We have proposed an innovative approach for assessing dental utilization. The proposed hurdle model allowed us to characterize differences between adults with and without diabetes on three dimensions of dental utilization: the use of dental services, the type of dental care used, and the intensity of each type of service. This model allowed us to decompose dental utilization into multiple components and to examine the impact of diabetes on each of these components separately. This decomposition allowed us to determine whether diabetes impacts the use of services by increasing or decreasing the probability of using any dental services or whether diabetes differentially impacts specific classes of treatment or the frequency with which specific classes of treatment are used. From the perspective of planning the availability of dental services, this is a useful approach because it gives us the ability to identify whether utilization is impacted at the person level or whether the intensity of use of certain procedures is affected.
It is interesting to note that the prevalence of diabetes mellitus in this population was 8.24%, which closely approximates the CDC-stated prevalence in the United States. Our study indicated that individuals with diabetes were less likely to visit a dentist than similar individuals without diabetes. This finding emphasizes the need to increase awareness and support for oral health care among diabetes patients. A similar pattern was reported by Tomar et al. 2000 , in a study on dentate adults that concluded that adults with diabetes were less likely than those without to have seen a dentist (65.8% vs. 73.1%) within the 12 month study period. Over the 5-year study period, our results showed a more minor, though statistically significant, difference (84% vs. 88%).
Our results suggest that adults with diabetes receiving dental care had an increased risk of oral complications as evidenced by substantial differences in the mix of services between the two groups. They were more likely to receive periodontal care – maintenance and non-surgical procedures, tooth extractions, and tooth replacing procedures like removable prosthetics. This is not surprising given the established link between diabetes and periodontitis and between periodontitis and tooth-loss.
Taylor et al. 2008 , in its review of adverse effects of diabetes on periodontal health, identified 13/17 observational studies that provided consistent evidence of greater prevalence, severity, extent, or progression of at least one manifestation of periodontal disease. Additionally, evidence from three observational studies supported periodontal disease increasing the risk for diabetes complications and no published reports refuted the findings. Kapp et al. 2007 , in a survey-based one-year study on a national sample of dentate adults, found that prevalence of tooth-loss among adults with diabetes was 38% and that adults with diabetes had 1.5 times greater odds of having at least one tooth removed compared to non-diabetics. These national level estimates of tooth loss within a single year are higher than our estimates of tooth-loss prevalence (33%) and odds ratio (1.4) in a five year period in the state of Washington. In addition to differences in the study populations, the survey based design of the earlier study may also contribute to differences between the results of the two studies. Studies from Bacic et al [34, 35] showed the number of extracted teeth to be significantly greater in the diabetic group than in the control group (14.1 vs. 10.4). Other studies have postulated tooth loss as an inevitable result of periodontal disease that is difficult to control [36–39]. Given that we measured incidence of extractions during the study period and we did not have a diagnosis measure of baseline edentulism among the study participants, we do not have a direct means of comparing our estimates with that of the earlier study. Most of the difference in our tooth loss estimates (measured by number of extractions) between the two groups came from more diabetes patients receiving an extraction rather than more extractions being performed per patient. Furthermore, our findings of higher odds of prosthetics and lower odds of restorations among patients with diabetes are consistent with their increased tooth loss [40–43].
Upon evaluating preventive oral health services, we found that among patients using dental services, those with diabetes were less likely than matched controls to receive overall preventive dental cleanings that included both prophylactic and periodontal maintenance services. When viewed individually, patients with diabetes were less likely to receive prophylaxes but more likely to receive periodontal maintenance codes that unlike the former represent procedures used in the treatment of active oral disease or its maintenance once stabilized . In addition, they received significantly fewer diagnostic procedures. Given previous findings that adults with diabetes had significant oral complications and that regular care provides opportunities for identification, prevention and treatment of periodontal disease, this finding suggests a significant prevention gap and emphasizes the episodic nature of use of dental care in this group.
Some of the statistically significant differences in Table 2 are quite small. The statistical significance of these results may be attributable to our large sample size. Given the substantial and increasing prevalence of diabetes in the population (8.24% in our cohort), even small significant differences in highly used procedures like diagnostic services and preventive care or in less utilized but more expensive procedures like fillings, extractions or removable prosthetics are likely to have meaningful effect on population oral health and health care cost and utilization.
Our study has several additional strengths. First, we have used a propensity score matching method which provides more flexible and robust adjustment for confounding than regression methods provide when the distribution of potential confounding variables differs substantially between comparison groups. As anticipated, adults with diabetes differed significantly from control patients at baseline. After matching procedures were completed, the two groups were found to be similar in the distribution of baseline characteristics. Second, unlike the traditional regression approach, propensity score matching allowed assessment of the sensitivity of results to possible hidden bias. Finally, the use of administrative data rather than self-reported oral care utilization eliminates the potential for recall bias present in survey-based studies.
The methodology used in this study has some limitations. First, the proposed hurdle model assumes independence between the use and intensity (counts) of dental services. This assumption may not be realistic. However, dependence between use and intensity that is attributable to covariates included in our matching procedure will be eliminated via matching. Second, the standard errors of post-matching estimates did not fully account for the estimation of propensity score matching weights, which were assumed fixed and known in these analyses. Replication methods (e.g. bootstrap) could be used to obtain standard error estimates that account for variability in the estimated propensity scores. However, bootstrapped standard errors are very unlikely to alter our results as the study employed a large sample size resulting in highly precise effect estimates. Because separate hypothesis tests were conducted for each class of dental utilization, significance tests should be considered with reference to the number of hypothesis tests that were conducted. Third, because our cohort consisted of patients with well-established diabetes, it is possible that these patients had severe periodontal disease or edentulism at baseline contributing to adverse oral outcomes or fewer dental visits during the study period. The claims data used contained no diagnosis data that detailed periodontal disease severity or the number of teeth at the baseline. However, the results from sensitivity analyses suggest that our inference for most outcomes is robust unless the effects of unobserved confounders such as baseline edentulism were high enough to approximately double the odds of diagnosis of diabetes in our study population. For a few outcomes such as restorations, our results were sensitive to moderate effects of unobserved confounders. The possibility of bias due to unobserved confounding is a limitation of observational studies.
The results of this study show that among adults seeking dental care, those with diabetes have higher utilization of dental surgeries and extractions and therefore may potentially benefit from increased prophylactic services. In future research, this could be directly tested via a randomized trial. Additionally, we plan to study differences in cost patterns associated with classes of dental services to gain insight into the economic consequences of the impact of diabetes on dental services utilization observed in this study.