If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
School of Environmental Science and Public Health, Wenzhou Medical University, Chashan University Town, Wenzhou, Zhejiang Province, People's Republic of China
School of Environmental Science and Public Health, Wenzhou Medical University, Chashan University Town, Wenzhou, Zhejiang Province, People's Republic of China
Two-thirds of patients in the United States with newly diagnosed lung cancer would not meet the current U.S. Preventive Services Task Force (USPSTF) screening criteria, which suggests a need for amendment of the definition of high risk. To provide evidence of additional high-risk subpopulations and estimated gains and losses from using different criteria for screening eligibility, we conducted a two-step study using three cohorts.
Methods
The two prospective cohorts comprised 5988 patients in whom primary lung cancer was diagnosed between 1997 and 2011 (the hospital cohort) and 850 defined-community residents (the community cohort); the retrospective cohort consisted of the population of Olmsted County, Minnesota, which was observed for 28 years (1984–2011). Subgroups of patients with lung cancer who might have been identified using additional determinates were estimated and compared between the community and hospital cohorts. The findings were supported by indirect comparative projections of two ratios: benefit to harm and cost to effectiveness.
Results
Former cigarette smokers who had a smoking history of 30 or more pack-years and 15 to 30 quit-years and were 55 to 80 years old formed the largest subgroup not meeting the current screening criteria; they constituted 12% of the hospital cohort and 17% of community cohort. Using the expanded criteria suggested by our study may add 19% more CT examinations for detecting 16% more cases when compared with the USPSTF criteria. Meanwhile, the increases in false-positive results, overdiagnosis, and radiation-related lung cancer deaths are 0.6%, 0.1%, and 4.0%, respectively.
Conclusions
Current USPSTF screening criteria exclude many patients who are at high risk for development of lung cancer. Including individuals who are younger than 81 years, have a smoking history of 30 or more pack-years, and have quit for 15 to 30 years may significantly increase the number of cases of non-overdiagnosed screen-detected lung cancer, does not significantly add to the number of false-positive cases, and saves more lives with an acceptable amount of elevated exposure to radiation and cost.
With the declining percentage of the U.S. population that smokes, the incidence of lung cancer and mortality due to lung cancer have been decreasing among men for the past three decades and, only recently, have begun showing a decrease among women.
As of 2014, use of low-dose computed tomography (LDCT) screening for lung cancer was recommended by the U.S. Preventive Services Task Force (USPSTF) for annual screening of people aged 55 to 80 years who have a history of smoking cigarettes at a rate of 30 or more pack-years and either are current smokers or have quit within the past 15 years.
However, our recent report showed that approximately two-thirds of patients with newly diagnosed lung cancer would not have met the current USPSTF criteria for being at high risk for development of lung cancer and thus eligible for LDCT screening.
In particular, we found a 24% falloff in meeting the eligibility criteria for screening (from 57% in 1984–1990 to 43% in 2005–2011), which exceeded the 17% decline in incidence of lung cancer (from 53 to 44 cases per 100,000 population) within the same time intervals. Herein we report our further investigations to delineate the high-risk subpopulations on the basis of evidence from two prospective cohorts of patients with lung cancer and a retrospective community cohort. Our goal was to improve the identification of individuals at high risk for development of lung cancer by (1) demonstrating the chronological patterns of patients who would have been the beneficiaries or “missed-outs” under the USPSTF criteria for lung cancer screening in two contrasting cohorts and (2) providing indirect evidence of a new subpopulation that should be included in the definition of high risk and the potential benefit versus harm and projected cost versus effectiveness of including them.
Methods
Study Population
This study included two steps: description and validation. Step 1 used two prospectively observed cohorts of individuals with lung cancer, one based on patients referred to Mayo Clinic (i.e., the hospital cohort, n = 5988) and the other consisting of residents of Olmsted County, Minnesota (i.e., the community cohort, n = 850). The hospital cohort included patients who had pathologically confirmed primary lung cancer diagnosed at Mayo Clinic in Minnesota during a 15-year period (between January 1, 1997 and December 31, 2011)
All cases were identified using the Rochester Epidemiology Project database, which has for more than 60 years maintained a comprehensive system linking the medical records of almost all persons residing in Olmsted County.
This population comprises approximately 140,000 persons, 83% of whom are non-Hispanic whites; it is socioeconomically similar to the white population of the United States and is representative of the population of the midwestern United States. More details were published previously.
This study was approved by the institutional review boards of Mayo Clinic and Olmsted County Medical Center.
The objective of step 2 was to provide indirect evidence supporting the findings in step 1. We have derived comparative benefit-to-harm and cost-to-effectiveness ratios for three sets of criteria—NLST, USPSTF, and the expanded criteria suggested by our study on the basis of the information provided in the models by de Koning et al.
Although hypothetical and indirect, the comprehensive models built by de Koning et al. are very helpful in initial evaluation of the impact (positive and negative) of a potential high-risk subpopulation given the lack of individual-level smoking history data or up-to-date and accurate smoking history information for entire populations of interest. Briefly, the modeling groups standardized input data on smoking histories and non–lung cancer mortality to simulate life histories of the U.S. cohort born in 1950, which uses an updated version of the National Cancer Institute’s smoking history generator. Their models assumed 100% adherence to screening criteria; the data derived from trials of short duration (e.g., 4 to 9 years) were extrapolated to lifetime follow-up, and smoking history data from one to two decades ago were assumed to be current.
: (1) total number of CT examinations, including screening examinations; (2) number of screening-detected cases; (3) reduction in lung cancer mortality; (4) total cases detected at an early stage; (5) average number of screening examinations per person screened; (6) screening examinations per averted death from lung cancer; (7) screening examinations per life year gained; (8) average number of false-positive results per person screened; (9) number of instances of overdiagnosis; (10) overdiagnosis as a percentage of screening-detected cases; and (11) radiation-related lung cancer deaths.
Table 1Demographic and Clinical Features of the Community and Hospital Cohorts of Individuals with Lung Cancer Diagnosed in 1997–2011
Characteristics
Prospectively Enrolled and Monitored Lung Cancer Patients (N = 6838)
Data missing on the race of 13 patients (0.2%) in the hospital cohort and on pack-years of smoking by one patient (0.1%) in the community cohort and 27 patients (0.5%) in the hospital cohort, respectively.
Data missing on the race of 13 patients (0.2%) in the hospital cohort and on pack-years of smoking by one patient (0.1%) in the community cohort and 27 patients (0.5%) in the hospital cohort, respectively.
0.0025
1–19
88 (11.4%)
781 (15.8%)
20–29
82 (10.6%)
583 (11.8%)
≥30
601 (78.0%)
3580 (72.4%)
Quit-years (for former smokers), n (%)
0.0001
<15 years
241 (57.1%)
1443 (47.2%)
≥15 years
181 (42.9%)
1613 (52.8%)
SD, standard deviation; Q, quarter.
a Data missing on the race of 13 patients (0.2%) in the hospital cohort and on pack-years of smoking by one patient (0.1%) in the community cohort and 27 patients (0.5%) in the hospital cohort, respectively.
We used a screening program most similar to the program A-55-80-30-25 in de Koning HJ, Meza R, Plevritis SK, et al.,6 namely, A-55-80-30-30 (in which A means “annual,” 55 is the start age, 80 is the stop age, 30 is the number of pack-years smoked, and 30 is the number of years since quitting), which is also supported by our previous work (Ebbert et al., 20033; Wang et al., 2015; and Aberle et al., 2011.7
Setting of LDCT annual entry criteria for screening and comparisons
1. Total CT examinations, including screening, setting NLST at 100% as reference
2. Screening-detected cases, setting NLST at 100% as reference
3. Reduction in lung cancer mortality, %
4. Total cases detected at an early stage, %
5. Average screening examinations per person screened, n
6. Screening examinations per lung cancer deaths averted, n
7. Screening examinations per life- year gained, n
8. Average false-positive results per person screened, n
namely, A-55-80-30-30 (in which A means “annual,” 55 is the start age, 80 is the stop age, 30 is the number of pack-years smoked, and 30 is the number of years since quitting), which is also supported by our previous work (Ebbert et al., 2003
Their comprehensive models standardized input data on smoking histories and non-lung cancer mortality to simulate life histories of the U.S. 1950 birth cohort.
We used an A-55-80-30-25 screening program as in the models of de Koning et al., in which A stands for annual LDCT, 55 is the start age, 80 is the stop age, 30 indicates a smoking history of 30 or more pack-years, and 25 indicates 25 or fewer years since quitting; this model is most similar to our findings as supported by our previous work.
For each patient, medical records were reviewed and abstracted for the following information: demographic characteristics (age, sex, and race), history of occupational exposure, history of tobacco exposure, histologic diagnosis of lung cancer, staging, treatment modality, family history of lung cancer, and other comorbid conditions. For the hospital cohort, information was also obtained from an interview, follow-up questionnaire, or both. The patient interview and/or annual follow-up questionnaire obtained detailed information on history of tobacco exposure, history of occupational exposure, and family history of cancer. The information on history of tobacco exposure included current or past use, duration, average number of cigarettes smoked per day, and number of years since quitting smoking. Current smokers were defined as those who were actively smoking, as well as those who had stopped smoking within 1 year before being receiving a diagnosis of lung cancer. Former smokers were defined as those who had quit smoking for at least 1 year before receiving their diagnosis. Never-smokers were defined as those who had smoked fewer than 100 cigarettes during their lifetime. Pack-years are calculated by multiplying the number of packs smoked daily by the number of years smoked. History of chronic obstructive pulmonary disease (COPD) was determined on the basis of an explicit diagnosis documented in the medical history with pulmonary function tests in the medical record. Family history of lung cancer was defined as having at least one first-degree relative (parent, sibling, or offspring) with lung cancer. Positive exposure to asbestos was based on self-reported direct contact with the asbestos-containing material for at least 1 year and corroborated by occupational history (job titles and tasks) at least 5 years before the diagnosis of lung cancer.
Statistical Analysis
In step 1, we performed three descriptive analyses: (1) calculation and sorting of the frequencies of selected key characteristics, including age groups (50–54, 55–80, >80 years), smoking status (ever-smokers versus never-smokers), smoking history in terms of pack-years (<20, 20–30, >30) and years since quitting smoking (quit-years; <15, 15–30, >30), history of COPD and/or lung cancer, history of asbestos exposure, and family history of lung cancer in first-degree relatives; (2) illustration by pie charts of subgroups of those not meeting the USPSTF criteria; and (3) calculation and illustration of the distribution of pack-years and quit-years in patients not meeting the USPSTF criteria.
In step 2, we performed three levels of hypothetical comparisons: (1) the relative gain achieved by using the USPSTF criteria versus the NLST criteria, (2) the relative gain achieved by using the criteria from our study versus those of the NLST, and (3) the relative gain achieved by using the criteria from our study versus those of the USPSTF. For clarity and simplicity in the comparisons, we set the estimates of the NLST as the standardized reference values at 100 or 100% when involving the actual numbers (e.g., total numbers of CT examinations and screening-detected cases). For parameters of common knowledge (e.g., reduction in lung cancer mortality and total cases detected at an early stage), the original model-based percentage for the NLST was used as the comparator, with the predicted change (%) resulting from using the USPSTF criteria and our study criteria. For the estimated means (e.g., average screening examinations per person screened per life year gained, average false-positive results per person screened, and over diagnosis of screening-detected cases), the original model-based average for the NLST was used as the comparator, with the predicted change (%) resulting from using the USPSTF criteria and our study criteria.
Results
Characteristics of Patients in the Two Prospective Lung Cancer Cohorts
Table 1 presents basic information and comparisons of 5988 hospital and 850 community lung cancer cohort members in whom lung cancer was diagnosed between 1997 and 2011. Patients’ characteristics differed between the two cohorts, reflecting the typical referral bias in a tertiary medical center. All variables except sex ratio were significantly different; specifically, compared with the community cohort, the hospital cohort was younger and characterized by a higher representation of never-smokers, lighter smokers, and long-term quitters. Despite these differences, the two cohorts showed remarkably consistent results in the following three respects:
Subgroups Outside the USPSTF Screening Criteria
Figure 1 illustrates the relative proportions of screening-eligible and screening-ineligible patients in the two patient cohorts in order of subgroup frequency. Individuals falling under more than one variable were grouped within the larger subgroup. The frequencies of the selected risk factors in each cohort are summarized in Table 1. When the community cohort was used as a standard reference group (Fig. 1A), 46% of cases were found to meet the USPSTF criteria, whereas only 38% in the hospital cohort were found to do so. In the areas of the pie charts other than that representing those patients meeting the USPSTF screening criteria, we compared eight factors characterizing the patients in terms of additional intensity of tobacco smoking exposure, age of 50 to 54 years at the time of diagnosis of lung cancer, family and personal history of lung cancer, and history of COPD and asbestos exposure. The two most frequently occurring characteristics in both cohorts were “quit smoking for 15 to 30 years” (12% versus 17%) and “smoking history of 20 to 30 pack-years” (6.3% versus 6.2%). In both cohorts, “history of COPD” (1.9% versus 2.2%) and “personal history of lung cancer” (0.3% versus 0.6%) were the lowest in frequency of all presented factors.
Figure 1Subgroup distribution by frequency of known risk factors in patients with lung cancer who were younger than 81 years and in whom lung cancer was diagnosed in 1997–2011: (A) community cohort and (B) hospital cohort.
Distribution of Pack-Years and Quit-Years in Patients Ineligible for Screening Under the USPSTF Criteria
We examined the distribution of pack-years and quit-years among those in both cohorts who were younger than 81 years and ineligible according to the USPSTF screening criteria, as depicted in Figures 2A and 2B; clearly standing out is the subgroup of those who had a smoking history of at least 30 pack-years and had quit smoking for more than 15 years. Therefore, we propose that being a former smoker who has quit smoking for 15 to 30 years and has a smoking history of at least 30 pack-years be added to the current USPSTF screening criteria.
Figure 2Distribution of pack-years and quit-years in patients younger than 81 years who were ineligible by the National Lung Screening Trial and U.S. Preventive Services Task Force criteria, 1997–2011: (A) community cohort and (B) hospital cohort.
Temporal Pattern Change According to the NLST and USPSTF Screening Criteria
Of those patients in the community cohort, 35.7% would have met the entry criteria of the NLST and 45.8% would have met the USPSTF screening criteria. Of those patients in the hospital cohort, only 33.4% and 38.2% would have met the respective criteria. Figures 3A and B illustrate the relative proportion of alternative criteria over three 5-year intervals for both cohorts (i.e., 1997–2001, 2002–2006, and 2007–2011). The bottom portion of each bar represents the NLST criteria (age 55–74 years, smoking history of at least 30 pack-years, and former smoker with 15 or fewer quit-years), the middle portion of each bar represents those aged 75 to 80 years (extended according to the USPSTF criteria), and the top portion shows our proposed addition of former smokers with a smoking history of at least 30 pack-years and 15 to 30 quit-years. More specifically, in the community cohort (see Fig. 3A), the percentage of patients who were eligible according to the NLST decreased from 40.8% to 31.9% (p = 0.017) and the percentage of those eligible according to the USPSTF decreased from 53.2% to 40.5% (p = 0.002) over the 15-year period. On the other hand, the proportion of those eligible according to the proposed criteria increased; particularly noteworthy is the fact that coverage of patients meeting the proposed criteria in the most recent time interval was at 52%.
Figure 3Temporal pattern of percentage of the community cohort (A) and hospital cohort (B) covered by the National Lung Screening Trial (NLST) and U.S. Preventive Services Task Force (USPSTF) screening criteria, 1997–2011.
The same trends toward decreasing eligibility were observed in the hospital cohort (see Fig. 3B): the percentage of those eligible according to the NLST criteria decreased from 35.2% to 31.0% (p = 0.003), and the percentage of those eligible according to the USPSTF criteria decreased from 40.3% to 35.1% (p < 0.001). More strikingly in the most recent time interval, the percentage of lung cancer cases added by the subgroup with 15 to 30 quit-years (17.6%) is more than four times higher than the 4.1% increase achieved by adding the USPSTF criterion of “those 75 to 80 years old” to the NLST criteria.
Potential Benefit versus Harm and Projected Cost versus Effectiveness
Table 2 provides hypothetical projections of benefit versus harm and cost versus effectiveness through 11 itemized comparisons (labeled columns 1–11), reflecting meaningful relative gains and losses with the three sets of criteria—the NLST, USPSTF, and our study—and purposefully not involving actual expense and productivity measures. Five illustrative points regarding pros and cons are listed in the following paragraphs.
1.
With regard to the balance of total numbers of computed tomography (CT) screening examinations and screening-detected lung cancer cases (columns 1 and 2), when the NLST estimates are set at 100 as a reference, using the USPSTF criteria would add 8.2% more CT examinations, for a 19.7% gain in screening-detected cases of lung cancer, whereas the criteria suggested by our study may add 29% more CT examinations, for a 39% gain in screening-detected lung cancer cases. With regard to our study criteria versus those of the USPSTF, 19% more CT examinations would be the trade-off for 16% more cases of lung cancer detected.
2.
In the models in which the NLST criteria reduce lung cancer mortality by 12.3%, a greater reduction in mortality may be achieved by using the criteria suggested by our study than by the predicted reduction according to the USPSTF criteria (column 3, i.e., 15.8% [12.3 + 3.5] versus 14% [12.3 + 1.7]). Our study criteria versus those of the USPSTF provide a 1.8% greater reduction in lung cancer mortality.
3.
In the models in which 48.4% of the total cases detected in the NLST were in an early stage, a higher rate of detection of early-stage lung cancer may be achieved with the criteria suggested by our study than the predicted increase by the USPSTF criteria (column 4, i.e., 52.1% [48.4 + 3.7] versus 50.5% [48.4 + 2.1]). Our study criteria versus those of the USPSTF result in a 1.6% increase in detecting lung cancer at an early stage.
4.
When the average numbers of screening examinations per person screened (column 5), per lung cancer death averted (column 6), and per life year gained (column 7) were weighed, the reference criteria of the NLST yielded 13.3, 577, and 49, respectively. Compared with the NLST criteria, the USPSTF criteria resulted in gains of 1.1, 3, and 3, whereas use of our study criteria resulted in gains of 3.1, 6, and 5, respectively. Use of our study criteria versus those of the USPSTF resulted in an increase in number of average screening examinations by two per person screened, three per lung cancer death averted, and two per life year gained, respectively.
5.
With regard to concerns regarding false positivity (column 8), overdiagnosed cases (columns 9–10), and radiation-related lung cancer deaths (column 11), the increases in false-positive results, overdiagnosis, and radiation-related lung cancer death are minimal by all comparisons.
Discussion
According to the latest data from the Surveillance Epidemiology and End Results Program and National Program of Cancer Registries, the incidence of lung cancer among men in the United States peaked in 1984 but continued to increase among women until 2006.
We found a similar trend in the population of Olmsted County (Fig. 4), thus supporting the validity of considering our data to be generalizable to the United States' general population. During 2005–2012, the percentage of heavy smokers in the United States who smoked 30 or more cigarettes per day declined significantly, from 12.6% to 7.0%.
National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines): Lung Cancer Screening. Version 1.2012. http://www.jnccn.org/content/10/2/240.full. Accessed December 1, 2014.
The American Association for Thoracic Surgery guidelines for lung cancer screening using low-dose computed tomography scans for lung cancer survivors and other high-risk groups.
Figure 4Trends in the incidence of lung cancer, Olmsted County, Minnesota, 1984–2011, by calendar year of diagnosis and adjusted by age (A) and by age (B).
Since 2002, former smokers have outnumbered current smokers, and from 1998 to 2012, the percentage of U.S. adults who were current cigarette smokers declined from 24.1% to 18.1%.
The fact that most lung cancer cases diagnosed in the United States today are in former smokers reflects the success of smoking cessation efforts and also reflects the continued high-risk status of former smokers.
We specifically evaluated quitting smoking for 15 to 30 years because our previous study showed that risk for the development of lung adenocarcinoma remained elevated for up to 30 years beyond smoking cessation for former heavy and light smokers alike.
We have also reported that in Olmsted County, the trend has been toward a decrease in the proportion of patients with lung cancer and a smoking history of 30 or more pack-years and an increase in lung cancer among former smokers who had quit smoking for at least 15 years.
A striking observation from the current study is the distribution of pack-years and quit-years in those ineligible for screening under the USPSTF criteria (as is shown in Fig. 2): we found that compared with other risk categories, “quit smoking for 15 to 30 years” accounted for the greatest percentage of those with lung cancer.
Our results also showed that, compared with when the NLST entry criteria were used, the number of lung cancer cases discovered among those in the community and hospital cohorts who met the USPSTF criteria increased by only 4.1% to 8.6% (see Fig. 3, 2007–2011). In contrast, when “former smokers with 15 to 30 quit-years” was added, the increase in those who met the criteria and in whom lung cancer actually developed ranged from 13.3% to 17.6%. Therefore, assuming that screening could reduce mortality and be cost-effective to the same extent as expected without increasing harm, high-risk subpopulations outside the USPSTF criteria need to be reconsidered—especially those who have sustained smoking cessation beyond 15 years. By considering the patients from the Olmsted County population, we were able to compare the local and referral patients whose lung cancer was diagnosed during the same time period (see Fig. 4).
We acknowledge several limitations. First, data from Olmsted County may not be generalizable to the entire United States in terms of racial distribution, disease patterns, and access to care. White individuals, who are known to have a lower incidence of lung cancer, are overrepresented in Olmsted County. Second, although we found that “quit smoking for 15 to 30 years” and “smoked 20 to 30 pack-years” constitute the two highest percentages of patient subgroups in both cohorts, the proportions of patients in the hospital cohort who were older than 81 years, were never smokers, had a history of COPD, and had history of asbestos exposure differed from those in the community cohort—likely because of referral practice and patients’ self-preference. Third, we were unable to develop a prediction model for individualized assessment of lung cancer risk, as reported by other studies,
Predictive accuracy of the Liverpool Lung Project risk model for stratifying patients for computed tomography screening for lung cancer: a case-control and cohort validation study.
because doing so would require complete data on all known risk factors in the entire Olmsted County population and such data are not currently available. On the basis of the modeling data provided by de Koning et al.,
however, we were able to indirectly project the impact of extending the screened population to include former smokers with more than 15 years of cessation, which may significantly increase the number of non-overdiagnosed screen-detected lung cancers, save more lives, and have an acceptable amount of increased scan exposure and cost yet still be unlikely to significantly add false-positive cases.
Nonetheless, although our two prospective patient cohorts differed significantly in many aspects, they provided consistent study results from a case detection perspective (i.e., the relative proportions of patients with known risk factors in whom lung cancer was diagnosed). This is one of the efficient designs for timely capture of whether the end results of a disease under screening reasonably reflects the predefined high-risk population, although obtaining the most definitive answer to justify a change in high-risk definition being cost-effectiveness requires calculating risks directly, having the denominator to know the number needed to screen, and knowing how screening is implemented in the real world.
In both the community and hospital cohorts, the trend in percentage of patients with lung cancer who met the USPSTF criteria for inclusion in screening decreased between 1997 and 2011. The trend toward decreasing eligibility for screening in both cohorts exceeded the decline in incidence over time, which demonstrates that the current criteria for entry into lung cancer screening did not identify those in whom lung cancer actually developed. Inclusion of identifiable high-risk subpopulations should be reconsidered to improve current screening criteria. Our current and previous studies provide evidence that former smokers with 15 to 30 quit-years remain at high risk and should be considered eligible for LDCT screening for lung cancer. The current USPSTF recommendation to stop screening after 15 years of smoking cessation is not reflective of the continued high risk, although participation of the expanded population in the screening setting needs to be further evaluated.
Acknowledgments
We thank Barbara A. Abbott, Jennifer St. Sauver, Walter A. Rocca, and Barbara P. Yawn, for their Rochester Epidemiology Project resource made available to this study. The authors appreciate Connie Edwards for her technical assistance with the manuscript.
National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines): Lung Cancer Screening. Version 1.2012. http://www.jnccn.org/content/10/2/240.full. Accessed December 1, 2014.
The American Association for Thoracic Surgery guidelines for lung cancer screening using low-dose computed tomography scans for lung cancer survivors and other high-risk groups.
Predictive accuracy of the Liverpool Lung Project risk model for stratifying patients for computed tomography screening for lung cancer: a case-control and cohort validation study.
Disclosure: This study was funded by grants R03 CA77118, R01 CA80127, and R01 CA84354 from the U.S. National Institutes of Health; by Rochester Epidemiology Project grant R01 AG034676 from the National Institute on Aging; and by the Mayo Clinic Foundation. The authors declare no other conflict of interest.
00038-6&id=gr1.jpg){kind=link}
00038-6&id=gr2.jpg){kind=link}
00038-6&id=gr3.jpg){kind=link}
00038-6&id=gr4.jpg){kind=link}