Celiac disease is an immune reaction to gluten — the storage protein of wheat, and its relatives in rye and barley — in genetically susceptible people. The reaction damages the lining of the small intestine, flattening the villi that absorb nutrients, and the consequences run from malabsorption and anaemia to bone disease, and to a long list of effects outside the gut.
It is not an allergy, and it is not an intolerance. Wheat allergy is a distinct, IgE-mediated reaction to ingested wheat [1], and the consensus nomenclature separates the three routes explicitly: allergic (wheat allergy), autoimmune (celiac disease, dermatitis herpetiformis, gluten ataxia) and possibly immune-mediated (gluten sensitivity) [2]. Celiac disease is an autoimmune disease: the immune system, having been shown a food protein, attacks one of the body's own enzymes. Gluten is the trigger, but the tissue being destroyed is the patient's own — which is why the damage continues as long as the trigger is present, and why the condition carries the same associations as other autoimmune diseases.
It is common. A nationally representative US survey of 7,798 people put prevalence at 0.71 percent (1 in 141), rising to 1.01 percent among non-Hispanic whites — and of the 35 people found, 29 were unaware of the diagnosis [3]. A large earlier US screening study of 13,145 subjects found 1 in 133 in not-at-risk individuals [4]. Prevalence has been rising over time [5], and the World Gastroenterology Organisation puts adult prevalence between roughly 1 in 100 and 1 in 300 in most of the world [6].
Most cases are undiagnosed, and the reason is that the classical picture — a wasted child with diarrhoea and a distended abdomen — is the minority presentation. Many patients have few symptoms or atypical ones [6], and the terminology has had to be standardised twice to keep up [7] [2]. Screening-detected celiac disease is, on closer questioning, usually not silent — simply unrecognised [8].
The treatment is a strict lifelong gluten-free diet, and it works. That combination — a known trigger, an effective treatment, and most patients undiagnosed — is what makes the diagnostic question in this review the central one.
Start here: why gluten, and why these people
The peptide that cannot be digested
Gluten is unusual as a food protein: glutamine is its most abundant amino acid and proline its second [9], and because gluten proteins are proline-rich they resist enzymatic digestion in the gastrointestinal tract — a property that probably contributes to their immunogenicity [10]. In 2002 a specific 33-amino-acid peptide from α-gliadin was identified with a remarkable set of properties: it is stable against all gastric, pancreatic and intestinal brush-border proteases; it reacts with tissue transglutaminase far more selectively than that enzyme's known natural substrates; it induced gut-derived T-cell lines from 14 of 14 celiac patients tested; and homologues of it are present in every food grain toxic to celiac patients and absent from every non-toxic one [11].
That last clause is the part worth pausing on. The peptide's distribution across grains matches the clinical toxicity of those grains exactly. It is about as close to identifying the molecular culprit as a disease of this kind gets.
Transglutaminase, and why the antibody exists
Tissue transglutaminase was identified as the autoantigen of celiac disease in 1997 [12]. Transglutaminases are cross-linking enzymes with a wide biological role [13] [14]; in the gut, tissue transglutaminase generates T-cell-stimulatory gluten peptides by deamidating glutamine — and only particular glutamine residues are modified, with the spacing between glutamine and proline determining which [9]. That selective deamidation strongly enhances gliadin-specific T-cell reactivity [15].
That modification is what completes the mechanism. The deamidated peptides bind far better to the HLA-DQ2 and HLA-DQ8 molecules on antigen-presenting cells, which display them to CD4 T cells [16] [17]; the adult intestinal T-cell response to α-gliadin focuses on a single deamidated glutamine targeted by transglutaminase [18]. Gliadin-specific T cells restricted by DQ2 were isolated from the small-intestinal mucosa of patients [19], the binding rules for gliadin epitopes on DQ2 have been refined [20], and the equivalent DQ8-restricted peptide was characterised [21] [22]. The autoantibody against transglutaminase is a by-product of that T-cell response — which is precisely why measuring it works as a test.
Genetics
HLA-DQ2 and DQ8 are necessary but not sufficient [23]. They are found in virtually all patients — close to 100 percent sensitivity — and also in 30 to 40 percent of the European population, the great majority of whom will never develop celiac disease [6]. That asymmetry is exactly why HLA typing rules the diagnosis out well and rules it in barely at all. Genome-wide association studies have added many further risk variants influencing immune gene expression [24] [25], with substantial overlap with type 1 diabetes [26] — shared and distinct variants across the two conditions, which is the genetic basis for their clinical co-occurrence. Environmental triggers have been investigated, including early-childhood rotavirus infection [27].
Pillar 1: measurement and diagnosis
Serology first
The first-line test is IgA antibody against tissue transglutaminase (tTG-IgA), and in a paediatric comparison against histology it was the best available single index — sensitivity 92.5 percent, specificity 97.6 percent [28]. Antibodies against deamidated gliadin peptides and endomysial antibodies are the other members of the panel [29] [30] [31].
Two practical traps matter more than the assay details.
The first is IgA deficiency. tTG-IgA cannot be positive in someone who makes no IgA. In one audit of 9,533 patients tested for endomysial antibodies, only 4,698 (49 percent) were also tested for IgA deficiency; deficiency was found in 35 of those 4,698 (0.75 percent), and only 19 of the 35 (54 percent) were then managed appropriately [32]. This is why a total IgA is ordered alongside the tTG, and why an IgG-based test is used when IgA is absent — in one study IgG anti-gliadin correctly identified all three children with celiac disease and total IgA deficiency whose IgA-based tests were negative [28]. Whether total IgA must be measured in everyone has been debated [33].
The second trap is the one patients most often fall into: the tests only work while you are eating gluten. A gluten-free diet normalises the antibodies and heals the mucosa, so someone who removes gluten before being tested may become untestable for months. This is not a technicality — antibody titres in treated patients fall below the cutoff over the following year [28], and a formal gluten-challenge study showed measurable histological damage returning within 14 days of resuming gluten, with antibody titres rising markedly by day 28 [34]. The practical rule is to test first and change the diet afterwards.
Biopsy, and when it can be skipped
Confirmation is by duodenal biopsy showing the characteristic changes — increased intraepithelial lymphocytes, crypt hyperplasia and villous atrophy, graded on the Marsh classification [35]. The European diagnostic criteria were formalised in 1990 [36] and revised as the antibody tests improved.
Two caveats about the biopsy itself. Villous atrophy in adults can be patchy, so multiple samples are needed and a single negative biopsy is not conclusive [37]; and lesser degrees of change — raised intraepithelial lymphocytes at the villous tip without atrophy — are hard to interpret [38].
The no-biopsy pathway. The 2011 ESPGHAN guidelines introduced an option that has changed paediatric practice: in a child with symptoms suggestive of celiac disease and tTG-IgA above ten times the upper limit of normal, the diagnosis may be made without duodenal biopsy, under a strict protocol with further laboratory tests including endomysial antibodies and HLA typing [39]. Later European guidance retains this for a defined group of children [40]. The logic is the subject of the figure below, and it is entirely about predictive value.
HLA typing is useful mainly in the negative direction: celiac disease is unlikely in someone with neither DQ2 nor DQ8 [39]. It is a good rule-out test and a poor rule-in one, for the same reason the antibody is a good rule-out test in low-prevalence settings.
What celiac disease is not
Wheat allergy is a distinct IgE-mediated condition, and the amounts of gluten that are safe differ between the two [41]. Non-celiac gluten sensitivity is the harder category. A double-blind, randomised, placebo-controlled rechallenge trial in patients with irritable bowel syndrome who had celiac disease excluded and were symptomatically controlled on a gluten-free diet found that gluten did induce symptoms: of 19 patients rechallenged with gluten, 13 (68 percent) reported that symptoms were not adequately controlled [42]. The mucosal immunology differs from celiac disease — IL-17 expression separates the two gliadin-induced disorders [43] — and the nomenclature has been formalised [2] [44]. Gluten-related neurological syndromes, including cerebellar ataxia, have their own literature and their own serological marker in transglutaminase 6 antibodies [45] [46] [47].
Centerpiece: what a positive antibody actually means
Sensitivity and specificity are properties of an assay. Predictive value is a property of the assay and the population together, and celiac disease is an unusually clean illustration of why the difference matters.
The inputs are all measured, in separate studies. tTG-IgA: sensitivity 92.5 percent, specificity 97.6 percent, in 161 children with biopsy-confirmed disease and 129 controls [28]. Prevalence: 1 in 22 in first-degree relatives, 1 in 39 in second-degree relatives, 1 in 56 in symptomatic patients and 1 in 133 in not-at-risk individuals, from 13,145 people screened [4]. The only operation applied to them is Bayes' theorem.
The check the arithmetic was never given. That validation paper reports a sensitivity, a specificity, a positive predictive value of 98 percent and a negative predictive value of 91.2 percent — and separately reports that it enrolled 161 cases and 129 controls. It never connects them. Applying Bayes at the study's own case-control ratio of 55.5 percent gives a PPV of 97.96 percent and an NPV of 91.25 percent, reproducing both headline figures to within a tenth of a percentage point. That is not a coincidence and it is not a compliment: it demonstrates that the reported predictive values are a property of the enrolment design — roughly half cases by construction — and therefore do not transfer to any clinic.
A second check, against a different source. The World Gastroenterology Organisation states that tTG has a "low positive predictive value in the general (low risk) population (with a prevalence of 1%)", and gives no number [6]. Using sensitivity and specificity measured by an unrelated group, Bayes puts it at 28 percent — so roughly seven in ten positives in an unselected population would be false. The guideline's adjective is confirmed by arithmetic that neither paper performed.
What follows from that. The same assay, unchanged, gives a positive predictive value of 98 percent in its validation study, 65 percent in a first-degree relative, 41 percent in a symptomatic patient, 23 percent in someone not at risk, and 28 percent in an unselected population at 1 percent prevalence. Nothing about the test changed. Only who was tested.
This is the whole justification for three things that otherwise look like bureaucracy. It is why biopsy confirmation is standard — even a 65 percent PPV in the highest-risk group is not a diagnosis. It is why guidelines recommend a serial algorithm adding the more specific endomysial antibody in low-prevalence settings [6]. And it is why the no-biopsy pathway requires a very high titre: raising the threshold to ten times the upper limit of normal trades sensitivity for specificity, and specificity is exactly the term that governs predictive value when prevalence is low [39].
The mirror image is the reason these tests are still excellent. At 1 percent prevalence the negative predictive value exceeds 99.9 percent. tTG-IgA is close to conclusive for ruling celiac disease out and weak for ruling it in — which is the correct shape for a first-line test in a common, under-diagnosed, easily-missed condition.
Three honest limits. The sensitivity and specificity come from a paediatric cohort and would differ in adults and across assay manufacturers. The figure assumes a single fixed cut-off, whereas the real diagnostic pathway uses the titre as a continuous variable — which is the whole point of the 10× rule, and this substrate does not contain titre-stratified predictive values, so the figure cannot show how the curve shifts at higher thresholds. And pre-test probability in practice is a clinical judgement, not one of the four tidy numbers plotted here.
Pillar 2: treatment
The gluten-free diet
There is one treatment and it is dietary: strict, lifelong exclusion of the gluten-containing cereals — wheat, rye and barley [6] [48]. It heals the intestine, resolves symptoms in most patients, and normalises the antibodies.
How completely it heals, and how fast, is worth knowing precisely. In 158 patients followed with serial biopsies over 15 years, histological remission among those with villous atrophy was 65.0 percent within 2 years, 85.3 percent within 5 years, and 89.9 percent in long-term follow-up — leaving 10.1 percent with persisting villous atrophy after 5 years. Children did far better: 95 percent recovered within 2 years and 100 percent in the long term [49]. Adults heal slowly and sometimes incompletely, and the gap between symptom resolution and mucosal healing is real.
Adherence is the limiting factor. A systematic review of 38 studies found strict adherence ranging from 42 to 91 percent depending on definition and measurement, lowest among ethnic minorities and among those diagnosed in childhood, and most strongly associated with cognitive, emotional and socio-cultural factors, membership of an advocacy group, and regular dietetic follow-up [50]. Notably, patients found by screening adhered no worse than those who presented with symptoms — an argument against the assumption that asymptomatic people will not comply. Adherence is also expensive: the economic burden of the diet itself has been quantified [51]. Objective measurement is now possible through gluten immunogenic peptides detectable in urine [52].
Nutrition and bone
Malabsorption leaves deficits that need looking for — and they persist. Poor vitamin status was documented in celiac patients who had been on a gluten-free diet for ten years [53], so the diet does not automatically fix nutrition.
Bone deserves specific attention. Bone mass and metabolism are abnormal in celiac disease [54] [55], the gluten-free diet improves bone mineral content in growing patients [56], and there are joint guidelines for osteoporosis in celiac disease and inflammatory bowel disease [57]. Conversely, celiac disease is found more often than expected among patients presenting with osteoporosis, prompting arguments for routine screening in that group [58].
Monitoring, and refractory disease
Follow-up tracks antibody normalisation and, where indicated, mucosal healing. Clinical and laboratory information have a high positive but low negative predictive value for persisting intestinal damage in treated patients, and dietary compliance assessed by interview was the best single marker of control [59] — another instance of the theme running through this review.
Refractory celiac disease — persisting villous atrophy and malabsorption despite a strict diet — affected 7.0 percent of that follow-up cohort, and 5 of those 11 patients developed enteropathy-associated T-cell lymphoma [49]. It is classified by whether the intraepithelial lymphocytes are phenotypically abnormal [60] [61] [62], the relationship to lymphoma is well described [63], and immunosuppression with azathioprine and prednisone has been used [64]. Lymphoproliferative malignancy risk tracks the small-intestinal histopathology [65], as does mortality [66] [67] [68] [69].
Associated conditions and relatives
Celiac disease travels with other autoimmune disease — autoimmune thyroid disease [70] and type 1 diabetes above all, with celiac disease appearing after the onset of type 1 diabetes in prospective follow-up [71] and the two sharing genetic variants [26]. Duration of gluten exposure has been examined as a risk factor for developing further autoimmune disorders [72]. Severe liver disease is another association in which the diet may reverse the abnormality [73].
Testing first-degree relatives is justified by the arithmetic above: at 1 in 22, they are the group in which a positive result carries the most weight [4].
Drugs
There are none in routine use. The gluten-free diet remains the only treatment [40] [74] [48].
Pillar 3: what is unresolved
Non-dietary therapy. Three strategies have reached patients in trials. Enzyme therapy follows directly from the 33-mer's protease resistance — the original paper showed the peptide could be detoxified by a bacterial prolyl endopeptidase and proposed oral peptidase supplementation on that basis [11], combination enzyme therapy for gastric digestion of dietary gluten has been tested in patients [75], and a prolyl endoprotease has been shown to degrade gluten efficiently [10]. Tight-junction modulation with larazotide acetate was studied for preventing celiac activation during gluten challenge [76] and for persistent symptoms despite a gluten-free diet [77] [78]. None is standard care, and the honest framing is that these are adjuncts to the diet rather than replacements for it.
Mass screening. The case is unusually strong on paper — the disease is common, most cases are undiagnosed [3], screening-detected patients are usually symptomatic once asked [8], and they adhere to the diet as well as anyone [50]. Screening has been done sequentially with serology and gastroscopy in North America [79], in primary care with rapid tests [80], and cost-effectiveness has been modelled in defined high-risk groups such as children with Down syndrome [81]. What the figure above adds is the counterargument: at population prevalence, most positives are false, so any mass programme must be a two-stage algorithm and must be costed as one.
Better diagnostics. Deamidated gliadin peptide and tTG antibodies have been meta-analysed against each other [29], urinary gluten immunogenic peptides offer objective adherence measurement [52], and the open question is a titre-based pathway validated in adults as the 10× rule has been in children [39] [40].
Who heals and who does not. One in ten adults still has villous atrophy after five years of a strict diet [49], and predicting them in advance is not currently possible.
Dig deeper in lmmol
Celiac disease belongs with the autoimmune cluster and the associations are clinical, not merely thematic. Thyroid disease is the commonest companion [70], and type 2 diabetes and glycemic control is the nearest published relative of the type 1 diabetes link, where the two conditions share genetic variants outright [26] [71]. Inflammatory bowel disease is the instructive contrast: another immune-mediated enteropathy, but one with no identified trigger — remove gluten and celiac disease resolves, whereas Crohn's and colitis are managed by suppressing immunity indefinitely, and the two conditions even share osteoporosis guidance [57]. Rheumatoid arthritis and psoriasis round out the autoimmune siblings, and celiac disease is the one among them where the antigen is known and avoidable. Osteoporosis is a downstream consequence worth reading in both directions, since undiagnosed celiac disease turns up among people presenting with low bone density [58] [57]. The full collection is at health.