IADR Abstract Archives

Cross-Center Validity and Generalizability in Predicting Tooth Loss in Periodontitis

Objectives: An increasing number of teeth are being retained in a growing population of older adults, resulting in expected increases in number of periodontally-affected teeth. The knowledge of an individual’s predicted probability of losing periodontally-affected teeth would be instrumental in formulating prophylactic treatment plans. Hence, we aimed to predict tooth loss during supportive periodontal therapy (SPT) across four German university centers.
Methods: In our entire cohort (n=897), mean (standard deviation) age was 45 (11) years, with 59% females. Tooth loss in four centers (Kiel(KI) n=391; Greifswald(GW) n=282; Heidelberg(HD) n=175; Frankfurt/Main(F) n=49) during SPT was assessed. Our outcome was annual tooth loss per patient. Multivariable linear regression models were trained on data from 75% patients from one center and used for predictions on the remaining 25% patients of this center and 100% patients from the remaining centers. This procedure was performed for each center. The models’ prediction error was assessed via root-mean-squared-error (RMSE) i.e., deviation of predicted estimates from the observed data.
Results: Annual tooth loss per patient differed across centers (median=0.00 (interquartile interval: 0.00, 0.17) in GW and 0.09 (0.00, 0.19) in F, p=0.001) Age, smoking status, and number of teeth present before SPT were associated with tooth loss (p<0.03) (Table 1). Predictions within centers showed RMSE from 0.14 to 0.30, and cross-center RMSE ranged from 0.15 to 0.31, indicating low generalizability (Table 2). Predictions had higher accuracy in F and KI than in HD and GW, while the center on which the model was trained had a less consistent impact on prediction estimates. None of the models evaluated showed useful predictive value.
Conclusions: Associations should be distinguished from predictions. Despite significant associations of covariates with annual tooth loss, a clinically useful prediction was not possible, highlighting the need for further research to identify predictors of periodontally-affected tooth loss in adults receiving periodontal treatment.

Division:
Meeting: 2021 Continental European and Scandinavian Divisions Meeting (Brussels, Belgium, Hybrid)
Location: Brussels, Belgium, Hybrid
Year: 2021
Final Presentation ID: 0043
Abstract Category|Abstract Category(s): Periodontal Research-Diagnosis/Epidemiology

Authors

Arsiwala, Lubaina ( Charité – Universitätsmedizin Berlin , Berlin , Berlin , Germany )

Graetz, Christian ( University of Kiel , Kiel , Germany )

Schwendicke, Falk ( Charité – Universitätsmedizin Berlin , Berlin , Berlin , Germany )

Krois, Joachim ( Charité – Universitätsmedizin Berlin , Berlin , Berlin , Germany )

Bäumer-könig, Amelie ( University Hospital Heidelberg , Heidelberg , Germany )

Pretzl, Bernadette ( University Hospital Heidelberg , Heidelberg , Germany )

Eickholz, Peter ( Johann Wolfgang Goethe-University Frankfurt/Main , Frankfurt/Main , Germany )

Petsos, Hari ( Johann Wolfgang Goethe-University Frankfurt/Main , Frankfurt/Main , Germany )

Kocher, Thomas ( University Medicine Greifswald , Greifswald , Germany )

Holtfreter, Birte ( University Medicine Greifswald , Greifswald , Germany )

Financial Interest Disclosure: NONE

SESSION INFORMATION

Oral Session IN PERSON
Periodontology
Friday, 09/17/2021 , 01:30PM - 03:30PM

TABLES

Table 1. Linear regression estimates (95% confidence interval) for the association of covariates with annual tooth loss per patient, in the whole cohort (n=897)

Covariate	Estimate	95% LCI	95% UCI	p-value
Male (ref.: female)	-0.02	-0.05	0.01	0.28
Age (years)	<0.01	<0.001	<0.01	<0.01
Former smoker (ref.: non-smoker)	-0.01	-0.04	0.03	0.64
Current smoker (ref.: non-smoker)	0.08	0.03	0.12	<0.001
Diabetes (ref.: no)	0.07	-0.01	0.15	0.07
Aggressive periodontitis (ref.: CP)	0.02	-0.03	0.07	0.39
Number of teeth before SPT	0.00	-0.01	0.00	0.02
Number of teeth before SPT	0.00	-0.01	0.00	0.02

Abbreviations: LCI= lower confidence interval, UCI=upper confidence interval, ref=reference group, CP=chronic periodontitis, SPT= supportive periodontal therapy Significant associations (p<0.05) are written in bold.
Table 2. Linear regression prediction models’ results

	Models built on data from
Estimates (95% CI)	Kiel	Greifswald	Heidelberg	Frankfurt/Main
Male (ref.: female)	-0.02 (-0.06-0.02)	0.03 (-0.05-0.10)	-0.11 (-0.23-0.01)	-0.01 (-0.11-0.09)
Age (years)	0.00 (0.00-0.01)	0.00 (0.00-0.00)	0.00 (0.00-0.01)	0.00 (-0.01-0.00)
Ever smoker (ref.: never)	0.04 (0.01-0.06)	-0.03 (-0.08-0.02)	0.09 (0.03-0.16)	0.04 (-0.04-0.12)
Diabetes (ref.: no)	-0.03 (-0.17-0.11)	0.18 (0.03-0.32)	-0.01 (-0.38-0.35)	0.11 (-0.03-0.26)
Aggressive periodontitis (ref.: CP)	0.03 (-0.03-0.09)	Not applicable	0.02 (-0.14-0.18)	-0.09 (-0.28-0.10)
Number of teeth before SPT	0.00 (-0.01-0.00)	0.00 (-0.01-0.01)	0.00 (-0.02-0.01)	0.00 (-0.02-0.02)

Prediction error (RMSE (95% CI)) for models trained in…	Kiel 0.17 (0.16-0.18)	Greifswald 0.25 (0.23-0.27)	Heidelberg 0.25 (0.22-0.29)	Frankfurt/Main 0.14 (0.13-0.16)	Mean RMSE for models tested on…..
Kiel	0.14	0.17	0.17	0.17	0.16
Greifswald	0.27	0.30	0.29	0.28	0.29
Heidelberg	0.29	0.31	0.24	0.30	0.29
Frankfurt/Main	0.16	0.15	0.15	0.16	0.16
Mean RMSE for models trained in….	0.22	0.23	0.21	0.23

Models were developed on 75% of the data available from one center and tested on 25% of the same center and 100% of data from the other centers. Abbreviations: ref=reference level, CP=chronic periodontitis, SPT= supportive periodontal therapy, RMSE=Root Mean Squared Error; lower RMSE values represent better model performance Mean RMSE on data from each center (right column) as well as for each model across centers (bottom line) are shown