Ganciclovir suppressed CMV reactivation in sepsis and mortality doubled

What was tried

NCT04706507 tested whether suppressing cytomegalovirus reactivation improves outcomes in critically ill adults who are not immunosuppressed. Fred Hutchinson Cancer Center ran it with the National Heart, Lung, and Blood Institute, under principal investigator Michael Boeckh. The design was randomized, triple-masked and placebo-controlled. Enrollment opened on 29 June 2021 and reached 205 participants, 106 on ganciclovir and 99 on placebo.

Participants were CMV IgG seropositive, in an ICU with acute respiratory failure expected to last at least two more days, with an infection judged to be the source and meeting the Sepsis-3 definition. Immunosuppression was an exclusion. The intervention was intravenous ganciclovir at 5 mg/kg twice daily for five days, then once daily until discharge, against matching saline. The primary endpoint was respiratory-support-free days through day 28.

Ganciclovir is a nucleoside analogue, ChEMBL CHEMBL182, maximum phase 4, curated as a herpesvirus DNA polymerase inhibitor. In cytomegalovirus that polymerase is encoded by UL54. Results were posted on 12 December 2025, and the record was updated on 20 October 2026 as TERMINATED, reason given as "due to increased mortality and futility analyses".

The biological hypothesis

Roughly a third of latently infected, non-immunosuppressed hosts reactivate CMV during critical illness, and reactivation tracks with worse outcomes. Mansfield and colleagues reviewed that literature in 2015 and were explicit that causality was unsettled. The trial tested the causal reading: reactivated CMV drives lung injury in sepsis, so blocking replication should shorten time on respiratory support. Forster and colleagues had supplied the animal support in 2009, showing that a septic trigger reactivated murine CMV, that ganciclovir suppressed it, and that the 10 mg/kg/day schedule this trial used minimised pulmonary fibrosis.

The clinical predecessor was the Phase 2 GRAIL trial, NCT01335932, reported by Limaye and colleagues in 2017 in 160 patients. Its primary endpoint, change in plasma IL-6 from day 1 to 14, was null, with a between-group difference of 0 log10 units, 95% CI -0.3 to 0.3. Ganciclovir did cut reactivation, 12 percent (10 of 84) against 39 percent (28 of 72), p less than 0.001. Median ventilator-free days favoured it as a secondary outcome, 23 against 20 overall at p equal to 0.05 and in a prespecified sepsis subgroup at p equal to 0.03. Mortality did not differ. The Phase 3 was built on that subgroup signal.

What actually happened

The drug did what it was designed to do. CMV reactivation by day 28 occurred in 5 of 85 ganciclovir recipients and 20 of 77 placebo recipients, p less than 0.001 by Fine-Gray competing risks regression.

Every clinical endpoint moved the other way. Mean respiratory-support-free days were 11.5 on ganciclovir (n=105) against 13.7 on placebo (n=96), p equal to 0.166. Ventilator-free days were 11.1 against 14.1, p equal to 0.097. ICU-free days were 10.3 against 12.8, p equal to 0.091. None reached significance, and every point estimate favoured placebo.

Mortality separated clearly. By day 28, 34 of 106 ganciclovir recipients had died against 13 of 99 on placebo, hazard ratio 2.65, 95% CI 1.40 to 5.02, p equal to 0.003. By day 180 the counts were 45 of 106 against 23 of 99, hazard ratio 2.11, 95% CI 1.28 to 3.50, p equal to 0.004. Computed from those counts, 180-day mortality was 42.5 percent against 23.2 percent, a difference of 19.2 percentage points, or one extra death per 5.2 patients treated. The posted adverse event module records zero serious events in both arms, which cannot be reconciled with 68 deaths and reflects an incomplete submission.

Failure mechanism, best guess

The archetype is a safety signal, and the mechanism is most consistent with CMV reactivation being a marker of illness severity rather than a cause of it. Gatto and colleagues showed this in 2022 across 431 critically ill COVID-19 patients. Reactivation occurred in 88 of 431 (20.4 percent), unadjusted hospital mortality was 67.0 percent with reactivation against 24.5 percent without, and after adjustment the association vanished, hazard ratio 1.141, 95% CI 0.757 to 1.721. Severity and secondary bacterial infection predicted reactivation. Treating a marker does not move the outcome the marker reports on.

The excess deaths need a second explanation, and the drug's own toxicity is the leading candidate. openFDA FAERS returns 10,641 serious reports naming ganciclovir, with neutropenia (611) and pancytopenia (569) among the leading reaction terms. Myelosuppression in a septic, neutrophil-dependent population is a plausible route from target engagement to harm. The posted results give no cytopenia or secondary infection rates by arm, so this stays a hypothesis.

Enrollment compounded both problems. Fernández and colleagues noted in 2025 that reactivation is not CMV disease, and that qPCR cannot separate shedding from tissue-invasive infection. Enrolling on serostatus meant treating many patients for a virus that would never have harmed them.

How to prevent this next time

Endpoint-level dispersion is unavailable for both trials, and the Phase 2 reported medians where the Phase 3 reported means, so no defensible Bayesian recalculation is possible. The usable levers are qualitative.

Base-rate discipline comes first. The Phase 2 primary endpoint was null, and the signal that justified a Phase 3 was a secondary endpoint in a subgroup of 160 patients at p equal to 0.03. That is close to the weakest evidence that routinely gets promoted, and the Phase 3 reversed its direction.

Biomarker enrichment on disease rather than exposure comes second. Serostatus selects everyone who has ever met CMV. Enrolling on quantitative viral load above a defined threshold would have concentrated any real effect and spared those with nothing to gain.

A pre-launch red team comes third. The observational literature already held adjusted analyses in which the mortality association attenuated toward the null. A review tasked with arguing that reactivation is a severity marker would have found the central risk and asked for mortality as a co-primary endpoint.

The single highest leverage change would have been to require a Phase 2 that met its own primary endpoint before committing a 205-patient Phase 3 to a secondary subgroup signal.

What this means for similar programs

Any program whose rationale rests on an infection or biomarker that correlates with poor outcomes in critical illness inherits this problem. Reactivating herpesviruses, translocating gut bacteria and rising inflammatory markers all appear in sicker patients, and suppressing them is not the same as treating what made those patients sick. The test that separates the two is whether an adjusted analysis preserves the association, and it was available before this trial began. Target engagement validates the molecule, not the hypothesis behind it.

Open questions

Were the excess deaths concentrated in patients who developed cytopenias, and did secondary infection rates differ between arms? The posted results do not say, and that analysis would separate drug toxicity from an inert drug in a dying population.

Did the direction of effect vary by baseline viral load? The trial enrolled on serostatus but measured reactivation, so a stratified analysis is possible from data already collected.

Is there a narrower population, defined by tissue-invasive CMV disease, in which the drug helps? This trial cannot say.

Sources

1. Journal articles below were retrieved via PubMed. All sources accessed 21 October 2026. - ClinicalTrials.gov, NCT04706507 record and posted results. https://clinicaltrials.gov/study/NCT04706507 - Limaye AP, et al. JAMA 318(8), 2017, 731-740. https://- Gatto I, et al. Intensive Care Medicine 48(6), 2022, 706-713. https://- Mansfield S, et al. Medical Microbiology and Immunology 204(3), 2015, 431-437. https://- Forster MR, et al. Antiviral Research 85(3), 2009, 496-503. https://- Fernández S, et al. Intensive Care Medicine 51(1), 2025, 39-61. https://- ChEMBL, molecule CHEMBL182 and its mechanism against target CHEMBL1872. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL182/ - openFDA drug adverse event endpoint, counts for ganciclovir. https://open.fda.gov/apis/drug/event/.