Type 1 diabetes (T1D) is a major disease of childhood, affecting over 1.5 million children worldwide, that is characterised by the progressive autoimmune destruction of insulin-producing β cells in the pancreatic Islets of Langerhans (1). T1D can result in acute metabolic disturbances such as hypoglycemia and ketoacidosis, and chronic cardiovascular, renal, retinal and neuronal complications. Along with the demands of daily management, these outcomes contribute to psychological stress and impaired quality of life. Additional to its burden on individuals and families, T1D imposes a substantial cost on the healthcare system (2). The incidence of T1D has been rising, due to changing environmental factors that increase penetrance of risk genes (3). While recent interventions with immune agents have slowed loss of β-cell function (4), long-term remission remains elusive. To alleviate the burden of disease it is important to focus on disease prevention and identify modifiable environmental risk factors.

Alterations in the gut microbiome have been described in T1D (5), including a decrease in bacterial taxonomic diversity, also seen after oral antibiotic administration. The connection with the gut microbiome underpins the proposition that early-life antibiotic exposure may be a potential environmental risk factor for T1D. This is supported by studies in the non-obese diabetic (NOD) mouse, a model of spontaneous autoimmune diabetes, which demonstrate that early-life antibiotic exposure influences the development of diabetes (6–8). Administration of vancomycin in drinking water from birth to weaning decreased the incidence of diabetes (6). Fecal bacterial diversity was decreased, leaving a single dominant species, Akkermansia muciniphila, and small intestinal proinflammatory CD4⁺ T cells were increased (6). In contrast, continuous exposure to vancomycin or neomycin from just before birth (to mothers) increased the incidence of diabetes (7). Again, ileal and colonic bacterial diversity were decreased with an increase in the abundance of Akkermansia muciniphila, as well as Enterobacteriaceae, and small intestinal proinflammatory CD4⁺ T cells were increased (7). In another study, the macrolide antibiotic tylosin tartrate, administered before and just after weaning, accelerated the onset of diabetes, especially in lower incidence males; the β-lactam, penicillin V, given at sub-therapeutic doses to mothers in pregnancy and offspring up to three months of age had no effect (8). These contrasting results suggest that the timing of antibiotic exposure influences the risk of T1D, pre-weaning being protective, in contrast to post-weaning.

Many cohort studies (9–28) have investigated an association between early-life antibiotic exposure and T1D risk but they have not been subjected to a comprehensive systematic review and meta-analysis. Here, we synthesise existing evidence in a systematic review and meta-analysis of the association between early-life antibiotic exposure and T1D. Because studies in the NOD mouse model suggest that timing of antibiotic exposure may be critical, the review encompassed four exposure windows: preconception (12 months before conception), prenatal (during pregnancy), neonatal (first two weeks after birth) and postnatal (0–6 months, 0–12 months, 0–24 months after birth). The following were of interest: any antibiotic, specific antibiotic classes (macrolides, sulfonamide/trimethoprim, tetracyclines) and antibiotic spectra (broad or narrow), as well as the number of antibiotic courses. Determining if antibiotic exposure in early life influences T1D risk could guide research on antibiotic-microbiota interactions in T1D and inform clinical decision-making about antibiotic use.

This meta-analysis found no evidence that exposure to antibiotics in pregnancy or early infant life was associated with the development of T1D. However, the odds of T1D development were increased by 23%, 34% and 26% after preconception exposure to macrolides, sulfonamides/trimethoprim and tetracyclines, respectively. In contrast, preconception exposure to any antibiotic (class agnostic) or cephalosporins was not associated with significantly altered odds. Similarly, phenoxymethylpenicillin and quinolone antibiotics showed no significant association, though their wide confidence intervals suggest substantial uncertainty. Across the two preconception studies, only 140 cases were exposed to phenoxymethylpenicillin and 77 to quinolones, limiting statistical power. A true association may exist but not be detected due to the small sample size.

The individual preconception studies provide context to these significant pooled results. In the Hakola 2024 study (28), preconception macrolide use was associated with both increased odds of T1D and increased rate, strengthening confidence in this association. In contrast, sulfonamides/trimethoprim and tetracyclines had significant ORs but non-significant adjusted HRs, weakening the certainty of their association with T1D. This discrepancy may stem from limited statistical power in the Hakola study, in which only 68 cases were exposed to sulfonamides/trimethoprim and 236 to tetracyclines, compared to 270 for macrolides. The meta-analysis increased the sample size from 68 to 79 for sulfonamides/trimethoprim and 236 to 275 for tetracyclines but lacks adjustment for confounders beyond case matching for age and sex, unlike the adjusted HRs in Hakola (28). Because the adjusted HRs in the Hakola study did not identify a significant association for preconception exposure to sulfonamides/trimethoprim or tetracyclines with T1D, larger, well-powered studies with robust confounder adjustment will be needed to clarify these relationships.

Given higher confidence of the association between preconception exposure to macrolides and T1D, it is relevant to consider possible mechanisms. Macrolides have rapid and profound effects on intestinal microbiome diversity and composition that may persist for many months, especially when administered early in life. The more consistent effects include a decrease in alpha diversity and in the beneficial taxa Bifidobacteria, Lactobacilli and Akkermansia muciniphila spp that produce anti-inflammatory short chain fatty acids (29–31). These ‘dysbiotic’ alterations are similar to those reported in the fecal microbiome of children with T1D (32). In addition, macrolides have anti-inflammatory and immunomodulatory properties independent of their bacteriostatic effects (33, 34). That macrolides could have anti-inflammatory effects on the maternal gut microbiome to increase the risk for T1D in offspring appears counterintuitive. However, it is widely accepted that immune activation is necessary to induce maturation of immune regulatory pathways that protect against autoimmune disease. In the non-obese diabetic (NOD) mouse model of T1D, immune activation by mycobacterial adjuvant (35) or Toll-like receptor agonists (36) promotes immune regulation that protects against diabetes. In humans, females with T1D have a significantly lower risk of having a child with T1D than males with T1D (37). One explanation for this ‘maternal protection’ may be promotion of immune regulation by the pro-inflammatory state of the maternal gut microbiome in pregnancy in women with T1D (38). We suggest therefore that macrolides could offset protection against autoimmunity afforded by specific gut microbes at an early stage of development. Different macrolides have common and type-specific effects on the gut microbiome. However, the preconception studies in this review did not categorise macrolides by type (e.g. azithromycin, clarithromycin, erythromycin, roxithromycin) or time of exposure within the 0–12 months before conception.

Despite the inclusion of large, cross-country cohorts and adherence to strict eligibility criteria, several caveats limit the applicability of our findings. First, antibiotic exposure characteristics were often incomplete. Exposure windows were broad. For example, the 0-12-month preconception period may have included periods that are irrelevant to T1D risk. Broad exposure windows may dilute meaningful signals and contribute to statistically non-significant findings, despite the presence of a true effect in a critical subperiod. Although the effects of macrolides on the gut microbiome may last many months, it is plausible that a narrower window before conception would be biologically more relevant. Second, the measurement of antibiotic exposure varied. The Hakola study (28) relied on prescription reimbursement records, which do not consider hospital- administered antibiotics. Similarly, the Kilkkinen study (12) used a nationwide drug prescription register, which does not consider actual consumption or unprescribed use. Nevertheless, both studies found significant positive associations between preconception macrolide exposure and T1D. Other studies used parental questionnaires, which are prone to recall bias. Both register- and questionnaire-based approaches risk exposure misclassification, potentially resulting in antibiotic-exposed individuals being included in the control group or unexposed individuals in the case group. Misclassification could shift estimates towards the null. Third, race/ethnicity could impact the interpretation and generalisability of the findings, especially as the risk of T1D is higher in individuals of Northern European descent who carry specific human leukocyte antigen (HLA) haplotypes (1). However, analysis of the influence of race/ethnicity was not feasible because minimal data were reported in the included studies. Also, the two preconception studies were both Finnish register-based, with potentially overlapping samples in a high-risk T1D population. Future studies from other countries and more diverse populations would strengthen validity for this exposure window. Finally, studies were observational in design and cannot address causality. In addition, as in many studies, potential publication bias, where null findings are less likely to be published and hence included in a review, may inaccurately shift results towards false significance. Taken together, these caveats indicate that the evidence can be accepted with only moderate certainty and should be strengthened by further studies.

Other limitations are related to the review process. First, only four literature databases were searched (Medline, Embase, Web of Science Core Collection, and Scopus). While these are major biomedical resources, other repositories (e.g., CINAHL, Global Health) may contain relevant studies not identified. Second, data extraction was performed by a single reviewer, increasing the risk of transcription errors. Third, no formal risk of bias assessment (e.g. ROBINS-I) was conducted beyond applying predefined inclusion and exclusion criteria.

In conclusion, the significant association between preconception exposure to macrolides, sulfonamides/trimethoprim and tetracyclines and the development of T1D suggests that exposure to these antibiotics in the preconception period may be a modifiable risk factor for T1D. However, given the moderate quality of the evidence, further well-powered, confirmatory studies are needed to inform clinical decision-making and public health policy. These would ideally explore narrower preconception exposure periods, e.g., 0–3 and 3–6 months, and document antibiotic subclasses. Prescription data, pharmacy dispensing records and self-reported adherence logs should be accessed where possible to minimize misclassification of cases and controls. Additionally, analysis of the gut metagenome of antibiotic-exposed and -unexposed women may provide insights into how disruption of maternal microbial communities could contribute to T1D risk in the offspring. The current review and meta-analysis can readily be adjusted to account for the results of future studies.