Home 
Efficacy and safety of oral insulin versus placebo for patients with diabetes mellitus: A systematic review and metaanalysis Correspondence Address: OBJECTIVE: Compliance to insulin injections is poor due to difficulty in subcutaneous administration. Hence, there is a need of an oral formulation of insulin. Oral insulin is currently under investigation. The present analysis aimed to evaluate oral insulin versus placebo for patients with diabetes mellitus (type1 and type2). MATERIALS AND METHODS: Results from PUBMED and MEDLINE were searched and compiled from January 1, 2000 to January 9, 2020. Postprandial blood glucose excursions (2PPG), glycated hemoglobin (HbA1c), Cpeptide levels, and immune antibody (IAA) levels were compared between the arms. In addition, time to diabetes and safety of oral insulin were discussed. RESULTS: Thirteen out of 1778 trials were included to the analysis. Oral insulin was found to induce significant reduction in mean 2PPG excursion (standardized mean difference [SMD]: −1.94, 95% CI: −3.20 to −0.68, I^{2} = 91.81, P < 0.005) and mean IAA levels (SMD:0.49, 95% CI: −0.82 to 0.16, I^{2} = 27.12, P < 0.005) compared with placebo. Mean Cpeptide levels were notably lower in the oral insulin arm. However, the difference was not statistically significant. No significant difference was observed in mean HbA1c levels. The rate of development of type1 diabetes was not significantly influenced by oral insulin. No deaths or treatmentrelated serious adverse events were reported. CONCLUSION: Oral insulin provided significant benefits for acute maintenance of diabetes mellitus. It elicited lower immune response and was well tolerated. This new formulation has potential to augment the management of diabetes mellitus. More studies are required to assess its longterm effects.
Introduction Diabetes mellitus, a metabolic disease, is the outcome of defective secretion of insulin, defective insulin action, or both. It results in chronic hyperglycemia leading to damage and dysfunction of various organs of the circulatory system, excretory system, and nervous system. Type1 diabetes usually emanates from the destruction of the pancreatic βcells by an autoimmune process, which leads to absolute insulin deficiency (immunemediated diabetes). Alternatively, it may lack established etiologies (idiopathic diabetes). Type2 diabetes results from insulin resistance. It may improve with weight reduction and pharmacological interventions; however, it is rarely restored to normal.[1] Some of the parameters and tests used for evaluating diabetes include fasting glucose, postprandial plasma glucose (PPG), fasting insulin, glycated hemoglobin (HbA1c), glucose infusion rate, Cpeptide levels, and level of insulin antibodies. PPG refers to plasma glucose concentrations after meals. PPG excursion, defined as the change in glucose concentration from before to after a meal, is a significant parameter for measuring the overall metabolic control and risk for diabetic complications.[2] HbA1c is hemoglobin linked to a sugar. Longterm (~60 days) glycemic status can be measured using this parameter in patients with diabetes mellitus.[3] Cpeptide, a short amino acid chain produced as a byproduct of the insulin synthesis process in the human body, is released into the blood in equal amount of insulin. It helps to measure endogenous insulin secretion.[4] The presence of type1 diabetes is indicated by the autoantibodies targeted to insulin immune antibody (IAA). They are one of the markers of βcell destruction. Current therapy for diabetes includes insulin injections or oral hypoglycemic agents (OHAs) such as biguanides, glucagonlike peptide 1 analogues, thiazolidinediones, sulfonylureas, dipeptidyl peptidase4 inhibitors, sodium–glucose cotransporter2 inhibitors, or combination therapy. While patients frequently develop resistance to biguanides and sulfonylureas, thiazolidinediones are associated with cardiovascular events.[5],[6],[7] The latter three newer agents have reported lower rates of adverse events (AEs), better lowering in HbA1c levels, enhance insulin sensitivity and amelioration of glucotoxicity.[8],[9],[10] Insulin is the most recommended firstline treatment for type1 diabetes and secondline treatment (with OHAs) for type2 diabetes. As insulin is degraded by digestive processes when administered orally, it is administered subcutaneously to bypass firstpass metabolism. However, this administration route is the major limiting factor of insulin therapy. Compliance to injections is poor due to injection site pain and needle phobia.[11] Around a quarter of patients do not comply with insulin injection treatment.[12] Such noncompliance may negatively impact the outcome of the prescribed therapy, and increase the complications of diabetes due to missed insulin doses. To overcome this hurdle, formulations that can be consumed via oral route, and prevent insulin degradation in the stomach are developed. Such formulations are currently under investigation in clinical trials. As many patients miss insulin doses or selfterminate the therapy, it is essential to compare the outcomes of oral insulin with absence of exogenous insulin. In this study, the efficacy of oral is compared with placebo for treatment or prophylaxis of diabetes mellitus. AEs associated with oral insulin are also discussed. Materials and Methods Search strategy and eligibility criteria PubMed, MEDLINE, and Science Direct were searched from January 1, 2000 to January 9, 2020. Potential trials were identified using the search terms “insulin,” “oral” and “diabetes mellitus” individually, and by using the search string “([insulin] AND oral) AND diabetes mellitus” under all fields. All interventional trials that included placebo as the comparator, provided data on 2PPG, HbA1c, Cpeptide, IAA, or time to diabetes for adult patients with diabetes mellitus who received ≥1 dose of insulin via oral route of administration (buccal spray, capsule, and tablet) were included. Trials published in nonEnglish languages, observational trials, and pharmacokinetic trials were excluded. Data extraction The characteristics of each trial, patient demographics, journal, year of publication, type of diabetes, trial phase, interventions, type of formulation, dose of insulin, number of patients who completed the trial, patients assigned to each intervention, key outcomes, AEs, and results were extracted from each trial. When multiple doses of oral insulin were administered in a trial, the data for the outcome from minimal effective dose, and/or from combined doses, where available, was used for analysis. Quality assessment The risk of bias for each of the 13 trials was assessed using the Cochrane Risk of Bias Tool. The assessment was based on five domains (selection, performance, detection, attrition, and reporting). The parameters included random sequence generation (selection bias), protocol variation (performance bias), blinding (detection bias), missing data (attrition bias), and selective outcomes reporting (reporting bias). Details are provided in [Supplementary File 1]. Key outcomes The key outcomes analyzed were mean 2PPG excursions, mean HbA1c, mean Cpeptide levels, and mean levels IAA for the oral insulin arm versus placebo arm. In addition, time to diabetes and safety data from the trials was also discussed. Statistical analysis Standardized mean difference (SMD), an effect size statistic, has been used to represent the analytical findings of the outcome measures from the oral insulin and placebo arms for each trial. In addition, a correction factor was applied to remove the upward bias which could have resulted into overestimation of the effect size statistic in small samples (n <20). This corrected or unbiased effect size estimate, Hedges' g, was used in all analytical computations for metaanalysis to assess the consistency of the effect across trials. Randomeffect metaanalysis method has been applied to obtain the most precise estimate of overall mean. i.e., The summary effect from a distribution of effect sizes. To measure the heterogeneity among these effect sizes, I2 statistic was calculated for each summary effect. P values were computed to analyze the magnitude of the effect size estimates for each study and also for the summary effects.[13],[14],[15],[16] R program was used for analysing all the data from the selected trials, and to create forest plots for graphical interpretations. Detailed summary on statistical analysis is provided in [Supplementary File 2] [Supplementary Table 1],[Supplementary Table 2],[Supplementary Table 3],[Supplementary Table 4]. Reporting guidelines The data are presented as per Preferred Reporting Items for Systematic Review and MetaAnalyses guidelines. Results Search results [Figure 1] demonstrates the study selection process. A total of 1778 studies (without duplicates) were identified. Of these, 69 and 1689 studies were excluded based on the screening of abstract, and full text, respectively. Twenty publications consisted of clinical trials on oral insulin for diabetes mellitus. Thirteen of the 20 trials that compared oral insulin versus placebo with or without pretreatment with OHAs were selected for pooled analysis [Table 1].{Figure 1}{Table 1} Quality assessment Based on judgment from the five domains (selection, performance, detection, attrition, and reporting), the quality of six, four and three trials were found to be good, fair, and poor, respectively [Table 2]. As measured by the five parameters per domain, the risk of bias was low for four trials, moderate for two trials, and high for seven trials.{Table 2} Findings from metaanalysis Mean postprandial plasma glucose excursion Five clinical trials that reported 2PPG excursions were selected for comparing the effect of oral insulin versus placebo.[17],[18],[19],[20],[21] Four trials were on type2 and one trial was on type1 diabetes. Overall, 61 patients received the minimum effective dose of insulin or the dose used for analysis.[20],[21] In comparison with the placebo arm, results demonstrated a significant reduction in the mean 2PPG excursion in the oral insulin arm (SMD: −1.94, 95% CI: −3.20 to −0.68, I2 = 91.81, P < 0.005) [Figure 2].{Figure 2} Mean glycated hemoglobin To assess the effect on mean HbA1c, three trials that reported mean HbA1c at 12 months were selected.[22],[23],[24] Data were available for 218 patients, of whom 115 received oral insulin and 103 received placebo. Based on results from metaanalysis, no significant difference was observed in the mean HbA1c levels between the insulin and placebo arms at 12 months (SMD: 0.16, 95% CI: −0.10 to 0.43, I2 = 0, P = 0.88) [Figure 3].{Figure 3} Mean Cpeptide levels Four trials that reported Cpeptide levels were included to compare oral insulin versus placebo.[18],[19],[22],[24] Results showed that the mean Cpeptide levels were lower in the oral insulin arm compared with the placebo arm. However, the difference does not yield statistically significant results (SMD:0.61, 95% CI: −1.39 to 0.16, I2 = 88.42, P = 0.06) [Figure 4].{Figure 4} Mean autoantibodies targeted to insulin Three trials that reported IAA levels were included to compare the effect of oral insulin versus placebo.[24],[25],[26] Results showed that the mean levels of IAA were substantially lower in the oral insulin arm (SMD: −0.49, 95% CI: −0.82 to −0.16, I2 = 27.12, P < 0.005) [Figure 5].{Figure 5} Time to diabetes Based on the data from three trials, a significant difference was not observed in the oral insulin versus placebo arms for the rate of development of type1 diabetes [Table 3].[27],[28],[29]{Table 3} Other outcomes Safety Among the 13 trials, four did not report safety data.[24],[25],[26],[29] There were no oral insulinrelated deaths reported in any trial that assessed the capsule formulation. Serious AEs or severe hypoglycemic events were not observed in any trial.[17],[27],[28] Three treatmentemergent AEs were observed in a patient who was his own control during oral HDVI treatment, and included forearm infiltrate and right forearm site tenderness.[17] In another trial, infection was reported as the most common AE. Overall, 134 AEs were reported in the oral insulin arm and 120 AEs were reported in the placebo arm.[27] The rate of chemical hypoglycemia was found to be 4.4 per 100 patientyears (PTYR) and 3.4 per 100 PTYR in the oral insulin and placebo arms, respectively (P = 0.387).[28] There were no AEs observed (insulin arm) in two trials.[22],[23] In the placebo arm, one patient reported orofacial angioedema and four AEs, headache (two cases), itching left ear, and hyperglycemia were reported in another trial.[17],[22] No deaths, serious AEs, or AErelated discontinuations were reported when oral insulin was administered in the form of hexylinsulin monoconjugate 2 (HIM2). A total of 58 AEs were reported in 15/16 patients in one trial, and 13/18 patients in another trial.[20],[21] Asthenia (n = 4) and headache (n = 4),[21] and headache (n = 6) and anemia (n = 3) were frequently reported.[20] One AE, urinary urgency (mild intensity) was considered to be of unknown relationship to oral insulin. Individual clinical laboratory value assessment demonstrated that mean pulse rates were higher in patients administered with 1.0mg/kg of HIM2 than those who received 0.5 mg/kg of HIM2; however, the difference was not found to be of clinical importance. Among the total of 12 hyperglycemic events that occurred throughout the trials, three were reported after HIM2 administration. One patient experienced hypoglycemia after HIM2 dosing but did not require any supplemental glucose.[21] One patient who received 1.0 mg/kg of HIM2 had mild AEs of dizziness, headache, and nervousness that were reported to have possible causality to oral insulin.[20] No AEs or severe hypoglycemic episodes that required medical intervention were reported in the trials which assessed the spray formulation (Oralin).[18] There were no serious AEs in the trials that evaluated insulin tablets. One discontinuation due to hypoglycaemia was reported. All six hypoglycemic events (n = 5) were related to IN105. One patient required oral glucose intervention. Elevated serum triglyceride level (n = 2) was another frequently observed AE. One episode each of dizziness and hyperhidrosis was reported in the patient with a relative hypoglycaemia. One increased triglycerides AE and hypoglycemia were considered to be related to IN105. Patients with dizziness and hyperhidrosis demonstrated complete recovery with the treatment.[19] Discussion In the current metaanalysis, the effect of administering insulin orally was determined based on the levels of glucose fluctuation postmeals, the levels of HbA1c in plasma, formation of insulin via measuring Cpeptide levels, and the destruction of βcells of pancreas via autoimmune antibodies produced targeting insulin. Pooled analysis was performed individually for determining the effect of oral insulin versus placebo using the four parameters, 2PPG excursions, Hb1Ac, Cpeptide levels and IAA levels. Prior to conducting metaanalysis on the above parameters in 13 clinical trials, the risk of bias and quality of all the trials was analysed using the Cochrane Risk of Bias Tool. Results of 2PPG excursions indicated that insulin could significantly reduce fluctuations in the levels of 2PPG compared with placebo. As such fluctuations have the potential to cause vascular complications, oral insulin may provide protection against hyperglycemic shock by acute elevation in blood glucose level postmeals. The trials by Clement 2004 and Kipnes 2003 provided data on 2PPG with both, the minimum effective dose of oral insulin, and pooled data calculated for all the doses. In the trial by Clement, 2PPG values were lower in insulin arm than placebo when data for all the doses were pooled,[21] while in the Kipnes trials, they were significantly lower with both, minimum effective dose and pooled dose data.[20] Although glucose levels and glucose excursion levels are the most frequently monitored parameters for diabetes mellitus, they do not reflect longterm glycemic control. To overcome this issue, glycated proteins are measured in the blood.[30] Based on results from metaanalysis, HbA1c levels 12 months after initiating therapy were not found to be significantly affected by oral insulin compared with placebo. However, the I2 statistic (measure of heterogeneity) value was zero, which implied that all the observed variance was spurious, and the dispersion could be attributed to random error. Hence, to conclude the efficiency of oral insulin for glycemic control in the long run data from more trials assessing the longterm effects of oral insulin is required. Another parameter frequently measured is levels of endogenous insulin. As Cpeptide is produced in equal amounts to insulin, it serves as one of the best parameters to evaluate endogenous insulin secretion.[4] Two trials, each, in our metaanalysis on Cpeptide levels included patients with diabetes (type1 and type2). Results from our metaanalysis indicate that oral insulin did not influence Cpeptide levels in patients with type1 diabetes who received capsule formulation.[22],[24] Among the two trials involving patients with type2 diabetes, Cpeptide levels were supressed to a greater extent in the Oralin treated arm of one trial indicating adequate Oralin absorption in the buccal mucosa thus relieving pancreatic burden to produce high quantity of insulin for controlling the level of blood PPG.[18] In the other trials, Cpeptide levels did not differ significantly from placebo arm at the minimum effective dose of oral insulin administered as a tablet formulation. However, Cpeptide levels decreased with an increase in the dose of oral insulin. Cmax of Cpeptide was significantly lower with 15 mg and 30 mg of oral insulin compared with placebo.[19] Overall results from metaanalysis indicated that oral insulin reduced the levels of Cpeptide compared with placebo. Although the difference was not statistically significant at minimum effective doses evaluated in this metaanalysis, administration of higher doses with more readily absorbable formulations provide significant reduction in Cpeptide levels that may decrease the burden and hence reduce the rate of destruction of pancreatic cells. This analysis hence suggested that the effect of oral insulin varies with the type of diabetes, dose of insulin, and formulation. Easily absorbable formulations that can bypass firstpass mechanism provided higher suppressive effect, while other formulations may require higher doses to provide equivalent suppression of endogenous insulin release in patients with type2 diabetes. It is known that subcutaneous (SC) insulin injections induce insulin antibodies.[31],[32] In our review, we assessed if oral insulin can increase/decrease the levels of IAAs in patients with diabetes. Results of metaanalysis demonstrated that levels of IAA with oral insulin were significantly lower compared with placebo. This is contrary to the common response observed with SC injections proving oral administration of insulin to be safer with respect to eliciting immune response.[31],[32] In addition to performing the metaanalysis, we also reviewed the time to develop type1 diabetes. However, based on data from three trials, no difference was reported in the onset of diabetes in patients administered with oral insulin. Overall, insulin was well tolerated with hypoglycemia being the most frequent AE. No deaths or oral insulinrelated serious AEs were reported in the studies. Conclusion Oral insulin provided significant benefits to patients with diabetes mellitus compared with placebo for amelioration of acute responses such as postprandial glucose excursions and insulin release. However, its efficacy varied with the dose and type of formulation. Overall, oral insulin was found to elicit a significantly lower immune response compared with placebo, which is contrary to the elevated levels of IAAs released after SC administration. It was safe and well tolerated. This new formulation has potential to augment and improvise the management of diabetes mellitus. More trials are required to assess the longterm effects of oral insulin. Financial support and sponsorship Nil. Conflicts of interest SD is an employee of Sanofi Healthcare India Pvt Ltd, and DD is an employee of Pharmaceutical Product Development, Inc. Supplementary File 1 [INLINE:1] Reference Cochrane Risk of Bias Tool; Appendix F. Available from: https://www.ncbi.nlm.nih.gov/books/NBK132494/bin/appffm1.pdf. [Last assessed on 2020 Feb 25]. Supplementary Files 2 The key outcome measures for the analysis were postprandial blood glucose excursion levels at 2 hours, glycated hemoglobin (HbA1c) after 12 months, levels of Cpeptide, and levels of immune antibodies (IAA). The index used to represent study findings in metaanalysis is an effect size statistic, which embodies information about both the direction and the magnitude of quantitative research findings. Standardized mean difference (SMD), an appropriate effect size statistic, has been used to represent the analytical findings of the outcome measures from the oral insulin and placebo arms for each trial. This approach was first suggested by Cohen (1969) in connection with describing the magnitude of effects in statistical power analysis. The sample estimate is often called Cohen's d in research synthesis.[1] The standardized mean difference (Cohen's d), is calculated from the statistical information reported in a research study according to: [INLINE:2] Where x̅1 and x̅2 are the sample means in the oral insulin and placebo groups, and spooled is the withingroups standard deviation, pooled across groups. [INLINE:3] Where n1 and n2 are the sample sizes, and s1 and s2 are the standard deviations in the two groups, respectively. The variance of d is given as, [INLINE:4] The standard error of d is the square root of Vd, [INLINE:5] However, this effect size estimate has been shown to be upwardly biased that tends to overestimate the absolute value of standardized mean difference parameter in small samples (n < 20). Hedges (1981) provided a simple correction for this bias and the corrected sample estimate is known as Hedges' g.[2] A correction factor, known as J, is used to convert from Cohen's d estimate to Hedges' g estimate. [INLINE:6] Where df is degrees of freedom used to estimate spooled, which for two independent groups is n1 + n2  2. Thus, the standardized mean difference (Hedges' g), was calculated according to: [INLINE:7] This corrected or unbiased effect size estimate, Hedges' g, was used in all the analytic computations for metaanalysis The variance of g is given as, [INLINE:8] The standard error of g is the square root of Vg, [INLINE:9] The effect size estimates for each study was computed. The estimates were then used to assess the consistency of the effect across studies and a summary effect based on randomeffects model was computed to evaluate the overall results[1] Confidence intervals indicate the range within which the sample mean effect size is likely to be, given the observed data. For example, a 95% confidence interval around a sample mean effect size indicates a 95% probability that the sample mean effect size is between these two values. This is useful in indicating the degree of precision of the estimate of the mean effect size. Additionally, if the confidence interval does not include zero, then the mean effect size is statistically significant at the level specified by the confidence interval (i.e., α = 0.05 for a 95% confidence interval). Assuming the effect size was normally distributed, 95% confidence intervals were computed based on standard normal (z) distribution. The confidence interval for a mean effect size is based on the standard error of the mean difference and a critical value from the standard normal (z) distribution (e.g., 1.96 for α =0.05), LLg = g  zα * SEg and ULg = g  zα * SEg Where Za denotes the Zvalue corresponding to confidence limits of α level of significance. zα = 1.96, where α = .05 A test statistic Zp was also computed to test the null hypothesis that the common true effect is zero as [INLINE:10] There is a necessary correspondence between the P value for Z and the confidence interval, such that, the P value will fall under 0.05 if and only if the 95% confidence interval does not include the null value. Therefore, by scanning the confidence intervals the statistically significant studies could be easily identified. For a onetailed test the P value is given by, [INLINE:11] Where we choose '+' if the difference is in the expected direction or '–' otherwise and Φ(Zp) is the standard normal cumulative distribution of Zp. Randomeffect metaanalysis method has been applied to obtain the most precise estimate of overall mean. i.e., the summary effect from a distribution of effect sizes. It assumes that the effect sizes are normally distributed. Thus, allows us to combine the effect sizes and evaluate the statistical significance of the summary effect.[3] In this approach, a weighted mean was calculated for each study, where the weight assigned is the inverse of that study's variance, given as, [INLINE:12] Where Vgi is the withinstudy variance for study i and T2 is the betweenstudies variance. Formulas for computing the withinstudy variance were presented earlier. Based on DerSimonian and Laird method, we computed [INLINE:13] Where [INLINE:14] df = (k1), k is the number of studies [INLINE:15] The weighted mean (M) was then computed as [INLINE:16] i.e., the sum of the products Wi gi (effect size multiplied by weight) divided by the sum of the weights Wi. The variance of the summary effect is estimated as the reciprocal of the sum of the weights, [INLINE:17] The estimated standard error of the summary effect is the square root of the variance, [INLINE:18] Then, 95% lower and upper confidence limits for the summary effect are estimated as LLM = M  1.96 * SEM and ULM = M  1.96 * SEM Finally, a Zvalue to test the null hypothesis that the mean effect is zero was computed using[3] [INLINE:19] For a onetailed test the P value is given by, [INLINE:20] Where we choose '+' if the difference is in the expected direction or '–' otherwise and Φ(Zp) is the standard normal cumulative distribution of Z.[4] Heterogeneity in effect sizes was assessed by quantifying the variation in true effect sizes using I2 statistic. However, the variation that we observed was partly spurious, incorporating both true heterogeneity and random error. According to Higgins et al. (2003), the I2 statistic, which reflects the ratio of true heterogeneity to total variance across the observed effect size estimates, is given as, [INLINE:21] This I2 statistic is not sensitive to both the metric of the effect size and the number of studies. If I2 is near zero, then we assume almost all the observed variance is spurious. By contrast, if I2 is large, then that signifies most of the observed variance is real. In the schematic, the effect size for each study and the summary effect were bounded by a confidence interval, reflecting the precision with which the effect size has been estimated in each case. Moreover, P values has been computed for each study and for the summary effect based on the magnitude of the effect size and the volume of information on which the estimate is based. For each outcome measure, the P value of the summary effect is substantially more compelling than that of any single study. [INLINE:22] [INLINE:23] [INLINE:24] [INLINE:25] Reference Borenstein M, Hedges L, Higgins J, Rothstein H. Effect sizes based on means. In: Introduction to MetaAnalysis. John Wiley & Sons Ltd, West Sussex, United Kingdom. 2009. p. 2132.Lipsey M, Wilson D. Selecting, computing, and coding the effect size statistic. In: Practical MetaAnalysis. SAGE Publications, California, United States of America. 2001. p. 4851.Lipsey M, Wilson D. Analysis issues and strategies. In: Practical MetaAnalysis. SAGE Publications, California, United States of America. 2001. p. 11625.Borenstein M, Hedges L, Higgins J, Rothstein H. Randomeffect model. In: Introduction to MetaAnalysis. SAGE Publications, California, United States of America. 2001. p. 6975. References


