Association Between Medicaid Dental Payment Policies and Children’s Dental Visits, Oral Health, and School Absences

Key Points Question Are increases in the ratio of Medicaid payment rates to dentist charges for an index of services associated with improvements in children’s outcomes? Findings This cross-sectional study used a difference-in-differences analysis to evaluate 15 738 Medicaid-enrolled and 16 867 privately insured children aged 6 to 17 years who participated in the 2016-2019 National Survey of Children’s Health. Increasing the Medicaid fee ratio was associated with significant but modest improvements in children’s dental visits and oral health and had no significant association with school absences. Meaning More generous Medicaid dental payment policies are associated with improvements in children’s outcomes.


eAppendix 1. Description of Fee Ratio Construction
We constructed an index of Medicaid fees using the most commonly billed codes for children's dental visits according to a recent American Dental Association (ADA) analysis. Services included in order of their weight in the index were periodic oral evaluationestablished patient (D0120), child prophylaxis (D1120), adult prophylaxis (D1110), topical fluoride application (D1208), sealant application (D1351), bitewing radiographic images (D0272 and D0274), topical application of fluoride varnish (D1206), and periapical intraoral radiographic images (D0220 and D0230). Medicaid payment rates for each state and year were obtained from fee schedules posted on state Medicaid websites. Since we only observed calendar year in the NSCH, we weighted state-level payment rates for each service by the number of months they were in effect in each year.
To account for geographic differences in the market for dental services and changes over time, we deflated this index by dentist-reported charges for the same services from the 2016, 2018, and 2020 ADA Survey of Dental Fees. The ADA's survey reports fees for 9 geographic regions in each year. We used linear interpolation between years to compute these charges by region for 2017 and 2019.

eAppendix 2. Sample Exclusions and Sample Sizes
The sample included publicly insured children and a control group of privately insured children.
Public health insurance coverage included "Medicaid, medical assistance, or any kind of governmental assistance plan for those with low incomes or a disability." Children who reported public health insurance alone or in combination with a second source of coverage were considered publicly insured. The sample was restricted to children ages 6-17 since our analysis examined school absences. We also limited the sample to children with family incomes up to 300% of the federal poverty level to increase comparability between the treatment group of Medicaid-enrolled children and the control group including privately insured children. Before making any additional exclusions, the sample included 36,921 children, including 17,866 publicly insured children and 19,055 privately insured children.
We excluded children who resided in Colorado, Delaware, Idaho, and Tennessee because Medicaid payment rate information was not available for these states (3,041 children). After making this exclusion, the sample included 33,880 children, including 16,380 children with public health insurance and 17,500 children with private health insurance. We also excluded 1,275 children (3.8%) who had missing information for any of the demographic characteristics included as controls in our regression models. The final analysis sample included 32,605 children, including 15,738 with public health insurance and 16,867 with private health insurance.

eAppendix 3. Description of Regression Model and Outcome Variables
Our main regression model was of the following form: Where is the outcome of interest for child i, who resides in state s, and is observed during interview year t; is the state-by-year continuous fee ratio, and is a binary indicator for Medicaid coverage. represents child and household characteristics and state-by-year policy variables, represents state indicators, and represents year indicators. The term is an error term, which is clustered by state. All models are weighted using sampling weights available from the National Survey of Children's Health, and models are estimated using linear probability models.
The coefficient of interest is , which provides the difference-in-differences estimate for the interaction term between the continuous fee ratio and the indicator for Medicaid coverage. Outcome variables included children's past-year preventive dental visits (at least 1 and at least 2), excellent oral health, and school absences (at least 4 and at least 7). Response options for the question about children's oral health status included excellent, very good, good, fair or poor. Response options for the question about children's school absences included none, 1-3 days, 4-6 days, 7-10 days, and 11 or more days. Table S1 presents associations between the fee ratio and each sample characteristic to test if the fee ratio is systematically correlated with the sample's composition. A separate regression was run for each sample characteristic as an outcome variable. The regressors included the fee ratio and its interaction with the Medicaid coverage indicator, state, and year fixed effects.

eAppendix 4. Description of Supplementary Tables and Figures
Regressions were weighted and model errors were clustered at the state level, as in our main analysis. Significant coefficients on the interaction term between the fee ratio and the Medicaid coverage indicator would suggest that the composition of Medicaid-enrolled children was changing relative to that for privately-insured children with changes to the fee ratio, which could bias our main difference-in-differences results. As shown in Table S1, none of the associations with the sample characteristics were statistically significant at conventional levels.  Table S3 shows the results of estimating our main models on specific preventive services that we considered as secondary outcomes. These outcomes include past-year dental cleaning, toothbrushing instructions, dental x-rays, fluoride treatment, and dental sealants. Our findings indicate a significant and positive association between the fee ratio and dental cleanings, toothbrushing instructions, and fluoride treatments. Associations with dental x-rays and sealants were not significant at conventional levels. Table S4 replicates our main regression analysis for the preventive dental visit outcome including children of all ages rather than restricting to those ages 6-17. Overall, estimates were similar to our main results when considering children of all ages. One important difference was that the estimate for any preventive dental visit for non-Hispanic Black children was statistically significant when including children of all ages (0.36 percentage point increase, P=0.04), which may be because of the increase in sample size. Table S5 presents the results of our test for preexisting outcome trends. To conduct this analysis, we augmented our main regression model by including the future change in the fee ratio and its interaction with the Medicaid indicator as explanatory variables. For example, the future change in the fee ratio for 2016 observations was the difference between the 2017 and 2016 fee ratio in the state of residence. A significant coefficient on the interaction terms would suggest a violation of the parallel trends assumption required for difference-in-differences analysis. As shown in Table S5, none of these coefficients were statistically significant at conventional levels. Table S6 presents the results of separate regressions that include observations from states that experienced a fee ratio change in 2017, 2018, and 2019, respectively, in addition to states that did not experience a change in any year. We defined a change in the fee ratio to be any change of at least 2 percentage points in absolute value. This value was selected to include most changes in the fee ratio while avoiding considering as changes very minor fluctuations that occur in most states in each year. It was necessary to define a threshold value for fee ratio changes to determine which states had changes in each year and which states could serve as control states.
We then re-estimated our main regression including all control variables except only including states with a change in the given year and states with no change in any year. The fee ratio was still included as a continuous variable. The first column of Table S6 shows the results of this exercise when only including states with a change of at least 2 percentage points in absolute value between 2016 and 2017, and states with no change larger than 2 percentage points in any year as a comparison group. Estimates shown in the table are the coefficients on the interaction between the continuous fee ratio and the Medicaid indicator. Results were similar to our main analysis including all states both qualitatively and quantitatively. The estimates for the preventive dental visit variables remained statistically significant at conventional levels. The estimate for excellent oral health was similar in magnitude to our main results but was no longer statistically significant at conventional levels.
The second and third columns repeat this analysis for states with changes in 2018 and 2019, respectively. While estimates are no longer statistically significant, estimates for the dental visit variables remain positive and for the at least 2 visits variable, are of a similar magnitude. It is not surprising that these estimates are no longer statistically significant since it was necessary to exclude many states with policy changes, and also because it may take time for changes in the fee ratio to influence outcomes. For example, fee ratio changes that occur in 2019 may not have immediate effects on children's dental visits and oral health in 2019, the final year of survey data in our study.
Overall, these results do not suggest that our main findings are biased by variation in treatment timing, Despite reducing our power by dividing states by the year of policy change and the need to define a specific threshold for policy changes, results comparing states with changes in 2017 to non-changer states are very similar to our main estimates and results for the 2018 and 2019 analyses are qualitatively similar for the dental visit variables. were missing Medicaid dental payment information. Results for the dental visit outcomes were statistically significant and larger in magnitude than our main analysis including all states. For example, when limiting to fee-for-service states, we estimated that a 1 percentage point increase in the fee ratio was associated with increases of 0.24 and 0.66 percentage points in at least 1 and at least 2 dental visits, respectively. By comparison, our main estimates for these outcomes were 0.18 and 0.27 percentage points, respectively. In contrast, the estimate for excellent oral health was somewhat smaller in magnitude and no longer statistically significant (P=0.16). Table S8 replaces the continuous fee ratio with indicators for having a fee ratio above or below the 75 th and 50 th percentiles of the distribution, respectively. These indicators are significantly associated with increases in dental visits and excellent oral health, which is consistent with our main results. Also similar to our main results, associations with school absences are not significant at conventional levels. Table S9 replicates our main analysis except the fee ratio is replaced with the inflationadjusted Medicaid fee index in dollars. The consumer price index was used to adjust fees for inflation. Findings are qualitatively similar to our main results, as shown in the Table. We estimate that a $1 increase in the inflation-adjusted fee index is associated with increases of 0.31, 0.45, and 0.30 percentage points in at least 1 preventive visit, at least 2 preventive visits, and excellent oral health, respectively. Similar to our main analysis, associations with the school absences outcomes are negative but not statistically significant at conventional levels. Table S10 presents the results of logistic regressions. Coefficients shown are odds ratios for the interaction between the continuous fee ratio and the Medicaid indicator. Estimates suggest positive and significant associations between the fee ratio and preventive dental visits (at least 1 and at least 2) and excellent oral health. Table S11 presents the results of several sensitivity analyses. The first column shows the results of including a control for family income as a percent of the federal poverty level in our main models, the second shows the results of excluding 6 states that had a change in Medicaid adult dental coverage during our study period (CA, CO, CT, ID, IL, LA) based on data from the Center for Health Care Strategies, and the third shows the results of excluding foreign born children. In general, the results of these sensitivity tests were very similar to our main results. Table S12 presents results when restricting the sample to Medicaid-enrolled children.
These regressions were similar to our main analysis, except the coefficient of interest is the fee ratio rather than its interaction with Medicaid status. Estimates are larger in magnitude with larger standard errors relative to the analysis including a control group, which is not surprising given smaller sample sizes and lower power to detect significant effects. The coefficient estimates for the dental visit and excellent oral health outcomes are positive but no longer statistically significant. Estimates for the absences variables are negative, and the result for four or more school absences is statistically significant (P=0.008).
Finally, Figures S1-S4 plot trends in all outcomes except for at least 1 dental visit (main text Figure 2) in the 9 states that first had a change of at least 2 percentage points in the fee ratio in 2018 or 2019, as described in the main text. In general, trends appear similar for Medicaidenrolled and privately-insured children, with the exception of at least 4 school absences.
However, estimates have wide confidence intervals and it is not possible to rule out similar pretrends. These visual representations complement our regression-based analysis of pretrends shown in Appendix  Notes: Estimates shown in the table are in terms of percentage points. The estimates from each column come from separate regression models where the sample is restricted to states with a fee ratio change in the year indicated and states with no fee ratio change in any year, as described in Appendix D. Difference-in-differences estimates represent the effect of a 1 percentage-point increase in the fee ratio. Models include state and year fixed effects in addition to child and time-varying state-level controls (unemployment rate, Medicaid managed care penetration rate, dentist supply per capita, Medicaid income eligibility limit for working parents Notes: Estimates shown in the table are in terms of percentage points and are for the coefficient on the continuous fee ratio variable. Difference-in-differences estimates represent the effect of a 1 percentage-point increase in the fee ratio. Models include state and year fixed effects in addition to child and time-varying state-level controls (unemployment rate, Medicaid managed care penetration rate, dentist supply per capita, Medicaid income eligibility limit for working parents). Estimates are weighted and model errors are clustered at the state level. The sample was restricted to publicly insured children.