At age 18, all Swedish male citizens are notified of their military obligation, and most are examined for military suitability the same or following year. In addition, a small number of naturalized Swedish males aged 18–24 are summoned for the enrolment test at the time of their naturalization. Only a small proportion of subjects – in 1998, the year of our study, 2%- are rejected before appearing at the enrolment center. These cases consist of imprisoned subjects and those who obviously cannot join the military force because of serious disability (such cerebral paresis). These are the only subjects not participating in the enrolment procedure. Of the other 48,502 who were summoned, 76% were aged 18, 22% were 19, and the remaining 2% were aged 20–24.
The enrolment procedure takes place throughout the year in one of five regional recruitment centers. Physical and psychological tests are conducted at these centers to determine both ability to perform military or civil service and type of service for which individuals are best suited. The procedure consists of a series of self-administered questionnaires, physical and medical examinations, and semi-structured psychological interviews. Each aspect of the testing procedure is undertaken independently of the others, and examiners are blind to the results of previously administered tests.
A determination about whether individuals are inducted is based on a scoring system that signifies a prognosis of ability to perform a worthwhile job in the military. All were subjected to an identical screening process, with the exception that, if at some point in the test procedure an individual's unsuitability for military service became apparent, no further tests might be administered, if it was obvious that the person would be rejected from military service. This would typically happen in some subjects who at the medical examination were found to be medically unfit, or in those who at the intellectual capacity test were found to be definitely mentally substandard. Therefore, the number of persons participating in each test varies somewhat throughout the report.
Variables of interest
Variables for this study were derived from the battery of test procedures taking place at enrolment. Each of these variables was collected independently of the others and without access to previously collected information. Both the independent and the dependent variables are described below:
Independent variables (in order of the test procedure)
▪Type of education is obtained with a questionnaire completed at the test center. Educational programs in Sweden vary in contents and difficulty. The choice of and acceptance into study programs in the last 3 years of schooling depend on previous school performance. The study programs for the last 3 years of schooling may be grouped into three broad classes: a) the natural sciences, which offers the greatest opportunities to continue directly into further higher studies, b) the social sciences/humanistic studies, which offer somewhat fewer opportunities for further higher studies, and c) "other studies" which include a broad spectrum of courses that not only lead to various trades/skilled jobs but are aimed also at keeping the less intellectually inclined occupied in a meaningful way. c) The "other studies" educational track provides only a few if any opportunities for further higher education. Included in the "other" group are also 4,236 young men not placed in the first two educational classes because they had only 9 years of education or they were refugees/immigrants whose educational status was unclassifiable. Separate analyses for this foreign subgroup revealed that its profile matched well with that of the whole group of "other studies".
▪Body height is measured by the medical personnel during the medical examination as well as body weight and body mass index (BMI) is calculated from body weight (kg) and height (cm).
▪Intellectual capacity is determined at the time of enrolment with a battery of tests. It consists of a total of 200 items, with separate sections on logical, verbal, spatial, and theoretical/technical aspects of reasoning. Individual test scores are summed into a final score, which in turn is divided into 9 subgroups, with 1 being the lowest and 9 the highest, the so-called stanine scale. Some subjects, already classified as obviously medically unfit for military service, will not reach this test station.
▪Psychological strength (coping with stress) is obtained from a structured interview with a psychologist. Based on the young men's recollected responses to past events, this variable is intended to reflect the level of adaptation in life, including psychological and physical endurance under stress. Its ultimate aim is to determine ability to cope during war-like situations. Some subjects, already classified as obviously medically or intellectually unfit for military service, will not reach this station.
Dependent (outcome) variables
A health-profile is obtained through a medical interview/examination procedure, based on a health-declaration by the enrollee. The health-declaration questionnaire, filled out at home before travel to a regional enrolment center, consists of information on health problems that are of importance in evaluating future performance as a soldier. Enrollees are also requested to bring copies of their previous medical records and other health-related documents to the enrolment center. At the enrolment center, they go through the questionnaire with a nurse, who also performs a brief physical exam to document obvious disease or impairment. A physician performs another, more thorough, anamnesis and exam. Afterward, with data from the various sources, the physician diagnoses any health problems. Diagnoses are coded in accordance with the WHO International Classification of Diseases, ICD10 categories.
In this study, a number of diagnoses are defined as "back problems" (M40–M55 according to the ICD10 classification system), including non-specific back or neck pain/disease. No consideration in our study was made of the "severity" of the diagnosis. Individuals who indicated that they had a back problem were "forced" into one of these diagnoses by the medical practitioner, although the problem in reality very well might have been of non-specific character.
Ill health was defined as any type of medical condition that was considered by the medical practitioner to be sufficiently to make military service unwarranted.
The study sample represents almost the whole 1998-cohort of 18-year old Swedish men (those 18-year olds who are not tested at the time they are summoned are tested the following year, thereby excluding the possibility of selection bias). The minute fraction (2%) who was not summoned to the test procedure was usually institutionalized for reasons of criminality or severe disability. Whether this resulted in a bias in terms of the associations tested in this study is not known, but the group is too small to be able to influence the results to a large extent. Individuals who did not complete the whole screening procedure are all found in the disadvantaged groups (poor health and/or low intellect), and – had they reached the psychological test station – are likely to also have poor coping skills). Their absence from our analyses would therefore result in an underestimation of the strength of associations rather than the opposite.
When recruits have completed the battery of tests and examinations, data are sent from the five regional test centers to a central data office for cleaning and storage. The quality of data is also checked and if data problems are found, including unusual reporting patterns from the medical and psychological examinations, regional centers are notified. The data-base is well maintained and repeat comparisons are made with previous years to assure standardization. No abnormalities were observed for the data from the year of our study and values for the examined cohort are consistent with the pattern observed over the years.
The Swedish educational system is uniform, and data from it are classified according to an official coding system and are unlikely to be misclassified.
The test instrument used to determine intellectual capacity is performed on a personal computer, computer-analyzed, and scored prior to and independently of any other test administered during the enrolment procedure. The test of intellectual capacity is unavailable to the public. It has been used by the Swedish army for several decades, however, and is under continuous scrutiny and development in order to meet the needs of the military organization. Scores obtained from this instrument have been shown to be relatively normally distributed , as is confirmed in our study. According to a validation study carried out on a sample of 15,195 young men of different types and abilities, scores from the present version of the test have been found to be well correlated with military grades at the time of completion of military service .
The coping variable is obtained through structured, standardized interviews, unavailable to the public, conducted by psychologists trained for this task. Feed-back from central office on its results is provided to the individual testing centers if any deviations from the national pattern are noted. Because not all individuals reach this stage of the induction process, some deviation from the perfect norm would be expected, as confirmed in our study. Coping determined in this manner also has been validated as explained in the paragraph above and found to correlate well with grades at completion of military service .
Analysis and presentation of data
Transformation of independent variables
Intercooled STATA 7.0 was used for the statistical analysis. Each of the 5 independent variables (height, BMI, type of education, intellectual capacity, and coping) was tested against the outcome variables (back problem s and ill health) using the polynomial regression method so as to determine the maximum power with statistical significance (up to the 4th power) of the polynomial . We used this procedure to identify whether associations follow a linear, quadratic, cubic or quadric pattern, or whether there is no pattern at all. This information was used to categorize the independent variables in a clinically relevant manner and to reduce data information loss.
Presence of effect modification may confuse the overall outcome between two variables. For this reason, potential interactions between the variables in relation to outcome were tested using the basic independent variables as described above. This was done by testing two of these at a time and their product in a logistic regression. The pre hoc level of significance was set at 0.01. In this way, any modification between specific subgroups of one variable in relation to specific subgroups of the other variables could be detected. Presence of effect modification would require data to be reported in the relevant strata, whereas absence of effect modification would allow for crude or adjusted overall estimates. This algorithm of analysis followed the recommendations by Gerstman .
Bivariate and multi variable analyses
Links between each independent variable and the outcome variables were tested, one by one, and reported as odds ratios. In order to reveal any associations between these variables, they were then tested against each other, except height and BMI, which by definition are dependent on each other.
Independent variables that were associated with each other were then introduced into the analyses one at a time, controlling for individual confounding. Thereafter, the association for each independent variable was again tested against the outcome variable, whilst controlling for all the other independent variables. Separate analyses were made for height and BMI, as BMI is a product of weight and height. Finally, the area under the receiver operated curve (ROC) was identified. This area denotes the proportion of individuals classified using a specific model. Values between 0.5 and 1 indicate an increasingly higher ability to differentiate true cases from true non-cases.
Due to the large study sample, confidence intervals were so small that they became meaningless to report and even small associations would be statistically significant making also p-values meaningless. The size of the estimate was therefore taken into account, rather than the size of the p-value.