Routine Islet Autoantibody Testing in Clinically Diagnosed Adult-Onset Type 1 Diabetes Can Help Identify Misclassification and the Possibility of Successful Insulin Cessation

Identifying type 1 diabetes in adults can be challenging. Clinical features that can help in distinguishing type 1 and 2 diabetes frequently overlap, and the high prevalence of type 2 diabetes means even classical features of type 1 diabetes, such as low BMI or ketoacidosis, may not confirm the diagnosis (1–5). As a result, misclassification in adults is common: studies using biomarker-based approaches suggest that approximately one in three adults developing type 1 diabetes are initially diagnosed as having type 2 diabetes and one in six diagnosed with type 1 diabetes do not have this condition (6–9).

Recommendations for education, treatment, and monitoring in different diabetes subtypes vary markedly. Misclassifying type 2, or maturity-onset diabetes of the young (MODY), cases as type 1 diabetes can therefore result in suboptimal clinical management, inappropriate education, potentially unnecessary insulin treatment, and reduced access to agents shown to have cardiovascular benefits in type 2 diabetes. Routine C-peptide testing in those with clinically diagnosed type 1 diabetes has recently been shown to lead to reclassification and insulin withdrawal in many patients, but C-peptide testing is likely to have limited utility at diagnosis, as levels may be retained at diagnosis in type 1 diabetes (6,10).

Islet autoantibody testing may assist differentiation of type 1 diabetes from other diabetes subtypes (11–13) and has maximum utility at diagnosis, as levels can decline in long-standing disease. Recent guidance by the American Diabetes Association (ADA) and European Association for the Study of Diabetes (EASD) recommended islet autoantibody testing at diagnosis in all adults with clinically suspected type 1 diabetes; however, the clinical impact of routine testing in this population has not been directly assessed (14).

We aimed to assess whether absence of positive islet autoantibodies in recent-onset clinically diagnosed adult-onset type 1 diabetes is suggestive of misclassification and whether feedback of routinely measured islet autoantibodies is associated with subsequent change in patient treatment.

We used longitudinal data from the prospective study StartRight: Getting the Right Classification and Treatment From Diagnosis in Adults With Diabetes (ClinicalTrials.gov, clinical trial reg. no. NCT03737799, to explore the effect of islet autoantibody status on clinical, biochemical, and genetic characteristics in patients with clinically diagnosed type 1 diabetes treated with insulin from diagnosis. A reference population from the same study, of islet autoantibody–negative participants diagnosed with type 2 diabetes and not treated with insulin from diagnosis, was also included in this analysis.

A total of 1,798 adult (≥18 years) participants diagnosed with diabetes (of any type, excluding gestational diabetes mellitus, known secondary diabetes, and pregnancy at the time of recruitment) within the previous 12 months were recruited from 55 National Health Service (NHS) sites in the U.K. Recruitment took place between August 2016 and February 2020. Participants were followed annually for 2 years from recruitment. Eligible participants were identified through routine clinical care appointments, screening of primary and secondary care records, and self-referral after advertisements in various clinical settings, with the majority (57%) of participants identified though secondary care specialist teams (who manage almost all of those with type 1 diabetes in the U.K.). To ensure a sufficient number of participants with late-onset type 1 diabetes, the study population was enriched for older individuals with type 1 diabetes through aiming for equal recruitment of those treated with and without insulin among those age ≥50 years at diabetes diagnosis. A study flow diagram is presented in Supplementary Fig. 1.

Clinically diagnosed type 1 diabetes was defined as a self-reported clinical diagnosis with concurrent treatment with insulin at recruitment (N = 722 of 1,798). Confirmed type 2 diabetes was defined as a self-reported clinical diagnosis of type 2 diabetes with absence of insulin treatment at recruitment and negativity for islet autoantibodies (N = 731 of 1,798). Participants were considered to have “suspected type 1 diabetes” when they were treated with insulin at recruitment, and the self-reported diagnosis indicated clinical uncertainty (for example, “uncertain type,” “possible type 1 diabetes,” “likely type 1 diabetes,” “possible type 2 diabetes”) (N = 91 of 1,798).

Participants were excluded from analysis (14.1% of recruited participants [N = 254 of 1,798]) 1) if they had a reported diagnosis of type 2 diabetes and were either receiving insulin at recruitment (N = 85) or not receiving insulin but were islet autoantibody positive (N = 84) or 2) where there was clinical uncertainty (or other diabetes type) in a participant’s self-reported diagnosis and they were not receiving insulin at recruitment (N = 85) (see Supplementary Fig. 1).

Diabetes type, initial diabetes treatment, symptoms at diagnosis (thirst, polyuria, and patient-reported weight loss) and concurrent autoimmune conditions were self-reported at recruitment/baseline (median duration of diabetes at recruitment 4.0 months). At baseline, participants’ medical notes and laboratory records were reviewed by a research nurse to confirm diagnosis details, and biochemistry (HbA_1c, glucose) at diagnosis. Diagnosis notes were examined for evidence of diabetic ketoacidosis (DKA), including discharge diagnosis, measured ketones, and pH. At study recruitment a nonfasted (within 1–5 h of a meal) blood sample was collected and analyzed for serum C-peptide, islet autoantibodies (glutamic acid decarboxylase antibody [GADA], zinc transporter 8 [ZNT8A], islet antigen 2 [IA-2A]) and DNA extraction performed for generation of a type 1 diabetes genetic risk score (T1D-GRS). Baseline BMI was calculated from weight and height measured at the baseline visit.

At each annual visit including recruitment, participants collected a urine sample for urine C-peptide–to–creatine ratio (UCPCR) measurement (15,16). Samples were collected post–home meals. Participants were asked to fully empty their bladder immediately premeal and collect urine in containers with boric acid preservative 2 h after meal completion. The sample was then posted directly to the Exeter Clinical Laboratory for analysis.

Participants were then contacted by telephone or e-mail 1 year (within a minimum/maximum range of 10–16 months) and 2 years (within a minimum/maximum range of 22–28 months) postrecruitment for recording of concurrent treatment, treatment changes, and health service use, including hospital admission for diabetes-related illness. Postrecruitment HbA_1c results were obtained by the research team from health care records.

Laboratory results at diabetes presentation, and postrecruitment HbA_1c values, were obtained from participants’ health care records as previously stated. All other biochemical analysis was undertaken by the academic department of blood sciences at the Royal Devon and Exeter Hospital, Exeter, U.K. GADA, IA-2A, and ZNT8A islet autoantibodies were measured with the RSR ELISA assay (RSR, Cardiff, U.K.) on the DYNEX DS2 automated ELISA system (DYNEX Technologies, Worthing, U.K.). Islet autoantibodies were considered positive if they exceeded the 97.5th centile of a cohort of 1,559 population control subjects without diabetes: GADA ≥11 units/mL, IA-2A ≥7.5 units/mL, and ZNT8A ≥65 units/mL in those aged up to 30 years and ≥10 units/mL in those aged ≥30 years (17,18). These assays and thresholds have 97.5% (GADA and ZNT8A) and 99% (IA-2A) specificity based on the above analysis of the 1,559 people without diabetes in our local population. In the 2020 international islet autoantibody standardization program certification of the Exeter clinical laboratory, these assays and thresholds have 99% specificity for all three islet autoantibodies (GADA, IA-2A, and ZNT8A). Sensitivity was 74% for both GADA and ZNT8A and 72% for IA-2A.

C-peptide (blood and urine) was measured with the automated Roche Diagnostics (Manheim, Germany) E170 immuno-analyzer (limit of detection 3.3 pmol/L, inter- and intra-assay coefficients of variation <4.5% and <3.3%, respectively). Urine creatinine (for UCPCR) was analyzed with the Jaffe method on the Roche P800 modular analyzer.

Islet autoantibody results were reported to both the participant and the treating primary and secondary (if applicable) care clinician following the study baseline visit. The result reporting forms included basic advice on results interpretation, with the reporting forms shown in Supplementary Figs. 5 and 6. Baseline C-peptide, and follow-up results such as UCPCR, were not reported. The management of each participant was left entirely to the discretion of the clinician responsible for their care.

T1D-GRS based on 67 single nucleotide polymorphisms associated with type 1 diabetes, and accounting for interactions between 18 HLA DR-DQ haplotype combinations, was generated as previously described (19). Single nucleotide polymorphisms were directly genotyped by LGC genomics, as previously described (20). In this study the T1D-GRS was only included for participants with White European ethnicity, as it has not been fully validated for other ethnicities. This result was not reported to the participant or clinician.

Due to the skewed nature of the C-peptide and UCPCR data, all values were natural log transformed for analysis in line with previous studies (21–24), which also allowed for a linear fit in assessing change over time after diagnosis. The geometric mean and 95% CIs for these variables are presented in the tables. For modeling annual UCPCR, the intercept and slopes were determined with use of mixed-effects models as previously described (21), with random effects at the patient level to allow each patient to contribute multiple C-peptide values at different time points. The benefit of this random-intercept, random-slope model is that it allows for variability between individuals in terms of both C-peptide level at diagnosis (the intercept) and in the percentage change in C-peptide over time (the slope). We separately assessed groups defined by reported diabetes type, treatment, and antibody status using an interaction term within the mixed-effects model.

Due to the slope being on a log scale, coefficients were interpreted in terms of the percentage change per year (calculated from the exponential of the β-coefficient − 1). The half-life of C-peptide was calculated from log_e(0.5)/β. We determined the variability of individual slopes in the longitudinal models using the SD range (which we calculated by backtransforming the β-coefficient ± 1 SD of the slope).

Comparisons of clinical characteristics between islet autoantibody–positive and –negative participants were made with use of two-sample t tests for continuous data and Pearson χ² test for categorical data. The relationship between T1D-GRS and longitudinal C-peptide loss (annual change in natural log UCPCR) was assessed with mixed-effects models with random effects at the patient level, with percentage change per year calculated from the β-coefficient as described above. Statistical significance was defined as P < 0.05. All analysis was carried out in Stata, version 14.2 (StataCorp, College Station, TX), and GraphPad Prism, version 9.1.2 (GraphPad Software, San Diego, CA).

Of included participants, 46.8% (722 of 1,544) were classified as having clinician-diagnosed type 1 diabetes. Of these participants 24.8% (179 of 722) were negative for all three islet autoantibodies. A detailed breakdown of the islet autoantibody composition is shown in Supplementary Fig. 2. Of the included participants 47.3% (731 of 1,544) were classified as having confirmed type 2 diabetes and 5.9% (91 of 1,544) as “suspected type 1 diabetes.” Treatment status at ∼1 year and/or 2 years postrecruitment was available in 91.1% (N = 1,407 of 1,544) of individuals (see Supplementary Fig. 1), with median follow-up duration 24 months (interquartile range [IQR] 22, 26).

Compared with islet autoantibody–positive participants diagnosed with type 1 diabetes, autoantibody-negative participants were older (mean age at diagnosis, autoantibody negative vs. positive, respectively, 42.7 years [95% CI 40.6, 44.8] vs. 38.2 years [95% CI 37.0, 39.4] years) were more likely to have a higher BMI (27.4 kg/m² [95% CI 26.6, 28.2] vs. 25.0 kg/m² [95% CI 24.7, 25.4]), had a lower prevalence of concurrent autoimmune conditions (3.9% [95% CI 1.1, 6.8] vs. 16.3% [95% CI 13.2, 19.4]), and were more likely to be male (72.9% [95% CI 66.3, 79.5] vs. 50.5% [95% CI 46.2, 54.7]) (all P < 0.001), as shown in Table 1. Rates of DKA and osmotic symptoms were broadly similar; however, glucose at diagnosis was higher in antibody-negative participants and reported weight loss lower (both P < 0.05). The characteristics of the control cohort diagnosed with type 2 diabetes are shown in Supplementary Table 1.

The distribution of genetic susceptibility to type 1 diabetes, assessed by T1D-GRS, was bimodal in participants with islet autoantibody–negative type 1 diabetes, with a mean value of 10.85 (95% CI 10.43, 11.27), and intermediate between those with autoantibody-positive type 1 diabetes (13.09 [95% CI 12.92, 13.25]) (P < 0.001) and type 2 diabetes (10.12 [95% CI 9.93, 10.32]) (P < 0.001), as shown in Fig. 1A.

In the islet autoantibody–negative type 1 diabetes cohort, geometric mean plasma C-peptide at recruitment was higher (607.6 pmol/L [95% CI 500.5, 737.5]) than in those with autoantibody-positive type 1 diabetes (413.0 pmol/L [95% CI 382.1, 446.4]) (P < 0.001) but lower than in the control subjects diagnosed with type 2 diabetes (1,792.3 pmol/L [95% CI 1,710.9, 1,877.6]) (both P < 0.001). This cohort also had a slower progression of C-peptide loss: the annual change in UCPCR in islet autoantibody–negative participants diagnosed with type 1 diabetes was −24% [95% CI −14, −32], with a half-life of 2.6 years, compared with an annual decline of −43% [95% CI −39, −47], in those who were autoantibody positive (half-life 1.2 years) (P < 0.001), as shown in Fig. 1B. Annual change in UCPCR in the control type 2 diabetes cohort was −6% [95% CI −1, −11], half-life 11.2 years.

In participants positive for islet autoantibodies, type 1 diabetes genetic susceptibility, as assessed by the T1D-GRS, was not associated with C-peptide (UCPCR) decline (P = 0.3). In contrast, in those with a clinical diagnosis of type 1 diabetes and negative for islet autoantibodies, higher T1D-GRS was associated with a higher rate of C-peptide loss (P = 0.026), with a 1-unit increase in T1D-GRS associated with a 5.8% (95% CI 0.67, 11.1) greater annual decline in UCPCR.

Follow-up treatment status was available for 90.2% (164 of 179) and 89.4% (487 of 543) of those with clinically diagnosed type 1 diabetes who were negative and positive for islet autoantibodies, respectively. Median follow-up duration was 24 months (IQR 22, 26). After feedback of a negative islet autoantibody result, treatment change was seen in 36.6% (60 of 164) of autoantibody-negative participants: 22.6% (37 of 164) discontinued insulin and 14.0% (23 of 164) added adjuvant agents to continued insulin. This is shown in the treatment flowchart presented in Fig. 2. Conversely, in participants where a clinical diagnosis of type 1 diabetes was reinforced by the presence of positive islet autoantibodies, only 7.8% changed therapy: 1.0% (5 of 487) discontinued insulin, while 6.8% (33 of 487) subsequently added adjuvant glucose-lowering therapies (31 of 33 metformin) to continued insulin.

At the latest follow-up (median 24 months [IQR 22, 26]), mean HbA_1c was comparable between islet autoantibody–negative participants discontinuing insulin and those continuing insulin treatment: 57.5 mmol/mol (95% CI 51.6, 63.5) (7.4% [95% CI 6.9, 8.0]) vs. 60.8 mmol/mol (95% CI 56.2, 65.4) (7.7% [95% CI 7.3, 8.1]) (P = 0.4), as shown in Fig. 3. These were both comparable with HbA_1c of those who had an adjuvant glucose-lowering therapy added to concurrent insulin therapy, with an HbA_1c at latest follow-up of 57.3 mmol/mol (95% CI 44.8, 69.8) (7.4% [95% CI 6.3, 8.5]) (P > 0.05 for both); however, this group had a notably higher baseline HbA_1c than those continuing insulin alone or discontinuing insulin therapy (78.9 mmol/mol [95% CI 67.5, 90.3], 9.4% [95% CI 8.3, 10.4], vs. 68.7 mmol/mol [95% CI 64.0, 73.3], 8.4% [95% CI 8.0, 8.9], and 67.9 mmol/mol [95% CI 59.8, 76.0], 8.4% [95% CI 7.6, 9.1], respectively). No ketoacidosis was reported after insulin withdrawal; however, one participant changing treatment (from basal-bolus insulin to basal insulin with oral agents) developed ketoacidosis concurrent with coronavirus disease 2019.

The characteristics of islet autoantibody–negative participants diagnosed with type 1 diabetes, stopping and continuing insulin, are shown in Supplementary Table 2. The mean T1D-GRS of those stopping insulin was 9.47 (95% CI 8.52, 10.42), comparable with that of the type 2 diabetes cohort (mean 10.12 [95% CI 9.93, 10.32]), and consistent with these participants having nonautoimmune diabetes and being initially misclassified. The characteristics of the five individual participants diagnosed with type 1 diabetes who stopped insulin following a positive islet antibody test are shown in Supplementary Table 3; three of five were positive for multiple islet autoantibodies.

In additional analysis we evaluated the 91 study participants considered to have suspected type 1 diabetes, based on receiving insulin at recruitment and having a self-reported diagnosis indicating clinical uncertainty. Follow-up treatment status was available in 90.1% (N = 82 of 91) of these individuals.

Of these participants, 68.1% (62 of 91) were islet autoantibody negative; their characteristics, based on autoantibody status, are shown in Supplementary Table 4. Islet autoantibody–negative participants with suspected type 1 diabetes had a higher baseline C-peptide than those who were autoantibody positive, mean 1,159.4 pmol/L (95% CI 967.0, 1,390.2) vs. 747.8 pmol/L (95% CI 550.9, 1,015.2) (P < 0.05), with lower T1D-GRS, 9.98 vs. 12.60 (P < 0.001).

Following feedback of islet autoantibody results there was treatment change in 74.5% (41 of 55) of autoantibody-negative participants with suspected type 1 diabetes: 52.7% (29 of 55) discontinued insulin, and 21.8% (12 of 55) added oral hypoglycemic agents to continued insulin (Supplementary Fig. 3). In those continuing and discontinuing insulin, HbA_1c at latest follow-up was 53.1 mmol/mol (95% CI 42.5, 63.8) (7.0% [95% CI 6.0, 8.0]) and 51.3 mmol/mol (95% CI 45.5, 57.2) (6.8 [95% CI 6.3, 7.4]%), respectively (P = 0.7)—both similar to values of those adding adjuvant glucose-lowering therapies to continuing insulin treatment, 57.3 mmol/mol (95% CI 44.8, 69.8) (7.4% [95% CI 6.3, 8.5]) (both P > 0.05), as shown in Supplementary Fig. 4. No islet antibody–negative participants with suspected type 1 diabetes reported ketoacidosis during the follow-up period.

Our findings demonstrate that adults diagnosed with type 1 diabetes who are negative for islet antibodies have genetic and C-peptide characteristics that are intermediate between type 1 and type 2 diabetes. This is suggestive of substantial misclassification within this group, with this group likely to include those who have islet antibody–negative autoimmune (type 1) diabetes and those with nonautoimmune (predominantly type 2) diabetes who have been misclassified. Following feedback of a negative islet autoantibody result to participants and their treating clinician, alteration of treatment regimen, including successful insulin cessation, was common. In contrast, treatment change was rare in antibody-positive individuals. In those treated with insulin from diagnosis with uncertain diabetes type, more than one-half of those who were autoantibody negative discontinued insulin therapy. Insulin cessation was not associated with deterioration in glycemic control.

To our knowledge this is the first study to examine the clinical impact of routine islet autoantibody measurement in adults with a clinical diagnosis of type 1 diabetes. Our finding suggesting misclassification of clinically diagnosed type 1 diabetes in adults is consistent with a previous study defining diabetes according to maintained insulin secretion (≥3 years’ diabetes duration). Foteinopoulou et al. (6) showed that routine C-peptide testing of those with long-standing diabetes, followed by islet antibody and type 1 genetic risk score testing in selected individuals, was associated with reclassification of 11% of those developing diabetes as an adult, with reclassification rare (0.8%) in childhood-onset diabetes. Of reclassified individuals, 22% successfully discontinued insulin therapy.

Analysis of clinical phenotypes associated with a diagnosis of type 1 diabetes in relation to islet autoantibody status has previously been explored. Bravis et al. (25) identified a level of diagnostic heterogeneity associated with 268 islet autoantibody–negative individuals (adults and children) in the multiethnic After Diagnosis Diabetes Research Support System-2 (ADDRESS-2) cohort within 6 months of a diagnosis of type 1 diabetes. Relative to the 1,510 islet autoantibody–positive participants, the autoantibody-negative participants tended to be older (median 31.4 vs. 20.1 years old), have a higher BMI (25.5 vs. 23.9 kg/m²), and be male (72% vs. 56%). Although this was a mixed age cohort with a focus predominantly on clinical features, those highlighted in that study reflect several findings within our cohort. Interestingly, Thomas et al. (26) looked at the genetic predisposition to type 1 diabetes in the ADDRESS-2 cohort and demonstrated a significant reduction in genetic predisposition to type 1 diabetes in antibody-negative adults with clinically diagnosed type 1 diabetes. This was consistent with non–type 1 diabetes in 67% of those aged ≥18 years at diagnosis who had negative GADA, IA-2A, and ZNT8A autoantibodies. Islet autoantibody status did not affect genetic predisposition to type 1 diabetes in children, which may reflect the lower frequency of nonautoimmune diabetes in this population.

A key strength of StartRight is the routine measurement of GADA, IA-2A, and ZNT8A in all eligible adults, from primary and secondary care, recruited within 12 months of diabetes diagnosis regardless of perceived diabetes type. This was accompanied by detailed assessment of biomarkers, clinical features, and treatment regimens from diagnosis and over the proceeding 2 years, including an annual UCPCR, allowing for follow-up of β-cell function. Routine feedback of islet autoantibodies, without specific clinical guidance, meant that we were able to observe whether the results of testing were associated with subsequent changes in standard clinical care. This is fundamental for the consideration of routine testing, as the impact of a test on patient care and outcomes will, alongside test cost and practicality, determine its utility.

A key limitation of this analysis is that there was no control arm with regard to participants and their treating clinicians who did not receive feedback of the islet autoantibody results. This means it is possible that participants may have changed treatment or stopped insulin without routine antibody testing; however, the rarity of attempted treatment change in antibody-positive participants does suggest that clinicians will take different actions where islet antibody status is known. A further limitation is that recruitment and antibody assessment were up to 12 months after diagnosis of diabetes. While this is unlikely to meaningfully affect findings of islet autoantibody prevalence in adults (27), it is likely that some participants will already have had islet autoantibodies measured in clinical care, potentially underestimating the impact of routine testing at diagnosis. This study relied on participants’ self-reporting of diabetes type (alongside insulin treatment from diagnosis) and annual treatment changes (including timings). It is possible that reported diagnosis could have differed from clinicians’ actual diagnosis, as this was not confirmed from medical records, and that insulin withdrawal may not have been captured if temporary. Of note, the study had enriched recruitment for older people with type 1 diabetes and therefore may not be reflective of the population age distribution for type 1 diabetes. However, despite this enrichment only 26% of our participants with clinically diagnosed type 1 diabetes were diagnosed after age 50 years, with U.K. population data suggesting that 30% of all incident adult-onset type 1 diabetes occurs in this age-group (1). Our study population was mostly of European ancestry, with T1D-GRS analysis limited to this group, and further studies to assess the impact of ethnicity would be of great interest: diagnosing type 1 diabetes in adulthood can be particularly challenging in those whose ethnicity is associated with higher rates of type 2 diabetes in the comparatively young and thin (28). Consistent with this, the results of our study, and of others, demonstrate higher rates of negative islet antibodies in non-White ethnicity participants diagnosed with type 1 diabetes (25). Lastly, there was no routine MODY testing, though the authors expect the rates of MODY in this cohort of subjects with insulin-treated diabetes, who are predominantly aged >30 years at diagnosis, to be extremely low: in a similar cohort the prevalence in islet autoantibody–negative adults diagnosed at age >30 years was only 1% (26).

Our findings are clinically important, as they suggest that routine islet autoantibody testing in adults with diagnosed or suspected type 1 diabetes helps to identify a significant proportion who may be misclassified and in some cases will be able to stop insulin treatment. In contrast, one or more positive islet antibodies will usually confirm type 1 diabetes in this setting. While only a modest proportion of those clinically diagnosed with type 1 diabetes in our study stopped insulin treatment after islet antibody testing, misclassification may potentially result in a lifetime of unnecessary insulin treatment and lack of access to effective treatment approaches in type 2 diabetes. Therefore, identifying possible misclassification through routine antibody testing at diagnosis has the potential to lead to marked benefit for individual patients and, given the high cost of a lifetime of insulin treatment for type 1 diabetes, to be cost-effective (6,29). However, a formal analysis of cost-effectiveness, and longer-term data on patient outcomes, would be needed to address this question. Our findings support recent ADA/EASD guidelines recommending the testing of islet autoantibodies in all adults with newly diagnosed suspected type 1 diabetes and suggestions in this guidance that positive islet autoantibodies will usually confirm type 1 diabetes in this setting (14). It is important to emphasize that many islet autoantibody–negative participants will still have autoimmune (or idiopathic) type 1 diabetes and develop severe insulin deficiency with absolute insulin requirement. Any attempts to withdraw insulin where there is uncertainty regarding diabetes subtype must be undertaken with marked caution and appropriate patient safeguarding to prevent ketoacidosis. It must also be recognized that some people with classical type 1 diabetes can temporarily discontinue insulin in the “honeymoon period” but go on to absolute insulin requirement. Therefore, careful ongoing monitoring of those discontinuing insulin following an initial diagnosis of type 1 diabetes is essential, and longer-term follow-up of this group is an important area for future research.

As this study did not include direct intervention to attempt insulin withdrawal, it is likely that there will be antibody-negative individuals who remain misclassified and/or continue to receive unnecessary insulin treatment. Therefore, it is important that, as recommended in recent ADA/EASD guidance, plasma C-peptide be assessed after 3 years’ diabetes duration in those with negative islet autoantibodies to further clarify diabetes diagnosis and treatment requirements (10,14).

For researchers our findings provide further evidence that clinically diagnosed type 1 diabetes in adults is likely to consist of a mix of diagnosis in those with and without autoimmune type 1 diabetes. This is likely to be because the very high prevalence of type 2 diabetes in adults will make robustly discriminating true type 1 diabetes from atypical presentations of type 2 diabetes challenging: low prior likelihood of type 1 diabetes may mean that atypical presentations of type 2 diabetes (e.g., with ketoacidosis, marked hyperglycemia, or low BMI) are relatively common in comparison with classical presentations of type 1 diabetes (5,26). Some reported characteristics of type 1 diabetes in older adults, such as low islet autoantibody prevalence, may potentially reflect the inadvertent study of those with and without autoimmune diabetes, and our findings, along with other research in this area, suggest a need to combine clinical diagnosis with a confirmatory antibody test in this setting (6,25,26,30).

In adult-onset clinically diagnosed type 1 diabetes, negativity for islet autoantibodies should prompt careful consideration of other diabetes subtypes. When routinely measured, negativity for antibodies is associated with successful insulin cessation. These findings support recent recommendations for routine islet autoantibody assessment in adult-onset type 1 diabetes.

Acknowledgments. The authors thank participants who took part in the study and the research teams who undertook cohort recruitment. The authors thank Rachel Nice of the Blood Sciences Department (Royal Devon and Exeter Hospital) for assistance with conducting laboratory aspects of the study and Richard Oram, Mike Weedon, and Diane Fraser (all University of Exeter) for assistance with deriving the T1D-GRS. The authors thank the ADDRESS-2 study team (Imperial College, London, UK) for support with participant recruitment.

Funding. StartRight is funded by the National Institute for Health and Care Research (NIHR) (CS-2015-15-018) and Diabetes UK (17/0005624). Genetic analysis was funded by the European Foundation for the Study of Diabetes (2016 Rising Star Fellowship). N.J.T. is funded by a Wellcome Trust–funded GW4 PhD fellowship. A.T.H. is supported by the NIHR Exeter Clinical Research Facility and a Wellcome Senior Investigator award and an NIHR Senior Investigator award. A.G.J. was supported for this work by an NIHR Clinician Scientist award (CS-2015-15-018).

The views given in this article do not necessarily represent those of the NIHR, the National Health Service, or the Department of Health.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. R.J.E., N.J.T., and A.G.J. designed the study. A.V.H., B.A.K., A.C., T.J.M., and A.G.J. researched data. R.J.E., N.J.T., B.M.S., and A.G.J. analyzed data. R.J.E., N.J.T., and A.G.J. wrote the first draft of the manuscript. All authors reviewed the draft and contributed to the revision of the manuscript. A.G.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at Diabetes UK Professional Conference 2022, London, U.K., 28 March–1 April 2022.