BACKGROUND

Research has demonstrated associations between antibiotic use in early life and higher weight in later childhood. Lingering questions regarding this association include potential heterogeneity of treatment effects: the timing, type, and number of antibiotic exposures in early life; and differential effects of antibiotics in subgroups, such as by sex and medical comorbidities. With electronic health record (EHR) and other health care data available for large populations, the National Patient-Centered Clinical Research Network (PCORnet) provided a unique opportunity to investigate these questions and help build the research capacity of this new network. The study was led by 3 principal investigators, and the study results include data from 36 PCORnet institutions in 10 clinical data research networks (CDRNs).

OBJECTIVES

The study aims were to examine the association between antibiotic use before 24 months of age and weight outcomes at ages 48 to <72 months (ie, approximately 5 years) and at 108 to 132 months (ie, approximately 10 years) and weight trajectories from early- to mid-childhood. To explore the association of maternal antibiotic use during pregnancy, we also incorporated maternal variables from the data of 7 participating institutions. The study had a qualitative aim to gauge opinions from parents and providers about their impressions of the potential relationship between antibiotics and childhood obesity, as well as an evaluation aim intended to identify accomplishments and needed improvements in the development of PCORnet.

METHODS

For all analytic aims, the study used PCORnet's infrastructure and common data model (CDM) to capture all relevant data from 2009 to 2016. The CDM allows each participating institution to store its data in a standardized manner, facilitating the combination of data across sites. Study cohort inclusion criteria were the availability of same-day height and weight measurements at 0 to <a12, 12 to <30, and >24 months of age. Using deidentified individual-level data, we determined the association between antibiotic use at <24 months of age with the primary outcomes of (1) body mass index (BMI) -score and overweight or obesity prevalence at 48 to <72 months (5 years) and 108 to 132 months (10 years) of age; and (2) weight trajectories from 24 months of age on. In 7 institutions, we were able to link data from mothers and children. For these sites, we assessed whether maternal antibiotics during pregnancy were associated with child BMI -score and overweight or obesity prevalence at 48 to <72 months of age. For analyses of BMI -score and overweight or obesity prevalence, we used mixed-effects regression models stratified by complex chronic condition status, accounting for clustering by network partner and adjusting for demographic and clinical factors, including sex, race, ethnicity, preterm birth, asthma, infections, corticosteroid episodes, encounters at <24 months of age, and age at outcome. We also conducted several secondary analyses, including examining differential effects on BMI -score by antibiotic spectrum (narrow and broad spectrum, with further analyses of specific classes of medications within the broad-spectrum category) and age period (<6 months, 6 to <12 months, and 12 to <24 months) of exposures. We incorporated data on maternal characteristics from the subset of institutions for which we had data available linking mothers to children. For the trajectory analyses, we used longitudinal rate regression to assess differences in the rates of weight change across exposure groups. The study's qualitative aim used parent focus groups at 4 participating institutions and telephone provider interviews using scripts developed by the study team and stakeholder advisory group. We audio-recorded and transcribed focus groups and interviews, coded themes identified during the review of transcripts, and analyzed the code lists using a content analysis approach. Because this study was one of the first observational epidemiologic studies to be conducted across PCORnet, an external evaluator completed an extensive review of it, along with the PCORnet Bariatric Study; we report some of the findings from that evaluation.

RESULTS

Among 1 792 849 children with a same-day height and weight measurement at <12 months of age, 362 550 children were eligible for the cohort for 5-year outcomes; 430 376 children were eligible for the trajectory analysis; 56 727 children were eligible for 10-year outcomes; and 53 320 mother–child pairs were eligible for the analysis of maternal antibiotics during pregnancy. Among children who received antibiotics at <24 months of age, we found that BMI -score was slightly higher at 5 years by 0.04 (95% CI, 0.03-0.05) with slightly higher odds of overweight or obesity (odds ratio, 1.05; 95% CI, 1.03-1.07) than among children who had not received antibiotics. We found modest evidence of a dose response, with more antibiotic exposure translating to higher weight; the BMI -score differences were 0.07 higher [0.06-0.08] for children exposed to ≥4 antibiotics vs children who had no antibiotics. For 5-year-old boys and girls who were at average height in this study, this difference amounted to 0.11 kg more weight if exposed to ≥4 antibiotic courses (vs 0) before 24 months of age. Broad-spectrum antibiotics were associated with slightly higher BMI scores than were narrow-spectrum antibiotics. We did not identify substantive differences by timing of exposure, sociodemographics, and whether children had chronic diseases. Effects were similar when we examined weight trajectories after 24 months and for weight outcomes at 10 years. We found no association between maternal antibiotic use during pregnancy and child weight outcomes. In the qualitative aim, we discovered that both parents and clinicians generally prioritized providing acute relief to children with an infection over the modest concern about long-term risks of weight gain.

CONCLUSIONS

Antibiotic use at <24 months was associated with a slightly higher weight in later childhood. Maternal antibiotic use during pregnancy was not associated with weight in children. These findings likely should not affect clinical decision-making regarding whether to give children antibiotics, but they may be important at the population level considering the widespread use of antibiotics. While completing this project, the study team helped create a process for the curation and use of EHR prescribing data in PCORnet.

LIMITATIONS

The PCORnet CDM focuses largely on EHR data, which includes prescribing data rather than dispensing data. In a limited number of participating institutions, our own analyses found missing data in PCORnet prescribing data compared with insurance claims and pharmacy dispensing data. In addition, at the time of this study, PCORnet did not have the ability to link patients across institutions. Thus, there was a possibility of duplicate patients in our study data set.