## Abstract

## Introduction

With a growing number of individuals retaining their teeth lifelong, often with periodontitis-induced root surface exposure, there is the need for cost-effective management strategies for root caries lesions. The present study aimed to assess the cost-effectiveness of root caries preventive treatments.

## Methods

Patients were simulated over 10 years using a Markov model. Four treatments were compared: No treatment, daily 225–800 ppm fluoride rinses, chlorhexidine (CHX) varnish (2×/year), silver diamine fluoride (SDF) varnish (2×/year). Data from a systematic review were submitted to network meta-analysis for inferring relative efficacies of treatments. The health outcome was years of teeth being free of root caries. A mixed public-private payer perspective within 2016 German healthcare was taken, with costs being estimated from fee item catalogues or based on market prices. Populations with different numbers of teeth and tooth-level risks were modelled. Monte-Carlo microsimulations, univariate- and probabilistic sensitivity analyses were performed.

## Results

In populations with 16 teeth at risk and low tooth-level risk for root caries, providing no preventive treatment was least costly, but also least effective (130 Euro, 144 years). SDF ranked next, being more costly (180 Euro), but also more effective (151 years). Payers willing to invest 8.30 Euro per root caries-free tooth-year found SDF most cost-effective. CHX varnish and fluoride rinse were not cost-effective. In populations with more teeth and high tooth-level risk, SDF was the most effective and least costly option.

## Conclusions

Root caries preventive treatments (like SDF) are effective and might even be cost-saving in high risk populations.

## Clinical significance

Application of SDF can be recommended as a cost-saving treatment for prevention of root caries in patients with high risk of root caries.

## 1

## Introduction

The prevalence and incidence of root caries lesions are high and increasing in many countries , as teeth are often retained lifelong, whilst being affected by periodontal disease and recession, with exposed root dentin being at risk for root caries induction, especially in older individuals with impaired dexterity, oral hygiene and reduced salivary flow . The treatment of root caries lesions is challenging, as restoration placement is hampered by difficulties in moisture control, dentin being the sole adhesive substrate, and lacking retention in saucer-shaped root cavities. Moreover, a growing share of patients who develop root caries have a limited mobility, with traditional restorative treatment often not being possible . Consequently, there is great need for effective root caries prevention strategies. A number of preventive treatments are available, like daily (self-applied) fluoride rinses, or professionally applied protective varnishes like chlorhexidine (CHX), which exerts antibacterial effects, or silver diamine fluoride (SDF), which is both remineralizing and antibacterial .

A recent review confirmed the efficacy of fluoride rinses or CHX or SDF varnishes for root caries prevention , but also found great uncertainty as to which of these treatments is preferable. One issue, which should be considered alongside clinical efficacy, is cost-effectiveness: Preventive measures incur costs, but might reduce the need for (costly) restorative measures, which might additionally fail repeatedly. If prevention is cost-saving, public health decision makers might want to actively support the provision of root caries prevention. Moreover, certain preventive treatments are likely to be more costly than others, but also have a different efficacy. Third, the application of preventive treatments might be more cost-effective in high risk patients, or patients with a large number of teeth (with preventive costs being distributed between teeth, and higher cost-savings when avoiding root caries). The present study aimed to assess the cost-effectiveness of different root caries preventive treatments.

## 2

## Methods

## 2.1

## Mode and horizon

A Markov model was used to follow a patient over 10 years (TreeAge Pro 2014, TreeAge Software, Williamstown, MA, USA). Populations with different numbers of teeth and tooth-level risk of root caries were modelled. The initial health state was always a tooth without root caries ( Fig. 1 ). The tooth could experience root caries, the probability of which was altered by various preventive treatments being applied (see below). A root caries lesion was treated with a one- to two-surfaced cervical glass ionomer restoration. This restoration could fail due to various reasons, which resulted in re-treatment. The risk of such restoration failure was estimated based on data from a randomized controlled trial .

## 2.2

## Setting and perspective

This study adopted a mixed private-payer-perspective in the context of German healthcare.

## 2.3

## Comparators

Four preventive treatments were compared: No treatment, 225–800 ppm fluoride rinses, self-applied daily (assuming 15 ml used for each daily rinse), CHX varnish applied twice yearly by a dentist or a dental care professional, SDF varnish, also applied twice yearly. The efficacy of the different treatments was estimated based on a recent systematic review . For our analysis, only trials comparing a minimum of one of the described treatments against each other or any (placebo) control, which reported on the mean (SD) caries increment, were included (Table S1). To make use of the fact that treatments were compared both directly and indirectly (Fig. S1), network meta-analysis was used for inferring relative efficacies between treatments . Details on the methodology and the mean differences between treatments can be found in the appendix (Table S2). We transformed mean differences into risk ratios (RR), i.e. the risk of a caries increment when applying one of the preventive treatments compared with untreated controls.

## 2.4

## Health outcomes

The health outcome was years of teeth being free of root caries (e.g. in patients with 16 teeth, a maximum of 160 undiscounted years of caries-free teeth could be achieved over the 10 years). The health outcome was estimated based on the risk of a tooth experiencing root caries if no treatment was applied. This was based on studies included in an updated systematic review on risk indicators for root caries (Table S3). These studies all reported on patient-level root caries incidence, with the sample-size weighted annual incidence being 24% (range: 17–44%). We assumed each individual in these studies to have had a mean of 20 teeth at risk (as this was the mean number of teeth retained in individuals aged 65–74 years old in Germany in the Fourth German Oral Health Survey) . This resulted in an annual tooth-level incidence of 1.2% (0.9–2.2%).

## 2.5

## Costs and discounting

German healthcare is characterized by a two-tiered insurance system. The majority of patients (87%) are enrollees of the public (statutory) insurance , while the remaining population is privately insured. Reimbursements for treatments follow a fee-per-item system. For members of the public insurance, most dental treatments are fully covered, with fees being defined according to the public fee catalogue Bewertungsmaßstab (BEMA). Treatments which are not covered (like preventive treatments in adults) are fully or partially paid privately by the patient (out-of-pocket expense), or reimbursed by additional private insurances. For estimating costs of these treatments, the private fee catalogue Gebührenkatalog (GOZ) is used .

This study employed a mixed public-private payer perspective, as is most common in Germany. Costs were estimated from BEMA, and only if not reimbursed by the public insurance, fee items of GOZ were used. For GOZ, factoring of item-points is common to determine prices, allowing to individualize fees. The present analysis used the standard multiplication factor (×2.3). Calculating costs using average national multiplication factors has not been found to greatly impact on cost-effectiveness . Note that in this framework, costs are identical if dentists or dental care professionals perform treatments.

We estimated the following costs via fee items: Application of varnish, restoration placement, and associated measures like anesthesia and moisture control. Fee items cover both staff and material costs. In contrast, costs for fluoride rinses will be paid privately by patients, with only material costs incurring. We estimated these using market prices in 2016. We assumed 15 ml of rinse being used daily. Details on cost estimations can be found in detail in the appendix (Table S4 and S5). Costs were calculated in 2016 Euro, and future costs and effectiveness were discounted at 3% per annum . Discounting accounts for opportunities forgone if spending money now instead of later, or gaining health benefits later instead of now .

## 2.6

## Populations

As described, different populations were modelled to assess the effects of heterogeneity on cost-effectiveness. First, we used the updated review on risk indicators of root caries incidence to assess how having more teeth at risk, or having had (any or high) previous root caries experience was associated with root caries incidence. Only studies which had performed multivariable regression analysis, with the outcome being root caries incidence, reporting a risk estimator (risk or odds ratio), an estimator of variance (confidence intervals, SD, SE, p-values) and a sample size were included in our estimation (Table S3). Two meta-analyses were performed (details can be found in the appendix). The risk ratio and 95% confidence interval for root caries incidence in patients having a high number of teeth or surfaces at risk (compared with those having low or no teeth or surfaces at risk) was 1.55 (95% CI: 1.39/1.73). Similarly, patients with (some or extensive) previous root caries experience (compared with those with no or minimal experience) had 1.47 (95% CI: 1.16/3.21) times the risk of root caries incidence. Note that while a number of different thresholds or definitions had been used for each risk indicator, there was low heterogeneity.

Implementation of these estimates into the model was performed as follows. We first constructed two groups of patients, one with a high number of teeth at risk (24 teeth), one with a low number of teeth at risk. This assumption was supported by epidemiologic data on elderly Germans (65–74 years old) from the Fourth German Oral Health Study (DMS IV) . If assuming that tooth-level risk for root caries is equal in these groups, patient-level risk differs by 1.5, i.e. in line with our meta-analytical estimates. Second, we modelled each of these two groups to either have low or high root caries experience, with the tooth-level risk for root caries being adjusted according to the estimate yielded by our meta-analysis (i.e. 1.47 [1.16/3.21]). That resulted in an annual tooth-level risk of 1.0% (0.7–1.9%) in low risk populations and 1.5% (1.1–2.6%) in high-risk populations, respectively.

## 2.7

## Analytical methods

Simulations were performed in discrete annual cycles. Cost-effectiveness was calculated as costs per years of teeth remaining root caries free. To calculate cost-effectiveness, we performed Monte-Carlo microsimulations. If possible, parameter uncertainty was introduced by randomly sampling from triangular or uniform distributions between 5 and 95% percentiles or ranges, as yielded by included studies or own calculations . Parameters submitted to this probabilistic sensitivity analysis were transition probabilities and costs for fluoride rinse.

To compare strategies, alternatives were first ranked according to their costs, with the least expensive option being ranked first. If the second ranked option was more expensive and less effective than the comparator, it was considered *dominated *. In case the second ranked option was more expensive but more effective, it was considered *undominated *. In the first case, incremental costs-effectiveness-ratios (ICER: cost difference per effectiveness difference) were negative (higher costs per lost effectiveness), in the latter case, the ICER was positive .

The net benefit (NB) of each treatment was calculated using the formula NB = λ × Δe − Δc , with λ denoting the ceiling threshold value of willingness-to-pay, i.e. the additional maximal costs a decision-maker is willing to sacrifice to gain one additional unit of effectiveness . If λ > Δc/Δe, an alternative is considered more cost-effective than the comparator despite possibly being more costly . Using this approach, cost-effectiveness-acceptability curves were generated, plotting the probability of being cost-effective against a decision-maker’s ceiling threshold.

To additionally check the robustness of findings, univariate sensitivity analyses were performed. First, annual tooth-level risk of root caries were increased to 3.0% (i.e. annual root caries increments were 0.72 and 0.48 on patient level in the groups with high and low number of teeth at risk, respectively). Given that other longitudinal studies reported annual increments of 0.4 teeth or surfaces , these estimate represent higher-risk populations (while our basecase estimates are at the lower end of the risk spectrum). Second, costs for fluoride rinse were set to zero (simulating the public-payer perspective). Third, application of varnishes 4×/year were modelled, as this is recommended in certain patients , but would increase the costs. Fourth, the risk of restoration failure was varied between 50 and 200% relative to the basecase, mainly as only one short-term randomized trial was underlying our estimate on restoration failure. Last, discounting rates were varied between 0 and 5%.

## 2

## Methods

## 2.1

## Mode and horizon

A Markov model was used to follow a patient over 10 years (TreeAge Pro 2014, TreeAge Software, Williamstown, MA, USA). Populations with different numbers of teeth and tooth-level risk of root caries were modelled. The initial health state was always a tooth without root caries ( Fig. 1 ). The tooth could experience root caries, the probability of which was altered by various preventive treatments being applied (see below). A root caries lesion was treated with a one- to two-surfaced cervical glass ionomer restoration. This restoration could fail due to various reasons, which resulted in re-treatment. The risk of such restoration failure was estimated based on data from a randomized controlled trial .

## 2.2

## Setting and perspective

This study adopted a mixed private-payer-perspective in the context of German healthcare.

## 2.3

## Comparators

Four preventive treatments were compared: No treatment, 225–800 ppm fluoride rinses, self-applied daily (assuming 15 ml used for each daily rinse), CHX varnish applied twice yearly by a dentist or a dental care professional, SDF varnish, also applied twice yearly. The efficacy of the different treatments was estimated based on a recent systematic review . For our analysis, only trials comparing a minimum of one of the described treatments against each other or any (placebo) control, which reported on the mean (SD) caries increment, were included (Table S1). To make use of the fact that treatments were compared both directly and indirectly (Fig. S1), network meta-analysis was used for inferring relative efficacies between treatments . Details on the methodology and the mean differences between treatments can be found in the appendix (Table S2). We transformed mean differences into risk ratios (RR), i.e. the risk of a caries increment when applying one of the preventive treatments compared with untreated controls.

## 2.4

## Health outcomes

The health outcome was years of teeth being free of root caries (e.g. in patients with 16 teeth, a maximum of 160 undiscounted years of caries-free teeth could be achieved over the 10 years). The health outcome was estimated based on the risk of a tooth experiencing root caries if no treatment was applied. This was based on studies included in an updated systematic review on risk indicators for root caries (Table S3). These studies all reported on patient-level root caries incidence, with the sample-size weighted annual incidence being 24% (range: 17–44%). We assumed each individual in these studies to have had a mean of 20 teeth at risk (as this was the mean number of teeth retained in individuals aged 65–74 years old in Germany in the Fourth German Oral Health Survey) . This resulted in an annual tooth-level incidence of 1.2% (0.9–2.2%).

## 2.5

## Costs and discounting

German healthcare is characterized by a two-tiered insurance system. The majority of patients (87%) are enrollees of the public (statutory) insurance , while the remaining population is privately insured. Reimbursements for treatments follow a fee-per-item system. For members of the public insurance, most dental treatments are fully covered, with fees being defined according to the public fee catalogue Bewertungsmaßstab (BEMA). Treatments which are not covered (like preventive treatments in adults) are fully or partially paid privately by the patient (out-of-pocket expense), or reimbursed by additional private insurances. For estimating costs of these treatments, the private fee catalogue Gebührenkatalog (GOZ) is used .

This study employed a mixed public-private payer perspective, as is most common in Germany. Costs were estimated from BEMA, and only if not reimbursed by the public insurance, fee items of GOZ were used. For GOZ, factoring of item-points is common to determine prices, allowing to individualize fees. The present analysis used the standard multiplication factor (×2.3). Calculating costs using average national multiplication factors has not been found to greatly impact on cost-effectiveness . Note that in this framework, costs are identical if dentists or dental care professionals perform treatments.

We estimated the following costs via fee items: Application of varnish, restoration placement, and associated measures like anesthesia and moisture control. Fee items cover both staff and material costs. In contrast, costs for fluoride rinses will be paid privately by patients, with only material costs incurring. We estimated these using market prices in 2016. We assumed 15 ml of rinse being used daily. Details on cost estimations can be found in detail in the appendix (Table S4 and S5). Costs were calculated in 2016 Euro, and future costs and effectiveness were discounted at 3% per annum . Discounting accounts for opportunities forgone if spending money now instead of later, or gaining health benefits later instead of now .

## 2.6

## Populations

As described, different populations were modelled to assess the effects of heterogeneity on cost-effectiveness. First, we used the updated review on risk indicators of root caries incidence to assess how having more teeth at risk, or having had (any or high) previous root caries experience was associated with root caries incidence. Only studies which had performed multivariable regression analysis, with the outcome being root caries incidence, reporting a risk estimator (risk or odds ratio), an estimator of variance (confidence intervals, SD, SE, p-values) and a sample size were included in our estimation (Table S3). Two meta-analyses were performed (details can be found in the appendix). The risk ratio and 95% confidence interval for root caries incidence in patients having a high number of teeth or surfaces at risk (compared with those having low or no teeth or surfaces at risk) was 1.55 (95% CI: 1.39/1.73). Similarly, patients with (some or extensive) previous root caries experience (compared with those with no or minimal experience) had 1.47 (95% CI: 1.16/3.21) times the risk of root caries incidence. Note that while a number of different thresholds or definitions had been used for each risk indicator, there was low heterogeneity.

Implementation of these estimates into the model was performed as follows. We first constructed two groups of patients, one with a high number of teeth at risk (24 teeth), one with a low number of teeth at risk. This assumption was supported by epidemiologic data on elderly Germans (65–74 years old) from the Fourth German Oral Health Study (DMS IV) . If assuming that tooth-level risk for root caries is equal in these groups, patient-level risk differs by 1.5, i.e. in line with our meta-analytical estimates. Second, we modelled each of these two groups to either have low or high root caries experience, with the tooth-level risk for root caries being adjusted according to the estimate yielded by our meta-analysis (i.e. 1.47 [1.16/3.21]). That resulted in an annual tooth-level risk of 1.0% (0.7–1.9%) in low risk populations and 1.5% (1.1–2.6%) in high-risk populations, respectively.

## 2.7

## Analytical methods

Simulations were performed in discrete annual cycles. Cost-effectiveness was calculated as costs per years of teeth remaining root caries free. To calculate cost-effectiveness, we performed Monte-Carlo microsimulations. If possible, parameter uncertainty was introduced by randomly sampling from triangular or uniform distributions between 5 and 95% percentiles or ranges, as yielded by included studies or own calculations . Parameters submitted to this probabilistic sensitivity analysis were transition probabilities and costs for fluoride rinse.

To compare strategies, alternatives were first ranked according to their costs, with the least expensive option being ranked first. If the second ranked option was more expensive and less effective than the comparator, it was considered *dominated *. In case the second ranked option was more expensive but more effective, it was considered *undominated *. In the first case, incremental costs-effectiveness-ratios (ICER: cost difference per effectiveness difference) were negative (higher costs per lost effectiveness), in the latter case, the ICER was positive .

The net benefit (NB) of each treatment was calculated using the formula NB = λ × Δe − Δc , with λ denoting the ceiling threshold value of willingness-to-pay, i.e. the additional maximal costs a decision-maker is willing to sacrifice to gain one additional unit of effectiveness . If λ > Δc/Δe, an alternative is considered more cost-effective than the comparator despite possibly being more costly . Using this approach, cost-effectiveness-acceptability curves were generated, plotting the probability of being cost-effective against a decision-maker’s ceiling threshold.

To additionally check the robustness of findings, univariate sensitivity analyses were performed. First, annual tooth-level risk of root caries were increased to 3.0% (i.e. annual root caries increments were 0.72 and 0.48 on patient level in the groups with high and low number of teeth at risk, respectively). Given that other longitudinal studies reported annual increments of 0.4 teeth or surfaces , these estimate represent higher-risk populations (while our basecase estimates are at the lower end of the risk spectrum). Second, costs for fluoride rinse were set to zero (simulating the public-payer perspective). Third, application of varnishes 4×/year were modelled, as this is recommended in certain patients , but would increase the costs. Fourth, the risk of restoration failure was varied between 50 and 200% relative to the basecase, mainly as only one short-term randomized trial was underlying our estimate on restoration failure. Last, discounting rates were varied between 0 and 5%.