Thread by @Garth_Nicholas1

19 Nov: Ipi-nivo-meso, stratified randomization

For Lung Cancer Awareness Month #LCAM I’m going to review 30 important lung cancer trials in 30 days. These posts are directed at non-medical professionals, with descriptions of the results and of what makes a good trial.#lcsm 1/18

We previously discussed a mesothelioma trial on 8 November. On 15 November I also briefly alluded to a study that established platinum/pemetrexed as the standard chemotherapy, in 2004. Today’s study expands the use of immunotherapy into this disease. 2/18

All of our previous immunotherapy trials (13, 16 Nov) have been about pembrolizumab, an antibody to PD-1. This trial uses the similar antibody, nivolumab, along with a second antibody, ipilimumab, which targets a different part of the immune response, called CTLA-4. 3/18

Combination immunotherapy has more side effects, notably diarrhea, rash, and pneumonitis. In some cancers, such as melanoma, the increased toxicity is balanced by improved outcomes with the combination. In others (like lung cancer) the combination doesn’t seem that much better.

In this study, people with previously untreated mesothelioma were randomized to either standard chemo or ipi/nivo immunotherapy. Randomization was stratified by gender and histology (more on this below). Primary endpoint was overall survival. 5/18

Overall survival was improved in the ipi/nivo group compared to chemo (median 14.1 vs 18.1 months, p=0.002). Note, however that survival was slightly better in the chemotherapy arm for the first 4 months or so. 6/18

This is more notable if we look at PFS. The chemo arm looks better early, they cross at 8 months, and ipi/nivo does better after that. This illustrates a pattern of response to immunotherapy: often taking a few months to benefit, then sometimes sustaining benefit for a long time.

Another noteworthy finding came when looking at the outcomes by histologic types. Mesothelioma is often divided into epithelioid and non-epithelioid (or sarcomatoid), with the sarcomatoid type having a worse prognosis. 8/18

In this trial most of the benefit was seen in the sarcomatoid group (panel C below). This wasn’t because they did better with immunotherapy than the epithelioid group (panel B), but rather they did so much worse with chemo, making the degree of improvement with ipi/nivo striking.

The toxicity of the combined immunotherapy was comparable or greater than chemotherapy. For people with sarcomatoid mesothelioma it’s an easy choice, but for those with epithelioid histology the two seem quite similar, with chemo being less toxic, cheaper, and of shorter duration

One wonders if a combination of chemotherapy and immunotherapy could see patients through the initial dip during the first few months when they are not yet responding to immunotherapy. This is the subject of ongoing trials. 11/18

Many of the studies we’ve looked at make use of stratified randomization. Simple randomization would be like flipping a coin to assign each subject to a study arm. This leaves to chance, however, the possibility that the groups could be imbalanced in some variable. 12/18

For some variables that doesn’t matter. No one cares if one group averages two inches taller than the other group. But imbalances in some variables might make final study results hard (or impossible) to interpret. 13/18

For instance, in this study it was known that sarcomatoid patients would do worse with chemo. If these patients were overrepresented in the chemo arm, then it would perform poorly, and it would make it harder to understand the relative performance of the ipi/nivo arm. 14/18

Rather than take this risk, investigators will stratify randomization on critical variables. Let’s call the arms of a study A and B. The trial statistician will generate random blocks of 4, 6, or 8 assignments for each strata, each block containing an equal number of As and Bs.

As people are enrolled each is assigned in the sequence corresponding to their histology. The first sarcomatoid patient will be arm A, the 2nd arm B, etc. This assures that when the trial ends there will be almost equal numbers of epithelioid and sarcomatoid in each arm. 16/18

Stratified randomization carries a statistical price in the need for slightly increased sample size. It also means that Cox proportional hazards models have to be used for analysis of stratified data, and these models are where Hazard Ratios come from (see 13 November) 17/18

Starting tomorrow I’ve got a two-fer of lung cancer screening studies. I hope you’ll join me again! 18/18