Mazzaglia 2016.

Study characteristics
Methods	Effects of a computerized decision support system in improving pharmacological management in high‐risk cardiovascular patients: a cluster‐randomized open‐label controlled trial Clustered RCT (230 clusters and 230 providers), conducted in 1) General practitioners (GPs) from the Health Search Network research group of nearly 800 GPs representative of each Italian geographic area in terms of patient population, which sent (from 1998) all clinical information from its patient list to Health Search Cegedim Strategic Data Longitudinal Patient Database (HS‐CSD‐LPD). 2) Intervention delivered at the general practitioners level through a computerised decision support system (CDSS). In Italy. 2 arms: 1. Control (standard software, usual care) (control arm) and 2. Intervention (CDSS: computerised decision support system) (intervention arm)
Participants	Control arm N: 9326 Intervention arm N: 11904, NA, NA Diabetes type: 2 Mean age: 70.44 ± 12.68 % Male: 52.98 Longest follow‐up: 24 months
Interventions	Control arm: (standard software, usual care) 1) Clinician education Intervention arm: (CDSS: computerised decision support system) 1) Electronic patient registry 2) Clinician education 3) Clinician reminder
Outcomes	Anti‐platelet drugs Lipid‐lowering drugs
Funding source	The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by funds from the Italian Medicines Agency (AIFA).
Notes	—
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Trial allocating participating GPs to one of the 2 groups by a computerised randomisation process. The random allocation of GPs was performed using STATA software, version 10.1 (STATA Corp., College Station, TX, USA).
Allocation concealment (selection bias)	Low risk	Cluster‐RCT.
Provider's baseline characteristics (selection bias)	Low risk	Randomisation process was stratified by age and geographic location (i.e. north‐east, north‐west, central, southern, major islands); such features are in fact the most relevant predictors of physician prescribing behaviour in Italy. Characteristics of GPs (i.e. age, gender and geographic location) did not significantly differ between intervention and control groups.
Patient's baseline characteristics (selection bias)	Unclear risk	Table 2. Outcome measurements were not stratified to the diabetes subgroup, but they represented 21,230 out of 25,491 patients randomised (83.3%). Some demographic data have significant P values (congestive heart failure comorbidity, hypertension comorbidity, COPD comorbidity, Charlson index and familiar anamnesis with diabetes).
Patient's baseline outcomes (selection bias)	High risk	Table 2. Outcome measurements were not stratified to the diabetes subgroup, but they represented 21,230 out of 25,491 patients randomised (83.3%). Some outcomes have significant P values (DBP, total blood cholesterol levels and mean number of concurrent drug use).
Incomplete outcome data (attrition bias)	High risk	230 GPs have been randomly allocated to receive either the CDSS (intervention group: 115 GPs) or paper‐based information (control group: 115 GPs). Only 197 GPs, 106 in the intervention group (8% lost) vs 91 in the control group (21% lost) sent all baseline information about eligible patients (14% lost overall). High and unbalanced numbers.
Blinding of participants and personnel (performance bias) and of outcome assessors (detection bias)	High risk	Use of recommended cardiovascular drugs reported by GPs (subjective outcome). GPs are not blinded (cluster‐randomised, open‐label controlled trial). Quote: "GPs voluntarily agreed to collect patient information and attend specified training courses for data entry into a specific designed software used for managing patient information during their routine practice."
Selective reporting (reporting bias)	Unclear risk	No registered or published protocol. They added sub‐analysis by baseline exposure (per GPs tertiles) on figures 2 and 3.
Risk of contamination (other bias)	Low risk	Cluster‐RCT. Unlikely that control GPs received alerts from the computerised decision support system integrated into standard software. However, since the randomisation was at the clinician level, it is possible that physicians working in the same clinic communicated together.
Other bias	Low risk	No evidence of other bias.