Efficacy and Safety of Curcumin Supplement on Improvement of Insulin Resistance in People with Type 2 Diabetes Mellitus: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Background Diabetes is a major public health concern. In addition, there is some evidence to support curcumin as part of a diabetes treatment program. Methods Data from randomized controlled trials were obtained to assess the effects of curcumin versus placebo or western medicine in patients with type 2 diabetes mellitus (T2DM). The study's registration number is CRD42018089528. The primary outcomes included homeostasis model assessment-insulin resistance (HOMA-IR), glycosylated hemoglobin (HbAlc), total cholesterol (TC), and triglyceride (TG). Results Four trials involving 453 patients were included. The HOMA-IR of curcumin group is lower in Asia (WMD: −2.41, 95% CI: −4.44 to −0.39, P=0.02) and the Middle East subgroups (WMD: −0.60, 95% CI: −0.74 to −0.46, P < 0.00001). The HbAlc in the curcumin group is lower than that in the control group (WMD: −0.69; 95% CI: −0.91, −0.48; P < 0.0001). The TC and TG levels of the curcumin group are lower in the Asia subgroup (TC: WMD: −23.45, 95% CI: −40.04 to −6.84, P=0.006; TG: WMD: −54.14, 95% CI: −95.71 to −12.57, P=0.01), while in the Middle East the difference was of not statistically significant (TC: WMD: 22.91, 95% CI: −16.94 to 62.75, P=0.26; TG: WMD: −4.56, 95% CI: −19.28 to 10.16, P=0.54). Conclusion Based on the current evidence, curcumin may assist in improving the insulin resistance, glycemic control, and decreased TG and TC in patients with T2DM.


Introduction
As a serious metabolic disease, diabetes affects about 5% of the world's people. Epidemiological data show that the number of people with diabetes is expected to increase dramatically to 592 million by 2035 [1]. 12% of global health expenditure is spent annually on diabetes and its complications [2]. Diabetes is divided into different types, wherein type 1 and type 2 diabetes accounted for more than 90% of all cases. Among these types of diabetes, type 2 diabetes mellitus (T2DM) causes metabolic abnormalities and serious complications that have a profound impact on the patient's lifespan and quality of life. T2DM is mainly characterized by insulin resistance and hyperglycemia [3]. Its common complications include microvascular disease (diabetic retinopathy, diabetic nephropathy, and diabetic neuropathy), macrovascular disease (diabetic heart disease, diabetic cerebrovascular disease, and peripheral vascular disease) [4,5], and increased risk of cancer [6,7]. At present, T2DM has a variety of therapeutic drugs, such as human insulin preparation, alpha glucosidase inhibitor, dipeptidyl peptidase-4 inhibitor, incretin analogue, biguanide, insulin secretagogue, insulin sensitizer, and intestinal lipase inhibitor [8,9]. However, the currently used therapies have many side effects such as hypoglycemia, gastrointestinal problems, and weight gain [8]. erefore, new drugs and natural compounds are constantly being tested to prevent and treat diabetes better [10].
Curcumin is a chemical component extracted from the rhizome of some plants. It has a series of effects such as blood lipid lowering, antitumor, anti-inflammatory, and antioxidation [11,12] and has been used as a food flavoring agent, preservative, and ancillary medication for some diseases (such as heart disease and tumors) [13,14]. In the treatment of diabetes, there is also evidence to support curcumin as a part of the diabetes treatment program [14,15]. At present, many randomized controlled trials (RCTs) on the treatment of T2DM with curcumin have been published [11,[16][17][18][19][20][21][22], but there is still no systematic review and meta-analysis to assess the effects and safety of curcumin. erefore, we decide to perform a systematic review and meta-analysis for the first time to evaluate the clinical effects of curcumin on T2DM.

Protocol.
Study selection, assessment of eligibility criteria, data extraction, and statistical analyses were performed based on a predefined protocol registered on PROSPERO CRD42018089528 (see supplementary materials) [23].

Search Strategy and Selection Criteria.
We searched the English database and the Chinese database from the beginning of their establishment to September 3, 2020. e English database includes EMBASE, Medline Complete, the Cochrane Library (until Issue 9, 2020), ClinicalTrials, PubMed, and Web of Science. e Chinese database includes the Chinese Science and Technology Periodical Database (VIP), Chinese Biomedical Database (CBM), Wan Fang Database (Chinese Ministry of Science and Technology), and China National Knowledge Infrastructure Databases (CNKI). e search strategy for PubMed is presented in Table 1 as an example.
Studies meeting the inclusion criteria would be included in this review: (1) participants: patients with type 2 diabetes mellitus; (2) intervention: curcumin with no limits on the type, dose, frequency, and so on; (3) comparisons: Western medicine, blanks, or placebo; (4) outcomes: primary outcomes: homeostasis model assessment-insulin resistance (HOMA-IR), glycosylated hemoglobin (HbAlc), total cholesterol (TC), and triglyceride (TG); secondary outcomes: body mass index (BMI), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), fasting glucose, and fasting insulin; (5) study type: randomized controlled trials (RCTs) with no limits on the manner by which randomization has been achieved on blinding or on the language of publication. Studies meeting the exclusion criteria would be excluded: (1) not T2DM patients; (2) the participant is not human; (3) nonoriginal research literature; and (4) non-RCT.

Data Extraction.
Two reviewers independently extracted the data and it was checked by a third reviewer. When there is a disagreement, a consensus is reached through mutual discussion and negotiation with all reviewers. e extracted data include basic information (author, publication time, age of the research object, etc.), sample size, intervention measures, dose, intervention time, outcomes, etc. [24].

Study Quality Assessment.
e risk of bias of RCTs was assessed by using the risk of bias assessment tool based on the Cochrane Handbook [25]. Two reviewers independently assessed the risk of bias. When there is a disagreement, a consensus is reached through mutual discussion and negotiation with all reviewers. e risk of bias is divided into three levels: high risk, low risk, and unclear. e content of the evaluation includes random sequence generation, allocation concealment, blinding, incomplete outcomes, selective reporting, and other bias.

Statistical Analysis.
e data were analyzed by RevMan 5.3 software. Cochrane's Q and I 2 test were used to judge the heterogeneity of different studies. If there is good homology between studies (I 2 < 50%, P > 0.1), the fixed-effects model is used for meta-analysis. If there is heterogeneity between studies (I2 > 50%, P < 0.1), we first find the source of heterogeneity, conduct subgroup analysis, and then choose random-effects model or give up meta-analysis [26]. e dichotomous variable measure was summarized by risk ratio (RR) with a 95% confidence interval (CI). e continuous outcomes underwent meta-analysis using weight mean differences (WMD) and 95% CI. If the units of outcomes are different, or the value difference between RCTs is more than 10 times, the standard mean differences (SMD) and 95% CI are used according to the situation.

Sensitivity Analysis and Publication Bias Detection.
STATA 15.0 was utilized for sensitivity analysis and publication bias detection. Studies with RCTs ≥5 were evaluated for publication bias. e outcomes with P > 0.1 were thought to have publication bias. e outcomes that meet the following conditions are all subjected to sensitivity analysis: (1) random-effects model is used; (2) number of included RCTs ≥ 3; and (3) the results of the fixed-effects model are inconsistent with the results of the random-effects model (whether it is a subgroup result or a summary result).

Results of the Search.
e total records identified through database searching and other sources were 431. Sixteen records were included after initial identification. Four records were excluded: Yang et al. is not RCT [27]; and the participants in the remaining 3 records are not only T2DM patients [28][29][30]. Eventually, 12 records were included to undergo analysis (Figure 1).

Allocation Concealment.
Khajehdehi et al. [22] and ota et al. [33] did not describe an acceptable method of allocation concealment; therefore, they were rated as having an unclear risk of bias. Panahi et al. [16][17][18][19], Na et al. [20,21], Asadi et al. [31,32], and Adibian et al. [34] utilized the capsules in the same shape, size, and color to contain curcumin and placebo; Chuengsamarn et al. [11] used opaque and consecutively numbered envelopes. Hence, these five RCTs considered to have allocation concealment were rated as having low risks of bias.

Blinding, Incomplete Outcome Data, and Selective
Reporting. All RCTs claimed to use blinding, but only Asadi et al. [31,32] described the implementation process for researchers and participants, so only its double blinding was rated as a low risk of bias. Four RCTs [11,[16][17][18][19][20][21]33] did not describe the implementation process for both researchers and participants, and they were rated as high risk of bias. Khajehdehi et al. [22] described blinding to researchers, so its blinding of outcome assessment (detection bias) was rated as low risk of bias. Adibian et al. [34] described blinding of participants, so the blinding of participants and personnel (performance bias) was rated as low risk of bias.

Incomplete Outcome Data and Selective Reporting.
e incomplete outcome data of all RCTs are rated as low risk of bias because the number of missing persons and the reasons for the missing between groups is balanced. One RCT (Khajehdehi et al. [22]) failed to provide all outcomes mentioned in its protocols; thus, we thought its risk of bias was high. e others 6 RCTs reported study's prespecified outcomes that are of interest in the review; and their risks of bias were low.

Other Potential Bias.
Panahi et al. [16][17][18][19] claimed that one of the authors had a conflict of interest and was therefore assessed as a high risk of bias. Other sources of bias in the other 6 RCTs were not found; therefore, the risks of other bias of the RCTs were low.  (Figure 3).

Body Mass Index.
ree RCTs [11, 16-19, 31, 32] reported BMI. Due to the high heterogeneity (I 2 � 89%, P � 0.0002), the statistical analysis was abandoned according to the Cochrane Handbook for Systematic Reviews of Interventions. Panahi et al. [16][17][18][19] found that there was a statistically significant difference in BMI changes between the curcumin group and the control group (−0.49 ± 0.52 versus 0.24 ± 0.73; P < 0.001). Chuengsamarn et al. [11] also found that there was a statistically significant difference in BMI changes between the curcumin group and the control group (−1.97 ± 5.38 versus 0.16 ± 4.32; P < 0.001). However, Asadi et al. [31,32] found that there were no significant differences in terms of BMI and weight between the two groups. [11,[16][17][18][19][20][21][22][31][32][33][34] reported fasting blood glucose. However, the data in Adibian et al. [34] cannot be extracted; hence, the data in 6 RCTs include 282 participants in the curcumin group and 282 participants in the control group. Due to the high heterogeneity, 6 RCTs were subdivided into two subgroups according to the region of the patients. After subdivision, the heterogeneity was low in the Asia Pacific subgroup but high in the Middle East subgroup (Asia:   8 Evidence-Based Complementary and Alternative Medicine I 2 � 0%, P � 0.79; the Middle East: I 2 � 74%, P � 0.02). e random-effects model was used. e fasting blood glucose in the curcumin group was lower than that in the control group Asia subgroup (SMD: −0.57, 95% CI: −0.79 to −0.36, P < 0.00001). However, the difference between the curcumin group and the control group in the Middle East subgroup was of no statistical significance (SMD: 0.04, 95% CI: −0.50 to 0.58, P � 0.89). e summary results also showed that the difference between two groups was of no statistical significance (SMD: −0.28, 95% CI: −0.62 to 0.06, P � 0.11) (Figure 7). Two RCTs [11,[16][17][18][19] reported fasting insulin. Due to the high heterogeneity (I 2 � 96%, P < 0.00001), the statistical analysis was abandoned according to the Cochrane Handbook for Systematic Reviews of Interventions. Panahi et al. [16][17][18][19] found no significant difference in fasting insulin levels between the two groups (P > 0.05), while Chuengsamarn et al. [11] found that the curcumin group had lower fasting insulin levels (P < 0.05). [11,[16][17][18][19][20][21][22]34] including 248 participants in the curcumin group and 249 participants in the control group reported LDL-C and HDL-C. Due to the high heterogeneity, 5RCTs were subdivided into two subgroups according to the region of the patients. After subdivision, the heterogeneity was low in each subgroup (Asia: I 2 � 6%, P � 0.30; the Middle East: I 2 � 29%, P � 0.24) among the RCTs.

Blood Lipid. Five RCTs
e fixed-effects model was used. e LDL-C in the curcumin group was lower than that in the control group in the Asia subgroup (WMD: −20.85, 95% CI: −28.78 to −12.92, P < 0.00001), but it was higher in the Middle East subgroup (WMD: 15.67, 95% CI: 5.91 to 25.43, P � 0.002) (Figure 8).
e RCTs in HDL-C were also subdivided into two subgroups according to the region of the patients for the same sake. After subdivision, the heterogeneity was still high in the two subgroups (Asia Pacific: I 2 � 91%, P � 0.001; the Middle East: I 2 � 76%, P � 0.02) among the RCTs. e random-effects model was used. e difference in HDL-C between two groups was of no statistical significance in each subgroup (the Middle East: WMD: −0.30, 95% CI: −3.78 to 3.19, P � 0.87; Asia: WMD: 6.01, 95% CI: −2.58 to 14.60, P � 0.17). e summary result also showed that the difference between two groups was of no statistical significance (WMD: 2.26, 95% CI: −2.03 to 6.55, P � 0.30) (Figure 9).  Evidence-Based Complementary and Alternative Medicine

Sensitivity Analysis
Results. Sensitivity analysis was performed for 5 outcomes: HOMA-IR, TC, TG, FBG, and HDL-C. (1) In the outcome "HOMA-IR," after we omitted Na et al. [20,21], we found that the estimate of the result moved out of the lower limit of 95% CI (Figure 11(a)). (2) In the outcome "TC," after we omitted Panahi et al. [16][17][18][19], we found that the estimate of the result moved out of the lower limit of 95% CI, while after omitting Chuengsamarn et al. [11], the estimate of the result moved out of the upper limit of 95% CI (Figure 11(b)). (3) In the outcome "TG," after omitting Chuengsamarn et al. [11], the estimate of the result moved out of the upper limit of 95% CI (Figure 11(c)). (4) In the outcome "FBG," no matter which study was removed, the results were not significantly changed, suggesting that the heterogeneity may not come from RCT (Figure 11(d)). (5) In the outcome "HDL-C," after omitting Chuengsamarn et al. [11], the estimate of the result moved out of the lower limit of 95% CI (Figure 11(e)). e abovementioned RCTs may be the source of heterogeneity of corresponding outcomes.

Main Findings.
e HOMA-IR of the curcumin group is lower in Asia and the Middle East subgroups. e HbAlc in the curcumin group is lower than the control group. e TC, TG, and fasting blood glucose level of the curcumin group is lower in the Asia subgroup, while in the Middle East, the difference was of no statistical significance. For BMI, fasting insulin, and HDL-C level, there is no strong evidence that which one is better. Interestingly, although the LDL-C level of the curcumin group in the Asia subgroup is lower than that of the control group, the LDL-C level of the curcumin group was higher than the control group in the Middle east subgroup.

Overall Completeness and Applicability of Evidence.
Most of RCTs come from Asia Pacific (especially Southeast Asia, like China and ailand) and the Middle East (mainly Iran). Due to the lack of RCTs from all over the world, the applicability of the findings is limited.

Discussion of the Source of Heterogeneity.
Most outcomes have heterogeneity, so this study conducted a sensitivity analysis to find the source of heterogeneity. Sensitivity analyses were performed for 5 outcomes: HOMA-IR, TC, TG, FBG, and HDL-C.
For the outcome "TC," "TG," and "HDL-C," Chuengsamarn et al. [11] prepared curcumin monomer, while the preparations of several other RCTs are curcuminoids or turmeric, which suggests that different drug preparation methods may be the source of heterogeneity. For the outcome "FBG," no matter which study was removed, the results were not significantly changed, suggesting that the heterogeneity may not come from RCT.
In addition to the above possible sources of heterogeneity, heterogeneity may also come from ethnic differences, regional differences, gender differences, body size differences, and so on. More relevant RCTs are needed to be conducted in more diverse subgroups to reduce the heterogeneity of the research and stabilize the conclusions.

Novelty of is Research.
is systematic review and meta-analysis showed that curcumin can improve HOMA-IR and HbAlc, especially in Asian patients and the Middle Eastern patients. In terms of blood lipids, curcumin can reduce TC, TG, and fasting blood glucose level in Asian patients, but does not improve in the Middle East patients significantly. Curcumin may lower the LDL-C    Evidence-Based Complementary and Alternative Medicine level of Asian patients. Interestingly, after curcumin intervenes in the Middle Eastern patients, their LDL-C level has increased. is is something that needs attention in the future. Demmers et al. [35] estimated the available scientific data on the effectiveness and safety of medicinal food plants in the treatment of impaired glucose tolerance. It included an RCT of curcumin extract intervention for diabetes and found that the fasting blood glucose at 2 hours after intervention showed statistical significance after 3, 6, and 9 months (P < 0.01). In addition, the curcumin extract intervention (HbA1c) value showed statistical significance after 3, 6, and 9 months (P < 0.01). HOMA-IR after curcumin extract intervention showed statistical significance after 6 months and 9 months (P < 0.05 and P < 0.01). It shows that curcumin has shown a reliable result that is effective in treating impaired glucose tolerance. Compared with Demmers et al.'s [35] research, our study is mainly devoted to exploring the interventional effects of curcumin and turmeric extracts on T2DM. We included more curcumin in the treatment of T2DM-related RCTs, and the evidence for improving the outcome of multiple related indicators of glucose tolerance is more sufficient. Compared to Marton et al.'s [36] review, (1) in this study, the RCT data (HOMA-IR, HbAlc, and so on) of curcumin treatment of T2DM were statistically analyzed (meta-analysis). (2) is study also conducted a subgroup analysis based on regions. (3) We adopted stricter screening criteria and merged the records belonging to the same RCT. (4) Sensitivity analysis was conducted in this study, which can more accurately locate the main source of heterogeneity. (5) e publication bias assessment was carried out in this study, and the publication bias of outcomes was excluded. In terms of improving metabolism, Roshanravan et al. [37] found the protective effect of supplementing Crocus sativus L. on hyperlipidemia and hyperglycemia through systematic reviews and meta-analysis studies. Our research also found that curcumin has the same clinical effect in improving the metabolism of diabetes.

e Strengths of is Review.
is registered systematic review and meta-analysis is the first one that comprehensively evaluated the previous RCT of curcumin on T2DM and underwent a subgroup analysis to assess the applicable  population. It also collected the detailed data from each RCT and a comprehensive assessment of risk of bias was conducted.

e Limitations of is Review.
RCTs from Asia and the Middle East account for a large proportion, which affects the applicability of the findings. e quantity and quality of RCTs are also not high in this review; several subgroups only include one RCT. Only 453 participants were included in this review, which may also impact the applicability of the findings. Meanwhile, half of RCTs [16][17][18][19]22] have an unclear risk of bias in randomization, and one of the RCTs [22] has a high risk of bias in selective reporting; this may also influence the interpretation of the results. In addition, most of the outcomes have a high heterogeneity (such as HOMA-IR, TC, and TG) that cannot be eliminated by appropriate subgroup analysis. is also affected the applicability of the findings. e heterogeneity may come from the potential discrepancies in the pharmacological effects of various curcumin preparations, which may result from different standardizations of curcumin manufacturing process, dosage, duration of treatment, units of laboratory tests, and races of the selected patients or other places.

Implications for Future.
is systematic review and meta-analysis found that curcumin may improve the HOMA-IR and fasting blood glucose of individuals in Asia   is may be related to ethnic differences, and more research is needed to amend or confirm it. In addition, in the aspect of decreasing BMI and fasting insulin, and improving HDL-C, curcumin may not have an advantage over the control group. However, due to the lack of evidence, more RCTs are needed. Last but not the least, for safety, there are no serious adverse events reported in RCTs, and the occurrence of adverse events in curcumin groups is the same as that of the placebo control group; therefore, it can be considered as a safety treatment based on current evidence.
In summary, curcumin may be more effective in Asia. In addition, for clinical practices, curcumin may be recommended as an adjunct to the treatment of T2DM patients to improve insulin resistance and glycemic control and reduce blood lipids. For Middle Eastern T2DM patients, the appropriate dosage, usage, and so on are yet to be confirmed. For future research, more RCTs about the adverse events and the T2DM-related outcomes are needed to revise or validate the findings in this review.
e RCTs containing the data of the patients of other    countries or regions around the world are also needed to expand the applicability of the results [38].

Conclusion
Based on the current evidence, curcumin may assist in improving insulin resistance, glycemic control, and decrease in TG and TC in patients with T2DM.

Data Availability
All data generated or analyzed during this study are included in this published article.

Disclosure
Tianqing Zhang and Qi He are co-first authors. Hengjing Hu is the corresponding author.

Supplementary Materials
"CRD42018089528" is the protocol of our systematic review and meta-analysis. e "PRISMA 2009 checklist" is a standardized checklist that make sure that systematic reviews and meta-analyses are implemented according to this standard. (Supplementary Materials)