Du Pen 1999.

Methods	Cluster‐RCT; Unit of allocation: Physicians' practice
Participants	Ambulatory patients with diagnostic evidence of locally invasive or metastatic solid tumours and with at least a 6‐month life expectancy and a screening pain score of at least 3 on a scale of 0 to 10. Setting / country: Practices of 13 Western Washington oncology physicians / USA Type of cancer: Any type Phase of care: Palliative care Sample size at randomisation: 96
Interventions	Treatment algorithms for cancer pain management in the community setting: The algorithm was based on the Agency for Health Care Policy and Research Guidelines for Cancer Pain Management. The process is operationalised with a set of tools, starting from the initial assessment. A clinic flow sheet is used to document the intensity of the pain, note the presence of any neuropathic pain character, and note the presence of any pain‐ or analgesic‐related side effects. A bulleted set of analgesic guiding principles for opioids, nonsteroidal antiinflammatory drugs, tricyclic antidepressants, and anticonvulsants is available for the oncology clinic staff for reference. The algorithm decision tree directs the oncologist/oncology nurse to comprehensive side effect protocols, equianalgesic conversion charts, and a primer for intractable pain. A flow sheet for each patient's chart was created to monitor significant pain and symptom indicators against their analgesic therapy. All of these tools were designed with the goal of maximum ease of use in the outpatient oncology setting. Patients had an initial clinic visit with the pain algorithm physician, at which time the intervention was initiated. The study nurse facilitated the assessment of pain and side effects as outlined by the algorithm and titrated medications under the direction of the algorithm physician. Patients were instructed regarding their role in the algorithmic process, and the importance of reporting increased and/or unrelieved pain or side effects was stressed. The pain intensity represented the first level of algorithmic treatment decision making, the pain character represented the second level. The algorithm also drove routine reassessment. The most recent pain intensity score determined frequency of contact. Control: Standard‐practice: Pain management by patients' community oncologists, who used their usual pain management and side effect strategies and documented in their usual fashion.
Outcomes	Patient: Pain, symptoms, QoL, satisfaction with pain management, pain Professional: Pain management
Notes	Length of follow‐up: 3 months
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote: "The patients were randomised within referring physicians’ practices in permuted blocks such that an approximate balance between treatment arms and the treatment assignment of each patient was not predictable."
Allocation concealment (selection bias)	Unclear risk	See quote, first item.
Blinding (performance bias and detection bias) All outcomes	Low risk	Quote: A data collection nurse who recorded outcome data, but was blinded to patient treatment randomisation, collected data for both the algorithm and standard practice groups."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: (See Table 1) Similar proportion of drop out in the 2 groups with similar reasons for dropping out.
Selective reporting (reporting bias)	Low risk	All outcomes in the methods section are reported in the results section.
Other bias	Low risk	No evidence of any other bias.
Baseline outcomes similar?	Low risk	Quotes: "There were no significant differences between the two groups on any of the baseline variables" "The algorithm treatment was the main effect influencing usual pain reduction, even when the two strongest confounders (chemotherapy and patient adherence) were introduced using analysis of covariance techniques." "Two‐way analysis of variance, using patient adherence as a covariate, indicated a significant confounding effect of non adherence on worst pain reduction (P < 0.02), whereas reduction in usual pain was statistically correlated with primary treatment effect (P < 0.02), despite the introduction of the adherence effect." "There was no statistically significant difference between the treatment groups in type of chemotherapy administered." "When chemotherapy was factored out, patients in the algorithm group had significantly lower worst pain scores than patients in the standard‐treatment group in both early (t = ‐2.70, P < 0.008) and late (t = ‐2.2, P < 0.04) phases of the study". Comment: The authors do not present baseline data for all the outcomes listed in Methods Section, although they state they are not different at baseline. See description of each outcome. They also acknowledge the influence of 2 confounding variables (chemotherapy and patient adherence) which they took into account in statistical analysis.
Baseline characteristics similar?	Low risk	Quote: "Baseline demographic and descriptive data were similar for patients in the algorithm and standard‐practice groups (Table 2). There were no significant differences between the two groups on any of the baseline variables."
Protected against contamination?	Low risk	Patients were cluster randomised by practice so contamination was prevented.