early detection in lung cancer

CANCER SCREENING AND DIAGNOSIS 0025-7125/96 $0.00 + .20

EARLY DETECTION IN LUNG CANCER

Case Finding and Screening

Daniel R. Wolpaw, MD

The case for early detection of lung cancer would appear to be simple and logical. Lung cancer is a common and deadly disease. Surgi- cal resection is the only approach that offers a potential cure, and it is effective only for localized disease. Localized (or stage I) disease is more likely to be identified before cancer-related symptoms appear. Early detection efforts have clearly been shown to pick up an increased number of these presymptomatic lesions. Therefore, it would seem obvious that a program of early detection for lung cancer saves lives. This line of reasoning is both attractive and intuitive and is supported by evidence through every step except the most important one-saving lives. Studies in large groups of patients have failed to show a decrease in mortality through early detection. The ”logical” conclusion is not necessarily wrong-although it may be; the issue is complex, and an understanding of this complexity is important to decision making in a medical practice.

Debates over screening have always generated a significant amount of heat as well as light, and this is particularly true of lung cancer. As described by Sackett and Holland,-’4 there are the ”evangelists”-advo- cates of screening who would build on existing data and common sense-and there are the ”snails”-methodologists who focus on rigor- ous scientific (and sociopolitical) criteria to evaluate any proposed intervention. In the literature on screening for lung cancer, there are a considerable amount of data, much of it flawed or indirect, and an even greater volume of interpretation and reinterpretation of that information.

~ ~~~ ~

From the Division of General Internal Medicine and Health Care Research, Case Western Reserve University and University Hospitals of Cleveland, Cleveland, Ohio

MEDICAL CLINICS OF NORTH AMERICA

VOLUME 80 - NUMBER 1 -JANUARY 1996 63

64 WOLPAW

It is particularly important to be able to sort out what is known versus what makes sense and to define as clearly as possible the question that is being asked.

The nature of the question is a key issue. Early detection, screening, and case finding are not equivalent activities. Early detection is an approach that may be applied in a variety of settings. It is a process that relies on a disease with an adequate window of opportunity for preclinical detection and on the availability of acceptable diagnostic tools. Terms such as screening and case finding define the specific setting and expecta- t i o n ~ ~ ~ for applying early detection (Table 1). Screening is a program offered to a large group of apparently healthy individuals as a sensitive tool for risk stratification in a disease in which earlier detection provides for easier, more effective treatment. There is implied benefit to the volunteering individual and minimal chance of harm. Case finding occurs when a patient seeks care for concerns or symptoms that may or may not be related to the target disorder. Such a visit generally carries the expectation of comprehensive health management. In this context, approaches to testing can and should be individually negotiated and discussed between patient and physician with clear understanding of risk and benefit.

Both screening and case finding use the process and requirements of early detection. With this in mind, the author addresses the applicability of early detection to lung cancer, discusses the available data on how it has been evaluated, and brings this information into focus for the individual patient in the office. An examination of the process of early detection must include consideration of tumor characteristics, including incidence, impact, risk groups, and biological behavior (latency), and test characteristics, including sensitivity, specificity, predictive value, cost, and acceptability. Moving from this process to a broadly applied program such as screening or an individualized intervention such as case finding requires studies of early detection and an evaluation of treatment options and applicability and benefits-improved disease-specific mortality and reliable disease-free reassurance outweighing the harm of false- positive results.

Table 1. SCREENING VERSUS CASE FINDING

Screening Case Finding

Symptoms

Contact/motivation cost Compliance/acceptance Benefit Sensitivity Specificity Predictive value

None

Physician/health system/policy lnsurancelsystem Variable Implied Crucial Moderately important Moderately important

Variable Not necessarily related to cancer Patient-initiated Patienthnsurance High Individually negotiated/discussed Very important Crucial Crucial

EARLY DETECTION IN LUNG CANCER 65

It is important to keep in mind that this process of evaluation is beset by multiple sources of bias. These include lead-time and length- time bias, overdiagnosis, and selection bias. These are significant con- founders in the available literature and must be kept in mind when reviewing everything from test characteristics to survival versus mortality results.

SCOPE OF THE PROBLEM

Lung cancer is the leading cause of cancer-related death for both men and women in the United States. Current estimates73 predict nearly 170,000 new cases of the disease in 1995, along with almost 158,000 deaths. Other statistical highlights include the following:

1. There has been a rapid rise in lung cancer deaths among women since 1960, whereas the age-adjusted death rates in men have leveled off over the last 10 years. U.S. women currently rank second in the world at 24.7 deaths per 100,000 population (Scot- land is first at 28 per lOO,OOO).”

2. It is estimated that greater than 90% of people who acquire the disease die from and each individual who succumbs loses an average of 15 years of life.33

3. Five-year survival remains dismal, although it has improved from 8% to 14% for whites between 1960 and 1990, while moving from 5% to 11% for Afri~an-Americans.~~ This improvement has mainly been ascribed to improved surgical techniques.

Clearly, lung cancer is a common and deadly problem and a reasonable disease target for efforts to improve outcome through early detection.

POPULATIONS AT RISK

Identifying populations at risk is an important step both in defining the applicability of early detection and in applying it clinically. Im- portant factors include the following:

1. Smoking. Relative risk reaches levels of 20 times or more for 2-packs-per-day smokers as opposed to nonsmokers. It is estimated that 80% of lung cancer occurs in smokers.‘ Although smoking rates have declined overall in the United States, it is estimated that 3000 teenagers become regular smokers each day in the United States.* Second-hand smoke is also a risk, account- ing for an estimated 3000 lung cancer deaths in nonsmoking adults each year.*

2. Age. Although difficult to separate out from smoking and a variety of other environmental factors, the incidence of lung cancer begins to climb rapidly at age 40 and increases dramati- cally between ages 40 and 80.43, 66

66 WOLPAW

3. Preexisting lung disease. This category includes airway obstruc- tion6* and bullous lung disease.28

4. Occupational exposures. These possibly account for up to 13% to 27% of cases of lung cancer.34 Agents include asbestos, radon (miners), alkylating compounds, nickel, chromates, halo ethers, and carcinogenic polyhydrocarbons.

5. Family history. Genetic risk is an intriguing and evolving area. Based on a retrospective generational study of families of patients dying of lung cancer, Sellers and suggest that there may be a genetic predisposition to lung cancer that is expressed in the presence of an environmental stimulus such as smoking.

6. Prior lung cancer. Most likely secondary to an individual's combination of other risk factors, patients with prior lung cancer have a significant risk of second primaries. Estimates place the risk as high as 3% per year.47

DISEASE

Lung cancer is a heterogeneous disease, both in terms of the variety of cell types and in the biologic behavior of any particular cell type. This heterogeneity creates difficulties in designing and evaluating approaches to diagnosis and treatment. Evidence suggests that the rate of growth and spread of any individual tumor can vary considerably,32 making the timing and selection of detection and potential intervention problematic. Despite this heterogeneity, it is useful to keep in mind the major cell types, as follows:

1. Squamous or epidermoid. This histologic type made up approximately one third of cancers in the National Cancer Institute (NCI) trials, but current estimates have dropped to 17%.36 Lesions may be central or peripheral. Malignant cells can be identified in sputum. These tumors are generally most amenable to resection.

2. Adenocarcinoma. This accounted for 27% to 46% of cancers in NCI trials and is currently the predominant cell type with an overall 40% preva len~e .~~ Lesions are usually peripheral. There is a nar- row window of detection before spread (first detectable at approximately 1 cm, likely to spread when greater than 2 ~ m ~ ~ ) . Sputum is not useful with standard techniques.

3. Large cell. Characteristics of this cell type are more similar to adenocarcinoma, and it is grouped with adenocarcinoma in some analyses of the NCI data.2o

4. Small cell. This histology accounted for 16% to 23% of cancers in NCI trials. It is aggressive, not generally resectable, and not considered a target for early detection or screening.

Understanding of the carcinogenic process that results in these histologic types continues to grow. Current models postulate multiple stages of genetic change beginning with pluripotential cells in the respi-


ratory tract and involving antioncogenes, proto-oncogenes, and local production of growth factors.46 Continued elucidation of this process may not only help to explain the observed mixed histology and biologic behavior, but also may contribute to the development of useful tumor markers.

The concept of early detection in screening or case finding relies on a presymptomatic phase of the disease in question that is available for identification and intervention resulting in favorable alteration of the natural history. This process depends on the nature of the disease, the sensitivity of the detection tools, and the applicability and success of treatment options. As described previously, the biologic nature of lung cancer is complex and varied, and understanding of the time intervals involved in any particular presymptomatic phase is hampered by this variability. Non-small cell carcinoma is estimated to have an early-stage detectable period that ranges from 6 months to 12 For any individual tumor, the rate of growth and potential for metastases are not known; from the biologic standpoint, however, there would seem to be a window of opportunity for detecting lung cancer that would lend itself to early detection.

DETECTION TOOLS

Chest X-Ray Film

For many years and through various improvements in technology and technique, the chest x-ray film (CXR) has been the initial diagnostic tool (beyond the history and physical examination) for detecting lung cancer. More sensitive imaging technology has been hampered by availability, acceptability, and expense, and the CXR is likely to remain on the front lines for the foreseeable future. This preeminence has been recognized by the NCI in its decision to look at the usefulness of CXR screening for lung cancer over the next 15 years as part of the Prostate, Lung, Colon and Ovary (PLCO)

CXR certainly fulfills many of the test characteristics required for screening. It is widely available, relatively inexpensive ($70 including reading for posteroanterior and lateral outpatient), acceptable to patients, and without significant radiation exposure or other morbidity. Other attributes, however, remain in question: What is the sensitivity, specificity, and predictive value? In other words, if the patient or physicians elect to obtain this test to look for lung cancer, what does the result mean? The answer remains unclear.

The sensitivity of the CXR in lung cancer detection is variably estimated from as high as 85% to as low as 16%, depending on whether the calculation is based on prevalence data? longitudinal incidence

65 or mathematical m~deling.'~ In their evaluation of a large retrospective database (approximately 300,000 subjects), Soda and col- leaguesh5 identified an overall sensitivity of 70%. This figure varied

68 WOLPAW

depending on tumor histology (adenocarcinoma 85%, squamous 52%, large cell 75%), and it is notable that centrally located squamous cell carcinoma was excluded from the analysis. The key to any sensitivity estimation is an accurate identification of false-negative films, and this can be obtained only through accurate follow-up information. In the Memorial Sloan-Kettering arm of the NCI trials, up to 80% of early- stage adenocarcinoma and large cell cancer had retrospective evidence of a lesion on a prior CXR, some up to 4 years before diagn~sis. '~ A significant number of advanced-stage cancers identified between screen- ings are not retrospectively identifiable, indicating rapidly growing le- s i o n ~ . ~ ~ Thus, although test sensitivity in a prevalence or one-time application may be lower for more indolent lesions, cumulative opportunity for identification of these lesions while still at an early stage is worth noting. These detection characteristics leave the door wide open for lead-time and length-time bias in studies of screening-that is, slow- growing, less aggressive tumors are more likely to be detected on repeated CXRs. Despite the fact that adenocarcinoma tends to be peripheral and more localizable by CXR, lesions are unlikely to be prospec- tively identified before they are 1 cm and are at significant risk (>50%) for metastasis by the time they reach 2 74 By the time of successful CXR imaging, these tumors have been estimated to have undergone more than 20 doublings and have a high potential for metastasis.1°

Estimates of specificity for the CXR range from 89'/0~~ to 99'/0.~~ Although specificity (as pointed out in Table 1) is of greater concern in case finding than in screening, there has been considerable emphasis focused on the consequences of a false-positive examination.l2S 52 In the NCI trial, there were 3361 false-positive films in 31,360 patients (10.7%). In a large percentage of these, the only further evaluation needed was a repeat CXR or review of an old film; however, a significant number required invasive diagnostic testing, and at least two patients died after thoracotomy for a benign l e~ ion .~

What does a single negative CXR mean? The data available from the NCI and Japan would indicate a high negative predictive value (>99%), even in a high-risk population of smokers. This figure goes down slightly with lower estimates of sensitivity and specificity but still remains greater than 95%. Note that predictive value is highly prevalence dependent and is different in an extremely high-risk group, such as asbestos workers. It is instructive to look at the impact of a test such as the CXR on the pretest probability that the patient has no diseuse. As shown in Table 2, the average smoker walking into a physician's office is likely already to have a 99% or greater chance of not having lung cancer (likelihood of no disease). Performing a CXR on this patient most likely changes this disease-free risk to at most 99.5% (50% sensitivity). If this patient is at higher risk, the impact of the CXR is greater but still not dramatic. The calculations in Table 2 are based on a specificity

What about a positive or suspicious CXR? Estimates of positive predictive value range from 1% to 23y0.'~. 40 In other words, there is a

of 90%.


Table 2. THE CHEST X-RAY FILM IN LUNG CANCER DETECTION: NEGATIVE PREDICTIVE VALUE

Before Test: After 20% Sensitivity Test: After 50% Sensitivity Test: Prevalence* Likelihood of Likelihood of No Disease Likelihood of No Disease

(%) No Disease (Yo) (W (%)

1 99 99.2 99.5 5 95 96 97.5

10 90 92 95

*Prevalence estimates are difficult. Prevalence screening in the National Cancer Institute trials revealed 223 cancers in 31,360 male smokers >45 years old (0.7%).45 True point prevalence must be higher (i.e., tests were not 100% sensitive). Over the 11 years of the combined trials, 1340 cancers were diagnosed in 29,638 patients for a cumulative prevalence of 4.5%. Figures in extremely high-risk groups (e.g., smoking asbestos workers) are undoubtedly higher but difficult to pin down.

Note: Calculations are based on a specificity of 90%.

1% to 23% chance that an abnormal CXR results in a diagnosis of lung cancer.

Considering these test characteristics, it would seem that 1. The average smoker walking into his or her physician’s office or

signing up for cancer screening has a low probability of having lung cancer, most likely less than 1%. A routine CXR done with the stated or implied goal of pronouncing the patient free of lung cancer does not change this probability significantly enough to fulfill that promise.

2. A positive CXR carries a low but extremely variable risk of true malignancy, depending on the risk potential of the individual under evaluation. Consequences of the resulting false-positive results vary but can be significant.

3. Although the CXR is an imperfect detection tool, its cost and patient acceptance profile make it a candidate for further study.

Sputum Evaluation

The other major tool of detection is sputum evaluation. This technique uses cytologic examination of induced or 3-day pooled collections of spontaneously produced material and currently costs about $90. It was intensively studied in the NCI trialsz6 and was found to be both sensitive and specific for squamous cell carcinoma. The sensitivity fell off considerably, however, when applied to all histologic types. Although specificity was excellent, workup of a false-positive sputum evaluation (without CXR findings) can be extensive. Newer techniques using monoclonal antibody recognition have shown promise for improving sensitiv- ity69 but have not been applied in a clinical trial that could demonstrate early detection efficacy. Considering that all lung cancer is believed to develop from bronchial epithelial cells,58 sputum remains an attractive potential vehicle for early detection and for the use of biomarkers in the

70 WOLPAW

growing field of chem~prevention.~" Reliable markers, whether cellular or molecular, could potentially shortcut the need for long-term randomized trials to determine the potential efficacy (reversal of carcinogenesis) of such early interventions as antioxidant vitamins and DFMO (an ornithine decarboxylase inhibitor).8, 30, 4x

It should be pointed out that in all the NCI trials, cancer diagnosed by sputum alone was almost invariably early stage and highly resectable. The addition of sputum evaluation to regular CXR screening, however, made no difference in cancer mortality. Several explanations have been advanced for this finding, including an unexpectedly low prevalence of squamous cell carcinoma and the indolent nature of these tumors, which is likely to allow presentation on CXR while still resectable. This has led some authors to suggest that sputum may be more useful as a one-time detection tool than as a periodically applied test.43

In its current form, sputum cytology does not appear to be a useful early detection tool. Much of this information comes from the NCI trials. Ongoing research and new techniques may change this conclusion.

TREATMENT

Pulmonary resection has remained the treatment of choice for early- stage lung cancer for many years. Advances in preoperative staging and surgical techniques have improved outcomes, but it remains an invasive treatment approach with significant morbidity. There is also the problem of early subclinical spread and second primary cancers-for example, new tumors developed in 21 of 122 patients completely resected in the NCI trials, and at least 11 of those 21 died of that di~ease.~ Nonsurgical approaches are focusing on new radiotherapy and chemotherapy techniques and combinations, including neoadjuvant treatment.", 56, To this point, however, nonsurgical therapies have remained palliative in nature. Endobronchial approaches to identification and laser treatment of early cancer lesions continue to be e~plored.~, l4

As mentioned previously, the whole area of chemoprevention is receiving considerable interest and attention. Although early results such as those reported in the ATBC trial (Apha-Tocopherol, Beta Carotene Cancer Prevention were disappointing, studies are 0ngoing,2~, 50

and molecular and genetic insights continue to grow. At the present time, generally available treatment options remain

limited and suboptimal. This is a crucial factor in the whole area of early detection and screening. The most amenable disease behavior linked with the most sensitive detection tools still requires available, effective therapy for a successful outcome. One of the concerns about identifying increasingly sensitive markers of premalignant lesions is that there is no standard treatment approach for these abn~rmalities.~~


EVIDENCE-STUDIES OF EARLY DETECTION

Taking into account these essential components (i.e., a significant disease with identifiable risk groups and a potential preclinical window, acceptable (questionably) early detection tools, and a potentially curative treatment approach), the next step is to look at the evidence that a screening (or case finding) program can be of benefit. Although most of the major studies have focused on the context of screening, the Mayo Clinic arm of the NCI trials recruited participants from the regular outpatient practice and consequently may more closely resemble a case- finding situation. The key is to focus on data that eliminate the variotis sources of bias through both study design and data analysis. The ideal study would

1. Eliminate selection bias through randomization of a recognizable and applicable subject group (i.e., for screening use volunteers, for case finding use practice patients) with adequate controls for known risk factors.

2. Include study and follow-up periods of adequate lengths to take into account the biologic behavior of the disease and decrease the impact of lead-time and length-time bias as well as overdiagnosis. As described by Fontana and col leag~es,~~ the successful screening intervention would be expected to increase the number of early-stage cancers diagnosed during early detection interventions and provide for adequate follow-up to reveal a corresponding increase in later-stage cancers in the control group (missed because of no screening). The final totals of the screened and intervention groups would ideally be equal; screening would improve disease-specific mortality because of the successful treatment interventions for early-stage disease.

3. Judge results by differences in disease-specific mortality rather than length of survival after diagnosis. Length of survival after diagnosis is subject to lead-time, length-time, and overdiagnosis bias, while appearing convincing because of its consonance with the intuition that earlier diagnosis is better.

4. Be powerful enough to detect significant differences in mortality. This includes consideration of overall numbers as well as such issues as contamination and compliance. For example, control patients may have routine CXRs, which are equivalent to screening (contamination). Noncompliance with the screening intervention results in a group of patients who must be counted in the intervention group to limit selection bias but acts to dilute any potential impact of the screening.

There have been many attempts to study the effectiveness of screening for lung cancer (Table 3), beginning in the 1950s with trials from Philadelphia, London, and the Veterans Administration and continuing through the large multicenter NCI trial of the 1970s and 1980s and the more recent Czechoslovakian randomized and German Democratic

3

Tabl

e 3. S

TUD

IES

OF

LUN

G C

AN

CE

R E

AR

LY D

ETE

CTI

ON

Stu

dy

Tool

VA

Stu

dy

Spu

tum

plu

s C

XR

q

6 m

onth

s P

hila

delp

hia

Neo

plas

m

CX

R e

very

6 m

o

Lond

on M

ass

Rad

iolo

gy

CX

R e

very

6 m

o

NC

I Tria

ls (C

ELC

DP

) P

reva

lenc

e M

ayo

Lung

Pro

ject

sc

reen

, C

XR

pl

us s

putu

m

ever

y 4

mo

NC

I Tria

ls

Pre

vale

nce

John

s H

opki

ns a

nd

scre

en;

CX

R

Mem

oria

l Slo

an-

ever

y ye

ar;

Ket

terin

g Lu

ng

sput

um e

very

4

Pro

ject

s m

o C

zech

oslo

vaki

an T

rial

Pre

vale

nce

scre

en; C

XR

pl

us s

putu

m

ever

y 6

mo

GD

R

CX

R (

part

of

tube

rcul

osis

sc

reen

ing

prog

ram

)

PLC

O

CX

R: i

nitia

l plu

s ye

arly

x

3 yr

Con

trol

Non

e

Non

e

Yes

; si

mila

r po

pula

tion

of

25,0

00

reco

mm

enda

tion

for y

early

CX

R

plus

spu

tum

X

l

Yes

; C

XR

eve

ry

year

Yes

: sta

ndar

d

Yes

; CX

R p

lus

sput

um a

t 3

year

s

4 co

ntro

ls fo

r ea

ch

lung

can

cer

case

; 2 in

sam

e ar

ea; 2

from

ho

spita

l em

erge

ncy

room

Y

es;

rand

omiz

ed

vs. u

sual

car

e

Des

ign

Non

rand

omiz

ed

Non

rand

omiz

ed.

Com

pare

to in

terv

al

dete

cted

cas

es

Non

rand

omiz

ed

Ran

dom

pro

spec

tive

stud

y of

sm

oker

s >4

5; p

reva

lenc

e sc

reen

Ran

dom

pro

spec

tive

stud

y of

sm

oker

s >4

5; p

reva

lenc

e sc

reen

Ran

dom

ized

tria

l

Cas

e-co

ntro

lled

stud

y

Ran

dom

men

plu

s w

omen

60-

74

yr

No.

14,6

00

61 00

29,4

00 s

cree

ned;

25

,000

con

trol

921 1

20,4

27

6346

130

148.

000

Pow

er (

20%

D

ecre

ase

in

Res

ult

Mor

talit

y)’

Ref

eren

ce

13%

rese

ctab

le; 1

2%

0.38

39

No

chan

ge 5

-yr

0.32

6

3-yr

sur

viva

l

surv

ival

No

sign

ifica

nt

0.39

7

No

chan

ge in

0.

5 4

diffe

renc

e in

5-y

r su

rviv

al

mor

talit

y.

Sig

nific

antly

im

prov

ed 5

-yr

surv

ival

mor

talit

y.

Sig

nific

antly

im

prov

ed 5

-yr

surv

ival

mor

talit

y; c

ance

r de

tect

ed m

ore

ofte

n be

fore

sym

ptom

atic

deat

h 0.

88-1

.09;

no

si

gnifi

cant

diff

eren

ce

No

chan

ge in

0.

5 4

No

diffe

renc

e in

0.

39

36, 3

8, 7

2

Rel

ativ

e ris

k of

can

cer

11

In p

rogr

ess

>0.9

9 (0

.89

49

for

10%

de

crea

se in

m

orta

lity)

Daf

a fr

om N

atio

nal C

ance

r In

stitu

te: C

once

pt p

aper

: Pro

stat

e. lu

ng, c

olo-

rect

al a

nd o

varia

n ca

ncer

scr

eeni

ng tr

ial.

1991

C

XR

= C

hest

x-r

ay.


Republic case-control studies. These inquiries have been summarized in a number of excellent reviews,'2, 20, 40, 52 and salient features are listed in Table 3. The NCI trials, often collectively referred to as the Cooperative Early Lung Cancer Detection Program, consisted of major randomized trials at three sites-Johns Hopkins, Memorial Sloan-Kettering, and the Mayo Clinic. The Johns Hopkins and Memorial Sloan-Kettering arms were similar in design and execution, whereas the Mayo Lung Project (MLP) varied in intervention and control characteristics. The MLP is worthwhile examining closer because this study most closely addresses the issue of screening versus not screening and demonstrates a number of important aspects of the screening problem.

MAY0 LUNG PROJECT

The MLP2", 24, 25 enrolled 10,933 Mayo Clinic outpatients between 1971 and 1976. Targeting a high-risk group, subjects were required to be men greater than 45 years old who smoked at least 1 pack of cigarettes per day. These patients were tested with CXR and 3-day pooled sputum in an attempt to identify preexisting cancer cases (prevalence screening). Ninety-one cases were diagnosed, with prevalence rising from 1 per 1000 in the 45 to 49 age group to 17 per 1000 for those older than 65 years.22 These patients were excluded from the longitudinal screening phase of the study as were patients who had a life expectancy of less than 5 years or were considered poor risks for potential surgical treatment (pulmonary resection). This left 9211 study patients. These individuals were randomly assigned to a screening group and a control group. The highlights of the 11-year study are depicted in Figure 1.

The MLP results are worth reviewing carefully. They have been the subject of considerable discussion and argument over the last 10 years, and the debate continues. The following points are worth noting:

1. An excess of 46 cases of lung cancer were diagnosed in the screened group, almost all early-stage, resectable squamous and adenocarcinoma, with a few advanced-stage adenocarcinoma.25 The numbers of late-stage tumors were almost identical in the two groups (112 in screened, 109 in control). The expected pattern of the successful screening intervention (i.e., greater numbers of early-stage cancer identified in the intervention phase, balanced by increasing or equalizing numbers of late-stage tumors in the control group) never appeared in the 11 years of this study. Interestingly enough, results from the Johns Hopkins and Memo- rial Sloan-Kettering arms of the NCI trials showed the same equality between study and control arms in the numbers of advanced-stage cancer.25

2. Of the 116 cancers in the intervention group diagnosed outside of routine screening, about 44% were found in the group that did

74 WOLPAW

Overall compliance = 75% Follow-Up (n=4606)

I 1 9211 Patients

I Follow-Up (17-4579) 1

c I Dlaanosed bv ScreeninQ I i Diaanosed Olrtside Screeninn i i 112 no%, with SvmDtoms I

116 (56.3%) (73 with symptoms;

- 90 (437%) - without symptoms

Between

(Deaths per I000 Patient-Years)l 3 2

I 48 (30%) kithout Symptoms 1

No Screenin 5-Year Survival Unreseclabie Cancer

Lung Cancer Death Rate

Figure 1. The Mayo Lung Project: data summary.

not comply with screening (25% of screening group).22 This compromises the power of the study to detect a decrease in mortality.

3. Approximately 30% of lung cancers diagnosed in the control group were found on screening CXRs (e.g., routine physical examinations, preoperative evaluation). About 50% of the control group had CXRs obtained on a yearly basis unrelated to cancer symptoms, and an estimated 80% received them over a 2-year period.25 This type of contamination of the control group reflects not only the common use of the CXR in medical culture, but also the fact that patients at risk for lung cancer are predisposed to other illnesses that prompt CXR evaluation (e.g., pneumonia). It blurs the distinction between intervention and control groups and directly compromises the power of the study.

4. Long-term (up to 11 years) follow-up of cases reveals screen- detected survival greater than 35% (sputum 50% and CXR 35%) and symptom-detected survival of 10%. These unfavorable survival data for symptom-detected patients are consistent through the literature.

The end result is that however seductive the survival data may be,


even intensive screening (every 4 months) failed to detect many cancers before they were considered unresectable. The number of late-stage cancers in the control group never rose to account for early-stage cancers missed because of no screening. This has often been ascribed to noncompliance, contamination, bad luck in the percentages of squamous cell cancer (detectable by sputum and more amenable to resection), and baseline insufficient power to detect a small but significant (10% to 20%) decrease in mortality.

The question of overdiagnosis bias has also received considerable attention-that is, the risk of identifying lesions on screening that never become clinically significant. The implication for the MLP is that the 46 excess cancers were in fact either not morbid lesions or, alternatively, so indolent that patients with significant health risks (i.e., smokers) were more likely to die of other causes before the cancers became a problem. Flehinger and colleagues18 attempted to examine this issue based on the NCI data. They looked at survival of early-stage cancer patients who did not have surgery secondary to refusal or medical contraindications. Forty-five patients fit into this category, 28 with surgical contraindications and 17 who refused. Five-year survival averaged 70% in the 434 early-stage patients undergoing surgery and 10% in those patients not resected. Results were recalculated after eliminating nodal status (which can change after surgery) without significant change in these findings. Sobue and coworkersW approached the same question somewhat differ- ently based on Japanese screening data and arrived at the same conclusion-overdiagnosis does not play a significant role in lung cancer studies.

Another potential factor in these results relates to the natural history of the disease. It is possible that missed early-stage cancers never blos- somed into unresectable tumors in the control group because the follow- up period was too As described earlier, the duration stage I for any particular patient or cancer is not known, but estimates of 4 to 12 years and beyond for adenocarcinoma have been advanced.I7 In addition, as pointed out earlier, early-stage cancers picked up on CXR are more likely to be slower growing. No study to date has used a washout period of more than 5 to 6 years. The ongoing PLCO is designed with a 3-year screening phase and a 7- to 10-year follow-up phase.

DISCUSSION

Argument for Early Detection Programs

As presented by a number of authors,31,59 the case for early detection

1. 40% of lung cancers can be detected on CXR in stage I. 2. 70% to 80% of resected stage I disease results in long-term sur-

efforts can be summarized as follows:

vival.

76 WOLPAW

3. 5-year survival in screened patients (NCI trials) was 35% com- pared with 13% for the United States as a whole.

4. Percent stage I cancer in the screened group dropped from 41% to 20% in the postscreen follow-up period, with a corresponding drop in 4-year survival.

5. Incidentally diagnosed cancer (interval CXR or postscreening CXR) had survival data similar to screen-detected cases, whereas symptomatically diagnosed cases dropped to the national 13% range.

6. The only large randomized trial comparing screening to no screening is the Mayo Clinic arm of the NCI trial, and this study lacks sufficient power to detect a 10% to 20% decrease in mortality, which could save 17,000 to 34,000 lives by 1994 data."

7. Mathematical modeling using NCI data projects a mortality benefit of up to 13% to 18%.17,1y

McDougal14' notes that it is "counterintuitive to believe that early treatment of at least some types of lung cancer will not decrease mortality." Despite the fact that these arguments rely heavily on detection and survival data and all the potential bias that accompanies them, they continue to be seductive when combined with this intuition. This can be viewed as the "evangelist" approach described by Sackett and Holland.54

Argument Against Early Detection Programs

There is the perspective of the "snail" or "method~logist."~~ It has been addressed in several notable reviews'*, 2i, 4u and relies on what is known about the disease, the tools, the treatment, and screening itself.

1. Lung cancer is a highly variable disease even within the same histologic classification, and little hard data are available on the window of opportunity for effective detection, potential revers- ibility, or curative treatment. EddyI2 concludes that, "The nature of lung cancer appears to be such that the probability or timing of death from lung cancer is not influenced by the timing of detection and treatment within this time window."

2. Currently applicable detection tools using CXR and sputum are limited in sensitivity and specificity. The combination of low sensitivity (probably <%YO) and low point prevalence makes a negative CXR almost meaningless, and even an optimistic specificity of 90% creates significant levels of false-positive results, with resulting emotional, physical, and economic costs.12 These characteristics threaten two principles of screening: implied benefit (balancing reassurance and decreased mortality) and minimal harm.

3. The available data from studies of various designs over the last 40 years demonstrate no beneficial effect of screening on lung cuncer


mortality. It may be that all these studies suffer from problems of design, insufficient power, and overcontamination or that mathematical modeling indicates otherwise. It also may be that regular CXRs do have a small favorable impact on mortality but that a formal program of screening does not add to the baseline level of testing routinely done in this risk population. The bottom line at this point is that the data do not support screening. Over the course of 11 years, the NCI trials demonstrated increased detection but not increased cure. The temptation to focus on more favorable staging at the time of diagnosis and improved survival statistics can be misleading secondary to unavoidable bias. Once again, in the combined arms of the NCI trials, despite significantly increased resectability (>50%) and improved 5-year survival in the intervention groups, there was no difference in lung cancer mortality rates.

4. The following organizations do not recommend screening: Ameri- can Cancer Society, American College of Radiology, National Cancer Institute, U.S. Preventive Services Task Force, Canadian Task Force on the Health Examination.

CONCLUSIONS AND SUMMARY

There is general agreement that the most effective approach to lung cancer is primary prevention-stop smoking. Richards5’ has proposed the MVROCST-the Monosyllabic Verbal Response Office Cancer Screening Test: “Do you smoke?” If “yes,” intervene. If ”no,” move on. Ample evidence exists that a clear message from a physician to a patient about the importance of stopping smoking makes a differen~e.~~ In contrast to the maze of arguments and data on early detection, this is something that each physician clearly can and should do. A reduced risk for lung cancer may begin as early as 5 years after cessation of cigarette use.36 H ~ u s k o n e n ~ ~ has proposed conceptualizing screening as a coordinated intervention with the goal of identifying populations at risk and working to modify that risk. Primary prevention should be central to any efforts to reduce mortality from lung cancer, and attention to this area needs to increase despite the difficulties and frustration. Despite declining percentages of smokers in the population as a whole, it is estimated that more than 3000 teenagers become regular smokers each day in the United States.* In this environment, the question of whether to recommend a CXR or sputum for early detection is not going to disappear in the near future.

The NCI has recognized the persistent and important nature of this debate and is currently funding the Prostate, Lung, Colon and Ovary Cancer Screening Trial.4y This is a large and powerful randomized study of men and women aged 60 to 74. The lung cancer arm is designed to look at the usefulness of a yearly CXR intervention in reducing cancer- specific mortality. The overall power of the study (based on national

78 WOLPAW

mortality data) is 0.99 for a 15% reduction in lung cancer mortality and 0.89 for a 10% reduction, with differentially better sensitivity in men than women. The study is currently in progress at multiple sites and will be completed over the next 12 to 14 years.

In the meantime, what is the right approach? It is useful in considering this question to return to the concepts of early detection, screening, and case finding.

1. Early detection in lung cancer remains a concept of uncertain applicability because of the unknowns and variability in the natural history of the disease. The available, accessible, and acceptable detection tools appear to be inadequate by current evidence. This is not a static field, however, and new work in the area of biomarkers carries promise for significantly more sensitive and specific techniques. Tockman and colleagues71 conclude that early detection is concep.pttially sotind, although not currently practical, and further research may expand the role of intervention. In the end, a judgment on early detection in lung cancer must be linked to the proposed setting-screening or case finding.

2. Screening, defined as the application of a test to the general population to define disease risk further with the implied benefit of improved treatment and 0utcome,5~ cannot be recommended for lung cancer. This is the perspective of the major organizations cited previously, and it is based on admittedly imperfect but nonetheless convincing data.

3. Case finding, the situation of the patient who seeks care and is available for informed discussion and negotiation on possible testing, is a potentially different situation. EddyI2 argues that the systematic application of a recommendation in the case of certain patients (e.g., smokers over 50 years) amounts to large-scale screening. Although this may be true if a recommendation is applied blindly to every patient of a particular profile, it does not take into account the real difference in the concept of case finding, particularly the potential for discussion on pros and cons and elicitation of patient preferences. When the evidence does not support the general application of a test, individuals must decide-individual patients and their physicians.** F0ntana,2~. one of the principal investigators in the NCI trials (MLP) and an opponent of mass screening, expresses concern over the complete elimination of periodic CXRs in high-risk patients negotiated in a case-finding setting. He raises an interesting and thought- provoking question: “If we are to assume that detection of early stage, potentially resectable lung cancer confers absolutely no mortality advantage, then early diagnosis and treatment of asymptomatic lung cancer presenting as a solitary circumscribed pulmonary nodule (’coin lesion’) cease to be important.”

An interesting and practical aspect of this issue relates to how the


subject of CXR early detection is raised in any particular patient visit. There would appear to be three main possibilities:

1. The physician routinely initiates the discussion of early detection in patients at risk (mostly smokers).

2. The physician engages in this discussion only if the patient inquires.

3. The physician-patient interaction focuses on primary prevention (stopping smoking or occupational exposure), and the risks and benefits of the CXR are incorporated as indicated into the discussion of risk modification (similar model to Huuskonen"*).

When viewed from this perspective, case finding is not a uniform activ- ity. Common sense would argue for the third approach based on the evidence for primary prevention and the evidence against the routine application of early detection tools in lung cancer.

The bottom line on early detection in lung cancer is not known. Clinicians may know in 15 years at the conclusion of the PLCO trial, but the biomedical technology landscape is likely to be different, and questions will undoubtedly remain. The key in practice would appear to be an emphasis on primary prevention and the opportunity for informed discussion between individual patients and their physicians.

ACKNOWLEDGEMENTS

The author thanks Carol Calloway for her help in manuscript preparation and Barbara Juknialis for her editorial assistance. He also thanks Philip Anderson, MD, Kenneth Covinsky, MD, C. Seth Landefeld, MD, and Terry Wolpaw, MD, for their comments on the manuscript.

References

1. American Cancer Society: Cancer facts and figures-1988. Atlanta, American Cancer Society, 1988

2. American Cancer Society: Cancer facts and figures-1994. Atlanta, American Cancer Society, 1994

3. Bailar JC 111: Screening for lung cancer-where are we now? Am Rev Respir Dis 130:541-542, 1984

4. Berlin NI, Buncher CR, Fontana RS, et al: National Cancer Institute's Cooperative Early Lung Cancer Detection Program: Results of the initial screen (prevalence). Am Rev Respir Dis 130545-570, 1984

5. Berlin NI, Buncher CR, Fontana RS, et al: The National Cancer Institute Cooperative Early Lung Cancer Detection Program: Early lung cancer detection: Introduction. Am Rev Respir Dis 130:454-549,1984

6. Boucot KR, Weiss W: Is curable lung cancer detected by semiannual screening? JAMA

7. Brett GZ: Earlier diagnosis and survival in lung cancer. BMJ 4:260-262, 1969 8. Castonguay A: Methods and strategies in lung cancer control. Cancer Res

9. Cortese DA: Endobronchial management of lung cancer. Chest 89(suppl 4):234~-236s,

224~1361-1365, 1973

52(~~ppl):2641~-2651~, 1992

1986

80 WOLPAW

10. DeVita VT Jr: The relationship between tumor mass and resistance to chemotherapy: The James Ewing Lecture. Cancer 51:1207-1220, 1983

11. Ebeling K, Nischan P: Nonexperimental evaluation of the effectiveness of a screening program for lung cancer. International Journal of Technology Assessment in Health Care 8:245-254, 1992

12. Eddy DM: Screening for lung cancer. Ann Intern Med 111:232-237, 1989 13. Edell ES, Cortese DA: Bronchoscopic phototherapy with hematoporphyrin derivative

for treatment of localized bronchogenic carcinoma: A 5-year experience. Mayo Clin Proc 629-14, 1987

14. Edes TE, McDonald PS: Retinol and retinol-binding protein in lung cancer screening. Cancer Detect Prev 15:341-344, 1991

15. Epler GR: Screening for lung cancer: Is it worthwhile? Postgrad Med 87181-186, 1990 16. Flehinger BJ, Kimmel M: The natural history of lung cancer in a periodically screened

population. Biometrics 43:127-144, 1987 17. Flehinger BJ, Kimmel M, Melamed MR Natural history of adenocarcinoma-large cell

carcinoma of the lung: Conclusions from screening programs in New York and Balti- more. J Natl Cancer Instit 80:337-344, 1988

18. Flehinger BJ, Kimmel M, Melamed MR The effect of surgical treatment on survival from early lung cancer: Implications for screening. Chest 101:1013-1018, 1992

19. Flehinger BJ, Kimmel M, Polyak T, et al: Screening for lung cancer. Cancer 72:1573- 1580, 1993

20. Flehinger BJ, Melamed MR Current status of screening for lung cancer. Chest Surg Clin North Am 4:l-14, 1994

21. Flehinger BJ, Melamed MR, Zaman MB, et al: Early lung cancer detection: Results of the initial (prevalence) radiologic and cytologic screening in the Memorial Sloan- Kettering Study. Am Rev Respir Dis 130:555-560, 1984

22. Fontana RS: Screening for lung cancer. 1n Miller AB (eds): Screening For Cancer. Orlando, Academic Press, 1985, pp 377-395

23. Fontana RS: Should asymptomatic cigarette smokers have annual chest x-rays after age 55 years? Debates in Medicine. Yearbook of Medicine 3:133-140, 1990

24. Fontana RS, Sanderson DR, Taylor WF, et al: Early lung cancer detection: Results of the initial (prevalence) radiologic and cytologic screening in the Mayo Clinic study. Am Rev Respir Dis 130:561-565, 1984

25. Fontana RS, Sanderson DR, Woolner LB, et a1 Screening for lung cancer. Cancer

26. Frost JK, Ball WC, Levin ML, et al: Early lung cancer detection: Summary and conclusions. Am Rev Respir Dis 130:565-570, 1984

27. Frost JK, Ball WC, Levin ML, et a1 Early lung cancer detection: Results of the initial (prevalence) radiologic and cytologic screening in the Johns Hopkins study. Am Rev Respir Dis 130:549-554, 1984

28. Goldstein MJ, Snider GL, Liberson M, et a1 Bronchogenic carcinoma and giant bullous disease. Am Rev Respir Dis 971062-1070, 1968

29. Goodman GE, Omenn GS, Thornquist MD, et al: The Carotene and Retinol Efficacy Trial (CARET) to prevent lung cancer in high-risk populations: Pilot study with cigarette smokers. Cancer Epidemiol Biomarkers Prev 2389-396, 1993

30. Greenwald P, Kelloff G, Burch-Whitman, et al: Chemoprevention. CA Cancer J Clin 45:3149, 1995

31. Heelan R: Lung cancer imaging: Primary diagnosis, staging, and local recurrence. Semin Oncol 18:87-98, 1991

32. Heelan RT, Flehinger BJ, Melamed MR, et al: Non-small cell lung cancer: Results of the New York Screening Program. Radiology 151:289-293, 1984

33. Horm JW, Sondik EJ: Person-years of life lost due to cancer in the United States, 1970 and 1984. Am J Public Health 79:1490-1493, 1989

34. Huuskonen MS: Screening for occupational cancer. Scand J Work Environ Health

35. Ihde DC, Lippman S Intervention approaches in early lung cancer and premalignant

67(~~~~1):1155-1164, 1991

~ ~ ( S U P P ~ 1):110-114, 1992

disease: Breakout group report. Cancer Res 52(suppl):2767~-2768s, 1992


36. Ihde DC, Minna JD: Non-small cell lung cancer. Part I: Biology, Diagnosis, and Staging. Curr Probl Cancer 15:64-104, 1991

37. Kubik A, Parkin DM, Khlat M, et al: Lack of benefit from serni-annual screening for cancer of the lung: Follow-up report of a randomized controlled trial on a population of high-risk males in Czechoslovakia. Int J Cancer 4526-33, 1990

38. Kubik A, Polak J: Lung cancer detection: Results of a randomized prospective study in Czechoslovakia. Cancer 12:2427-2437,1986

39. Lilienfeld A, Archer PG, Burnett CH, et al: An evaluation of radiologic and cytologic screening for early detection of lung cancer: A cooperative pilot study of the American Cancer Study and the Veterans Administration. Cancer Res 26:2083-2121, 1966

40. Marfin AA, Schenker M: Screening for lung cancer: Effective tests awaiting effective treatment. Occup Med State of the Art Reviews 6:111-131, 1991

41. McDougall JC: Lung cancer: To screen or not to screen. Arch Intern Med 154:945, 1994 42. Melarned MR, Flehinger BJ: Screening for lung cancer. A m Intern Med 111:764-765,

1989 43. Melarned MR, Flehinger BJ: Early lung cancer as a potential target for chemopreven-

tion. J Cell Biochem 17F(suppl):57-65, 1993 44. Menck HR, Garfinkel L, Dodd GD: Preliminary Report of the National Cancer Data-

base. CA Cancer J Clin 41:7-18, 1991 45. Mulshine JL: Overview: NCI workshop on investigational strategies for detection and

intervention in early lung cancer. Cancer Res 52(suppl):2639~-2640s, 1992 46. Mulshine JL, Linnoila RI, Jensen SM, et al: Rational targets for the early detection of

lung cancer. J Nat Cancer Instit Monogr 13:183-190, 1992 47. Mulsliine JL, Tockman MS, Smart CR Commentaries: Considerations in the develop-

ment of lung cancer screening tools. J Nat Cancer Instit 81:900-906, 1989 48. Mulshine JL, Tockman MS: Diagnostic research Breakout group report. Cancer Res

52(suppl):2766s, 1992 49. National Cancer Institute: Concept paper: Prostate, lung, colo-rectal and ovarian cancer

screening trial. 1991 50. Omenn GS, Goodman GE, Thornquist MD, et al: The Carotene and Retinol Efficacy

Trial (CARET) to prevent lung cancer in high-risk populations: Pilot study with asbestos-exposed workers. Cancer Epidemiol Biomarkers Prev 2:381-387, 1993

51. Richards JW Jr: Cancer screening in the office. Am Fam Physician 431528, 1991 52. Richert-Boe KE, Humphrey L L Screening for cancers of the lung and colon. Arch

Intern Med 152:2398-2404, 1992 53. Russell MA, Wison C, Taylor C, et al: Effect of general practitioners’ advice against

smoking. BMJ 2:231-235, 1979 54. Sackett DL, Holland WW: Controversy in the detection of disease. Lancet ii:357-359,

1975 55. Sagawa M, Saito Y, Takahashi S, et al: Clinical and prognostic assessment of patients

with resected small peripheral lung cancer lesions. Cancer 66:2653-2657, 1990 56. Salazar OM: Combined modality treatment of small cell lung cancer. Chest 96(suppl

1):74S78S, 1989 57. Sellers TA, Potter JD, Bailey-Wilson JE, et al: Lung cancer detection and prevention:

Evidence for an interaction between smoking and genetic predisposition. Cancer Res 5 2 ( ~ ~ ~ ~ 1 ) : 2 6 9 4 ~ - 2 6 9 7 ~ . 1992

\ , I ,

58. Shepherd FA: Screening, diagnosis, and staging of lung cancer. Curr Opin Oncol 5~310-322, 1993

59. Shields TW: Screening, diagnosis, and staging of non-small cell lung cancer and consideration of unusual primary tumors of the lungs. Curr Opin Oncol 4:299-307, 1992

60. Shields TW: Screening, staging, and diagnostic investigation of non-small cell lung cancer patients. Curr Opin Oncol 3:297-305, 1991

61. Skillrud DM, Offord KP, Miller DR Higher risk of lung cancer in chronic obstructive pulmonary disease. Ann Intern Med 105:503-507, 1986

62. Smart CR Annual screening using chest x-ray examination for the diagnosis of lung cancer. Cancer 72:2295-2298, 1993

82 WOLPAW

63. Sobue T, Suzuki T, Matsuda M, et al: Sensitivity and specificity of lung cancer screening in Osaka, Japan. Jpn J Cancer Res 82:1069-1076, 1991

64. Sobue T, Suzuki T, Matsuda M, et al: Survival for clinical stage I lung cancer not surgically treated: Comparison between screen-detected and symptom-detected cases. Cancer 69:685-691, 1992

65. Soda H, Tomita H, Kohno S, et al: Limitation of annual screening chest radiography for the diagnosis of lung cancer. Cancer 72:2341-2346, 1993

66. Sondik E (ed): Annual Cancer Statistics Review, 1988. Washington, DC, Division of Cancer Prevention and Control, National Cancer Institute, 1989

67. The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group: The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med 330:1029-1035, 1994

68. Tockman MS, Anthonisen NR, Wright EC, et al: Airways obstruction and the risk for lung cancer. Ann Intern Med 106:512-518, 1987

69. Tockman MS, Gupta PK, Myers JD, et al: Sensitive and specific monoclonal antibody recognition of human lung cancer antigen on preserved sputum cells: A new approach to early lung cancer detection. J Clin Oncol 6:1685-1693, 1988

70. Tockman MS, Gupta PK, Pressman NJ, et al: Considerations in bringing a cancer biomarker to clinical application. Cancer Res 52(supp1):2711~-2718s, 1992

71. Tockman MS, Mulshine JL, Smart CR To the Editor. Ann Intern Med 111:765-766,1989 72. Walter SD, Kubik A, Parkin M, et al: The natural history of lung cancer estimated from

the results of a randomized trial of screening. Cancer Causes Control 3:115-123, 1992 73. Wingo WA, Tong T, Bolden S Cancer statistics, 1995. CA Cancer J Clin 452-30, 1995 74. Woodring JH: Pitfalls in the radiologic diagnosis of lung cancer. Am J Radio1 1541165-

1175, 1990

Address repvint requests to

Daniel R. Wolpaw, MD Division of General Internal Medicine and Health Care Research

University Hospitals of Cleveland 11100 Euclid Avenue

Cleveland, OH 44106-6033

early detection in lung cancer

Documents