3.1 Basic characteristics of included literature

Figure 1 displays the flow diagram of PRISMA. A total of 1811 related studies were retrieved from Cochrane Library Central Register of Controlled Trials (n = 561), PubMed (n = 351), Embase (n = 424), and Web of Science (n = 475). After the removal of duplicates, the titles and abstracts of 1284 documents were comprehensively reviewed. Finally, 11 full-text articles involving 1250 patients were finally included in this meta-analysis [26,27,28,29,30,31,32,33,34,35,36]. All the included literature were prospective RCTs. Table 1 displays the characteristics of each included study. Eleven included studies were prospective RCTs, mainly from three regions, Asia (n = 5) [26,27,28, 34, 36], Europe (n = 5) [29, 30, 32, 33, 35], and America (n = 1) [31]. Eight studies included lung cancer patients of all ages [26, 28,29,30,31,32,33, 35], and three studies included elderly lung cancer patients [27, 34, 36]. In 6 studies, the lung cancer patients who planned to have pneumonectomy underwent pulmonary rehabilitation training before operation [28, 31, 32, 34,35,36], and in 5 studies, the pulmonary rehabilitation plan ran through the whole hospitalization length of patients [26, 27, 29, 30, 33].

Fig. 1

PRIMSA study flow diagram of the selection process. PRISMA: preferred reporting items for systematic reviews and meta-analysis

Table 1 Baseline characteristics of the included randomised controlled trials3.2 Risk of bias assessment results

Figure 2 shows the quality evaluation of all included RCTs. All included literature has high implementation bias because the research subjects and investigators were not blinded, but they have a low bias in loss to follow-up and reporting. Although investigators and subjects couldn’t be blinded, these studies are highly reliable in other respects. Of the 11 included studies, 7 had selection bias [26, 28, 31,32,33,34,35] and 1 had detection bias [33].

Fig. 2

A Risk of bias assessment, B Risk of bias summary

3.3 Primary outcomes3.3.1 FEV1

Two studies reported FEV1 in the two groups, involving 211 patients [34, 35]. The pooled analysis found no significant difference in FEV1 after pairwise comparison (SMD: − 0.17; 95% CI: − 0.44, 0.10; P = 0.22), and the heterogeneity was insignificant (I2 = 0%, P = 0.43) (Fig. 3A).

Fig. 3

Meta-analysis of pulmonary rehabilitation comparison of primary outcomes in preoperative studies A FEV1 B FVC C PEF D 6WMD E The incidence of postoperative pulmonary complications. CI confidence interval, SMD standard mean difference, MD mean difference

3.3.2 FVC

Two studies involving 211 patients were included in FVC analysis [34, 35]. This pooled analysis showed that the difference in FVC after pairwise comparison was not statistically significant (SMD: − 0.13; 95% CI: − 0.40, 0.14; P = 0.35), with insignificant heterogeneity (I2 = 0%, P = 0.80) (Fig. 3B).

3.3.3 PEF

Two studies reported PEF of the two groups, involving 161 patients [34, 36]. The results showed that there was no significant difference in PEF after pairwise comparison (SMD: 0.10; 95% CI: − 0.21,0.41; P = 0.52), and the heterogeneity was insignificant (I2 = 0%, P = 0.87) (Fig. 3C).

3.3.4 6MWD

Seven studies involving 719 patients reported 6MWD of the two groups [26, 29, 31,32,33,34, 36], with the pooled results suggesting that there was no statistical significance in 6MWD between the two groups (MD: 25.36; 95% CI: − 0.76, 51.48; P = 0.06), but significant heterogeneity was observed (I2 = 87%, P < 0.00001) (Fig. 3D).

3.3.5 3.3.5 The incidence of PPCs

A total of 587 patients from 5 studies were included in the analysis of PPCs [27, 28, 33, 34, 36]. The results showed that the patients who received systematic PPRT had a lower risk of experiencing PPCs compared to the control group (RR: 0.39; 95% CI: 0.25, 0.60; P < 0.0001), with a statistically significant difference. There was no significant heterogeneity in the results across the included studies (I2 = 0%, P = 0.94) (Fig. 3E).

3.4 Secondary outcomes3.4.1 Borg dyspnea score

Three studies reported Borg dyspnea score [33, 34, 36], involving 301 patients. The summary results revealed that there was no significant difference in Borg dyspnea score between the intervention group and the control group (MD: − 0.35; 95% CI: − 0.74, 0.04; P = 0.08), no significant heterogeneity (I2 = 0%, P = 0.99) was observed (Fig. 4A).

Fig. 4

Meta-analysis of pulmonary rehabilitation outcomes in preoperative studies. A Borg dyspnea score B Borg fatigue score C Hospitalization length D Quality of life E Hamilton anxiety and depression scores. CI confidence interval, SMD standard mean difference, MD mean difference, RR risk ratio

3.4.2 Borg fatigue score

Two studies reported the data of Borg fatigue score [34, 36], concerning 161 patients. The pooled results showed that there was no significant difference in Borg fatigue score after pairwise comparison (MD: − 0.14; 95% CI: − 0.85, 0.57; P = 0.70), and no significant heterogeneity was observed (I2 = 0%, P = 0.89) (Fig. 4B).

3.4.3 Hospitalization length

Seven articles regarding 815 patients were included in the analysis of the hospitalization length [27, 29,30,31, 33, 34, 36]. The results showed that there was no statistical significance in the hospitalization length after pairwise comparison (MD: − 0.23; 95% CI: − 1.05, 0.58; P = 0.58), but statistical heterogeneity was observed (I2 = 83%, P < 0.00001) (Fig. 4C).

3.4.4 Quality of life

Five studies involving 614 patients reported the quality of life between the two groups [27, 31, 33, 34, 36]. The pooled results showed that compared with the patients receiving routine nursing care, the patients receiving PPRT exhibited significant differences in quality of life (SMD: 0.16; 95% CI: 0.01, 0.32; P = 0.04), and there was no significant heterogeneity (I2 = 0%, P = 0.56) (Fig. 4D).

3.4.5 Hamilton anxiety and depression score

Two studies concerning 160 patients reported HAD scores between the two groups [26, 31]. The results showed no significant difference concerning this indicator between the two groups (SMD: − 0.00; 95% CI: − 1.49, 1.48; P = 1.00), but obvious heterogeneity in statistical analysis was observed (I2 = 95%, P < 0.00001) (Fig. 4E).

3.5 Sensitivity analysis and publication bias

We executed the sensitivity analysis for 6MWD and the hospit