The Effect and Safety of Thunder-Fire Moxibustion for Low Back Pain: A Meta-Analysis of Randomized Controlled Trials

Background Low back pain (LBP) is considered the leading cause of people living with years of disability worldwide. Notably, thunder-fire moxibustion (TFM) is a new type of moxibustion, which has been widely applied to treat pain syndromes for thousands of years. This study aims to provide evidence to evaluate the effect and safety of TFM in treating LBP. Methods A systematic search of PubMed, Web of Science, the Cochrane Library, Embase, EBSCO, CNKI, Wanfang Data, CBM, and VIP (until April 2021) was used to identify studies reporting pain intensity, disability, Japanese Orthopedic Association (JOA) score, and quality of life in patients with LBP. Randomized controlled trials (RCTs), which compared TFM and other therapies in LBP, were included. Meanwhile, methodological quality was evaluated using the Cochrane criteria for risk of bias, and the level of evidence was rated utilizing the GRADE approach. Results Twenty-one RCTs, including 2198 patients, satisfied the inclusion criteria. Compared with other therapies, the effect of TFM was statistically significant, pain intensity decreased (SMD = 0.94; 95% CI (0.74, 1.14); p < 0.00001), disability improved (SMD = 1.39; 95% CI (0.19, 2.59); p=0.02), and the JOA score increased (SMD = −1.34; 95% CI (−1.88, −0.80); p < 0.00001). It was also reported that the patient's quality of life improved after treatment for a period of 4 weeks (SMD = −0.29; 95% CI (−0.42, −0.16); p < 0.0001) and after a follow-up of 1 month (SMD = −0.20; 95% CI (−0.34, −0.07); p=0.003). The evidence level of the results was determined to be very low to low. Conclusions Based on the existing evidence, it can be concluded that TFM may have a better effect than other treatments on LBP. However, it is not yet possible to assess the safety level of TFM therapy. Due to the universal low quality of the eligible trials and low evidence level, rigorously designed large-scale RCTs must be conducted in order to further confirm the results in this review.


Introduction
Low back pain (LBP) is a symptom, not a disease, which results from several different known or unknown abnormalities or diseases [1]. LBP is the most common musculoskeletal health problem with the highest prevalence in the adult population [2]. e estimated lifetime prevalence of LBP is up to 80%, meaning many adults will experience an episode of LBP at least once [3]. According to the Global Burden of Disease Study 2017, LBP was classified as the leading cause of years lived with disability (YLDs) globally. Specifically speaking, the global YLDs for LBP were 42.5 million in 1990 and increased 52.7% to 64.9 million in 2017 [1,4,5]. Notably, annual healthcare costs attributed to LBP in the United States are estimated to be $100 billion. Particularly, two-thirds of which were indirect costs of lost wages and productivity, which imposes an economic burden on the healthcare system [6,7]. Disability and costs attributed to LBP are projected to increase in the coming decades, and this is particularly true in low-income and middle-income countries [1].
Clinicians and researchers have used conventional drugs and surgery to treat LBP for many years, but a large proportion of patients still continue to suffer from LBP [8,9]. e most commonly used method to relieve pain syndrome of LBP is a nonsteroidal anti-inflammatory drug (NSAID), but it should be noted that their long-term use may increase gastrointestinal and renal risks [6,10]. At best, surgery has a minimal impact on LBP, which is also more costly and carries a greater risk of adverse effects than nonsurgical management [11]. roughout the past three decades, changes have been made to critical recommendations in national clinical practice guidelines in Denmark, the United States, and the UK [6,12,13]. Greater emphasis is now placed on selfmanagement, physical and psychological therapies, and some forms of complementary medicine, and less emphasis has been placed on pharmacological and surgical treatments.
According to the guidelines of the American College of Physicians, nonpharmacological treatment of superficial heat is recommended for patients who suffer from low back pain (moderate-quality evidence) [6]. As one of the most complementary therapies for LBP, moxibustion is a traditional Chinese medicine (TCM) therapy that has a history of thousands of years in China. Specifically speaking, moxibustion refers to igniting moxa velvet or sticks and then burning or fumigating them on corresponding acupuncture points to prevent and treat diseases by means of heat or medicine [14]. As a new type of moxibustion therapy, thunder-fire moxibustion (TFM) was ameliorated by Prof. Zhao Shibi based on her decades of medical practice experience [15]. Notably, it is widely applied in China to treat diseases such as eye diseases, otolaryngological diseases, osteoarthropathy, gynecological diseases, and pain caused by any other disease [16][17][18][19][20]. Notably, TFM was listed as a critical new technology promotion project by the State Administration of Traditional Chinese Medicine in 2010 [21]. Compared with conventional moxa sticks, TFM has a larger diameter of not only moxa but also agarwood, frankincense, woody, dried ginger, and other TCM [22]. e temperature of TFM can reach up to 240°C when burning, and its average temperature is 142°C higher than that of ordinary moxibustion. Moreover, its warm stimulation involves the epidermis and affects the subcutaneous and muscle layers. When TFM burns, near-infrared rays can also penetrate the deep tissues of the human body, and the penetration depth is more than 10 mm, while conventional moxibustion is about 10 mm [23,24]. Additionally, TFM can be combined with various manipulations to improve the curative effect, such as pecking and rotating, and arrays can also be used.
Although several clinical trials have been conducted on TFM for treating LBP, based on our understanding, no systematic review and meta-analysis of TFM or TFM combined with other treatments for treating LBP have been reported. Consequently, the aim of this study focused on evaluating the quality of these randomized controlled trials (RCTs) to assess the effect and safety of TFM in treating LBP and better guide clinicians.

Methods
is meta-analysis was performed according to the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guidelines [25].

Data Sources.
A systematic literature search was conducted in the following databases from their inception to the period of April 18, 2021: PubMed, Web of Science, the Cochrane Library, Embase, EBSCO, China National Knowledge Infrastructure (CNKI), Wanfang Data, Chinese Science and Technology Periodical Database (VIP), and Chinese Biology Medicine (CBM) disc. e search strategies for PubMed, Embase Cochrane Library, Web of Science, and EBSCO are presented in the Appendix. Other databases were also searched using these terms, but they were slightly modified. Two researchers searched independently and imported the identified literature into EndNote software to delete the duplication and select potential articles by reviewing the titles and abstracts. e full texts of the chosen articles were reviewed according to inclusion and exclusion criteria.

Types of Participants.
Patients with LBP regardless of gender, age, ethnicity, education, and economic status who meet the diagnostic criteria were included in the study [26,27].

Types of Interventions.
e experimental group adopts a single TFM or TFM combined with other therapies. e control group receives other therapies besides TFM, such as usual care, acupuncture, moxibustion, medication, or physical therapy.

Types of Outcome Measures.
e outcomes included are pain intensity (including Visual Analogue Scale (VAS) [28]) and disability (on Roland-Morris Disability Questionnaire (RMDQ) [29] and Oswestry Disability Index (ODI) [30]). It should also be noted that other outcomes in this review were the Japanese Orthopedic Association (JOA) score [31] and quality of life (36-item short-form health survey (SF-36) [32]). Two outcome measures were considered to be positive indicators, such as the JOA score and SF-36, while all others were negative indicators. Among the positive indicators, the higher the score, the better the effect of the intervention. On the contrary, among the negative indicators, the lower the score, the better the effect of the intervention.

Data Extraction.
A data collection form was created to record selected studies such as the first author, published year, sample size, age, course of the disease, intervention regimens, treatment duration, follow-up duration, and outcomes before extracting the valuable information. Two researchers (Yao and Chen) independently completed the data extraction and the extracted information was reviewed once again upon completion. e divergence of opinion was 2 Evidence-Based Complementary and Alternative Medicine resolved by consulting the senior reviewer (Sun). If related data were deficient, one researcher (Yao) contacted the writers of the articles for lost information either through telephone or e-mail.

Assessment for Risk of Bias.
Two independent reviewers assessed the risk of bias following the Cochrane Handbook for Systematic Reviews of Interventions [33], including the following items: (1) random sequence generation (selection bias), (2) allocation concealment (selection bias), (3) blinding of participants and personnel (performance bias), (4) blinding of outcome assessment (detection bias), (5) incomplete outcome data (attrition bias), (6) selective reporting (reporting bias), and (7) other bias. e evaluation on these items was rated as "low," "high," or "unclear." Meanwhile, an Egger's test could be applied to appraise the extent of publication bias. Divergences were resolved by discussion. If the two investigators were unable to reach an agreement, the third and fourth reviewers (Sun and Du) were consulted for a final decision.

Data Synthesis and Analysis.
e meta-analysis was implemented by using RevMan 5.3 (available from the website: https://community.cochrane.org/tools/reviewproduction-tools/revman-5). Change values evaluated efficacy from baseline to endpoint data on each outcome in this meta-analysis [34]. In terms of parallel trials, net changes in measurements (change scores) for the trials were calculated by subtracting the postintervention data from the baseline value. For crossover studies, it was recommended that paired t-test data were extracted, which separately evaluated the value of "measurement on intervention" minus "measurement on control" for each participant. However, because this type of data was rarely provided, we resorted to using mean and SD [35]. If SDs were not reported directly, it was calculated from SEM or 95% CI using the following formulas: .92, where n represents the number of subjects. Change-from-baseline SD was estimated using the equation: rough a conservative estimate, a minimum correlation coefficient of 0.5 was used [36]. Notably, χ 2 test and I 2 test were used to measure the heterogeneity among studies. A fixed-effect model was adopted if I 2 < 50% and P> 0.1; otherwise, a random-effects model was employed. Dichotomous outcomes were reported as risk ratio (RR) and continuous data as weighted mean difference (WMD) and standard mean difference (SMD). Additionally, we conducted metaregression and subgroup analysis to explore the source of heterogeneity [37]. Sensitivity analysis was performed to evaluate the stability of analysis using different effects models and examining the effects of individual factors on the overall combined effect size. e potential publication bias was tested by employing an inverted funnel chart developed by Egger (Egger's test) when the number of eligible RCTs was more than 10 [38]. e sensitivity analysis and the Egger's test were carried out by STATA 12.0 software (Stata Corp, College Station, TX, USA).

Level of Evidence.
e level of evidence was evaluated with the help of the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) [39]. e level of evidence from low to high was classified into four grades: very low, low, moderate, and high. Particularly, RCTs started with a high level of evidence. en, the level of evidence was lowered gradually from the five aspects, including risk of bias, indirectness, inconsistency, imprecision, and publication bias. On the contrary, the level of evidence was gradually derived from three factors, which were doseresponse gradient, large effect, and plausible confounding.

Study Selection.
A total of 298 potential studies were identified through initial database searching. One hundred and fifty-nine articles were deleted due to duplication. After reviewing the titles and abstracts, 95 studies were excluded because of ineligible patient populations (n � 42), ineligible intervention (n � 35), and duplicates (n � 18). en, the eligibility of the remaining 44 studies was evaluated by reviewing the full text. Particularly, 23 studies were excluded due to non-RCT (n � 12), inappropriate grouping method (n � 2), and the absence of data (n � 1). Finally, a total of 21 RCTs  satisfied the inclusion criteria and were included in the systematic review. e selection process and reasons for exclusion are shown in Figure 1.

Study Characteristics.
All included trials were conducted in China and published from the period of 2011 to 2021. One article [53] was published in English, and twenty were published in Chinese. One RCT adopted a 3-arm parallelgroup design [45], and 20 trials used a 2-arm parallel-group design. Sample sizes varied from 53 to 420 participants, and a total of 2198 patients were included. Eighteen RCTs [41-45, 48-57, 59, 60] used VAS to assess pain intensity. Meanwhile, three RCTs [46,48,50] selected ODI, and one RCT [47] used both ODI and RMDQ to assess disability. Notably, two RCTs [41,53] adopted SF-36 to assess quality of life, and seven RCTs [40,43,48,49,55,57,58] reported a JOA score. Table 1 lists the details and characteristics of the included RCTs.

Risk of Bias.
Based on Cochrane criteria, the risk of bias assessment is shown in Figures 2 and 3. Fourteen [40,41,46,[48][49][50][51][52][53][54][55][58][59][60] of all 21 studies used a random table for randomization, and the remaining seven trials [42-45, 47, 56, 57] did not provide the methods of sequence generation. Only four trials [50,53,58,60] reported using sequential numbering and opaque sealed envelopes to conduct allocation concealment and the remainder did not provide concealment methods. Although both groups in a study [60] used moxibustion boxes to compare the effects of thunder-fire moxibustion and pure moxibustion and Evidence-Based Complementary and Alternative Medicine avoided revealing relevant grouping and treatment information to the subjects, the author was not able to ensure the reliability of the blinding method. Consequently, this study is judged to be unclear, and the rest are considered high risk. Four trials [46,52,58,60] reported employing the blindness of the assessor. Two studies [44,56] only stated that the baseline was not statistically significant but failed to present specific data.

Metaregression.
A pooled analysis of improvement was conducted in the pain intensity with TFM treatment using the meta-analysis method. Severe heterogeneity was detected among studies (I 2 � 75%, c 2 � 67.47, df � 17, p < 0.00001), which demonstrates that it was necessary to conduct the metaregression. Particularly, the metaregression was employed to identify the heterogeneity factor from the possible factors (such as treatment duration, moxibustion method, combined use, and sample size) that may cause heterogeneity.
e regression results illustrated that the moxibustion method was the source of heterogeneity p � 0.032(p � 0.032). erefore, a subgroup analysis was employed based on the moxibustion method (array or manipulation).

Pain Intensity.
e forest plot illustrating the results of the meta-analysis for pain intensity is shown in Figure 4. e pain intensity was reported in eighteen studies [41-45, 48-57, 59, 60] with 993 participants in the experimental groups and 989 in the control groups to evaluate the curative effect of TFM. All of these eighteen studies applied the Visual Analogue Scale as the outcome measurements. Despite the use of manipulation or array, the result indicated that TFM was able to significantly reduce pain compared with the control group on LBP (SMD � 0.94, 95% CI (0.74, 1.14), p < 0.00001p < 0.00001). e subgroup differences test indicated no potential differences between the manipulation group and the array group.

Disability.
e forest plot illustrating the results of the meta-analysis for the disability is shown in Figure 5. Four studies measured the level of disability [46][47][48]50], they all utilized an Oswestry Disability Index, and one simultaneously used a Roland-Morris Disability Questionnaire [47]. erefore, the latest data were not used. In total, the

Quality of Life.
e forest plot illustrating the results of the meta-analysis for the quality of life is shown in Figures 7 and 8. ere were two RCTs [41,53]     Heterogeneity: tau 2 = 1.42; chi 2 = 56.00, df = 3 (P < 0.00001); I 2 = 95% Test for overall effect: Z = 2.27 (P = 0.02) Test for subgroup differences: chi 2 = 2.19, df = 1 (P = 0.14), I 2    Evidence-Based Complementary and Alternative Medicine local skin, which was relieved after approximately two days. Besides such symptoms, there were no other uncomfortable reactions in the two groups. It should be noted that the adverse events of the two groups were tolerable and did not require specific interventions.

Sensitivity Analysis.
Sensitivity analysis was carried out as a means to evaluate the stability of meta-analysis by using STATA 12.0 software, such as pain intensity (Figure 9). After the sequential exclusion of individual studies one by one, the WMDs were recalculated to identify any significant change in our results. Sensitivity analysis showed that the exclusion of any single study was unlikely to overturn our findings.

Publication Bias.
Based on the pain intensity of the STATA 12.0 software, publication bias was analysed through Egger's test, which is shown in Figure 10.
e results demonstrated a p value of 0.504. is is more significant than 0.05 and reflected no publication bias (from a statistical significance perspective) for this present meta-analysis.

Level of Evidence.
e results of GRADE analysis revealed that the evidence quality of all outcome indicators was determined to be low or very low, which was not conducive to our result recommendation. As listed in Table 3, we lowered the levels mainly by the risk of bias, imprecision, and inconsistency.

Discussion
We intend to appraise the curative effect and the safety of TFM on LBP. About 21 RCTs were included for metaanalysis after searching and screening the major domestic and foreign databases by evidence-based medicine. e result revealed that TFM had favourable effects for LBP in comparison with TFM and other active treatments or TFM combined with other active treatments with active treatments alone. Notably, TFM can relieve pain and disability caused by LBP. It should also be noted that in terms of the JOA score, TFM had favourable effects for LBP in the comparison of TFM and other active treatments or TFM combined with other active treatments with active treatments alone. Particularly, TFM significantly improved the quality of life in the RP and BP dimensions compared to the control group after a 4-week treatment. Meanwhile, it only improved in the BP dimension relative to the control group after a follow-up of 1 month. We recommend the effect of TFM in LBP because of the low-to-very low level of evidence. e curative effect of TFM is closely related to moxibustion methods, and there are various methods that are suitable for different diseases. e moxibustion methods of TFM include manipulations (such as bird pecking, circling, and spiral moxibustion) and array method. Specifically  Evidence-Based Complementary and Alternative Medicine speaking, the array method refers to the use of single, double, or multihole moxibustion boxes. Based on the condition of different patients, two or more moxibustion boxes are placed on the patients in horizontal array, vertical array, oblique array, T-shaped array, etc. Notably, the majority of articles included in this study used the array method. rough the strong thermal stimulation of the moxibustion stick burning, the array method gathers the heat and expands the heated area, which increases homogeneity to a certain extent. Impressively, our regression analysis also verified this result. Here, we used the treatment duration, the moxibustion method, combined use, and sample size as possible Heterogeneity: chi 2 = 0.01, df = 1 (P = 0.92); I 2 = 0% Test for overall effect: Z = 0.64 (P = 0.52)

Mental Health (MH)
Pain intensity, disability, and JOA score of LBP were statistically significant with substantial heterogeneity. As heterogeneity across studies is expected in meta-analyses [61], it is not surprising that there was considerable heterogeneity in the effect of TFM on the LBP. Although a subgroup analysis was performed based on the regression results, heterogeneity still existed in these comparisons. e variety of acupoint selection schemes, treatment frequencies, and courses may have caused unresolved heterogeneity. Specifically speaking, the frequency is usually once a day, but it also includes every other day and once every three days.

Evidence-Based Complementary and Alternative Medicine
Intervention time also varied from 15 min to 60 min. It should be noted that these conditions may be related to the cause and duration of LBP. TFM has unique thermal and infrared effects during burning so it may produce various adverse effects, such as burn wounds, blister, and pruritus [62]. Seven of the 21 studies mentioned adverse events, and only 2 cases experienced local skin redness and miliary rash, which was related to TFM. Moreover, neither of these two patients requires particular medical intervention. Nevertheless, the safety of TFM cannot be definitively concluded due to a relative lack of studies providing details of the adverse events. However, the issue of whether moxibustion-induced burns are actually considered an adverse event still remains controversial [63]. Traditional Chinese moxibustion is also known as scarring moxibustion. It has long been taken for granted that it causes minor burns, scarring, and purulence during treatment, as various ingredients enter the body through burn-damaged skin [64].
Due to the following limitations, we were unable to reach an exact conclusion regarding the effect of TFM.
is is especially attributed to the fact that the methodological quality of inclusive studies was low and that there was no multicenter study, and the outcome indicators were subjective. Additionally, the sample size of most studies was small and an inappropriate random method was used. Moreover, there was allocation concealment and a lack of blinding of most studies, which exaggerated the results of the outcome measures. In this study, the correct reporting of allocation concealment and blinding of outcome measurers were both 19.05% of the literature. e blinding of participants and subjects was not successfully performed due to the particularity of the TFM treatment, which could lead to overestimation. e potential mechanism of TFM for LBP is not yet distinct, but it does have a positive therapeutic effect. Compared to thermal therapy, TFM is based on the TCM meridian theory. Specifically, it uses the heat, thermal infrared radiation, and physicochemical factors produced by drug combustion through meridian and acupoints feeling in achieving WenTong meridian and adjusting human body's energy to treat disease [53]. WenTong meridian means promoting the dredging function of meridians by warming. Chen [65] reported that TFM had an anti-inflammatory effect on model rats with knee osteoarthritis. Its therapeutic mechanism may be related to reducing the contents of TNFα and IL-1β in the serum of model rats. Notably, some studies even demonstrated that TNF-a and IL-1β seemed to play a significant role in patients suffering from LBP [66,67]. However, these theories have not yet been fully established. Consequently, there is still a great distance to go before the mechanism involved with TFM is fully understood.
In TCM theory, the most commonly used acupoints for LBP were located in the bladder, gallbladder meridian, and the governor vessel-all of which pass through the waist. In our statistical results of acupoints, the vast majority of acupoints were located on these three meridians. According to textbooks and clinical practice, the acupoints of BL23, BL25, BL40, GV3, GV4, and GB30 were globally used to treat nonspecific and chronic LBP, as reported by Yuan et al. [68,69]. In addition, Yuan reported that Ashi acupoints are usually reported from all sources. e above statements are consistent with our research results.
is illustrated that when the interveners used TCM therapy to treat LBP, such as acupuncture and moxibustion, they followed the TCM theories in selecting acupoints.
is review presented several limitations. First of all, we collected a significant amount of literature through a comprehensive search strategy of nine different databases, Huang (2018) Liu (2017) Mao (2019) Sun (2019) Tian (  ⊕⊕○○ Low E: experimental group; C: control group; CI: confidence interval; RCT: randomized controlled trial; SMD: standard mean difference; MD: mean difference. (1) Blind method is missing, allocation hidden report is insufficient, and random method description is not clear; (2) statistical heterogeneity and clinical heterogeneity were more significant; (3) the total sample size was small; (4) blind method is missing and allocation hidden report is insufficient. ⊕⊕xx represents the low level of evidence. ⊕xxx represents the very low level of evidence.
Evidence-Based Complementary and Alternative Medicine 13 without any language restrictions. However, only articles published in Chinese and English were retrieved and all the studies were conducted in China. is may be due to the facts that thunder-fire moxibustion belongs to a category of TCM and that less foreign studies were found in this area. Second, given that the methodological quality of most qualified trials was low, it may lead to serious selection performance and detection bias.
To some extent, this weakened the authenticity and reliability of the evidence for TFM treatment of LBP in this study. ird, although some sources of heterogeneity were identified through regression and subgroup analysis, significant heterogeneity still existed among studies. Finally, the course of TFM was short term (less than 12 weeks) among the included studies so it is unclear whether the long-term practice of TFM is beneficial for LBP patients.
While this systematic review and meta-analysis had some limitations, it nonetheless demonstrated some glaring advantages. Although an increasing number of studies reported TFM to successfully treat LBP patients ranging from case report studies to cohort studies to RCTs, there was no systematic review. is is especially in those that primarily referred to its effectiveness in treating LBP. Hence, this meta-analysis was designed to evaluate the efficacy of TFM for LBP. In addition, we conducted this systematic review and meta-analysis in strict accordance with the PRISMA guidelines, and the content met the criterion. erefore, we speculated that the results of this review could provide evidence on the efficiency and safety of TFM in treating LBP, which would benefit both patients as well as practitioners.

Conclusion
is review provided a comprehensive assessment of the quality of the methodology and the level of evidence. Existing evidence indicates that TFM is able to effectively treat LBP. However, the findings should be cautiously interpreted because of universally low-quality eligible trials and low evidence level. e safety of TFM cannot be definitively concluded due to a relative lack of studies that provide details of its adverse effects. In the future, more welldesigned, rigorous, large sample, and multicenter prospective randomized controlled trials are needed on this subject to confirm the validity of the results.