The Medline database of medical publications, the Cochrane library of randomized controlled trials, and EMBASE database were carefully searched by the five authors for English publications comparing artificial pancreas with its control group (any control group). Reference lists of selective publications (most relevant ones) were also carefully reviewed for appropriate articles. In addition, official websites of specific journals such as Lancet Diabetes and Endocrinology, Diabetes Care, and Cardiovascular Diabetology were also searched for any relevant publication.
During this process, we searched for the terms “artificial pancreas”, “bionic pancreas”, “closed loop glucose control”, “type 1 diabetes mellitus”, “diabetes mellitus”, “glucose control”, and “glucose monitoring”, one at a time and in combination. Publications comparing artificial pancreas with its control group were considered relevant to this analysis.
Studies were included if they compared artificial pancreas with its control group (any relevant control group); they reported glucose control parameters or other outcomes related to glucose monitoring (during daytime and overnight/24-h basis); they involved patients with T1DM.
Studies were excluded if they did not compare artificial pancreas with its control group; they did not report glucose monitoring parameters as their endpoints or they reported only daytime or overnight measurement, but not both in combination; they involved patients with type 2 diabetes mellitus instead of patients with T1DM; they were duplicates.
Type of Participants, Endpoints, and Follow-Up Periods
This analysis included patients with T1DM.
The endpoints (during daytime and nighttime) which were analyzed included:
Mean and median glucose concentration.
Time spent outside the euglycemic phase (outside the glucose range 3.90 to 8.0 mmol/L).
Time spent in the hypoglycemia phase (blood glucose < 3.9 mmol/L).
Insulin required/delivered per day.
Time spent in the hyperglycemia phase (blood glucose > 10.0 mmol/L).
Number of hypoglycemic events.
The follow-up period ranged from less than 1 week to 2 months.
Data Extraction, Review, and Statistical Analysis
After the search process, which was conducted in accordance with the PRISMA guideline , the same reviewers assessed the titles and abstracts and independently selected the most suitable articles which satisfied the inclusion and exclusion criteria of this analysis and then data were extracted. The studies which were included in this analysis were judged as having low to moderate risk of bias .
This is a meta-analysis and therefore inconsistency across the studies was evident . Hence, heterogeneity was assessed by two statistical methods:
The Q statistic test, whereby a P value less or equal to 0.05 was considered statistically significant.
The I2 statistic test; a high percentage value indicated high heterogeneity (whereby a random effects model was used) and low percentage value denoted low heterogeneity (whereby a fixed effects model was used).
Since continuous data was used in this analysis, i.e., mean and standard deviation (SD), data were evaluated by means of weighted mean differences (WMDs) and 95% confidence intervals (CIs). In case the SD value was not provided, but a p value was given, SD was calculated using the formula SD = √n × p × (1 – p). The analysis was carried out by RevMan 5.3 software.
Publication bias was assessed by visually observing funnel plots.
Compliance with Ethics Guidelines
This meta-analysis is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.