Study selection

We identified a total of 6185 records from six databases (figure 1) and found 1133 records from forward citation. After removing duplicates, a total of 5857 titles and abstracts were screened, and 78 full-text articles were considered for inclusion by the pairs of reviewers. All disagreements and decisions were discussed within the multidisciplinary team and 62 articles were excluded (supplementary table S1). Thus, we included 16 articles in our review [50–65]. No additional papers were added from the pre-publication update.

Characteristics of included studies

Of the 16 included studies, 15 were individually randomised trials, and one was a cluster randomised implementation trial [59]. The latter, while relevant to our inclusion criteria, had a very different trial design informing the challenge of implementing home-PR within routine COPD care, rather than providing evidence of effectiveness, and we therefore did not include it in the meta-analysis. Eight studies compared home-PR versus usual care [50, 53, 58, 59, 61, 63–65] and seven studies compared home-PR versus centre-PR [51, 54–57, 60, 62]. One study compared home-PR against two different comparators (centre-PR and usual care) and is therefore included in both analyses [52] (supplementary table S2 presents key characteristics of included studies, main findings and interpretation).

The trials were conducted in Australia (n=3) [50, 56, 59], Brazil (n=2) [52, 63], Spain (n=2) [54, 65], the United Kingdom (n=2) [57, 58], Canada (n=1) [60], China (n=1) [51], Denmark (n=1) [55], Egypt (n=1) [53], India (n=1) [64], Iran (n=1) [61] and Turkey (n=1) [62]. Of these, nine were high-income countries [50, 54–60, 65], four were upper-middle-income countries [51, 52, 62, 63] and three were lower-middle-income countries [53, 61, 64].

All studies were in people with COPD. In total, 1800 people with a range of severities were recruited to the included trials (range 39–314 participants). Out of the 1733 participants with reported baseline demographic data, 1048 (62%) were male and the mean age ranged from 56 to 79 years.

All pulmonary rehabilitation programmes included either aerobic and/or resistance exercises (aerobic (n=15) [50, 52–65], resistance (n=13) [50–60, 62, 63], both (n=12) [50, 52–60, 62, 63]). Stretching exercises were included in two trials [52, 63] and inspiratory muscle training in one trial [53]. All studies except one [57] had 24 or more exercise sessions; five trials had more than 48 sessions of exercise [50, 53, 56, 58, 64]. All but two [55, 62] of the home-PR programmes included face-to-face training sessions either as inpatients [53, 61], outpatients [51, 52, 54, 57, 58, 60, 63–65] or home visits [50, 56, 59]. Most of the programmes described some form of supervision of the home-based sessions, most commonly telephone calls [52, 56–59, 61, 63, 65] although one used videoconferencing [55] and one study in housebound individuals provided repeated home visits. Other strategies included provision of a manual or written information [51, 57, 58, 61, 62] activity diaries [50, 52, 55, 56, 60, 62, 63, 65], pedometers [54, 56, 65] and heart rate monitors [52].

Risk-of-bias assessment

Only three studies were at overall low risk of bias [55–57]. Two were at unclear/moderate risk of bias [58, 60] and 11 were at high risk of bias [50–54, 59, 61–65] (supplementary figure S1). Blinding of participants and personnel is impossible due to the nature of the intervention, but only six studies ensured outcome assessors were blind to allocation [54–58, 60]. Computer-generated randomisation sequence was used in 10 studies [51, 52, 55–61] and allocation concealment was described in seven [50, 54–58, 60]; the remaining studies did not provide sufficient information on randomisation [53, 62–65]. We were able to compare reported outcomes with published protocols or trial registrations for six studies [51, 52, 55–57, 59], all of which were judged to be at low risk of selective reporting bias. Without a protocol for comparison, the remaining studies were designated as unclear risk of bias [50, 53, 54, 58, 60–65].

Uptake, engagement, completion and trial attrition (objective 3)

Table 3 shows details of recruitment, uptake, engagement, completion of pulmonary rehabilitation sessions and trial attrition.

TABLE 3

Recruitment, uptake, engagement and completion of pulmonary rehabilitation (PR) sessions and trial attrition

Screening and eligibility for the trials

Nine studies [51, 52, 55–61] provided details of the eligibility screening process, reporting recruitment rates between 12% and 56%. Five trials cited the presence of comorbidity as a reason for excluding between 3% and 14% of screened participants [51, 55–58]. Three studies reported that approximately one in five (22.8%, 18.3% and 12.0% [55–57]) potentially eligible patients declined to participate because of a strong preference for centre-PR. In contrast, one trial comparing home-PR versus centre-PR excluded 55% because they definitely wanted home-PR [55]. Distance/travel was cited as a reason for nonparticipation in two trials [51, 60].

Uptake of pulmonary rehabilitation

The implementation cluster RCT reported an uptake of 66% among the 107 patients referred by their general practitioner [59]. Two trials reported that two patients did not attend any pulmonary rehabilitation sessions [50, 65].

Engagement with the programme

Only four studies defined “engagement” as a pre-determined proportion of pulmonary rehabilitation sessions attended [55, 56, 59, 60]. Using the widely cited 70% threshold [67], Holland et al. [56] showed that engagement with home-PR was nearly twice that of centre-PR (91% versus 49%; relative risk of nonengagement in centre-PR: 1.91, 95% CI 1.52–2.41). In contrast, two studies [55, 60] showed no between-group difference, although the latter used a lower threshold (≥60%) and reported that >90% of the participants in both groups achieved this threshold. The implementation cluster RCT reported 46% engaged with ≥70% of the pulmonary rehabilitation programme.

Completion of post-PR assessment and trial attrition

In the trial context, completion of the post-PR assessment was generally reported as attrition (i.e. loss to trial follow-up). Rates of attrition at the post-PR follow-up assessment ranged from 0% to 51%, but with no consistent pattern to suggest that mode of delivery affected follow-up.