Asia stymies Columvi's expansion

The potentially confirmatory Starglo trial of Roche's Columvi looked like a relatively innocuous item in this week's US advisory committee meeting, but it has proved highly controversial. On Tuesday the adcom voted against the Starglo results, leaving Columvi's accelerated approval for third-line B-cell lymphoma hanging in the balance.

The reason is that nearly half of Starglo's patient population was Asian, versus just 9% from North America. This has been problematic before, in studies of China-developed anti-PD-(L)1 drugs, but here there was an added issue: survival data differed wildly between the two regions, showing that the Asian population drove an overall positive result, while Americans did especially badly. 

That said, the possibility of Columvi's accelerated nod being rescinded as a result seems low, especially given that at least one ongoing phase 3 study, the first-line Skyglo trial, might be capable of confirming the approval.

Hero to zero?

Before the adcom was convened Starglo, concerning Columvi plus chemo in second-line diffuse large B-cell lymphoma, looked surprisingly positive: data at last year's EHA conference showed a 38% reduction in risk of death on its OS primary endpoint.

However, Roche omitted to say whether it thought the result had clinical relevance, and doubts intensified with the revelation of the adcom's key voting question: are the Starglo population and trial results applicable to the proposed US patient population? 

Subgroup analyses provided by the FDA painted a damning picture, not only in the fact that just 25 of Starglo's 274 patients were from North America, but also because these subjects did numerically worse on Columvi plus chemo than on Rituxan plus chemo in terms of OS and PFS.

The larger subgroup of all 113 non-Asia patients yielded a 1.06 hazard ratio for OS. However, among just European participants median OS was 21.2 months for Columvi plus chemo versus 13.8 months for Rituxan plus chemo – still statistically non-significant, but enough for the drug to secure EU approval last month.

Selected Starglo data (Columvi + chemo vs Rituxan + chemo) in 2nd-line DLBCL

Source: FDA & Roche.

While small patient numbers suggest that the US result is a freak finding, the FDA's Dr Margret Merino told the adcom: "If the OS results in the non-Asian region were considered true, this could indicate a signal for harm for patients in non-Asian regions."

Curiously, Starglo's protocol specified a minimum enrolment target for China, but not for other regions. Roche issued a statement arguing that the Starglo population's "clinical and disease characteristics ... are representative and applicable to US patients". It didn't point out the negative adcom vote, merely saying a decision on second-line approval was expected by 20 July.

Shifting stance

FDA reviewers disclosed that Roche's first explanation for Starglo's emerging regional divergence was a lag in US enrolment. Then, after the inconsistency persisted with longer follow-up, Roche's case switched to suggesting regional differences in available subsequent therapy for progressing patients.

However, this would only have scuppered the OS result. Yet Starglo's PFS and response rate data were similarly inconsistent in the North American versus Asian populations. 

At the adcom Roche argued that Starglo wasn't powered to detect subgroup differences, and cited a multivariable model of the intent-to-treat population showing data consistency. Charles Fuchs, its head of oncology and haematology drug development, also pointed out that the 27.2-month median OS for control in ex-Asia patients was well above 8-13.5 months in historical numbers.

The adcom wasn't asked whether Starglo should be used to convert Columvi's accelerated to a full approval, but on the specific voting question panellists voted eight to one that its result wasn't applicable to a US population.

The likely outcome is that Columvi's second-line label is off the table, but the drug looks set to remain in accelerated limbo in the third line.