{
  "abstract": "Introduction Semaglutide, a glucagon-like peptide-1 receptor agonist (GLP-1RA) used to treat type 2 diabetes mellitus (T2D), has potential associations with higher rates of diabetic retinopathy (DR) complications including proliferative DR (PDR) and diabetic macular edema (DME). The purpose of this study was to determine whether an association exists between semaglutide and PDR and treatment-requiring DR/DME.Research design and methods This was a retrospective cohort study of 14 databases (six administrative claims and eight electronic health records) in the Observational Health Data Sciences and Informatics Evidence Network. Adults with T2D on semaglutide, other GLP-1RA (dulaglutide, exenatide), or non-GLP-1RA medications (empagliflozin, sitagliptin, glipizide) from 1 December 2017 to 31 December 2023 were included. The association between semaglutide and PDR or treatment-requiring DR/DME was assessed using an active-comparator cohort design comparing new users of semaglutide as second-line T2D treatment to those on other GLP-1RAs and non-GLP-1RAs. Propensity score-adjusted Cox proportional hazards models were used to estimate hazard ratios (HRs). Network-wide HR estimates were generated using a random-effects meta-analysis.Results The study included 810 390 new semaglutide users for T2D. PDR risk for semaglutide was similar to dulaglutide (HR 0.81, 95% CI 0.42 to 1.54, p=0.51), empagliflozin (HR 0.83, 95% CI 0.53 to 1.30, p=0.41) and sitagliptin (HR 0.83, 95% CI 0.45 to 1.55, p=0.57) but was lower than glipizide (HR 0.59, 95% CI 0.39 to 0.88, p=0.01). The risk for treatment-requiring DR/DME for semaglutide was similar to empagliflozin (HR 0.66, 95% CI 0.43 to 1.02, p=0.06) but lower than dulaglutide (HR 0.53, 95% CI 0.31 to 0.91, p=0.02), sitagliptin (HR 0.46, 95% CI 0.26 to 0.81, p=0.008) and glipizide (HR 0.55, 95% CI 0.33 to 0.91, p=0.02).Conclusions and relevance We did not identify increased risk for either PDR or treatment-requiring DR/DME comparing semaglutide with other GLP-1RAs or non-GLP-1RAs. Patients with T2D should still undergo close eye care follow-up, particularly when initiating new antihyperglycemic medications.",
  "authors": [
    {
      "affiliations": [
        "Wilmer Eye Institute, Johns Hopkins School of Medicine, Baltimore, Maryland, USA",
        "Biomedical Informatics and Data Science, Division of General Internal Medicine, Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD, USA"
      ],
      "name": "Cindy Xinji Cai"
    },
    {
      "affiliations": [
        "Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA"
      ],
      "name": "Akihiko Nishimura"
    },
    {
      "affiliations": [
        "Viterbi Family Department of Ophthalmology and Shiley Eye Institute, University of California San Diego, La Jolla, California, USA",
        "Division of Biomedical Informatics, Department of Medicine, University of California San Diego, La Jolla, CA, USA"
      ],
      "name": "Sally Baxter"
    },
    {
      "affiliations": [
        "National Eye Institute, National Institutes of Health, Bethesda, Maryland, USA"
      ],
      "name": "Kerry Goetz"
    },
    {
      "affiliations": [
        "National Eye Institute, National Institutes of Health, Bethesda, Maryland, USA",
        "Casey Eye Institute, Oregon Health & Science University, Portland, Oregon, USA",
        "Department of Medical Informatics and Clinical Epidemiology, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "Michelle Hribar"
    },
    {
      "affiliations": [
        "Roski Eye Institute, Keck School of Medicine, University of Southern California, Los Angeles, California, USA"
      ],
      "name": "Brian Toy"
    },
    {
      "affiliations": [
        "Department of Ophthalmology, Mayo Clinic, Rochester, Minnesota, USA"
      ],
      "name": "Andrew Barkmeier"
    },
    {
      "affiliations": [
        "Byers Eye Institute, Department of Ophthalmology, Stanford University, Palo Alto, California, USA"
      ],
      "name": "Sophia Wang"
    },
    {
      "affiliations": [
        "Department of Ophthalmology, Bascom Palmer Eye Institute, University of Miami Miller School of Medicine, Miami, Florida, USA"
      ],
      "name": "Swarup Swaminathan"
    },
    {
      "affiliations": [
        "Vanderbilt Eye Institute, Vanderbilt University Medical Center, Nashville, Tennessee, USA"
      ],
      "name": "Alexis Flowers"
    },
    {
      "affiliations": [
        "Vanderbilt Eye Institute, Vanderbilt University Medical Center, Nashville, Tennessee, USA"
      ],
      "name": "Eric Brown"
    },
    {
      "affiliations": [
        "Roski Eye Institute, Keck School of Medicine, University of Southern California, Los Angeles, California, USA"
      ],
      "name": "Benjamin Xu"
    },
    {
      "affiliations": [
        "Department of Ophthalmology, Mayo Clinic, Rochester, Minnesota, USA"
      ],
      "name": "John Chen"
    },
    {
      "affiliations": [
        "Casey Eye Institute, Oregon Health & Science University, Portland, Oregon, USA",
        "Department of Medical Informatics and Clinical Epidemiology, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "Aiyin Chen"
    },
    {
      "affiliations": [
        "Byers Eye Institute, Department of Ophthalmology, Stanford University, Palo Alto, California, USA"
      ],
      "name": "Theodore Leng"
    },
    {
      "affiliations": [
        "Department of Ophthalmology, Mass Eye and Ear, Boston, Massachusetts, USA"
      ],
      "name": "Michael Boland"
    },
    {
      "affiliations": [
        "Department of Clinical Practice, Jazan University, Jazan, Jazan, Saudi Arabia",
        "Pharmacy Practice Research Unit, Jazan University, Jazan, Saudi Arabia"
      ],
      "name": "Thamir Alshammari"
    },
    {
      "affiliations": [
        "Department of Biostatistics, University of Michigan, Ann Arbor, Michigan, USA"
      ],
      "name": "Fan Bu"
    },
    {
      "affiliations": [
        "Department of Biomedical Informatics, Columbia University, New York, New York, USA"
      ],
      "name": "Thomas Falconer"
    },
    {
      "affiliations": [
        "Biomedical Informatics and Data Science, Division of General Internal Medicine, Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD, USA"
      ],
      "name": "Benjamin Martin"
    },
    {
      "affiliations": [
        "Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA"
      ],
      "name": "Erik Westlund"
    },
    {
      "affiliations": [
        "Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA"
      ],
      "name": "Nestoras Mathioudakis"
    },
    {
      "affiliations": [
        "Institute for Informatics, Data Science and Biostatistics,Department of Medicine, Washington University in St Louis, St. Louis, Missouri, USA"
      ],
      "name": "Linying Zhang"
    },
    {
      "affiliations": [
        "Institute for Informatics, Data Science and Biostatistics,Department of Medicine, Washington University in St Louis, St. Louis, Missouri, USA"
      ],
      "name": "Ruochong Fan"
    },
    {
      "affiliations": [
        "Institute for Informatics, Data Science and Biostatistics,Department of Medicine, Washington University in St Louis, St. Louis, Missouri, USA"
      ],
      "name": "Adam Wilcox"
    },
    {
      "affiliations": [
        "Institute for Informatics, Data Science and Biostatistics,Department of Medicine, Washington University in St Louis, St. Louis, Missouri, USA"
      ],
      "name": "Albert Lai"
    },
    {
      "affiliations": [
        "Department of Internal Medicine, University of California Davis, Davis, California, USA"
      ],
      "name": "Jacqueline C Stocking"
    },
    {
      "affiliations": [
        "Vanderbilt Eye Institute, Vanderbilt University Medical Center, Nashville, Tennessee, USA"
      ],
      "name": "Yangyiran Xie"
    },
    {
      "affiliations": [
        "Vanderbilt Eye Institute, Vanderbilt University Medical Center, Nashville, Tennessee, USA"
      ],
      "name": "Lok Hin Lee"
    },
    {
      "affiliations": [
        "Department of Medical Informatics and Clinical Epidemiology, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "David Dorr"
    },
    {
      "affiliations": [
        "Oregon Clinical and Translational Research Institute, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "Izabelle Humes"
    },
    {
      "affiliations": [
        "Oregon Clinical and Translational Research Institute, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "David McCoy"
    },
    {
      "affiliations": [
        "Oregon Clinical and Translational Research Institute, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "Mohammad Adibuzzaman"
    },
    {
      "affiliations": [
        "Department of Ophthalmology and Visual Neurosciences, University of Minnesota, Minneapolis, Minnesota, USA"
      ],
      "name": "Raymond Areaux, Jr"
    },
    {
      "affiliations": [
        "IQVIA, Real World Solutions, Brighton, England, UK"
      ],
      "name": "James Brash"
    },
    {
      "affiliations": [
        "Department of Medical Informatics and Clinical Epidemiology, Oregon Health & Science University, Portland, OR, USA"
      ],
      "name": "Nicole Weiskopf"
    },
    {
      "affiliations": [
        "Stanford School of Medicine and Stanford Health Care, Stanford, California, USA"
      ],
      "name": "Hannah Morgan-Cooper"
    },
    {
      "affiliations": [
        "Stanford School of Medicine and Stanford Health Care, Stanford, California, USA"
      ],
      "name": "Priya Desai"
    },
    {
      "affiliations": [
        "Wilmer Eye Institute, Johns Hopkins School of Medicine, Baltimore, Maryland, USA"
      ],
      "name": "Diep Tran"
    },
    {
      "affiliations": [
        "Wilmer Eye Institute, Johns Hopkins School of Medicine, Baltimore, Maryland, USA"
      ],
      "name": "Zainab Rustam"
    },
    {
      "affiliations": [
        "Wilmer Eye Institute, Johns Hopkins School of Medicine, Baltimore, Maryland, USA"
      ],
      "name": "Gina Zhu"
    },
    {
      "affiliations": [
        "Janssen Research & Development, Titusville, New Jersey, USA"
      ],
      "name": "Joel Swerdel"
    },
    {
      "affiliations": [
        "Janssen Research & Development, Titusville, New Jersey, USA",
        "Department of Medical Informatics, Erasmus University Medical Center, Rotterdam, Netherlands"
      ],
      "name": "Anthony Sena"
    },
    {
      "affiliations": [
        "Biomedical Informatics and Data Science, Division of General Internal Medicine, Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD, USA"
      ],
      "name": "Paul Nagy"
    },
    {
      "affiliations": [
        "Department of Biostatistics, University of California Los Angeles School of Public Health, Los Angeles, California, USA",
        "VA Informatics and Computing Infrastructure, US Department of Veterans Affairs, Salt Lake City, UT, USA"
      ],
      "name": "Marc Suchard"
    },
    {
      "affiliations": [
        "Department of Biostatistics, University of California Los Angeles School of Public Health, Los Angeles, California, USA",
        "Johnson and Johnson, Horsham, PA, USA"
      ],
      "name": "Martijn Schuemie"
    },
    {
      "affiliations": [
        "Department of Biomedical Informatics, Columbia University, New York, New York, USA"
      ],
      "name": "George Hripcsak"
    },
    {
      "affiliations": [
        "Department of Biomedical Informatics, Columbia University, New York, New York, USA",
        "Johnson and Johnson, Horsham, PA, USA"
      ],
      "name": "Patrick Ryan"
    }
  ],
  "full_text": "WHAT IS ALREADY KNOWN ON THIS TOPIC Semaglutide, a glucagon-like peptide-1 receptor agonist (GLP-1RA) used to treat type 2 diabetes mellitus, has been associated with higher rates of diabetic retinopathy complications.WHAT THIS STUDY ADDS In this retrospective study of adults with type 2 diabetes mellitus across 14 databases in the Observational Health Data Sciences and Informatics Evidence Network, we did not identify an increased risk for proliferative diabetic retinopathy or treatment-requiring diabetic retinopathy comparing 810 390 new users of semaglutide with other GLP-1RAs and non-GLP-1RAs.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY In contrast to some prior studies, real-world evidence does not suggest an increased risk of diabetic retinopathy complications with semaglutide. Patients with diabetes should still be monitored for vision-threatening complications.Introduction Semaglutide is a glucagon-like peptide-1 receptor agonist (GLP-1RA) approved by the US Food and Drug Administration (FDA) for the treatment of type 2 diabetes mellitus (T2D). 1–5 Given its well-documented beneficial effects on cardiovascular and kidney outcomes, it is one of the most commonly used agents for T2D treatment.6 Evidence of semaglutide’s impact on ophthalmic complications is conflicting. Across the Semaglutide Unabated Sustainability in Treatment of Type 2 Diabetes (SUSTAIN) clinical trials, SUSTAIN-6 demonstrated higher rates of diabetic retinopathy (DR) complications (including vitreous hemorrhage, blindness or conditions requiring treatment with an intravitreal agent or photocoagulation) (HR 1.76; 95% CI 1.11 to 2.78; p=0.02) when comparing semaglutide to placebo.3 A similar trend was seen with the development of DR among patients taking oral semaglutide compared with placebo in the Peptide Innovation for Early Diabetes Treatment (PIONEER)-6 trial.7 Other trials such as Semaglutide Cardiovascular Outcomes Trial (SOUL) excluded patients with DR of maculopathy.8 Importantly, meta-analyses across multiple clinical trials suggest that this elevated risk of DR is specific to semaglutide and not associated with other GLP-1RAs.9 10 Despite these findings, retrospective studies in routine clinical practice have shown mixed results in the risk of DR complications associated with GLP-1RAs.11–18 Many of these retrospective studies focus on GLP-1RAs as a class of medications without an investigation of the drug-specific effects of semaglutide.11–13 15–18 There is therefore a need to better understand the risk of DR complications associated with semaglutide exposure in routine clinical practice. The Observational Health Data Sciences and Informatics (OHDSI) Evidence Network is an international open-science collaborative centered around the Observational Medical Outcomes Partnership Common Data Model with a focus on generating reliable real-world evidence with open-source analytics.19–21 This study leveraged the OHDSI Evidence Network to systematically and thoroughly assess at scale whether semaglutide exposure was associated with increased risk of DR complications compared with other GLP-1RAs and non-GLP-1RAs.Methods Study design This was a retrospective study of 14 databases (6 administrative claims and 8 electronic health records) in the OHDSI Evidence Network using an active-comparator new-user cohort design. 22 23 Database details are in online supplemental eTable 1. All participating sites had local institutional review board (IRB) exemptions for use data for secondary research or approvals (eg, IRB at Johns Hopkins University approved the study IRB00296724, waiver of consent was granted). The study adhered to the tenets of the Declaration of Helsinki and followed Strengthening the Reporting of Observational Studies in Epidemiology guidelines.24 SP110.1136/bmjdrc-2025-005424.supp1Supplementary data Subjects and exposures Adults ≥18 years old who were new users of semaglutide (GLP-1RA), dulaglutide (GLP-1RA), exenatide (GLP-1RA), empagliflozin (sodium-glucose cotransporter-2 (SGLT2) inhibitor), sitagliptin (dipeptidyl peptidase-4 (DPP4) inhibitor) or glipizide (sulfonylurea) as second-line treatment for T2D during the study period (1 December 2017–31 December 2023) were included. These medications were identified by prior OHDSI network studies as the most commonly prescribed in each drug class. 2 5 New users were defined as previously described.2 5 In brief, adults with T2D were included if they were on metformin monotherapy, had at least 1 year of prior observation, initiated treatment with one of the medications of interest, had no prior exposure to a comparator diabetes medication and had at most 30 days of prior insulin use23 (online supplemental eTables 2 and 3). Time-at-risk began with medication initiation until the end of continuous drug exposure, defined as a gap in exposure of >30 days or the end of the continuous observation period.DR outcomes A consortium of board-certified ophthalmologists, including fellowship-trained retina specialists, developed two primary definitions for DR complications ( online supplemental eTables 2 and 3). (1) Proliferative DR (PDR) was defined as the first occurrence of a diagnosis code for PDR. (2) Treatment-requiring DR or diabetic macular edema (DME) defined as the initiation of intravitreal anti-vascular endothelial growth factor (VEGF), focal laser photocoagulation, panretinal photocoagulation or pars plana vitrectomy for a diagnosis of DR or DME noted in the preceding 90 days. Patients with diagnoses for age-related macular degeneration or retinal vein occlusion were excluded.Statistical analysis Baseline characteristics of patients within each T2D medication exposure cohort were summarized by database. 25–27 Unadjusted incidence proportions and incidence rates for each outcome were also calculated.25 28 Active-Comparator New-User Cohort Analysis Propensity scores were estimated using a large-scale propensity score approach, applying regularized regression using all baseline characteristics (eg, demographic characteristics, pre-existing conditions, medications and procedures).29 30 Propensity-score adjusted HRs for PDR and treatment-requiring DR/DME were estimated using Cox proportional hazards models comparing new semaglutide users with users of comparator medications: other GLP-1RAs (dulaglutide and exenatide) and non-GLP-1RA drugs (empagliflozin, sitagliptin and glipizide) while on-treatment.31 Patients in each target and comparator exposure cohort were matched 1:1 using propensity scores.A set of study diagnostics was employed including pre-exposure covariate balance after propensity score adjustment, empirical equipoise, minimum power requirement and evaluation of systematic error and bias using negative control outcomes.32–36 A total of 97 outcomes believed to be causally unrelated to the exposures under investigation were included to detect and adjust for systematic errors including misclassification, surveillance and selection biases, in addition to residual confounding23 34 36 (online supplemental eTable 4). Only databases that passed prespecified study diagnostics reported HR estimates.22 A random-effects meta-analysis combined each database’s estimates into a network-wide estimate.37 38 Sensitivity analyses Three sensitivity analyses were performed to test the robustness of our findings against cohort specification and study design ( online supplemental eTables 2 and 3). First, a temporal split was employed in which the main cohort was restricted to three distinct time windows (December 2017 to January 2020, February 2020 to June 2021, July 2021 to December 2023). This addressed potential biases due to decreased healthcare utilization from the COVID-19 pandemic, and the 60% increased prescription of semaglutide from 2021 to 2023 after FDA approval for obesity.39 Second, as a sensitivity analysis to cohort specification, we created a second definition to include new users of each T2D medication, regardless of prior exposure to metformin or a comparator drug. Third, a sensitivity analysis of the study design was also employed. A self-controlled case series (SCCS) in which patients acted as their own controls was used to estimate the incidence rate ratio (IRR) of the DR outcomes associated with each T2D medication exposure compared with non-exposure control time.32 40–42 The observation period was restricted to when patients had T2D, and the patients’ first 365 days in the database were also excluded. Conditional Poisson regression models were used to compare the outcome IRRs.43 A separate pre-exposure time window was defined as the 30 days prior to treatment initiation and separately adjusted.43 44 Models also adjusted for calendar months.43 44 A set of study diagnostics were employed, and only databases that passed diagnostics were included in the random-effects meta-analysis.23 33–38 Results A total of 14 databases were included in the study, which, in total, included patients with T2D who were new users of semaglutide (810 390), dulaglutide (326 282), exenatide (25 936), empagliflozin (715 802), sitagliptin (493 563) and glipizide (832 295) ( online supplemental eTable 1, 5–8). Baseline characteristics of patients in the Optum’s deidentified Clinformatics Data Mart Database are presented in table 1 as a representative example; all other databases are in online supplemental eTable 9. All results are provided via an interactive web application.45 46 Table 1Baseline patient characteristics in each type 2 diabetes (T2D) drug exposure cohort (semaglutide, dulaglutide, exenatide, empagliflozin, sitagliptin, glipizide) during the study period in the Optum’s deidentified Clinformatics Data Mart Database—Socio-economic Status (Optum Clinformatics)Patients, number (%)Semaglutide (GLP-1 RA)Dulaglutide (GLP-1 RA)Exenatide (GLP-1 RA)Empagliflozin (SGLT2 inhibitor)Sitagliptin (DPP4 inhibitor)Glipizide (sulfonylurea)N=43 620N=14 923N=1414N=43 302N=19 581N=44 092Age (in years)*  ≤29453 (1)168 (1)23 (1)165 (0)29 (0)174 (0)  30–498110 (19)3095 (21)315 (22)5122 (12)875 (4)4478 (10)  50–6924 473 (56)8238 (55)832 (59)21 614 (50)7292 (37)20 330 (46)  ≥7010 586 (24)3426 (23)248 (18)16 401 (38)11 389 (58)19 111 (43)Race  Asian1179 (3)332 (2)37 (3)2333 (5)961 (5)2008 (5)  Black or African American5748 (13)1939 (13)166 (12)5228 (12)3020 (15)5458 (12)  White28 635 (66)9976 (67)900 (64)27 337 (63)11 283 (58)26 464 (60)Ethnicity  Hispanic or Latino4920 (11)1703 (11)228 (16)5329 (12)2892 (15)6836 (16)  Not Hispanic or Latino35 562 (82)12 247 (82)1103 (78)34 898 (81)15 264 (78)33 930 (77)Sex  Female26 699 (61)8343 (56)797 (56)17 872 (41)10 683 (55)20 587 (47)  Male16 921 (39)6580 (44)617 (44)25 430 (59)8898 (45)23 505 (53)Medical conditions  DCSI (SD)†2.57 (1.97)2.64 (1.97)2.72 (2.08)3.25 (2.25)3.57 (2.39)3.26 (2.31)  CCI (SD)†4.04 (2.37)4.05 (2.33)3.92 (2.23)4.69 (2.67)5.03 (2.71)4.64 (2.68)  Macular edema due to diabetes mellitus248 (1)146 (1)19 (1)458 (1)229 (1)541 (1)  Mild nonproliferative retinopathy due to diabetes mellitus811 (2)335 (2)38 (3)1249 (3)634 (3)1391 (3)  Moderate nonproliferative retinopathy due to diabetes mellitus127 (0)61 (0)5 (0)251 (1)133 (1)290 (1)  Severe nonproliferative retinopathy due to diabetes mellitus44 (0)26 (0)<5 (0)74 (0)42 (0)96 (0)*In each database, age groups span 5 years, and counts fewer than 5 are reported as<5. When combining age groups for reporting, all such categories were treated as having 5 patients.†The mean and SD are shown.CCI, Charlson Comorbidity Index—Romano adaptation; DCSI, Diabetes Complications Severity Index; DPP4, dipeptidyl peptidase-4; GLP-1 RA, glucagon-like peptide 1 receptor agonist; SGLT2, sodium-glucose cotransporter-2.Among all patients with T2D, the mean incidence of PDR was 675.9 per 100 000 persons and 356.0 per 100 000 person-years, and the mean incidence of treatment-requiring DR/DME was 408.8 per 100 000 persons and 218.8 per 100 000 person-years (online supplemental eTable 10). Among new semaglutide users, the mean incidence of PDR was 11.2 per 100 000 persons and 22.6 per 100 000 person-years, and treatment-requiring DR/DME was 5.4 per 100 000 persons and 11.0 per 100 000 person-years.The risk for PDR for new users of semaglutide as second-line therapy was similar to dulaglutide (GLP-1RA) (HR 0.81, 95% CI 0.42 to 1.54, p=0.51), empagliflozin (SGLT2 inhibitor) (HR 0.83, 95% CI 0.53 to 1.30, p=0.41) and sitagliptin (DPP4 inhibitor) (HR 0.83, 95% CI 0.45 to 1.55, p=0.57) but was lower compared with glipizide (sulfonylurea) (HR 0.59, 95% CI 0.39 to 0.88, p=0.01) (figure 1). The risk for treatment-requiring DR/DME for new users of semaglutide as second-line therapy was similar to empagliflozin (SGLT2 inhibitor) (HR 0.66, 95% CI 0.43 to 1.02, p=0.06) but lower compared with dulaglutide (HR 0.53, 95% CI 0.31 to 0.91, p=0.02), sitagliptin (0.46, 95% CI 0.26 to 0.81, p=0.008) and glipizide (HR 0.55, 95% CI 0.33 to 0.91, p=0.02) (figure 1).Figure 1Forest plot for the active-comparator new-user cohort analysis. HR and 95% CI estimates for the risk of proliferative diabetic retinopathy and treatment-requiring diabetic retinopathy while on treatment with a second-line type 2 diabetes (T2D) medication comparing between semaglutide and other GLP-1RAs (dulaglutide, exenatide) and non-GLP-1RA medications (empagliflozin, sitagliptin and glipizide). Only results from databases and comparisons that passed study diagnostics are provided, as well as the meta-analytic estimates. *The total number of patients in each comparison is shown. Panel on the left shows the results for proliferative diabetic retinopathy (PDR), while the panel on the right shows the results for treatment-requiring diabetic retinopathy (DR) or diabetic macular edema (DME). GLP-1 RA, glucagon like peptide-1 receptor agonist. CCAE, Merative MarketScan Commercial Claims and Encounters Database‬; IQVIA, IQVIA Open Claims‬; ‭Optum EHR, Optum de-identified Electronic Health Record data set‬; PharMetrics, PharMetrics Plus‬.In the sensitivity analysis with a temporal split, when restricted by calendar time, the risk for treatment-requiring DR/DME for semaglutide users was lower compared with empagliflozin (SGLT2 inhibitor) but only from 2021 to 2023 (HR 0.54, 95% CI 0.31 to 0.95, p=0.03) (online supplemental eFigure 1). In the sensitivity analysis to cohort specification, for new users of semaglutide, when second-line therapy was not specified, the risk for PDR and treatment-requiring DR/DME was no longer different comparing semaglutide with any of the other GLP-1RAs or non-GLP-1RAs (online supplemental eFigure 2).In the SCCS sensitivity analysis, there was no increased IRR of PDR associated with semaglutide exposure (IRR 1.10, 95% CI 0.97 to 1.26, p=0.13), but there was an increased rate for treatment-requiring DR/DME associated with semaglutide exposure (IRR 1.23, 95% CI 1.05 to 1.43, p=0.01) (online supplemental eFigure 3). The IRR for the two DR outcomes for the other medications is shown in online supplemental eFigure 3. The pre-exposure IRR in the 30 days prior to medication exposure is shown in online supplemental eTable 11.Discussion In this retrospective cohort study across 14 databases in the OHDSI Evidence Network, there was no increased risk for either PDR or treatment-requiring DR/DME comparing semaglutide with other GLP-1RAs or non-GLP-1RAs. In the primary active-comparator new-user cohort analysis, there was evidence of a lower risk of PDR when comparing second-line use of semaglutide with glipizide, and a lower risk of treatment-requiring DR/DME when comparing semaglutide with other GLP-1RAs (dulaglutide) and non-GLP1-RAs (sitagliptin, glipizide).The potential for worsening retinopathy with intensive glycemic control has been known since the 1980s from the landmark Diabetes Control and Complications Trial (DCCT), which studied intensive and conventional treatment among patients with insulin-dependent diabetes. The DCCT assessed retinopathy severity using stereoscopic fundus photographs graded using the Early Treatment Diabetic Retinopathy Study scale. This method of DR assessment is still considered the gold standard in modern clinical trials.47 48 The DCCT found that the risk of significant worsening of retinopathy severity at the 6 and/or 12-month visit was higher among patients assigned to the intensive treatment (13.1%) compared with patients assigned to the conventional group (7.6%).49 50 The magnitude of the hemoglobin A1c reduction was an important risk factor for early DR worsening.49 The Oslo study similarly found that rapid glycemic control was associated with the development of cotton wool spots concerning for transient retinopathy worsening in patients with insulin-dependent diabetes.51 In both studies, these changes appeared transient and were not associated with serious or irreversible vision loss. Importantly, in the long term, the risk of progression of microvascular complications, including retinopathy, was reduced with tighter glycemic and blood pressure control.49 50 52–54 These results are relevant to semaglutide and other GLP-1 RAs, which are considered high-efficacy glycemic therapy, resulting in greater reductions in hemoglobin A1c compared with other second-line T2D medications.55 The potential for early worsening of DR with reductions in hemoglobin A1c as identified in DCCT, combined with findings from SUSTAIN-6 and PIONEER-6 and other studies, has fueled ongoing concern for potential worsening of DR with semaglutide use.3 7 9 Large-scale observational health studies of DR often have to use proxy measures since fundus photographs for grading severity are not typically available. Administrative claims databases do not include retinal imaging, and photos are challenging to share across sites.56 Diagnosis codes to identify DR, without consideration of severity, as has been used in some studies,16–18 is prone to outcome misspecification as less severe DR, for example, non-proliferative DR, is often not coded.57 58 We chose to focus on DR complications—one on a diagnosis of PDR and another on treatment—requiring DR/DME. PDR is typically well coded;57 58 however, by using diagnosis codes alone and not requiring a prior instance of non-proliferative DR, we could be capturing prevalent cases of PDR that were previously treated.59 As a proxy for incident DR complications, we created another definition requiring procedural interventions such as laser, intravitreal anti-VEGF injections and retinal surgeries, as has been done in prior studies.12 The treatment-requiring DR/DME includes not only PDR but also other complications, such as DME requiring intervention. A limitation of both DR outcomes is that we do not know the precise changes in retinopathy staging. The ongoing prospective clinical trial evaluating how semaglutide affects diabetic eye disease compared with placebo among patients with T2D (FOCUS) will help clarify the precise impact of semaglutide on DR staging.60 In the primary analysis, comparing new users of semaglutide as a second-line treatment for T2D with other GLP-1RAs and non-GLP-1RAs, we did not identify an increased risk for PDR or for treatment-requiring DR/DME. There appeared to be a lower risk of PDR comparing semaglutide with glipizide and a lower risk of treatment-requiring DR/DME when comparing semaglutide with other GLP-1RAs (dulaglutide) and non-GLP-1RAs (sitagliptin, glipizide). We chose to restrict our analysis to new users of these medications as second-line treatment after metformin use, based on the American Diabetes Association (ADA) guidelines at the time of study design.55 By restricting our analysis to new second-line users, we are more likely to compare patients with T2D at similar points of their disease course.5 61 Attempting to control for potential confounding from the duration of T2D is critical since it is a major risk factor for both PDR and DME.62–64 By removing the requirement for second-line therapy in one of our sensitivity analyses, we no longer identified a decreased risk of the DR outcomes with semaglutide compared with other GLP-1RAs and non-GLP-1RAs. Not restricting our analysis to second-line users of T2D medications allowed the inclusion of patients with more severe T2D who could be on semaglutide as third or even fourth-line treatment, particularly given the expanding indications of semaglutide for the reduction of cardiovascular and kidney outcomes.1 3 4 The SCCS sensitivity analysis suggests a small increase in treatment-requiring DR/DME associated with semaglutide exposure compared with non-exposure, though elevated rates were also observed with other T2D medications including glipizide. However, these findings should be interpreted cautiously due to potential time-varying confounding by disease duration and severity,43 44 and indications of bias44 in elevated pre-exposure IRR estimates across all T2D medications. The elevated IRR for both DR outcomes in the 30-day window immediately prior to T2D medication exposure could reflect healthcare utilization patterns whereby patients are more likely to be referred for further management of systemic T2D after significant retinopathy is identified. Finally, the SCCS may reflect transient changes in DR paralleling the rapid decrease in hemoglobin A1c associated with semaglutide use3 52; however, these changes are challenging to capture in this analysis due to the inadequate specificity of diagnosis codes for capturing DR progression.57 58 Given these biases, the SCCS findings should be considered exploratory rather than conclusive.The lack of elevated PDR or treatment-requiring DR/DME risk comparing semaglutide to other GLP-1RAs and non-GLP-1RAs should reassure the prescriber. Our findings are in direct contrast to other observational studies identifying an increased risk of DR outcomes with GLP-1 RA exposure compared with SGLT-2 inhibitors.13 15 These studies were limited to examination of class effects (GLP-1RA) without investigation of drug-specific effects (semaglutide) or only examined patients on GLP-1RA or SGLT-2 inhibitor monotherapy, thus excluding patients prescribed these medications as second-line treatment for T2D as recommended by the ADA.55 Extrapolating findings from DCCT, it should be reassuring that any early worsening is transient in nature and does not affect visual acuity outcomes. With sufficient ophthalmic monitoring, evidence suggests that treating vision-threatening complications (ie, center-involved DME, PDR) after they appear leads to similar vision outcomes as attempting to prevent those complications.48 65 Prescribers of semaglutide and other T2D medications can help ensure that their patients with T2D are closely monitored by eye care providers for the development of vision-threatening complications.The incidence of PDR and treatment-requiring DR/DME among patients with T2D varies across studies. Using administrative claims from the USA, an estimate for the development of PDR at 5 years following initial diagnosis of T2D is 1.74%, which is similar to our findings.59 A study leveraging administrative claims from Japan estimates the incidence of treatment-requiring DR/DME ranging from 149 to 1722 per 100 000 person-years, depending on baseline characteristics.66 We identified a much lower incidence of PDR and treatment-requiring DR/DME among new users of semaglutide compared with other studies.12 13 The restriction of our analysis on new users of semaglutide as a second-line treatment for T2D and differences in the underlying base populations could be major contributors to a lower incidence rate.This study has limitations and strengths. Since this was a retrospective large database study, we were unable to assess the precise staging of DR using fundus photographs or other imaging modalities. Although we used a robust and previously leveraged definition of treatment-requiring DR/DME for DR complications, this definition did not capture patients who might require treatment but did not receive it. We were also unable to evaluate whether differences in treatments resulted in differences in visual acuity outcomes. This highlights a broader limitation of many databases for ophthalmic research, as eye examination data are not routinely included, and work is ongoing to standardize these data and incorporate them into the databases to enable future analyses.67 68 Hemoglobin A1c is a major confounder for both PDR and treatment-requiring DR/DME that could not be adjusted for. We also are unable to adjust for duration of diabetes. Despite these limitations, this is, to our knowledge, the largest retrospective observational health study investigating the association of semaglutide with DR complications. Other strengths of the study include its use of sensitivity analyses to test the robustness of our findings against study choices and the use of diverse databases that include both administrative claims and electronic health records.In conclusion, despite prior evidence and concern for DR worsening, these real-world data do not suggest an increased risk for either PDR or treatment-requiring DR/DME comparing semaglutide with other GLP-1RAs or non-GLP-1RAs. Patients with T2D should undergo close follow-up with eye care providers for DR screening and follow-up, particularly when initiating new medications.",
  "title": "Semaglutide and diabetic retinopathy: an OHDSI network study",
  "uid": "0024965f-9e92-5164-b1a8-71e0b2403ee6"
}