I ing therapy, the most commonly used index

WHEN MAY ENDOMETRIAL CANCER BE CONSIDERED CURED? RICHARD R. MONSON, MD,* BRIAN MACMAHON, &ID,* AND JAMES H. AUSTIN, MD+ To assess when a woman may be considered cured following treatment for endometrial cancer, two methods have been used. First, using the relative survival ratio, the mortality of 761 women treated for endometrial cancer was compared to that of the general population; in the fourth and fifth years after treatment, the annual relative survival ratio was 98%. After 7 years, the ratio was 10096, i.e., subsequent mortality did not differ from that of the general population of the same age. Second, the causes of death of women who died 5 or more years after treatment were reviewed. Among 478 women who survived to the seventh anniversary and were not known to have clinical recurrence at that time, there were at most 15 deaths that could have been attributable to the initial cancer. In none of these late deaths was there autopsy evidence that uterine cancer was, in fact, a contributing cause of death. N STUDIES OF THE COURSE OF CANCER FOLLOW- I ing therapy, the most commonly used index of outcome is the proportion of patients who survive to the fifth anniversary-the 5-year survival rate. It is clear, however, that for certain forms of cancer the attributable mortality is not all encompassed within this 5-year period. The use of this end-point appears to be based on convenience and convention, rather than on any demonstration that it represents a particularly significant point in the natural history of the disease. We examine here a series of patients with endometrial cancer seen at the Boston Hospital for Women and attempt to answer two questions: Is there any point following initial diagnosis at which, if a patient survives to that point, she may be considered cured? If so, when is that point reached? METHOD In evaluating whether patients with a particular disease may be considered cured, either of two questions may be asked: 1. whether the mortality rate of a group of treated individuals is distinguishable from that of the general Supported by Grant 5 PO1 CA 06373, from the National Cancer Institute, U.S. Public Health Service. * Department of Epidemiology, Harvard School of Public Health, Boston, Mass. t Boston Hospital for Women, Parkway Division, and Department of Obstetrics and Gynecology, Harvard Medical School, Boston, Mass. Received for publication March 6, 1972. 419 population; or 2. whether the deaths that do occur among treated individuals appear to result from the disease for which they were treated. The first procedure depends on the evaluation of group mortality rates, the second on consideration of causes of death of individual patients. The two methods are inde- pendent of each other and should give independent answers to the questions posed. We have used both procedures. The study population consisted of 761 histologically confirmed cases of endometrial cancer in Caucasian residents of Massachusetts who were first treated at the Boston Hospital for Women, Parkway Division, between 1920 and 1959. The series has been described in more detail elsewhere.1 Of the 761 women, 14 were lost to follow-up. Cause of death was determined from the death certificate; in only 11 yo (53J463) was the cause confirmed by autopsy. However, if a death was reported to be due to cancer, we sought confirmation from biopsy or surgical specimens obtained prior to death. As previously reported,3 histologic confirmation was obtained in 86y0 (66/77) of the second primary cancers, excluding skin, which occurred after the corpus cancer. Of the 11 women whose second primary cancer was not confirmed, seven died at least 5 years after diagnosis of their corpus cancer. The relative survival ratio is the basic measure of survival used. It is the observed percentage of survivors expected if the group had experienced the same mortality rates as oc-

420 CANCER A?~g71.\1 1972 Vol. 30 curred in the general population. The expected survival percentages were computed using mortality rates for Massachusetts women, specific for 5-year age and calendar year groups. The most frequently used relative survival ratio is cumulative-that is, the probability of surviving from the time of diagnosis to a specified anniversary. Here, however, we are more concerned with the intewul ratio, i.e., the probability of surviving from one anniversary (e.g., the fourth) to the next (the fifth). Interval survival ratios, as well as cumulative ratios, have been computed for each year up to the 20th anniversary of diagnosis. In the tables, ratios whose confidence limits do not include 100~o-ratios indicating mortality rates in the study signilficantly different from those of the general populationare denoted by asterisks. RESULTS Surviuul ratios: Fig. 1 presents interval relative survival ratios for each year following diagnosis. The shaded area represents the 95% confidence limits of the estimates. The survival ratio was 89.9% in the first year. The survival ratio increased until the fifth year when the observed survival was 98.8% of that expected. After about the seventh anniversary of diagnosis, the interval survival ratios did not differ significantly from loo%, indicating that the subsequent survival did not differ from that of the general population. These findings suggest that the first 7 years may encompass more completely than the conventional 5 years all the excess mortality of patients with endometrial cancer. However, most of the excess mortality has occurred by the fifth year. Furthermore, in these data, the trend between the fifth and seventh year is irregular. While it definitely can be stated that there is no demonstrable excess after the seventh anniversary, the possibility that the point of "no excess" was reached somewhere between the fifth and the seventh anniversaries cannot be excluded. Survival ratios will be considered here taking cognizance of stage and grade of tumor. For reasons given earlier,l the term stage is used to refer to extent of tumor spread as evaluated at the end of the first regimen of treatment, rather than to the usual clinical stage evaluated prior to treatment. Stage I-which is the only stage with sufficient cases for separate evaluation-consists of those cases with disease confined to the corpus uteri. Tumor grade is examined because of its strong predictive value among the large proportion of patients with Stage I disease.1 Women with Stage I cancer had a relative survival of 96.5% or higher in each of the first 5 years after treatment (Table 1). Since this is a large group of patients, the standard errors FIG. 1. Annual relative survival ratios with 95y0 confidence intervals. 0 S 10 IS 20 YEARS OF SURVIVAL

~ No. 2 TABLE 1. ENDOMETRIAL CANCER * Monson et al. 421 Relative Survival Ratios According to Stage of Tumor* Stage+ Survival period after diagnosis ( yrs. ) I 11-IV Not operated on 0-1 97.2% 76.3% 65.2% 1-2 96.5' 83. 78.3: 2-3 98.1: 85.7% 83.3% 3-4 98.7 92.4t 84.6: 4-5 99.1 99.7 95.4 _- ~ ~ 0-5 90.0% 48.9: 34.3: 5-10 10-15 15-20 5-20 All patients 89.9% 93.0: 95.4: 97.2' 98.8 76.5: 100.7 92.5 86.0 99.1 99.8 100.2 (76.7) 99.0 97.1 89.1-94.3 -- ~ 97.6 82.6 32. 1% 92.5 Total number of patients 535 143 83 761 * Ratios in parentheses are based on 20 or fewer patients. Ratios based on 10 or fewer patients are not given. + Stage as assessed postoperatively in patients who underwent surgery. * Ratios that differ from 100% at the 5% level of statistical significance. of the survival ratios are small and even these high survival ratios differed significantly from 100% until the fourth year. For all women with Stage I cancer, the relative survival ratio after the fifth anniversary was very close to 1007~. Relative survival in women with Stage 11-IV carcinoma, and in those who did not come to surgery, was considerably less than IOOyO during the first 4 years following treatment. Even for these groups, however, the relative survival ratio was close to looyo by the end of the fifth year. However, the first 5 years should be regarded as a minimum estimate of the duration of excess mortality, since the number of women with advanced disease was small and the standard errors of the relative survival ratios large. For example, the fact that the survival ratio of 82.6'y0 for the interval 5 to 20 years after diagnosis did not differ significantly from 100% may be attributable to the small number of patients in the group rather than to the absence of an excess mortality among them. Among the 83 women who did not come to surgery, only 22 survived for 5 years. This group continued to experience an excess mortality after the fifth anniversary which, in spite of the small size of the group, was statistically significant. This group contained many patients who did not undergo surgery because of medical problems unrelated to the tumor itself, and their continued excess mortality may be related to their underlying medical condition rather than to persistence of their endometrial cancer. As shown in Table 2, women with Stage I carcinoma, whose tumor was graded 1 or 2, experienced essentially no excess mortality even in the earliest years after diagnosis. The 5-year relative survival ratio for these two groups was somewhat less than lo@%, but not significantly so. Thereafter, with increasing malignancy of tumor grade, the duration of excess mortality became progressively longer and the 5-year relative survival ratios progressively smaller. Women with Stage I, Grade 4 disease had a 5-year survival similar to that of women with Stage 11-IV, Grade 1-2 disease. While, for women with spread outside the corpus and the most malignant tumor grades, mortality was not significantly in excess during the fourth year, the 3-4 and 4-5 year relative survival ratios were only about 90%. It seems likely that with larger numbers in this category, significant excess mortality would have been observed until at least the fifth anniversary. Table 3 shows survival in 5-year intervals after diagnosis according to age, year, and pay status. These are the three demographic variables which were found to be associated with significant survival differences at 5 years. Of interest is the fact that after the fifth year the survival differences observed prior to that time were no longer evident. After the fifth anniversary, the only 5-year interval survival to differ significantly from 100% was the

422 CANCER August 1972 Vol. 30 TABLE 2. Relative Survival Ratios According to Stage of Tumor and Nuclear Grade* Survival period Stage I, by grade Stage 11-IV, by gradet after diagnosis (yrs.) 1 2 3 4 1-2 3-4 0-1 1-2 2-3 3-4 99.7 98.1 98.9 99.0 98.3 99.5 98.7 98.6 95.8 92.8t 95.8 97.4 85.4: 78.9% 98.3 102.9 83.8: 84.3' 95.4 90.2: 4-5 100.7 100.5 95.1 93.8-100.7 ~_ 0-5 5-10 10-15 15-20 96.5 -_ 97.3 106.9 107.8 _ 95.6 101.3 97.6 91.4 102.9 99.3 101.4 63.9: _ (101.9) (107.1) (93.2) -- 61.2: -_ 94.3 95.5 (83.4) - 5-20 112.2 _- 90.4 103.6 101.7 _ 75.1 Total number of pateints 126 239 127 36 102 * See footnotes to Table 1. Eight patients with unknown nuclear grade are excluded. t Includes 83 women who were not operated on. 62.5' 79.5t 72.4' 90.2 91.3 29.6t -- 84.1 (93.0) (76.7) 60.0 -_ 123 15-20 year survival for ward patients (p = 0.05). Further, only two 5-20 year interval survivals differed from 100%-for ward patients and for women treated during the 1930's. These two interval survival ratios are correlated, for 85% of women treated in this decade were ward patients, as compared to 50% in the other three decades. CAUSES OF INDIVIDUAL DEATHS The results described above indicate that in each year prior to the fifth anniversary after diagnosis there was an excess mortality among patients with endometrial cancer. This must be attributed to the disease, or to associated factors such as its treatment, regardless of the pathologic cause to which the individual deaths were ascribed. In considering the causes of individual deaths to evaluate the possibility of cure of this disease, attention, therefore, is focused on deaths after the fifth anniversary. Among the 521 women who survived 5 years, 158.5 deaths were expected and 169 were observed in the subsequent 15 years. While these values do not differ significantly, it is possible that some of the observed deaths might have been due to endometrial cancer. There were 35 deaths after the fifth anniversary in which suspicion of death from uterine cancer could be raised. These include: 1. TABLE 3. Five-year Interval Survival Ratios According to Selected Demographic Characteristics* Interval survival (yrs.) 5-20 year deaths Total Characteristic number 0-5 5-10 10-15 15-20 5-20 Observed Expected Age at treatment 20-54 271 83.97 98.3 96.1 103.4 97.7 39 33.7 55-64 301 80.47 95.3 101 6 93.0 90.0 81 70.5 65 + 189 56.37 112.4 96.2 - - 49 54.3 Year of treatment 1920-29 92 64.97 101.9 87.0 104.4 92.5 25 23.6 1930-39 148 68.7t 91.1 97.4 90.6 80.4' 45 33.6 1940-49 256 79.27 100.6 103.4 92.6 96.3 68 70.4 1950-59 265 82.17 100.4 98.9 - - 31 30.9 Pay status+ Ward 434 72.87 99.0 95.7 87.2: 82.6: 108 89.2 Private 325 81.9 99.2 103.5 105.3 108.1 61 68.5 TOTAL 761 76.5+ 99.1 99.0 99.0 94.8 169 158.5 * See footnotes to Table 1. t Excludes two patients with unknown pay status.

No. 2 seventeen deaths reported on the death certificate to be due to uterine cancer, 2. eight deaths reported to be due to other causes but with uterine cancer stated to be present either in autopsy records or on death certi ficates, 3. seven deaths due to an unconfirmed primary cancer of another site, and 4. three deaths attributed to metastatic cancer of unstated primary site. Fifteen of these 35 deaths occurred prior to the seventh anniversary. In all but one of these, there was evidence that the uterine cancer contributed to-or at least was present at the time of-the death. Eleven of these 15 patients had clinical evidence of recurrence prior to the fifth anniversary; two had clinical recurrences in the fifth or sixth years and one had no clinical evidence of recurrence, but pelvic recurrence was found at autopsy. The remaining patient died of an unconfirmed primary cancer of the pancreas that was diagnosed, but not biopsied, at a palliative operation 6 years after treatment for uterine cancer. The role of the original cancer in the 20 suspect deaths that occurred after the seventh anniversary is less certain. The details of these deaths are presented in Table 4. Six of these patients (# 1-6) were judged to have died of second primary cancers, but there was no histologic confirmation. Among these 20 deaths, only two (# 7, 8) were examined at autopsy. Metastatic uterine cancer was discovered in each, but there was no indication that it played a role in the death of the patient. Four patients (# 9-12) had no known clinical recurrence prior to death, so it is possible that their death was due to another cancer. Patient 9, in fact, had breast cancer 10 years before her uterine cancer, and the cause of death on the death certificate was listed as breast cancer. Of the remaining eight patients, three (# 13, 15, 16) had clinical recurrence 6 or less years after the initial treatment. The remaining five had evidence of metastatic cancer 7 or more years following their initial uterine cancer, but there was no histologic confirmation of the type of cancer in any of these. In short, it is probable that a few deaths after the seventh anniversary were due to the uterine cancer, but the number is very small. Among 478 women who survived to the seventh anniversary and were not known to have clinical recurrence at that time, probably five, but not more than 15, deaths were due to the uterine cancer. It should be noted that not all ENDOMETRIAI, CANCER - Monson et al. 423 the study patients had a clinical evaluation around the seventh anniversary, so that even these numbers probably represent high estimates of the number of late deaths among patients clinically tumor-free at the seventh anniversary. DISCUSSION The methodologic problems attending the two ways of evaluating cure in malignant disease are well illustrated in this material. In the statistical comparison of observed and expected numbers of deaths, a major problem is that as these values approach each other it becomes difficult-because of chance varia- tion-to detect the occurrence of a small number of deaths due to the malignancy against a background of a relatively large number of deaths due to unrelated causes. In the alterna- tive method-the consideration of causes of individual deaths-it is usually necessary to assemble retrospective data to have sufficient cases for analysis. Such data rarely include the detail and closeness of observation that is desirable, For example, definitive data on late recurrence of endometrial carcinoma may come from populations with much higher autopsy rates than those prevailing among the population from which these cases derived. Nevertheless, the fact that the findings from these two approaches tend to converge enables some tentative conclusions: 1. If a patient with endometrial cancer survives to the seventh anniversary without clinical evidence of recurrence, she is unlikely to die of uterine cancer or its sequelae. 2. The conventional 5-year survival rate is a reasonable measure of the likelihood of cure of endometrial cancer, since deaths prior to the fifth anniversary encompass approximately 92% of the total deaths likely to occur as a result of the cancer. 3. There does appear to be a small number of women who die of, or with, endometrial cancer at intervals up to 20, and possibly more, years after initial diagnosis and treatment. This number is so small that its primary importance is in terms of understanding the biology of this tumor, rather than of any implication to clinical follow-up or therapy. These findings appear superficially to be at variance with those reported by Bailar2 from an analysis of 2,394 cases of endometrial cancer reported to the Connecticut Cancer Registry, between 1935 and 1951. Closer examina-

Age at Year of Years of Category Case no. Stage Grade treatment treatment survival Cause of death* Remarks TABLE 4. Details of Late Deaths Possibly Due to Carcinoma of Uterus A. Probable second 1 I 2 69 1933 11 Carcinoma of pancreas No autopsy or clinical information primary cancer deaths I 3 63 1932 12 Carcinoma of stomach Operative diagnosis; no biopsy 3 I 2 74 1946 13 Carcinoma of rectum Operative diagnosis; no biopsy 4 I 1 60 1927 15 Carcinoma of stomach Liver metastases; clinical diagnosis 5 I 2 55 1938 18 Carcinoma of lung Second primary by clinical diagnosis 6 I 2 57 1941 20 Carcinoma of pancreas Operative diagnosis; no biopsy B. Non-cancer deaths; 7 I1 3 66 1959 7 Cirrhosis of the liver Autopsy done; uterine cancer present in uterine cancer present 8 I 2 56 1933 15 Myocardial infarction pelvis Autopsy done; uterine cancer present in lungs F C. Possible uterine cancer deaths-no information 9 I 3 66 1938 7 Carcinomatosis due to cancer of breast Primary breast cancer in 1928 z n m P on clinical recurrence 11 I I 2 1 68 46 1925 1923 12 26 Carcinoma of uterus Carcinomatosis due to cancer No clinical information after 1929 No clinical information after 1939 b e 0s E of uterus c) 12 I 3 53 1925 33 Bronchopneumonia; carcino- No clinical information after 1947 W matosis 4 rc D. Probable uterine can- 13 I 2 62 1948 7 Uremia due to carcinoma of cer deaths-clinical uterus Pelvic nodules-6 yearst recurrence 14 I 2 49 1935 9 Carcinomatosis Concurrent cancer of cervix; lung nodules-8 years 15 I 2 60 1922 9 Carcinoma of uterus Radiation therapy only; uterine enlargement-2 years 16 I 2 65 1929 10 Carcinoma of uterus Vaginal metastasis-1 year 17 I 2 58 1950 11 Carcinoma of urethra Urethral obstruction-9 years 18 I 1 66 1940 13 Carcinoma of uterus Pelvic nodules-11 years 19 I 2 62 1949 15 Superior mesenteriu artery thrombosis Pelvic nodules and enlarged liver-15 years 20 I1 3 64 1947 19 Carcinoma of uterus Pelvic nodules-14 years * Cause of death as listed on death certificate. + Interval between initial treatment of uterine cancer and recurrence. I

No. 2 tion suggests, however, that the two sets of data may not be incompatible. One apparent difference is that the relative survival ratios reported from Connecticut are generally lower than those in this study-both before and after the fifth anniversary. Of particular relevance is that, in Connecticut, annual relative survival ratios between the fifth and twentieth anniversaries remained constant at about 97yo, i.e., they did not reach 100~o as they did in this series. The explanation for this discrepancy may lie in the appropriateness of the general population rates used to compute expected survival values in the two series. The Connecticut data included all cases of endometrial cancer diagnosed in the state. Thus, the use of state mortality data might be expected to provide a more accurate estimate of expected deaths than in the Massachusetts study, since the patients of the particular Massachusetts hospital may be of somewhat higher average socioeconomic status than the population at large. Since mortality rates are inversely related to socioeconomic status, the expected mortality might have been overestimated, and the relative survival ratios underestimated, in the Massachusetts study. Note that, if this were true, the survival of the Massachusetts patients after the \fifth anniversary would be even more favorable than suggested by the data presented; in other words, the experience would be even more different from the Connecticut data. On the other hand, endometrial cancer is a condition which has a somewhat higher incidence rate in women from the higher economic strata and, therefore, it may be that state mortality rates tend to overestimate the expected deaths among affected individuals. In this case, both sets of relative survival rates may be underestimated. As Bailar points out, accurate assessment of relative survival ratios requires that such determinants of overall mortality as income, marital status, and place of residence be taken into account in the estimation of the deaths expected. This is particularly true when one is concerned with the level of accuracy required to differentiate a ratio of 97y0 from one of 100%. In any event, ENDOMETRIAL CANCER * Monson et al. 425 the difference between the two series in this respect is small and could well result from this kind of problem. A more substantial difference exists with respect to the proportion of women who died after the fifth anniversary with cancer present at the time of death. In Connecticut, 38.4% (104/271) decedents after the fifth anniversary were reported to have cancer present at the time of death.2 In Massachusetts, a maximum of 15.3y0 (35/228) of deaths after the fifth year had uterine cancer present. Bailar noted that 46y0 of uterine malignancies present at death had not been microscopically confirmed, but this problem would be expected to be present in both states and is an unlikely explanation of the difference. Part of the explanation, however, may lie in the shorter follow-up period in the Connecticut study, and the consequence that a higher proportion of the women known to have died after the fifth anniversary died shortly after the fifth anniversary. Thus, in Connecticut, 66y0 (179/271) of the deaths after 5 years were in the interval 5-10 years after diagnosis; the corresponding figure in our study was 29% (65/228). The cases with cancer present at death tend, of course, to cluster in this interval. Among the 178 Connecticut deaths in this interval, 85 were reported to have cancer present at the time of death, whereas among the 93 women who died after the 10th year, cancer was said to be present in only 19. Among the 163 deaths in the Massachusetts series after the 10th year, there were nine in the group considered suspicious of the presence of cancer. Although there is still a substantial difference between the two series in suspected prevalence of cancer at death, even after the 10th year-20'~o and 6%, respectively-the numbers are small and differences of this size might well be accounted for by differences in the method of recording or reviewing the evidence as to the presence of cancer. The Connecticut data do not permit examination of the question of whether the deaths with cancer in the interval 5-10 years after diagnosis are clusterd in the first 2 years of this interval, as they are in our series. REFERENCES 1. Austin, J. H., and MacMahon, B.: Indicators of prognosis in carcinoma of the corpus uteri. Surg. Cynecol. Obstet. 128:1247-1252, 1969. 2. Bailar, J. C. 111: Uterine cancer in Connecticut: late deaths among 5-year survivors. 1. Natl. Cancer Inst. 27:239-256, 1961. 3. MacMahon, B., and Austin, J. H.: Association of carcinomas of the breast and corpus uteri. Cancer 23275-280, 1969.