Kousgaard 2003.

Methods	Cluster‐RCT; Unit of allocation: General practitioner
Participants	Cancer patients newly referred to the department of oncology and scheduled for treatment or attendance for control. Setting / country: Department of Oncology of Aarhus University Hospital / Denmark Type of cancer: Any type Phase of care: Pre‐treatment, treatment, discharge, surveillance, recurrence, Second‐primary‐cancer Sample size at randomisation: 199 GPs (248 patients)
Interventions	Shared care program: The patient was instructed to see his/her own GP about questions and problems. A discharge summary letter was written for the GP by the department of oncology in accordance with specially developed guidelines. The discharge summary included specific information on the disease and its treatment, general information about chemotherapy, radiotherapy, pain treatment, information about treatment of induced nausea and sickness and information about some acute oncological conditions (knowledge transfer). It also stated names and phone numbers of doctors and nurses responsible for the patient in the discharge summary letter to the GPs (improved communication channels). It also aimed to improve patient involvement in their own care by providing patients with oral as well as written information about the information package to their GP, and by encouraging patients to contact their GP when facing problems they assumed could be solved in this setting. Control: Normal procedure which included no procedure for informing the GP about newly diagnosed patients. The participating practitioner received the traditional information from the department, i.e. the discharge letter of an extract from the hospital record.
Outcomes	Patient: Performance status, QoL, attitudes of patients towards contacts with the GP Process: Patient perception of cooperation within the healthcare system, number of contacts with GP (patient interview), number of contacts with patient (GP interview)
Notes	Length of follow‐up: 6 months
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote from ref #2: "A project secretary outside the hospital premises kept a list of numbers from 1 to 250 randomly arranged into two groups. After obtaining written informed consent from the patients, the investigator opened an envelope with a random number of 1 to 250. This number was communicated to the project secretary who informed the investigator of the group to which this number (patient) belonged. Once a patient was randomised to a particular group, any further patients from the same general practice were automatically assigned to the same group." Comment: From the quote above, it is not clear how the 250 numbers were randomised into 2 groups.
Allocation concealment (selection bias)	Low risk	See quote above.
Blinding (performance bias and detection bias) All outcomes	High risk	Quote from ref #2: "The study was unblinded. Patients in both groups were informed of the group to which they had been assigned as active involvement of the patients in the intervention group was part of the strategy". Comment: Data were collected though a self‐report questionnaire so assessors could not have been blinded.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: Proportions of attrition were similar in control (26%) and experimental (33%) groups. Reasons for attrition also.
Selective reporting (reporting bias)	Low risk	All outcomes described in Methods are reported in Results.
Other bias	High risk	Quote from ref #2: "We used random allocation without blinding because we wanted to involve the patients by informing them about their possibilities. This implies a risk of information bias by the GP which may have influenced the time 0 scores. Patients in the intervention group may have had more positive expectations of their GP, knowing that they would be better informed about the disease and its treatment, and this may explain why the next two assessments by the GP were relatively less positive." Quote from ref #2: "The absence of regular baseline data is a problem as the patients actually received information about the intervention before answering the first questionnaire. They had 14 days to answer the questions at time 0. We had not foreseen this bias due to positive expectations in the intervention group. It would have been a real baseline if the patients had answered the time 0 questionnaires before randomisation, but we had a practical problem with time." Comment: Baseline assessments might have been biased, but not more than the other assessments made later during follow‐up, because of unblinded trial. However, this bias might have been prevented or adjusted for using statistical adjustments.
Baseline outcomes similar?	Low risk	Patients attitude towards their GP (many items)‐global assessment: NO No. of contacts with their GP: YES EORTC: YES Performance: YES
Baseline characteristics similar?	Low risk	Quote from ref #2: "Randomisation yielded an almost equal number of patients and an almost equal distribution according to disease and sociodemographic parameters in each group (table 1). However, the randomisation produced an imbalance in age with more young patients (18 to 49 years) in the intervention group." Comment: Sensitivity analysis were performed to see if results were different between younger and older patients.
Protected against contamination?	Low risk	Comment: Patients were randomised by GP.