1. Participant numbers in trials.

Participant numbers	Agar 2017	Au 2012	Bernacki 2019	Clayton 2007	Epstein 2017	Lautrette 2007	Reinhardt 2014	Walczak 2017
Eligible for inclusion	Nursing homes: 111 eligible Participants: 148 UC 171 intervention	1292 1173 mailed introductory letter	Clinicians: 133 recruited Patients: 9182 screened UC (8530 ineligible) 9395 screened intervention (8842 ineligible)	196	Physicians: 38 eligible Patients: 453 eligible	132	214	363
Excluded	Nursing homes: 91 (55 did not meet criteria; 36 declined) 14 UC 11 intervention	30 screened out (did not meet criteria) 21 declined participation prior to screening	Clinicians: 6 pilot clinicians ineligible Patients: 8987 UC 9211 intervention	22	Patients: 137 excluded (38 ineligible, 99 refused)	‐	‐	‐
Refused to take part	‐	645 (did not wish to participate) 101 (did not keep appointment)	Clinicians: 36 Patients: unclear	18	99	6	104	253
Randomised to intervention group(s)	10 homes, 160 participants 156 received allocated intervention	194	Clinicians: 48 (20 clusters) Patients: 184 (20 clusters)	92	Physicians: 19 Patients: 139 Carers: 105	63	58	61
Randomised to control (usual care) group	10 homes, 134 participants 130 received allocation intervention	182	Clinicians: 43 (21 clusters) Patients: 195 (19 clusters)	82	Physicians: 19 Patients: 142 Carers: 99	63	52	49
Excluded post‐randomisation (for each group; with reasons if relevant)	UC 66 Intervention 89 All excluded because did not die during the study period	‐	96 (intervention) Death; lost to follow‐up; declined further surveys; completed 2 years 118 (UC) Death; lost to follow‐up; declined further surveys; completed 2 years Intervention: 184 (20 clusters) enrolled: 134 analysed (18 clusters); (50 total: no family/friend response, no baseline survey, withdrew) 34 (13 clusters) analysed: 74 died, 11 lost to follow‐up; 17 declined further surveys; 32 completed 2 years UC: 195 (19 clusters) enrolled: 144 analysed (17 clusters); (51 total: no family/friend response, no baseline survey, withdrew) 26 patients (13 clusters) analysed: 77 died. 12 lost to follow‐up, 21 declined further surveys; 34 completed 2 years	‐	‐	‐	‐	‐
Withdrawn (for each group; with reasons if relevant)	UC 4 Intervention 4 (died before intervention)	Control 27 (14.5%) 3 patient‐clinician relationship changed 6 refuse to continue 9 not contactable 8 no target visit 1 too ill/deceased Intervention 43 (22.2%) 4 patient‐clinician relationship changed 15 refuse to continue 8 not contactable 10 no target visit 3 too ill/deceased 3 other	Intervention 12 Control 8	None	Intervention 6 Control 5	Intervention 11 Control 7	‐	‐
Lost to follow‐up (for each group; with reasons)	‐	‐	Intervention: 11, no reason given Control 12, no reason given	Intervention: 2, 1 mistakenly seen by a junior physician who was not participating in the study, 1 due to mechanical failure of tape recorder Control: 2, 1 mistakenly seen by a junior physician who was not participating in the study 1 due to mechanical failure of tape recorder	Intervention: 3 died Control: 1 died 1 lost to follow‐up	Intervention: did not answer telephone n = 4; experiencing severe emotional distress n = 5; refused interview n = 2 Control: did not answer telephone n = 3; experiencing severe emotional distress n = 1; refused interview n = 2; patient still alive n = 1	Total numbers reported at each time point N = 96/110 completed 3‐month measures N = 90/110 completed 6‐month measures Plus an additional 3 where complete data were not available NB also for several outcomes data for slightly lower numbers in total are available i.e. table 3 – ranges from n = 65 to n = 81. Not clear what happened to these missing measures	Intervention: 21 (34%) Unclear; data collection hampered by declining health, attrition high (higher in intervention group) but no systematic reasons for differential dropout identified. Reasons for loss not reported specifically Control: 9 (18%) Reasons for loss not reported specifically
Included in the analysis (for each group, for each outcome)	Intervention 67 Control 64 These numbers were analysed throughout, with losses for particular scales/assessments noted where applicable	Intervention 194 Control 182	Intervention 38 patients (13 clusters) Control 26 patients (13 clusters)	Intervention 90 Control 80	Intervention 19 physicians, 130 patients Control 19 physicians, 135 patients	Intervention 52 (83%) Control 56 (89%)	6‐month data (longest time point) Intervention 47 Control 40	Intervention 39 Control 40
Assessment of attrition bias for RoB ratings	Assessed as low risk Large proportion of data (participants randomised) missing. However, these were comparable for the 2 study groups, and was due to participants not dying during the study period (outcomes assessed for this study were focused on those around death) Withdrawal rates for other reasons were low and comparable across groups	Assessed as unclear risk Withdrawal 15% to 22% respectively control and intervention arms Reasons for withdrawal/dropout were reasonably comparable except that more (15 versus 6) refused to continue participation in the intervention group Authors report no differences in baseline characteristics regarding whether patients completed the study or were lost to follow‐up ITT analysis was used; effect of imputed data on results was examined in analysis models with authors reporting similar results where imputed and non‐imputed data were used in analysis	Assessed as unclear risk Patient participation rates and numbers analysable were low but comparable between arms Authors note non‐participants and those not analysed were not significantly different from those who were analysed, and groups were still comparable (based on randomisation), although non‐participants were older, and less likely to have breast cancer than participants; and those patients with analysable data were more likely to be married and have higher incomes than those with non‐analysable data	Assessed as low risk Low levels of loss to follow‐up 4/174; balanced across groups (n = 2 each), with comparable reasons	Assessed as low risk "Fewer than 3% of follow‐up questionnaires were missing" (page 95) Data seem otherwise complete for outcomes reported in main paper and in supplement 3	Assessed as low risk Loss to follow‐up and withdrawals were acceptably low and comparable across study groups: 52/63 (83%) completed interviews at 90 days intervention group, 56/63 (89%) in control group Reasons for withdrawal/loss were similar across groups (not answering telephone, refused interview), although higher rates of severe emotional distress in intervention group (n = 5) than control group (n = 1)	Assessed as unclear risk Missing data were reported 110 were randomly assigned and completed baseline interviews; 96 (87%) completed 3 month outcomes; 90 (82%) completed 6 month outcomes Losses were fairly comparable across groups and no major differences between those who completed the study and those who dropped out (on key demographic features) were noted by the authors However, numbers were lower for some outcomes (such as ratings of care management) where n = 65 (numbers fairly comparable between the 2 groups) Not clear what impact this may have had on the result, or what the reasons for this missing data were	Assessed as high risk Attrition was high (31/110 (28%) lost to follow‐up), possibly largely explained by declining health of participants (patients) Higher in intervention group. No systematic reasons for differential attrition were identified by authors but 34% dropout in intervention group is substantial and may introduce bias Authors state that ITT analysis was used (according to group assignment) but dropout rates may be problematic
Additional notes	The study was underpowered for the primary outcome because fewer participants died during the study period than predicted Predicted sample size required recruitment of 272 participants (assuming 10% dropout rate) (17 per site)	‐	Sample size calculated as 200 evaluable patients per arm for required power, assuming 6% dropout	‐	‐	‐	‐	Trial is underpowered (sample size calculated at 140; 110 recruited; 79 completed) to detect differences between groups

ITT: intention to treat; RoB: risk of bias; UC: usual care.