On Jan. 8, the American Cancer Society published its annual estimates of new cancer cases and deaths, declaring that the latest data—from 2016 to 2017—show the “largest ever single-year drop in overall cancer mortality of 2.2%.”
Much hoopla followed.
The next day, President Donald Trump tweeted: “U.S. Cancer Death Rate Lowest In Recorded History! A lot of good news coming out of this Administration.” Beyond the White House, too, the biggest questions revolved around credit, as medical oncologists and cancer epidemiologists pondered the 2.2% drop.
“Is this a blip in the data?” some asked. Or, perhaps, is this a sign that survival benefits and delay in progression from new drugs for lung cancer—targeted therapies and immunotherapies, etc.— are finally starting to show up as a mortality drop on the population level?
One thing was beyond dispute: the overall drop in mortality was driven by an unprecedented decline in lung cancer deaths. Since lung cancer continues to cause more deaths than breast, prostate, colorectal, and brain cancers combined, the single-year drop in lung cancer mortality is responsible for 0.8% of the decrease in overall mortality, accounting for more than a third of the total 2.2% decline.
“Immunotherapy is showing such a dramatic impact in the treatment of locally advanced and advanced non-small cell lung cancer that this effect elevates the statistics for all lung cancer and—this I find astonishing—you can even see its effect in age-adjusted cancer mortality overall,” opined Otis Brawley, the Bloomberg Distinguished Professor of Oncology and Epidemiology at Johns Hopkins University (The Cancer Letter, Jan. 8).
In conversations with The Cancer Letter, key cancer epidemiologists interpreting the data agree that Brawley’s interpretation could well hold up—with the caveat that more data and further analyses are needed.
“We saw this acceleration in the pace of the decline in lung cancer mortality just within the past decade,” Rebecca Siegel, scientific director of surveillance research at ACS, said to The Cancer Letter. “From 2% per year decline, to 4% per year decline. So, 10 years ago, lung cancer mortality was declining by 2% per year, but in the past five years, it declined by 4% per year.
“That’s because of improvements in treatment as well as the continued declines in incidence. But now, the mortality declines are outpacing incidence, because there’s been improvements in survival.”
A conversation with Siegel appears here.
While more data are needed to test the hypothesis, preliminary calculations show a dramatic increase in the use of immuno-oncology agents in recent years, said Lynne Penberthy, associate director of NCI’s Surveillance Research Program, which supports the institute’s Surveillance, Epidemiology, and End Results (SEER) Program.
“We see increased survival in the last few years for lung cancer,” Penberthy said to The Cancer Letter. “And particularly, we see slightly greater increase in survival for non-small cell lung cancer versus small cell lung cancer. That is consistent with the ACS hypothesis—although, we always like to see a longer trend.”
A conversation Penberthy appears here.
Survival → Mortality
How do these improvements in survival translate into a reduction in mortality in lung cancer? With the rate of screening remaining relatively low—studies suggest that fewer than 2% of eligible current and former smokers are getting screened—changes in survival may have a more direct impact on lung cancer mortality.
“The relationship between survival and mortality is quite complex when screening is involved,” said Eric “Rocky” Feuer, branch chief of the NCI’s Surveillance Research Program. “Because of lead-time bias, survival could improve with no improvement in population mortality. If we assume that for lung cancer that screening rates are very low, then the relationship between survival and mortality is more straightforward—but still not completely straightforward.”
Feuer is the overall project scientist for the institute’s Cancer Intervention and Surveillance Modeling Network (CISNET), a collaborative group of investigators who use simulation modeling to guide public health research and priorities. In particular, CISNET models excel at studying latency, which measures the time lapse between the initiation of an intervention and its effect on mortality.
“If incidence has been declining at the exact same rate for many years and the composition of the cases (in terms of histology and stage) stays constant, and survival (for each stage and histology) has not been improving at all for many years, then incidence and mortality should be declining at the same rate,” Feuer said to The Cancer Letter. “However, these exact conditions are never precisely met. Because the rate of decline of incidence varies somewhat, the composition of cases changes, because of changing smoking habits and other factors—and diagnostic methods change. When these conditions are not met, mortality declines must be offset from incidence by an appropriate amount in accordance with survival by stage and histology.
“Of course, on top of all this, when survival is improving (by stage and histology), then mortality declines will be larger,” Feuer said. “Under the ideal conditions mentioned above—i.e. incidence has been declining at the exact same rate for many years and the composition of the cases (in terms of histology and stage) stays constant, and survival (for each stage and histology) has not been improving at all for many years—then the introduction of a new improved treatment should cause a clear Joinpoint in mortality with a faster decline.
“We never have these ideal conditions, and that is where modeling really helps. Under flat survival, the model allows us to translate incidence trends (which may be complex) into mortality trends (Model run #1). We then add improvements in survival on top of the incidence trends, and see how much mortality declines over and above run #1).”
The role of treatment in the decrease in lung cancer mortality will become clearer over time and would be teased out through modeling. CISNET and affiliated investigators are working on unpacking lung cancer statistics, as they have already done for breast cancer, Feuer said.
“While the decline in the mortality rate from 2016 to 2017 seems impressive and important, I would rather not spend too much time trying to interpret a drop from one year to the next,” Feuer said. “Instead, I prefer to interpret a change in trend as measured by a statistical model like Joinpoint [which enables the user to test that an apparent change in trend is statistically significant].”
Drug treatment effect?
Skeptics caution against linking lung cancer mortality with drug treatment effect at this time, arguing that:
Mortality is dropping largely because of aggressive smoking cessation efforts. For many decades, lung cancer incidence trends have been considered a proxy for lung cancer mortality trends, because there were so few effective treatments. Models continue to project considerable decreases in lung cancer mortality and the overall number of deaths over the next 50 years—largely because of the lower smoking initiation rates in current cohorts, leading to lower smoking prevalence at older ages, when lung cancer risk is greatest.
Screening is likely to have some impact on mortality rates, even with low uptake. The results of the National Lung Screening Trial—demonstrating that low-dose computed tomography screening has the potential to reduce risk of death by 20%—were published in 2011, leading to a 2013 recommendation by the U.S. Preventive Services Task Force in favor of low-dose CT screening, and a 2014 coverage decision by the Centers for Medicare and Medicaid Services.
It’s too soon to observe a signal from new drugs on the population level. Targeted therapies and immunotherapy agents for non-small cell lung cancer in particular were approved within the past decade. For instance, the EGFR-targeted therapies joined the NSCLC toolkit in 2013, and the first PD-1 checkpoint inhibitor didn’t show up until 2015. Thus, the databases used by ACS include only two years of data on checkpoint inhibitors and about four years for EGFR-targeted therapies.
Improvements in drug treatments largely benefit only a subset of lung cancer patients. The majority of drugs approved by FDA for the treatment of lung cancer over the past two decades were indicated for locally advanced or metastatic non-small cell lung cancer. The survival rates for lung cancer, overall, remain low. According to 2009-2015 data from SEER, 24% of NSCLC patients are alive five years after diagnosis vs. 6% of SCLC patients. For distant, metastatic disease, the five-year survival rates are 6% for NSCLC and 3% for SCLC.
While the decline in the mortality rate from 2016 to 2017 seems impressive and important, I would rather not spend too much time trying to interpret a drop from one year to the next. Instead, I prefer to interpret a change in trend as measured by a statistical model.
Eric “Rocky” Feuer
Since more than three quarters of patients with stage IV lung cancer—regardless of histologic subtype—die within two years, experts say it’s possible that these improvements in survival rates may not translate into a significant, sustained decline in mortality rates in long-term data.
“One of the reasons we need to look at the longer-term data and continue to monitor trends is that if the treatments are impacting mortality, they’re really increasing survival, right? It seems unlikely that new treatments are curing the disease per se,” NCI’s Penberthy said. “So, we’ll see people who live longer, which is good, but that doesn’t mean that we still don’t really need to continue to focus on prevention, smoking cessation, screening, etc. Most patients with stage IV disease aren’t going to be cured by treatment, even if it is prolonging their survival.”
That said, the survival rates have improved for lung cancer overall.
“Specifically for NSCLC, five-year relative survival has increased from 19%, in the mid-90s, to 25% for more current diagnoses,” ACS’s Siegel said. “Well, for SCLC, it’s reading steady at just 6%. There’s definitely a need for better treatment for these patients.”
Focusing on NSCLC
This begs the question: Is it possible to observe treatment-related improvements in lung cancer survival at the population-level, even if progress is largely being made in one histologic subtype, NSCLC?
The first step is to determine whether the lung cancer mortality rate is dropping faster than the incidence rate, and then characterizing the contribution of all interventions—smoking, early detection, treatment—before a study or model is able to isolate and account for the effect of drugs, said CISNET’s Feuer.
“For a highly fatal cancer like lung cancer, if incidence and mortality are declining at about the same amount per year, it generally indicates that primary prevention (e.g. tobacco control) is responsible for the decline,” Feuer said. “When the trends in incidence and mortality become discordant, then early detection through screening (secondary prevention), which can impact both incidence and mortality) and improvements in treatment (which generally only impacts mortality) may be playing an additional role.”
Since screening may only be slightly affecting the recent decline in lung cancer mortality, and although the time from the introduction of screening until it impacts lung cancer mortality is shorter than for other cancers, it may take a few years until a decline can be observed, Feuer said.
“Treatment, especially targeted therapy, seems like it may be related to the recent declines in mortality, especially since EGFR TKI was approved for stage IV NSCLC first-line therapy in 2013,” Feuer said. “Use of other new therapies, including immunotherapy, could also be responsible for a part of the decline especially depending on the degree of any off-label use prior to formal approval. But these arrived more recently, and thus, are less likely to be related to the faster decreases in mortality since 2013.”
The mortality rate for lung cancer has been falling faster than incidence for at least five years, Siegel said.
“So, if you look at the annual percent change in lung cancer mortality from 2008 to 2013, it’s 2.4% per year,” Siegel said. “But then, from 2013 to 2017, it jumps to 4.3% per year.
“I think it can be supported by the fact that survival rates for lung cancer, until around 2000, moved very little. Which is why you could look at lung cancer incidence trends, and say this is what’s going on with mortality, because few people survive very long.
“But since 2000, survival rates, we actually looked at one-year relative survival rates, and they’ve improved by about 10 percentage points since 2000, for every stage of lung cancer diagnosis.”
Not just drugs for lung cancer
Improvements in treatment interventions for lung cancer aren’t exclusive to targeted therapies or immuno-oncology drugs, Siegel said.
“It tells you that it’s better treatments, across the board, which there have been for lung cancer—improvements in surgery, surgical techniques, because of video-assisted thoracic surgery,” Siegel said. “There’s been improvements in staging for lung cancer because of integrated PET/CT imaging.
“Tyrosine kinase inhibitors, particularly, have contributed, but because you have these improvements in treatments across the board, that’s reflected in a similar improvement in survival for every stage at diagnosis.
“The percentage of patients diagnosed with localized stage lung cancer in SEER areas hasn’t changed over time. And you would expect to see the uptake of screening reflected in a shift in stage.”
This is different from, for example, melanoma, another cancer where the most rapid decline in mortality has been observed over the past five years. In melanoma, decreases in mortality are driven largely by immunotherapy—the death rate dropped by 6.4% from 2013 to 2017, and by as much as 7% per year in individuals under 65.
“For melanoma, there’s only a jump in survival in recent years for metastatic disease,” Siegel said. “And that is because of immunotherapy. Or rather, that suggests, we can’t say for sure again, because this is ecologic data, but it strongly suggests.
“That suggests that it is because of these drugs that were approved by the FDA in 2011 to treat metastatic disease.”
Also, it’s important to note that many of the newer drugs in lung cancer received accelerated approval from FDA based on response rate and their ability to delay progression, as opposed to overall survival, which in many cases was subsequently demonstrated in confirmatory trials.
In melanoma, nearly all the drugs that are now presumed to be driving the improvement in mortality were approved based on metrics of delay in progression—progression-free survival and recurrence-free survival.
A feast for modelers
According to Feuer, four factors make it difficult to speculate about—without actually doing the modeling—the underlying cause of the accelerated decline in lung cancer mortality, or to guess at the exact magnitude of effect attributable to targeted therapies and immuno-oncology agents:
The strong background decline in lung cancer incidence and mortality—associated with prior improvements in the initiation and cessation rates of cigarette smoking—makes interpretation of any possible additional declines in lung cancer mortality more difficult to tease apart.
To understand the size and timing of the impact of any new intervention—e.g. screening or treatment—on cancer mortality, modelers need to reconstruct when the intervention was first available, how quickly its use disseminated, how effective it is, and what is the latency between its use and impact on mortality.
New treatments—i.e. targeted therapy and immunotherapy—mostly have impacted only metastatic NSCLC through 2017, but the mortality trend includes all histologies and stages of disease.
There are usually synergies between the intervention types—tobacco control, screening, and treatment—making it difficult to ascribe the cause of a decline to any single factor. For example, if screening detects more early-stage cancers, and there are treatment improvements for early stage cancer, the mortality decline associated with both interventions together may be more than the sum of each working separately.
“I want to mention that the CISNET models are able to capture and forecast the latency between the use of an intervention and its impact on mortality,” Feuer said. “Those show, generally, a large difference between latency of the three types of interventions impacting lung cancer mortality.
“Tobacco control (primary prevention), especially, declines in smoking initiation but also cessation, takes the longest—many, many years for it to influence mortality. The middle case is screening, and the fastest, especially for stage IV disease, is treatment.”
If a treatment is not curative—which is the case for many of the newer drugs approved for metastatic lung cancer—the decline in mortality may be temporary and, eventually, the rates could return close to the background status quo trend, Feuer said.
“Using modeling, we can partition the impact of each of these interventions on the decline in mortality as time passes. In observed data, we only can observe the impact of the sum total of all interventions,” Feuer said. “So, for example, if the new national Tobacco 21 initiative, increased screening rates, and advances in therapy all start influencing lung cancer mortality rates in the years to come, without modeling, it may become increasingly difficult to tease apart the contribution of each.”
The ACS paper is the first hint that something is having an effect, said Rafael Meza, associate chair and associate professor of epidemiology, co-leader of the Cancer Epidemiology and Prevention Program, and associate professor of global public health at the University of Michigan Rogel Cancer Center.
“I agree with Rocky. The faster decrease since 2013 is most likely due to treatment improvements than screening, given when these came into the picture,” Meza said to The Cancer Letter. “Nonetheless, we should see the impact of screening in the next few years, and we’ll get more data to try to tease their relative contributions in more detail.”
Meza is the coordinating principal investigator for the CISNET Lung Working Group.
“I think, hopefully, this decrease in mortality will continue as even newer treatments and screening are adopted and disseminated, and as more data becomes available, we’ll be able to pinpoint the most likely causes,” Meza said.
Modeling can allow researchers to predict, year by year, the impact of these interventions on mortality through 2030, or even 2040, Feuer said.
“As we assemble the components of each intervention—tobacco control, screening, and treatment—we can see how they play out over the short and long term,” Feuer said. “If we, or others, question some of the assumptions, we can change them and see how it modifies the picture in the future.
“Because the impact of some interventions may take a long time to influence mortality, we can appreciate the full benefits of each intervention as they play out. It should also be remembered that while lung cancer screening and treatment only impact lung cancer mortality, the impact of tobacco control is much broader, preventing not only the onset of lung cancers, but of other health conditions as well.”
Modeling requires a structured assembly of multiple components—i.e. initiation of new interventions, dissemination, and efficacy—via a systematic process that makes all the assumptions explicit, and thus, easier to judge and evaluate.
“The population impact of all these new interventions is certainly coming, but exactly when and how large the impact will be, we can’t say for sure at this point,” Feuer said. “As screening rates hopefully increase, that will play a role in mortality declines as well.
“As I always tell Rafael, it’s a good time to be a lung cancer modeler.”
