Table 4.
Logistic Regression Models Used to Identify Predictors of OA Status.
| Model | Candidate Predictors | Final Predictors** Odds Ratios with 95% Confidence Interval | Final Model Goodness of fit p-value* | Area Under the Curve with 95% Confidence Interval |
|---|---|---|---|---|
| 1. Subject Characteristics Only | • Gender • BMI |
BMI • OR=2.37, 95%CI (1.30, 4.31) |
0.203 | 0.71 (0.60, 0.82) |
| 2. Metabolites Only | • 2,998 of the 18,850 peaks from the untargeted analysis associated with OA (p < 0.1) | Unknown Peak (3.74_231.6499 m/z) • OR=4.54, 95%CI (2.00, 10.29) High vs Low intensity for Unknown Peak (6.61_391.2138 m/z) † • OR=5.90, 95%CI (1.92, 18.13) |
0.253 | 0.82 (0.73, 0.90) |
| 3. Microbiome-Related Metabolites Only | • 27 identified/annotated metabolites related to the microbiome associated with OA (p < 0.1) | Hippuric Acid (PDb) • OR=0.53, 95%CI (0.32,0.90) Propionic Acid (PDc) • OR=0.44, 95%CI (0.25, 0.77) |
0.292 | 0.75 (0.64, 0.85) |
| 4. Micobiome-Related Metabolites, Microbiome genus†† | • 27 identified/annotated metabolites related to the microbiome associated with OA (p < 0.1) • 13 out of 236 microbiome taxa associated with OA (p < 0.1) |
Hippuric Acid (PDb) • OR=0.45, 95%CI (0.26, 0.80) Propionic Acid (PDc) • OR=0.43, 95%CI (0.25, 0.76) k__Bacteria;p__Firmicutes;c__Clostridia;o__Clostridiales;f__;g_ • OR=0.31, 95%CI (0.13, 0.78) |
0.664 | 0.80 (0.70, 0.91) |
| 5. Micobiome-Related Metabolites, Microbiome genus††, and BMI | • 27 identified/annotated metabolites related to the microbiome and associated with OA (p < 0.1) • 13 of the 236 microbiome genus†† associated with OA (p < 0.1) • BMI |
N-Acetyl-D-Glucosamine (OL1) • OR=0.48, 95%CI (0.27, 0.85) Hippuric Acid (PDb) • OR=0.49, 95%CI (0.25, 0.94) Propionic Acid (PDc) • OR=0.40, 95%CI (0.20, 0.78) k__Bacteria;p__Firmicutes;c__Clostridia;o__Clostridiales;f__;g__ • OR=0.28, 95%CI (0.11, 0.67) BMI • OR=4.35, 95%CI (1.73, 10.97) |
0.278 | 0.88 (0.81, 0.96) |
*
Hosmer-Lemeshow Goodness of Fit Test
**
Final predictors selected using stepwise logistic regression with entry criteria p-value < 0.1 and exit criteria p-value > 0.05
†
peak 6.61_391.2138m/z was dichotomized at the median to improve model fit.
††
If genus level information was not available, the lowest available taxon was used.
Unmatched peaks are displayed in the format retention time_mass.
AUC in bold is significantly higher than the AUC for the subject characteristics model.
BMI, body mass index; OR, odds ratio; CI, confidence interval