We conducted a prevalence study based on registry data collected in the Fifth Military Conscription District of Denmark, which includes the former counties of North Jutland and Viborg (population approximately 700 000). We included all men born as singletons from January 1, 1977, to December 31, 1983, who registered in the Fifth Military Conscription District from 1997 to 2003. Evaluation at the military draft board is mandatory for all Danish men aged 18–20 years. Men must register with authorities in one of the country’s conscription districts, which were determined by their place of residence at age 18. During registration, men can report conditions that potentially preclude military service. Draft board physicians verify these reports and men with a verified condition are exempted from military duty without further examination. Documentation for the diagnosis leading to exemption is filed in the Conscript Registry and coded according to the 10th revision of the International Classification of Diseases (ICD-10) .
We used the civil registration number (a unique personal identifier assigned at birth) to link data from the Conscript Registry to information on previous hospitalisations for OM recorded in the Danish National Registry of Patients (DNRP) . The DNRP contains data on all hospitalisations in Denmark since 1977, including civil registration numbers and diagnoses coded according to the International Classification of Diseases eighth revision (ICD-8) until 1994, and the tenth revision thereafter .
We defined hospitalisation with OM as at least one hospital diagnosis of OM (ICD-8 codes 381 and 382) registered before the age of 8, since the prevalence of OM among Danish children is highest in this age group . Thus, we included cases of OM with and without mastoiditis, acute OM, OME, and chronic suppurative OM. Hereafter, OM refers to any of these different types of OM.
Cognitive function and educational level
All men who attend the evaluation at their military draft board must take a 45-minute group intelligence test, the Boerge Prien test (Danish Børge Prien Prøve, BPP), which has been used since 1957 by Danish military draft boards . The test includes four time-limited subtests: letter matrices, verbal analogies, number series, and geometric figures. The single final score is the sum of correctly answered items (range: 0–78). BPP scores correlate to a large extent with scores in the Wechsler Adult Intelligence Scale (WAIS) (correlation coefficient = 0.82), although the BPP is a group test and the WAIS is an individually-administered test. Nevertheless, both tests are assumed to measure general intelligence . From the conscript records, we also obtained information on whether the conscripts had achieved or were currently achieving the General Certificate of Secondary Education (GCSE), which is typically achieved at the age of 18 or 19 in Denmark.
We obtained information on gestational factors from the Danish Medical Birth Registry, which has tracked all births in Denmark since January 1, 1973 . The information is derived from birth notification forms completed by midwives attending the birth. Variables include civil registration number, date and place of birth, gestational age, birth weight, maternal parity, and, since 1978, information on maternal marital status and Apgar score.
Given that permanent hearing impairment may be a consequence of OM in developing countries  and hearing impairment is also associated with cognitive function [7, 8], we obtained information on hearing from conscript records. At the conscript examination an audiometric test is performed. The recording of the audiogram measurements are undertaken at frequencies 500, 1000, 2000, 3000, 4000, 6000 and 8000 Hz. Initial screening are carried out equivalent to 20 decibel . Hearing is classified on a scale from 0–5 (severe hearing impairment to no hearing impairment). To improve the precision of the estimates, we collapsed the scale into severe hearing impairment (0–3), moderate hearing impairment (4), and no hearing impairment (5).
We first constructed a box-and-whisker plot of BPP score distributions according to age at first hospitalisation for OM. We defined low cognitive function as a BPP score in the bottom quartile and used a log-binomial regression to estimate crude and adjusted prevalence ratios (PRs) for a BPP score in this quartile, according to age at first hospitalisation for OM . PRs for a BPP score in the bottom quartile were estimated and stratified according to hearing impairment measured at the time of conscription examination.
The following possible confounding factors were included as identified from the literature: birth order (0, 1, 2, 3+), maternal age (≤20, 21–35, >35 years), marital status (married/unmarried), 5-minute Apgar score (<7, 7–10), gestational age, a composite variable derived from gestational age and birth weight [small for gestational age (SGA) or not] , and hearing score at conscription (severe, moderate, or no hearing impairment). In the regression analysis all variables were entered as sets of indicator variables. In a separate analysis we additionally adjusted for a previous diagnosis of febrile seizures or epilepsy. We examined PRs for OM hospitalisation and GCSE stratified into four quartiles by BPP score. To avoid loss of observations, missing values were estimated using multiple imputations, such that five imputed datasets were created and analysed with their averages serving as the estimates [24, 25]. The regression model used for imputation included variables for gestational age, SGA or non-SGA, birth order, maternal age, marital status, and Apgar score 5-minutes after birth . Confidence intervals (CI) around these estimates reflect uncertainty both about the value of the PRs and about the imputed values.
There was a lack of cognitive data on the exempted men because many of the conscripts (11.9%) were exempted from draft board examination. To quantify this potential selection bias, a sensitivity analysis was performed in which the regression analyses were repeated with imputation of BPP scores for all the 2470 exempted men. To assess the maximal impact of a potential selection bias, the regression analyses for the worst case scenario and best case scenario was repeated assuming that all exempt men had a BPP score in the bottom quartile and no exempt men had a BPP in the bottom quartile, respectively. Model-estimated PRs were compared with corresponding pooled Mantel-Haenszel (non-parametric) estimates. In an additional analysis, all men with a previous diagnosis of meningitis were excluded (68 men out of 18 412 in the study population).
Data were analysed using Stata software, version 11.2 (Stata, 4905 Texas, USA).
The study was approved by the Danish Data Protection Agency (http://www.datatilsynet.dk/english/, record no. 2011-41-5807). Data were obtained from the Conscript Registry, Danish Medical Birth Registry, and the DNRP. According to Danish legislation the study did not require permission from a Scientific Ethical Committee.