Pancreatic fat tracks with future diabetes, but the pooled estimate is fragile

Design

Systematic review and meta-analysis of longitudinal observational studies

Population

Ten studies of adults with baseline pancreatic fat measurement and subsequent glycaemic follow-up

Primary outcome

Incident type 2 diabetes and glycaemic progression

Effect

Pooled estimate 2.56 (95% CI 1.27–5.14, I² 93.3%); 1.45 (1.23–1.73, I² 10.2%) after excluding one influential study

Ten longitudinal studies were pooled to test whether pancreatic fat at baseline predicts later dysglycaemia. In the primary binary analysis of five studies, higher pancreatic fat burden was associated with incident type 2 diabetes with a pooled estimate of 2.56 (95% CI 1.27–5.14) — but heterogeneity was I² = 93.3%, which is close to the point where pooling stops being meaningful.

Removing one influential ultrasound-based study changed the picture. The association survived at 1.45 (1.23–1.73) with I² falling to 10.2%. A separate continuous analysis of three studies gave 1.17 (1.04–1.32, I² = 79.2%), and two studies linked pancreatic fat to glycaemic progression at 1.96 (1.10–3.48).

The honest reading is that the direction is consistent and the magnitude is not established. Much of the spread comes from how pancreatic fat was measured — ultrasound, CT and MRI are not measuring the same thing to the same precision. This is a research-stage marker, not a reason to start requesting pancreatic fat quantification in clinic.

• Do not order imaging to quantify pancreatic fat outside a study — no threshold, no intervention and no outcome data support it.
• Where a scan done for another reason reports a fatty pancreas, treat it as one more prompt for standard metabolic screening, not a new diagnosis.
• Screen with HbA1c or fasting glucose, which have thresholds and evidence behind them.
• Note the measurement method whenever a study reports ectopic fat — it is the main source of disagreement between them.

Why it matters

An imaging marker with no threshold and no intervention behind it is a research finding, however biologically plausible it looks.

Don't overread it

These were observational studies — they cannot show that reducing pancreatic fat prevents diabetes.

The statistics, in plain English

I² of 93.3% means almost all the variation between studies is real disagreement rather than chance, so the pooled 2.56 is an average of results that do not belong together. That one study's removal both halved the estimate and collapsed heterogeneity to 10.2% tells you the headline number was largely that study's number.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.