Sorry for the delay folks! Trying to move to New York while doing my offboarding for residency has eaten my time and created an unintentional cliffhanger. But, back to it!
When we last left our hero, if you’ll allow that lung cancer screening (LCS) is our hero, we had seen a fairly impressive mortality benefit and some evidence that the intervention was cost-effective. Still, we had concerns that comparing to chest X-ray instead of no screening had underestimated the amount of adverse events to which implementation of LCS on a population level could lead, and complete uncertainty as to whether the benefits in a controlled RCT environment would hold up in the real world. If you want a longer review, click on back and read.
So, on to this report from the VA! As 8 large VA centers took it on themselves to implement LCS on their high-risk populations, Linda Kinsinger and colleagues documented the relevant outcomes. This was not designed as a research protocol per se (really more of a QI project cum prospective cohort study), best illustrated by the fact that it was exempt from IRB review. Still, in practice reads enough like it that the data will feel pretty familiar.
If You’re Only Going to Read One Paragraph
This report on the implementation and first two years of a formalized program for lung cancer screening at eight large VA centers, somewhat in the style of a prospective cohort study, revealed several pitfalls in the implementation process. Despite dedicated coordinators handling the logistics, patient participation in screening was low, and lack of meaningful data on patient smoking histories very much limited their ability to assess whether patients met criteria. In patients who were screened (N = 2106), 1257 (59.7%) had nodules, of whom only 42 went on to need biopsy and 31 were found to have lung cancer (1.5% of the population screened). This is a significantly higher proportion of nodules but similar rates of cancer diagnosis to the NLST in this older, heavy smoking VA population. While uncertainty over the long-term toxicity of low-dose CT and concern for spillover of screening into lower-risk populations persist (and seem to be well-founded), there clearly is a defined population who will benefit from this intervention in a cost-effective manner. We should continue to work to define best practices for screening to maximize benefit and minimize risk.
The VA LCS Implementation
So, what does LCS look like when the VA tries to do it on a population level? Here’s what they brought to the table:
- 35 major centers volunteered, of which the 8 felt to have the highest degree of preparedness and administrative support were chosen for inclusion.
- They leveraged CPRS, the long-standing and universal electronic health record of the VA system, to mine patient data about smoking history, much as most EHR software can recommend preventative health services on the basis of charted patient characteristics. Patients were excluded for other malignancies or a documented life expectancy less than 6 months. Their primary care physicians were also given the option of excluding them based on severe chronic disease limiting their life expectancies, which led to the exclusion of 15.7% of patients otherwise eligible.
- Each center had a physician champion for the LCS program, and also hired a full-time coordinator for their efforts. So that’s 8 coordinators among a total of 4246 patients who were offered screening, which I think is a non-absurd ratio, but still a resource that will be scarce in the community.
And what did they get for their trouble? Well, the study starts with 93033 candidate patients for screening based on age and life expectancy, and then immediately loses 36555 of them due to a lack of sufficient data about their smoking history. And here we stumble across our first major limitation: The EHR can only tell you what you have already told it. If you have never recorded a smoking history, or not recorded a pack-years analysis, or not updated it since you first saw the patient 5 years ago, population screening efforts that rely on EHR to find eligible patients will be severely limited.
This leaves 56478 patients, of whom 38395 did not meet the lifetime smoking criteria from the National Lung Cancer Screening Trial (NLST). And then of the remaining 18083 patients, we lose 13048 of them because…
Wait, hold on. This is weird. “Patients were excluded from screening because they were not assessed by their clinician.” The authors have a footnote explaining that phased rollout and variable implementation of clinician reminders may contribute to this, but all that this explains is that we are dealing with a non-random exclusion of more than two thirds of the patients who were otherwise eligible for screening. So while there are still lessons to be learned from this paper, if we were ever considering calling NLST data into question on the basis of our findings here, this should be the end of that idea.
The VA patients were older, with 52.5% greater than 65 years of age compared to only 26.6% of the NLST patients. Slightly more patients were active smokers, and, shockingly, the VA study had a larger proportion of male patients.
Lastly, after 15.7% of the remaining patients were excluded on the basis of physician discretion, 4246 patients remained. Of these, 57.7% (2106) agreed to undergo screening, of whom 85.7% of patients actually completed at least one low-dose CT (LDCT). So of 4246 patients actually eligible, 2106 completed CT. That gives us an adherence rate of 49.6%, compared to 95% in the NLST. So as we point out limitations of real-life implementation, let’s note that even with a long-standing EHR and dedicated coordinators, just over half of eligible patients made it from eligibility to the scanner. Might this significantly greater falloff change the characteristics of the patients we are screening? Hard to say. Perhaps sicker patients are more likely to want screening, or alternatively are less likely to adhere. It’s a question without an answer, but we will need to keep it in mind as we look at whether LCS produces results equivalent to those in the NSLT.
Of those patients screened, 56.2% had a nodule requiring follow up CT, compared to only 27.3% in the first round of the NSLT. Despite this, only 1.5% of patients had lung cancer, compared to 3.9% in the NSLT (this is my math, not reported, but 1060 cases in 26722 patients). So, more nodules requiring follow up for overall fewer cancer diagnoses (and presumably, as such, a lower mortality benefit, if indeed one exists). Interestingly, the proportion of patients with nodules requiring follow up varied impressively between centers, from 30.7% to 85.0%, likely indicating differing practice among radiologists.
OK, So What Does This Mean for My Patients?
Given that this study was more of a fact-finding mission than an experiment designed to answer a specific question, we can’t be surprised that there is a ton to unpack here. But I think we can focus on three specific lesions that we should learn from the VA’s experience in trying to buy into LCS.
1) It is really hard to do. The VA did not mess around here, hiring full time coordinators and selecting the centers best prepared to ramp up a population-level screening program. And yet, they not only lost a tremendous proportion of potentially eligible patients to insufficient ability to evaluate their candidacy, but also could convince barely half of those they were certain were eligible. As I mentioned above, it is really difficult to say what effect this will have on the mortality benefit of screening, but it’s not unreasonable to assume it will have some effect.
2) We need to figure out what to do about these nodules. As we focus more and more on the high-risk patients who will benefit most from this intervention (see for example the VA’s older, heavier smoking population), we may see the higher proportion of nodules they experienced here. The high inter-center variation in frequency emphasizes the degree to which we still lack a standardized method of evaluating CT scans for this purpose. The radiologists are working on it, and have managed to increase the positive predictive value by 2.5 fold (to 17%, as opposed to the ~4% in this VA trial), but this needs to be implmented on a larger scale.
3) It’s important that we figure out how to make this work. This is a way to stop people dying from lung cancer, and it’s not every day we have a demonstrated mortality benefit and the ability to bring that to the people who need it. Do not write off lung cancer screening just because there are major issues in implementation.
There is no question that LCS leads to early diagnosis of potentially curable lung cancer that might be inoperable if it were not screen detected. Lung cancer does not seem to have the same issues in lead time bias as we have encountered in breast and prostate cancer screening – a more aggressive tumor, if lung cancer is there it needs to come out yesterday if not sooner. The concerns about the health and fiscal impact of a test with a relatively low signal-to-noise ratio are totally legitimate, but that is not a reason to throw out the test. It’s a reason to work hard to minimize these risks. We can start with not screening patients who do not meet criteria — an article published in the same issue of JAMA notes that while we are seeing an increase in appropriate use of CT for lung cancer screening, we are simultaneously seeing more CT screening of patients who do not in fact meet the inclusion criteria. To say it mildly, this is a problem.
Lung cancer screening remains a huge bugaboo for those of us who want to give our patients evidence-based cancer screening but want to avoid low-yield interventions that subject them to needless testing and anxiety. As with PSA before it, there are wrinkles to be ironed out in how we harness low-dose CT to reduce the burden of lung cancer mortality in the population at large. But when used as directed it’s already as cost-effective and safe as many other interventions we already do, so while the details may change, the smart money says LCS will be making its way onto more health maintenance checklists, not fewer, over the next 5 years.
Coming Up on IMHEAT
Today is my 30th birthday and I’m moving to New York tomorrow, so while I’ll try to get something written for Thursday, I don’t want to make a promise I can’t keep. But, we will be back on Monday for sure! In case I do find time to write something or get up a great guest piece, make sure you’re all set to find out about it as it happens: