Table 3 Details of scenarios 1 and 2

1

Whether the risk of diabetes among SGA users is higher than that among FGA users?

Glucose metabolic disorders are a known risk of SGA [28, 29]. Therefore, the diabetes risk of SGA is compared to that of FGA.

Inclusion criteria:

• Patients initiated with any SGA^a or FGA^b during the study period (January 01, 2015, to December 31, 2017).

• New users of monotherapy for SGA or FGA: Patients with no prescription for the drugs for > 180 days prior to the first prescription of SGA or FGA.

Exclusion criteria:

• Patients with a diagnosis of dementia (ICD10: F01, F03, G30, G310, etc.) in the 180 days before the first prescription of any SGA or FGA.

• Patients with a diagnosis of diabetes (ICD10: E10, E11, E12, E13, E14, O24) in the 180 days before the first prescription of any SGA or FGA.

Blood glucose

The blood glucose or HbA1c before prescribing the drug was considered to be the potential confounder. The missing proportions at each hospital cohort were approximately 30–60% for HbA1c and approximately 5–40% for blood glucose. We chose blood glucose because we were interested in the effects of moderate missing data.

Blood glucose

≧200 mg/dL

or HbA1c(NGSP)≧6.5%

Note:

• We convert HbA1c (JDS) to HbA1c (NGSP)

From the date of the first prescription of SGA or FGA to the first date of the following: occurrence of diabetes, end of exposure, end of study period (December 31, 2017), change to a different antipsychotic drug, or addition of a different antipsychotic drug, end of observation period^c.

Sex, age, year of cohort entry, hospitalization, first visit, emergency care (at the date of the first prescription), class number of concomitant medications, complications^e, concomitant medication^f (180 days prior to the date of the first prescription of any antidiabetic drug)

Overall cohort = 3,430 (SGA users: 1,087, FGA users: 2,343)

Complete cases of blood glucose = 2,990 (87.2%)

2

Whether users of rosuvastatin have a different risk of hepatic injury than atorvastatin users?

Liver damage is a known risk of HMG-CoA reductase inhibitors (statins). Both rosuvastatin and atorvastatin are contraindicated in patients with impaired liver function and have similar levels of alertness for liver damage [30, 31]. Although the results of some observational studies support this [32], atorvastatin (especially at high doses) has been reported to have a higher risk of causing liver damage than other statins [33]. Therefore, the risk of liver damage in rosuvastatin is compared with that in atorvastatin.

Inclusion criteria:

• Patients initiated with any rosuvastatin or atorvastatin during the study period (January 01, 2015, to December 31, 2017).

• New users of monotherapy for statin as in Scenario 1.

Exclusion criteria:

Patients with a diagnosis of hepatic injury. (ICD10:B18, K70-K76, K770, K778) within the 180 days before the first prescription of any statin.

ALT, ALP,

LDL-chol, TG

We selected the above four laboratory tests, thinking that the values of the liver function-related and lipid metabolism-related laboratory tests before prescribing the drug would be potential confounders.

Diagnosis of hepatic injury. (ICD10: K71-K76, K770, K778).

From the date of the first prescription of rosuvastatin or atorvastatin to the first date of the following: occurrence of hepatic injury, end of exposure, end of study period (December 31, 2017), change to another statin, or addition of another statin, end of observation period^c.

Sex, age, year of cohort entry, hospitalization, first visit, emergency care (at the date of the first prescription), class number of concomitant medication, complications^g, concomitant medication^h (180 days prior to the date of the first prescription of any antidiabetic drug)

Overall cohort = 4,119 (rosuvastatin users: 2,538; atorvastatin users: 1,581)

Complete cases of ALT, ALP, LDL-chol, and TG = 3,761 (91.3%), 2,996 (72.7%), 3,176 (77.1%), and 3,306 people (80.3%), respectively.