13 November: Our 1st Keynote, HR
This year for Lung Cancer Awareness Month #LCAM I’m going to summarize 30 important lung cancer trials over 30 days. These posts are directed at non-medical professionals, with descriptions of the results and of what makes a good trial.#lcsm 1/17
This year for Lung Cancer Awareness Month #LCAM I’m going to summarize 30 important lung cancer trials over 30 days. These posts are directed at non-medical professionals, with descriptions of the results and of what makes a good trial.#lcsm 1/17
The uptake of immunotherapy in lung cancer has been rapid and widespread. In 2014 it was research only: in 2021 it is used in all types of lung malignancies, in almost all stages of disease. We will discuss at least six immunotherapy studies in the remainder of the month. 2/17
Lymphocytes are white blood cells that should lead an immune response to tumours. In some cases, tumour cells have a protein on the surface called PD-L1 that binds to a protein called PD-1 on lymphocytes, and serves to “turn off” that part of the immune response. 3/17
Our commonest immunotherapies are antibodies that bind to either PD-1 or PD-L1 to inhibit this interaction. Once freed from this downregulation, the immune system can cause profound and sometimes long-lasting decrease in tumour size. 4/17
This trial randomized people with non-squamous metastatic NSCLC (without EGFR or ALK) to one such antibody, pembrolizumab, or placebo. All patients got standard chemo with platinum-pemetrexed (see 11 Nov). Responding patients could remain on treatment for up to two years. 5/17
There were 2 primary endpoints, PFS and OS (tested sequentially p=0.025), and enough patients to have 90% power to detect a PFS hazard ratio of 0.7 (more on this below). The trial was stopped at the first interim analysis due to the already apparent difference between the arms.
Total enrollment was 616, two thirds randomized to pembrolizumab, one third to placebo (plus chemo for all).
Median PFS was improved from 4.9 to 8.8 months (p<0.001). 7/17
Median PFS was improved from 4.9 to 8.8 months (p<0.001). 7/17
Overall survival was also significantly improved, as shown below.
Survival was improved in all subgroups, including all levels of PD-L1 expression. We will talk more about PD-L1 expression in three days. 8/17
Survival was improved in all subgroups, including all levels of PD-L1 expression. We will talk more about PD-L1 expression in three days. 8/17
This trial established platinum-pem-pembro as the first-line treatment of choice for metastatic non-squamous NSCLC.
Remember back to November 1 and the median overall survival of 17 weeks in that study. Look at this study: 22 months median survival. That’s 30 years of progress!
Remember back to November 1 and the median overall survival of 17 weeks in that study. Look at this study: 22 months median survival. That’s 30 years of progress!
Our methodology concept today is the Hazard Ratio (HR). This is a complex and frequently misunderstood and misapplied statistic. Despite this, hazard ratios seem to be increasingly featured in clinical trial reports, usually in favour of easier to understand statistics. 10/17
First, what the HR isn’t. Look at the KM curve above. HR of 0.56 is often interpreted as a “44% reduction in the risk of death”. But it is not. Check for yourself: at none of the described points (median, 1 yr, 2 yr) is the control arm survival 44% lower than the pembro arm.
What is a hazard? It is the probability an event will happen at time x, given that it had not happened earlier. If the event is progression of cancer, the hazard of progression on day 100 is the probability of it happening that day, given that it had not happened on days 1-99.
The hazard on any day is pretty small, but all these little hazards add up to give the shape of the survival curve. If the hazard is estimated at 2% in the standard arm and 1% in the experimental arm, the hazard ratio would be 0.5. 13/17
So an HR is something like the ratio of an estimated daily event rate in the two arms. That doesn’t sound very useful or intuitive. So why do we use them? 14/17
Advantages of HRs
1. They give a measure that describes the whole curve, not just one point (eg median)
2. They are easy to create: they're the natural output of common statistical methods called Cox Hazard Models
3. Give a single number to compare treatment effects across trials
1. They give a measure that describes the whole curve, not just one point (eg median)
2. They are easy to create: they're the natural output of common statistical methods called Cox Hazard Models
3. Give a single number to compare treatment effects across trials
The disadvantages of HRs is that they are quite abstract, as described above.
How should doctors communicate these abstract numbers to patients? My sense is that we probably just shouldn’t. Medians or landmark survivals (% at 1- or 2-years) are more understandable and relatable.
How should doctors communicate these abstract numbers to patients? My sense is that we probably just shouldn’t. Medians or landmark survivals (% at 1- or 2-years) are more understandable and relatable.
Tomorrow we’ll use our knowledge of adjuvant therapy (Nov 2), selection of endpoints (6 November) , and cross-over (7 November), among other concepts, to understand the discussion of a recent controversial trial. 17/17
• • •
Missing some Tweet in this thread? You can try to
force a refresh
