New CT lung cancer screening rules save more lives than NLST

By Eric Barnes, AuntMinnie.com staff writer

Lung cancer screening with CT would detect more cancers and save more lives if current screening criteria were modified to include a wider range of individuals at risk, concludes a new study in the February 21 New England Journal of Medicine.

The study found that the new screening criteria, validated on the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial cohort, are more accurate than the criteria from the National Lung Screening Trial (NLST), which showed an average 20% reduction in deaths.

Sensitivity rose from 71% to 83% (p [ 0.01) when conventional NLST criteria were replaced with a model based on the experience of PLCO trial participants, with no corresponding decrease in specificity, wrote researchers from Brock University in St. Catherines, Ontario, and several U.S. institutions.

The PLCO-based criteria “predicted the six-year risk of lung cancer with high accuracy and was more efficient at identifying persons for lung cancer screening, as compared with the NLST criteria,” wrote Dr. Martin Tammemägi, PhD, professor of epidemiology at Brock, and colleagues (NEJM, February 21, 2013, Vol. 368:8, pp. 728-736).

Better guidelines?

“If you were starting a lung cancer screening program, the data here show that it makes more sense to use the [PLCO-based] risk prediction model to decide who should be enrolled,” Tammemägi told AuntMinnie.com in an interview. “That would lead to your having to screen fewer people to get more lung cancers, and we anticipate that there will be a substantial number of lives saved using the model as opposed to using the NLST criteria.”

In 2011, results of that trial showed that CT screening could reduce mortality for long-term smokers by 20%. However, NLST’s strict selection criteria required participants to be current smokers or have quit within the past 15 years, be between 55 and 74 years of age, and have a smoking history of at least 30 pack-years.

“These selection criteria were intended to increase the yield of lung cancers, but they exclude many known risk factors for lung cancer, and with dichotomization of continuous data, much valuable information is not included,” Tammemägi and colleagues wrote.

Applying an accurate lung cancer risk prediction model can identify those at the highest risk and, with CT screening, increase the number of lung cancers detected, while reducing the number who need to be screened. The present study tried to create a better model by modifying PLCO data to ensure their applicability to NLST data, defining risk as the probability of a lung cancer diagnosis within six years.

The researchers developed the new modified logistic-regression model to predict lung cancer in the PLCO control group of smokers; the model was introduced in a 2011 paper (Journal of the National Cancer Institute, July 6, 2011, Vol. 103:13, pp. 1058-1068).

For the present study, the researchers validated the model in the PLCO intervention group of smokers (i.e., PLCO participants who met the model criteria for inclusion), as well as NLST participants. The model was then revalidated in the PLCO intervention group stratified as to whether or not participants met the NLST criteria. Analysis of follow-up was halted at six years, as was NLST, to create consistency between the models for comparison, the group noted.

Unlike NLST, the PLCO-based model used in the study, dubbed PLCOM2012, is multifactorial, combining estimated risks from a number of variables. The NLST model, in contrast, demands dichotomized yes/no answers to a handful of questions, such as whether the screening candidate quit smoking within the past 15 years, Tammemägi told AuntMinnie.com.

“What makes this [PLCO-based] model more predictable is the detail of information captured is much greater than just saying ‘have you smoked more or less than 30 pack-years,’ or ‘did you quit within 15 years or not,’ which is basically what the current standard has been,” he said.

Rather than NLST’s yes or no responses, “this model actually puts in the numbers, and it has more numbers describing smoking,” he said.

For example, PLCO has four variables describing smoking history, as opposed to two simple toggled variables in NLST. There are also seven additional variables for risk factors for lung cancer: age, race, socioeconomic status measured by education, body mass index (BMI), chronic obstructive pulmonary disease (COPD), personal history of cancer, and family history of lung cancer.

“So we’re using more information to capture a person’s real risk,” he said.

Still, the age criteria, 55 to 74 years, were the same for both NLST and PLCO.

The model’s calibration and ability to discriminate lung cancer cases was assessed as area under the receiver operator characteristics curve (AUC). In the validation data, 14,144 (37.9%) of 37,332 individuals met the NLST screening criteria, while for comparison, 14,144 of the highest risk individuals were considered eligible for screening under PLCOM2012 criteria. Finally, Cox models were used to determine whether the 53,202 individuals undergoing low-dose CT screening in NLST differed according to risk.

More cases detected

The results showed that the PLCOM2012 model was a better predictor of lung cancer than the NLST model, with an AUC of 0.803 in the development dataset and 0.797 in the validation dataset.

Compared with NLST criteria, PLCOM2012 criteria had better sensitivity for predicting the development of lung cancer (71.1% in NLST versus 83% in PLCOM2012, p [ 0.001) and better positive predictive value (4% in NLST versus 3.4% in PLCOM2012, p = 0.01), without loss of specificity (62.9% and 62.7%, respectively; p = 0.54). In addition, 41.3% fewer lung cancers were missed.

Accuracy of lung cancer classification using PLCOM2012 criteria
Criteria Participants with lung cancer
(n = 678)
Participants without lung cancer
(n = 36,654)
Total participants
(n = 37,332)
Predictive value
NLST
Criteria positive 482 true positive (3.4%) 13,662 false positive (96.6%) 14,144 PPV 3.4%
Criteria negative 196 false negative (0.8%) 22,192 true negative (99.2%) 23,188 NPV 99.2%
Sensitivity 71.1%
Specificity 62.7%
PLCOM2012
Criteria positive 563 true positive (4%) 13,581 false positive (96%) 14,144 PPV 4%
Criteria negative 115 false negative (0.5%) 23,073 true negative (99.5%) 23,188 NPV 99.5%
Sensitivity 83%
Specificity 62.9%
PPV = positive predictive value; NPV = negative predictive value.
NLST criteria for study entry included a smoking history of at least 30 pack-years, and cessation within 15 years for smokers who had quit. For the PLCOM2012 criteria, positivity was defined as a probability of lung cancer greater than 1.3455% over a period of six years. Table courtesy of NEJM.

The NLST screening effect did not vary according to PLCOM2012 risk (p = 0.61 for interaction), according to the authors.

Put another way, among the 37,332 smokers in the PLCO intervention group, the PLCOM2012model chose 81 more individuals for screening who received a lung cancer diagnosis at follow-up compared to the NLST criteria.

Assuming a 15% overdiagnosis rate, 69 of the 81 people can be considered to have true life-threatening lung cancer, the group wrote, and based on a five-year survival rate of 15%, the expected number of deaths among persons who did not undergo screening would be 59.

Overall, PLCOM2012 makes better use of the data compared to NLST. But how do you decide who meets the criteria for screening? It depends on how much money you have and how many people you want to screen, starting with the highest-risk individuals.

Deciding whom to screen

When data are plugged into the model, it outputs the probability of the individual developing lung cancer within six years. It is then up to each screening program to determine how many patients it can afford to screen based on probability of disease according to the model.

“If you use the cutoff of 1.6% risk of lung cancer and screen everybody that is above that, you will have to screen one-third of all smokers and you’ll pick up 80% of all lung cancers in that population,” Tammemägi said. “If you want to do better than that, you can go down to 1% and screen everybody who has 1% risk or higher; you would have to screen half the smokers, and you would be capturing 90% of the lung cancers in the population. So it’s a bit of a trade-off, and it depends on what each program can afford to do.”

Under NLST criteria, for example, a white 55-year-old man with a BMI of 27, no COPD, and no personal or family history of cancer, who smoked 20 cigarettes a day for 30 years and quit 15 years ago, would have the lowest risk eligible under NLST — with about 0.5% chance of developing lung cancer, Tammemägi said. So even though the risk is very low for this individual, NLST criteria would allow screening, and a program using the criteria would pay for it.

On the other hand, the current study shows that the PLCOM2012 model, “given equal numbers of people selected, will miss 41% fewer lung cancers, and compared to NLST criteria, it’s 17% more sensitive in picking up lung cancers overall,” he said.

The new criteria are currently being tested in a real-world setting. The Pan-Canadian Early Detection of Lung Cancer Study, in which Tammemägi is involved, is using a prototype of the PLCO-based criteria exclusively to select individuals for screening. The group enrolled 2,537 individuals in the trial, with 113 lung cancers detected during the first three years, he said.

“We have identified 4.5% with cancers in three years; that is way over double what the National Lung Screening Trial had at this point,” he said. “So the Canadian trial corroborates the findings of this paper, and indicates that the model does work.”

Tammemägi said the group has been contacted by several different research teams in Canada, Australia, and the U.S. who want to use the model. Some have asked that it be altered to include asbestos risk.

An online calculator (brocku.ca/cancerpredictionresearch) can be used to determine an individual’s lung cancer risk according to the PLCOM2012 model.