Study population and sampling procedure
The study population was students attending public secondary schools in Kampala and Lira. Kampala is the capital city, which accounts for nearly 45% of all urban residents in Uganda. Lira district is typically rural and situated 350 km from Kampala. Although the proportion of the urban-rural population in Uganda is 1:5, approximately 80% of all secondary schools are located in urban areas. Thus, there was no strict justification for a proportional sampling of students from the two areas to reflect the urban-rural population ratio. The fluoride concentration of the drinking water in Kampala is 0.3 mg/L. In Lira, the fluoride concentration in the water was unknown at the time of the survey. Water samples were subsequently collected in tight-lidded plastic bottles for fluoride analysis. The fluoride concentration in Lira ranged from 0.10 mg/L to 1.20 mg/L with an average of 0.50 mg/L .
The STATA statistical program was used to estimate the sample size, allowing for a design factor, an assumed caries proportion in the Ugandan population and required precision. The sample size was calculated following the guidelines of our statistician. Briefly, using a software package that takes account of complex sample designs (STATA), and assuming a design factor of 1.7, a standard error of 0.02 and a proportion of sugar consumption of 25 % (p = 0.25), yielded a sample size of 1.300 students.
Twenty schools were listed in Lira and 10 considered for sampling, the inclusion criteria being schools with at least 250 students and placed at least 10 km from Lira town center. In Kampala, 30 schools within the radius of 10 km from the city center (main Post Office building) were considered for sampling. A total of 10 secondary schools (5 from Kampala and 5 from Lira) were selected by simple random sampling. A total of 1146 out of 1324 eligible students, 52% urban, mean age 15.8 (SD1.6, range 13–19), response rate 87%, completed structured questionnaires in the autumn of 2001. A sub-sample for a subsequent survey including a clinical examination was selected by systematic random sampling from the list of the participants of the main survey. The follow-up survey was conducted three months after completion of the main survey. A total of 372 students out of 515 sampled, 48% urban, mean age 16.3 (SD 1.7), response rate 72%, completed a short version of the original questionnaire and were examined for dental caries following the WHO diagnostic criteria . Absenteeism from school was the main reason for non-response. Written informed consent to participate in the study was obtained from students and their parents/guardians. Ethical clearance to conduct the study was granted by the Ministries of Health and Education in Uganda, the local administration, school authorities and the ethical research committees in Uganda and Norway.
The questionnaire was constructed and completed in English, the language of instruction in Ugandan secondary schools. This made the translation and back-translation of the questionnaire unnecessary. Sensitivity to culture and selection of appropriate words were considered. Oral health professionals and methodologists reviewed the survey instrument for semantic equivalence, experiential equivalence and conceptual equivalence. It was pilot tested before use. Some modifications concerning clarification of the content and simplification of the wording was considered necessary after the pilot study. The students completed the questionnaires in their respective classrooms under the supervision of trained research assistants and in the absence of the teacher to ensure confidentiality and to reduce response bias. The questionnaire of the main-survey had 78 questions assessing socio-demographic characteristics, the OIDP inventory, oral health related behaviours, self-assessed oral health and social and psychological factors related to intake of sugared snacks and drinks. The questionnaire of the follow-up survey had 32 questions. It addressed sugar intake and related socio-psychological factors a second time in addition to the main socio-demographic characteristics.
Socio-demographic characteristics were assessed in terms of gender, age and place of residence. Father's and mother's highest level of education were assessed on scales ranging from 1 = "have not gone to school" to 5 = "university". Two dummy variables were constructed yielding the categories 0 = "lower education" (including no education/primary school and adult education) and 1 = "higher education" (including secondary school, college, university) and added into a sum score of Parental education yielding the categories 0 = low, 1 = medium, 2 = high. Religious groups was assessed to probe into culturally different lifestyles (e.g. eating habits) that are evident between the main religious belongings; 1 = Catholic, 2 = Protestant and 3 = Muslims/ others. Thus, religious affiliation was used as a social marker in the analyses.
Oral impact of daily performance was obtained by adding scores for eight frequency items. "During the past 6 months how often have problems with your mouth and teeth caused you any difficulties with, 1) eating, 2) speaking and pronouncing clearly, 3) cleaning teeth, 4) sleeping and relaxing, 5) smiling without embarrassment, 6) maintaining emotional state, 7) enjoying contact with other people and 8) carrying out major school work. The scale used was in the range: (0) "never affected", (1) "less than once a month", (2) "once or twice a month", (3) "once or twice a week" (4) "3–4 times a week", (5) "every or nearly every day". For analysis, dummy variables were constructed yielding the categories 0 = "never affected" (including the original category 0) and 1 = "affected less than once a month or more often" (including the original categories 1–5). Simple count scores (SC) were created by adding the 8 dummy variables. Additive scores (ADD) were created by adding the 8 OIDP items as assessed originally. Finally the OIDP SC frequency scores were dichotomised, yielding the categories (0) " no daily performance affected" and (1) "at least one daily performance affected". Proportions of missing cases ranged from 0.5 (difficulty with cleaning teeth) to 2.3% (difficulty with carrying out major school work).
Received oral health care was assessed by one question: "During the past 2 years have you attended a dental clinic in order to receive treatment?" The response categories were 1 = yes, 0 = no. Satisfaction with dental appearance/oral condition was assessed by one item each, i.e. "Are you satisfied or dissatisfied with the appearance/condition of your teeth"?. A 5-point response scale was used ranging from (1) "very satisfied" to (5) "very dissatisfied". For analyses two dummy variables were constructed yielding the categories (0) "Satisfied with dental appearance/ oral condition" and (1) "Dissatisfied with dental appearance/oral condition". Last dental appointment painful was assessed by one item, i.e. " If you have attended a dental clinic, was your last visit painful?" using the categories (0) Not painful at all and (1) at least slightly painful
One trained dentist conducted an oral examination under field conditions with an assistant recording the observations. Caries was assessed using the decayed, missing and filled tooth index (DMFT) as described by the World Health Organisation, WHO . The child was seated on a chair in the shade outside the school building. Cotton rolls were used to clean the teeth and to control saliva. Natural light was the source of illumination. Sharp dental probes and plane mouth mirrors were employed to assess carious lesions. Except third molars, teeth with any part visible through the gum were examined. Caries was recorded as being present when a lesion in a pit or fissure or on a smooth surface had a detectable softened floor, undermined enamel, softened wall or a temporary filling. On proximal surfaces, the probe had to enter a lesion with certainty. A tooth was considered missing, if there was a history of extraction due to pain and / or the presence of a cavity.
Data were analysed using SPSS (version 10.0). Non-parametric statistics were the primary choice because the OIDP frequency scores were not normally distributed. Cohen's kappa and Cronbach's alpha was used to test for intra-examiner agreement and internal consistency reliability, respectively. To assess discriminant validity, multivariate analyses were performed by logistic regression and with the dichotomised OIDP SC scores as dependent variable, checking for all possible 2-way interactions between independent variables. GLM ANOVA was conducted to assess construct validity after using log transformation (non-linear transformation) of the OIDP ADD scores. As the clinical examination was restricted to a sub-sample, the multivariate logistic regression analyses comprised only 372 students. Age, gender, parental education, religious affiliation and place of residence were forced into multivariate analyses independent of statistical significance with the outcome variable to control for potential confounding effects. To correct for effects of the cluster design, logistic regression analyses were re-analysed with STATA (7.0) using the svylogit command. The statistically significant relationships observed initially were left essentially unchanged.