On 2 September FDA published a draft guidance titled Pharmacokinetics in Patients with Impaired Hepatic Function: Study Design, Data Analysis, and Impact on Dosage, and in the same notice withdrew the May 2003 final guidance it replaces. Comments are due by 1 December under docket FDA-2026-D-8693. The 2003 document was one of the longest-serving clinical pharmacology guidances still in force, and a good number of the hepatic impairment protocols on ClinicalTrials.gov cite it by name, which as of this month means they cite a withdrawn document. That is a small instance of a provenance problem we wrote about in July, though it is not what we noticed first.
What we noticed first was the title. The 2003 guidance was "Study Design, Data Analysis, and Impact on Dosing and Labeling." In the Federal Register notice, and on the cover of the PDF, the 2026 draft is "Study Design, Data Analysis, and Impact on Dosage." (FDA's guidance search page was still listing the new draft under the old title when this was written, so the notice is the reference to cite.) Labeling has been dropped from the name, and the labeling section inside the document has shrunk to four bullet points. We do not read that as FDA caring less about the label. Having gone through the whole draft, what has happened is that the evidentiary work behind the label is now spelled out upstream, in the protocol and the study report, in considerably more detail than the 2003 guidance ever asked for, and the label section is short because it is meant to summarise work that is fully written up elsewhere.
What the 2003 guidance did for a labeling writer
The old guidance was, among other things, a source of sentences. It told sponsors that where no dedicated study was warranted, labeling should say the impact of hepatic impairment had not been studied and that effects requiring a dosage adjustment were unlikely. It described a reduced study design, comparing subjects with normal hepatic function to a single moderate impairment group, and it set out what the label could then say: no adjustment for mild or moderate impairment, and caution for severe. FDA clinical pharmacologists have presented that decision tree in training material with the severe-impairment branch ending in the words "use with caution in Child-Pugh C."
That shortcut helped support a good deal of nonspecific labeling. A 2013 review in the Journal of Clinical Pharmacology by Chang and colleagues looked at 157 new molecular entities approved between 2004 and 2011 and found that 27 percent of the labels examined said only to use the drug "with caution" in liver disease, with no specific dosing recommendation, and that specific information on hepatic metabolism was missing from around 90 percent of them. The authors concluded that the guidance itself needed updating, and thirteen years on, this draft is the update.
The four bullets
Section VI of the new draft, in full, says the prescribing information should include "a summary of essential scientific information needed for the safe and effective use of the drug in patients with HI," and then lists four items: pharmacokinetics and pharmacodynamics in patients with hepatic impairment, information on hepatic elimination of the drug and relevant active metabolites, the clinical effects of the pharmacokinetic or pharmacodynamic changes, and "recommendations to prevent, mitigate, monitor for, or manage risks in patients with HI (e.g., different dosage or monitoring recommendations)."
There is no model language anywhere in the section, and nothing to reach for when the study was not done. What the bullets ask for instead is an account of the elimination pathway, which happens to be what the 2013 review found missing from nine labels in ten, together with a recommendation the prescriber can act on. A standalone "use with caution" sentence, with nothing on elimination, exposure change, clinical effect or what to do about it, would not meet the spirit of those four elements. That is the same editorial direction FDA gave for Section 7 in its drug interaction labeling draft, where "avoid concomitant use" is preferred over "use with caution", applied here to Sections 2, 8.7 and 12.3 without spelling it out.
The old moderate-impairment shortcut is no longer the emphasised design
Section IV.B of the draft says the sponsor "should conduct a study in subjects with mild, moderate, and severe HI, as well as a control group without HI." The single-group moderate impairment study that the 2003 guidance allowed does not appear, and the alternative the draft does offer runs the other way: "a study design that only investigates the effect of severe HI could be a reasonable approach to begin characterization," and if a significant effect is seen in severe impairment, "FDA recommends that the other HI subpopulations be studied."
For anyone who drafts labels, this shifts where the evidence gap sits. The moderate-only design tended to leave a label silent on severe impairment because severe impairment had not been studied, and the 2003 wording accommodated that. Under the draft, severe impairment moves from being the group most likely to remain unstudied to the group FDA may want characterised first, rather than one routinely left as "not studied." The label can only describe the gap that development left it, and this moves the gap earlier.
What the study report now has to say
This is where the draft gets specific in a way the 2003 guidance was not, and every item on the list is a sentence someone has to write.
On classification, the draft says "the study report should provide both the total CP score and the values of each of the five components as well as a clear description of how ascites and hepatic encephalopathy severity or grade were determined, including the condition of assessment during scoring (e.g., ascites severity assessment with or without diuretic use)." Where a sponsor uses an alternative score such as MELD or ALBI, "the CP classification should also be assessed, and results should be reported for all classification methods." A subject whose severity changed between screening and dosing "should be included in the severity group identified closest to pharmacokinetic sampling," which is a population assignment rule that the report has to apply and state, and it is the kind of rule where a correct number can still sit under the wrong population heading if nobody checks.
On dose selection, "the rationale for the dosage selection should be included in the study protocol and study report." On protein binding, the draft asks for unbound concentrations for highly extracted, highly bound drugs, and adds that "a rationale for not measuring unbound drug concentrations should be included in the study protocol and study report," with parameters "reported in terms of both unbound and total concentrations of drug, where applicable, in the study report."
On analysis, the sponsor "should provide the geometric and arithmetic means and measures of variability for pharmacokinetic parameters, the point estimate for the ratio of the means of the impaired and control groups for total and unbound AUC and Cmax, and the associated 90 percent confidence interval," and for the regression of pharmacokinetic parameters against the components of the classification system, "the sponsor should calculate, and report estimates of the parameters of the chosen model as well as measures of their precision." The etiology of the underlying liver disease is to be collected and reported. The sample size needs a written justification, and the draft offers one form it could take.
Two of those items say "protocol and study report" together, and we read that as a consistency requirement as much as a content one. The rationale that appears in the report has to be the rationale the protocol gave, which is straightforward if the report is drafted with the protocol open and easy to get wrong if it is drafted from the analysis outputs and a memory of what the protocol said.
Pooling across studies
The draft accepts a population pharmacokinetic approach in place of a dedicated study where subjects with varying degrees of hepatic impairment are "adequately represented in clinical trials," and then sets conditions. Among them is "using the same method to classify hepatic function when data are pooled across studies," alongside accurate dosing and sampling records, adequate samples per subject, and exposure-response analyses that treat hepatic impairment as an independent predictor.
The classification rule has consequences beyond the popPK report. Child-Pugh is recommended for cirrhosis, the NCI Organ Dysfunction Working Group criteria for cancer patients except those with hepatobiliary cancers, and the draft says FDA "does not recommend the use of the NCI Criteria to assess hepatic function for non-cancer patients." A programme that pooled across phase 2 and phase 3 trials using different classification methods cannot simply harmonise in the summary. Which method each trial used is a fact recorded in each protocol and each clinical study report, and the Module 2.7.2 summary has to say what they were and whether the pooling condition was met. That is a cross-document reconciliation of the sort that a late change to an analysis population also forces, and it is cheaper to write when the classification method is a field the drafting tool can read from every study than when it is a sentence someone has to find in each one.
The dose in the label may now rest on an explicit exposure-matching argument
Section V.C says that where a different dosage is needed, "the dosage recommendations should be based on exposure-matching to a reference group with an acceptable benefit-risk profile, often the normal hepatic function group," and offers two ways to get there: simulations that project exposures "within the 5th and 95th percentiles of those achieved in the reference group," or no-effect boundaries, with a cross-reference to the ICH M12 drug interaction guidance for the latter.
Where a sponsor takes either route, the number that lands in Section 2 of the label is the output of an argument that has to be written up in the clinical pharmacology summary and referenced from 12.3, and the label bullet asking for "clinical effects of the pharmacokinetic or pharmacodynamic changes" is asking the writer to state the exposure-response reasoning that connects them. It is the same demand that FDA's meeting guidance makes of a package, that a result be described "with some degree of quantification" rather than merely called significant.
Timing, and the eligibility guidances behind it
The draft says the effect of hepatic impairment "should be assessed early in drug development to adequately guide trial eligibility and identify a recommended dosage for patients with HI," and it footnotes the July 2020 guidance on cancer trial eligibility for organ dysfunction and the December 2025 guidance on enhancing participation in clinical trials. Read with the three oncology eligibility guidances FDA finalised in July, the direction is that hepatic impairment data is meant to feed into phase 3 inclusion criteria rather than be generated after the pivotal trial to satisfy a labeling requirement. For orphan drugs the draft asks sponsors to discuss timing with the review division early, which for a small programme is another item on a calendar that designation already compresses.
One more change, to the guidance itself
The last page of the draft carries something new. A "Guidance History" table lists the September 2026 draft and the May 2003 final, with a footnote that the table "was implemented beginning August 2026, and previous guidance history may not be captured in totality." FDA guidances have not historically carried their own version history inside the document, so this is new, and for a writer citing guidance in a submission it is quietly useful: the document now records its predecessor on its own last page, which, given how often guidance changes underneath a long development programme, is the kind of provenance detail we would rather have in the source than have to reconstruct.
What to do with it
The draft is nonbinding and open for comment until December, and a final version may restore some model wording. In the meantime our practical reading is that the evidence behind hepatic impairment labeling now has to be written down more explicitly, and earlier, than most programmes are used to. The dose rationale goes in the protocol as well as the report. The report has to carry the classification details down to the individual Child-Pugh components, the binding rationale, the parameter estimates with their variability and confidence intervals, the disease etiology and the analysis choices, and any pooled analysis has to show that hepatic function was classified the same way in every study it draws on. By the time the dosage recommendation reaches Section 2 of the label it should be traceable back through the clinical pharmacology summary to the exposure-matching argument that produced it. The four label bullets are a rendering of all that, which is one more reason we keep saying the document was never the source of truth.
The draft guidance is at fda.gov/media/71311/download, which is the same URL the 2003 guidance occupied until this month.