Validity of Denture Usage Definitions Based on Claims Data in Japanese Older Adults

Abstract

Background:

Administrative claims data are increasingly used in oral health research, but their validity for identifying denture use has not been established.

Objective:

To evaluate the accuracy of algorithms based on dental claims codes for identifying denture users, dentist-reported oral health screening records were used as the reference standard.

Methods:

We analyzed data from 4,053 adults aged ≥65 y in the Longevity Improvement and Fair Evidence (LIFE) Study in Japan. Twelve algorithms incorporating 56 denture-related diagnosis and procedure codes were developed from claims in the 12 mo preceding the screening. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Sensitivity analyses used a 6-mo claims window. A post hoc bias analysis was also performed.

Results:

During screening, 59.9% of the participants were classified as denture users. Algorithms based solely on diagnosis codes showed very high specificity (98.3%–100%) but low sensitivity (1.4%–20.8%). Among the procedure-based definitions, oral rehabilitation codes (algorithm f; denture adjustment or instruction) had the highest sensitivity (63.6%). The combined algorithm using both diagnosis and procedure codes (algorithm C) achieved the best balance, with 65.3% sensitivity, 96.6% specificity, 96.6% PPV, and 65.1% NPV. Similar findings were observed using the 6-mo claims data. Bias analysis indicated that the risks for denture use could be underestimated by 26% to 59%, with algorithm C showing the least bias.

Conclusions:

Denture use can be identified from dental claims data with moderate accuracy. Algorithms combining denture-related diagnosis and procedure codes or using oral rehabilitation codes provide practical definitions for research on oral and systemic health outcomes.

Knowledge Transfer Statement:

The result of this validated study can be identified as denture use from dental claims with moderate accuracy. In combination with validated denture use and the number of remaining teeth, it can expand opportunities to assess oral health status and its relationship with systemic health outcomes in large-scale epidemiologic studies using claims data.

Background

Administrative claims data have been increasingly used in epidemiological and health service research in recent years because they provide large sample sizes, population representativeness, and insights into real-world effectiveness and clinical practice patterns (Schneeweiss and Avorn 2005). However, because claims are generated for fee-for-service reimbursements rather than for research purposes, data inaccuracies may lead to misclassification and bias (Terris et al 2007; Koram et al 2019). Therefore, validation against a reference standard is essential.

The use of dentures may mitigate the risk of various systemic health outcomes that increase with tooth loss. Previous studies have shown that compared with nondenture users, denture users have a lower risk of various outcomes associated with tooth loss, indicating physical or cognitive health, such as mortality, need for long-term care, and dementia (Kino et al 2024); psychological health, such as depressive symptoms (Nakazawa et al 2024) and low frequency of laughter (Tamada et al 2024b); and health behaviors, such as reduced protein intake (Kusama et al 2023). Denture use has also been reported to moderate the risk of incident pneumonia associated with dysphagia (Takeuchi et al 2019). In light of these findings, it has recently been suggested that denture use may also contribute to the maintenance of quality of life in older adults (Hoshi-Harada et al 2025).

Previous validation studies have demonstrated that although the number of remaining teeth can be ascertained using dental claims data (Ono et al 2021; Tsuneishi et al 2022; Tamada et al 2024a), whether denture use can be accurately identified has not yet been evaluated. If denture use could be reliably measured through administrative claims data, this would substantially expand the opportunities for oral health research. For example, it would enable large-scale studies of the role of prosthetic rehabilitation in aging populations, facilitate health service evaluations of access to and equity in denture provision, and support investigations into the impact of denture use on systemic health outcomes and long-term care needs.

The present study aimed to examine the validity of claims-based algorithms in identifying denture use in Japanese older adults. We compared claims-based definitions based on denture-related diagnosis and procedure codes with dentist-assessed denture usage from oral health care screening records, which served as the reference standard.

Methods

Setting

This validation study used the Longevity Improvement and Fair Evidence (LIFE) Study database from 1 Japanese municipality, covering the period from April 2018 to March 2020, which includes 2 health insurance (the National Health Insurance [NHI] and the Latter-Stage Older Persons Health Care System) claims data and oral health care screening data (Fukuda et al 2023).

In the Japanese health insurance system, NHI covers the self-employed or not employed, including part-time workers and retired individuals, and their dependents aged <75 y; the Later-Stage Older Persons Health Care System covers individuals aged ≥75 y. The details of the Japanese health insurance system are described elsewhere (Sakamoto et al 2018). This health insurance system covers various denture-related procedures, including the fabrication of new dentures and maintenance services such as adjustments, repairs, and relining. Regarding denture materials, coverage is limited to conventional dentures such as acrylic resin dentures and excludes metal plate dentures and implant-supported dentures (Zaitsu et al 2018). The present validity study targeted individuals wearing conventional dentures, such as removable partial dentures and complete dentures, not including fixed partial dentures.

In the target municipality, oral health care screening is provided free of charge to residents aged 65 y and older to promote their oral health and prevent the need for long-term care. The program was implemented in 2018 and had a participation rate of approximately 5%.

The target population included individuals who underwent oral health care screening in the fiscal year 2019, had at least 1 dental claim in the 12 mo prior to the screening month, and were continuously enrolled in either the NHI or the Latter-Stage Older Persons Health Care System during the study period (April 2018 to March 2020). The following records were excluded: nonassigned ID, duplicate records, and incorrect dates. A flowchart of the study participants is shown in the Figure.

Figure.

Flowchart of analyzed participants.

Reference Standard and Algorithms

The record of using or not using dentures, based on an assessment by a dentist during oral health care screening, was used as the reference standard.

We used 12 algorithms with 56 denture-related diagnosis and procedure codes in the 12 mo of dental claims data prior to the screening month (Appendix Table 1). Appendix Table 2 shows the distribution of the diagnosis and procedure code records. These algorithms comprised 4 diagnosis records (a: ill-fitting denture, b: broken denture, c: missing teeth, d: other denture-related diseases), 5 procedure records (e: management of new dentures, f: oral rehabilitation, g: denture repair, h: denture relining, i: other denture-related procedures), and 3 combinations of diagnosis and procedure records (A: denture-related diagnoses [a–d], B: all denture-related procedures [e–i], and C: either denture-related diagnoses or procedures [A or B]).

Statistical Analysis

Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) with 95% confidence intervals (CIs) were calculated for each algorithm. Sensitivity and specificity were defined as the proportions of true denture users and nonusers, respectively, correctly identified by each algorithm. The PPV was defined as the proportion of individuals identified as denture users based on claims that were confirmed as users at screening. The NPV was defined as the proportion of individuals identified as nonusers based on claims that were confirmed as nonusers during screening. For sensitivity analyses, we defined the algorithms using the participants’ 6-mo dental claims data prior to the screening month. Additionally, we evaluated the accuracy of the algorithms after restricting the study population to individuals with fewer than 28 teeth.

To estimate the potential impact of misclassification of exposure when denture use was applied as an exposure variable in epidemiological studies, we performed a post hoc bias estimation analysis for algorithms A–C (Chubak et al 2012). Percent bias, defined as the difference between the observed relative risk (RR_O) and the true relative risk (RR_T) divided by the RR_T using the sensitivity and specificity obtained from the present study, was simulated by setting the RR_T to 2.00 and 3.00 and by using the prevalence of denture use determined based on the results from this screening and the 2022 Survey of Dental Disease (Ministry of Health, Labour and Welfare 2022).

All analyses were performed using STATA software (version 17.0; Stata Corp.). This study was conducted according to the checklist of reporting criteria for studies validating health administrative data algorithms (Benchimol et al 2011).

Ethical Approval

This study was approved by the Kyushu University Institutional Review Board for Clinical Research (approval No. 226530) and the Ethics Committee of Tohoku University Graduate School of Dentistry (approval No. 40589). Approval for data use was obtained from the municipality’s Personal Information Protection Review Board.

Results

Overall, 4,053 participants were included in the present study, and 2,427 participants (59.9%) were defined as denture users based on oral health care screening. The participants’ characteristics are listed in Table 1. The mean age of the participants was 78.2 years (SD = 6.3), and 61.4% were women. Denture users had fewer teeth than nonusers (mean: 13.7 vs. 25.0).

Table 1.

Characteristics of Study Participants (n = 4,053).

Characteristic	Nondenture Users (n = 1,626)	Denture Users (n = 2,427)	Total (N = 4,053)
Age, n (%)
65–74	509 (55.4)	410 (44.6)	919 (22.7)
≥75	1,117 (35.6)	2,017 (64.4)	3,134 (77.3)
Sex, n (%)
Men	618 (39.5)	948 (60.5)	1,566 (38.6)
Women	1,008 (40.5)	1,479 (59.5)	2,487 (61.4)
Number of teeth, mean (SD)	25.0 (3.7)	13.7 (7.7)	18.2 (8.5)

SD, standard deviation.

Table 2 shows the number of participants with true positives (TPs), false negatives (FNs), false positives (FPs), true negatives (TNs), and the accuracy of the algorithms in identifying denture users, based on dental claims data from 12 mo prior to oral health care screening. In the algorithms using denture-related diagnosis codes (a–d), the sensitivity ranged from 1.4% (b: broken denture) to 20.8% (a: ill-fitting denture). The specificity and NPV ranged from 98.3% (a) to 100% (b–d) and from 40.5% (b–d) to 45.4% (a), respectively. The PPV ranged from 94.7% (a) to 100% (d: other denture-related diseases). Among the algorithms using denture-related procedure codes (e–i), algorithm f (oral rehabilitation) demonstrated the highest sensitivity (63.6%) and NPV (64.0%), but its specificity (96.7%) and PPV (96.7%) were slightly lower than those of the other procedure-based algorithms. Among the combined algorithms, algorithm C (any denture-related diagnosis code or procedure code) achieved the highest sensitivity (65.3%) and NPV (65.1%) along with moderate specificity (96.6%) and PPV (96.6%). Its overall performance was comparable to that of algorithm B, which used all denture-related procedure codes (sensitivity, 64.3%; specificity, 96.7%; PPV, 96.7%; and NPV, 64.5%).

Table 2.

Accuracy of Algorithms in Identifying Denture Users from 12-mo Claims Data before Oral Health Care Screening (N = 4,053).

Algorithm	TP, n	FN, n	FP, n	TN, n	Sensitivity, % (95% CI)	Specificity, % (95% CI)	PPV, % (95% CI)	NPV, % (95% CI)
a	505	1,922	28	1,598	20.8 (19.2–22.5)	98.3 (97.5–98.9)	94.7 (92.5–96.5)	45.4 (43.7–47.1)
b	35	2,392	1	1,625	1.4 (1.0–2.0)	100 (99.7–100)	97.2 (85.5–100)	40.5 (38.9–42.0)
c	39	2,388	1	1,625	1.6 (1.2–2.2)	100 (99.7–100)	97.5 (86.8–100)	40.5 (39.0–42.0)
d	40	2,387	0	1,626	1.7 (1.2–2.2)	100 (99.8–100)	100 (91.2–100)	40.5 (39.0–42.1)
e	8	2,419	1	1,625	0.3 (0.1–0.6)	100 (99.7–100)	88.9 (51.8–100)	40.2 (38.7–41.7)
f	1,543	884	53	1,573	63.6 (61.6–65.5)	96.7 (95.8–97.5)	96.7 (95.7–97.5)	64.0 (62.1–65.9)
g	37	2,390	0	1,626	1.5 (1.1–2.1)	100 (99.8–100)	100 (90.5–100)	40.5 (39.0–42.0)
h	10	2,417	0	1,626	0.4 (0.2–0.8)	100 (99.8–100)	100 (69.2–100)	40.2 (39.0–42.0)
i	12	2,415	0	1,626	0.5 (0.3–0.9)	100 (99.8–100)	100 (73.5–100)	40.2 (38.7–41.8)
A	602	1,825	30	1,596	24.8 (23.1–26.6)	98.2 (97.4–98.8)	95.3 (93.3–96.8)	46.7 (45.0–48.3)
B	1,561	866	54	1,572	64.3 (62.4–66.2)	96.7 (95.7–97.5)	96.7 (95.7–97.5)	64.5 (62.5–66.4)
C	1,586	841	55	1,571	65.3 (63.4–67.2)	96.6 (95.6–97.4)	96.6 (95.7–97.5)	65.1 (63.2–67.0)

Each algorithm was constructed as follows: a, ill-fitting denture; b, broken denture; c, missing teeth; d, other denture-related diseases; e, management of new dentures; f, oral rehabilitation; g, denture repair; h, denture relining; i, other denture-related procedures; A, denture-related diagnoses (a–d); B, all denture-related procedures (e–i); and C, either denture-related diagnoses or procedures (A or B).

CI, confidence interval; FN, false negative; FP, false positive; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive.

In the first sensitivity analysis using 6-mo dental claims data prior to the screening month, similar estimates of accuracy were observed, as in the main result of the 12-mo analysis (Appendix Table 3). In the second sensitivity analysis, restricted to participants with fewer than 28 teeth, the performance of the algorithms remained consistent with the main findings (Appendix Table 4).

The results of the post hoc bias estimation analysis, based on the sensitivity and specificity of algorithms A–C from the present study, are presented in Appendix Table 5. In the case where the exposure of denture usage was determined based on the results from this screening (the proportion exposed to X was set to 0.599 in scenarios 1 and 2), the percent bias in RR ranged from –36.1 to –26.4 when the RR_T was 2.00 and –52.9 to –41.5 when the RR_T was 3.00. If the exposure of denture usage was determined based on the 2022 Survey of Dental Disease (the proportion exposed to X was set to 0.748 in scenarios 3 and 4), the percent bias in RR ranged from –41.5 to –33.9 when the RR_T was 2.00 and –58.6 to –50.6 when the RR_T was 3.00. Algorithm C yielded the smallest bias across the scenarios.

Discussion

This validation study examined the accuracy of algorithms for identifying denture users from Japanese dental claims data using dentist-reported screening records as the reference standard. Among the algorithms tested, the combination of denture-related diagnosis and procedure codes (algorithm C) achieved the highest sensitivity (65.3%) and NPV (65.1%), with moderate specificity (96.6%) and PPV (96.6%). This approach demonstrated greater agreement with the reference standard than diagnosis- or procedure-based algorithms alone.

Notably, algorithm f (oral rehabilitation) also showed strong performance, with relatively high sensitivity (63.6%) and accuracy across all measures. In the Japanese dental health insurance system, the oral rehabilitation code is recorded as a procedure code when dentists provide denture adjustments, maintenance, or patient instruction aimed at optimizing denture use. Because these services are commonly provided to denture users, this code accounted for the largest number of participants in the claims data. Together with algorithms B and C, this represents a practical approach for identifying denture users from claims data, particularly in studies where sensitivity is prioritized.

Although some algorithms (e.g., broken denture or denture repair) exhibited very high specificity and PPV, they were limited by their low sensitivity because they capture only a small subset of denture users. Patients who use dentures but do not experience problems may not receive denture-related treatment or routine maintenance during the observation period, resulting in no records of denture-related diagnoses or procedures in the claims data. Consequently, a relatively large number of FN cases may occur when relying solely on claims records. By contrast, algorithms C, B, and f performed consistently across metrics, suggesting that broader code sets or specific procedure codes with high coverage are best suited for identifying denture use in claims data.

The sensitivity analysis, which used 6 mo of dental claims data before the screening month, yielded results broadly consistent with those of the main analysis, which used 12 mo of data. This suggests that denture use can be identified within a relatively short observation period. However, previous studies have reported that most denture users do not visit the dentist regularly for denture maintenance every 6 mo or annually (Algabri et al 2024; Tosun and Uysal 2025). Therefore, a longer follow-up period may be more appropriate for research contexts in which maximizing case ascertainment is critical. An optimal observation window should be selected based on the objectives of the study.

The post hoc bias analysis highlighted the implications of misclassification when denture use was applied as an exposure variable in epidemiologic research. Across scenarios, the algorithms underestimated the true RRs by 26% to 59%, with algorithm C producing the smallest bias. These findings underscore the importance of accounting for exposure misclassification when interpreting the results based on claim-derived denture data.

This study had some limitations. First, participants were restricted to a single municipality, and individuals were continuously enrolled in the NHI or the Latter-Stage Older Persons Health Care System. Exclusions owing to age-related transitions in insurance coverage (Sakamoto et al 2018) may have introduced a selection bias. Second, the algorithms were validated within the Japanese claims system, which may differ in coding practices from other countries, limiting generalizability. Therefore, external validation in other settings is required. Despite these limitations, this study provides important evidence of the feasibility of using claims data to identify denture use. Furthermore, when combined with information on the number of remaining teeth and Eichner’s classification, these measures may help identify individuals who require dentures but do not currently use them. Considering that the use of prosthodontic treatments not covered by public insurance, such as dental implants, remains extremely low among individuals with dental prostheses in Japan (Ministry of Health, Labour and Welfare 2024), the potential impact of such treatments on the validity of this approach is likely minimal. Accordingly, our findings may expand opportunities to assess oral health status, including both denture use and unmet prosthetic needs, and their associations with systemic health outcomes in large-scale epidemiologic studies.

Conclusion

This validation study demonstrated that denture use can be identified from dental claims data with moderate accuracy. The combined algorithm using both denture-related diagnosis and procedure codes (algorithm C) achieved the highest sensitivity and NPV while maintaining high specificity and PPV. Algorithm f (oral rehabilitation) also showed high accuracy, indicating its potential utility in claims-based studies. Although misclassification remains a concern, particularly for exposure assessments in epidemiological research, these algorithms provide a practical approach for incorporating denture use in the analyses of oral systemic health.

Author Contributions

A. Kinugawa: contributed to conception and design, data analysis and interpretation, drafted the manuscript; Y. Tamada: contributed to conception and design, data analysis and interpretation, critically revised the manuscript; T. Kusama, M. Hoshi-Harada, S. Ono, N. Yoda, K. Osaka: contributed to conception and design, data interpretation, critically revised the manuscript; F. Oda, M. Maeda: contributed to acquisition and interpretation, critically revised the manuscript; H. Fukuda: contributed to conception and design, data acquisition and interpretation, critically revised the manuscript; K. Takeuchi: contributed to conception and design, data interpretation, drafted and critically revised the manuscript. All authors have their final approval and agree to be accountable for all aspects of work.

Supplemental Material

sj-docx-1-jct-10.1177_23800844261444341 – Supplemental material for Validity of Denture Usage Definitions Based on Claims Data in Japanese Older Adults

Supplemental material, sj-docx-1-jct-10.1177_23800844261444341 for Validity of Denture Usage Definitions Based on Claims Data in Japanese Older Adults by A. Kinugawa, Y. Tamada, T. Kusama, M. Hoshi-Harada, S. Ono, F. Oda, M. Maeda, N. Yoda, K. Osaka, H. Fukuda and K. Takeuchi in JDR Clinical & Translational Research