lung cancer screening for smokers, an individual risk-based approach

The USPSTF strongly recommends low-dose chest CT (LDCT) annual screening for ever-smokers with >30 pack-year smoking history aged 55-80 or until they are 15 years after stopping smoking, based on the 3-year National Lung Screening Trial (NLST). JAMA just published an article evaluating the use of risk models/individual risk-based strategies to help focus the LDCT intervention (see lung cancer CT screen risk model jama2016 in dropbox, or doi:10.1001/jama.2016.6255). By looking at individual lung cancer risk beyond the criteria of NLST, they actually found relatively higher risk in some patients with a low risk by NLST (and therefore no screening done in NLST or offered by USPSTF) but a low risk for many included in the USPSTF guidelines. of note, there is no currently accepted validated risk tool for lung cancer for population screening. details:

--they looked at 3 databases: the CXR-only wing of the NLST (2002-2009), the ever-smokers control group of the Prostate, Lung, Colorectal, and Ovarian cancer screening trial (PLCO, 1993-2009), and the National Health Interview Survey (NHIS, 1997-2001)

    --PLCO: 155K US men and women 55-74 yo had 4 annual  CXRs and found no benefit from screening CXR in smokers

    --NLST randomized 53.5K smokers aged 55-74 with at least 30 pack-years smoking to LDCT vs CXR and found 20% decrease in lung cancer mortality by LDCT

    --NHIS: 87.5K people followed in annual cross-sectional US group from 2004, with linkage to the National Death Index. This database from 1997-2001 was used to validate the lung cancer death model. Then the model was applied to a more contemporary US population (NHIS 2010-2012) looking at all ever-smokers aged 50-80, which included 18,643 people, 52% male, 72% white/15% black/9% Hispanic, 55% post-high school education, 32% with BMI>30, 36% current smoker and 26% <10 pack-yrs, 21% 10-20, 15% 20-30, 14% 30-40, 24% >40; 67% had quit >15 years; 98% no family history, 7% had emphysema.

--they assessed an array of characteristics from these databases (including age; education; sex; race; smoking intensity, duration, and quit-years; BMI; family history of lung cancer; self-reported symptoms of emphysema) to develop risk-based models for lung cancer incidence and deaths, then applied that to the later NHIS database.

--results:

    --Hazard Ratios for lung cancer incidence (will only mention the really significant ones)

        --age, very highly correlated with lung cancer incidence (HR 80) and lung cancer death (HR 432)

        --pack-years smoking:

            --30-40: HR 1.63 for lung cancer incidence; 1.74 for lung cancer death

        --90% of the CT-preventable lung cancer deaths are likely preventable by screening only 49% of US ever-smokers aged 50-80

        --one remarkable finding: in the risk-based model to USPSTF, 36% of the USPSTF-eligible smokers would not need to be screened (5-year lung cancer risk, 1.3%; NNS, 647), and would be replaced by 36% high-risk smokers (5-year lung cancer risk, 3.2%; NNS, 226) who did not meet the USPSTF criteria (mostly because they were African-American, lower BMI, less educated; 22% smoked <30 pack-years but tended to be longer-term smokers (>45 years, but 61% smoked <1/2  pack-per-day), and 14% quit > 15 years ago but were high intensity smokers, almost all having >30 pack-years and 53% >45 pack-years ). ie, following the USPSTF recommendations, instead of an individual risk-based approach (incorporating more than just the smoking history), would both over-screen many low-risk people and not screen some of the high risk ones.

--the lung cancer incidence model was validated by the chest xray groups of the NLS and PLCO of ever-smokers; the lung cancer death model was validated in the 1997-2001 NHIS and in the PLCO xray group of ever-smokers

--lung cancer mortality, by this model, was 24% lower than expected in the NLST xray group

--based on the NHIS 2010-2012 data, there was an estimated 43.4 million ever-smokers aged 50-80 in the US.

--screening the 9 million ever-smokers eligible by the USPSTF criteria for LDCT: the estimate was to prevent 46,488 deaths over 5 years

--but screening  9 million of the highest risk group by the risk-based population would prevent 55,717 deaths (9229 more)

    --so, in this risk-based model NNS (number-needed-to-screen to prevent a death) was 162 vs 194 with USPSTF, with fewer false-positives (116 vs 133), all with p<0.001

    ​--and if one used the USPSTF NNS of 194 and applied that to the highest risk of the risk-based group (with risk >1.9%) [ie, increasing the screening by the risk-based analysis to equal the NNS from USPSTF], then 3.1 million more people would be screened and then 62,382 deaths would be prevented. This increase to 12.1 million would also not have an increase in numbers of false positives.

commentary:

--so, this study did a good job developing risk models and validating them through several US cohorts, thereby suggesting that they are pretty robust and transportable. The models seemed to be more efficient than the USPSTF recommendations in terms of identifying higher risk individuals, decreasing lung cancer incidence and mortality, and decreasing the false-positive rates from LDCTs.

--in the NLST itself, 88% of CT-prevented lung cancer deaths occurred in the 60% of those at highest risk and there were 64% with false positive results; only 1% of the lung cancer deaths occurred in the 20% at the lowest risk (see lung cancer CT screening risk based models nejm2013 in dropbox, or Kovalchik SA. N Engl J Med 2013; 369: 245).​ In fact, if one looks at those with normal LDCT in the initial NSLT, there was a dramatically lower risk of lung cancer development or death (see blog at end), to the point that I am strongly considering stopping LDCT after a couple of negative scans.

--a few limitations of the risk-based study:

--there is the assumption that the 20% decreased mortality in NSLT-eligible would apply to the NSLT-ineligible that would be included in the risk-based strategy.

--the NHIS cohort only has data on lung cancer mortality not incidence, so the above risk-based model for mortality was validated only by the 2 research studies (PLCO and NLST xray only groups)

--confining the LDCTs to the high risk only, by using the risk-model, may be associated with more complications from procedures and surgery (these people are generally at higher surgical risk), so might distort the risk:benefit calculations from NLST

--the risk-based approach does have some improvement in decreasing lung cancer deaths from 46488/9 million screened to 55717/9 million, which is increasing absolute numbers from 5.2 to 6.1/1000 screened. still pretty small numbers.

--so, as per my prior blogs on the USPSTF recommendations, I think they way-overinterpreted NLST by extending a 3 year study (with decreasing pickup of incident lung cancers by the 3rd year), to the potential for 25 years of annual screening and the attendant high radiation exposure, to extending the upper age limit from 74 in NLST  to 80, and then to codifying a single but good study into routine practice though the absolute risk reduction was relatively small (62 deaths per 100,000 person-years).  The current individual risk-based approach suggests that there are many high risk patients who would not be screened by the USPSTF criteria, and many on the USPSTF list who are actually quite low risk and likely do not benefit much from the screening. I hope (and sort of expect) that the current USPSTF guidelines will be reviewed and reconsidered at some point in the next couple of years… And, it is important to remember that, as pointed out in the other blogs, lung cancer is not even close to being the primary killer associated with smoking (heart disease from smoking being much more prevalent, and COPD, etc rating pretty high as well). And focusing on lung cancer screening may dilute the bigger message (ie, it would be pretty awful if patients who had a normal LDCT felt that smoking was not really so bad for them, andit was okay to continue smoking…) see older blogs below for more on this.

see: http://gmodestmedblogs.blogspot.com/2016/03/need-annual-low-dose-chest-cts.html which analyzes a retrospective study from NLST finding that those with an initial normal LDCT had about 35% lower rates of lung cancer incidence and mortality

http://gmodestmedblogs.blogspot.com/2014/10/uspstf-lung-cancer-screening-revisited.html  is a critique of NLST, especially its perhaps overenthusiastic acceptance and extensions by USPSTF