The Incidence and Risk of Celiac Disease in a Healthy US Adult Population

Mark S. Riddle, MD, DrPH; Joseph A. Murray, MD; Chad K. Porter, PhD


Am J Gastroenterol. 2012;107(8):1248-1255. 

In This Article


Database Information

Data were obtained from the Defense Medical Surveillance System, the main repository for medical data of Department of Defense beneficiaries maintained by the Armed Forces Health Surveillance Center.[10,11] All subjects were active duty US military personnel who served between 1999 and 2008. Medical information was derived from ambulatory and inpatient claims data for care provided within the Military Health System. Demographic data were derived from personnel information; deployment data were derived from deployment rosters and deployment health assessments. Included data were linked at the individual level and compiled into a single de-identified data set, which was provided to the study investigators. The study was reviewed and approved by the Naval Medical Research Center Institutional Review Board in compliance with all applicable regulations governing the protection of human subjects.

CD Case Identification and Control Selection

A CD case was identified when the service member had at least two ICD9-CM (579.0) specific medical encounters for a CD diagnosis within 6 months. Subsequent medical encounters were determined to be related to the CD if the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9-CM) code was included in any of the diagnostic positions. Each case was matched by time (a medical encounter within 1 calendar year), sex, and age (within 1 year) with up to four controls with an unrelated diagnosis randomly selected from the population that produced the cases (i.e., active duty military personnel). Baseline medical encounters for the control subjects included vaccinations, procedures, or other unrelated diagnoses. Incident CD was defined as the first documented ambulatory or inpatient medical encounter with the ICD9-CM code listed above. Given that the military generally would not enlist or commission a service member with known CD, it is unlikely that many of the incident CD diagnoses would include individuals who have a prior diagnosis of CD, though undiagnosed disease could be prevalent among both cases and controls selected.


The primary exposure variable of interest was infectious gastroenteritis (IGE) at any time before a diagnosis of an incident CD (cases) or censure (controls). An IGE exposure was defined by ICD9-CM codes for specific pathogens and non-specific infectious enteritis as previously reported.[12] Non-specific IGE codes were included due to the lack of routine microbiology performed on patients with IGE in similar health-care settings.[13] Both specific and non-specific codes were utilized to classify an exposure as a specific bacterial, non-specific bacterial, protozoal, or viral etiology. We also evaluated exposures at 24, 18, 12, and 6 months before CD diagnosis or censure (controls) to assess the temporal relationship between exposure and outcome. While previous studies have evaluated associations of exposure up to 1 year, we chose to extend exposure windows up until 24 months to account for potential in diagnostic delay due to patient's care seeking behavior and/or provider diagnosis,[14] and we saw no significant changes in effect estimates with differential exposure windows. Due to similarities in clinical presentation between the primary exposure (IGE) and symptoms of the CD outcome of interest, we evaluated a 6-month diagnostic delay window whereby exposures occurring within 6 months before CD diagnosis (or censoring for controls) were excluded.[12] In addition to IGE exposure, other demographic variables available in the data were evaluated, including race, military rank, educational attainment, marital status, and branch of service.


CD incidence was estimated using the number of incident cases in a given year and the total number of active duty service members for that same year. The associations between CD, IGE, and covariates were initially explored by univariate methods. All analyses evaluated each CD independently. Univariate and multivariate conditional logistic regression models were used to evaluate the relationship between IGE, other covariates, and CD. For multivariate models, a backwards elimination approach was used, whereby all variables were initially added to the models. The variable with the largest insignificant P value was removed, and the models were refit. This process was continued iteratively until all variables retained in the models were significant at the α = 0.15 level.[15] Effect modification was assessed statistically utilizing a multiplicative approach. The association of psychological comorbidity among cases and controls was also evaluated, but limited to concomitant diagnosis for initial CD visit of incident diagnosis and at censure visit for controls.

Statistical analyses were performed using SAS v. 8.2 for Windows (SAS Institute, Cary, NC). Two-tailed statistical significance was evaluated using an α of 0.05.