Skip to main content

Table 3 Logistic regression models for predicting total periodontitis (moderate and severe combined) and their performance

From: A rapid, non-invasive tool for periodontitis screening in a medical care setting

Predictor Model 1:
Questionnaire, demographic data and biomarkers
Model 2:
Questionnaire and demographic data
Model 3:
Questionnaire only
Contributing to model Reference
category
B Contributing to model Reference
category
B Contributing to model Reference
category
B
Predictor
 Q1. Gum disease          
 Q2. Own teeth/gum health + Negative 1.800 + Negative 1.692 + Negative 1.686
 Q3. Gum treatment + Yes 1.367 + Yes 1.286 + Yes 1.584
 Q4. Loose teeth + Yes 1.774 + Yes 1.560 + Yes 1.969
 Q5. Lost bone
 Q6. Tooth appearance
 Q7. Floss use
 Q8. Mouthwash use + 1–7 days 1.005 + 1–7 days 1.075 + 1–7 days 0.846
 Age (years) + > 39 2.206 + > 39 2.209    
 Sex
 Smoking
 Albumin concentration
 Chitinase activity + n/a 0.394       
 Protease activity + n/a 0.094       
Model performance
 AUROCC (95% CI) 0.91 (0.86–0.96) 0.88 (0.82–0.93) 0.81 (0.74–0.88)
 Predicted probability cut-off 0.662 0.678 0.623
 Hosmer-Lemeshow 0.777 0.910 0.937
 Sensitivity (95% CI), (%) 80 (71–87) 78 (69–86) 85 (78–92)
 Specificity (95% CI), (%) 88 (76–95) 84 (71–93) 63 (49–76)
 PPV (95% CI), (%) 93 (86–97) 91 (84–95) 82 (75–89)
 NPV (95% CI), (%) 69 (59–77) 66 (57–74) 68 (55–81)
  1. The table lists all candidate predictors (data presented in Table 1, Table 2, and Fig. 1). The predictors marked with a + are the ones that remained in the prediction model after stepwise backward regression modeling (see Methods section of main text). The reference category represents the response which was coded 1 in the analysis. B is the regression coefficient of the predictor, indicating its weight.
  2. n/a not applicable, AUROCC area under receiver operator characteristic curve, PPV positive predictive value, NPV negative predictive value