figure b

Introduction

Insulin allergy can significantly complicate diabetes management. Although rare, with an estimated prevalence of 0.1–3% [1,2,3], the number of cases of insulin allergy may reach 800,000 out of approximately 90 million insulin-treated patients worldwide.

Clinical manifestations range from cutaneous reactions, which are either immediate (type I; pruritic urticarial papules) [4,5,6] or delayed (type IV; subcutaneous, inflammatory, non-pruritic and inconstantly painful nodules) [7,8,9], to less frequent systemic manifestations, including life-threatening anaphylaxis [10, 11]. Although insulin itself is the main allergen, excipients, the nickel used in needles and the latex used in vial caps and cartridge plungers can also be involved.

Allergen avoidance is challenging in insulin-requiring patients, calling for a rigorous diagnostic and treatment workflow. To this end, we analysed a retrospective cohort to validate clinical criteria that can guide diabetologists about the need for specialised allergology referral. We subsequently performed a case–control study to measure the diagnostic performance of allergology tests.

Methods

Retrospective cohort

Fifty-two consecutive patients were referred from 2000 to 2010 for suspected insulin allergy to our diabetology clinic, which is the regional reference centre for insulin allergy. A non-opposition form signed at hospital admission allows anonymised data from these patients to be used for research purposes.

Clinical assessment in the diabetology clinic

After verifying the use of proper subcutaneous injection techniques, standardised medical records were used to classify patients into three likelihood categories [12] based on four clinical criteria derived from our own experience [13] and general guidelines on evaluating suspected drug allergies [14]:

  1. (1)

    local or systemic reactions suggestive of insulin allergy (i.e. immediate [<15 min] or delayed [>6 h] hypersensitivity reactions);

  2. (2)

    reactions elicited by each injection of the index insulin(s);

  3. (3)

    reactions centred on the injection sites;

  4. (4)

    reactions observed by the physician (i.e. a positive insulin challenge test).

Clinically likely, possible and unlikely insulin allergy were defined by the presence of all, some or none of these four criteria, respectively.

Case–control study

Ten consecutive patients with a likely insulin allergy, 24 insulin-treated non-allergic patients and 21 insulin-naive patients were recruited over 3 years (2014–2016). Exclusion criteria were concomitant anticoagulant therapy (interfering with skin tests), severe renal failure (GFR <30 ml/min) and ongoing pregnancy. The study was completed by 10/10 participants with insulin allergy and 22/24 insulin-treated and 18/21 insulin-naive control participants. The main reason for dropping out of the study was an inability to attend the allergology appointment for skin tests; in one case the participant was undergoing concomitant anticoagulant therapy. All participants were examined by a single diabetologist and by an allergologist. The Ile-de-France II ethics committee approved the study (2009-A00954-53; ClinicalTrials.gov NCT01407640). Participants gave written informed consent.

IDR and skin prick tests

Ten insulin formulations were selected for testing, based on the reported allergenic formulation(s) and their excipients (see electronic supplementary material [ESM] Table 1). As isolated excipients are not available, test panels included combinations to rule out allergies to excipients (e.g. insulin glulisine, which does not contain zinc, to exclude zinc allergy). All participants were examined by the same allergologist in the morning, after a breakfast preceded by their usual insulin injection(s). No participants were on antihistamine drugs or corticoids at the time of investigation.

To measure intradermal reactions (IDRs), insulins (100 U/ml) were injected intradermally in a volume of 20 μl at increasing tenfold concentrations in 0.9% saline (154 mmol/l NaCl), from 1/10,000 (2 × 10−4 U) to 1/10 (2 × 10−1 U) or until a local reaction was observed. Protamine (1000 U/ml; Sanofi, France) was tested in parallel at concentrations from 1/10,000 (2 × 10−4 U) to 1/10 (2 × 10−1 U), because of non-specific reactions at 2 U [15]. The whole procedure included up to 40 intradermal injections of ten formulations and lasted 90–120 min, with a cumulative insulin dose of <3 U under glucose monitoring. Index formulations (i.e. those reported to elicit reactions) were tested last. Skin prick tests were performed with undiluted preparations using the same insulin panel and latex (Stallergenes, France). IDR and skin prick test panels included glycerol saline (Stallergenes) and histamine (only for skin prick tests, 10 mg/ml; Stallergenes) as negative and positive controls, which were negative and positive, respectively, in all cases.

Reactions were measured after 20 min and after 24 and 48 h, when delayed reactions were reported. In the retrospective cohort, reactions were scored as positive if wheal-and-flare responses occurred, that is, both redness ≥10 mm and a papule ≥3 mm in diameter [16]. In the case–control study, reactions were scored as positive if the diameters of papules increased by >5 mm for IDRs and >3 mm for skin prick tests compared with the end of the injection, or if the diameter of the papules was >80% of that seen with the histamine positive control. Following IDR tests, punch skin biopsies were obtained for some insulin allergy participants and stained with haematoxylin–eosin–saffron.

In vitro anti-insulin antibody assays

Serum anti-insulin IgE and IgG4 and anti-protamine IgE were measured using the Pharmacia CAP System FEIA (Phadia/ThermoFisher, Sweden). Anti-insulin IgE levels were further quantified using an in-house radio-binding assay [17] using Streptavidin Sepharose High Performance beads (GE Healthcare, Germany) coated with biotin-labelled mouse anti-human IgE (RRID:AB_396180). Briefly, sera (5 μl) were incubated with 1159 nU 125I-labelled insulin (13.43 MBq/μg; Sanofi) in 25 μl 50 mmol/l Tris and 1% Tween-20 (pH 8; TBT) at 4°C for 72 h before adding 50 μl anti-IgE-coated bead suspension, incubating for 1 h at 4°C under agitation, washing in cold TBT and counting. Non-specific binding was determined using beads coated with biotin-labelled mouse anti-rat IgM (RRID:AB_395115). Results are expressed as nU bound insulin/ml, calculated as [(anti-IgE counts − control anti-IgM counts)/(total counts per tube)] × 1159 × 200. Anti-insulin antibodies were measured as described previously [18].

Statistical analysis

Data were analysed using SAS software v9.4 (SAS Institute, USA) and two-tailed statistical tests, as detailed in the table legends.

Results

Patients with clinically likely insulin allergy have a shorter but variable duration of insulin treatment

Table 1 summarises the clinical presentation of the 52 patients referred for suspected insulin allergy. Patients were first classified based on the presence or absence of the following clinical characteristics: (1) local or systemic reactions suggestive of insulin allergy, that is, immediate (<15 min) hypersensitivity (urticaria, i.e. erythematous, pruritic papules [Fig. 1a–c]; laryngeal angioedema) or delayed (>6 h) hypersensitivity (subcutaneous, inflammatory, non-pruritic and inconstantly painful nodules [Fig. 1d, e]); (2) reactions elicited by each injection of the index insulin(s); (3) reactions centred on the injection sites; and (4) reactions observed by the investigator in response to an insulin challenge test.

Table 1 Clinical presentation and characteristics of the 52 patients referred for insulin allergy
Full size table
Fig. 1
figure 1

Representative allergic skin reactions to insulin. (a) Immediate-type reaction to insulin lispro. (b) Positive IDRs, 20 min readout. (c) Negative IDRs, 20 min readout. (d) Delayed-type reaction to insulin detemir. (e) Positive IDR to insulin detemir, 24 h readout

Full size image

Reactions were categorised as clinically likely insulin allergy when all four criteria were present (26/52 patients, 50%). In this group, reactions were mostly local (19/26, 73%); reactions were systemic in 7/26 (27%) patients (i.e. generalised urticaria/pruritus [n=5] and laryngeal angioedema [n=2; graded as severe anaphylaxis] [19]). Immediate-type reactions were more common (77%) than delayed reactions (23%).

Clinically possible insulin allergy was assigned when some but not all criteria were met (9/52 patients, 17%). All patients in this group reported pruritus centred on the injection sites, which was mostly immediate (78%) but inconsistent and without a skin reaction that could be verified by the physician.

Patients meeting none of the clinical criteria were classified as being clinically unlikely to have an insulin allergy (17/52 patients, 33%). Manifestations were variable and not consistently observed at each injection: either local reactions (5/17, 29%; pruritus or pain in the area of the injection sites, but not centred on the injection sites) or, more frequently, non-specific systemic reactions without local manifestations (11/17, 65%). Non-specific and delayed reactions were more common in this group than in the clinically likely and clinically possible insulin allergy groups.

Overall, patients in the clinically likely and clinically possible insulin allergy categories were considered clinically positive for insulin allergy; those assigned to the clinically unlikely insulin allergy group were considered clinically negative.

The clinical characteristics of the 52 patients are further summarised in Table 1 (additional details are provided in ESM Table 2). There was a higher percentage of women in the clinically unlikely group than in the clinically likely group. BMI was lowest in the clinically possible group. The distributions of age, type 1 and type 2 diabetes, disease duration, HbA1c values and prior history of atopy were similar across the groups. The delay between the first injection of index insulin(s) and clinical manifestations was heterogeneous but was most commonly 3–4 months and was similar across the groups. However, patients with clinically likely insulin allergy had a shorter treatment duration.

Index formulations covered all major types of insulin but were more frequently insulin detemir, protamine-containing insulins and/or multiple insulins, irrespective of clinical likelihood of insulin allergy.

Collectively, the most relevant feature in the clinical history of patients with clinically likely insulin allergy was a shorter yet variable duration of insulin treatment.

Clinical likelihood criteria identify patients with an insulin allergy subsequently confirmed by an IDR test

The IDR results are summarised in Table 2 (further details of the IDR tests are provided in ESM Table 3). Using the IDR test as the gold standard, the sensitivity of the clinical likelihood criteria was 92% (24/26 positive IDR tests) for likely insulin allergy patients, 33% (3/9) for possible insulin allergy patients and 77% (27/35) when considering both together, with a positive predictive value of 100%. The two IDR-negative patients with clinically likely insulin allergy reported delayed local reactions to insulin detemir. The specificity of the clinical likelihood criteria (i.e. IDR-negative patients identified as clinically negative [unlikely insulin allergy]) was 100% (14/14).

Table 2 IDR results and diagnostic performance of clinical likelihood criteria
Full size table

Besides confirming insulin allergy, an IDR test is useful for identifying allergenic formulations. Indeed, all positive IDR tests were positive for index insulin(s) and often also for other formulations. In all but one case (PIA08), the reported reaction type (immediate or delayed) was concordant with the IDR. Most delayed reactions (75%, 3/4) were associated with insulin detemir, as reported previously [9, 13]. Six patients testing positive to protamine also reacted to insulin formulations without protamine, indicating a concomitant allergy to protamine and insulin. No reaction to other excipients was observed. Reactions to the glycerol saline control were negative in all cases, thus ruling out an allergy to nickel used in needles.

Skin prick tests were performed in a subset of 24 patients (11 with likely insulin allergy, nine with possible insulin allergy and four with unlikely insulin allergy [ESM Table 3]). All positive reactions were in the likely insulin allergy group (5/11, 45%). Moreover, all skin prick-positive patients were also IDR positive. Anti-insulin IgE was measured in a subgroup of patients; 12/15 likely insulin allergy patients, 2/4 possible insulin allergy patients and 1/4 unlikely insulin allergy patients were positive. Similar to the skin prick test, all IgE-positive patients were also IDR positive (except for the clinically unlikely insulin allergy patient). Anti-protamine IgE measurements were negative in all patients (data not shown).

Collectively, positive clinical likelihood criteria identified patients with an insulin allergy that is subsequently confirmed by an IDR test with 77% sensitivity and 100% specificity. IDR tests identified immediate- and delayed-type reactions to index insulins and other allergenic insulin formulations, with no added value for skin prick and anti-insulin IgE tests.

Case–control study to define the diagnostic performance of skin prick and anti-insulin IgE tests

The retrospective cohort results indicated that only patients with clinically likely or clinically possible insulin allergy should undergo skin tests to confirm or disprove diagnosis. To measure the diagnostic performance of allergology tests, we conducted a case–control study by recruiting patients with clinically likely insulin allergy in parallel with insulin-naive type 2 diabetic patients and non-allergic insulin-treated type 1 diabetic patients (Table 3). Although diabetes duration was shorter in the insulin allergy group, insulin treatment duration and doses were similar in insulin allergy and insulin-treated participants. In the insulin allergy group, type 2 diabetic participants were treated with slow-acting insulin and oral hypoglycaemic agents (OHAs), except for two participants who were weaned off insulin before study entry. History of atopy was not associated with insulin allergy. Considering that insulin-treated and insulin-naive control participants all had type 1 and type 2 diabetes, respectively, HLA haplotypes were compared between the insulin allergy group and the insulin-treated and insulin-naive groups combined, thus correcting for the potential bias of type 1 diabetes-associated haplotypes. HLA-A2 showed a slightly but significantly higher representation among insulin allergy participants than among the combined group (80% vs 42%, p=0.04).

Table 3 Characteristics of case–control study participants and clinical reactions
Full size table

Table 3 further summarises the timing of clinical reactions in insulin allergy participants (individual participants are described in ESM Table 4). Immediate-type reactions were most common (8/10, 80%). Participants were enrolled after a median time of 10 months (range 1–124 months) following their first reaction. Most (7/10) were still experiencing reactions at study entry (median time since last reaction 0 months, range 0–96 months); 2/10 were switched to OHAs and/or glucagon-like peptide 1 (GLP-1) analogues 2–3 weeks before enrolment. One participant (46-PS) had been on continuous subcutaneous insulin infusion (CSII) for 9 years to treat insulin allergy but still experienced skin reactions 96 weeks before study entry.

Diagnostic performance of IDR and skin prick tests and histopathology

The diagnostic performance of the IDR and skin prick tests was assessed by comparing clinically likely insulin allergy participants with either insulin-treated or insulin-naive control participants (Table 4). Both tests displayed 100% specificity and positive predictive values for both comparisons, but the sensitivity of the IDR test was far superior to that for the skin prick test (80% vs 10% for both comparisons). Negative predictive values were 67–71% for skin prick tests and 90–92% for IDR tests. As before, all positive IDRs were positive for index insulin formulation(s), alone or in combination with others (ESM Table 4). No reaction to excipients or latex was detected.

Table 4 Diagnostic performance of different assays in the case–control study
Full size table

Punch skin biopsies performed at the IDR site in 4/10 insulin allergy participants showed superficial and deep urticarial perivascular lymphocytic infiltrates with some eosinophils (Fig. 2).

Fig. 2
figure 2

Histopathology of a punch skin biopsy at the reaction site in an insulin allergy patient. Representative haematoxylin–eosin–saffron-stained section from participant 28-ET from the site of reaction following allergology tests showing superficial and deep dermal perivascular infiltrates composed of lymphocytes and a few eosinophils. (a) Original magnification ×40, scale bar 1 mm. (b) Original magnification ×100, scale bar 250 μm. (c) Original magnification ×400, scale bar 50 μm. Eosinophils are indicated by arrows

Full size image

Collectively, IDR tests provide the best diagnostic performance (80% sensitivity, 100% specificity) and identify index insulin formulation(s), while prick tests are highly specific (100%) but poorly sensitive (10%).

Diagnostic performance of anti-insulin IgE assays

A commercial anti-insulin IgE fluorimetric assay displayed 86% or 90% specificity, depending on the comparison, and 56% sensitivity (Table 4), with no correlation between anti-insulin IgE and IgG4 levels (data not shown) and no added diagnostic value of IgG4 testing, as reported previously [20]. To improve specificity, we employed an in-house anti-insulin IgE radio-binding assay. With a positive cut-off value of >150 nU/ml, determined after carrying out a receiver operating characteristic analysis (AUC 0.819 [ESM Fig. 1a]), specificity reached 100% and sensitivity was 50% for the comparison between the insulin allergy group and either of the control groups (Table 4). The two anti-insulin IgE assays correlated well (Spearman r=0.818, p=0.011 [ESM Fig. 1b]). Anti-insulin antibodies were positive in all cases. Notably, anti-insulin IgE levels were inversely correlated with the index insulin concentration testing positive in the IDR tests, for both commercial assays (r=−0.839, p=0.033 [ESM Fig. 1c]) and in-house assays (r=−0.772, p=0.043 [ESM Fig. 1d]).

Collectively, the in-house anti-insulin IgE assay achieved 100% specificity and 50% sensitivity for detecting clinically likely insulin allergy participants.

Clinical correlates of IDR and anti-insulin IgE assays in individual participants

IDR tests were negative in two insulin allergy participants (ESM Table 4). One (52-PC) reported immediate-type reactions to insulin detemir and was anti-insulin IgE positive in the in-house assay. The other (46-PS) was enrolled 96 months after her last skin reaction, although her IDR was positive for the index insulin and several others 9 years earlier, at the time of the first reactions and subsequent switch to CSII. Both anti-insulin IgE assays were negative, with high anti-insulin antibody levels. The two participants who were weaned off insulin shortly (2–3 weeks) before study entry (28-ET and 29-HD) still tested positive on the IDR test and both anti-insulin IgE assays.

The case of participant 38-RO is noteworthy. High anti-insulin IgE levels were detected with the commercial assay and, to a larger extent, with the in-house assay. This type 2 diabetic participant was first treated at diagnosis, 10 years before enrolment, with insulin detemir, which was discontinued 2 months later because of the presence of urticarial lesions. Insulin NPH was reintroduced 10 years later because of poor glycaemic control and was discontinued 3 days later on insulin allergy relapse with similar lesions. It was then resumed 6 months later because of persistent hyperglycaemia, followed by another immediate insulin allergy relapse and study entry 3 months later. The very high anti-insulin IgE levels observed likely reflect these three previous insulin challenges.

Therapeutic management

Based on published literature [3, 10, 11, 21, 22], we undertook a stepwise therapeutic approach for patients with positive IDRs to index insulin: 31 patients with clinically likely insulin allergy (23 from the retrospective cohort [one was lost to follow-up] and eight from the case–control study) and three patients with possible insulin allergy, as detailed in ESM Tables 3 and 4 and summarised in Table 5. Mean follow-up duration was 4.7 years (range 0.5–12 years). First, spontaneous resolution was observed in 3/34 (9%) patients. Second, replacement of insulin detemir with OHA/GLP-1 or another insulin formulation resolved symptoms in 4/34 (12%) patients. Third, a re-evaluation of insulin dependency led to a switch to OHA/GLP-1 in 5/34 (15%) patients. Fourth, switch to another insulin formulation resolved insulin allergy in 3/34 (9%) patients. Fifth, antihistamine treatment (cetirizine, desloratadine) was attempted in those with immediate-type reactions and was sufficient in 4/34 (12%) patients in total. The remaining patients were switched to CSII with either aspart or lispro (15/34, 44%), often combined with squared boluses. At CSII initiation, a void subcutaneous catheter was left overnight to exclude reactions to the material. CSII was started at the usual dose, except for three patients with systemic manifestations, who started at a dose of 0.1 U/h, with 0.1 U/h increments every hour until normoglycaemia. In most patients, CSII prandial boluses up to 4 U were well tolerated from the start; otherwise, they were omitted during the first few weeks.

Table 5 Therapeutic management of patients with a positive IDR to index insulin
Full size table

Collectively, insulin allergy resolution can be achieved using a stepwise strategy that includes a switch to OHA/GLP-1 or other insulin formulations, antihistamine treatment and, eventually, CSII.

Discussion

We describe a large retrospective cohort of 52 consecutive patients referred for suspected insulin allergy, complemented by a case–control study of participants with and without insulin allergy. As described previously [5, 23], prior medical history was not associated with insulin allergy: insulin allergy was observed in both type 1 and type 2 diabetes, often in those without an atopic background, and with a variable delay after the introduction of index insulin(s), usually of a few months. While local reactions were more common, systemic reactions were observed in 27% of patients with clinically likely insulin allergy, including two cases of laryngeal angioedema. Non-specific and delayed reactions were more frequent in the clinically unlikely insulin allergy group.

Our first aim was to define whether clinical likelihood criteria can inform the need for further allergology workup at first referral to the diabetology clinic, where it is sometimes difficult or unsafe to perform challenge tests. The presence of all (clinically likely insulin allergy) or only some criteria (clinically possible insulin allergy) displayed 77% sensitivity and 100% specificity when considering an IDR test as the gold standard, suggesting that the criteria can guide decisions on whether to pursue further investigations.

Our second aim was to define the diagnostic performance of the confirmatory tests available on subsequent allergology referral. To this end, the presence of all four clinical likelihood criteria that define the likely insulin allergy group was taken as the most suitable reference, because such criteria include an insulin challenge test, which is the recommended diagnostic gold standard [14]. The IDR test displayed the best diagnostic performance (100% specificity, 80% sensitivity); the skin prick test had a low sensitivity of 10% (100% specificity) [10] and the anti-insulin IgE assays had an intermediate sensitivity of 50% and 56% (100% specificity using the in-house assay) [24]. These latter tests had no added diagnostic value, as all positive participants were also IDR positive. Moreover, IgE assays are relevant only for immediate-type (IgE-mediated) reactions. Indeed, insulin allergy participants with IDR-confirmed delayed reactions were either anti-insulin IgE negative or marginally positive.

IDR tests were also used to identify the main allergen [25], which was insulin in all cases. Multiple index insulins were often reported and tested positive in IDR tests. The single most frequently used formulations were detemir and protamine-containing insulins. Although allergies to protamine, insulin excipients or latex have been described previously [21, 22], our cohort confirms their rarity.

Despite the introduction of general guidelines for evaluating drug allergy [14], there is a lack of robust evidence on insulin allergy based on large series [25], and case–control studies are missing. Based on our results, we propose a systematic diagnostic workflow (Fig. 3, blue). First, patients referred for suspected insulin allergy should be classified based on the four clinical likelihood criteria. In their absence, insulin allergy is unlikely and further investigations are not warranted, as IDR tests to index insulin(s) are invariably negative in this scenario. IDR tests may still be useful in selected cases to avoid unjustified insulin withdrawal.

Fig. 3
figure 3

Proposed diagnostic (blue) and treatment (green) workflow for insulin allergy. Switching to OHA/GLP-1 should be guided by C-peptide measurements. IA, insulin allergy

Full size image

Conversely, the presence of these criteria meets the definition of positive clinical manifestations, with two different degrees of likelihood (likely or possible insulin allergy) depending on whether all or only some criteria are present. The single most important criterion is the observation of a positive reaction after an insulin challenge test, which should be performed whenever possible and, albeit not identifying the allergen(s), is considered diagnostic [14]. All patients with a positive test result should undergo further allergology assessment to establish the diagnosis (in the absence of a challenge test) and to identify allergenic and non-allergenic formulations. Nonetheless, switching to other insulin formulations or OHA/GLP-1 should not be delayed if possible and/or if needed, as both the IDR and IgE assays still yielded positive results 2–3 weeks after insulin withdrawal. Although poorly sensitive, skin prick tests are easier to perform and should be used for screening, because a positive result means that an IDR test is unnecessary, thus shortening the diagnostic workup and limiting the risks of systemic reactions. Anti-insulin IgE assays are not recommended for diagnosis [14, 25], but could be relevant to monitor insulin allergy desensitisation. Haastrup et al [26] screened 144 patients with suspected insulin allergy using the Phadia anti-insulin IgE assay, with only 34 (24%) testing positive, of whom 33 tested positive on a subsequent IDR test. Interestingly, 12/71 (17%) IgE-negative patients were also IDR positive. The 54% lower insulin allergy prevalence found in this study compared with our study (27/52 [52%] with IDR-confirmed insulin allergy) may reflect the sensitivity of the Phadia IgE assay (56%) and the fact that delayed (non-IgE-mediated) reactions are not detected by IgE assays. Moreover, our clinical likelihood criteria allow one third of patients to be excluded from further workup rather than testing for anti-insulin IgE in all patients.

Insulin detemir is associated with higher frequencies of cutaneous, mostly delayed, reactions [5, 9, 13]. Delayed reactions with other formulations have been described [27,28,29] but are rare, including with degludec (and liraglutide), despite it having a similar 16-carbon fatty di-acid chain. The crystallised structure of degludec (possibly masking some antigenic determinants) and slower release kinetics may explain its lower allergenicity [30]. Detemir reactions can be resolved by replacement with another insulin without the need for skin tests in the absence of relapse.

Guidelines can also be proposed for the subsequent management of IDR-confirmed insulin allergy cases (Fig. 3, green). First, replacement of insulin with OHA/GLP-1 should be evaluated in C-peptide-positive type 2 diabetic patients. Second, replacement with another insulin formulation should be attempted. As insulin itself was the allergen in our study and multiple index insulin formulations were often identified, replacement with another, preferably IDR-negative, insulin [25, 26] was seldom possible. Antihistamine agents (or topical sodium cromoglicate) should therefore be considered as third-line treatment for immediate-type reactions. Although its efficacy has been described previously [31], we did not consider local corticoid treatment [25] because its vasoconstrictive effects [32] can reduce systemic insulin bioavailability. As reported previously [26], most patients (44%) ultimately required CSII, which should be the first-line treatment for anaphylactic reactions. CSII benefit is twofold [3, 10]. First, diluted insulin exposure limits allergenicity, especially with prolonged (30–60 min) prandial boluses. Second, although other protocols have been described [10, 22, 25], CSII favours insulin desensitisation [13] by providing an escalating, continuous low-dose antigen exposure [33]. Although intravenous insulin infusion [34] or other exceptional procedures [35, 36] were not required in our study, CSII did not resolve insulin allergy in one patient. Another three patients were not able to tolerate boluses higher than 4 U and required additional measures (e.g. low carbohydrate/glycaemic index diets, alpha-glucosidase or sodium–glucose cotransporter 2 inhibitors, bolus fractionation).

Our study has some limitations. First, the highest insulin concentration used for the IDR tests was 0.2 U, as recommended previously [25] to limit hypoglycaemia risk, which might explain the negative results in 4/36 (11%) patients with clinically likely insulin allergy (with two displaying delayed reactions to detemir). Of note, these clinically likely insulin allergy patients may have been correctly classified because they underwent a challenge test, typically with a higher dose of insulin of 1–2 U. Nonetheless, another IDR test with a higher, undiluted 2 U dose of insulin taken with carbohydrates should be carried out in cases of negative IDR tests. Second, the IDR sensitivity of 80% was probably underestimated by the omission of this further testing and by inclusion of one patient already under CSII who experienced her last clinical reaction 96 months before enrolment. Another IDR-negative patient (52-PC) reported immediate-type reactions to insulin detemir. Although reactions to detemir are more frequently delayed, the allergic nature of detemir reactions has been questioned [37], which justifies the proposed first-line replacement trial without allergology referral. Third, although allergies to excipients (e.g. zinc, metacresol) were excluded with the proposed test panel, latex allergy may have been associated with insulin allergy but was not tested in all patients. The diagnostic workup should therefore include a latex skin prick test. Fourth, the number of patients analysed is small because of the rarity of insulin allergy, and the proposed workflow requires validation in larger studies.

The short time frame available to treat insulin allergy patients without compromising glycaemic control and the frequent ultimate resort to CSII for patients not otherwise requiring it underline the need for novel therapeutics (e.g. omalizumab) [35]. Testing of such strategies would be facilitated by the short follow-up needed to assess clinical outcomes. It may be beneficial to consider insulin allergy as a proof-of-concept setting for strategies to restore immune tolerance to insulin for preventing type 1 diabetes [38]. Of note, allergy desensitisation protocols are being considered to this end [39, 40]. While prevention trials require large screening efforts to stratify disease risk and years of follow-up to assess outcomes, trialling the same therapeutics in insulin allergy patients would require only a few months of follow-up. On the one hand, this could provide treatment options for an otherwise rare condition that is not attractive for dedicated drug development. On the other hand, these trials could allow effective agents to be repurposed for type 1 diabetes prevention.

In conclusion, clinical likelihood criteria can be used by diabetologists to guide insulin allergy diagnosis and decisions about the need for specialised skin tests. A stepwise management strategy has been proposed. CSII is eventually required in most patients, underlining the need to develop alternative desensitisation strategies.