Isotretinoin: a rare item here with no placebo-controlled trial on the page — and two reviews that came apart

A 2026 meta-analysis in Front Med carries this entry's badge, and we carry a 2023 systematic review in Curr Drug Saf alongside it. The randomised trials they looked at are almost the same. And their conclusions read differently.

First, the 2026 meta-analysis.

It pooled four randomised controlled trials, 210 randomised (202 analysed), comparing low dose (≤0.5 mg/kg/day) with conventional dose (0.5–1.0 mg/kg/day). The primary outcome was the change in GAGS (Global Acne Grading System) score at 24 weeks, with certainty of evidence assessed by GRADE.

The results.

• at 24 weeks the pooled mean difference favoured neither regimen [MD = −1.87, 95% CI (−4.38, 0.64); p = 0.15; I² = 85.67%]
• heterogeneity was predominantly attributable to one trial — it enrolled more severe cases and used adjunctive antibiotics and corticosteroids. Excluding it reduced heterogeneity (I² = 44.97%) and still showed no significant difference [MD = −0.92, (−2.62, 0.77); p = 0.28]
• post-treatment worsening did not differ overall, and in the sensitivity analysis excluding that trial showed no difference with no heterogeneity [MD = 0.10, (−1.67, 1.86); p = 0.91; I² = 0%]
• low-dose regimens had higher patient satisfaction

Next, the 2023 systematic review.

Out of 921 articles found electronically, after excluding those not measuring with GAGS, four randomised controlled trials were included, assessed with the Cochrane Risk of Bias Tool.

• across all trials, low-dose regimens were preferable in all types of acne — with similar efficacy to conventional dose but fewer side effects and better patient satisfaction and compliance
• a continuous low-dose regimen had the best efficacy among low-dose regimens
• the authors listed limitations — a slight difference in dosage between the selected studies, some studies not explaining side effects and relapse rate thoroughly, and not stating the compliance scoring method used
• conclusion: it recommends continuous low-dose treatment as the chosen regimen. However, further evaluation regarding relapse rate compared with the conventional dose is needed

Now how to read it. The content of this entry is the divergence between the two.

First, set the two sentences side by side.

• 2023 — 'low dose is preferable'
• 2026 — 'neither regimen was favoured (p = 0.15)'

They do not contradict. Both say 'the efficacy is similar'. What differs is what each added next. The 2023 review started from similar efficacy and added side effects and compliance to get to 'therefore low dose'. The 2026 meta-analysis looked at efficacy alone and stopped at 'no difference'.

That is why this site carries both. Read one alone and it reads as 'low dose is better' or 'the dose does not matter', when what the material actually says is 'the efficacy is similar, and on everything else low dose is easier'.

Second, look at I² = 85.67%. Heterogeneity is very high — close to meaning the four trials were measuring different things. And the 2026 meta-analysis named the source: one trial that enrolled more severe cases and used antibiotics and corticosteroids alongside. The question this site keeps returning to appears here too — what was it added on top of.

Third, 210 people. Four randomised trials pooled come to 210. That number says how thin the material on this question is. One trial on the finasteride page here was 1,553.

Fourth, the endpoint is GAGS. A severity score assigned by a person looking. Not counted lesions but a grade. So this entry's endpoint is marked 'graded'.

Fifth, both reviews regretted the relapse data. The 2023 review notes that some studies did not explain relapse rate thoroughly, and its conclusion says further evaluation of relapse rate against conventional dose is needed. In acne, relapse is the central question. That place is empty.

Sixth, what this entry compared is dose against dose. There is no placebo. The fact set out at the top of this page applies here unchanged.

So it sits at 'moderate'. A good shape — two meta-analyses attached — but 210 people even pooled, no placebo, empty relapse data, and two reviews whose concluding sentences diverge.

A cohort survey in N Engl J Med, 1995, from the Slone Epidemiology Unit, Boston University School of Public Health. It is the largest entry by numbers on this site.

First, why this is up here as a badge. This site usually badges efficacy only. But for this drug the risk of exposure in pregnancy is heavier than any other item. Showing efficacy as a badge and leaving this as a line under risks would be a screen that leans. The finasteride page made the same call.

Background. The drug is effective for acne but it is teratogenic. To minimise pregnancies among exposed women, the manufacturer together with the U.S. Food and Drug Administration implemented a multicomponent Pregnancy Prevention Program in 1988. This paper reports an ongoing survey designed to assess compliance with that programme.

Methods. Treated women enrolled through their physician, by filling out a form in the medication package, or by calling a toll-free number. They were randomly assigned to follow-up by telephone or by mail. Telephone interviews were at the start of therapy, in the middle, and six months after it ended; mailed questionnaires six months after it ended. Median duration of therapy was 20 weeks.

The results.

• between 1989 and 1993, 177,216 eligible women enrolled in the survey
• among 24,503 interviewed within one month of enrollment, 99% had been told to avoid pregnancy
• at that time about 54% were not sexually active (of whom 37% used contraception), 42% were sexually active (of whom 99% used contraception), and 4% were infertile
• among 124,216 women with completed telephone or mail follow-up, there were 402 pregnancies during therapy — 3.4 per 1,000 courses of isotretinoin
• of the pregnant women, 72% had elective abortions, 16% spontaneous abortions, 3% ectopic pregnancies and 8% live births
• conclusion: the pregnancy rate among women receiving isotretinoin was substantially lower than in the general population and was compatible with the programme working

Now how to read it.

First, we write the number without inflating it in either direction. The authors' conclusion leans towards the programme having worked. The pregnancy rate was lower than in the general population. At the same time there were 3.4 per 1,000 courses. Both sentences are true together.

Second, so what this entry says is not 'the programme is useless'. What it says is 'even with the programme it did not reach zero'. And that means there is work left to the person taking it.

Third, there is a reason the distribution of outcomes is written out. 72%, 16%, 3%, 8%. Those numbers show what follows when a pregnancy does occur. This site does not omit them. Because that weight belongs in the decision to start this drug. And this site writes no judgement whatsoever about those choices.

Fourth, it is 1995 and the United States. The programme has been revised several times since, and systems differ between countries. What is checked, and how, where you are now is to be confirmed in a clinic and in the product information. This site does not guide that procedure.

Fifth, the design's limits. This is not a randomised controlled trial but an enrollment-based follow-up survey. It is self-reported by people who chose to enroll, and of 177,216, only 124,216 had completed follow-up. What happened among those lost to follow-up is unknown. Under-reporting is possible.

Sixth, look at the contraception figures. Of the 42% who said they were sexually active, 99% used contraception. And still there were 402. No contraception is 100%. That is why these programmes have required two methods at once.

So it sits at 'moderate'. A large cohort of 124,216, but not a randomised trial, based on self-report, and from 1995.