Risk of bias for analysis 1.7 Healthcare utilisation: respiratory‐related hospital admissions (mean number per participant) (primary analysis).

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Bösch 2007

Some concerns

Information from the author regarding allocation concealment: “Pick of envelope. Enrolment and selection were right before the start of the study‐ a selection bias cannot be fully excluded". Comment: The method used to generate the random sequence generation was not clearly reported. The information received from the author regarding allocation concealment is too concise to assess risk of bias.

High risk of bias

Effect of assignment: Comment: Blinding of participants and personnel was not reported. No information reported whether patients were analysed in the wrong intervention group or were excluded from the analysis. No intention‐to‐treat analyses were performed. Effect of adhering: Comment: Eight (21%) participants in the intervention group and one (8%) participant in the control group dropped out, without clearly reported reasons.

Some concerns

Comment: Only the patients who completed the study follow‐up were included in the baseline characteristics and analysis (per protocol analysis). Eight (21%) participants in the intervention group and one (8%) participant in the control group dropped out. Reasons for drop‐out were not clearly reported. Furthermore, the trial report provides no information about the extent of missing outcome data.

Some concerns

Comment: There was no protocol available. Some of the outcomes were self‐reported outcomes (e.g. frequency of avoidance of physical exertion, frequency of sport activities, frequency of impairment of lifestyle), which could have been influenced by knowledge of the received intervention. Blinding of outcome assessment was not reported.

Some concerns

Comment: No signs for selective outcome reporting, results were reported extensively; however, no protocol was available.

High risk of bias

Unpredictable direction of the outcome.

Bucknall 2012

Low risk of bias

“We used a minimisation technique to stratify randomisation of participants by demographic factors (deprivation category of area of residence,11 age and sex, FEV1 per cent predicted at the time of randomisation, smoking status, participation in pulmonary rehabilitation within two years, and number of previous admissions) to control for key aspects of disease severity and predictors of readmission. We constructed a computer generated sequence by using the method of randomised permuted blocks of length four, with two allocations being made at random and two by minimisation. Treatment group allocations were obtained by telephone, after baseline assessments had been made.” p 2. Comment: Random sequence allocation was adequately performed.

High risk of bias

Effect of assignment: Comment: No blinding of patients and personnel. Data for the primary outcome (time to first acute COPD hospital admission or death) were available for all patients and analysed with a Cox proportional hazards model adjusting for the stratification variables. Effect of adhering: Comment: No blinding of patients and personnel. Seventy‐five (42%) of 180 intervention group participants who completed 12 months of follow‐up were judged to have become successful self managers.

Low risk of bias

Comment: Only for secondary outcome measures (e.g. SGRQ, HADS, CSES) not all data were available for all patients randomised (Table 3).

Low risk of bias

‐

Some concerns

‐

High risk of bias

Unpredictable direction of the outcome.

Tabak 2014

Low risk of bias

"Patients were randomized using a computer‐generated randomization list (Blocked Stratified Randomization version 5; Steven Piantadosi), where randomization was applied in blocks of two and four.” p. 936. “Participants were allocated by a data manager in order of inclusion following the randomization list, placed in a sealed envelope.” p. 936. Comment: There was a significant difference in dyspnea (MRC scale) but this seems by chance. All other parameters seem evenly distributed.

Some concerns

Effect of assignment: "To present the outcome measures over time in both groups, a mixed‐model analysis for repeated measures was performed (intention to treat)." p. 938. Comment: Patients and personnel were not blinded. No information was reported regarding deviations from the intervention that arose because of the trial context. Effect of adhering: "Participants in the control group received only usual care. Patients in the control group were allowed to attend regular physiotherapy sessions if this was prescribed as part of usual care." p. 937. “Although some patients in the intervention group quit the physiotherapy modules ( exercising and activity coach) due to weak (n=l, < T l ) or unstable condition (n=l, <T2) and personal circumstances (n=2, <T4), they persisted in using the web portal and triage diary til! T4.” p. 939. Comment: Most outcome measures are reported for 3 months follow‐up, whereas there was a total of 9 months follow‐up. There was a high number of withdrawals for the 9 months follow‐up. However, there were more drop‐outs in the usual care control group than in the intervention group.

High risk of bias

“A large number of patients were not able or willing to continue study participation: 33% in the intervention group and 86% in the control group.” p. 939. Comment: Most outcome measures are reported for 3 months follow‐up, whereas there was a total of 9 months follow‐up. There was a high number of withdrawals for the 9 months follow‐up (more dropouts in the control group). Reasons reported for patients from the control group: too much effort (n=2), exacerbation (n‐1), hospitalisation (n‐2), kidney problems (n‐1), exacerbation, and cancer (n‐1).

Low risk of bias

‐

Low risk of bias

Comment: Not all outcome measures were reported for the 9 months follow‐up. Exacerbations (duration) was not reported. Also, no information or results provided for the use of diaries in the usual care control group.

High risk of bias

Unpredictable direction of the outcome.

Titova 2015

High risk of bias

Randomisation process: “They were randomly allocated to either integrated care (IC) or usual care (UC) based on their address of permanent residence. In order to create two pairs of districts with approximately equal numbers of citizens, a pair‐wise matching of districts was carried out. It was decided by lottery that participants from District Pair 1 were assigned to the UC group, and participants from District Pair 2 were assigned to the IC group.” Titova 2015, p. 2. “There were no significant differences in baseline characteristics between the IC and the UC groups, but when comparing those who died during follow‐up with those who were still alive after two years, those who died had a significantly higher PaCO2 and lower BMI at baseline (Table 1). Those who died during follow‐up also had significantly more HA and HD during the year prior to study start.” Titova 2015, p. 4. Comment: No information provided about the allocation concealment. Timing of identification or recruitment of participants: "They were randomly allocated to either IC or UC group based on their address of permanent residence. Patient data were registered at discharge and 6, 12, and 24 months after discharge from the TUH." Titova 2017, p 3. Comment: If allocation was known before study inclusion, suitable patients could be selected for the intervention.

High risk of bias

Effect of assignment: “The study was a prospective, open, single‐centre intervention study.” Titova 2015, p. 2. "The principal comparisons of changes were performed on an intention‐to treat basis." Titova 2017, p. 4. Comment: No blinding of participants and personnel. Effect of adhering: ‐

High risk of bias

“172 patients were randomly allocated to the IC group (n = 91) or the UC group (n = 81), respectively. After two years of follow‐up, a total of 100 patients (58% of the included patients), 51 patients in the IC group and 49 patients in the UC group, were available for evaluation (Figure 1).” Titova 2017, p. 5. "Those who died had a significantly higher PaCO2 and lower BMI at baseline (Table 2). Those who died during follow‐up also had significantly more HA and HD during the year prior to study start." Titova 2015, p. 4. Comment: No mortality reported; however, Figure 1 shows higher mortality for the IC group, n=35 (38.4%), compared to the UC group, n=21 (25.9%).

High risk of bias

“Information concerning the number and duration of the COPD exacerbations, as well as the time from onset of symptoms until the start of self‐initiated treatment is insufficient due to many incomplete registrations in “My COPD book.” Titova 2015, p. 9. "“Data concerning HA (hospital admissions) and HD (hospital days) were collected from the hospital registry database’s medical charts.” Titova 2015, p. 3. Comment: Unclear who was the outcome assessor.

Some concerns

Trial register reported number of hospitalisations as the primary outcome. Secondary outcomes were quality of life, lung function, use of medication, cost‐effectiveness. However, the article reported not all secondary outcomes (e.g. cost‐effectiveness).

High risk of bias

Unpredictable direction of the outcome.

Jolly 2018

High risk of bias

"When a patient is identified as eligible for the study, and has given written, informed consent to take part, the researcher randomises the patient using a web‐based programme hosted by the Primary Care Clinical Trials Unit (PC‐CTU), University of Birmingham. Centre specific randomisation lists were produced by a statistician at the trials unit. Participants are individually randomised, stratified by centre. The four recruitment centres are Birmingham and West‐Midlands South; Greater Manchester; North West Midlands and Oxfordshire/Gloucestershire. Only the PC‐CTU have access to the allocation sequence." Sidhu 2015 (protocol paper), p. 4. "The study groups were generally well balanced in terms of patient characteristics, although there was a higher proportion of current smokers in the telephone health coaching group. The usual care group reported a higher level of self reported moderate and vigorous physical activity, but this was not reflected in the accelerometry data at baseline." Jolly 2018 (results paper), p. 5.

High risk of bias

Effect of assignment: Comment: Patients were informed of their allocation at the end of the recruitment appointment; they were not masked to treatment allocation. All data were analysed by intention to treat. Effect of adhering: "We achieved good fidelity of delivery of the intervention, with 75% of intervention participants receiving all four calls and only four patients receiving none." p. 11. "The dose and coverage of intervention delivery were high: 86.4% (999/1156) of the scheduled calls were delivered and 75.4% (218/289) of participants received all four calls." No information regarding pulmonary rehabilitation, oxygen use.

High risk of bias

"There was imbalance in the follow‐up rates between telephone health coaching (82.7%; 37 withdrawals) and usual care (96.2%; 7 withdrawals)." p. 4. "Participants who did not provide data at 12 months were more likely to be in GOLD stage 3, to be smokers, had lower levels of self‐reported physical activity, and to live alone than responders. " p. 5.

Some concerns

"Patients were informed of their allocation at the end of the recruitment appointment; they were not masked to treatment allocation. Data were entered into the study database by researchers at the University of Birmingham who were masked to the treatment allocation." p. 3. Comment: No information about outcome assessors.

Low risk of bias

‐

High risk of bias

Unpredictable direction of the outcome.

Lenferink 2019

Low risk of bias

Comment: Allocation was stratified on potential confounders. Only a baseline difference in 6MWD was reported.

High risk of bias

Effect of assignment: Comment: It was not possible to blind the treatment for patients and people delivering the treatment. Intention‐to‐treat analyses were performed. Effect of adhering: "The 12‐month follow‐up was completed by 83.3% of the individuals in the self‐management group and 84.8% in the UC group (figure 1). Patients completed 81.3% (self‐management: 82.3%; UC: 80.2%) of the diary data." p. 6. Comment: The effect of adherence to the intervention was not analysed. However, it was observed that a high number of patients in completed follow‐up and their symptom diaries. Adherence to action plans was measured, but not yet reported in the current publication.

Low risk of bias

Comment: Relatively low number of drop‐outs. The imputated data analyses showed similar results as the observed data analyses.

Low risk of bias

"Wherever possible, though, assessors of outcomes were blinded to the treatment group." p. 3.

Low risk of bias

Comment: Some small changes in final analyses compared to analysis plan that was published with the study protocol.

High risk of bias

Additional bias: "Based on the exacerbation rates/patient/day from the COPE‐II study (intervention: 0.116; control: 0.176, both with an estimated standard deviation of 0.17) and allowing for overdispersion and an attrition rate of 10%, 105 patients per group provided 80% power to detect an effect of this size, with a two‐tailed p<0.05." p. 4. Comment: The study included less patients (intervention: n= 102; usual care: n= 99) than aimed by power calculation (both groups n= 105). Unpredicted direction of the outcome.

Wang 2019

Low risk of bias

"The allocation sequence was generated and released to the interventionist on a case‐by‐case basis by an independent department that specialised in supplying randomly generated sequences for research. The randomisation sequence was recorded." p. 151. "Participants were randomly assigned to groups after consent was obtained and baseline data were collected." p. 151. "There were no significant differences in demographic and clinical data between the two groups at baseline (Table 1)." p. 151.

High risk of bias

Effect of assignment: Comment: It is not possible to blind the treatment for patients and people delivering the treatment. Intention‐to‐treat analysis was used. Effect of adhering: "154 participants were enrolled, among whom, 143 completed the 12‐month study program. The attrition rate was 7.14%." p. 151. "...adherence to the self‐management intervention was reported by the participants but not monitored by the nurses." p. 155. Comment: No info on reasons of adherence by patients reported in the current publication.

Low risk of bias

Comment: Relatively low attrition rate.

Low risk of bias

"To minimize researcher bias, the interventionists who provided care were blinded to the participants’ baseline and allocation sequence. The statistician was blinded to the participants’ results during the trial." p. 151. Comment: Not possible to blind the treatment for patients and people delivering the treatment.

Some concerns

Comment: No pre‐specified analysis plan published which can be checked. However they mention: The statistician was blinded to the participants’ results during the trial.

High risk of bias

Unpredictable direction of the outcome.