A patient-reported questionnaire developed in a German early arthritis cohort to assess periodontitis in patients with rheumatoid arthritis

Patients from the early arthritis cohort study Course And Prognosis of Early Arthritis (CAPEA) were asked to participate in this project. CAPEA is a prospective, multicentre, non-interventional, observational study in which patients were enrolled between 2010 and 2013 [14]. Eligible patients had arthritis for less than 6 months. They were consecutively enrolled in rheumatology clinics and practices in Germany and observed for 2 years in order to investigate the prognostic value of early symptoms for the development of a chronic course of disease. Ethical approval for CAPEA was obtained from the Ethics Committee of the Charité University Medicine, Berlin, in May 2009 with an amendment for the periodontitis project in May 2012.

All patients enrolled in CAPEA until January 2013 were sent a questionnaire including 12 questions about their PD status and other items considering dental replacement, comorbidities, current medication and pain. The questionnaire items were as follows: “number of teeth” (0–28), “receding jaw bone” (0, no; 1, yes), “receding gums” (0, no; 1, at up to 3 teeth; 2, at 4 to 10 teeth; 3, at over 10 teeth), “presence of gum pockets” (0, none; 1, at up to 3 teeth; 2, at 4 to 10 teeth; 3, at over 10 teeth), “loose teeth” (0, no, never; 1, I had loose teeth in the past; 2, yes, I currently have loose teeth), “tooth extractions because of inflammation and deep gum pockets” (0, no; 1, at up to 3 teeth; 2, at 4 to 10 teeth; 3, at more than 10 teeth), “more dentist visits because of inflammation than because of caries” (0, no; 1, yes), “more tooth/gum problems than other persons of the same age and sex” (0, less than others or comparable to others; 1, more than others; 2, a lot more than others), “inflammation of the gums/bleeding” (0, never; 1, every few years; 2, in many years; 3, (nearly) every year), “magnitude of suffering from dental problems in total during the last 6 months” (0, not at all; 1, a little bit; 2, quite a bit; 3, severe problems), “cold- or heat sensitivity” (0, no, never; 1, yes, in the past; 2, yes, currently) and “use of antibiotics to treat inflammation in the jaw bone” (0, never; 1, once; 2, two to five times; 3, more than five times). Most of the questions were illustrated with pictures to demonstrate the appearance of a radiograph with receding jawbone for example. The questionnaire is available from the authors upon request.

Patients were asked for the permission to contact their dentists. For all patients who returned a written consent, their dentists were then contacted by mail. They were asked to report whether or not the patient had been diagnosed with PD and to assess the PD status semiquantitatively with the possible answers “no”, “mild (< 30% bone loss)”, “moderate (30–50% bone loss)” or “severe PD (> 50% bone loss)”. Additionally, the number of teeth was reported. Furthermore, the dentists were asked to send any radiographs not older than 5 years for evaluation, if available.

The obtained radiographs were scored independently by three dentists at the School of Dentistry at the University of Birmingham, UK. The dentists were blinded to the clinical data of the patients. Disagreements were resolved by discussion. The confidence in the diagnosis of PD based on the available radiographs was rated as “certain”, “pretty certain” or “uncertain”.

The PD status reported by the patients’ dentists was defined as the reference standard for PD for all analyses.

Correlations between the patient-reported items, the dentists’ assessment and the blinded external assessment of the radiographs were analysed using Spearman’s correlation coefficient. Confirmatory factor analysis was used to test the one-dimensional factor structure of the questionnaire. Items with similar content may result in correlated measurement errors [15] as indicated by large modification indices. Therefore, correlated residuals were assumed in the confirmatory factor model to avoid this method error. The evaluation of the model fit was based on the cut-offs as recommended by Hu and Bentler [16] (root mean square error of approximation (RMSEA) ≤ 0.06, comparative fit index/Tucker-Lewis index (CFI/TLI) ≥ 0.9).

These items were used to calculate a final score for the detection of PD. Since age strongly correlates with the number of teeth and the probability to have PD, we always included age in the score [1].

The diagnostic properties of the score were evaluated by determining the sensitivity, specificity and the area under the receiver operator characteristic curve (AUC). Possible values for the AUC range from 0.5 to 1: 0.5 meaning a random classification of patients as having PD or not and 1 meaning perfect discrimination of the score between the groups. As the PD status was not assessed binary but with several levels of severity, different classifications of patients were performed. This resulted in three binary classifications of PD status: no versus mild/moderate/severe PD, no/mild versus moderate/severe PD and no/mild/moderate versus severe PD. To include all classifications of PD into a single model, an ordinal regression was performed so that it was possible to use the resulting score to assign patients to the most likely level of PD without having to choose which severity of PD should be detected.

The AUCs resulting from applying the model based on the whole dataset on the same data are likely too optimistic. We corrected for this overoptimism with bootstrap methodology. For 500 bootstrap samples of the size of the original dataset, models for the PD score were estimated. The resulting models were applied to both the original dataset and the respective bootstrap samples. Differences in the resulting AUCs were calculated, resulting in an estimator for the mean overoptimism. This estimator was subtracted from the original AUCs, resulting in bias-corrected AUCs. Additionally, the models based on the dentist’s assessment of PD were applied to the subsample of patients with a radiographic assessment of PD, using this as the reference standard.

A total of 512 patients completed the patient questionnaire and gave permission to contact their dentists. We received 353 data sets with the dentist’s assessments of the PD status and 253 data sets with additional radiographs. Radiographs of 4 patients were excluded due to insufficient quality. Figure 1 shows a flowchart of the respective patient numbers. The clinical characteristics at baseline of the different patient groups are shown in Table 1. The subgroups were comparable to the CAPEA cohort except for a slightly higher mean number of teeth in the patients with available radiographs.

According to their dentists, 30% of the patients had no, 33% mild, 26% moderate and 11% severe PD. Of the 253 patients with radiographic evaluations, 23% had no, 25% mild, 29% moderate and 23% severe PD. For 41% of the patients, the three independent dentists rated the security of their PD assessment as certain; in 49% of cases, they were moderately certain; and in 11% of cases, they were uncertain, meaning that at least two dentists rated the PD status of the respective patients as uncertain. Certainty was higher for the assessment of no or severe PD than for the assessment of mild or moderate PD.

The correlations between the patient-reported items and the dentists’ assessments of PD are shown in Table 2. They were highest for the patient-reported number of teeth, receding jaw bone, receding gums, presence of gum pockets and loose teeth. Those items also correlated highest with the independent assessment of PD via radiographs. The strength of the correlation is only moderate with the highest correlation coefficient of − 0.49 between the number of teeth and the dentist’s assessment of PD.

The factor structure of two models was tested by confirmatory factor analysis (CFA): (a) a one-factor model in which all items were included and (b) a one-factor model in which six items were included (number of teeth, receding jaw bone, receding gums, presence of gum pockets, loose teeth, tooth extractions because of inflammation and deep gum pockets), which correlated among each other in preliminary analyses. The CFA including all questionnaire items did not result in an acceptable model fit (RMSEA = 0.096, CFI = 0.86, TLI = 0.88, WRMR = 0.96). The model that included six selected items did not fit the data well (RMSEA = 0.148, CFI = 0.89, TLI = 0.82, WRMR = 1.05). The modification indices suggested correlated residuals between the items “number of teeth” and “loose teeth” (modification indices = 26.5). The resulting model with correlated residuals yielded an acceptable model fit (RMSEA = 0.06, CFI = 0.94, TLI = 0.90, WRMR = 0.77).

Six variables (number of teeth, receding jaw bone, receding gums, presence of gum pockets, loose teeth, tooth extractions because of inflammation and deep gum pockets) were identified as having a prognostic value for PD. These variables and three additional demographic variables (age, sex and formal education) were used to calculate several binary scores for the assessment of PD. As there are three possible levels of disease severity, several results have to be considered for every possibility to classify patients with the score.

The item which correlated best with the different assessments of PD was the number of the remaining teeth. Therefore, the first proposed possibility to classify patients was to use only age and the self-reported number of the remaining teeth. This resulted in a sensitivity of 86/80/86% (no versus mild, moderate or severe PD/no or mild versus moderate or severe PD/no, mild or moderate versus severe PD), a specificity of 49/64/78% and an AUC of 0.73/0.78/0.86 (Table 3). For all models, the bias-corrected AUCs are only marginally lower. The AUCs of the models based on the dentist’s assessment applied to the radiograph scoring data differ most from the original model for the two models in which severe PD is detected (0.66 versus 0.85 and 0.77 versus 0.90).

When using all items from the patient questionnaire that were identified as being useful for the classification of PD, all models improved the diagnostic properties compared to the simple model only using age and the number of teeth. The AUCs of these models range between 0.82 and 0.92 depending on the severity level of PD that shall be detected. The models including sex and formal education of the patients did not show more favourable properties than those only including age as a demographic variable (data not shown). Therefore, sex and formal education were not included in the score.

For the ordinal regression model, there was again a simple version with only the number of teeth and age, and one model including the five additional patient-reported items mentioned above. A likelihood ratio test showed that the model with the additional items is better than the simple version.

The following were the results for the score:

PD score = 2.8 + 0.033 × age + 0.37 × gum pockets + 0.30 × receding gums + 0.45 × loose teeth + 0.84 × receding jaw bone − 0.40 × tooth extractions − 0.12 × number of teeth.

The cut-offs were 1.83 for mild PD, 3.91 for moderate PD and 6.26 for severe PD. For example, a patient with an age of 40 years, no reported tooth or gum problems and all 28 teeth would have a score of 2.8 + 1.32–3.36 = 0.76 and would be classified as having no PD.

The following were the results for the simple version of the score:

PD score (simple version) = 2.5 + 0.036 × age − 0.11 × number of teeth. The corresponding cut-off values were 1.16 for mild PD, 2.88 for moderate PD and 4.91 for severe PD.

Table 4 shows the classification of the patients by the score compared to the reference standard. Patients with severe PD are only detected in less than 30% of the cases and most often classified as “moderate” (Table 5).

The longer version of the score had considerably more specificity for the detection of PD than the short version (57% versus 40%). It also had a higher sensitivity for detecting moderate or severe PD.