Isotretinoin and Mental Health: Understanding Apparently Conflicting Research

By Michael Wiederkehr, M.D.

A person considering isotretinoin can encounter two headlines on the same day: one says acne treatment improves depression, while another links the medicine to psychiatric adverse events. The headlines seem to demand a choice. Either the medicine is psychologically beneficial or it is dangerous. The evidence is more useful when the questions behind those headlines are separated.

Average symptom change is not the same outcome as a new diagnosis. A report of suicidal thoughts is not the same event as a suicide attempt. An association in a healthcare database is not the same finding as a causal drug reaction established in an individual. These distinctions do not minimize distress. They explain why several apparently conflicting observations can coexist.

The relevant clinical literature includes a depression review, supplementary diagnosis-code definitions for a large comparative cohort, and adverse-event reporting material. This article uses those sources alongside further primary studies and current official safety information. It explains what the research can support, what remains uncertain, and how that uncertainty can be handled in ordinary clinical care.

The central practical point is consistent across these different sources: population findings should inform prescribing, while new symptoms in an individual still require attention. A reassuring average should never become a reason to dismiss a person's experience. Equally, the existence of reported adverse events should not be translated into an unsupported claim that everyone taking isotretinoin will develop psychiatric problems.

Begin by naming the outcome

The phrase "mental health risk" can refer to many different outcomes. A study might measure a depression questionnaire, a clinician-recorded diagnosis, a prescription, a hospitalization, suicidal thoughts, self-harm, or a suicide attempt. Each outcome has a different definition and a different relationship to the underlying experience.

These are related experiences, but combining them carelessly makes the evidence harder to understand. A study can find less recorded depression and more recorded suicidal ideation without a mathematical contradiction. The categories have different definitions, ascertainment, and possible influences.

The first question to ask of a headline is therefore specific: "What was counted?" The next is "How was it counted?" A self-completed questionnaire, a structured interview, and an administrative code each reveal something different. None provides a complete picture on its own.

This also explains why a patient can experience distress that does not appear in a database study. The absence of a code may mean no diagnosis was made, no consultation occurred, or the information was recorded elsewhere. It does not necessarily mean that the person experienced no symptoms.

Improvement in average scores and individual worsening can coexist

An average describes the group, not every member. Imagine a hypothetical group of ten patients completing the same symptom questionnaire before and after treatment. Eight improve modestly, one remains unchanged, and one worsens substantially. The group average can improve even though the final person's experience is clinically important.

This is not a special problem unique to isotretinoin. It is a general property of averages. A summary that helps describe a population can conceal differences in direction, magnitude, and timing among individuals. The appropriate response is to examine both the overall effect and the distribution of experiences.

An invented numerical example makes the distinction visible. Suppose five people each improve by four points, while another worsens by ten points. The total change is a ten-point improvement across six people, so the average improves. That calculation does not erase the worsening person's experience. It merely answers a different question.

The reverse is also possible. A group with little average change can include substantial improvement in some people and substantial deterioration in others. "No average change" does not mean "nothing happened." It means the chosen summary did not show a clear overall movement under that study's conditions.

When reading about isotretinoin, distinguish three claims:

• The average symptom score changed during treatment.
• The frequency of a defined outcome differed between comparison groups.
• A particular person's symptoms were caused by the medicine.

Evidence supporting one does not automatically establish the others. Keeping the claims separate makes it possible to discuss the benefits of successful acne treatment and the need to respond to psychiatric symptoms without forcing either concern out of the conversation.

What the 2019 depression review found

A 2019 systematic review included 20 studies examining isotretinoin and depression. Seventeen contributed to an analysis of symptom scores before and after treatment. On average, symptoms improved. Four studies examined depressive disorder risk; their pooled estimate was not statistically significant, and retrospective and prospective findings differed. ^[1]

The review is useful because it shows that symptom change and disorder risk were analyzed separately. It should not be shortened to "isotretinoin treats depression." The included treatment was for acne, and changes in psychological well-being occurred in that context. The review did not establish an antidepressant indication.

Nor does a non-significant pooled risk estimate prove that a rare individual reaction is impossible. A confidence interval describes the uncertainty around an estimate. A result that does not cross a statistical significance threshold can still be compatible with several clinically different possibilities, especially when the evidence is limited or heterogeneous.

The before-and-after design introduces another interpretive issue. If symptoms improve over a course, possible contributors include changes in acne, expectations, support, life events, and other care. Without a suitable comparison, attributing the entire change to a direct effect of the medicine goes beyond what that design can establish.

The most defensible use of this review is therefore specific: average depressive symptom improvement during acne treatment is a recurring finding in the included studies, while a separate causal conclusion about psychiatric disorders requires different evidence. That is a more informative statement than declaring the drug universally protective or universally harmful.

What the large global comparison adds

A later retrospective study compared 75,708 people with acne treated with isotretinoin with 75,708 treated with oral antibiotics. Recorded depression was less frequent in the isotretinoin group. Major depressive disorder and suicide attempts had comparable risks, while suicidal ideation was more frequently recorded, with a hazard ratio of 1.41. ^[2]

That combination is exactly why endpoint definitions matter. The result for one category should not be substituted for another. Saying that the study found "no mental health risk" omits the ideation finding. Saying it found a 41% rate of suicidal thoughts misreads a relative measure as an absolute percentage.

The supplementary document accompanying the global study identifies the diagnosis codes used to construct the categories. It defines depression through a depressive episode code and major depressive disorder through a recurrent disorder code. It also lists separate codes for suicidal ideation, attempts, anxiety, and other conditions. ^[3]

Those definitions make the study reproducible, but they also shape what the labels mean. A database category reflects the specified coding rules, not every possible clinical meaning of its title. The supplementary methods therefore provide essential context for interpreting a result that appears simple in an abstract.

The study offers a large real-world comparison. It does not randomize treatment choice, measure every unrecorded symptom, or remove every difference between treatment groups. Its findings deserve attention without being turned into a final answer about causation in every individual.

Why the comparison group changes the answer

A study needs a comparison, but choosing one is consequential. People receiving isotretinoin may differ from people using topical treatment, oral antibiotics, or no prescription treatment. Those differences can include acne severity, previous treatment failure, healthcare access, and factors that are incompletely recorded.

A 2022 propensity-matched electronic-record study illustrates this issue. Its estimate for any incident neuropsychiatric outcome was lower when isotretinoin was compared with oral antibiotics, while comparisons with topical treatment or no anti-acne prescription produced estimates closer to no difference. ^[4]

This does not mean the investigators could choose any answer they wanted. It means the comparison defines the question. "Compared with whom?" belongs beside every relative risk statement. A medicine can look different against populations with different underlying characteristics and care pathways.

The table is a guide to interpretation, not a ranking in which one design answers every question. Different designs can be complementary. Confidence increases when findings make sense across several approaches and their limitations are examined directly.

The broader meta-analysis and its population-level conclusion

A meta-analysis published in JAMA Dermatology included 25 studies with more than 1.6 million participants. It did not find epidemiological evidence of increased suicide or psychiatric disorder risk at a population level. Its pooled one-year depression risk was 3.83%, while the reported pooled risks for individual suicide-related outcomes were each below 0.5%. The authors still recommended monitoring for mental distress. ^[5]

These absolute figures describe outcomes among users in the included evidence. They should not be labeled the percentage caused by isotretinoin. To estimate an attributable effect, a suitable comparison and valid causal assumptions are necessary. The presence of an outcome during exposure is not enough.

A large participant total also does not mean every outcome was measured in every participant. Different analyses within a review can include different subsets of studies. Readers should avoid attaching the full headline sample size to each individual estimate unless the methods support that interpretation.

Meta-analysis combines information, but it cannot automatically repair weaknesses shared by the underlying studies. If multiple studies depend on incomplete coding or treatment selection, pooling can produce a more precise estimate of an association while leaving causal uncertainty. Precision and validity are related but distinct qualities.

The practical message is proportionate reassurance about the available population evidence, combined with continued individual assessment. That combination is coherent. Monitoring is not an admission that the broad analysis is false; it addresses a different level of uncertainty and the consequences of missing important symptoms.

Why events before treatment matter

A Swedish cohort study followed 5,756 people treated for severe acne and examined suicide attempts before, during, and after treatment. The investigators observed that risk was already rising before isotretinoin exposure, which complicated attribution of the later pattern to the drug. ^[6]

The timeline matters because treatment does not begin in an otherwise empty history. People may reach isotretinoin after a prolonged period of distressing acne, unsuccessful treatment, or deteriorating quality of life. A study that begins counting only on the first prescription date can miss what was happening beforehand.

An illustrative timeline clarifies the issue:

1. A person's acne becomes more burdensome.
2. Psychological distress increases.
3. Earlier treatments do not provide enough benefit.
4. Isotretinoin is prescribed.
5. A psychiatric event occurs during follow-up.

This sequence establishes temporal association with the prescription, but it contains several plausible contributors. The correct conclusion is not that the medicine played no role. It is that a valid assessment needs the earlier trajectory as well as the date of exposure.

The same principle helps clinical conversations. If symptoms began before treatment, say so. If they appeared after a dose change or during another major event, include that timing. A detailed sequence can support assessment without pretending it proves causation by itself.

Adverse-event reports answer a different question

Adverse-event reporting systems provide another type of evidence. A 2025 FAERS analysis examined isotretinoin reports and looked for statistical reporting signals. ^[7] Such work can identify concerns worth investigating, particularly when a possible event is uncommon or not well captured in trials.

A report is not the same thing as a confirmed causal reaction. FDA explains that spontaneous reporting data can contain incomplete or duplicate information, and that a report does not establish that the product caused the event. These data also cannot establish incidence rates because the denominator and reporting process are unsuitable for that purpose. ^[8]

Imagine an invented database with 200 reports mentioning a particular symptom. To calculate incidence, a reader would need to know how many exposed people were at risk, how consistently events were identified, whether reports were duplicated, and how follow-up was handled. The report count alone does not supply those answers.

Nor can the proportion of psychiatric reports among all reports be treated as the proportion of treated patients experiencing psychiatric events. The denominator is a collection of reports, not all users. This is a common and consequential error in online summaries.

The appropriate comparison is between functions. Spontaneous reporting is useful for detecting possible signals. Cohort research can help estimate associations in defined populations. Clinical assessment addresses the individual presentation. None makes the others unnecessary, because each provides a different kind of information.

What a reporting signal means

A disproportionality analysis asks whether a drug-event pair is reported more often than expected within a reporting database under a particular statistical method. It does not directly compare all treated people with all untreated people. FDA describes such data mining as hypothesis-generating rather than proof of a causal association. ^[9]

The distinction can be understood through a news analogy. A rise in stories about an event may mean the event became more common. It may also mean public attention increased, reporting improved, or the threshold for mentioning it changed. Counting stories can identify a question, but it cannot resolve every explanation for the increase.

This does not mean reports are worthless or that patients should not submit them. A well-documented report can contribute meaningful information, especially about timing, other exposures, and clinical course. The limitation concerns the conclusions drawn from aggregated reports, not the legitimacy of the person's experience.

When reading a paper, look for words such as "signal," "reporting odds," "suspected," and "disproportionality." These terms should not silently become "caused," "incidence," or "proven risk" in the article summarizing it. Accurate translation from research terminology to ordinary language is essential.

A signal also should not be ignored merely because a cohort study is reassuring. The productive question is whether additional evidence clarifies it. That may include case review, alternative explanations, more suitable comparative studies, and consistency across sources. A responsible interpretation keeps investigation open without announcing a conclusion the method cannot deliver.

Matching helps, but it does not create randomization

Propensity-score matching is a way to make observed treatment groups more comparable on recorded characteristics. A study might match people by age, recorded diagnoses, or other available information. The objective is to reduce confounding from differences that could influence both treatment choice and the outcome.

The limitation is contained in the word "recorded." A factor cannot be adequately matched if it is absent, poorly measured, or represented by a crude category. A database may record that acne exists without capturing its full severity, duration, or emotional effect. It may record a diagnosis without describing whether symptoms were stable at the time of prescribing.

An invented example illustrates residual confounding. Suppose clinicians are more cautious about prescribing isotretinoin to people they consider at high psychiatric risk. Some of that judgment may not be coded. The treated group could then begin with a different risk profile even after matching on documented diagnoses.

The reverse kind of bias is also possible. People receiving a more intensive treatment may be asked about psychiatric symptoms more often, increasing the chance that symptoms are documented. This could increase recorded outcomes without a corresponding increase in underlying symptoms. These are possible interpretive mechanisms, not proven explanations for any particular study result.

The appropriate response is to ask how a study addressed such issues and whether sensitivity analyses changed the findings. Matching improves a comparison when well used. Calling it equivalent to random assignment overstates what the design accomplishes.

Relative measures are not personal probabilities

The global study's hazard ratio of 1.41 for recorded suicidal ideation is a relative comparison over follow-up. It does not mean that 41% of users experienced ideation, and it does not directly provide one person's probability. ^[2]

Consider a separate hypothetical example. If an event occurred in 10 of 10,000 people in one group and 14 of 10,000 in another, the second proportion would be 40% higher in relative terms. The absolute difference would be four events per 10,000 people. Those invented counts illustrate arithmetic; they are not the study's event rates, and a risk ratio is not identical to a hazard ratio.

A good article provides the measure's name and avoids converting it into a more dramatic statement. A good consultation goes further by discussing whether the studied population resembles the person being treated. Even a correctly reported group estimate cannot capture every feature of an individual situation.

Why a personal history deserves an individual plan

A history of depression, anxiety, self-harm, or another psychiatric condition is relevant information for prescribing. It is not a reason to hide symptoms for fear of automatically losing access to acne treatment. The clinician needs an accurate history to weigh options and arrange appropriate care.

The broad evidence does not justify a universal rule that every person with a psychiatric history must be excluded. Equally, a reassuring population estimate does not justify proceeding without considering a person's current condition, previous episodes, existing care, or ability to obtain follow-up.

A useful baseline discussion separates past and present:

• What conditions or episodes occurred previously?
• What symptoms are present now?
• Which treatments and clinicians are already involved?
• What changes would the patient or a trusted person be likely to notice?
• Who should be contacted if symptoms emerge?
• How will follow-up be coordinated if more than one service is involved?

These are planning questions, not a psychiatric diagnosis. Their purpose is to make the course reviewable and reduce uncertainty about whom to contact. A person receiving ongoing mental health care may benefit from ensuring that the relevant clinicians know the treatment timeline.

The plan should also address practical access. "Contact us if anything changes" is incomplete if the patient does not know the service's response time or the route for urgent concerns. A clear route makes it easier to report a problem early rather than waiting for the next routine appointment.

What current safety information asks clinicians and patients to do

The current United States isotretinoin label calls for assessment of psychiatric conditions before and during treatment. It directs stopping treatment and contacting the prescriber promptly if relevant psychiatric symptoms develop, and notes that further evaluation may be necessary. ^[10]

United Kingdom safety guidance also emphasizes assessment and monitoring of mental health. A January 2026 update changed an additional prescribing requirement for patients under 18, while retaining risk-minimization measures addressing mental health and sexual function. ^[11] Local processes should be checked in their own jurisdiction rather than copied between countries.

The important practical distinction is between following safety instructions and proving causation. A clinician may recommend stopping or pausing exposure while assessing a concerning development without having established that the medicine caused it. Taking protective action and continuing diagnostic assessment are compatible.

A patient should not interpret the label as a request to manage a psychiatric event alone. Stopping a medicine may be one part of the response, but it does not necessarily address the underlying symptoms or immediate safety needs. The clinical assessment remains important.

If someone cannot remain safe or has an immediate intention to harm themselves, emergency care is appropriate rather than waiting for a dermatology message to be answered. This statement concerns urgent symptoms in an article's audience; it does not imply that the reader currently has such symptoms.

Monitoring should describe changes, not demand a perfect score

A useful mental health review asks about meaningful changes in daily life. It should not become a performance in which the patient feels pressure to report that everything is fine so the prescription can continue. Honest reporting is part of safe treatment, not evidence of failure.

A short record can include when a change began, what it feels like, how persistent it is, and whether it affects sleep, work, study, relationships, or normal activities. If a questionnaire is used, it can support the discussion, but the number should not replace the person's description.

The examples are communication suggestions rather than diagnostic criteria. They help preserve information without requiring the patient to decide the cause. A clinician can then assess urgency and determine what additional support or investigation is needed.

The role of family, friends, and trusted people

A patient may choose to involve someone they trust, particularly if that person is likely to notice changes in behavior or functioning. The role should be supportive and agreed, not surveillance. The patient remains a person receiving care rather than a problem to monitor.

A useful conversation can be simple: "I am starting a medicine with mental health monitoring. If you notice a marked change, please tell me directly and help me contact the clinician if needed." The exact arrangement depends on age, preferences, relationships, and the care setting.

The trusted person should know the contact plan. Without that, they may notice a change but not know whether to call a primary care service, the prescribing clinic, or urgent care. Advance clarity reduces the chance of an important concern becoming an unresolved family argument about whether the medicine is responsible.

At the same time, every ordinary disagreement, tired day, or emotional reaction should not automatically be attributed to isotretinoin. Overattribution can make a patient feel disbelieved in another way, as if their circumstances no longer matter. The useful task is to notice significant changes and support assessment.

For adolescents, the care team should explain how the young person can raise concerns privately as well as how caregivers can contribute. The goal is accurate communication and appropriate support. A patient who fears punishment or automatic exclusion may withhold precisely the information the monitoring plan is intended to uncover.

Five scenarios that demonstrate different questions

Symptoms improve as acne improves

A hypothetical patient reports less embarrassment and greater willingness to socialize during a successful course. This is a meaningful benefit to discuss. It does not prove that isotretinoin has a direct antidepressant effect, because the person's skin, expectations, and activities changed together.

The appropriate interpretation preserves the benefit without claiming a mechanism that has not been established. The patient can still be monitored for later changes.

New distress appears after treatment begins

Another patient develops persistent symptoms during treatment. The timing matters and should be reported promptly. It does not by itself identify the cause. The response should follow safety instructions and obtain assessment rather than wait for certainty about causation.

A population-level reassuring result should not be used to tell this person that their symptoms cannot be related or do not deserve care.

A pre-existing condition worsens

A third patient has a known psychiatric condition and notices deterioration. The earlier condition is relevant, but it should not become an excuse to dismiss a possible treatment contribution. The clinicians need the previous pattern, current symptoms, treatment timeline, and other changes.

Coordination is particularly useful here because the dermatologist and mental health clinician may each hold information the other needs.

A questionnaire and a conversation disagree

A fourth patient has a relatively unremarkable screening score but describes a major change in daily functioning. The discrepancy deserves exploration. The questionnaire is one source of information, not an instruction to ignore the account.

Conversely, a high score needs interpretation and assessment rather than an automatic conclusion about a specific diagnosis or cause.

An online report creates intense fear

A fifth patient reads a dramatic adverse-event account and becomes frightened before starting treatment. The concern should be discussed respectfully. Explaining study design and the individual monitoring plan can be more useful than either dismissing the report or repeating its most alarming interpretation.

These scenarios show why care must address both evidence and experience. Neither alone supplies every answer.

How to judge a headline or article

A useful article states the study design, population, outcome, comparison, follow-up, and main uncertainty. It uses verbs that fit the evidence. "Associated with" should not become "causes" without additional justification. "No statistically significant difference" should not become "proved completely safe."

Before accepting a claim, ask:

1. Does the article distinguish symptoms, diagnoses, thoughts, and attempts?
2. Does it identify the comparison group?
3. Is the number an absolute rate, relative estimate, or report count?
4. Does the time period match the claim?
5. Were symptoms present before treatment?
6. Does the source examine individual reports or a defined population?
7. Are important limitations described near the conclusion?
8. Does the practical advice follow current official instructions?

These questions are not a demand that every reader become an epidemiologist. They are a way to identify when a dramatic sentence has omitted information necessary to understand it. A careful headline can still be readable without pretending the evidence is simpler than it is.

For example, "Large review finds no increased population risk but supports monitoring" is less absolute than "Drug cleared of all psychiatric risk." The first preserves the level of the finding and its practical consequence. The second asserts a certainty that no observational synthesis can provide for every individual.

What further research would help

Research becomes more useful when it captures the trajectory before treatment, meaningful differences in acne burden, psychiatric history, concurrent treatments, and changes during follow-up. Better information about these factors can reduce uncertainty that remains after large database comparisons.

Patient-reported symptoms and clinician diagnoses also provide complementary information. A future study that combines them can investigate experiences that never receive a diagnosis code while retaining clinically meaningful outcomes. It can also examine whether increased monitoring changes detection rates.

Rare outcomes require especially careful interpretation. A study can be large overall but still contain few events in a particular subgroup or time window. The resulting estimate may be unstable even when the total sample looks impressive. Reporting the event count and uncertainty helps readers see that limitation.

Research should also distinguish several different decisions: whether to start treatment, how to monitor it, what to do after symptoms emerge, and what happens after stopping. Evidence answering the first decision does not automatically answer the others.

These are methodological priorities inferred from the limitations discussed above, not claims that any single design would eliminate uncertainty. The aim is to make future evidence more directly relevant to decisions patients and clinicians actually face. The practical care plan cannot wait for a perfect study, but its uncertainty can be stated honestly.

Frequently asked questions

A practical framework for shared decisions

A prescribing conversation can be organized around four decisions: what benefit the treatment is expected to provide, what relevant history needs to be considered, how changes will be detected, and what happens if a concern emerges. This framework makes the discussion concrete without pretending to calculate a perfect individual probability.

The benefit discussion should describe the actual acne problem. A person with continuing scarring, repeated treatment failure, or major social impairment may weigh the course differently from someone whose acne is controlled by another option. The psychiatric discussion belongs within that overall treatment decision rather than appearing as an isolated warning with no context.

The history discussion should distinguish an old resolved episode from active symptoms and ongoing care. The monitoring discussion should specify the review process. The response discussion should establish who makes the next decision and how quickly the patient can obtain advice.

The framework avoids two unhelpful extremes. One treats the consent process as a list of frightening possibilities without a plan. The other treats reassurance as sufficient and leaves the patient unsure what to report. A useful discussion provides both proportion and action.

Patients can also ask how decisions will be documented. If several services are involved, a short written plan can prevent contradictory instructions. Documentation is particularly helpful when a concern arises outside the usual clinician's working hours and another professional needs to understand the course quickly.

What uncertainty should sound like in plain language

Uncertainty is easier to manage when it is expressed precisely. "We do not know anything" is inaccurate when large studies exist. "The evidence proves there is no possible risk" is also inaccurate. A more useful statement is: "The broad evidence is reassuring, but it cannot rule out every individual reaction, so we will assess your history and respond to changes."

Similarly, "This happened after treatment began" is a clear observation. "This was definitely caused by treatment" is a causal conclusion. The first statement can justify prompt assessment without prematurely committing to the second. Patients should not feel they need to prove the causal conclusion before being taken seriously.

The same standard applies when a clinician considers another explanation. Saying that acne itself or a life event may contribute should open the assessment, not close it. Multiple contributors can coexist. A person can experience distress related to acne and a possible medication effect at the same time; the presence of one does not logically exclude the other.

The aim is an honest account with an actionable plan. Research establishes what is known about groups. A clinical assessment addresses the person's current situation. Clear language connects the two while preserving the limits of each. That is how apparently conflicting research becomes useful information rather than a contest between reassuring and alarming headlines.

References

1. Li C, et al. Use of isotretinoin and risk of depression in patients with acne: a systematic review and meta-analysis. BMJ Open. 2019. https://pubmed.ncbi.nlm.nih.gov/30670500/
2. Kridin K, Ludwig RJ. Isotretinoin and the risk of psychiatric disturbances: A global study shedding new light on a debatable story. Journal of the American Academy of Dermatology. 2023. https://pubmed.ncbi.nlm.nih.gov/36273659/
3. Supplementary diagnosis-code definitions for the global psychiatric cohort. https://data.mendeley.com/datasets/ywz6dgr6rz/1
4. Paljarvi T, et al. Isotretinoin and adverse neuropsychiatric outcomes: retrospective cohort study using routine data. British Journal of Dermatology. 2022. https://pubmed.ncbi.nlm.nih.gov/35174880/
5. Tan NKW, et al. Risk of Suicide and Psychiatric Disorders Among Isotretinoin Users: A Meta-Analysis. JAMA Dermatology. 2024;160:54-62. https://jamanetwork.com/journals/jamadermatology/fullarticle/2812525
6. Sundstrom A, et al. Association of suicide attempts with acne and treatment with isotretinoin: retrospective Swedish cohort study. BMJ. 2010;341:c5812. https://www.bmj.com/content/341/bmj.c5812
7. Mining and analysis of adverse event signals of isotretinoin based on the real-world data of FAERS database. Frontiers in Medicine. 2025. https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2025.1559972/full
8. FDA. Adverse event reporting dashboard: limitations of reported data. https://www.fda.gov/drugs/fdas-adverse-event-reporting-system-faers/fda-adverse-event-reporting-system-faers-public-dashboard
9. FDA. Data Mining at the Center for Biologics Evaluation and Research. https://www.fda.gov/science-research/data-mining/data-mining-center-biologics-evaluation-and-research
10. Isotretinoin capsules prescribing information, revised August 2026. https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=1cf11710-f966-4529-8e08-02175f588bca
11. MHRA. Isotretinoin: changes to prescribing guidance and additional risk minimisation measures. January 22, 2026. https://www.gov.uk/drug-safety-update/isotretinoin-changes-to-prescribing-guidance-and-additional-risk-minimisation-measures