ISEE 2014 Microbiome Session Seattle, WA Antibiotic Use and Childhood Body Mass Index Trajectories Brian S. Schwartz, MD, MS Co-authors: Jonathan Pollak, Lisa Bailey-Davis, Claudia Nau, Annemarie Hirsch, Thomas A. Glass, Karen Bandeen-Roche August 26, 2014
Overview 1. Using electronic health records for (environmental) epidemiologic research 2. The Geisinger Health System and its electronic health record 3. Study design, study data, antibiotic exposure assessment, data analysis 4. Results, limitations, conclusions 5. Questions
The Geisinger Clinic 40+ community practice clinics and 4+ hospitals 450,000+ primary care patients representative of the general population in the region (PA) Need not have Geisinger insurance to use health system; 35%+ of encounters on medical assistance Electronic health record (EHR) since 2001 Across a large, varied geography (40+ counties) EHR can provide: demographics, encounters (OPT, IPT, ED), physician orders, laboratories, procedures, medications, vitals, ICD-9 codes, health insurance, problem list, clinical notes, dates
Downstream Consequences of Antibiotic Use When antibiotics are given to animals They gain weight They carry antibiotic-resistant bacteria Humans get MRSA and skin and soft tissue (SST) infections Accounting for animal equivalent units, manure produced, manure exported, acreage of crop fields, and distance from residence to farms and crop fields, persons in highest quartile of swine crop field manure exposure had increased odds (95% CI) of MRSA (1.38, 1.13-1.69) and SST (1.37, 1.18-1.60) infections (Casey J, JAMA Int Med 2013; Casey J, EHP 2014) When antibiotics are given to humans Resistant strains are selected for; patients have large increased risk of MRSA too (Casey J, Epidemiol Infect 2013) Do they gain weight? CA-MRSA, Medscape.com 4
What is Known About Antibiotics and Childhood Weight Gain? Recent review of human studies that evaluated associations between antibiotics and weight change (Million et al. Clin Microbiol Inf 2013) 17 published papers and 4 published abstracts Median sample size = 113; only 4 with > 300 people Cystic fibrosis (N = 6 studies); H. pylori infection (N = 2); poor nutrition or feeding intolerance (N = 5); 12 were randomized trials; rest highly varied: endocarditis, early life infections, prematurity, prophylaxis after measles Effect of antibiotic or improved health after resolution of infection? 16 evaluated only one antibiotic class, most commonly tetracyclines or macrolides Most reported significant weight gain with antibiotic use 5
3 to 6 years Key Quintiles of mean BMI in community 1 st 2 nd 3 rd 4 th 5 th Johns Hopkins Systems-oriented Childhood Obesity Center (U54 HD070725) Project 1 163,820 children 3-18 years in 1,289 communities Health data from 2001-12, average of 3 annual BMIs per child 7 to 12 years 910 communities with 10+ children Mean BMI z-scores by community Spatial variation in BMI; transition to higher average BMI with age New primary data collection from kids, parents, communities in Phase 2 13 to 18 years 6 Schwartz BS, et al. ADHD, stimulants, and BMI. Pediatrics 2014.
Antibiotic Exposure Assessment Reversible and Statistical Modeling Antibiotic order in 365 to 1 day window before BMI i, yes vs. no Evaluated two other windows (2.5 to 1.5 y, 1.5 to 0.5 y) Persistent Cumulative count of antibiotic orders up to BMI i, divided into groups (0, 1, 2-3, 4-6, 7+, with 4 dummy variables) Progressive Cumulative count of antibiotic orders up to BMI i-1, divided into groups (0, 1, 2-3, 4-6, 7+, with 4 dummy variables) Each with cross-products with age, age 2, and age 3 Random intercept for kid; random effects for age, age 2 ; group effect to account for non-stationarity of variances Adjusted for sex, race/ethnicity, medical assistance, each with age cross-products; & by antibiotic classes 7
Schematic of reversible, persistent, and progressive effects Progressive 25 BMI, kg/m 2 Lifetime cumulative count to BMI i-1 or BMI i Reversible Persistent Average BMI trajectory without antibiotics All three are time-varying 15 Any order in 365 to 1d window before BMI i 3 Age, years 18 8
Study Variable All Children Children Under Observation in 1.0y to 1d before at least one BMI Number 163,820 142,824 Age at first BMI, years, mean (SD), range 8.9 (5.0), 3-18 8.5 (5.0), 3-18 Race/ethnicity, percent White Black Hispanic Other or Missing 91.3 4.7 1.1 2.9 92.5 4.5 0.9 2.2 Sex, female, percent 49.7 49.8 Duration between first and last BMI, years, mean (SD), range 2.9 (3.1), 0-11.1 3.3 (3.1), 0-11.1 Number of annual BMIs available for analysis, mean (SD), range 3.2 (2.4), 1-13 3.5 (2.4), 1-13 BMI, first, kg/m 2, mean (SD), range 19.7 (5.3), 10.5-54.5 19.5 (5.2), 10.5-54.5 BMI, last, kg/m 2, mean (SD), range 21.5 (6.0), 10.7-55.6 21.6 (6.0), 10.7-55.6 No. of antibiotic orders, lifetime, age 0-18 years, mean, median, SD, range 4.0, 2, 6.1, 0-123 4.6 (6.3), 0-123 Lifetime antibiotic count to last BMI, % Zero One Two to three Four to six Seven or more 40.8 13.3 15.2 12.5 18.3 32.0 15.2 17.4 14.4 20.9 9
Evidence of a REVERSIBLE EFFECT that was STRONGER at younger ages (all antibiotic classes) Comparing Different Time Windows Variable Model 1a Beta (SE) Model 1b Beta (SE) Model 1c Beta (SE) Model 1d Beta (SE) Group and sample size UO 116,769 UO 116,769 UO 116,769 UO 142,824 WINDOW 1.0 y to 1 d 0.056 (0.009)*** 0.054 (0.008)*** Abx_PRE_period x age 0.013 (0.003)*** 0.016 (0.002)*** Abx_PRE_period x age 2-0.001 (0.0003)*** -0.001 (0.0003)*** Abx_PRE_period x age 3-0.0002 (0.00007)*** -0.0003 (0.00006)*** WINDOW 1.5 to 0.5 y 0.057 (0.009)*** Abx_PRE_period x age 0.007 (0.003)* Abx_PRE_period x age 2-0.0004 (0.0003) Abx_PRE_period x age 3-0.0001 (0.00006)* WINDOW 2.5 to 1.5 y 0.056 (0.009)*** Abx_PRE_period x age 0.005 (0.003)* Abx_PRE_period x age 2-0.0009 (0.0003)** Abx_PRE_period x age 3-0.0002 (0.00007)** Font this color = age cross-product (on this and following slides) UO = under observation in relevant period *** p < 0.001; ** 0.001 p < 0.01; * 0.01 p < 0.05 10
Evidence of a PERSISTENT EFFECT that was STRONGER with increasing age (all classes) Variable Beta (SE) Group of children and sample size UO 142,824 Antibiotic order 1.0 y to 1 d before BMI i, yes vs. no 0.044 (0.008)*** Abx_PRE_period x age 0.015 (0.003)*** Abx_PRE_period x age 2-0.001 (0.0003)*** Abx_PRE_period x age 3-0.0004 (0.00006)*** Lifetime antibiotic orders to BMI i, count One 0.020 (0.011) One x age 0.0009 (0.002) Two or three 0.048 (0.012)*** Two/three x age 0.005 (0.002)* Four to six 0.084 (0.014)*** Four-six x age 0.009 (0.003)*** Seven or more 0.146 (0.017)*** Seven/more x age 0.013 (0.003)*** UO = under observation in relevant period *** p < 0.001; ** 0.001 p < 0.01; * 0.01 p < 0.05 Cross-products with age 2 and age 3, NS 11
Evidence of a PROGRESSIVE EFFECT that did not vary by age Variable Beta (SE) Group of children and sample size UO 79,752 Antibiotic order 1.0 y to 1 d before BMI i, yes vs. no 0.044 (0.010)*** Abx_PRE_period x age 0.013 (0.003)*** Abx_PRE_period x age 2-0.001 (0.0004)*** Abx_PRE_period x age 3-0.0004 (0.00009)*** Lifetime antibiotic orders to BMI i, count One 0.020 (0.014) One x age 0.005 (0.003) Two or three 0.051 (0.017)** Two/three x age 0.011 (0.003)*** Four to six 0.082 (0.020)*** Four-six x age 0.017 (0.003)*** Seven or more 0.112 (0.025)*** Seven/more x age 0.024 (0.004)*** Lifetime antibiotic orders to BMI i-1 [LAG], count One 0.025 (0.011)* Two or three 0.021 (0.012) Four to six 0.047 (0.015)** Seven or more 0.090 (0.018)*** 12
Average (95% CI) POUNDS of additional weight at age 15y compared to children who did not receive antibiotics Boys & girls with the average height in our data All abx, all observed* BMIs All abx, 6 observed BMIs Macrolides only, all observed BMIs Macrolides only, 6 observed BMIs Model # children 142,824 26,228 142,824 26,228 PERSISTENT 1.6 (1.3, 1.9) 2.4 (1.8, 3.0) 2.5 (1.9, 3.2) 3.3 (2.2, 4.4) # children 79,752 26,228 79,752 26,228 PROGRESSIVE 2.2 (1.8, 2.6) 3.0 (2.3, 3.6) 2.7 (1.9, 3.4) 3.1 (1.9, 4.3) Abbreviations: Abx = antibiotics; BMI = body mass index * Observed indicates patient record documents was under observation in relevant window for each BMI 13
Strengths and Limitations Strengths Large sample size, longitudinal, several antibiotic dose metrics, effect modification by age, by antibiotic classes, among mainly children without chronic diseases, full age range Limitations Cannot exclude possibility that children received antibiotics outside of health system Systematically undercounted cumulative dose measure as it gets larger Cannot verify compliance with medication orders All of these would likely bias associations towards the null 14
Conclusions In large, longitudinal study of general population representative sample, children given antibiotics for a variety of indications evidenced gains in weight Complex associations: reversible effects that were stronger at younger ages; persistent effects that got stronger with age; progressive effects that did not vary by age (the latter two a function of cumulative dose) Magnitude seems to be important; comparable in magnitude to weight loss interventions Effects may be larger for certain antibiotics classes Future studies should evaluate whether and how these effects are due to microbiome changes Evidence that should add to calls for more judicious use of antibiotics in clinical practice 15
Acknowledgements: Project 1: Dynamics of childhood obesity in Pennsylvania from community to epigenetics Johns Hopkins Project Team: Thomas A. Glass (Epi), Project leader Brian S. Schwartz (EHS), Project co-leader Karen Bandeen-Roche (Bio), Biostatistician Joseph Bressler (EHS), epigenetics Ann Liu (EHS), Research associate Tak Igusa (Engineering), Systems scientist Claudia Nau (IH), Post-doctoral fellow, trainee Mehdi Jalalpour (Eng), Pre-doctoral fellow, trainee Jonathan Pollack (EHS), Data analyst Geisinger Health System Team: Annemarie Hirsch, subcontract PI Lisa Bailey-Davis, co-investigator Dione Mercer, Project coordinator Sy Landau, Research assistant Joseph DeWalle, GIS analyst Statement of funding support: This research was funded in part through a cooperative agreement from the National Institutes of Health (U54 HD070725). Contributing partners include the Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD) and the Office of Behavioral and Social Sciences Research (OBSSR).
Thank you for listening Feel free to contact me: bschwar1@jhu.edu