2.1

Pelotas cohort study 1982

Pelotas is a medium-sized city located in a relatively affluent area in the south of Brazil. In 1982, all infants born at three maternity hospitals in the city were identified, and the 5914 live-born infants and their mothers were weighed and measured; the mothers were also interviewed. This population has been followed several times, and further information is available elsewhere . In 1997 (when the children were 15 yrs of age), a systematic sample of 70 census tracts (27% of the total) was selected, and every household in these tracts was visited; 1076 cohort members were interviewed. Of these, 900 were randomly selected for the Oral Health Study (OHS-97), which consisted of an interview and dental examination. In 2006, when those individuals were 24-years-old (OHS-2006), 888 subjects investigated in the OHS-97 (98.7%) were contacted for another assessment comprised of interviews and oral health exams. From those, 720 were clinically examined. In 2013, the same 888 individuals investigated in the OHS-97 were again invited to participate in a new oral health assessment (OHS-13). In this last evaluation individuals were examined for quality of posterior restorations, presence of periodontal disease (bleeding on probing, probing depth and gingival margin), dental caries and other oral conditions. Oral examination was performed at home, using mirror, probe and artificial light, by six dentists previously trained and calibrated. Inter-examiner reliability was calculated and the lowest kappa for quality of restorations was 0.70. In order to assure the quality of the study, 15% of interviews were repeated.

2.2

Outcome – quality of restorations

All direct posterior restorations were clinically evaluated and classified as satisfactory or unsatisfactory based on criteria used previously on the same cohort (OHS-2006) . Restorations were classified as clinically satisfactory when they did not need an intervention (repair or replacement). This classification was originated from modified United States Public Health Service criteria (USPHS).

2.3

Tooth-level variables (Clinical factors)

At tooth-level, posterior restoration were assessed by: 1) Tooth type (premolars or molars); 2) Number of surfaces involved on restoration (one; two; three or more surfaces); 3) Restorative material used (amalgam or composite); 4) Estimated time in mouth reported by individuals (up to 10 years and more than 10 years).

2.4

Individual level variables

Individual variables were obtained from previous assessments of the cohort. Individual educational level at age 30 was collected by the number of years of study and categorized in three groups (“more than 11 years”, “from 9 to 11 years”, “less than 9 years”). Family income at age 30 was continuously collected in Brazilian currency (US Dollar 1 = BRL 2.20) and analysed in tertiles (“Higher” (3120–25000 BRL), “Intermediate” (1751–3100 BRL), Lower (85–1747 BRL)). Use of dental services at age 31 was collected through the questions “Have you been to dentist in the last 12 months?” and “Where have you been attended?”. Subsequently, the variable was categorized in “private, no insurance”, “Public service, free of charge” and “Private, reimbursed by dental health insurance”.

To assess periodontal status of individuals, clinical attachment loss was clinically measured by probing depth (distance in millimeters between the gingival margin and gingival sulcus) and gingival level (distance in millimeters between the gingival margin and the cementoenamel junction). Periodontal examinations included a full-mouth protocol, probing six sites per tooth using a PCP2 probe. Individuals were classified following the criteria of Baelum & Lopez , where periodontal diseases were identified when individuals had at least one site with simultaneous presence of attachment loss of 4 mm or more and bleeding on probing.

Dental caries experience was investigated at age 15, 24 and 31 by DMFT index . The Decayed component (D) collected at these three moments was organized as a discrete variable. Through a group-based trajectory analysis, the trajectory of caries presence from age 15 to 31 was established.

The risk for occlusal stress due to parafunctions was assessed by seven questions to detecting possible signs of occlusal disorders . In addition, the presence of facets parallel to the normal planes of contour, noticeable flattening of cusps or incisal edges and/or total loss of contour and dentinal exposure when identifiable were investigated. Individuals were classified as having a high risk for occlusal stress when answered positively for two questions and presented at least one of clinical aspects investigated .

2.5

Data analysis

Software STATA version 14.2 was used for data analysis. Group-based trajectory modeling was used to identify different trajectories for presence of caries throughout individuals life course. The method was designed to identify groups of individuals (clusters) that share similar trajectories of variables of interest measured over time. The models were estimated with the “traj” command .

Decayed component (D) of the DMFT index collected in these follow-ups as discrete variables were used to create a life course trajectory for experience of dental caries. The parameters for trajectory model were determined based on the maximum likelihood of the quasi-Newton method . The model selection method considers the estimation of the latent number of categories and the order of the polynomial for each latent trajectory. The final number of trajectories was established when the sequential comparisons of the Bayesian information criterion (BIC) and the adjusted BIC between the model with k and k + 1 trajectories did not result in any more substantial difference in the BIC value. For each group, a quadratic trajectory was considered, starting with only one group in a null model. The BIC analysis supported a model of 2 trajectory groups, being classified as high and low untreated caries during life-course.

Poisson multilevel regression models were used to analyze factors associated to quality of restorations, considering mixed effects and two levels of variables organization: tooth-level (level 1) and individual level (level 2). Variables selection in the model followed a theoretical model, based on the model proposed by Correa . Independent variables were ordered in four blocks to determine their entry in the multilevel model. Demographic and socioeconomic variables were inserted in the most distal block in relation to the outcome, followed by use of dental service at 31 years and oral health variables. Tooth level variables were positioned in a different block as proximal determinants of restoration failures. All associations were adjusted for covariates positioned in the same and in the upper levels of the model. Goodness of fit of models was assessed using deviance (-2 log likelihood). Individual level variance of the models was assessed using the command “xtmrho”. This command provides the Median Incidence Rate Ratio (MIRR). However, due to cross-sectional characteristic of the outcome we have estimated the Prevalence Ratio as effect measure and, consequently, the individual level variance measure was the Median Prevalence Ratio (MPR).

2.6

Ethical issues

All phases of this study were approved by the Ethics Committee of the Federal University of Pelotas, Faculty of Medicine, #384.332. Written informed consent was obtained from all enrolled individuals or from their parents or legal guardians.

2.1

Pelotas cohort study 1982

Pelotas is a medium-sized city located in a relatively affluent area in the south of Brazil. In 1982, all infants born at three maternity hospitals in the city were identified, and the 5914 live-born infants and their mothers were weighed and measured; the mothers were also interviewed. This population has been followed several times, and further information is available elsewhere . In 1997 (when the children were 15 yrs of age), a systematic sample of 70 census tracts (27% of the total) was selected, and every household in these tracts was visited; 1076 cohort members were interviewed. Of these, 900 were randomly selected for the Oral Health Study (OHS-97), which consisted of an interview and dental examination. In 2006, when those individuals were 24-years-old (OHS-2006), 888 subjects investigated in the OHS-97 (98.7%) were contacted for another assessment comprised of interviews and oral health exams. From those, 720 were clinically examined. In 2013, the same 888 individuals investigated in the OHS-97 were again invited to participate in a new oral health assessment (OHS-13). In this last evaluation individuals were examined for quality of posterior restorations, presence of periodontal disease (bleeding on probing, probing depth and gingival margin), dental caries and other oral conditions. Oral examination was performed at home, using mirror, probe and artificial light, by six dentists previously trained and calibrated. Inter-examiner reliability was calculated and the lowest kappa for quality of restorations was 0.70. In order to assure the quality of the study, 15% of interviews were repeated.

2.2

Outcome – quality of restorations

All direct posterior restorations were clinically evaluated and classified as satisfactory or unsatisfactory based on criteria used previously on the same cohort (OHS-2006) . Restorations were classified as clinically satisfactory when they did not need an intervention (repair or replacement). This classification was originated from modified United States Public Health Service criteria (USPHS).

2.3

Tooth-level variables (Clinical factors)

At tooth-level, posterior restoration were assessed by: 1) Tooth type (premolars or molars); 2) Number of surfaces involved on restoration (one; two; three or more surfaces); 3) Restorative material used (amalgam or composite); 4) Estimated time in mouth reported by individuals (up to 10 years and more than 10 years).

2.4

Individual level variables

Individual variables were obtained from previous assessments of the cohort. Individual educational level at age 30 was collected by the number of years of study and categorized in three groups (“more than 11 years”, “from 9 to 11 years”, “less than 9 years”). Family income at age 30 was continuously collected in Brazilian currency (US Dollar 1 = BRL 2.20) and analysed in tertiles (“Higher” (3120–25000 BRL), “Intermediate” (1751–3100 BRL), Lower (85–1747 BRL)). Use of dental services at age 31 was collected through the questions “Have you been to dentist in the last 12 months?” and “Where have you been attended?”. Subsequently, the variable was categorized in “private, no insurance”, “Public service, free of charge” and “Private, reimbursed by dental health insurance”.

To assess periodontal status of individuals, clinical attachment loss was clinically measured by probing depth (distance in millimeters between the gingival margin and gingival sulcus) and gingival level (distance in millimeters between the gingival margin and the cementoenamel junction). Periodontal examinations included a full-mouth protocol, probing six sites per tooth using a PCP2 probe. Individuals were classified following the criteria of Baelum & Lopez , where periodontal diseases were identified when individuals had at least one site with simultaneous presence of attachment loss of 4 mm or more and bleeding on probing.

Dental caries experience was investigated at age 15, 24 and 31 by DMFT index . The Decayed component (D) collected at these three moments was organized as a discrete variable. Through a group-based trajectory analysis, the trajectory of caries presence from age 15 to 31 was established.

The risk for occlusal stress due to parafunctions was assessed by seven questions to detecting possible signs of occlusal disorders . In addition, the presence of facets parallel to the normal planes of contour, noticeable flattening of cusps or incisal edges and/or total loss of contour and dentinal exposure when identifiable were investigated. Individuals were classified as having a high risk for occlusal stress when answered positively for two questions and presented at least one of clinical aspects investigated .

2.5

Data analysis

Software STATA version 14.2 was used for data analysis. Group-based trajectory modeling was used to identify different trajectories for presence of caries throughout individuals life course. The method was designed to identify groups of individuals (clusters) that share similar trajectories of variables of interest measured over time. The models were estimated with the “traj” command .

Decayed component (D) of the DMFT index collected in these follow-ups as discrete variables were used to create a life course trajectory for experience of dental caries. The parameters for trajectory model were determined based on the maximum likelihood of the quasi-Newton method . The model selection method considers the estimation of the latent number of categories and the order of the polynomial for each latent trajectory. The final number of trajectories was established when the sequential comparisons of the Bayesian information criterion (BIC) and the adjusted BIC between the model with k and k + 1 trajectories did not result in any more substantial difference in the BIC value. For each group, a quadratic trajectory was considered, starting with only one group in a null model. The BIC analysis supported a model of 2 trajectory groups, being classified as high and low untreated caries during life-course.

Poisson multilevel regression models were used to analyze factors associated to quality of restorations, considering mixed effects and two levels of variables organization: tooth-level (level 1) and individual level (level 2). Variables selection in the model followed a theoretical model, based on the model proposed by Correa . Independent variables were ordered in four blocks to determine their entry in the multilevel model. Demographic and socioeconomic variables were inserted in the most distal block in relation to the outcome, followed by use of dental service at 31 years and oral health variables. Tooth level variables were positioned in a different block as proximal determinants of restoration failures. All associations were adjusted for covariates positioned in the same and in the upper levels of the model. Goodness of fit of models was assessed using deviance (-2 log likelihood). Individual level variance of the models was assessed using the command “xtmrho”. This command provides the Median Incidence Rate Ratio (MIRR). However, due to cross-sectional characteristic of the outcome we have estimated the Prevalence Ratio as effect measure and, consequently, the individual level variance measure was the Median Prevalence Ratio (MPR).

2.6

Ethical issues

All phases of this study were approved by the Ethics Committee of the Federal University of Pelotas, Faculty of Medicine, #384.332. Written informed consent was obtained from all enrolled individuals or from their parents or legal guardians.