcome, it is important to acknowledge the complexity underlying prescribing of antibacterials in young children. None of the included variables in our regression analysis is sufficient nor necessary factors for prescribing antibacterials in children.

Strengths and limitations
Data in NorPD are considered valid as law requires registration, it minimizes selection bias, data is unaffected by recall-or information bias, and gives almost complete data. We also had data from a long period in time, which allowed us to describe temporal trends in use. We have only included records with a valid patient ID. This was necessary to follow individuals over time and explore the number of prescriptions dispensed in their first three years of life, and to explore the effect of birth season on age at first dispensed prescription. Excluding those without ID may imply some selection bias, as the proportion of records without ID was highest early in the study period, and most noticeable for the youngest children (0-year-olds). This may contribute to some underestimation in dispense rate and one-year prevalence. Records without an ID normally represents those without a national identity number, such as new-borns, tourists or immigrants.
To avoid overestimating dispense rate, we aggregated prescriptions to individuals having received two or more prescriptions on the same day on the same antibacterial agent. Possibly, we should also have investigated change in therapy 1-3 days after the first dispense. A recent Norwegian study found that 2-yearold children most frequently received a new prescription for liquid (2.7%) and solid antibacterials (10.7%) (34), suggesting that dispense rate may still be overestimated.
Another weakness is the method of data extraction. We measured outcome for the period 2008-2017, and this includes children born in 2005-2014. This means that our newest data unfortunately only contains data for 2-year-olds. We chose this approach to be able to observe each child for a full three years of life. Consequently, individuals being born in 2005-2007 had to be excluded in the multiple regression, as we could not capture the outcome variable with certainty.
Another weakness to consider with register data is the lack of information on use in hospitals, or supply directly from prescriber. Possibly, direct supply from prescriber is more frequent in places with long distances to pharmacies. On a national level, it is unlikely that this would affect our estimates. We decided to aggregate counties into dummy variables in our multiple regression model, this leads to loss of information, but reduces the complexity of the analysis. It does not account for those who moved county or abroad at a later time. Geographical variation is most likely more complex than residence county can capture. A Norwegian study showed that antibacterial dispense declined with increasing latitude along the South-North axis, and was contingent on municipality population size (35).
A study from the Norwegian Neonatal Network assessed early sepsis and use of systemic antibacterials in infants. The authors reported that 6% of term born infants was admitted to hospital during their first week of life, and 39% of these received intravenous antibacterials, equivalent to 23 per 1000 term born infants in Norway (36). This suggest a slight overestimation of age at first dispensed prescription in our multiple linear regression, and may possibly give a small selection bias, because of the exclusion of the youngest and sickest infants being treated in hospital.

CONCLUSION
Our study identified an association between birth seasons and subsequently dispensed antibacterials. Winter as a dispensing season was a predictor for if and when a child received an antibacterial drug. Children born in autumn collected their first prescription at a younger age than the other birth seasons. Further studies should include clinical variables and microbiome data to explain the variance in time to first dispensed prescription and possible consequences of early use of antibacterials.
Children aged 0-2 years are among the most frequent users of antibacterials in Norway, and both dispense rate and prevalence has fallen over the period. 1-yearolds had the highest dispense rate and prevalence. We have identified large variation in age-specific rates, showing that aggregating the youngest children into wider age groups lead to loss of information.