Three ordinary drugs track less sight-threatening retinopathy

Design

retrospective propensity score-matched cohort study with positive and negative drug controls, eight negative-control outcomes, E-values and nine sensitivity analyses

Population

adults with type 2 diabetes and documented eye care in a United States network, 2005–2025; matched pairs from 18,762 to 86,846 per drug

Primary outcome

incident diabetic macular oedema and proliferative diabetic retinopathy within three years

Effect

macular oedema hazard ratios 0.62 cetirizine, 0.63 ibuprofen, 0.46 prednisone; proliferative retinopathy 0.59, 0.48, 0.36; fenofibrate control 0.57 and 0.59; gabapentin control 0.95 and 1.04

Inflammation is an accepted part of diabetic retinopathy biology, and there has never been a practical way to test whether damping it generally makes any difference. A retrospective cohort in a large United States network took an unusual route: identify adults with type 2 diabetes taking cetirizine, ibuprofen or prednisone for their ordinary indications, propensity-match each cohort 1:1 against non-users on 44 covariates, and count incident diabetic macular oedema and proliferative retinopathy over three years.

All three drugs were associated with less disease. Cetirizine: macular oedema hazard ratio 0.62 (95% confidence interval 0.51 to 0.74), proliferative retinopathy 0.59 (0.46 to 0.77). Ibuprofen: 0.63 (0.56 to 0.69) and 0.48 (0.41 to 0.56). Prednisone: 0.46 (0.41 to 0.52) and 0.36 (0.30 to 0.43). Fenofibrate, included as a positive control because its effect is already established, gave estimates of similar size (0.57 and 0.59). Gabapentin, the negative control, gave nothing (0.95 and 1.04). Matched pairs ranged from 18,762 to 86,846, and the findings held across nine sensitivity analyses including a new-user design and glycaemic strata.

The design deserves credit — a positive control, a negative control, eight negative-control outcomes and E-values from 2.57 to 4.97 is more confounding control than most observational pharmacoepidemiology attempts. It still cannot establish cause. Prednisone in particular raises glucose and is given for acute illness, and its being associated with less retinopathy than non-users should provoke suspicion rather than prescribing. The honest reading is that this justifies a trial, and nothing else.

• Do not start any of these drugs to protect the retina; none is indicated for that.
• Do not stop an antihistamine or a non-steroidal in a diabetic patient on the assumption it is irrelevant either.
• Keep glycaemic and blood pressure control, and screening interval, as the interventions with trial evidence.
• Note the negative control result — gabapentin showed nothing, which is what makes the positive findings harder to dismiss.
• Watch for a prospective trial; these authors say that is what the data justify.

Why it matters

If damping general inflammation slows diabetic retinopathy, the intervention would be cheap and already licensed — which is exactly why it needs a trial rather than adoption.

Don't overread it

This is observational — an association with a drug taken for another reason cannot show that the drug protects the retina.

The statistics, in plain English

An E-value of 2.57 to 4.97 is the key number: it says an unmeasured confounder would have to be associated with both the drug and the outcome at that strength, above everything already adjusted for, to explain the result away. Those are large values, which makes simple confounding a harder explanation — but the prednisone finding, with the largest effect, also showed mild protective bias on the negative-control outcome panel, meaning some residual bias was detectable. Propensity matching on 44 covariates balances what was recorded; it cannot balance what was not, and in a network database the reason a drug was prescribed is often not recorded.

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.