| Sign In to gain access to subscriptions and/or personal tools. |
Direct and Indirect Effects of Interdental Hygiene in a Clinical Trial
1 Department of Periodontology, Leeds Dental Institute, University of Leeds, Clarendon Way, Leeds, LS2 9LU, UK Correspondence: * corresponding author, y.k.tu{at}leeds.ac.uk
Many randomized controlled trials (RCTs) in dental research test the efficacy of interventions on more than one outcome variable. Univariate methods, such as the t test or analysis of covariance, cannot evaluate the efficacy of interventions on multiple outcomes simultaneously. The aim of this study was to use structural equation modeling (SEM) to re-analyze a RCT, comparing the effects of pre-curved interdental brushes and flossing on probing pocket depth (PPD), plaque indices, and bleeding on probing (BOP) measured at baseline, intermediate, and final examinations. Results of SEM showed that the observed greater reduction in PPD and BOP in persons using interdental brushing than in those flossing is due mainly to the greater efficiency in plaque removal with the interdental brushes (indirect effect) rather than to the compression of the interdental papillae (direct effect). In contrast, smokers showed less BOP at baseline but also less improvement in BOP through direct effects.
Key Words: randomized controlled trials • structural equation modeling • interdental brushes • flossing
Many randomized controlled trials (RCTs) in dental research test the efficacy of interventions on more than one outcome variable, for full evaluation of the practical values of the interventions. For instance, RCTs on the treatment of persons with chronic periodontal diseases usually use probing pocket depth, clinical attachment level, plaque indices, and bleeding on probing as the treatment outcomes (Apatzidou and Kinane, 2004; Preshaw and Heasman, 2005). To understand the relationships among the changes in the outcomes requires the use of sophisticated statistical methods to undertake multivariate analysis (Gilthorpe and Cunningham, 2000), but most dental RCTs use univariate methods, such as the ttest or analysis of covariance, which cannot evaluate the efficacy of interventions on multiple outcomes simultaneously. Another limitation of univariate statistical methods is that when outcome variables are measured more than once following interventions, these methods cannot take into account the intermediate measurements while evaluating overall treatment effects. In contrast, by incorporating the intermediate observations, multivariate methods can provide a more informative and reliable evaluation of overall treatment effects in the analysis of RCTs. A previous study on the efficiency of interdental cleaning methods found that interdental brushing was more effective than flossing in plaque removal and pocket depth reduction (Christou et al., 1998). However, it was not clear whether the greater reduction in pocket depth was due to the greater efficiency in plaque removal or to mechanical depression of the interdental papilla by the interdental brush, leading to greater marginal gingival recession. The aim of this study was to re-analyze data from a RCT which compared two interdental cleaning methods with multiple outcomes measured at three occasions. Latent growth curve modeling (LGCM) (Byrne and Crombie, 2003; Curran and Willoughby, 2003; Bollen and Curran, 2006; Duncan et al., 2006), a special application of structural equation modeling for longitudinal data (Schumacker and Lomax, 2004; Kline, 2005), was used in this study to test whether the greater treatment effects (total effects) of interdental brushing compared with flossing on the outcomes were due mainly to plaque removal (indirect effect) or to mechanical depression of the interdental papillae (direct effect). In addition, LGCM was used to investigate the treatment outcomes for smokers and to determine whether the outcomes could be attributed to direct or indirect effects.
This RCT compared the effects of a novel customized interdental brushing technique, with pre-curved brushes, and a customized flossing technique. Detailed accounts of study design, conduct, and data collection of this RCT have been reported elsewhere (Jackson et al., 2006), in accordance with the CONSORT guidelines (Altman, 1996; Begg et al., 1996; Moher et al., 2001). In short, 77 persons with chronic periodontitis and meeting specific inclusion and exclusion criteria were recruited; 39 were randomly allocated to the test (interdental brushing) and 38 to the control (flossing) groups. Clinical measurements were recorded by a single examiner (MJ) at baseline, an intermediate examination at 6 wks, and the final examination at 12 wks. In this study, three main outcome variables were chosen for statistical analysis by latent growth curve modeling (LGCM): probing pocket depth (PPD), plaque indices scored as 0 to 3 (Silness and Löe, 1964), and bleeding on probing (BOP) recorded as absent or present (0 or 1). All outcome variables were measured at 4 sites per tooth (disto-buccal, mesio-buccal, disto-lingual, and mesio-lingual). The data were then aggregated across all sites, and the individual was used as the unit of statistical analysis. Previous analyses of this RCT indicated that the outcome variables followed normal distributions, and results from univariate analysis can be found in the previous study (Jackson et al., 2006).
Model-building and Research Hypotheses
The direct and indirect effects in multivariate LGCM are illustrated conceptually (Figs. 1
The regression weights for the intermediate measurements were close to 1 for all 3 outcomes (0.993 for pocket depth, 1.033 for plaque indices, 0.828 for BOP), indicating that most changes occurred during the first 6 wks (Table 1  ). It is noted that 1.033, the regression weight for the intermediate measurement of plaque indices, means that the final average plaque indices showed a slight relapse. Persons in the test group (using pre-curved interdental brushes) had a significantly greater reduction in pocket depth (0.264 mm, P < 0.001), plaque indices (0.258, P < 0.001), and percentage sites with bleeding on probing (0.091, P = 0.008) than those in the control group (using floss) (Table 1). No substantial differences in the changes in pocket depth and plaque scores were found between smokers and non-smokers (Table 1), but smokers had a significantly smaller reduction in the percentages of sites with bleeding on probing (0.07, P = 0.046). The P-values of the Chi-square test for the 3 univariate LGCMs were all greater than 0.05, indicating acceptable model fit.
The results of multivariate LGCM (mLGCM-1) for testing the relationship between pocket depth and plaque indices (Table 2 , Appendix Fig. 2) showed that persons with higher baseline plaque indices had, on average, greater pocket depth (1.037mm per one-unit increase in plaque indices, P < 0.001), and persons with greater improvement in average plaque indices also achieved greater average pocket depth reduction (0.944 mm, P = 0.061). Persons using the pre-curved interdental brushes showed greater reduction in the average plaque indices (0.256, P <0.001) than those flossing, and this in turn gave rise to greater pocket depth reduction (indirect effect). The direct effect of different cleaning methods (interdental brushing vs. flossing) on pocket reduction was relatively small (0.02 mm, P = 0.879) compared with the indirect effects through changes in plaque indices (the regression weight for the path, Brush to F2, multiplied by that for the path, F2 to F4 = 0.256*0.944 = 0.24 mm, P = 0.092). The total (direct plus indirect) effect was 0.26 mm (P < 0.001). The direct, indirect, and total effects of smoking on pocket depth or plaque indices change were small, and none of them was statistically significant. The P-value of the Chi-square test for the model was 0.538, RMSEA = 0, and CFI = 0.999, indicating good overall model fit.
The results of the final multivariate LGCM (mLGCM-2) for testing the relationship among pocket depth, plaque indices, and bleeding on probing (Table 2  , Appendix Fig. 3) showed that persons with higher baseline plaque indices had, on average, greater pocket depth (by 1.041 mm, P < 0.001) and greater tendency for bleeding on probing (0.274, P < 0.001), and persons with greater improvement in average plaque indices also achieved greater average pocket depth reduction (0.947 mm, P = 0.049) and greater improvement in bleeding on probing (0.294, P = 0.081). As shown in the previous model, persons using the interdental brushing technique showed greater reduction in the average plaque indices than those flossing, and this in turn gave rise to greater pocket depth reduction and greater improvement in bleeding on probing. However, the direct effect of different cleaning methods on pocket reduction was relatively small (0.02 mm, P = 0.895) compared with the indirect effects through changes in plaque indices (0.24 mm, P = 0.08). The total effect (direct plus indirect) was 0.26 mm (P < 0.001). The direct effects of different cleaning methods on bleeding on probing were relatively small (0.015, P = 0.789) compared with the indirect effects through changes in plaque indices (0.075, P = 0.112). The total effect was 0.091 (P = 0.007). None of the direct, indirect, or total effects of smoking on pocket depth or changes in plaque indices was statistically significant. Smokers had a lower percentage of sites with bleeding on probing at baseline (direct effect: 0.116, P = 0.019), but smokers showed less improvement in bleeding on probing (direct effect: 0.08, P = 0.032). The direct effect of smoking on the improvement of bleeding on probing was much greater than the indirect effect (which is in the opposite direction). The P-value of the Chi-square test for the model was 0.41, RMSEA = 0.022, and CFI = 0.995, indicating good overall model fit.
Previous analysis of this RCT by univariate statistical methods found that persons using the interdental brushing technique showed greater improvement in pocket depth, plaque indices, and bleeding on probing than those using the flossing technique (Jackson et al., 2006). Other studies also showed that interdental brushing was the preferred method for interdental cleaning in terms of plaque removal (Kiger et al., 1991) and resolution of bleeding on probing (Bergenholtz and Olsson, 1984; Christou et al., 1998). One study also found greater pocket depth reduction in persons using interdental brushing than in those flossing (Christou et al., 1998). Since no difference was found in the bleeding scores between interdental brushing and flossing, it has been hypothesized that the difference in pocket depth reduction was attributed to marginal gingival recession induced by the use of interdental brushing (Christou et al., 1998). However, univariate statistical methods cannot disentangle the direct and indirect effects of different oral hygiene methods on the change in pocket depth. For instance, a greater resolution in the gingival inflammation due to a more effective oral hygiene technique might cause a greater gingival recession. Therefore, the observed greater gingival recession and greater pocket depth reduction in the test group do not necessarily prove that interdental brushing induced greater mechanical trauma to the interdental tissues than flossing. In this study, we used LGCM to show that the greater reduction in pocket depth and bleeding on probing in persons who used the interdental brushing technique was due mostly to the greater efficiency of interdental brushing in removing dental plaque. The direct effects of oral hygiene techniques were relatively small compared with the indirect effects via their different performances in plaque removal. These results showed that improved oral hygiene performance can reduce gingival inflammation and therefore yield greater pocket reduction and less bleeding on probing. The lack of a substantial direct effect of Brush indicated that the greater pocket depth reduction in persons using interdental brushing techniques was unlikely to be explained by the mechanical depression of interdental papillae and gingival recession caused by interdental brushing, as suggested in the literature (Christou et al., 1998). Bleeding on probing is usually considered as a measure of periodontitis and is associated with the inflammatory change at the base of periodontal pockets (Christou et al., 1998), and it is therefore not surprising to find a small direct effect of Brush on bleeding on probing. Nevertheless, although oral hygiene measures are not active periodontal treatments, reduction in bleeding on probing could still be achieved by the improvement in plaque removal. This was shown in the indirect effect of Brush on the change in bleeding on probing. Pocket depth and bleeding on probing at baseline and the changes of these two variables were positively correlated, indicating that persons with deep pocketing were also more likely to show bleeding on probing, and those persons with greater reduction in pocket depth also showed greater improvement in bleeding on probing. In contrast, smoking seemed to have a direct effect upon bleeding on probing, but little effect on both changes in pocket depth and plaque indices, directly or indirectly. Smokers tended to have a lower percentage of sites with bleeding on probing, but showed less improvement in bleeding on probing. This is comparable with the findings from previous studies, which showed that although smoking is a risk factor for advanced chronic periodontal diseases, smokers tended to exhibit less marginal gingival inflammation and bleeding at baseline and showed reduced response to periodontal treatments (Grossi et al., 1997; Renvert et al., 1998; Kinane and Chestnutt, 2000). These results demonstrated that our models have a good overall model fit statistically, and the relationships among the variables in these models are consistent with both biological theory and clinical knowledge. In conclusion, in this study, we analyzed the relationships among variables in the aggregated data at the person-level, and demonstrated direct and indirect effects. Our research on using LGCM to analyze multilevel data is ongoing.
The second author was self-funded. The other authors were funded by the United Kingdom government’s Higher Education Funding Council for England (HEFCE). The first author currently holds a UK Research Council Fellowship. The authors thank the three companies who provided support: Colgate-Palmolive, for toothbrushes, dental floss, and toothpaste; Dentsply, for dental instruments; and Dental Health Boutique, Oral Healthcare, Leatherhead, United Kingdom, for interdental brushes. The authors’ work is independent of the funding resources.
A supplemental appendix to this article is published electronically only at http://jdr.iadrjournals.org/cgi/content/full/87/11/1037/DC1. Received for publication May 8, 2007. Revision received June 5, 2008. Accepted for publication July 5, 2008.
Journal of Dental Research, Vol. 87, No. 11,
1037-1042 (2008)
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
|
|
 |
|
Sign In to gain access to subscriptions and/or personal tools.
TOP
ABSTRACT
INTRODUCTION
Plaque Index 
