Document RjKQ9exEXRQxKe95v7NpqKe97

'p X 2 .0 3 *7 1226 THE NEW ENGLAND JOURNAL OF M EDICINE May 8, 1986 SPECIAL ARTICLES PROGRESS AGAINST CANCER? J ohn C. Bailar III and Elaine M. Smith A bstract We assessed the overall progress against cancer during the years 1950 to 1982. In the United States, these years were associated with increases in the number of deaths from cancer, in the crude cancer-related mortality rate, in the age-adjusted mortality rate, and in both the crude and the age-adjusted incidence rates, whereas reported survival rates (crude and relative) for cancer patients also increased. In our view, the best single measure of progress against cancer is change in the age-adjusted mortality rate associ ated with all cancers oombined in the total population. Ac cording to this measure, we are losing the war against can cer, notwithstanding progress against several uncommon forms of the disease, improvements in palliation, and exten sion of the productive years of life. A shift in research em phasis, from research on treatment to research on preven tion, seems necessary if substantial progress against cancer is to be forthcoming. (N Engl J Med 1986; 314:1226-32.) T HE primary purpose of this article is to assess the overall progress against cancer during the years 1950 to 1982, the most recent year for which reliable data are available. During this time there was very rapid and extensive growth of private and governmen tal support of research on cancer, and toward the end of the period there was also substantial emphasis on the effective delivery of research results to physicians, patients, and the public. It is time for an open debate to take stock of past achievements and to consider what levels of funds should be invested in what kinds of future efforts. We offer some observations and inter pretations relevant to such a debate. In 1962, cancer was the recorded cause of death for 278,562 Americans. In 1982, just 20 years later, 433,795 persons died of cancer -- a 56 percent in crease (Table 1). But the population was growing, and the proportions of persons in older age categories were changing. Crude mortality rates, which adjust for population size, increased by 25 percent (from 151.0 to 188.8 per 100,000) in this 20-year period, and ageadjusted mortality rates, which adjust for changes in age distributions as well as population size, increased by only 8.7 percent (from 170.2 to 185.0 per 100,000). Mortality data do not tell the whole story. We might ask, not how many Americans die of cancer, but how many contract the disease. From 1973 to 1981 the crude incidence rate for all neoplasms combined rose by 13.0 percent, and the age-adjusted incidence rate by 8.5 percent (Table 1). Or, we might focus on neither incidence nor mortal ity, but on the long-term survival of patients who have had a diagnosis of cancer. Unadjusted five-year sur vival rates for patients with all forms of cancer com bined increased by 4.2 percent from 1973 to 1978 (from 38.5 to 40.1 percent), while rates adjusted for "expected" mortality from all other causes of death rose by 5.1 percent (from 46.8 to 49.2 percent). From the Harvard School of Public Health. Bostun, and the U niversity of Iowa M edical C enter, Iowa C ity. Address reprint requests to Dr. Bailar at the D epart ment o f BioMalislics. Harvard School of Public Health. Bm tim . MA 0 2 H 5 . Supported in part by grants from the Alfred P. Sloan Foundation, the M obil Corporation, and the University of lo u a Revrarth Foundation Which of these conflicting pictures of change, if any, captures the "truth" about recent advances in the control of cancer? More specifically, what yardstick should we use to measure the overall success of the long and intense effort to control and eventually elimi nate these diseases? Interest in this matter is sharp ened by the recent announcement that the goal of the National Cancer Institute is a 50 percent reduction in cancer-related mortality (on an age-adjusted basis) by the year 2000.3 To answer these questions, we first discuss the kinds of data that are available and the methods used to reduce them to simple index fig ures.'1'5 We then give our views on which measures arc most appropriate arid 'what they indicate. Mortality Data In the United Slates, nearly all national cancerrelated mortality data are derived from death certifi cates submitted through local and state channels to the National Center for Health Statistics. The Death Registration Area has included the entire United Slates since 1933. Major changes since then include five revisions of the standard system for coding causes of death, as well as continual improvement in medical procedures for antemortem diagnosis.'"11 However, these changes have had less effect on the certification of deaths from cancer than on certification of deaths from other major causes. Mortality records can be used in many ways, each of which is best suited for specific purposes. Some times there is a need for information about changes n mortality that are independent of demographic changes such as shifts in the age distribution of the population or shifts in place of residence (for geo graphically related cancers). "Adjusted" rates may be used to remove the effects of the variable or variables adjusted, so that oilier effects can be more easily de tected and measured."1'5 One common method of adjustment for age is the "direct" method, which is a simple weigh led average of observed agc-spccific rates, with weights deter mined by some fixed "standard" population, such as the U.S. population of 1980. For this paper all adjust ments were made by the direct method widi reference Vol. 314 No. 19 PROGRESS AGAINST CANCER? -- BAILAR AM) SMITH H2) Table 1. Cancer in the United States: Selected Measures of Recent Changes.*lo M easure Mortality No. of deaths Crude m et Age-adjusted m et V im m 2 m2 Total * Chance *A v e r a g e C h a n g e /Y * 278.562 433.795 151.0 188.H 170.2 185.1 ms m i 55.7 25.1 8.7 + 7.8 + 1.3 +0.4 plaleaued, and recently began to rise again among white females; rose rapidly and steadily among nonwhile males; and declined slightly .and recently plaicaued among nonwhile females. In all race and sex groups combined, there was a moderate increase in age-adjusted mortality. (The small discontinuity in 1957 is a result of a change in methods of classifying a cause of death on death certificates; more recent changes have had only a minor influence on canccrrclalcd mortality ratcs.B*B) Incidence Crude raect Age-adjusied ratet 365.2 368.2 412.7 399.4 13.0 8.5 + 1.6 + 1.1 Site-Specific Mortality Data Although mortality from all forms of cancer com Five-year survival (ifc)t Absolute survival rate Relative survival n(c I97S 38.5 16.8 1978 40.1 49.2 4.2 5.1 +0.8 + 1 .0 bined provides the most important information, study of specific sites (Fig. 2) can both illuminate the over all changes and show why the siic-spccific analyses alone may be misleading. To preserve comparability 'Sources: MfKiy al..1 ihc National Cent for Health Stannic i,J and unpublKhed data ftom (he SEER Program. National Cancer Institute. across sites, Figure 2 shows rates of each cancer (in cluding the scx-spccific cancers) relative to the total -Ric> arc per 100.000 population. Age adjustment] art to the 1980 U.S. population. Incidence data for 1981 include two areas not in the 1973 data: the base population reflect] this change. population. Rates of breast cancer among women only and rates of prostatic cancer among men only are ap White population only. proximately twice the rates given in this graph. ^Relative to survival o f the U.S. white population with the same age distribution. There has been no apparent change in mortality from breast cancer among white or nonwhite women lo tlie U.S. population of 1980. There has been recent since 1950. Rates among nonwhites (not shown) vary discussion about whether, in view or diagnostic errors about their mean more than the rates among whites, at older ages, age-adjusted mortality rates should in but this appears to be due to the effect of smaller clude the entire age span.9,10 Ours do, because cancer numbers of deaths and, hence, larger random vari is a common cause of death and because (contrary to ability. the situation with some other causes of death) the The sharp and continuing rise in deaths from lung available data do not suggest that net errors are so `cancer (Fig. 2), nearly all from cigarette smoking, is high as to make the figures unreliable for overall eval now widely recognized as a medical, social, and po uation. Furthermore, changes in mortality rales for litical scandal. The increase was evident before 1950 persons in specific age categories may be useful for among white and nonwhite men, and it has been evi understanding causes of cancer, but cannot measure dent among white and nonwhite women since the late progress against cancer in all age groups. We believe that to study overall trends in cancer-related mortality {how they have changed in recent years and how they could change by the year 2000), the best single measure of mortality is the age- adjusted death rate associated with all cancers combined, supplement ed by age-adjusted rates and some times age-specific rates, for specific sex and broad racial categories. These measures remove the effect of changing population size and changing distribution according to age, sex, and race.- Figure 1 shows age-adjusted mor tality rates for all forms of cancer from 1950 to 1982 in the entire pop ulation and according to sex and race, with age adjusted to the 1980 population. Cancer-related mortal ity, measured in this way, rose stead Figure 1. Mortality from All Malignant Neoplasms, 1950 through 19B2. in the U.S. White Population and According to Race (White or Nonwhite) and Sex. ily among white males; fell slightly, Age was adjusted to the U.S. population of 1980. 1228 THE NEW ENGLAND JOURNAL OF MEDICINE May 8, 1986 1960s. These changes in death rates from Tung cancer have substantially affected mortality rates from all cancers combined (Fig. 1). (Later on we will discuss the effect of excluding lung and other cancers from the trends shown in Figure 2.) Data on nationwide mor tality trends with smokers and nonsmokers separated are not available. Mortality from cancer of the prostate (Fig. 2) has not changed appreciably in the entire male popula tion, despite continual increases among nonwhile men since 1950. Mortality from stomach cancer (Fig. 2) has steadily declined in all four race and sex groups. This decline reflects changes in incidence rather than better meth ods of treatment, earlier diagnosis, or changes in defi nition.6*8 Mortality from cervical cancer (not shown) has also declined dramatically as a result of wide spread screening programs, improved standards of living, and a high rate of hysterectomy.11 Mortality from colorectal cancer (Fig. 2) has been declining slowly and steadily for reasons not fully un derstood but probably including better diagnostic pro cedures and improvements in treatment. These data, taken alone, provide no evidence that some 35 years of intense and growing cflbrts to im prove the treatment of cancer have had much overall effect on the most fundamental measure of clinical outcome -- death. Indeed, with respect to cancer as a whole we have slowly lost ground, as shown by the rise in age-adjusted mortality rates in the entire popula tion (Fig. 1). This is not to say that without these efforts at treatment the trends would have been the same, but overall, the effort to control cancer has failed -- so far -- to attain its objectives. This generally dismal picture obscures some strik ing successes, however. For example, age-adjusted mortality from all cancers combined has dropped no tably in patients under the age of 30, though such deaths account for only about 1 to 2 percent of total mortality from cancer.12,13 In older persons, mortality from small-cell lung cancer and from non-seminoma testicular cancer has also decreased (data not shown). Incidence Data The possible measures for the incidence of cancer are similar to those for mortality from the disease -- counts, crude rates, and several kinds of adjusted rates, each of which may be limited to particular de mographic segments or particular forms of cancer.-**5 The incidence statistic that wc chose for a measure of overall progress against cancer is the direct ageadjusted rate for all cancers combined (U.S. 1980 standard), but supplemented by rates for certain nar rower segments that illuminate specific problems. Table 1shows cancer incidence data from the SEER (Surveillance, Epidemiology, and End Results) Pro gram, which was developed under the auspices of the National Cancer Institute. One can compare these statistics with the mortality data in Table 1, but keep ing in mind that most superficial skin cancers arc ex cluded, that the SEER data arc for a nonrandom sam ple of about 10 percent of the U.S. population from 10 diverse geographic areas (4 states, 5 metropolitan areas, and Puerto Rico), that the scries begins only in 1973 for 8 of 10 areas (the others were added in 1974 and 1975), and that the incidence data arc subject to substantial shifts in diagnosis and reporting during that time.14 Data on cancer incidence arc limited to the white population because the number of nomvhitcs in the SEER population was too small to provide reliable estimates of risks and be cause the distribution of nonwbites across specific racial categories was substantially different from that in the United States as a win tie. Cancer incidence rates are shown n Figure 3. Overall trends arc up ward among both white males and white females, suggesting a failure to prevent or control new or current causes of cancer. YE AH Figure 2. Mortality from Cancer of Selected Sites. 1950 through 1982. in the Total U.S. Population. Age was adjusted to the U.S. population ot 1980. Site-Specific Incidence Data The reported incidence rates for breast cancer show a distinct fineyear peak in 1974 and a slower rise in more recent years (Fig. 4). The reported incidence of cancer of the prostate (Fig. 4) lias in creased slightly among while men and more sharply among nonwhite men (data not shown). Incidence rates for lung cancer have been ris- Vol. 314 Ni. )`J l*RO(JRKSS AGAINST CANCKR? -- IIA1LAR AND SMITH 1229 V 8 5 *< 11/ iVf VC Ml '1W ') Figure 3. Incidence of All Cancers, 1973 through 1981, According to Sex, in the White Population of the SEER Registry Area. Age was adjusted to the U.S. population of 1980. ing rapidly in white men and white women, largely in response to changes in tobacco smoking in recent decades. Again, we sec no reason for optimism about overall progress during recent years. There is no reason to think that, on the whole, cancer is becoming any less common. Survival Data There are many divergent measures of case surviv al, just as there are for mortality and incidence. One can count a group of patients with cancer, then count the number who are alive at some specific time after diagnosis (e.g., 2, 5, or 10 years) and calculate the percentage surviving at that time. However, that mixes the lethal effects of cancer with deaths from unrelated causes. One might instead compute the per centage who are alive and appear to be free of cancer at five years, or exclude those who have died of causes other than cancer during the period, or try to calculate the lifetime probability that someone with cancer will eventually die of it. A common device is to avoid diffi cult judgments about the presence of recurrent cancer or the cause of death and, instead, adjust for "expect ed" survival estimated from rates in the general popu lation with the same age and sex distribution. The ratio of observed survival (cancer patients) to expect ed survival (general population), called the relative survival rate, is a commonly reported measure of case survival.15 Any of these survival measures can be applied to cancer overall, to specific forms of cancer, or to specif ic demographic groups of patients. Again we have many measures, with none of them clearly best. The difficulty in interpreting survival rates after cancer is illustrated by recent congressional testimony stating that the United Stales is on the verge of attaining a five-year survival rate of 50 percent. News stories did not always make it clear that the computation of such a high rate required exclusion of the nonwhitc popula tion and the use of relative rather than absolute sur vival rates."* Problems in Interpreting Recent Incidence . and S urvival Data Changing standards of diagnosis and medical care of patients with cancer may affect incidence and sur vival rales substantially more than they affect mortal ity rales. At one time, a cancer was a cancer, and it could be assumed that a truly malignant neoplasm would eventually appear in hospital records (for treat ment) or in death records (if treatment was unsuccess ful or not attempted). The major exception, most forms of superficial skin cancer, could be excluded from the registry system by definition (the biologic behavior of superficial skin cancer is unlike that of other neoplasms because metastatic spread,-the main reason for death from cancer, is uncommon). Other neoplasms lacking metastatic behavior used to be con sidered infrequent and were not regarded as a source of serious bias in the interpretation of trends: That assumption can no longer be made. The implications are substantial. The 1974 peak in the incidence of breast cancer (Fig. 4) corresponded to the occurrence of public dis closures that the wives of the U.S. President and Vice President had breast cancer and a major public effort to promote screening for the disease by mammog raphy. Although the 1974 peak was well beyond the limits of random variation, there has been no appar ent corresponding change in mortality from breast cancer (Fig. 2) or in case survival rates (Table 2). We believe that the 1974 peak in incidence is spurious and reflects the inclusion of a proportion of benign and borderline lesions that in other years would not have been delected and reported. That such shifts in diag nostic criteria do occur, and specifically for breast can cer, is well documented.7-l7,ia After the 1974 peak the HU Figure 4. Incidence ol All Cancers and Cancers of Selected Sites. 1973 through 1981, in the While Population ol ihe SEER Registry Area. Age was adjusted to the U.S. population ol 1980. 1230 THE NEW ENGLAND JOURNAL OF MEDICINE May 8, 1986 rates plateaued at a lower level, then started to rise ing of "cancer" of the breast, prostate, and lung, nei slowly but steadily among both white and nonwhite ther incidence rates nor case survival rates for these women. A recent resurgence of screening programs diseases can be taken as reliable indicators of change may account for some of this latest increase, but pres in the overall progress against cancer. One must won ent data do not permit a definitive conclusion about der whether similar problems affect the data on other whether it is artifactual or represents a true increase in forms of cancer. Mortality data do, in contrast, meas incidence. ure biologic behavior rather directly. That is mainly Cancer of the prostate is a common incidental find why we believe that mortality rales, age-adjusted to a ing when unselected tissue specimens of old men are current standard, arc the best single measure of over examined, whether at autopsy (after death from an all progress. Specifically, we disagree with the decision other cause) or biopsy (at surgery for a benign condi of the National Cancer Institute to emphasize survival tion). Reports of prevalence rates in the 25 percent (and the short-range goal of a five-year overall relative range are not rare.7 It is less widely recognized that case survival rate of 50 percent), because it is subject such lesions, especially those found incidentally at to substantial bias from changing standards of diag prostate surgery, are commonly reported as cancer in nosis and reporting. A reported survival rate of 50 per the incidence statistics. There appear to be no data on cent, if many of the patients do not have the biologic what proportion of these prevalent prostatic "can disease in question, would only mislead and confuse the cers" had shown evidence of malignant behavior. In public, the news media, governmental representatives, cidence rates for this disease do not exceed 1.2 per and health professionals who are not sophisticated in cent per year even in the oldest age groups, including biostatistical and epidemiologic analysis. some proportion of patients with incidental diag Enstrom and Austin22 have also discussed the prob noses; the incidence of clinically apparent prostatic lems of interpreting cancer survival rates. Although cancer must be lower, and mortality rates are lower this matter needs further study, the uncertainties are still. We must conclude that the prevalence rates arc great enough to make case survival an inappropriate seriously inaccurate and that most of 'the tumors measure of progress. found, which do have the microscopical appearance of Colleagues have argued that the overall picture of malignancy, do not have the behavior we associate cancer mortality is dominated by.rising rates of death with the word "cancer." Such an interpretation, com from lung cancer and that this disease should therefore bined with an increasing frequency of incidental be omitted from any summary measure of progress tissue diagnosis, would be consistent with the rapid against cancer. Reasons for such an omission have not changes in survival after prostatic cancer shown in been clearly stated, although it conveniently reverses Table 2. the overall rise in mortality from cancer. Lung cancer Lung cancer has increased rapidly in all major pop is in fact the best illustration of our primary conclu ulation segments. As a result, several kinds of screen sion that despite great effort over many years, research ing programs have been developed and tested. Find on cancer treatment has failed to deal effeciivclv with ings tend to be that in comparison to a randomized control group, the screened group has more cancers Table 2. Absolute and Relative Survival Rates.- detected, the cancers arc in earlier stages, more arc considered suit able for curative treatment, and Fiv e-Y e** S u * v tv*!. 1913 1974 Ye** in D u o m jm s 1975 1976 1977 1978 case survival rates arc substantially prrerni higher. However, overall mortality Absolute raiet is little affected.iy*21 This again All neoplasms 38.5 40.6 41.0 41.2 40.8 40.1 seems to be a result of detecting Colorectal cancer 36.2 37.8 38.2 39.6 39.4 38.7 and reporting lesions that have the microscopical appearance of cancer but not its biologic behavior. As a result of adding these benign condi tions, the pool of real "cancers" is Lung cancer 9.0 9.8 9.7 10.3 10.6 11.0 Breast cancer 64.3 65.2 67.0 66.0 66 0 65 0 Ptostaic cancer 41.2 43.9 45.9 47.7 4 7 .8 47 4 Hodgkin's disease 57.7 63.3 66.6 71.9 69 5 67 3 Non-Hodgkin's disease 34.3 37.9 40.0 4J.0 39 0 5 9 0 diluted, and we find high detection rates, early stage, resectability, and improved case survival, but with little or- no change in-outcome as measured by deaths. 'Thus, the incidence and case sur vival data for three major forms of cancer may not mean what they at first suggest, because of these un certainties about the current mean Relative raiet A ll neoplasms 4 6 .8 49.3 50.0 50.3 50 0 Colorectal cancer 46.4 48.8 49.4 51.1 51.5 Lung cancer lt.0 11.9 ll.K 12.5 15 (1 Breast cancer 72.3 73.6 75.8 74.7 75 l Prostate cancer 60.7 65.0 67.6 70.2 6 9 .9 Hodgkin's disease 61.5 67.6 71.0 76.5 74 3 Non-Hodgkin's disease 40,9 45.1 47.3 48.6 4 6 8 "SiHiiic iiita fti'tn the Skl'Jt Neunmal Institute fm hue rrulc* *nJ hue fertulrs ml), rcLint hi survival of ihc V S hue 49 ) 50 3 13 4 74 ,1 69.4 72 tl 47 l VoL the ihel talii will ove can per ere: of 1 rau rea: met ity: in 1 p. CCS: the I ten ma Cai cen the irei an< hof c sho los: era am car C ag car arc the vet tic no: ant Sr< tut go: cot pn su* evi sei im pn rc< th. un its ca fix Vol.314 No. 19 PROGRESS AGAINST CANCER? -- BAILAR AND SMI TH 1231 the cancer problem. We have never theless calculated agc-adjusled mor tality rates excluding lung cancer; with this exclusion the change in overall agc-adjuslcdmortaliiy from cancer since 1950 shifts from an 8 ' percent increase to a 13 percent de crease. If one also excludes cancer of the stomach and cervix, whose rates have also been changing for reasons largely unrelated to treat ment (Fig. 5), age-adjusted mortal ity shifts from 130.1 in 1950 to 128.9 in 1980 -- a change of less than I percent. It is difficult to claim suc cess in the war against cancer on the basis of these figures. In Figure 5 the lime scale is ex tended to the year 2000. We have marked on the figure the National Figure 5. Mortality from Cancer of All Sites and Selected Sites. 1950 through 1982, in the U.S. Population. Age was adjusted to the U.S. population of 1980. Extension to the year 2000 is shown to reflect the stated goal of the National Cancer Institute. Cancer Institute goal of a 50 per cent reduction in mortality by that year. It is clear that These comments about lack of progress arc in no the goal will not be attained unless the present upward way an argument against the earliest possible diagno trend is reversecTverv soon and there is a precipitous sis and the best possible treatment of cancer. The and unprecedented decline. We do not believe that problem is the lack of any substantial recent improve hopes for such a change arc realistic. ment in treating the most common forms. Conclusions Cairns23 has also discussed the results of the effort to develop cures for cancer. His approach is largely Some measures of efforts to control cancer appear to clinical and biologic; ours is largely epidemiologic ana show substantial progress, some show substantial statistical, yet we come to similar conclusions about losses, and some show little change. By making delib the poor rate of success to date and the need to recon erate choices among these measures, one can convey sider present directions in both research and applica any impression from overwhelming success against tions. His paper should be read in conjunction with cancer to disaster. ours for a more comprehensive view of the matter. O ur choice for the single best measure of progress The main conclusion we draw is that some 35 years against cancer is the mortality rate for all lorms of of intense effort focused largely on improving treat cancer combined, age-adjusted to the U.S. 1980 stand ment must be judged a qualified failure. Results have ard. This measure removes the effects of changes in not been what they were intended and expected to be. the size and age composition of the population, pre We think that there could be much current value in a vents the selective reporting of data to support par comprehensive, consolidated, objective review- of the ticular views, minimizes the effects of changes in diag technical reasons for this failure. What forces led to nostic criteria related to recent advances in screening overlapping waves of interest and program emphasis, and detection, and directly measures the outcome of such as chemotherapy screening, virology, immunolo greatest concern -- death. The National Cancer Insti gy, and perhaps now molecular biology, that have ap tute has also adopted this standard for its prospective peared to hold more promise than they have fulfilled? goal of halving cancer mortality by the year 2000, but Why were hopes so high, what w-ent wrong, and can continues to use relative case survival rates to assess future efforts be built on more realistic expectations? progress in years past.3,16 Why is cancer the only major cause of death for which Age-adjusted mortality rates have shown a slow and age-adjusted mortality rates are still increasing?24 steady increase over several decades, and there is no A lull analysis oi current program plans and direc evidence of a recent downward trend. In this clinical tions would require substantial expertise, time, and sense we are losing the war against cancel.'SubstantiaT support. On the basis of past medical experience with increases in our understanding of the nature and infectious and other nonmalignant diseases, however', properties of cancer have not led to a corresponding we suspect that the most promising areas are in cancec. reduction in incidence or mortality. On the basis of prevention rather than treatment^ Although no one the age-adjusted trends that we have presented, it is can be certain about the benefits of preventive efforts, unlikely that the National Cancer Institute will attain history suggests that savings in both lives and dollars its stated goal of reducing age-adjusted mortality from could be great. For example, opinions that attempts to cancer by 50 percent by the year 2000-- just 14 years prevent smoking have been discouraging arc wrong. from now. In scarcely 20 years of half-hearted effort, this country 1232 THE NEW ENGLAND JOURNAL OF MEDICINE May 8, 1986 has reversed historic trends in smoking and altered its casual tolerance of smokers. Societal antismoking norms have changed, and those who use tobacco are now on the defensive. Research opportunities in other areas of cancer prevention may well merit sharp in creases in support, even if this requires that current treatment-related research must be substantially cur tailed. Certainly, the background of past disappoint ments must be dealt with in an objective, straight forward, and comprehensive manner before we go much further in pursuit of the cure that always seems just out of reach.24 We are indebted to Jesse Berlin and Julia Bailey for help in processing the data presented here, to Dr. Michael Shimkin for encouragement, and to many colleagues for helpful comments on earlier drafts. R eferences 1. McKay FW, Hanson MR, Milter RW. Cancer mortality in the Untied States: 1950-1977. Bcthesda. Md.: National Cancer Institute. 1982, (National Can cer Institute monograph no. 59). (NIH publication no. 82-2435J. 2. Vital statistics in the United States. Vol. 2. Mortality. Pan B. Annual. Rockville, Md.: National Center for Health Statistics, 1950-82. 3. Division of Cancer Prevention and Control. Cancer control: objectives for the nation. 1965-2000. Monograph Scries (in press). 4. Hill AB. Principles of medical statistics. 9th ed. New York: Oxford Univer sity Press. 1971. 5. Armitage P, Statistical methods in medical research. New York: John Wj. ley. 1971. 6. Faust NM. Dal Iman AB. Comparability of mortality statistics for the sixth and seventh revisions. United Slates. 1958. (Vital Statistics special report 51(4)). Rockville. Md.: National Center for Health Statistics. March 1965. 7. Klcbba AJ, Doll man AB. Comparability of mortality statistics for the sev enth and eighth revisions of the International Classiheat ion of Diseases, United States. (Vital and Health Statistics 2(66)). Rockville, Md.: National Center for Health Statistics. October 1975. 8. Klcbba AJ. Scott JH. Estimates of selected comparability ratios based on dual coding of 1976 death certificates by the eighth and ninth revisions of the International Classification of Diseases. (Vita! Statistics repon 28(11). Sup plement). Hyattsville, Md.: National Center for Health Statistics. February I960. 9. Doll'R. Pcio R. The causes of cancer: quantitative estimates of avoidable risks of cancer in the U.S. today. New York: Oxford University Press, 1981. 10. Davis DL, Liltenfcld AD, Giltelsohn A. cl al. Increasing cancer in elderly Americans: fact or artifact? Am J Public Health (in press). 11. Kelsey JL. Hildreth' NC. Breast and gynecologic cancer epidemiology. Boca Raton, Fla.: CRC Press, 1983. 12. Miller RW. McKay FW. Decline in the US childhood cancer mortality: 1950 through I960. JAMA 1984; 251:1567-70. 13. Hammond D. Chard RL Jr, D'Angio GJ. el al. Pediatric malignancies. In: Hoogstraten B. ed. Cancer research: impact of the cooperative groups. New York: Masson, 1980:1-23. 14. Horro JW. Asire AJ. Young JL Jr. el al.. cds. SEER Program: cancer inci dence and mortality in the United Slates. 1973-81. Bcthesda. Md.: Depart ment of Health and Human Services, 1984. (NIH publication no. 85-1837). 15. Edcrcr F, Axtell LM, Culler SJ, The relative survival rate: a statistical methodology. Bcthesda. Md.: National Cancer Institute. 1961. (National Cancer Institute monograph no. 6). 16. DeVita VT. Cancer patient survival statistics. (NCI update). Bethesda, Md.: National Cancer Institute, November 26, 1984. 17. Thomas LB. Ackerman LV. McDivitt RW. Hanson TAS. Hankcy BF. Prorok PC. Report of NCI ad hoc pathology working group to review the gross and microscopic findings of breast cancer cases in the HIP study (Health Insurance Plan of Greater New York). JNC1 1977; 59:495-541. 18. Bcahn OH. Shapiro S, Smart C. Detailed report of the working group's review of BCDDP. Section III. Pathology review of minimal breast cancers delected in BCDDP. JNCI 1979; 62:673-64. 19. Berlin N'l. Buncber CR. Fontana RS, ct al. The National Cancer Institute cooperative early lung cancer detection program: results of the initial screen (prevalence): introduction. Am Rev Rcspir Dis 1984; 130:545-9. 20. Earlv lung cancer detection; summary and conclusions. Am Rev Respir Dis 1984; 130:565-70, 21. Bailar JC IN. Screening for lung cancer -- where are we now? Am Rev Respir Dis 1984; 130:541*2. 22. Enstrom JE. Austin DF. Interpreting cancer survival rates. Science 1977; 195:847-51. 23. Cairns J. The treatment of diseases and the war against cancer. Sri Am 1985; 253(5):5l-9. 24. Bailar JC IN. The case for cancer prevention. JNCI 1979: 62:727-31. Massachusetts Medical .Society Registry on Cominuim; Medical lCdtication To obtain information on continuing medical education courses in the New nt>lawl area, write or call, inti tea tiny field (s) or specialty in which information is desired, lo the Commit tee on Medical Kducation, 1440 Main St., Waltham. MA 0225-1; telephone (G17) 893-4GI0 (Metropolitan Huston) or W ATS 1-800-322-2303 (Massac Iniset is).