Machine Learning Models for Glycemic Prediction in Type 1 and Type 2 Diabetes: Systematic Review and Meta-Analysis.

What was studied

This review looked at machine learning models that predict blood sugar in people with diabetes. It covered type 1, type 2 and mixed groups. The team searched 9 databases. They included 299 reports, which covered 270 studies. 19 reports could be pooled.

What they found

The pooled error for glucose forecasting was 1.18 mmol/L. The pooled area under the curve for event prediction was 0.926. Of 28 assessment units, 24 were at high risk of bias. External validation was done in 10 of 251 studies. Prospective evaluation was done in 2 of 270 studies.

What it means for patients

The authors say these studies show forecasting and discrimination signals. They say pooled numbers cover only the part of the literature that could be analyzed. They say future studies need clearer reports on validation, calibration and use.

Why it is not certain yet

The authors say the studies differed and had high risk of bias. Reports of uncertainty were incomplete. External and prospective validation was limited. The authors say this prevents deployable benchmarks.

Source

Machine Learning Models for Glycemic Prediction in Type 1 and Type 2 Diabetes: Systematic Review and Meta-Analysis.

We restate facts from the source record in our own words. The original wording belongs to its publisher.