Table 2.

Performance characteristics of single lipid species combined with CA125, age, and BMI, in classifying early-stage (I-II) cases versus controls.

	Class	Metabolite	spec90	Δspec90	PΔs	AUC	ΔAUC	PΔA
1	DAG	DAG(16:1/18:1)	0.79	0.19	0.04	0.92	0.06	0.02
2	CE	CE(18:1)	0.67	0.07	0.08	0.87	0.00	0.38
3	CE	CE(17:0)	0.69	0.09	0.11	0.88	0.01	0.31
4	DAG	DAG(20:0/20:0)	0.71	0.10	0.13	0.88	0.01	0.26
5	TAG	TAG52:2-FA20:2	0.81	0.21	0.17	0.91	0.05	0.06
6	TAG	TAG54:3-FA20:3	0.79	0.19	0.20	0.91	0.04	0.08
7	PE	PE(P-16:0/20:3)	0.74	0.14	0.20	0.88	0.01	0.32
8	DAG	DAG(18:0/18:2)	0.76	0.16	0.22	0.90	0.03	0.10
9	TAG	TAG56:4-FA18:0	0.81	0.21	0.23	0.90	0.04	0.12
10	CE	CE(20:0)	0.67	0.07	0.23	0.87	0.01	0.38

Using data from the RP cohort, logistic regression was used to model the log odds of case status as a linear function of log₂ CA125, age, and BMI, with or without a given lipid metabolite (M), after imputing missing values for each species to ½ the minimum concentration detected for that species across all participants. Receiver operating characteristic (ROC) curve analysis was conducted to estimate area under the curve (AUC) and specificity at 90% sensitivity (spec90) conferred by the joint model (with M) versus the base model (without M: AUC = 0.865, spec90 = 0.603). Lipids with Δspec90 > 0.05 were selected and ranked by corresponding P value (P_Δs). Performance metrics listed for top 10 lipid species.