Modeling longitudinal oral health status and pneumonia risk: secondary data analyses of an integrated dental-medical cohort

The source population consisted of individuals regularly receiving care at Marshfield Clinic Health System (MCHS), a large, healthcare system of over 50 medical and dental centers offering integrated care delivery to a predominantly rural region of central, northern, and western Wisconsin, and part of Michigan’s upper peninsula. MCHS offers dental care through affiliation with the Family Health Center of Marshfield (FHC-M) Dental Centers, a network of 10 strategically established, regional dental centers within its service area. An estimated 85% of patients receiving dental care through the 10 FHC-M dental clinics also seek healthcare from MCHS. Because the MCHS features an integrated medical-dental electronic health record (iEHR), holistic oversight of medical-dental healthcare delivery to the observational cohort is feasible. MCHS also operates Security Health Plan of Wisconsin, a health maintenance organization (SHP HMO) that served as an additional health data source for the current study.

The analytical cohort included patients who were: 1) age ≥ 21 years, 2) received an oral health examination at an FHC-M dental center between 2007–2019, and 3) had reasonably complete capture of their medical care within MCHS data systems. This latter criterion included medically homed patients who receive primary medical care at MCHS medical centers documented by ≥ 2 ambulatory visits over the previous three years, or had an assigned MCHS primary care provider, were SHP HMO members, and/or residents of the Marshfield Epidemiologic Study Area [9, 10]. Data from 2020 were excluded due to potentially confounding factors secondary to the COVID-19 pandemic, including temporary suspension of dental care at FHC Dental Centers and very restricted ambulatory visits at MCHS Medical Centers. For any given participant, follow-up began when all eligibility criteria were met, and ended when an outcome or censoring event occurred, as defined further below.

Identification of dental visits associated with oral health evaluations was facilitated via application of CDT codes including D0120: periodic examination; D0150 comprehensive evaluation and D0180 comprehensive periodontal evaluation. Five measures of OH status at such visits were extracted from the iEHR. Measures included:

Analyses were conducted separately for each of the 5 key OH measures. Since the OH measures were ordinal, with generally skewed or peaked distributions, the measures were partitioned for analysis into ‘best’, ‘moderate’ and ‘worst’ OH groups. Group ranges were determined by observed patient numbers, with the assumption that most health system patients would have relatively good OH. Worst OH group designation among the cohort was arbitrarily assigned to the 20% exhibiting poorest OH based on review of data from clinical OH examinations evaluating the five key OH measures.

Documented pneumonia events/episodes were initially abstracted using International Classification of Diseases Clinical Modification (ICD-CM 9/10) codes captured in the iEHR for clinic patients meeting eligibility criteria including documentation of OH status in the iEHR preceding pneumonia incidence. Patients were evaluated for subsequent pneumonia events identified and validated by searches for pneumonia-specific ICD codes in the iEHR and supporting documentation offered by Current Procedural Terminology (CPT) or other coding relevant to pneumonia validation for up to 365 days after the index dental visit. Restricting this evaluation to 365 days was an attempt to limit analyses to a window of time where, given potential changes in oral health, measures from a single dental evaluation should have relevance. Using a full year had the advantage of always including a full range of seasons. Sensitivity analyses varying the period (e.g., to 3 years) showed similar results.

With the exception of ventilator-associated pneumonia, pneumonia ICD coding is not specific to pneumonia subtypes, but mainly classified using a non-subtype-specific code or codes specific to causal pathogens. A rule-based classification algorithm derived by our study team via informatics approaches applied to data from the same cohort analyzed in the current study defined pneumonia subtype classification for: community acquired (CAP), hospital acquired (HAP), health care acquired (HCAP), ventilator-acquired (VAP) or aspiration pneumonia (ASP) [12]. Classification was undertaken only on cases validated for pneumonia diagnosis. Subtype designation required ≥ 2 diagnostic codes entered by physicians or radiologic confirmation of the diagnosis [12]. Natural language processing (NLP) for identifying terminology validating or negating presence of pneumonia documented in free text of MCHS radiology notes was further developed to leverage non-structured pneumonia-associated terminology to confirm or negate true pneumonia cases [13]. Availability of these informatics tools supported conduct of time to event modeling for: 1) ‘any pneumonia’ and 2) the CAP subtype, observed with highest frequency (55%) during the study’s temporal window.

Time-to-event analysis explored the temporal association of comorbid oral/dental disease history within a defined temporal window preceding incident pneumonia. The temporal period initiated with identification of an index dental visit documented by evaluation of OH status (T₀). Confirmation of no diagnosis in a 90-day window preceding the dental visit was required to rule out on-going pneumonia. From T₀, the observational window extended over a 365-day period and ended at the earliest identification of either: 1) a pneumonia event; or 2) a censoring condition (e.g., 365 days without pneumonia, loss to follow-up, or death unrelated to pneumonia). OH-related data from the index dental visit and any subsequent visits within the observational window were combined in a time-weighted mean.

Patient groups defined by stratification of an OH measure likely also differ with respect to other health-related variables with potential for confounding. In order to reduce the risk of confounding and provide high-dimensional balance between the OH groups prior to analysis, propensity score matching was carried out. A propensity score was developed using logistic regression to create a predictive score for ‘worst’ OH, separately for each of the 5 key OH measures. Propensity score modeling targeted variables historically examined as potential risk factors for pneumonia based on literature review to identify potential risk factors. Historical risk variables identified included: age, gender, year of visit, chronic liver disease, heart failure, cerebrovascular disease, Type II diabetes mellitus, elevated HbA1c or glucose measure; chronic kidney disease; neoplastic disease and/or chemo-/or-radiotherapy: hypertension; elevated cholesterol/lipid measures; pharmaceutical exposure for control of hypertension, hyperglycemia or hyperlipidemia; and smoking history for a subset of the cohort for whom these data were available in the iEHR as structured data.

The logistic model included all individual covariates, with potential addition of important interaction terms through forward selection (p < 0.001 to enter the model). Those assigned to the ‘worst’ group were individually matched 1:1 by propensity score to patients in the ‘best’ and ‘moderate’ groups (arbitrary allowable caliper width 20% of the estimated propensity score standard deviation). Initial comparisons between matched OH groups regarding pneumonia incidence after the index dental visit were based on product-limit (Kaplan–Meier) estimators and log-rank tests. The primary analyses of pneumonia incidence were based on proportional hazards (PH) models [14]. These PH models included adjustment for the matching variables and were fit both with and without adjustment for prior history of pneumonia as a binary (yes/no) variable. Confidence limits for hazard ratios (HR) and tests of significance were based on robust covariance estimates to account for the matching. Modeling was carried out for the full cohort with any validated pneumonia, irrespective of subtype and separately for pneumonia classified as CAP applying a rule-based pneumonia subtype classification algorithm [12].

SAS was used for all statistical analyses. Test significance was assessed at level 0.05.

The present study has notable strengths. The study sample size was large (e.g. 9,796 unique patients in each of three matched groups in Table 2B) due to availability of rich, ‘real-world’, clinical, ‘big data’ captured in the iEHR and enterprise data warehouse of MCHS for their large, well-defined population-based cohort. These data supported comparisons between the full MCHS patient population accessing medical and/or dental healthcare, including patients with and without prior pneumonia incidence over a 12-year temporal window. Notably, the FHC-M dental clinic network represents a regional a dental safety net. Thus, available data is representative of a broad patient sample spanning diverse economic backgrounds with dental access provision irrespective of third-party coverage or income status.

All incident pneumonia cases analyzed were subtyped applying rule-based classification algorithms [12] and validated through radiographic confirmation and [13], mitigating classification error. Propensity score matching, while reducing the full sample size for analysis, appeared to successfully abate confounding from other established potential pneumonia risk factors, supporting discernment of the true relative risk contributed by the OHS to pneumonia emergence. To further reduce confounding potentially associated with smoking history and pneumonia recurrence, data were adjusted when proportional hazard (PH) models revealed imbalances across our OH subgroups. Adjustment was possible only for electronically extractable smoking data in structured data fields in the iEHR and more limited for early years when data were captured in clinical notes. Higher rates of current smokers and lower rates of non-smokers were noted among individuals stratified to the poor OHS group when compared to those assigned to good or intermediate OHS groups, and were statistically significant. Smoking status adjustment relative to OH status tended towards attenuation, marginally reducing the estimated OH risk.

No other OH variables examined, including dental caries/restorations emerged as risk factors for pneumonia. Notably, dental caries have been less frequently related to systemic health outcomes, including pneumonia, in other analytical studies [18]. However, a recent systematic review by Broers et al. (2022) identified caries as a major underlying factor contributing to tooth loss and estimated that 36–53% of tooth extractions may be attributable to underlying caries while 24 to 38% of tooth loss was associated with periodontal disease [19].

Some study limitations require acknowledgement. A selection bias is possible because the cohort only included patients with OH data documented by an FHC-M dentist. However, 75% of FHC dental center patients also sought medical care at our integrated healthcare system. Since cohort eligibility required availability of both medical and dental data, selection bias risk was mitigated by eligible cohort definition. Prior pneumonia history may be unknown if patients sought care outside of MCHS prior to becoming a MCHS patient, raising potential for classification error. However, multiple episodes of pneumonia were frequently documented for individual cohort members during the observational window, thereby supporting time-to-event analysis adjusted for prior pneumonia events. Initial OH status in combination with time-weighted analysis of additional documented OH examinations within the observational window, made OH status knowable for up to a year preceding a pneumonia event.

Interpreting OHS surrounding history of tooth loss predating the temporal window may be unknowable. However, number of missing teeth and clinical PD status documenting presence of severe PD emerged consistently as risk factors for any pneumonia when not adjusted for prior pneumonia. Taken together, these data support high rates of tooth loss as a surrogate for poor OHS over the lifespan.

Although propensity score matching was used to reduce potential confounding contributed by 15 pneumonia risk factors (Fig. 1), currently unknown factors may exist that could contribute to confounding among OH groups. Another potential limitation to consider is inaccuracy of data entry in the iEHR. However, to the greatest extent possible, algorithms developed to support this study [12, 13] were applied for pneumonia validation and sub-classification, required vetting of multiple levels of evidence in the iEHR to support pneumonia validation and sub-classification, while tooth level data were electronically extractable for all dental variables, with the exception of CAL which was frequently entered in clinical notes. Any records that were not classifiable were excluded from analyses. Manual validation of algorithm performance also found high accuracy for the classification tools developed to support this study [12, 13], supporting that data sets were well vetted for accuracy in determining presence of pneumonia, pneumonia sub-classification and dental variables modeled to stratify pneumonia risk across the three OHS strata.

Recently updated guidelines for PD assessment published by Tonetti et al. in 2018, recommend use of clinical attachment loss (CAL) as a key clinical variable [20]. However, since this study represents secondary analysis of data collected during a retrospective observational window, guidelines operative during the temporal observational window of this study (AAP consensus guidelines and 2014 updates [11]) were applied. While CAL was one of several components considered for clinical staging of PD, capture of CAL in dental health records throughout most of the observational window was more commonly entered in clinical notes, precluding electronic retrieval. Thus, independent evaluation of CAL held bias potential associated with large amounts of missing data since manual abstraction from thousands of records over 12 years was cost prohibitive. However, since CAL was available to dental practitioners, CAL would have been reflected in the composite clinical periodontal staging score which considers multiple variables. Notably, higher clinical periodontal staging scores were significantly associated with increased pneumonia risk. Thus, likelihood that independent evaluation of CAL would alter modeling outcomes is low.

The OH variable most associated with pneumonia was tooth loss largely attributable to infectious processes involving advanced destructive dental caries and PD as the most common causes for tooth loss [19]. As PD advances in severity, eventual tooth loss is attributed to complex interactions of the oral microbiome and host response that drives chronic inflammation, and destruction of periodontal tissue, the root and underlying bone supporting the tooth as the tooth loses its supportive infrastructure. In the context of dental caries, bacterial metabolites erode and destroy the enamel exposing the pulp and if untreated, the tooth cannot be repaired. In the current study, clinical periodontal status showed relatively strong associations with pneumonia incidence whereas percent of teeth with restorations did not. However, our study group posits that the presence of restorations also indicates that the patient had accessed dental care to arrest the destructive processes, potentially confounding examination of the true impact of untreated dental caries on tooth loss. Therefore, the impact of dental caries on tooth loss remains equivocal in the literature.

Notably, our dental centers represent a dental safety net that serves a largely rural disparity population with high representation of Medicaid patients who meet definitions of federal poverty levels, are uninsured, elderly, include ethnic/racial minorities who have had historically low access to dental care across their lifetimes, contributing to higher rates of poor OHS in this population. In the current study, number of missing teeth lost likely consequential to underlying infectious processes emerges as a representative surrogate of poor OHS and long-term OH neglect with high clinical relevance, especially in the context of oral-systemic disease interactions, including increased risk for pneumonia for individuals with a history of poor OHS.

Significant difference in rates of emergent pneumonia were consistently noted among individuals meeting classification criteria stratifying them to the ‘worst’ oral health group. Notably, older age was also associated with the ‘worst’ OH group. Age and poor dental health, risk factors identified for CAP in the systematic review by Almirall et al.(2017) [21], were also documented risk factors in the present study.

Pneumonia risk reduction among high-risk and aging populations with low OH literacy reflected gaps in understanding of the following: 1) gaps in knowledge concerning oral cavity care; 2) patient awareness of systemic health risks associated with poor OH, and why promotion and achievement of integrated health care delivery are important; 3) assurance of patient access to dental care in compliance with the periodicity recommended by the dental provider. Gaps in knowledge regarding any of these are potentially indicative of low oral-systemic health literacy and require interventional education, and in some cases, triage to appropriate providers to get health issues addressed. Notably, a survey conducted to examine knowledgeability and practice behaviors surrounding oral health assessment among medical primary care providers documented inadequate attention to patients’ OH [22]. Providers further reported lack of training and competency in making OH assessments [22].

With mounting evidence linking poor OH as a risk factor for pneumonia emergence, proactive attentiveness among medical providers to OH status of their patients is vital. Recognizing OH as a modifiable risk factor and educating patients regarding importance of engaging good oral hygiene is important especially among vulnerable patient populations experiencing dental care access barriers who are not receiving this education or preventive care from dental professionals. Proactive establishment of integrated medical/dental delivery featuring interdisciplinary communication, effective referral systems, and access to dental safety net operations to improve dental care access for vulnerable patients with no dental access will increase the probability of achieving better OHS across their lifetime, Finally, dental and medical professionals require continuing education on evidence-based oral-systemic health associations to ensure that patients receive appropriate interdisciplinary care and education.