Association of Sodium-Glucose Cotransporter 2 Inhibitor vs Dipeptidyl Peptidase-4 Inhibitor Use With Risk of Incident Obstructive Airway Disease and Exacerbation Events Among Patients With Type 2 Diabetes in Hong Kong

Key Points Question Are sodium-glucose cotransporter 2 inhibitors (SGLT2Is) associated with risk of incident obstructive airway disease (OAD) and exacerbation events in clinical settings among patients with type 2 diabetes compared with dipeptidyl peptidase-4 inhibitors (DPP4Is)? Findings In this cohort study using electronic health data for 30 385 patients with type 2 diabetes in Hong Kong, we found that SGLT2I use was associated with a 35% reduced risk of incident OAD and a 46% reduced rate of exacerbations compared with DPP4I use. Meaning These findings suggest that SGLT2I use may provide additional protective effects against OAD for patients with type 2 diabetes and encourage further investigation.


Data Source
The Clinical Data Analysis and Reporting System (CDARS) is a territory-wide representative electronic medical database from the Hospital Authority (HA) of Hong Kong. The HA manages all 42 public hospitals and 120 public outpatient general and specialist clinics in Hong Kong. More than 90% of the known patients with diabetes in Hong Kong are under the HA's care 1 . The CDARS is an ethnically homogeneous population of about 92% Han Chinese 2 . It contains clinical records from outpatient, emergency, and inpatient visits, including diagnosis, dispensing, clinical procedures and operations, laboratory tests, and death registry records. Since CDARS is a territory-wide database covering all public hospitals and outpatient clinics in Hong Kong, lost to follow-up would be unlikely.

Exclusion Criteria
The exclusion criteria were: 1) patients in the DPP4i control group with any uses of SGLT2is before index date; 2) patients with DPP4i and SGLT2i first initiated on the same date; 3) patients of type 1 diabetes; 4) patients with prescription records of index drugs for only one day; 5) patients with diseases routinely treated with systemic corticosteroids 3 , or with diseases affecting the respiratory system, except for OAD 3 (eTable 1); 6) patients with heart failure; 7) patients who received dialysis, kidney transplant, or other tissue and organ transplant within one year before index date; and 8) patients with no laboratory measurements of HbA1c or eGFR within one year before index date. Type 1 diabetes was defined according to a previous validation study 4 : 1) the number of type 1 diabetes diagnosis records to the number of type 2 diabetes diagnosis records ratio ≥4 4 ; 2) prescribed with insulin and no other glucose-lowering agents within the first year of diabetes diagnosis 4 ; or 3) age at diagnosis <30. eGFR was estimated using the new Asian modified CKD-EPI equation 5 .

"Prevalent New-User" Design
The design matched study participants based on the length of previous exposure to DPP4is in a time-dependent manner 6 . The patients in the SGLT2i group were patients who were new users of SGLT2is or ongoing users of DPP4is who switched to/added SGLT2is. The "prevalent new-user" design matched these SGLT2i users with ongoing users of DPP4is (but have not switched to SGLT2is) based on their length of previous use of DPP4is in a time-dependent manner, meaning that a DPP4i user could be matched with multiple SGLT2i users at different time-points over the course of DPP4i use. The baseline characteristics of the matched pairs were then assessed for propensity score (PS) calculation. Unlike the conventional new-user design, it allowed SGLT2is users with previous or ongoing use of DPP4is to be included in the cohort. For patients first initiating SGLT2is without previous use of DPP4is (i.e. new users), they were matched with patients first initiating DPP4is. The "prevalent new-user" design allowed an unbiased comparison between patients who switched to or added SGLT2is from DPP4is and patients who stayed on DPP4is.

Propensity Score (PS) Calculation and Matching
PS was calculated using conditional logistic regression stratified by the pairs matched in the "prevalent newuser" design. PS matching was done within each "prevalent new-user" matched pair using sequential greedy matching 7 with a calliper of 0.2 standard deviations (SD). The patients were matched 1:4 (SGLT2is:DPP4is) without replacement within and across "prevalent new-user" matched pairs.

Clinical History
No. of glucose-lowering agents