Source system

The registry’s data source is the proprietary Electronic Dental Record (EDR) used at NDCS—developed by Integrated Health Information Systems (IHiS, now known as Synapxe). The extracted registry’s domain-specific EDR data captures structured clinical information, including demographics, diagnoses, procedures, tooth condition, and periodontal charting (probing pocket depth [PPD], recession [REC] and bleeding on probing [BOP]). It does not include radiographic, administrative or pharmacy data, and free text notes are not extracted for the registry. Fig 1 illustrates the data architecture and workflow from EDR to ODR.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Data architecture.

Abbreviations: DB, database; EDR, Electronic Dental Record; ODR, Oral Disease Registry.

https://doi.org/10.1371/journal.pone.0341766.g001

Administration and governance

The ODR currently includes all patients who had a dental visit at NDCS between 01/01/2013 and 30/06/2025. The registry was established under CIRB exemption 2022/2167 with a waiver of informed consent because only routinely collected, anonymized clinical data were used. Governance follows Singapore’s Personal Data Protection Act, the Human Biomedical Research Act (HBRA), and the Declaration of Helsinki. Each patient is assigned a unique identifier (NRIC for citizens and permanent residents), which serves as a linkage key across visits and enables integration with other healthcare databases within SingHealth. NRICs were encrypted before storage; a linkage file permitting re-identification requires approval from the SingHealth Office of Deputy Group Chief Medical Informatics Officer (Research); and all data are stored in secure, access-controlled environments. Analysts access de-identified data through virtual desktops within the SingHealth network using corporate credentials and devices, ensuring that no data are stored or processed outside the secure infrastructure.

A formal governance process has been established for the access, storage, and use of the data. All approved data are stored within the SingHealth Research Control Tower, a secure, cloud-based institutional data platform. Access is restricted to the institutional network via a secure platform, and the system is commercially hosted in compliance with institutional data governance and security policies. Access is limited to authorized users through role-based permissions, and storage of data on personal devices is not permitted unless explicitly approved and secured in accordance with institutional guidelines. Researchers seeking to use ODR data follow a structured and transparent process. They begin by submitting a proposal outlining their intended analysis, which is shared with ODR investigators for feedback and potential collaboration. Once endorsed, the researcher submits a formal data request to the ODR Executive Committee for approval. External collaborators may also participate by contacting the ODR Principal Investigator, who reviews and circulates the proposal among ODR investigators for consideration.

System development

Initial discussions were held with the Principal Investigator, data manager, NDCS Chief Medical Informatics Officer, and Synapxe engineers to define the project scope and data requirements. EDR Subject area variables were identified and mapped to EDR’s metadata repository or data dictionary to understand the relationships between its elements to other data, usage and formats. Screens from the NDCS EDR were systematically reviewed to identify and confirm the exact front-end fields required for extraction at record-level from a larger dataset. These were subsequently traced to their corresponding back-end tables and compiled into a User Requirements Document outlining clinical variables and inclusion rules, and a Functional and Technical Specification detailing data elements, linkage keys (tooth, site, and patient level), and data flow architecture.

The team designed the registry schema and implemented a two-step extract–transform–load process:

1. (i). Cohort identification, which defined eligibility criteria (i.e., patient with an initial assessment visit note), and
2. (ii). Data import and mapping, which integrated the relevant clinical elements.

A data harmonization step reconciled discrepancies between front-end variable names and back-end storage formats. ETL scripts were implemented and validated through internal integration testing to ensure data lineage and field-level accuracy. User acceptance testing involved three independent testers—two clinicians and a data manager—who evaluated both functionality and data fidelity using a synthetic test cohort of 45 mock patients, each with 10 scheduled visits. The UAT primarily focuses on validating data constraints and rules established across the EDR datasets, checking on the consistency and completeness of data to ensure data integrity is maintained. Test visits were distributed across fixed weekdays to emulate real operational patterns such as clinic scheduling and end-of-day data loads. Testing comprised three key domains: a) functional verification: ensuring screens and queries retrieved the expected fields, and that diagnostic and odontogram logic operated as designed (the odontogram refers to the graphical charting interface in the EDR that records each tooth’s surface condition and periodontal charting; b) data verification: confirming that record counts and patient-timepoints in the ODR matched EDR extracts under identical inclusion criteria, with field-level checks on presence, surface condition, PPD, REC, BOP, and c) soak testing: performed 10 weeks after UAT initiation to confirm that newly generated visits on the scheduled days were correctly captured by the ETL runs, simulating ongoing production operations.

Data collected

For each eligible visit, the ODR captured:

• Demographics: age, date of birth, sex, ethnicity, residency status, Singapore Housing Index, a proxy measure of socioeconomic position based on the type of public housing (e.g., 1–2 room, 3 room, 4–5 room, or private housing) [27].
• Smoking history
• Diagnosis codes
• Examination chart variables: complaints, dental history, extra-oral and intra-oral findings
• Odontogram data: presence or absence of each tooth, tooth and surface condition (including caries, restorations, crowns, root-filled teeth, and prosthetic status), and periodontal charting (PPD, REC and BOP for six sites per tooth)

Data quality assessment

Characteristics of the ODR cohort

From 2013 to mid-2025, the ODR captured 229,249 unique patients (Table 1). The sex distribution was roughly balanced throughout, and the cohort’s composition was predominantly Chinese (~78%) and Singapore citizens (~93%). The cohort aged over time: mean age rose steadily from the late-40s in the to the mid-50s by 2023–2025. Smoking status was dominated by never-smokers across all years, with a modest decline in current smoking over time. Most clinical variables showed right-skewed distributions: median values were zero and means were low, indicating that while most patients had no recorded events for these measures, a smaller proportion exhibited higher counts (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of patients in the NDCS Oral Disease Registry, 2013–2025.

https://doi.org/10.1371/journal.pone.0341766.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Summary statistics of clinical variables recorded.

https://doi.org/10.1371/journal.pone.0341766.t002

Periodontal diagnoses and service activity

The overall distribution of periodontal diagnoses is presented in Table 3, with the full list in S1 Table in S1 File. Among cases classified with the 2018 EFP/AAP system, Stage III disease predominated, with both localized and generalized presentations represented across grades. For diagnoses recorded under the 1999 CPDC Classification, localized and generalized severe chronic periodontitis were the most common categories. Periodontal service activity by year and visit type is summarized in S2 Table in S1 File. Visit counts increased over time, with relatively lower counts observed in the earlier years of the registry period, followed by a marked increase from 2019 onwards. A transient drop in visits was observed in 2020, consistent with reduced clinical activity during the COVID-19 pandemic, followed by recovery in subsequent years.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Distribution of periodontal diagnoses in the ODR.

https://doi.org/10.1371/journal.pone.0341766.t003

Diagnosis agreement

Agreement between examiner-derived and recorded periodontal diagnoses varied by period (Table 4). The highest concordance was observed during the final years when the 1999 CPDC classification was still in use and in the most recent period, with lower agreement noted in the earliest ODR years and during the initial implementation of the 2018 EFP/AAP classification. After combining all diagnoses, unweighted Cohen’s kappa indicated substantial agreement: κ = 0.84 (95% CI 0.78–0.90).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Agreement between examiner-derived and recorded periodontal diagnoses across three time periods.

https://doi.org/10.1371/journal.pone.0341766.t004

Data completeness

Data completeness was low, with REC showing the highest proportion of missing data (Table 5). Completeness improved at follow-up compared with initial visits, likely because second visits typically corresponded to periodontal treatment appointments, who recorded more comprehensive charting. Nevertheless, PPD and BOP were only slightly more than half complete. At the site level, REC similarly exhibited the greatest degree of incompleteness across parameters (S3 Table in S1 File).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Completeness of periodontal charting.

https://doi.org/10.1371/journal.pone.0341766.t005

Lessons learnt

Electronic health records often operate in fragmented and inefficient systems, with limited interoperability and substantial data processing required to extract usable information. The lack of consistent structure and standardization further compounds these inefficiencies that hinder integration and secondary analysis [45]. Our experience revealed several operational challenges. First, initial assessment visit and treatment notes must be standardized and completed consistently. The relatively low number of recorded periodontal visits prior to 2019 likely reflects inconsistent use of structured periodontal note templates rather than true service volume. Second, data extraction depended on structured fields; free-text notes and radiographs were excluded, limiting capture of clinical data. Radiographs and prescriptions are stored in separate systems, further complicating data linkage.

Third, electronic health records are primarily designed for patient care rather than research, and substantial effort was required to harmonize variable names, map relationships from tooth to site to patient, and resolve inconsistencies. Finally, registry development requires close involvement of diverse stakeholders. This includes researchers, data managers, engineers, and clinicians familiar with the electronic health records system. Strong administrative leadership is also critical to navigate governance processes and ensure timely approvals. Clear governance structures and sufficient resource support are also necessary to maintain the registry’s long-term viability. The results from the registry can also be utilized for clinical audits, for example, by identifying gaps in documentation, and standardizing charting practices.

Limitations and strengths

The ODR is not without limitations. The ODR represents a single national specialty center and may not represent the broader community disease burden; the cohort is skewed toward citizens, and individuals seeking specialist care. Data completeness is limited. Whether missing charting data are outcome-dependent – for example, if clinicians are more likely to record sites with deeper pockets while omitting apparently healthy sites – cannot be determined from the registry, as the unrecorded values are by definition unknown. If such selective recording occurs, observed disease estimates may overestimate the true proportion across the full mouth and inflate the apparent prevalence of severe disease at the individual level. A particular challenge in dentistry is the inherently hierarchical structure of the oral cavity, where each patient has multiple teeth, each with several charted sites. This making comprehensive charting time-consuming in routine workflows. Moreover, the 1999 CPDC and 2018 EFP/AAP classifications require complete site-level data, a standard difficult to achieve in everyday clinical practice. Standardizing charting procedures and clinician training at NDCS will therefore be essential to improve data completeness and diagnostic reliability. A substantial proportion of visits carried unspecified periodontal codes, which is expected as many of these encounters were likely managed by clinicians from other specialties as part of routine triage, rather than by periodontists assigning definitive diagnoses. Smoking history was self‑reported and may also have been under‑recorded. Clinician-level identifiers are not captured in the ODR, precluding assessment of whether diagnostic coding differs across clinician groups. As a national specialty center operating within an academic medical center, cases managed within specialty units at NDCS, including the periodontal and restorative dentistry units, and are subject to consultant oversight, case presentations, and calibration discussions with general dental practitioners and residents, providing a degree of diagnostic standardization for both periodontal and caries diagnoses. However, encounters recorded during triage by clinicians from other dental specialties, which likely account for a substantial proportion of unspecified periodontal codes, do not undergo the same level of diagnostic review. Inter-clinician variability in recording practices may introduce information bias, particularly if diagnostic thresholds differ between specialists and generalists, for example, in the recording of early carious lesions or the assignment of periodontal case definitions.

Nevertheless, this study has strengths. It covers more than a decade of care for a large number of unique patients. The use of a unique national identifier for each citizen or permanent resident (NRIC) enables linkage to other SingHealth and national registries, providing opportunities to explore oral–systemic relationships. Standardization of data elements within the ODR enhances data quality and facilitates interoperability across healthcare domains. The calibrated diagnosis demonstrated substantial agreement, suggesting that clinical charting can be used reliably for epidemiological research.

Another strength of this study is the use of the SHI as a proxy for socioeconomic position [27]. In Singapore, home ownership and housing type serve as well-established socioeconomic indicators [46]. Individual-level data such as income or education are not routinely captured in electronic health records and may be subject to reporting and social desirability biases when self-reported. Area-level indices, such as neighborhood-based socioeconomic position measures, have been developed locally, but remain prone to ecological fallacy [47]. The SHI overcomes these limitations by providing building-level granularity, which retains privacy while approximating individual socio-economic status [48]. It is derived from open data Application Programming Interfaces (APIs) and represents a weighted average number of rooms per residential unit, scaled from 1 to 6. While it does not distinguish between private housing types (e.g., condominiums or landed properties), it preserves the ordinal housing-value gradient across the population (S4 Table in S1 File). The SHI has been validated and shown to correlate with population-level indicators such as bystander cardiopulmonary resuscitation rates [49] and vaccination uptake [50].

Future directions

To our knowledge, the ODR is the first large‑scale oral disease registry in Asia established within a national specialty center. Through this, we hope to contribute to the growing literature on real‑world oral health research and encourage collaborations that span registries and health systems. Ongoing work will extend the ODR to include data from other dental specialties (e.g., pediatric dentistry). Incorporation of clinical images and prescription data, as well as natural language processing of free‑text notes will enhance data richness and research utility. We hope for the registry to be refreshed at six-month intervals, subject to operational feasibility and resource availability.

The ODR's longitudinal structure and linkage potential also position it as a valuable resource for developing and externally validating prognostic models for oral disease outcomes such as tooth loss, an area where existing models have been limited by small sample sizes, inappropriate handling of missing data, and lack of external validation [51]. The utilization of the ODR represents a large-scale effort to integrate electronic oral health records within a major academic medical center in Singapore. This registry provides a valuable platform for research and healthcare service evaluation. Future expansion of its data scope and linkages will further enhance its utility, supporting quality assessment, care improvement, and targeted interventions aimed at reducing the burden of oral disease nationally.