Validation of a rapid type 1 diabetes autoantibody screening assay for community‐based screening of organ donors to identify subjects at increased risk for the disease

Summary

The Network for Pancreatic Organ donors with Diabetes (nPOD) programme was developed in response to an unmet research need for human pancreatic tissue obtained from individuals with type 1 diabetes mellitus and people at increased risk [i.e. autoantibody (AAb)‐positive] for the disease. This necessitated the establishment of a type 1 diabetes‐specific AAb screening platform for organ procurement organizations (OPOs). Assay protocols for commercially available enzyme‐linked immunosorbent assays (elisas) determining AAb against glutamic acid decarboxylase (GADA), insulinoma‐associated protein‐2 (IA‐2A) and zinc transporter‐8 (ZnT8A) were modified to identify AAb‐positive donors within strict time requirements associated with organ donation programmes. These rapid elisas were evaluated by the international islet AAb standardization programme (IASP) and used by OPO laboratories as an adjunct to routine serological tests evaluating donors for organ transplantation. The rapid elisas performed well in three IASPs (2011, 2013, 2015) with 98‐100% specificity for all three assays, including sensitivities of 64–82% (GADA), 60–64% (IA‐2A) and 62–68% (ZnT8A). Since 2009, nPOD has screened 4442 organ donors by rapid elisa; 250 (5·6%) were identified as positive for one AAb and 14 (0.3%) for multiple AAb with 20 of these cases received by nPOD for follow‐up studies (14 GADA+, two IA‐2A⁺, four multiple AAb‐positive). Rapid screening for type 1 diabetes‐associated AAb in organ donors is feasible, allowing for identification of non‐diabetic, high‐risk individuals and procurement of valuable tissues for natural history studies of this disease.

Introduction

The JDRF Network for Pancreatic Organ donors with Diabetes (nPOD) programme was established to investigate the human pancreas in the context of type 1 diabetes mellitus, wherein the insulin‐producing beta cells in the islets of Langerhans are destroyed 1. While the onset of the disease may be acute, in the vast majority of cases it is preceded by a variable period of months to decades where non‐diabetic individuals possess autoantibodies (AAb) against one or more of a series of type 1 diabetes‐associated autoantigens. Over recent decades, these AAb have been utilized for a variety of purposes, including identifying those at high risk for developing type 1 diabetes mellitus 2, 3. Indeed, investigators within the type 1 diabetes research community have developed and sought actively to standardize assays to detect these AAb 4, 5, 6, 7, 8, 9, 10, 11, the majority in the format of radiobinding assays (RBA) for AAb against glutamic acid decarboxylase (GADA) 12, insulinoma‐associated protein‐2 (IA‐2A) 13, 14, zinc transporter‐8 (ZnT8A) 15 and insulin (IAA) 16. However, emerging commercially available enzyme‐linked immunosorbent assays (ELISAs) have provided alternative means for type 1 diabetes AAb screening 17, 18, 19 and possess logistical benefits compared to RBA.

The logistics and biobanking efforts for nPOD have been described previously 1, 20. Herein, we convey information regarding nPOD's novel AAb‐screening programme, an effort designed to identify non‐diabetic organ donors at risk for type 1 diabetes mellitus. This effort, perhaps deemed simple, in actuality required performance in situations unlike those undertaken in any setting to date. Specifically, screening needed to be performed within the general population (i.e. low prevalence of the disease), in rapid fashion (< 3 h) and in environments outside typical research laboratories (i.e. clinical laboratories). To achieve this, nPOD modified a series of commercial ELISA‐based assays and monitored their performance during a period of 6 years. This report conveys those developmental efforts and describes an effective means for rapid AAb screening of organ donors for the purpose of identifying highly valuable pancreata for addressing key research questions pertaining to the natural history of type 1 diabetes mellitus.

Materials and methods

ELISA core laboratory

Implementing nPOD AAb screening. The nPOD ELISA core laboratory (University of Florida, Gainesville, FL, USA) served as the central quality assessment laboratory for the commercial ELISAs developed by RSR Ltd (Cardiff, Wales, UK), which was chosen as the screening platform. The nPOD ELISA core also developed the modified assays. ELISAs for GADA and IA‐2A were modified from the manufacturer's protocol to ensure results in a total of 3 h by shortening the primary incubation step to 1 h, along with shaking at 500 rpm. The nPOD ELISA core laboratory provided training and equipment (spectrophotometer with 450 nm capability and plate washer) to the relevant organ procurement organization (OPO) screening laboratories, with follow‐up consultation as necessary.

Beginning in 2009, ELISA kits were distributed directly from the US distributor, Kronus Inc. (Boise, ID, USA) to OPO screening laboratories to measure GADA and IA‐2A using the modified assay protocol. In addition to a standard curve with five calibrators, positive and negative controls were supplied by the manufacturer. While six unknown samples could be run in duplicate from each of the 96‐well GADA and IA‐2A kits, samples were run one at a time because the nature of organ donation processes does not allow for batching of test material. Based on receiver operating characteristic (ROC) curves and testing of 270 healthy controls and 102 new‐onset type 1 diabetes subjects, cut‐offs defining AAb positivity were assigned for GADA and IA‐2A (Fig. 1a,b). This initial protocol was refined further, as described below.

Figure 1.

Receiver operating characteristic (ROC) curve analyses are shown for (a) glutamic acid decarboxylase (GADA), (b) insulinoma‐associated protein‐2 (IA‐2A) and (c) zinc transporter 8 (ZnT8A). Plots are 100 – specificity % (x‐axis) versus sensitivity % (y‐axis). The following cut‐offs were selected to maximize specificity: (a) 20 IU, 98% specificity and 70% sensitivity [area under the curve (AUC) = 0·88], (b) 60 IU, 99% specificity and 63% sensitivity (AUC = 0·80 and (c) 20 IU, 97% specificity with 57% sensitivity (AUC = 0·75).

Updates to the AAb screening programme. In 2011, after 2 years of experience and an evaluation by the international Islet AAb Standardization Program (IASP), the cut‐off for AAb positivity was raised to increase specificity. The result of this action was to increase concordance between the OPO screening laboratories and the nPOD ELISA core laboratory, as well as to improve concordance between the ELISA assays and confirmatory RBA assays conducted at the nPOD central AAb core (Denver, CO, USA), described below (Fig. 2). In 2013, an ELISA assay for ZnT8A became available from the same manufacturer. Again using the modified protocol described above (shortened primary incubation with 500 rpm shaking), the nPOD ELISA core laboratory validated this assay with 134 controls and 62 new‐onset type 1 diabetes samples with ROC curve analysis (Fig. 1c), and chose a positivity cut‐off favouring specificity over sensitivity to minimize the likelihood of procuring tissues from donors yielding a false positive result. The ZnT8A assay was then added to the OPO screening laboratories' nPOD portfolio.

Figure 2.

Overview of the Network for Pancreatic Organ donors with Diabetes (nPOD) autoantibody against insulin (AAb) screening programme over time. The overall concordance of the screening enzyme‐linked immunosorbent assay (elisa) with nPOD elisa core laboratory is shown for glutamic acid decarboxylase (GADA) (blue line), IA‐2A (green line) and zinc transporter 8 (ZnT8A) (black line; left y‐axis). The number of total subjects screened (bars) and those screened under 30 years of age (filled bars) are shown (right y‐axis). Changes to the programme are indicated with text box and arrows for each new event above the graph.

Reconfiguring the assay. In 2013, all ELISA kits were delivered to the ELISA core laboratory, where they were reconfigured into what is now termed the combination nPOD kit for AAb. Design changes were implemented to obtain 12 individual runs per purchased kit, compared to the six runs per kit allowed previously. Calibration curves were eliminated, and instead calibrators were combined from each kit (GADA, IA‐2A and ZnT8A) to create a single calibrator/cut‐off solution with a value equal to the previous positivity cut‐offs favouring specificity. A single‐negative control solution for all three analytes and a single‐positive control containing known AAb to all three autoantigens were also included. A strip well from the GADA kit was placed into column 1 of a strip well holder; then a strip from the IA‐2A kit was placed into column 2; and finally, a strip from the ZnT8A kit was placed into column 3. The entire holder was placed in a sealed foil pouch with a desiccating packet and labelled as a ‘single‐run kit’, suitable for all samples, calibrator and controls to be tested in duplicate. The streptavidin–horseradish peroxidase (HRP) (SAV) from each of the three original kits was combined and diluted to working concentration, and aliquots were prepared for use as a single run. The calibrator, controls and SAV aliquots were stored at 4°C. Biotinylated detection reagents for GADA, IA‐2A and ZnT8A were prepared using buffers of varying colour designations, and aliquots sufficient for a single‐run kit were stored at −20°C. Similarly, ready‐to‐use tetramethylbenzidine (TMB) and stop solution (1N H₂SO₄) single‐use aliquots were prepared and stored at 4°C. Thus, for each run, the calibrator solution, controls and unknown(s) were common to all three strips while the biotinylated detection reagents were specific, with the buffer colour‐coded to match each strip of GADA, IA‐2A and ZnT8A. The SAV, TMB and stop solutions were also common to all wells. Taken together, these kits containing the plate and all reagents diluted to working‐strength (in amounts sufficient for testing a single unknown sample) were distributed to the OPO screening laboratories. The combination kit expiration date was determined to be 20 weeks from the date of reconstituting the reagents.

Following each run, the results are read from a spectrophotometer at 450 nm, and pasted into an Excel (Microsoft, Redmond, WA, USA) template worksheet that calculates AAb titres based on an index relative to the calibrator/cut‐off [sample optical density (OD) divided by calibrator OD], coefficients of variation and pass/fail of the controls. AAb positivity is defined as an index > 1. Results are e‐mailed to the nPOD ELISA core laboratory and reported to the relevant OPO along with any other serological results that are usually sent with a transplant organ case. If the results are positive for any of the tested AAb, the OPO may then offer this case to nPOD, unless the organ is accepted for transplant or fails to meet nPOD inclusion criteria 1, 20. From 2009 to 2012, nPOD screened all research‐consented cases, and from 2012 to the present, all consented cases that were under the age of 30 years (Fig. 2: open bars, all subjects; filled bars, those under the age of 30).

Screening versus recovery serum samples

When possible, the screening laboratories send an aliquot of each AAb‐positive serum sample to the nPOD ELISA core laboratory. If a case is referred to nPOD and accepted, the pancreas and other tissues are recovered as described previously 1, 20. An additional serum sample is obtained at the time of organ recovery for both quality control and research purposes. Thus, for the majority of cases, the nPOD ELISA core laboratory can compare AAb results from the screening and recovery samples. It is important to note that nPOD works with a large number of OPOs, but not all of them are equipped to screen for AAb and, for some cases, only recovery samples are available. Therefore, some nPOD cases (type 1 diabetes mellitus, type 2 diabetes mellitus, controls and other categories) are not screened for AAb prior to organ procurement but may have a recovery serum sample for AAb testing after the fact. All available samples are tested in the ELISA core laboratory, and confirmed by RBA at the nPOD central AAb core laboratory.

nPOD central AAb core laboratory

The nPOD central AAb core laboratory has a long history of excellence in the type 1 diabetes AAb field 10, 13, 15, 21, 22, participating routinely in the Diabetes AAb Standardization Program (DASP), now renamed IASP. The nPOD central AAb core began testing for ZnT8A in nPOD cases using RBA prior to the introduction of the ZnT8A ELISA. This core also tests for IAA as there is currently no reliable ELISA for this analyte. Every case with available serum that is referred to nPOD is tested via RBA for GADA, IA‐2A, ZnT8A and IAA for either confirmation of the ELISA screening results, or for determination of final AAb status. In cases of discrepancy, the RBA supersedes the ELISA as the result reported on the nPOD website (www.jdrfnpod.org).

Statistical analysis

ROC curve analysis was performed using GraphPad Prism software (San Diego, CA, USA).

Results

Modified ELISA performance in screening and standardization programmes

Using in‐house samples (i.e. University of Florida) from healthy controls and subjects with new‐onset type 1 diabetes mellitus (Fig. 1 shows ROC curves; Supporting information, Fig. S1 shows the portion of the ROC curves for 95–100% specificity), the modified GADA ELISA has a sensitivity of 76% and a specificity of 95% at a cut‐off of 5 IU and a sensitivity of 70% and specificity of 98%, with a cut‐off of 20 IU. For the modified IA‐2A ELISA, the sensitivity was 64% and specificity of 98% with a cut‐off of 15 IU; raising the cut‐off to 60 IU revealed a sensitivity of 63% with a specificity of 99%. The modified ZnT8A ELISA showed sensitivity of 64% and specificity of 90% at a cut‐off of 10 IU; with an increased cut‐off of 20 IU, the sensitivity and specificity was 57 and 97%, respectively. The overall coefficients of variation (CV) were median 2·8% (range 0·2–12·2%) for GADA, 1·8% (0·1–12%) for IA‐2A and 3·1% (0·1–15%) for ZnT8A. Assay runs yielding CV > 15% were repeated unless the available serum volume was insufficient, in which case the run was rejected and the sample excluded from the study. Importantly, the CVs for low titre samples near the positivity cut‐off (5–100 units for each analyte) were comparatively lower than the entire cohort: 1·7% (0·1–5·4%) for GADA, 1·6% (0·1–11·4%) for IA‐2A and 4·1% (0·1–15%) for ZnT8.

The modified ELISAs were enrolled into IASP workshops (2011, 2013 and 2015), and under blinded conditions the assays performed with similar sensitivity and specificity profiles to the in‐house validation samples. Specifically, with nPOD ELISA core laboratory‐defined cut‐offs, IASP 2011 results revealed GADA sensitivity of 64% with 100% specificity; IA‐2A sensitivity of 64% with 100% specificity; and ZnT8A had 68% sensitivity and 98·9% specificity. Similarly, with nPOD‐defined cut‐off for GADA, IA‐2A and ZnT8A, IASP 2013 resulted in sensitivities of 64, 60 and 68% and specificities of 98·9, 100 and 98·9%, respectively. The most recent IASP 2015 with nPOD defined cut‐offs yielded sensitivities of 82, 64 and 62% with specificities of 98·9, 98·9 and 100% for GADA, IA‐2A and Znt8A, respectively (Table 1).

Table 1.

Results of the autoantibody against insulin (AAb) standardization programmes, now called Islet Autoantibody Standardization Program (IASP)

Standardization year	Analyte	nPOD cut‐off sensitivity (%)	nPOD cut‐off specificity (%)	Adjusted sensitivity with Specificity 95%	Area under curve‐ROC
2010/11	GADA	64	100	70	88·9
	IA‐2A	64	100	64	82·0
	ZnT8A	68	99	72	85·8
2012/13	GADA	64	99	72	89·1
	IA‐2A	60	100	62	81·0
	ZnT8A	68	99	72	84·7
2014/15	GADA	82	99	88	93·7
	IA‐2A	64	99	64	81·8
	ZnT8A	62	100	66	82·8

Each row represents a given analyte for each participating year in the programme. Columns 3–4 contain sensitivity and specificity as calculated by the programme organizers based on Network for Pancreatic Organ donors with Diabetes (nPOD) core enzyme‐linked immunosorbent assay (elisa) laboratory cut‐offs. The programme then also provides a sensitivity calculation with specificity fixed at 95% (column 5). The final column reflects the area under the curve of the receiver operating characteristic (ROC) for each analyte. GADA = glutamic acid decarboxylase; IA‐2A = insulinoma‐associated protein‐2; ZnT8A = zinc transporter 8.

The modified ELISA has identified successfully AAb‐positive donors without diabetes and type 1 diabetes organ donors

To date, nPOD has screened a total of 4442 samples from 17 OPOs (Fig. 2). In total, 5·6% were positive for single AAb and 0·3% were positive for multiple AAb. In 2009–12, when two AAb were used in the screen, frequencies were 6·7% for single and 0·2% for multiple AAb, and when thresholds were raised and three AAb were used post‐2012, 3·1% single and 0.5% multiple AAb‐positive samples were detected (Table 2). Of the 46 offered and accepted cases that were AAb‐positive at screening, 20 confirmed as AAb‐positive without diabetes, four of whom were multiple AAb‐positive, while 15 confirmed as AAb‐positive and were found to actually have existing T1D. Eleven did not confirm by RBA, with nine of those occurring prior to 2012 (27% of the 33 AAb‐positive accepted cases within that time‐frame) and two post‐2012 (15% of the 13 AAb‐positive accepted cases). In one particular case, the screening programme identified a double GADA plus ZnT8A‐positive individual who was later determined to have ketoacidosis, and was diagnosed posthumously as an at‐onset type 1 diabetes subject.

Table 2.

The number of screens performed each year with the number and frequency (%) of single and multiple autoantibody against insulin (AAb)‐positive subjects identified over time since the creation of the Network for Pancreatic Organ donors with Diabetes (nPOD) AAb screening programme

Year	2009	2010	2011	2012	2013	2014
Total screened, n	130	827	1352	802	674	657
Screened under 30 years of age, n	52	330	541	802	674	657
Single AAb+, n (%)	10 (7·7)	65 (7·9)	91 (6·7)	43 (5·4)	22 (3·3)	19 (2·9)
Multiple AAb+, n (%)	0 (0·0)	1 (0·1)	4 (0·3)	2 (0·2)	6 (0·9)	1 (0·2)

The lowering of the upper age limit to 30 years reduced dramatically the total number of donors screened per year; however, the total number under the age of 30 has since stabilized at approximately 700 per year (Fig. 2; filled bars, right‐hand y‐axis). Raising the positivity cut‐off and the introduction of the combination kits improved the concordance of the screening ELISA assays with the core laboratory ELISA assays from 70 to 98% (Fig. 2; line graph, left‐hand y‐axis).

Results from the modified ELISA can be confirmed by RBA

The result of the RBA assay ultimately determines the AAb status of each nPOD case, but due to time and technical requirements the RBA is not feasible for initial screening by clinical laboratories. When comparing the results between ELISA and RBA using recovery serum samples, it is apparent that while there is agreement in the majority of cases, there remains some discordance between these orthogonal formats for each of the analytes tested (GADA, IA‐2A and ZnT8A; Fig. 3). Over the entire duration of the AAb screening programme, discordant results include seven positive samples screened by ELISA that were not confirmed by RBA (four GADA, two IA‐2A and one ZnT8A). For type 1 diabetes cases, RBA identified two GADA, eight IA‐2A and four ZnT8A subjects who were negative by ELISA. Conversely, ELISA found five GADA, two IA‐2A and eight ZnT8A type 1 diabetes subjects who were negative by the RBA. RBA found one GADA‐positive type 2 diabetes case despite a negative ELISA result, while ELISA identified two GADA‐positive type 2 diabetes subjects with negative RBA results. ELISA also found one type 2 diabetes subject positive for ZnT8A, with negative RBA. One sample was IA‐2A‐positive via ELISA at both screening and recovery, while both samples were negative for IA‐2A but positive for GADA as assessed by RBA (Fig. 3).

Figure 3.

Venn diagrams are shown comparing the enzyme‐linked immunosorbent assay (elisa) (left) with the radiobinding assay (RBA) (right) results for glutamic acid decarboxylase (GADA) (top), insulinoma‐associated protein‐2 (IA‐2A) (middle) and zinc transporter 8 (ZnT8A) (bottom) within the various categories of organ donors. The RBA is set as the gold standard, upon which the autoantibody against insulin (AAb) status is determined.

The nPOD ELISA core laboratory also assessed the stability of AAb in the screening sample in comparison to the recovery sample (Fig. 4). While a few subjects showed variations in AAb titre, few demonstrated discordance in classification as positive versus negative for GADA or IA‐2A by either method (ELISA or RBA). However, two cases were positive for GADA by both ELISA and RBA assays on the screening sample but were subsequently negative by both ELISA and RBA on the recovery sample. This was thought to be attributed to haemodynamic management of the subjects during the intensive care unit (ICU) stay.

Figure 4.

Screening serum sample versus the organ recovery serum sample results by (a,b) enzyme‐linked immunosorbent assay (elisa) and (c,d) radiobinding assay (RBA) are compared for (a,c) glutamic acid decarboxylase (GADA) and (b,d) insulinoma‐associated protein‐2 (IA‐2A).

Discussion

The nPOD AAb screening programme demonstrates the feasibility of finding pre‐type 1 diabetes subjects among the US organ donor population using an ELISA‐based platform for three common disease‐associated AAb, GADA, IA‐2A and ZnT8A. The observed frequency of these AAb among the organ donor population is consistent with published literature for the general population 8, 9, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33. Raising the positivity cut‐off simultaneously improved the concordance within ELISAs (OPO screening versus nPOD ELISA core laboratories) and decreased the frequency of single AAb‐positive cases from 7·7 to 2·9%. While it is most common to maximize sensitivity in screening programmes, the nPOD programme has deliberately maximized assay specificity instead, for both ethical and logistical reasons. There is no evidence in the literature supporting or prohibiting transplant of pancreata from AAb‐positive organ donors, but we recognize that AAb screening results could influence the decision to transplant. It is our goal to keep organs available for the primary transplant mission and when no matched recipient exists, to procure precious tissue from true AAb‐positive individuals without diabetes; thus, we have structured our screening programme to err on the side of assay specificity and minimize the number of false positive samples. Thankfully, nPOD's experience has been that the presence of AAb has not swayed decisions about transplant, and a number of pancreata from our screened AAb‐positive individuals went to transplant.

While the prevalence of type 1 diabetes mellitus in the US general population is low (1 : 250), the nPOD AAb screening programme provides the best available means to identify organ donors with subclinical disease pathogenesis, and by confirming with orthogonal assays there is a higher level of confidence that these cases reflect pre‐diabetes, particularly with cases that are multiple AAb‐positive. Importantly, studies on tissues from these AAb‐positive cases are yielding results that suggest these individuals are probably on the path towards the development of type 1 diabetes mellitus 20. We acknowledge, however, that these observations are indirect, as follow‐up for progression to hyperglycaemia is not possible. It is conceivable that some subjects with a single AAb could have been non‐progressors; again, without follow‐up, we cannot be certain.

The comparison of ELISA and RBA results has led to the following observations. There was agreement between the two assays for the majority of samples; some exceptions were observed, however, including instances where either ELISA or RBA distinctly identified GADA, IA‐2A or ZnT8A in individuals diagnosed as having type 1 diabetes mellitus. It is currently not known if these discrepancies are the result of different epitopes that are identified uniquely by the format of the ELISA versus RBA 34. The programme is designed with the RBA as the gold standard and therefore, within the nPOD database, the AAb‐positive cases are those confirmed by RBA. In order to draw any conclusion about ELISA‐positive RBA‐negative results in diabetes‐free subjects, one would need to compare larger prospective cohorts of subjects with both the ELISA and RBA, in terms of disease prediction.

Comparing AAb results in screening versus recovery samples (separated by only a few days) has revealed that, in a few samples, the titre has either decreased or increased. It is not known definitively whether the management of these individuals in the ICU has resulted in either dilution or concentration of AAb in circulation. Indeed, in two particular cases, the screened sample was positive for GADA in both the ELISA and RBA assay while the recovery sample was found negative by both methods. It is true, however, that most samples were relatively constant in titre between the two samples, but it is important to keep in mind the possible dilution or concentration due to effects of patient management efforts in the hospital. nPOD has implemented a programme that has identified AAb‐positive individuals successfully among the general organ donor population in clinical serology laboratories previously inexperienced in type 1 diabetes screening. This was made possible because of the development of a sufficiently sensitive and specific commercial ELISA that was modified for our purposes for the detection of three relevant AAb. The final configuration of these kits has improved the concordance between screening and core laboratories and specificity of the programme. By implementing a conservative approach, favouring specificity in the ELISA and ultimately having the RBA determine final AAb status, we are confident that the programme is able to identify rare AAb‐positive cases accurately. At the time of writing, the ElisaRSR 3 Screen ICA (Cardiff, Wales, UK) has become newly available for the detection of GADA, IA‐2A, and ZnT8A in a single assay, and further testing of this new application is warranted for potential implementation as part of the nPOD OPO AAb screening programme. As these AAb‐positive donors are thought to be in the pre‐type 1 diabetes phase of disease development, this programme has extended the ability to perform cross‐sectional investigation of the target organ in humans, a largely unexplored resource for understanding the disease pathogenesis. Of note is the observation that an AAb‐positive donor pancreas, in particular multiple AAb, may already contain compromised islets. This facet warrants additional exploration as to whether these organs should be transplanted. At this time, we have insufficient data to make that proclamation definitively, but given the potential clinical implications, it should be the focus of additional studies.

Author contributions

C. W. researched the data and wrote the manuscript, E. M. researched the data and reviewed/edited the manuscript, L. Y. researched the data and reviewed/edited the manuscript, A. M. researched the data and reviewed/edited the manuscript, R. G. researched the data and reviewed/edited the manuscript, A. P. conceived of the study and reviewed/edited the manuscript, C. N. contributed to discussion and reviewed/edited the manuscript, J. S. K. analysed the data and reviewed/edited the manuscript, D. A. S. contributed to discussion and reviewed/edited the manuscript, E. B. conceived of the study and reviewed/edited the manuscript, and M. A. A. conceived of the study and wrote the manuscript. M. A. A. is the guarantor of his work and, as such, takes responsibility for the integrity of the contents herein.

Disclosure

The authors declare that no disclosures exist.

Supporting information

Additional Supporting information may be found in the online version of this article at the publisher's web‐site:

Fig. S1. Receiver operating characteristic (ROC) curve analyses are shown for (a) glutamic acid decarboxylase (GADA), (b) insulinoma‐associated protein‐2 (IA‐2A) and (c) zinc transporter 8 (ZnT8A). Plots, showing only 95–100% specificity, are 100 – specificity % (x‐axis) versus sensitivity % (y‐axis). The following cut‐offs were selected to maximize specificity: (a) 20 IU, 98% specificity and 70% sensitivity [area under the curve (AUC) = 0·88], (b) 60 IU, 99% specificity and 63% sensitivity (AUC = 0·80) and (c) 20 IU, 97% specificity with 57% sensitivity (AUC = 0·75).