Type 1 and type 2 diabetes share a name and a blood test, and almost nothing else. The confusion this causes is not trivial — it shapes what people assume about cause, blame and treatment — so it is worth being precise at the outset.
Type 1 diabetes is an autoimmune disease. The immune system destroys the beta cells of the pancreatic islets, the only cells in the body that make insulin. As they are lost, insulin production falls and eventually stops. Without insulin, cells cannot take up glucose from the blood; glucose accumulates while the body, unable to use it, starts burning fat instead and generating acidic ketones. Untreated, this is fatal within weeks. Insulin replacement is not one option among several — it is the only thing keeping the person alive, and it is needed for life.
Type 2 diabetes is a different problem. There, insulin is being made but the body responds to it poorly, and beta-cell function declines against that background. It is treatable with drugs that improve insulin sensitivity or secretion, and often with weight loss. None of that applies to type 1, where there is simply no insulin to sensitise anyone to.
This distinction is why the two need separate reviews. The glycaemic-control machinery — HbA1c, glucose targets, complications — is shared, and is covered in the companion review. What is specific to type 1 is the autoimmunity, the total insulin dependence, the 24-hour self-management burden, and one thing that is genuinely new: a therapy that delays the onset of the disease before it becomes clinical.
It usually starts young, but not always. In United States youth, overall diabetes incidence was 24.3 per 100,000 person-years, with type 1 predominating in everyone under 10 regardless of ethnicity and rates in non-Hispanic white youth of 18.6, 28.1 and 32.9 per 100,000 at ages 0–4, 5–9 and 10–14 [1]; childhood incidence across Europe has been rising [2]. But adults develop it too, and are frequently misclassified as type 2 — a presentation named latent autoimmune diabetes in adults. Among 65 people with diabetes presenting after age 30, 73.7 percent of those who came to require insulin were anti-GAD positive, against 4.3 percent of those who did not [3].
It is relentless in a way that is easy to underestimate. Every meal requires an estimate of carbohydrate and a dose calculation; every episode of exercise, illness or sleep changes insulin requirements; and the two errors have opposite and immediate consequences — too little insulin risks ketoacidosis, too much risks hypoglycaemia. In the T1D Exchange registry, HbA1c plateaued at 7.5–7.8 percent across adulthood [4], and among 13,316 youth, the American Diabetes Association HbA1c target was met by 64 percent of under-6s, 43 percent of 6–13-year-olds and just 21 percent of 13–20-year-olds [5]. That is not a failure of effort. It is the difficulty of manually replacing a control system that normally runs itself.
And the disease begins long before the diagnosis — which is the subject of the centerpiece, and the reason this review is more optimistic than one written twenty years ago would have been.
Centerpiece: a disease you can see coming, and now delay
The insight that reorganised this field is that type 1 diabetes is not an event. It is a process that progresses through identifiable stages, and the immune system announces itself years before symptoms do. A joint scientific statement of JDRF, the Endocrine Society and the American Diabetes Association formalised the taxonomy: stage 1 is two or more islet autoantibodies with normal glucose, stage 2 is beta-cell autoimmunity with dysglycaemia, and stage 3 is symptomatic disease. Stages 1 and 2 are presymptomatic [6].
Note what that classification asserts. Stage 1 is not a risk factor for type 1 diabetes. It is type 1 diabetes, at a stage before the person notices.
The evidence that justifies it is the left panel. Pooling prospective cohorts from Colorado, Finland and Germany that followed children at genetic risk from before seroconversion, progression to clinical disease at ten years after seroconversion was 69.7 percent (95% CI 65.1–74.3) in 585 children with two or more islet autoantibodies, 14.5 percent (10.3–18.7) in 474 children with a single autoantibody, and 0.4 percent (0.2–0.6) by age 15 in children with none [7].
The check the arithmetic was never given. The paper prints those three numbers and never divides them. Going from no autoantibodies to one multiplies ten-year risk by 36; going from one to two multiplies it by a further 4.8; the two steps together are a 174-fold difference. Very few clinical markers stratify risk across two orders of magnitude, and the practical consequence is that the count matters more than the identity — which is exactly what the prevention trials found independently, where autoantibody number was highly predictive while insulin autoantibodies alone predicted nothing (no participant with insulin autoantibodies as their single antibody developed diabetes) [8].
Progression is faster with seroconversion before age 3 (74.9 versus 60.9 percent), with HLA DR3/DR4-DQ8 (76.6 versus 66.2 percent) and in girls (74.8 versus 65.7 percent) [7].
The right panel is what you can now do about it. TN-10 randomised 76 non-diabetic relatives at high risk — 44 to a single 14-day course of teplizumab, an Fc-receptor-non-binding anti-CD3 monoclonal antibody, and 32 to placebo, with progression tracked by oral glucose-tolerance tests every 6 months. Median time to diagnosis was 48.4 months with teplizumab against 24.4 months with placebo; diabetes was diagnosed in 19 of 44 (43 percent) against 23 of 32 (72 percent); the hazard ratio was 0.41 (95% CI 0.22–0.78, P = 0.006); and the annualised diagnosis rates were 14.9 percent per year against 35.9 percent [9].
An internal consistency check the paper never performs. It reports an adjusted Cox hazard ratio and, separately, two annualised rates. Dividing the rates gives 14.9 / 35.9 = 0.415, reproducing the reported hazard ratio of 0.41 to two decimal places. Two analyses, one number — which is reassuring about both.
And a second check that says something clinical. If risk accrued at a constant rate, the median time to diagnosis would be ln(2) divided by that rate. For placebo this predicts 23.2 months against the 24.4 reported — agreement within 5 percent, so untreated progression through this window really does behave like a constant hazard. For teplizumab the same identity predicts 55.8 months against the 48.4 observed: the real median arrives about 13 percent earlier than a constant hazard would give. The same discrepancy appears from the other direction — the ratio of medians is 1.98, short of the 2.41 the hazard ratio implies.
The natural reading is that a single 14-day course does not hold the hazard down indefinitely; the protection wanes. That is not a criticism of the result. It is a specific, quantitative statement about what one course buys, and it is the argument for studying redosing.
Why this matters beyond the delay itself. Two years of normal childhood without insulin injections is worth having on its own. But the deeper significance is that it converts type 1 diabetes from a disease you can only manage into one you can, partially, intervene in before it arrives — which requires that people be screened for autoantibodies while still well. Teplizumab also preserves C-peptide when given at new onset [10] [11], and the approach descends directly from the observation that anti-CD3 antibody induces long-term remission of overt autoimmunity in the non-obese diabetic mouse [12].
Two honest limits. TN-10 randomised 76 people, 72 percent of them 18 or under [9]; this is a small phase 2 trial in relatives of patients, not a population-screening study. And the delay is a delay: in the teplizumab arm, 43 percent were still diagnosed during follow-up [9].
The teaching point. Type 1 diabetes develops through a measurable presymptomatic autoimmune phase. Autoantibody count stratifies ten-year risk 174-fold, staging turns that into a diagnosis, and there is now a therapy that delays the clinical endpoint — which together move the field from management toward prevention.
Pillar 1: measurement and diagnosis
Telling type 1 from type 2
The distinction is usually obvious in a thin child with weeks of thirst, weight loss and ketones. It is frequently not obvious in an adult, and getting it wrong means someone with an absolute insulin deficiency is treated as though they have a relative one.
Islet autoantibodies are the marker of the autoimmune process. The four in routine use are against glutamic acid decarboxylase 65 (GAD65), insulinoma-associated antigen 2 (IA-2), zinc transporter 8 (ZnT8) and insulin itself [13] [14]. ZnT8 was identified as a major autoantigen later than the others and adds predictive information in relatives already positive for the rest [14] [15] [16] [17]. GAD antibodies were shown to discriminate the major types of diabetes early — present in 69 percent of short-duration and 59 percent of long-duration insulin-dependent diabetes [18] — and combinations outperform any single marker [13] [19] [20]. Because these assays underpin both diagnosis and trial eligibility, they have been formally harmonised and standardised across laboratories [21] [22].
C-peptide measures what is left of the person's own insulin. It is cleaved from proinsulin in equimolar amounts with insulin, so unlike insulin itself it is not confounded by injected insulin — which is precisely why it is the test used [23]. Low or absent C-peptide in someone with diabetes and autoantibodies is the biochemical signature of type 1.
The acute presentation is the third element: hyperglycaemia, weight loss and, in a substantial minority, diabetic ketoacidosis — insulin deficiency severe enough that fat breakdown generates enough ketones to acidify the blood. It remains common. In T1D Exchange adults, DKA frequency rose with HbA1c, with 21.0 percent of those at HbA1c 10 percent or above having an event in the past 12 months [24], and international consensus exists on its risk management — particularly relevant since SGLT inhibitors used as insulin adjuncts can produce euglycaemic ketoacidosis, where near-normal glucose masks the diagnosis [25] [26] [27] [28].
The staged, presymptomatic disease and how it is detected
Beyond the staging framework itself [6], the practical question is who to screen and when. In the TrialNet natural history study, the risk of developing islet autoantibodies declined with age — 11 percent lower per year of age for insulin autoantibodies — with cumulative seroconversion of 2 percent in those under 10 against 0.7 percent in those 10 and over, supporting annual screening in young children plus one screen in adolescence [29]. Autoantibodies appear early in life in the offspring of affected parents [30] [31] [32], and the large birth-cohort studies built to observe this — TEDDY, DAISY and BABYDIAB — define the field's evidence base [33] [34] [35] [36] [37].
Autoantibody character, not just presence, carries information. High-affinity insulin autoantibodies — on average a hundredfold higher affinity in children who went on to multiple antibodies than in those who did not — associate with HLA DRB1*04, early appearance and progression [38].
And glucose starts drifting before it crosses any diagnostic threshold. In children with multiple autoantibodies, a 10 percent rise in HbA1c between samples taken 3–12 months apart predicted diagnosis (hazard ratio 5.7) at a median 1.1 years later; two consecutive samples at or above 5.9 percent gave a median 0.9 years to diagnosis (hazard ratio 11.9) [39]. In a separate cohort, rising but still-normal HbA1c predicted progression with a hazard ratio of 4.8 per 0.4 percent, and did so independently of random glucose, which was only marginally predictive [40]. A value inside the normal range, moving, means more than a single value near the top of it.
Day-to-day monitoring, and the shift from HbA1c to time in range
HbA1c integrates glucose over roughly three months and is the measure against which the complication evidence was built. Its limitation is that it is an average: two people with identical HbA1c can differ enormously in how much time they spend hypoglycaemic or severely high.
Continuous glucose monitoring replaced the single number with a distribution, and time in range — conventionally the proportion of the day between 70 and 180 mg/dL — with it. The randomised evidence in type 1 diabetes is strong and consistent. In DIAMOND, 158 adults on multiple daily injections with baseline HbA1c 8.6 percent had an adjusted treatment-group difference of −0.6 percentage points at 24 weeks, with median time below 70 mg/dL of 43 versus 80 minutes per day [41]. In GOLD, a 161-patient crossover trial, HbA1c was 7.92 versus 8.35 percent [42]. And in the JDRF hypoglycaemia trial, time below 70 fell from 0.97 to 0.48 hours per day (ratio 0.49) while time in the 70–180 range rose from 16.0 to 17.6 hours per day [43] [44] [45] [46]. CGM detects hypoglycaemia that nobody noticed [47], works in pregnancy [48], and can largely replace routine fingersticks [49]. Optimal sampling duration for estimating long-term control has been quantified [50].
Pillar 2: management
Insulin, which is the foundation and cannot be omitted
Modern therapy is basal–bolus: a long-acting insulin to cover background requirements, plus rapid-acting doses matched to carbohydrate at each meal and to correct highs. It is delivered either by multiple daily injections or by continuous subcutaneous insulin infusion — an insulin pump. A Cochrane review of 23 trials and 976 participants found a HbA1c difference of −0.3 percentage points favouring pumps, with severe hypoglycaemia appearing reduced and quality of life preferred [51] [52] [53]. Insulin analogues, and ultra-long-acting basal insulins such as degludec, have improved the pharmacology and reduced nocturnal hypoglycaemia relative to older preparations [54] [55] [56], with once-weekly basal insulins now in trials [57] [58].
The evidence that tight control matters is the DCCT, and it is among the most consequential trials in medicine. Intensive therapy reduced development of retinopathy by 76 percent, slowed its progression by 54 percent, and reduced microalbuminuria by 39 percent, albuminuria by 54 percent and clinical neuropathy by 60 percent [59]. Its benefits persisted for years after the trial ended — the phenomenon of metabolic memory — across retinopathy and nephropathy [60] [61] [62] [63] [64], extended to cardiovascular outcomes [65] [66], and finally to survival: over a mean 27 years, 107 deaths, 64 in the conventional and 43 in the intensive group, hazard ratio 0.67 (0.46–0.99) [67]. The relationship between glycaemic exposure and retinopathy risk was quantified directly [68].
The DCCT also named the price, and it is the central tension of this disease: intensive therapy roughly tripled severe hypoglycaemia [59] [69].
Hypoglycaemia, the constraint on everything else
Hypoglycaemia is what makes tight control hard, and repeated episodes make it harder still. A single episode of hypoglycaemia blunts the neuroendocrine and symptomatic responses to the next one [70] [71], producing impaired awareness of hypoglycaemia — a state in which warning symptoms are lost. Prospectively, impaired awareness carried a sixfold increase in the frequency of severe hypoglycaemia, much of it at home during waking hours [72] [73]. The syndrome is reversible: meticulous avoidance of hypoglycaemia restores the thresholds [71] [74]. The broader framework of hypoglycaemia-associated autonomic failure organises these observations [75] [76] [77].
In T1D Exchange adults, severe hypoglycaemia was most frequent in those with 40 or more years of diabetes and, notably, was higher at both low and high HbA1c — leading the authors to suggest modifying targets particularly in patients with very low HbA1c [24].
Automated insulin delivery
This is the technology that changed daily life most. A hybrid closed-loop system links a CGM to a pump through an algorithm that adjusts basal insulin automatically, while the user still announces meals — hence "hybrid".
The randomised results are consistent. In a six-month multicentre trial, time in range rose from 61 to 71 percent in the closed-loop group and was unchanged at 59 percent in controls (adjusted difference 11 percentage points, 95% CI 9–14), with time below 70 mg/dL lower by 0.88 points and HbA1c by 0.33 points [78]. In children, time in range rose from 53 to 67 percent against 51 to 55 percent — again 11 percentage points, equivalent to 2.6 additional hours per day in range — with no episodes of ketoacidosis or severe hypoglycaemia in either group [79]. In-home hybrid closed-loop use lowered HbA1c from 7.7 to 7.1 percent in adolescents and 7.3 to 6.8 percent in adults [80] [81], and the pattern replicates across systems, settings and pregnancy [82], [83] [84] [85] [86] [87] [88] [89], with agreed outcome measures for such trials [90]. Earlier steps in the same direction — sensor-augmented pump therapy and threshold-suspend features — showed the incremental gains [91] [92] [93].
Real-world data now match the trials. In an NHS study of children and young people across three systems, 12 months of hybrid closed-loop use produced an average HbA1c reduction of 7 mmol/mol, a 13.4 percentage-point increase in time in range and a 50 percent reduction in hypoglycaemia frequency, with improvements in hypoglycaemia fear and sleep quality for patients and their parents [94] [95].
Education, which is not an add-on
Structured training in carbohydrate counting and flexible dosing has effects comparable to technology. DAFNE improved glycaemic control and quality of life without worsening severe hypoglycaemia [96] [97]; in routine practice at one year, HbA1c fell by 0.44 percentage points from a baseline above 8.5 percent, severe hypoglycaemia fell from 1.7 to 0.6 episodes per person per year, hypoglycaemia awareness improved in 43 percent of those reporting unawareness, and psychological distress fell significantly [98] [99] [100]. Across a larger cohort, structured training was associated with a 61 percent reduction in ketoacidosis risk and a 64 percent fall in emergency treatment costs for DKA and severe hypoglycaemia [101]. Adherence and family context are part of outcome, not separate from it [102] [103].
Screening for the other autoimmune diseases
Type 1 diabetes clusters with other organ-specific autoimmunity, so screening is routine rather than opportunistic.
Celiac disease runs in both directions: children with celiac disease had a hazard ratio of 2.4 (1.9–3.0) for subsequent type 1 diabetes, and also 2.3 for ketoacidosis or diabetic coma [104], while non-diabetic celiac patients show an increased prevalence of islet autoantibodies [105] [106]. The interaction is practical as well as immunological — a gluten-free diet changes carbohydrate intake and therefore insulin requirements [107] [108].
Thyroid autoimmunity is the other main association; postpartum thyroid dysfunction occurred in 10 of 40 women with type 1 diabetes (25 percent, 95% CI 12.7–41.2), prompting routine postpartum thyroid screening in this group [109].
Complications, and where type 1 differs
The microvascular complications — retinopathy, nephropathy, neuropathy — are shared with type 2 and are the outcomes the DCCT moved. Their prevalence rises with duration [110] [111], the renal course of microalbuminuria has been followed long-term [112], and chronic kidney disease predicts all-cause mortality in this population [113]. Cardiovascular risk is substantially elevated [114] [115] [116]. One instructive comparison: among adolescents, retinopathy was more common in type 1 (20 versus 4 percent) while microalbuminuria and hypertension were markedly less common (6 and 16 percent versus 28 and 36 percent), despite the type 2 group having shorter duration and lower HbA1c [117]. The two diseases damage the same organs at different rates and by partly different routes.
Pillar 3: progress
Automated delivery is still improving, with tubeless on-body systems, customisable targets and fully closed-loop and bihormonal designs under study [84] [118] [89].
Prevention and disease modification is the most consequential frontier, and teplizumab is the proof of concept [9]. It sits in a long line, most of which failed or fell short: cyclosporine produced remission at the cost of continuous immunosuppression [119]; oral and nasal insulin did not prevent disease overall [120] [121] [122] [123]; nicotinamide did not [124]; rituximab [125] [126], abatacept [127] [128] and GAD65 antigen therapy [129] preserved C-peptide transiently without altering the trajectory. Abatacept has since been tested for delaying progression in stage 1 relatives [130], and risk scores exist to identify the preclinical stage for trial entry [131]. The honest summary is that one intervention has delayed onset, several have slowed C-peptide loss for a time, and none has stopped the disease.
Beta-cell replacement has a real but constrained history. The Edmonton protocol achieved insulin independence in seven of seven patients, though all required islets from two donor pancreases [132], and outcomes improved over the following decade [133] [134] [135] [136]. A phase 3 trial in people with severe hypoglycaemia met its primary endpoint — HbA1c below 7 percent and freedom from severe hypoglycaemic events — in 87.5 percent at one year and 71 percent at two, with median HbA1c 5.6 percent and restored hypoglycaemia awareness, against 10.4 percent experiencing bleeds requiring transfusion and a significant fall in glomerular filtration rate on immunosuppression [137] [138]. The constraints have always been donor supply and lifelong immunosuppression.
Stem-cell-derived islets address the first constraint. Functional human pancreatic beta cells were generated in vitro [139], and the clinical translation is now reporting. Device-encapsulated stem-cell-derived cells, which would address the immunosuppression problem too, remain limited: with optimised devices and higher cell doses, 3 of 10 patients with undetectable baseline C-peptide reached levels of 0.1 nmol/L or more from month 6, correlating with improved CGM measures and reduced insulin dosing [140] [141] [142]. Unencapsulated, fully differentiated islets with immunosuppression have gone considerably further: among 12 participants followed at least 12 months, all were free of severe hypoglycaemic events with HbA1c below 7 percent and more than 70 percent time in range, and 10 of 12 (83 percent) were insulin independent — against two deaths in the trial, from cryptococcal meningitis and from progression of pre-existing neurocognitive impairment, with neutropenia the commonest serious adverse event [143] [144] [145] [146] [147]. Autologous haematopoietic stem-cell transplantation at new onset produced insulin independence in some patients but with the risks of the conditioning regimen [148] [149].
That contrast is the field in one paragraph: remove the immunosuppression and the cells underperform; keep it and they work, at a price that only severe disease currently justifies.
Screening programmes are the logical consequence of everything in the centerpiece. Autoantibody screening identifies stage 1 disease, HbA1c trajectory refines the timeline [39] [40], and there is now something to offer. The evidence base for who and when is in place [29] [7] [150].
A note on what this substrate does not contain. Adjunctive SGLT inhibitors appear here mainly through their ketoacidosis risk [25] [28], and there is little on smart insulin pens or on glucose-responsive "smart" insulins, both of which the field is actively pursuing. This review does not characterise them.
Dig deeper in lmmol
Diabetes and glycemic control is the companion review and covers the measurement machinery the two diseases share — HbA1c, glucose targets, the complications that follow from chronic hyperglycaemia — which this review deliberately does not duplicate. Read together, the pair make the central distinction concrete: type 2 is a disease of insulin action, type 1 of insulin absence, and the DCCT is where the case for tight control in type 1 specifically was established [59] [67]. Celiac disease is the autoimmune association with the strongest bidirectional evidence — a hazard ratio of 2.4 for subsequent type 1 diabetes after childhood celiac disease [104], raised islet autoantibodies in non-diabetic celiac patients [105], and a gluten-free diet that changes insulin requirements [107]. Thyroid disease is the other, with postpartum thyroid dysfunction in a quarter of women with type 1 diabetes [109]. Chronic kidney disease is where diabetic nephropathy leads; the DCCT reduced microalbuminuria by 39 percent and albuminuria by 54 percent [59], microalbuminuria's long-term renal course has been followed directly [112], and chronic kidney disease predicts all-cause mortality in this population [113]. Coronary artery disease is the macrovascular endpoint, elevated in type 1 diabetes [115] [114] and reduced by intensive glycaemic treatment in DCCT/EDIC [65]. The full collection is at health.