Irwin 2015.

Methods	RCT; duration: 12 months
Participants	Inclusion Postmenopausal women, HR‐positive, stage I‐III BC diagnosed 0.5‐4 years prior to enrolment On AI for at least 6 months Arthralgias for ≥ 2 months, with BPI‐SF score ≥ 3/10 Pre‐existing joint pain allowed if worsened after AI Physically inactive: baseline < 90 min exercise/week, no strength training Recruited from 5 hospitals in Connecticut, USA from June 2010‐December 2012 n = 121 (61 in exercise and 60 in usual care group). Lost funding during the study, so not all participants able to complete entire 12‐month programme. (45 participants completed programme in exercise arm vs 38 in control arm completed 12‐month programme) Mean age: exercise group: 62 ± 7 years; usual care 60.5 ± 7 years No significant differences between groups: 85% vs 84% non‐Hispanic white; 1.9 vs 1.8 years since starting AI; 52% vs 42% on pain medication; 32 vs 49% had pre‐existing arthritis
Interventions	Intervention Supervised resistance training twice a week at local health club, plus 150 min/week home‐based aerobic exercise (brisk walking, cycling) for 12 months Participants wore heart rate monitors during each workout Physical activity log book Comparator Usual care Given written information regarding cancer topics Monthly phone calls (attendance to monthly phone calls was 53%)
Outcomes	Primary outcome (Questionnaires done baseline, 3, 6, 9, 12 months) Modified BPI assessed worst pain, pain severity and pain interference. Altered 'pain' phrase to 'joint pain/stiffness.' WOMAC index DASH questionnaire Grip strength Adherence to exercise intervention (Arem 2016) Secondary outcome Pain medications at 6 and 12 months AI adherence Change in weight and physical activity at 6 and 12 months Safety data available Collected QoL data using SF‐36, FACT‐G, FACT‐B and FACIT‐Fatigue Preplanned subgroup analysis of those with pre‐existing joint pain
Notes	Study funding from National Cancer Institute, Breast Cancer Research Foundation, Yale Cancer Centre, and National Center for Advancing Translational Science.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "Permuted block randomization (at 1:1 ratio) with random block size was performed, stratified by joint pain before AI therapy and current bisphosphonate use"
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not described in article/protocol.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and personnel unable to be blinded due to nature of the intervention.
Blinding of outcome assessment (detection bias) All outcomes	High risk	PROs completed by unblinded participants. Grip strength may be influenced by motivational encouragement by assessors who are unblinded.
Incomplete outcome data (attrition bias) All outcomes	High risk	Incomplete outcome data. Quote: "Given funding cuts, the last 25 of the 121 patients recruited were enrolled into a 6 month rather than a 12 month trial. Thus their study compliance was based on the 6 month data". Dropout imbalanced between arms: 95% completed 6 months in intervention arm, vs only 82% in usual care group. 94% vs 80% at 12 months. Dropout reason not stated.
Selective reporting (reporting bias)	Unclear risk	Protocol published prospectively with many outcomes. Most are published in the primary publication and a variety of secondary publications. Eg, QoL reported in Baglia 2019. Some primary psychological outcomes are not reported, and stiffness subscale not reported from WOMAC. Stiffness was one of the outcomes of our review.
Adherence	Low risk	Intervention arm averaged 119 min/week of aerobic exercise, with average 70% of strength training sessions completed.
Contamination	High risk	Increase in physical activity levels in the control arm. Women randomly assigned to exercise increased their physical activity an average of 159 min/week, compared with 49 min/week in the usual care group (P = 0.001)
Other bias	Low risk	No other sources of bias identified