Discussion

We identified conditions constituting complex MM associated with the presence and severity of PD using machine learning approach in a nationally representative sample of people who were 60 years of age or older. Accounting for conditions constituting complex MM in our CTree model, we showed that sociodemographic factors and limitations in IADLs were better predictors for the presence and severity of PD than chronic conditions or geriatric syndromes. Participant’s sex, age, race, education and income levels were the most important sociodemographic variables for the severity of PD. The prevalence of PD in our study was relatively high, according to the 2017 case definition of PD. When compared with the prior criteria based on a combination of CAL values and/or probing depth values, the alone use of CAL values can partially explain the higher prevalence of PD. In addition, because all teeth were examined in the NHANES 2013–2014, the full-mouth periodontal examination approach is more likely to accurately assess and classify PD.13 The elderly, on the other hand, have a lower prevalence of PD because they are more likely to have fewer existing teeth and, as a result, better periodontal status. In addition, the 2013–2014 NHANES excludes edentate people who may have already lost their teeth due to PD, representing a biased sample of a dentate subgroup of older people.

Among individuals 60 years of age or older, hypertension, hyperlipidaemia, arthritis, functional limitations in general physical activities and urinary incontinence were the most prevalent conditions, regardless of their periodontal status. Interestingly, these common conditions represent our expanded definition of complex MM, including functional limitations and geriatric syndromes beyond chronic conditions. Although those chronic conditions and geriatric syndromes failed to show in our tree, limitations in IADLs variables in addition to sociodemographic factors emerged as strongly associated with the severity of PD. This is evidenced by the fact that they are the splitting variables in the CTree and top-ranked factors in random forest analysis, highlighting the importance of sociodemographic factors and functional limitations variables in predicting PD severity. On the other hand, among those without PD, coronary heart disease, kidney dysfunction and cognitive impairment were the most common conditions constituting MM. The use of self-reported data may underestimate the true prevalence of these conditions if these conditions have not yet been diagnosed, and/or the participant is unable to recall.

Limitations in IADLs measure the participants’ difficulties in doing such chores around the house, preparing meals and managing finances. Participants with combinations of sociodemographic factors and limitations of IADLs had a higher prevalence of no PD in the CTree than those who did not have IADLs. This may be explained by the low number of participants with IADLs (n=23) compared with those without IADLs (n=104). Although the prevalence of limitations in IADLs was 14% among the study population (n=131), 66.4% of those people had PD. Older people with IADL limitations are more likely to decline their ability to practice adequate oral hygiene and do not comply with regular periodontal maintenance visits, leading to PD. In addition, they may prioritise their general health over their oral health, given that older people are more likely to have conditions constituting complex MM. Therefore, functional limitations seem to play a more significant role in the severity of PD alongside sociodemographic factors.

Several research and clinical implications for these findings can apply in the periodontal field. Regarding research implications, these findings indicate the importance of accounting for functional limitations in addition to the sociodemographic factors when assessing PD. Older people are more likely to have functional limitations beyond chronic conditions and geriatric syndromes; therefore, they may be associated with higher severity of PD. No previous studies have identified complex MM associated with PD or its severity to the best of our knowledge. Instead, research in this area had solely assessed the relationship between PD and chronic conditions, which may explain the conflicting findings between studies that evaluated the PD using single or co-occurring chronic conditions. Indeed, the recent findings indicate the need to redirect our focus when studying PD to include functional limitations variables, rather than only chronic conditions or geriatric syndromes.

In clinical practice, identifying people with the most common combinations of conditions constituting complex MM associated with the presence and severity of PD will help healthcare providers target those by screening people with complex MM for PD and referring them for periodontal care. In addition, our findings argue for the development of integrated multidisciplinary approaches between medical and periodontal filed in the care of people with complex MM, specifically among older adults who are more likely to have conditions composing complex MM. Moreover, applying personalised periodontal medicine is crucial in managing potential etiological factors among the most vulnerable people to the development or severity of PD. Furthermore, rather than focusing on conditions constituting complex MM, clinicians should pay more attention to sociodemographic and behavioural characteristics, as they were the top-ranked variables for PD in the variable iportance plot.

To our knowledge, this is the first study to identify conditions composing complex MM that are associated with the presence and severity of PD. This study expands our prior work13 to incorporate geriatric syndromes and functional limitations beyond chronic conditions to evaluate its association with PD. Second, the use of innovative machine learning methods, CTree and random forest, is also a significant strength in this study. Both CTree and random forest using automatic variable selection to identify interactions and non-linear relationships. The ability of CTree to capture the complex relationship allows us to identify specific combinations of conditions composing complex MM that are highly associated with the presence and severity of PD without any prior assumptions by producing a tree that is easy to interpret. Random forest uses a bootstrap aggregation method to rank the importance of several variables simultaneously. The availability of comprehensive data among a nationally representative sample of the US older population that includes functional limitations, geriatric syndrome and chronic conditions is another strength of this study that further investigates the potential etiologies associated with developing PD.

This study has several limitations. First, our model failed to recognise that chronic conditions or geriatric syndromes were stronger predictors for PD. A large sample size may identify additional predictors for the presence and severity of PD. Second, CTree and random forest identified several variables for the PD; however, CTree produces only a single tree, whereas random forest is a bootstrap aggregation method that produces multiple trees without identifying the most common combinations of conditions that affect PD. Third, the cross-sectional design of our study does not allow us to assess the progression of PD as complex MM conditions arise or worsen, or even to assert the causal direction of associations. Fourth, all conditions constituting complex MM are self-reported data. Fifth, because other factors that may affect probing depth accuracy, such as gingival inflammation and the presence of local factors on the tooth surface, are not included in the public dataset, we did not include periodontitis-associated CAL that could occur in mid-buccal or mid-lingual surfaces in combination with a pocket depth of 4 mm or greater. Sixth, we were unable to determine whether participants received periodontal treatment to have clinical gingival health or gingivitis on reduced periodontium due to the nature of the NHANES dataset. Seventh, due to the small number of Asian and Hispanic/non-Hispanic ethnicities, we included them in the ‘other’ category. We had three racial/ethnic groups: white, black and other. Lastly, we note that people with PD who had dental insurance are less likely to benefit from it, because unlike dental caries, the treatment of PD is usually not covered by dental plans. In addition, the number of people with dental insurance coverage was very low in the present study. Therefore, future research focusing on the role of dental insurance plans that provide coverage for PD treatment would be beneficial.