The electronic database searches identified 3890 records, from which we removed 1298 duplicate records. From the remaining 2592 references, 2557 were excluded by title and abstract. This resulted in 35 records for the full-text review. After this, we excluded six records for failing to meet the inclusion criteria (three studies did not report an evaluation of functional capacity measured by the 6mWT. Two of them reported the endurance shuttle walk test27,28 and found no differences between groups in this test, and one study evaluated physical activity29. Three studies did not have an initial PR programme or the programme only included physical exercise, three were not a PR intervention, and one study was still ongoing30) (Fig. 1). We were unable to contact the authors of 5 records. Six citations reported duplicated data, referring to the same studies. Finally, 17 studies were kept after careful inspection. The k value was 0,85, indicating an almost perfect agreement13.

PRISMA Flowchart for Study selection.

The risk of bias full details may be found in “Supplementary Information 3 - Quality Assessment and Risk of Bias Table”. All studies had no blinding of participants and personnel, which was due to the nature of the intervention itself. Four studies presented a lack of blinding as the only source of bias, representing the best bias-free design possible for this kind of intervention. In addition, more than half of the included studies did not refer to the blinding of the outcome assessment, which is an important step that may overcome the difficulty upon blinding the intervention (Fig. 2).

The funnel plot analysis detected no relevant publication bias for all outcomes (see “Supplementary Information 4 – Complete meta-analysis data”).

This review included 17 studies,14 of which were randomised controlled trials31,32,33,34,35,36,37,38,39,40,41,42,43,44, and 3 were quasi-experimental trial (two studies had no control group45,46, and one was not randomized47). Detailed information on included studies is available in “Supplementary Information 2 - Complete data of selected studies”.

The publication year varies from 200231,32 to 202140. The mean participants’ ages ranged between 62 and 69 years old, and the mean percentage of predicted FEV1 was between 34 and 66%.

The studies included 1222 participants, all receiving an initial PR programme, varying between three39 and twelve weeks33,34,36,38 duration. Among these, 1151 participants were randomly assigned to either a control or an intervention group. Forty participants were allocated to one of the groups, though not randomly47.

The intervention group had access to a maintenance PR programme, varying between 231 and 3636 months. The presence of a healthcare provider supervision was variable (Table 1); in five studies, the intervention was always supervised31,33,43,47. In the other five, there was no supervision32,34,39,40,42, in six studies some of the sessions were supervised35,37,38,41,44.

In six studies, intervention was assessed through a home-based programme31,34,39,41,43, three took place in a community centre33,37,47, four studies used programmes that incorporated both home-based and community-based sessions32,35,38,44, and two studies assessed telerehabilitation40,42.

Five studies showed a significant improvement in functional capacity31,35,36,43,47 after the maintenance programme. In general, they used strategies for measuring programme adherence through daily exercise records performed by patients36. Within this group, three studies included home-based supervised programmes31,36,43.

In five studies36,40,42,43,47, education for self-management was provided. Only one study controlled the exercise intensity by the patient’s maximum heart rate and symptoms43. Pedometers were used in two studies38,39, and an experiment with music during exercise training and walks was explored in one study31.

One study36 included structured healthcare provider supervision regarding long-term programme transitions, and another study43 modified the intervention over time, reducing the number of sessions supervised in stages and alternating with telephone contacts.

Detailed Risk of Bias assessment in included studies according to GRADE system.

This review included 15 studies in the meta-analysis31,32,33,34,35,36,37,38,39,40,41,42,43,44,47, two of which were excluded45,46 because they lacked a control group. The two studies with more weight were Wetering et al. and Brooks et al. 32,36. We analysed the changes after the initial PR until the end of the maintenance and calculated the MD between these two moments.

Most studies showed an effect favouring maintenance PR programmes over usual care, of which five studies showed high certainty evidence (Fig. 3). The pooled estimate for the maintenance group is a MD of 27 meters (CI: 10.4 to 43.8; I2 = 93%; p < 0.0001) (Fig. 3), which lies between the minimal important difference confidence interval that ranges between 25 and 3314.

Forest plot of MD in 6mWT with maintenance PR programme, according to supervision type.

Considering the severe heterogeneity detected across studies (I2 = 93%), we performed a sensitivity analysis with the leave-one-out method. Only one study showed an effect favouring usual care over maintenance PR programmes with high certainty evidence32. Although it revealed a slight reduction of heterogeneity to I2 = 81%, no significant changes were found in the pooled estimate. Supplementary Information 4 summarises the main findings of clinical outcomes from selected studies.

Nine studies33,34,37,38,39,40,41,42,44 showed low certainty evidence of the effect, in which 439,40,41,44 the effect favoured the maintenance PR group over usual care. These studies included interventions with health technologies and telerehabilitation and were mainly unsupervised.

Subgroup and sensitivity analyses are available in detail in “Supplementary Information 4 - Complete meta-analysis data”.

The sub-analysis according to maintenance PR programme duration found a difference in the 6mWT favouring maintenance compared to the usual care group in programmes with more than 12 months36,41 with a MD of 14 meters with high certainty evidence and a significative reduction of heterogeneity (CI 12.8 to 15.0; I2 = 0%; p < 0.38). However, although programmes with eight weeks favoured maintenance PR programmes over usual care, they had high heterogeneity. Shorter programmes had low certainty evidence and high heterogeneity, although they also favoured maintenance PR programmes over usual care.

Subgroup analysis according to the initial PR programme duration revealed a difference in the 6mWT favouring maintenance compared to the usual care in studies with the initial PR programme of 12 weeks