Cost-Effectiveness of Oral Antidiabetic Drugs: A Prospective Multicenter Study of Real-World Patients

Table 12

Variable selected for post-LASSO OLS.

	HbA1c < 6.5%	Hypoglycemia	EQ-5D HRQoL	Inpatient cost	Outpatient cost	OTC drug cost	Total cost
60>age ≥ 50					Y
70>age ≥ 60					Y
Age ≥ 70			Y		Y		Y
Male
Lower secondary education
Upper secondary education					Y	Y
Tertiary education
Shenyang	Y	Y	Y	Y	Y	Y	Y
Chengdu	Y	Y	Y	Y	Y	Y	Y
Nanjing	Y	Y	Y	Y	Y	Y	Y
Guangzhou	Y	Y	Y	Y	Y	Y	Y
2000 > income ≥ 1000					Y
Income ≥ 2000			Y		Y
UEBMI
NRCM	Y	Y	Y	Y	Y	Y	Y
Govern. insur.
Currently smoking
Currently drinking			Y
Any physical exercise			Y
Diet control
10 > duration ≥ 5
Duration ≥ 10	Y	Y	Y	Y	Y	Y	Y
Heart disease					Y	Y
Hypertension	Y	Y	Y	Y	Y	Y	Y
Dyslipidemia
Stroke			Y
Baseline BMI
Baseline HbA1c	Y	Y	Y	Y	Y	Y	Y
Baseline FBS	Y
Baseline hypoglycemia
Baseline SBP	Y
Baseline DBP			Y
Baseline TC	Y	Y	Y	Y	Y	Y	Y
Baseline TG
Baseline HRQoL		Y	Y
(60 > age ≥ 50) × (B/L HbA1c ctrl)
(70 > age ≥ 60) × (B/L HbA1c ctrl)	Y
(Age ≥ 70) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(Male) × (B/L HbA1c ctrl)
(Lower secondary education) × (B/L HbA1c ctrl)
(Upper secondary education) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(Tertiary education) × (B/L HbA1c ctrl)
(Shenyang) × (B/L HbA1c ctrl)	Y
(Chengdu) × (B/L HbA1c ctrl)					Y
(Nanjing) × (B/L HbA1c ctrl)
(Guangzhou) × (B/L HbA1c ctrl)
(2000 > income ≥ 1000) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(Income ≥ 2000) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(UEBMI) × (B/L HbA1c ctrl)
(NRCM) × (B/L HbA1c ctrl)
(Govern. insur.) × (B/L HbA1c ctrl)
(Currently smoking) × (B/L HbA1c ctrl)
(Currently drinking) × (B/L HbA1c ctrl)
(Any physical exercise) × (B/L HbA1c ctrl)
(Diet control) × (B/L HbA1c ctrl)
(10 > duration ≥ 5) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(Duration ≥ 10) × (B/L HbA1c ctrl)	Y
(Heart disease) × (B/L HbA1c ctrl)
(Hypertension) × (B/L HbA1c ctrl)
(Dyslipidemia) × (B/L HbA1c ctrl)					Y
(Stroke) × (B/L HbA1c ctrl)
(Baseline BMI) × (B/L HbA1c ctrl)
(Baseline FBS) × (B/L HbA1c ctrl)	Y	Y	Y	Y	Y	Y	Y
(Baseline hypoglycemia) × (B/L HbA1c ctrl)		Y
(Baseline SBP) × (B/L HbA1c ctrl)
(Baseline DBP) × (B/L HbA1c ctrl)
(Baseline TC) × (B/L HbA1c ctrl)
(Baseline TG) × (B/L HbA1c ctrl)
(Baseline HRQoL) × (B/L HbA1c ctrl)					Y

Note: B/L HbA1c ctrl is the abbreviation for baseline HbA1c control rate. The post-lasso linear regression is a two-step procedure where firstly confounding variables are selected using lasso and then they are used as controls in multivariate linear regression. To minimize the risk of overpenalization, variables selected for post-lasso regression are taken as the union of variables selected using both outcome and treatment arms as dependent variables of lasso. The choice of the tuning parameter of lasso is based on empirical BIC.