Design: Systematic review and meta-analysis of 85 double-blind randomised trials
Population: 70,543 adults with diabetes, heart failure, chronic kidney disease or cardiovascular disease
Primary outcome: Change in haemoglobin and haematocrit
Effect: Haemoglobin +0.60 g/dL (95% CI 0.55–0.66); haematocrit +2.36 points (2.19–2.52) vs control

A JAMA meta-analysis pooled 85 double-blind randomised trials with 70,543 participants with diabetes, heart failure, chronic kidney disease or cardiovascular disease, mostly against placebo and with a median duration of 24 weeks. Compared with control, SGLT2 inhibitors raised haemoglobin by 0.60 g/dL (95% CI 0.55–0.66) and haematocrit by 2.36 percentage points (2.19–2.52). Erythropoietin rose slightly and transferrin saturation fell slightly (by 0.84 percentage points).

The practical value is in interpretation. A modest rise in haemoglobin after starting an SGLT2 inhibitor is a known drug effect and does not on its own call for a work-up for polycythaemia. Conversely, a haemoglobin that falls on treatment is moving against the drug's expected direction and deserves a reason.

The small fall in transferrin saturation fits increased red cell production drawing on iron stores. In patients with chronic kidney disease or heart failure, where iron deficiency is common, the rise in haemoglobin could make an underlying deficiency look less obvious. The authors describe the clinical significance as unclear, and no guideline has changed on the basis of this analysis.

Expect a rise of about 0.6 g/dL in haemoglobin within months of starting an SGLT2 inhibitor.
Do not refer for polycythaemia on a modest haematocrit rise that follows starting the drug.
Treat a falling haemoglobin on an SGLT2 inhibitor as unexplained until proven otherwise.
In kidney disease or heart failure, consider checking ferritin and transferrin saturation rather than relying on haemoglobin alone.
Note the start date of the SGLT2 inhibitor on the record so a later rise is read in context.

Why it matters

It changes how a routine blood count is read in a large share of diabetes, kidney and heart failure clinics.

The statistics, in plain English

Across trials the difference was 0.60 g/dL with a narrow 95% confidence interval (0.55 to 0.66), so the size is estimated precisely. Statistically certain does not mean clinically important: a 0.6 g/dL change is small for an individual, and the authors say its importance is unclear.