Katz 2004.

Methods	Study design: Cluster randomised controlled trial. The study report describes this as a randomised controlled trial. Duration of study: 40 weeks. Recruitment: Invited all practising physicians in the four participating clinics.
Participants	Description: Physicians (faculty and resident) in internal medicine and family practice in primary care clinics. Setting: Four university affiliated primary care clinics, Michigan, Mid‐West America. There were no inclusion and exclusion criteria. Physician sample consisting of 132 physicians (41 faculty physicians, 91 resident physicians) randomised, 65 to intervention group, 67 to control group.
Interventions	Intervention: Secure web‐based patient ‐provider communication tool to allow communication with clinic staff. Web site contained educational content addressing appropriate message content, expected response times and message handling by clinic staff. Patients prompted through regular e‐mail to enter the Website to read responses from the staff. Patients of intervention physicians were encouraged to use the Web system through promotion, including cards distributed by intervention physicians and brochures mailed to patient homes. Control: Did not have access to the secure web‐based patient ‐provider communication tool.
Outcomes	Number and type of web‐messages (from website) . Email volume (physician recall) Telephone call volume by type of call and physician (staff logs) (all measured during a one‐week collection period across the 40 week duration of the study, no detail on frequency of one‐week period) Attitudes toward web and e‐mail communication, preferences for different modes of communication and satisfaction with communication (via physician questionnaire at end of study).
Notes	Study also featured an end of study patient satisfaction questionnaire, however this was not part of the randomised controlled trial. Participant groups for this survey (intervention and control) were derived from a random sample of 900 patients (425 who had seen an intervention physician 1 or more times and a control physician no more than once during the study period and 425 patients who has seen a control doctor one or more times during the study period and an intervention physician no more than one time during the study period). Therefore this data was not included in the review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	The study randomised by half‐day clinic block; all physicians seeing patients during a certain half day window were randomised to treatment or control. This is a cluster randomisation technique. A coin was flipped, assigning 'heads' to treatment, 'tails' to control.
Allocation concealment (selection bias)	High risk	Contact with the author confirmed that the study did not conceal the random allocation sequence.
Blinding (performance bias and detection bias) All outcomes	High risk	Contact with the author confirmed that staff and investigators were not blind to group assignment.
Incomplete outcome data (attrition bias) All outcomes.	High risk	The response rate to the physician survey was 71.2% however no information was given on characteristics of non‐responders. An intention to treat analysis was not carried out. For email volume on average 91.7% of staff physicians responded vs 67.4% of residents. Missing estimates for residents were imputed to zero 'because feedback suggested that these residents had low email use with patients'
Selective reporting (reporting bias)	Low risk	Physician preferences for mode of communication for selected health issues are presented for the whole physician sample and not by intervention and control groups. All other outcomes are presented by intervention and control group.
Other bias	Unclear risk	Baseline comparability: Potential concerns with cluster randomised trials include baseline imbalances between the randomised groups, with regard to individuals and groups. No significant differences were observed between the intervention and control groups at individual level for the demographic data that is presented (faculty physicians, female sex and mean number of scheduled visits per week) however because the authors were not aware that they had carried out a cluster randomised trial they did not assess comparability at group level. Validation of measures: No information is given on whether the measures are validated. It is unclear whether the utilisation variable created for analysis purposes is validated despite the claim that volume of communication is highly correlated with the level of clinical activity of individual physicians. Reliability of measures: Cronbach's alpha coefficient for internal consistency reliability is presented for the different scales used in the physician survey: 'web benefits scale' 4 items, α =.88, general communication scale, 4 items, α = .82.As a higher score means higher reliability, and a score of 0.7 is a recommended threshold for reliability, then the reliability of these scales is acceptable. Recall and reporting bias: for email volume physicians were asked to recall the number of email messages received directly from patients during the previous week. The physician survey was also self‐administered. Information on phone call volume and type of call was collected from staff logs, these may be subject to some reporting unreliability.