Disease-modifying therapies and features linked to treatment response in type 1 diabetes prevention: a systematic review

Abstract

Background: Type 1 diabetes (T1D) results from immune-mediated destruction of insulin-producing beta cells. Prevention efforts have focused on immune modulation and supporting beta cell health before or around diagnosis; however, heterogeneity in disease progression and therapy response has limited translation to clinical practice, highlighting the need for precision medicine approaches to T1D disease modification.

Methods: To understand the state of knowledge in this area, we performed a systematic review of randomized-controlled trials with ≥50 participants cataloged in PubMed or Embase from the past 25 years testing T1D disease-modifying therapies and/or identifying features linked to treatment response, analyzing bias using a Cochrane-risk-of-bias instrument.

Results: We identify and summarize 75 manuscripts, 15 describing 11 prevention trials for individuals with increased risk for T1D, and 60 describing treatments aimed at preventing beta cell loss at disease onset. Seventeen interventions, mostly immunotherapies, show benefit compared to placebo (only two prior to T1D onset). Fifty-seven studies employ precision analyses to assess features linked to treatment response. Age, beta cell function measures, and immune phenotypes are most frequently tested. However, analyses are typically not prespecified, with inconsistent methods of reporting, and tend to report positive findings.

Conclusions: While the quality of prevention and intervention trials is overall high, the low quality of precision analyses makes it difficult to draw meaningful conclusions that inform clinical practice. To facilitate precision medicine approaches to T1D prevention, considerations for future precision studies include the incorporation of uniform outcome measures, reproducible biomarkers, and prespecified, fully powered precision analyses into future trial design.

Fig. 1. PRISMA flow diagram.

Flowchart displaying studies screened and excluded as part of abstract screening, then via full text review/eligibility assessment. 75 total papers were included in the extraction. AUC area under the curve; T1D Type 1 diabetes. This image was generated using Biorender.

Fig. 2. Relative effect of prevention therapies in individuals at risk for T1D.

Forest plot showing hazard ratio with 95% confidence intervals for primary prevention studies in genetically at-risk individuals and secondary prevention studies in individuals with elevated islet autoantibody titers. Primary prevention studies are divided by outcome—either time to islet autoantibody positivity or time to diabetes. All secondary prevention studies used time to diabetes as a primary outcome. DPT-1 Diabetes Prevention Trial Type 1 ; GAD glutamic acid decarboxylase.

Fig. 3. Precision analyses focused on treatment response were mostly part of primary trial papers, tended to be post hoc, and were biased toward positive findings.

a Stacked bar graphs showing relative frequencies and percentages of papers with precision analyses that were defined as prespecified, post hoc, or included both prespecified and post hoc analyses in the manuscript text. b Stacked bar graph displaying relative frequencies and percentages of papers reporting positive findings related to associations with treatment effects. c For papers that listed sample sizes of subgroups tested for differential treatment effects (only 53% of all papers with precision analyses), the smallest sample size reported is displayed, with mean and SEM indicated. F/u follow-up; n = 9 for precision papers; n = 16 for primary trials; n = 5 for f/u papers.

Fig. 4. Precision analyses tested many features, most commonly age and beta cell function, infrequently corrected for multiple comparisons, and typically tested for differential impacts on a C-peptide-based measure.

a Total number of features tested for association with each treatment response, with mean and SEM indicated, for all papers with precision analyses. b Stacked bar graph showing relative frequencies and percentages of papers that did or did not correct for multiple comparisons. c Frequencies of individual features tested for associations with treatment response. d Frequencies of outcomes utilized to assess for the presence of any features associated with differential treatment response. The C-peptide measure category was inclusive of any measure of beta cell function, including mixed meal area under the curve, stimulated C-peptide values, fasting C-peptide values, etc. F/u follow-up, fx function, Hba1c hemoglobin A1c, Aab autoantibody, HLA human leukocyte antigen, BMI body mass index, T1D type 1 diabetes, AGT abnormal glucose tolerance, CRP C-reactive Protein, DPTRS diabetes prevention trial-type 1 risk score, DKA diabetes ketoacidosis, Dx diagnosis.

Fig. 5. Risk of bias assessments for each paper category. Bias was assessed using Covidence’s Cochrane risk of bias tool.

For sequence generation, allocation concealment, and blinding categories, raters had the option of selecting high quality (green), low quality (orange), not reported (red), or that a decision could not be made because of primary trial was referenced in methods (yellow). For incomplete outcome data, raters only had the option to choose high quality/data provided (green) or low quality/data not provided (red). For selective reporting, raters had the option to select high-quality/primary endpoint predefined (green), low-quality/primary endpoint not defined (orange), or low-quality/not reported (red). For other sources of bias, raters had the option to select high quality/none (green), low quality/bias present but identified and considered (orange), or low quality/obvious bias present and not addressed (red). Data are shown as absolute frequencies.