A total of 14,055 studies were identified through the initial search. After removing duplicates 8,169 publications were screened without the use of filtering based on title and abstract. There were 54 conflicts giving an interrater agreement of 98% (Cohen’s κ = 0.64) after the first screening. The conflicts were resolved by consensus after discussion, and 55 articles were read in full text. This second screening process resulted in 16 conflicts giving an interrater agreement of 71% agreement (Cohen’s κ = 0.40). The full-text review of articles led to discussions, and conflicts were solved. The screening and selection process is shown in the PRISMA flow diagram (Fig. 1). The updated, supplementary screening and selection process is shown in another PRISMA flow diagram (Appendix 1). No further studies were included.

PRISMA flow diagram of articles screening and selection process

Finally, 14 studies were identified that met the inclusion criteria (Table 2). Studies included were published between 2008 and 2022. There were twelve randomized controlled trials [30,31,32,33,34,35,36,37,38,39,40,41], one quasi-experimental trial [42], and one prospective lagged controlled trial [43]. Three articles [35, 39, 43] referred to detailed descriptions of the interventions in previous publications and were considered duplicate data sources, even though they are distinct articles [44,45,46]. Collectively, the 14 studies included 1,399 unique patients with the mean reported ages of 42.3 to 62.3 years. The pathologic diagnoses reported in the included studies were: Lumbar spinal stenosis [32, 33, 43], nerve root compression [35], degenerative lumbar disease [31, 37, 39] or degenerative conditions [41], lumbar disc herniation [30, 32, 34,35,36, 38, 40, 42], and spondylolisthesis [32, 39, 42]. The surgical techniques presented included discectomies [30, 34, 38, 40] including microdiscectomies [30, 34], decompressions [33, 43], and lumbar spinal fusions [33, 37, 39, 42]. The surgical techniques being used in some studies were not specified [31, 32, 35, 36, 41].

Due to the heterogeneity of the included studies concerning the setting, population, intervention, controls, and outcome measures we did not perform quantitative analyses.

The risk of bias of the 12 included RCTs are summarized in Fig. 2. Eight of the studies were overall rated as at some concerns [31,32,33,34,35,36, 39, 40], while the remaining four studies were overall rated as at high risk [30, 37, 38, 41]. Ten of the RCTs were rated as at low risk concerning bias arising from the randomization process [30, 32,33,34,35,36, 38,39,40,41]. Two of the RCTs had some concerns according to this domain [31, 37]. Concerning the domain of bias due to deviations from the intended intervention, two studies met the criteria for low risk of bias [33, 35], seven studies had some concerns [30,31,32, 34, 36, 39, 40], and three studies were assessed as at high risk of bias [37, 38, 41]. All of the RCTs, except for one [30], had a low risk of bias due to missing outcome data. Conversely, except for one study [31], all of the RCTs had some concerns due to the measurement of outcome. Finally, three RCTs were rated as at low risk of bias [35, 38, 41], while the other nine studies had some concerns [30,31,32,33,34, 36, 37, 39, 40], due to bias in the selection of the reported result.

Summary of risk of bias of the included RCT's

Of the two non-randomized studies (Fig. 3), the study of Skolasky et al. [43] was overall rated as moderate risk. The study of Guo et al. [