Gill 2019.

Study characteristics
Methods	Using electronic clinical decision support in patient‐centered medical homes to improve management of diabetes in primary care: The DECIDE Study Clustered RCT (12 clusters and 59 providers), conducted in 1) The main setting was small‐ to medium‐sized independent primary care practices in Delaware that were already participating in statewide PCMH projects, of which there were 39 at the time the study was initiated. In addition, 10 offices in Maryland that were in a joint Delaware‐Maryland Accountable Care Organization (which assisted offices to implement PCMH principles) were also eligible. There was 1 cluster of practices from the state of Maryland and 5 clusters of practices from the state of Delaware. Only offices that already had EHR systems in place were eligible. 2) The CDS system generated reports for the practice staff and clinicians before each appointment in United States of America. 2 arms: 1. Control (no intervention) (control arm) and 2. Intervention (CDS: electronic health record–based clinical decision support tools) (intervention arm)
Participants	Control arm N: 1902 Intervention arm N: 4484, NA, NA Diabetes type: 3 Mean age: 61.09 ± 14.46 % Male: 50.67 Longest follow‐up: 12 months
Interventions	Control arm: (no intervention) Intervention arm: (CDS: electronic health record–based clinical decision support tools) 1) Audit and feedback 2) Case management 3) Electronic patient registry 4) Clinician reminder 5) Facilitated relay of clinical information
Outcomes	Glycated haemoglobin Low‐density lipoprotein
Funding source	This study was funded by Sanofi US, Inc.
Notes	—
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Method not reported. Overall, 41 practices were eligible for inclusion, of which 15 agreed to participate, and 12 were randomised. Of the remaining 10 offices (52 clinicians), 5 (23 clinicians) were randomised as controls. Clustered randomisation was based on practice size (no minimum) to reduce bias by office and by US state.
Allocation concealment (selection bias)	Low risk	Clustered trial.
Provider's baseline characteristics (selection bias)	Unclear risk	Practice characteristics not reported. Clustered randomisation was based on practice size (no minimum) to reduce bias by office and by US state. This required collection of pre‐enrollment surveys that provided precise information on practice characteristics. They just mention that the control offices had fewer patients per clinician and fewer patients with diabetes.
Patient's baseline characteristics (selection bias)	High risk	Table 1. Age group and mean age have significant P values (both < 0.0001). Quote: "Patients in the CDS group were significantly older than those in the control group". The 2 groups were similar in terms of gender and diabetes complications.
Patient's baseline outcomes (selection bias)	High risk	Table 1. HbA1c and LDL have significant P values (< 0.0001 and 0.0086, respectively). Quote: "Patients in the CDS group had better glycemic control and lower LDL‐C than those in the control group."
Incomplete outcome data (attrition bias)	High risk	Of the 15 offices that agreed to participate, 3 were excluded prior to randomisation and 2 were excluded after randomisation. Only 10 offices were analysed. At baseline, they have HbA1c data for 1068/1902 and 2732/4484 for control (44% lost) and intervention (39% lost) patients randomised. At 1 year, they have HbA1c data for 723/1902 and 2041/4484 for control (62% lost) and intervention (55% lost) patients randomised. Quote: "However, the main limitation of this study was the amount of missing A1C and LDL‐C data. Although some of these missing data were due to patients not undergoing the relevant tests at suitable time points (as would be expected in a real‐world study), most were due to problems of interoperability. A lack of interoperability also resulted in the exclusion of 5 offices."
Blinding of participants and personnel (performance bias) and of outcome assessors (detection bias)	Low risk	HbA1c and LDL were objectively measured.
Selective reporting (reporting bias)	Unclear risk	No registered or published protocol. Results match methods.
Risk of contamination (other bias)	Low risk	Clustered trial. Unlikely that control clusters had POC CDS systems implemented as an add‐on product to their EHRs. It was not possible to identify whether patients moved between practices (and therefore potentially between systems).
Other bias	Low risk	No evidence of other risk of bias.