The sample (n = 162) consisted of low socio-economic status English or Afrikaans-speaking adolescents (ages 12 to 16 years) from schools within a 25 km radius of Tygerberg Hospital located in the greater metropolitan area of Cape Town, Western Cape, South Africa. Screening included a structured psychiatric diagnostic interview, a developmental and medical history (from participants and at least one biological parent or legal guardian), a detailed physical and neurological examination assessing developmental delays and urine analysis and breathalyser testing (to confirm sobriety of participants during testing procedures). The Schedule for Affective Disorders and Schizophrenia for School Aged Children (6 to 18 years) Lifetime Version (K-SADS-PL) [33] was used to screen for psychiatric diagnoses. The Semi-Structured Assessment for the Genetics of Alcohol (SSAGA-II) [34] was used to confirm AUD diagnosis and to derive detailed substance use histories (alcohol, tobacco and all other drugs).

Participants were recruited for an AUD group meeting DSM-IV criteria for alcohol dependence or alcohol abuse [35] and for a light/non-drinking non-AUD group (with lifetime dose of <100 standard drinks of alcohol). Exclusion criteria for both groups were: mental retardation, lifetime DSM-IV diagnoses other than AUD, as defined in the KSADS-PL, current use of sedative or psychotropic medication, current signs or history of foetal alcohol spectrum disorders or exposure to heavy antenatal alcohol, sensory impairment, history of traumatic brain injury with loss of consciousness exceeding 10 min, presence of diseases that may affect the CNS (e.g., meningitis, epilepsy), HIV (tested using the enzyme linked immunosorbent assay (ELISA)), less than 6 years of formal education, and lack of proficiency in English or Afrikaans. Prior to consent being obtained for participation in the study, a social worker obtained collateral information from consenting parents, verifying the absence of medical, psychiatric and psychosocial problems. Participants in the two groups were individually matched for age (within 1 year), gender, language, socio-economic status and level of education (within 1 year). A total socio-economic status score was calculated for each participant by summing the category scores for family income (1–6), reversed employment category of participant's parent with the highest employment rank (Hollingshead reversed) (1–9), parent education (0–6), total assets (0–7), dwelling type (1–6) and bedroom cohabitation (1–7), with a maximum score of 41. Scores increased as indicators of socio-economic status increased, thus a higher score was associated with a higher SES. During recruitment it was attempted to match the samples for smoking status, but this was not feasible since smoking was much more prevalent in the AUD participants. This positive association of smoking and alcohol use is well documented [36].

Substance Use: A revised version of the Timeline Follow-back procedure (TLFB) [37], a semi-structured, psychiatrist-administered assessment of lifetime history of alcohol use and drinking patterns (i.e., frequency, quantity and density of alcohol consumption, including every phase from when participants first started drinking at least once per month to the present, including all periods of abstinence) was used in combination with the K-SADS-PL to elicit alcohol-use data. A standard drink was defined as one beer, cider or wine cooler (340 mL), one glass of wine (150 m) or a 45 mL shot of liquor. A similar procedure was carried out for each substance that the research participant acknowledged using.

Biochemical Determination of Vitamin D Status: Circulating s-25(OH)D is widely regarded as the most appropriate measure of vitamin D status with concentrations reflecting medium to long term vitamin D availability from both endogenous and dietary sources [11]. EDTA venous blood samples for the s-25(OH)D assays were collected from each consenting participant via venipuncture. Recruitment of subjects took place across all four seasons of the year resulting in a spread of blood sampling for assessment of vitamin D status over all the seasons. The LIAISON 25(OH) vitamin D assay (DiaSorin, Stillwater, MN, USA) was performed on the LIAISON analyser according to the manufacturer’s instructions. The analytical runs for control and sample specimens were done in duplicate.

For the interpretation of s-25(OH)D concentrations, it is important to note that there is no consensus in the literature on concentrations of s-25(OH)D that define vitamin D insufficiency in infants and children [38]. Holick and Chen have assumed that children have similar requirements to adults [39]. In line with this assumption, a s-25(OH)D concentration of greater or equal to 20 ng/mL was used in establishing the Institute of Medicine (IOM) Recommend Dietary Allowance for vitamin D to meet the requirements of nearly all children and adolescents [20,38,40]. The cut-offs for s-25(OH)D concentrations used in adults and in this study to define vitamin D status, have been used to classify vitamin D status in children [41]. These categories are based on data showing that intestinal calcium absorption is maximised above 32 ng/mL [8] and that parathyroid hormone concentrations in adults continue to decline and reach their nadir at between 30 and 40 ng/mL [42,43,44]. The following categories were thus used to define vitamin D status based on s-25(OH)D concentrations: vitamin D deficiency (<20 ng/mL), insufficiency (20 to 29.9 ng/mL) and sufficiency (≥30 ng/mL and greater) [9,45,46]. These categories are also in line with the reference ranges recommended for the LIAISON 25 OH Vitamin D assay.

Energy, Calcium and Vitamin D Intake: Energy, calcium and vitamin D intakes were estimated using three 24-h recall questionnaires per participant. This method of dietary intake assessment has been shown to be appropriate for quantifying dietary intake of groups in developing countries [47,48] and internal and external validity has been found to be acceptable in adolescents aged ten years and older [49,50].

The energy (kilojoules (kJ)), calcium (milligrams (mg) and vitamin D (International Units (IU)) intakes for each participant for each day were calculated using the South African Food Data System (SAFOODS) [51]. The average intakes over the three 24-h recall interviews were calculated to represent the observed intake distributions. The dietary data include nutrient intake estimates from food (both naturally present and fortified) and water only and exclude nutrient intake estimates contributed by dietary supplements and medications or that obtained from sunlight. It is important to bear in mind the limitations of the vitamin D intake estimates in this study, namely the relatively high percentage of missing values for vitamin D (approximately 30% to 40%) in the SAFOODS database [52].

The energy intake variable obtained from the 24-h recalls did not include estimated energy from alcohol intake. Average daily alcohol energy intake of the AUD group was estimated from average daily alcohol intake (g) per participant in the AUD group using the alcohol-use data from the most recent phase of drinking as follows: (1) frequency of alcohol use (days per month) multiplied by average quantity of alcohol consumed (standard drinks per drinking day) to obtain average monthly standard drinks of alcohol consumed; (2) average monthly standard drinks of alcohol consumed was divided by 28 days to obtain average daily standard drinks of alcohol consumed; (3) average daily standard drinks of alcohol consumed was multiplied by 13.6 g of alcohol per standard drink to obtain average daily alcohol intake in grams, which was converted to average daily alcohol energy (29 kJ/g) to obtain average daily alcohol energy (kJ). Average daily alcohol energy was added to daily energy intake from the observed intake distributions for each AUD participant to represent total estimated energy intake. Daily alcohol energy for the n = 48 light drinking participants in the non-AUD group was not calculated as their alcohol life dose was negligible (mean 5.77; SD 12.46 standard drinks), and the contribution of alcohol energy to total estimated energy intake would therefore also be negligible. The total estimated energy intake variable was reported in this study only for use when regression-adjusted differences between groups were determined, since differences in energy intake between groups may confound differences in nutrient intakes. The 24-h recall data was further used to determine the top five foods foods/energy-containing beverages (excluding alcoholic beverages) that contributed to calcium intake in each group.

Eating Behaviour (Frequency of Intake of Indicator Foods): The weekly frequencies of intake of foods reflecting healthy and poor food choices were estimated using a non-quantitative food frequency questionnaire. The questionnaire consisted of a list of 37 food categories, with each food category consisting of either single or multiple food items that were grouped based on shared nutritional characteristics. Indicator foods/categories were identified by firstly listing foods most commonly consumed by South Africans in the Western Cape using scientific reports, publications as well as unpublished dietary assessment information generated in small research projects/compilation of community profiles for nutrition interventions. Identified foods/categories were then classified as either healthy choices (offer protective effects against non-communicable diseases (NCDs)) or as poor choices (would increase NCDs risk). A panel of nutrition and NCD health experts advised this process. The response categories included “eaten in the past month” (yes/no), and if yes, “times eaten per week” or “times eaten per month”.

The frequency of intake recorded for each one of the 37 food categories was converted to reflect the number of times eaten per week per participant. The food categories “milk/sour milk/yoghurt” and “cheese” were selected as indicators of calcium intake for the purposes of this paper, and weekly frequencies of intake for only these two food categories are reported.

Recruitment procedures included oral presentations at schools and advertisement via word-of-mouth. A convenience sample of alcohol users was selected and non-users and light users for each of these adolescents were selected individually according to matching criteria. At the pre-screening stage, adolescents who did not meet eligibility criteria for possible inclusion in the AUD or non-AUD groups were excluded. Participants who met eligibility criteria were transported from their homes or schools to the testing site for complete physical and psychiatric screening for possible allocation to one of the groups or exclusion.

After confirmation of inclusion in the study, demographic information was obtained, the first 24-h recall interview was conducted and blood samples for the biochemical determination of s-25(OH)D were obtained in the morning from each participant by a phlebotomist. The remaining two 24-h recall interviews were completed on a Monday to obtain Sunday intakes and one other week day thereafter. The 24-h recall interviews were all conducted by a trained researcher, versed in relevant terminology and locally available food and beverages. The procedure for the 24-h recalls included the following consecutive steps: (a) listing of foods and beverages (including water) consumed by the participant in the previous 24 h, starting from time of waking and proceeding chronologically until time of going to sleep; (b) collection of detailed description of foods, preparation methods and brands where relevant and the amounts consumed; and (c) final checking to recall forgotten foods. Commonly used household measures and food pictures from the Dietary Assessment and Education Kit, developed by Steyn and Senekal [53] were used to assist with food portion size estimation. Estimated food portions were converted to grams using the MRC Food Quantities Manual [54]. The 24-h recall interviews in this study were conducted over a period that included all seasons of the year to account for seasonal dietary variations. Dietary data could not be collected for Fridays and Saturdays as it was not feasible to conduct interviews on Saturdays and Sundays.

The indicator food frequency questionnaire was administered by the trained researcher during the second contact session that also involved administration of the second 24-h recall interview.

The Committee for Human Research of Stellenbosch University approved all study procedures (N06/07/128). After eligibility was established, written consent from parents and written assent from participants was obtained. Participants were compensated for their time with gift vouchers. Confidentiality of all study information was maintained with the exception of statutory reporting requirements in newly-identified or ongoing threats to the safety of minor participants.

Descriptive statistics, including inspection of data for adherence to normal distributions, and group comparisons were computed using Stata/IC Version 11.1 for Windows [55]. Suitable transformations were applied to all variables with skewed distributions. Statistical significance was defined at a level of p ≤ 0.05. The smoking variable used in all regressions included the smoking group (light smokers (lifetime < 100 cigarettes) and regular smokers (lifetime > 100 cigarettes)) and the non-smoking group (participants who have never used tobacco).

Descriptive statistics (frequency distributions/means and standard deviations) of socio-demographic and substance use variables in the non-AUD and AUD groups were computed and compared for confirmation of group allocation and matching. For comparison of continuous variables, the Mann Whitney U Test was used and for categorical variables the Chi-square or Fisher’s exact tests were used.

Descriptive analyses for s-25(OH)D concentrations, dietary intakes of energy, calcium and vitamin D as well as weekly frequency of intake of foods in the two food categories, namely “milk/sour milk/yoghurt” and “cheese” included medians and interquartile ranges for each group. Due to the paired nature of the data, multi-level mixed effects linear regression was used to compare these variables between the two groups, with adjustment for gender, smoking status and vitamin D intake. A pairing variable was created according to the matched pairs in the sample and served as the level variable in the model. The s-25(OH)D, energy, calcium and vitamin D variables (transformed) were used as the dependant variables in the regressions. The distribution of the weekly frequency of intake of “milk/sour milk/yoghurt” could not be suitably transformed and a rank variable was created for use as the dependant variable in this particular regression.

The frequencies of participants in the non-AUD and AUD groups with vitamin D sufficiency, insufficiency and deficiency (according to the defined vitamin D status categories) were computed and compared between groups, including gender groups, using the Chi-square test.

As no specific reference intakes are available for the South African population, the Institute of Medicine’s (IOM) Dietary Reference Intakes (DRIs), were used to assess adequacy of dietary intake of calcium and vitamin D [20]. The adequacy of calcium and vitamin D intakes were determined using usual intake distributions in accordance with the recommendation of the IOM concerning the need to determine the usual nutrient intake distributions when assessing adequacy of dietary intake of groups in relation to the DRIs [56]. Adjustments to the observed intake distributions to obtain usual intake distribution estimates were made using the National Research Council (NRC)/IOM method [57,58].

Using the EAR cut-point method the frequencies of intakes below the EARs for calcium and vitamin D were computed in each group to reflect the risk of inadequate intake [56,59]. For these purposes, the EARs for 14 to 18 year-olds were used for calcium (1100 mg) and vitamin D (400 IU) [20]. The frequencies of usual intakes below the EARs (14 to 18 years) for each nutrient were compared between groups using the Chi-square test.

Using the 24-h recall data, the average calcium intake (milligrams) for each food code was computed for each participant, after which the average intake for each group was computed and ranked. This was done to obtain the top five foods/energy-containing beverages (excluding alcohol) that contributed to calcium intake in each group.

A total of 184 adolescents were recruited and screened, of whom 22 were excluded as screen failures due to a range of exclusion criteria, including cannabis and methamphetamine use and DSM-IV Axis I diagnoses, resulting in a final sample of 162. The non-AUD and AUD groups were successfully matched for age, education level, gender, language and socio-economic status. All except two participants from the non-AUD group were from the mixed ancestry ethnic group (Table 1).

As expected, AUD adolescents had dramatically greater alcohol exposure than those without AUD (Table 1). Almost all (95%) adolescents in the AUD group had a “weekends-only” style of alcohol consumption. The regular drinking frequency (days per month) and regular drinking quantity (standard drinks per month) indicates an approximate consumption of 13 drinks per drinking day, which suggests a binge drinking pattern. A greater proportion of adolescents in the AUD group smoked compared to the non-AUD group, and lifetime tobacco dose (total number of cigarettes smoked in lifetime) was greater in the AUD group (Table 1).

Blood samples for the biochemical determination of vitamin D status could be obtained from all participants in the sample, except for one AUD participant. Serum 25(OH)D was significantly lower in the AUD group compared to the non-AUD group (Table 2), although it is important to note that median s-25(OH)D concentrations in both groups were within the vitamin D insufficiency category (20 to 29 ng/mL) (Table 2). Vitamin D insufficiency and deficiency combined was found in almost 90% of AUD adolescents and approximately 70% of non-AUD adolescents (Table 2). A significantly greater percentage of adolescents in the AUD group had insufficient and deficient vitamin D status compared to the non-AUD group. There were no significant gender-by-group interaction effects for the frequency of adolescents with sufficient, insufficient and deficient vitamin D status for comparisons (Table 2).

Complete dietary intake data was collected for 160 participants, with two participants (one participant per group) refusing participation. Regressions showed significantly higher dietary intake of energy and calcium in the AUD group than in the non-AUD group, while vitamin D intake did not differ significantly between the two groups (Table 3). According to the EAR cut-point method all adolescents in the sample were at risk of inadequate calcium and vitamin D intakes, with intakes below the EARs for 14 to 18 year-olds (Table 4).

The first four foods contributing most to calcium intake were the same in both groups and included whole milk, white bread, cheddar cheese, savoury maize and wheat crisps (Table 5). Frequencies of intake of “milk/sour milk/yoghurt” (times per week) in the non-AUD (median 7; IQR 7–7) and AUD (median 7; IQR 3–7) groups were not significantly different. There was also no difference in the frequencies of intake of “cheese” in the non-AUD group (median 2; IQR 1–3) compared to the AUD group (median 3; IQR 0.75–3).