REPORTS ON SEMINARS

(1) Seminar on “Biostatistics in Cancer Research”
The seminar was held on November 11-13, 1988, at the Research Institute of Nuclear Medicine and Biology, Hiroshima University, Hiroshima, Japan. The organizers were Dr. Robert W. Miller, National Cancer Institute; Dr. Kunio Aoki, Department of Preventive Medicine, Nagoya University; Dr. David G. Hoel, Division of Environmental Risk Assessment, National Institute of Environmental Health Sciences; and Dr. Takashi Yanagawa, Department of Mathematics, Kyushu University. The proceedings of the conference will be published in Environmental Health Perspectives.
There were seven participants from the U.S. and thirteen participants from Japan. Dr. Miller, Program Area Coordinator, was unable to attend. The objective of the conference was to discuss the effective application of statistical methods to cancer research bringing together American biostatisticians and Japanese mathematicians, statisticians, and epidemiologists. In particular, the conference was planned to demonstrate the importance and fascination of biostatistics in cancer research, aiming to get the interest of young mathematical talents to strengthen the discipline in Japan. To achieve the objective, the conference was opened to any researchers and postgraduate students and a satellite symposium was also organized and held the day before the conference to supplement the conference and to bring together as many mathematical scientists and epidemiologists as possible for the conference. Altogether 20 participants and 68 observers gathered at the conference. The keynote speech prepared in part by Dr. Miller, was delivered by Dr. Hoel. It provided an excellent review of the impact of the first and second Biostatistics Conferences under the US-Japan program--in particular its promotion of interdisciplinary dialogue and the exchange of ideas among scientists in the two countries for the development of new and improved statistical methods in cancer epidemiology.
Session 1: Data Management
Dr. Frank Starmer discussed the concept of a self documenting file that contains annotations or comments that aid visual inspection of the data file. Dr. Starmer showed the access of data from annotated files, and illustrated data analysis with a few examples derived from the UNIX operating environment. This method provides investigators with both a useful representation of the primary data and a repository of comments that describe some of the context of data capture.
When analyzing data from cancer clinical trials, in which the treatment allocation uses dynamic balancing methods such as the minimization method for balancing the distribution of important prognostic factors in each arm, a conservative result occurs if the randomization scheme is ignored and a simple unstratified analysis is carried out. Dr. Yasuo Ohashi assessed the conservativeness by computer simulation, exploited a permutation test of the log-rank statistics for clinical trial data where the allocation is done by the minimization method and/or stratified permuted block design, and developed a computer program to carry out this test. Dr. Ohashi reported the difference between the systems which have been adopted in U.S./Europe and in Japan based on his experience in developing the international guidelines on data management in clinical trials of bladder cancer.
Session 2: Cohort Study
Dr. Suminori Akiba reported on the relationship of cigarette smoking and cancer in a large scale cohort study in Japan. By using survival analysis based on Poisson regression models adjusted for age, prefecture of residence and occupation, Dr. Akiba found with Dr. Takeshi Hirayama a statistically significant dose-response relationship with smoking and mortality rates for cancers of the liver, pancreas, and lung in both sexes, as well as for cancers of the mouth and pharynx esophagus, stomach, larynx, and bladder in men, and for cancer of the uterus in women. The magnitude of relative risks was only slightly affected by adjustment for prefecture of residence or occupation. Using linear relative risk models, Drs. Akiba and Hirayama revealed that the inclusion of a quadratic term for the amount of daily cigarette consumption in addition to the linear term improved the fit of the model significantly for cancers of the esophagus, stomach, and lung in men. Also, Dr. Akiba reported that when the follow-up period was divided into four four-year intervals, the relative mortality risk associated with lung cancer among males increased significantly during these time periods; in particular, the observed change was most marked among farmers and was not explicable by increased cumulative cigarette consumption with advancing age.
Dr. Duncan C. Thomas discussed the issue "Can familial aggregation of disease be explained by familial aggregation of environmental risk factors?" He introduced a proportional hazards model for the effects of measured covariates and of one or more components of "frailty" that are unmeasured but assumed to have common distribution and known covariance structure within each family; derived the posterior expectation of the hazard for each individual given covariate values and the observed and expected disease history of the family up to each instant of time. He furthermore reported the application to a study of breast cancer in twins and in families of bilateral premenopausal breast cancer probands and of lung cancer in relation to family smoking habits.
Dr. Atsuko Shibata reported the preliminary results from a cohort study of the association of family history of cancer with mortality which may be summarized as follows: Among 2,000 patients ( 1,912 males and 288 females) gastrectomized due to benign gastric disease, 274 male patients and 40 female patients had family histories of cancer. Twenty-two patients with a family history were found to have died during 2,750 person-years of observation as compared with 111 patients without a family history during 17,527 person-years, giving a relative risk of 1.26 (not significant).
Session 3: Clinical Studies and Screening
Epidemiologic study of the natural history of disease can provide information on the preclinical duration of disease. The preclinical duration is called the latency period or incubation period. Estimates of the latency period distribution are important for assessing the potential benefit of a screening program and for projecting future numbers of cases of disease. However, a number of important censoring, truncation and length-biased sampling issues arise in estimating the latency distribution from natural history studies. Dr. Ron Brookmyer discussed these problems in the context of two applications. The first application was the natural history of cervical cancer. A matched case-control methodology was proposed for joint estimation of the latency period distribution and the error rates of the pap smear screening test based on the data from a screening program for cervical cancer. Also a partial likelihood was proposed for the analysis of prospectively collected screening data. The second application concerned the natural history of AIDS. Dr. Brooknyer proposed a full likelihood to evaluate jointly the effect of covariates on risk of infection and risk of progression to AIDS once infected.
Dr. Tetsuo Kuroishi discussed the effectiveness of mass screening for uterine cancer in Japan by comparing the changes in the age-adjusted rates of PYLL (potential years of life lost) due to cancer of the uterus from 1969- 1972 to 1973-1977 between the high coverage rate (intensively screened) areas and the comparable control areas. Dr. Kuroishi reported that the percent reduction in the average age-adjusted rate of PYLL due to cancer of the uterus, and the years of life saved per 100,000 females were greater in the high coverage-rate areas than in the control areas.
Dr. Naohito Yamaguchi discussed a mathematical model of growth kinetics which enables one to estimate the functional relationship between the pathological variables of a tumor and the probability of detection of the tumor either through symptoms or by a screening test. The maximum diameter of a tumor was chosen to represent the pathological variables. Applying the model to data on gastric cancer, Dr. Yamaguchi reported that the doubling time of the maximum diameter in gastric cancer was 7.27 months and the probability of detecting the tumor of diameter x by indirect X-ray was p(x) = 0.323x²/[1 + 0.323x²].
Dr. Tomotaka Sobue reported the results from a case-control study for evaluating cervical cancer screening in Osaka, Japan. Dr. Sobue estimated that 78 percent of cervical cancer mortality and 59 percent of invasive cervical cancer incidence among non-screened women could be prevented by screening for cervical cancer. Various problems pertaining to case control studies for evaluating cancer screening were also presented for discussion.
Session 4: Measurement Errors
In epidemiologic studies, misspecification of the main exposure variable and the other explanatory variables is not uncommon. Such misspecification can greatly affect tests of the hypothesis of no association between the exposure of interest and response. Dr. Stephan W. Lagakes reviewed some recent results on the consequence of model misspecification on the validity and power of tests of association between exposure and response. Dr. Lagakos discussed a variety of cases in which misspecification occurred including mismeasurement of an explanatory variable, mismodeling of the functional relationship between a continuous covariate and response, discretization of a continuous covariate, or the complete exclusion of an important covariate from the model.
Dr. Takashi Yanagawa, on the other hand, discussed the mismeasurement of a response variable which resulted from the use of screening tests in estimating the cancer risks. Dr. Yanagawa developed methods for estimating the size of the high risk group adjusted for the characteristics of screening tests and of estimating the incidence rates of cancer among the high-risk group adjusted for the characteristic of the test. He demonstrated with Dr. Shinkan Tokudome the usefulness of the methods by applying them to estimate the risks of adult T-cell leukemia/lymphoma in the Saga prefecture, Japan.
Session 5: Statistical Inference of Cancer Risks
Dr. Michael Vaeth discussed some of the major issues and uncertainties involved in calculation of the excess lifetime risks and related risk projection methods. Calculation of excess lifetime risks usually requires rather complicated life table computations. Dr. Vaeth proposed a simple approximation to the excess lifetime risk and justified the validity of approximation for low dose levels of radiation exposure empirically and theoretically. This approximation provides important guidance in understanding the influence of the various factors involved in the calculations.
In the analysis of epidemiologic data, the parameters of interest are the odds ratio in case control studies and the rate ratio and difference in cohort studies. Through these parameters one can estimate the instantaneous-incidence-rate ratio and difference which are the most meaningful effect measures in cancer epidemiology. When there exist confounding factors, stratified analysis and summary measures for the effect parameters are needed. Dr. Toshiya Sato discussed confidence intervals of these important parameters. Specifically, Dr. Sato discussed those confidence intervals recently developed based on the Mantel-Haenszel estimator.
Dr. Robert Woolson reported on the performance of three combined odds ratios estimators (Mantel-Haenszel, maximum conditional likelihood, weighted least squares) and four tests of association (likelihood ratio, Mantel-Haenszel, Pearson, least squares test) using Monte Carlo techniques. Dr. Woolson concludes that the mean square error of the weighted least squares estimators is usually less than those of the other estimators; although, in many situations, this least square estimator suffers severe bias; that with sparse data only the likelihood ratio and Mantel-Haenszel tests for association maintain their size. In addition, Dr. Woolson reported the performance of several tests for homogeneity of k 2x2 tables; he concluded that the Breslow-Day test is reasonable for use in nonsparse data settings when taking into account both the size and power of the test; in sparse data settings the T4 statistic of Liang and Self performs the best when all tables, regardless of sample sizes, have odds ratios generated from the same distribution.
Session 6: Nonparametric Smoothing
One of the most promising and rapidly developing fields of statistics is the smoothing method based on the empirical Bayes approach. The usefulness of the method is already familiar in econometrics and engineering, but its use in medical science is rare. Dr. Takemi Yanagimoto reviewed recent development of smoothing methods in relation to standard statistical methods. Applying the method to cancer mortality rates of stomach, lung, and pancreas in Japan and also epidemiologic data on SMON, Dr. Yanagimoto demonstrated the superiority and usefulness of the method over the standard methods.
Dr. Megu Ohtaki reported on an innovative idea in estimating the interaction effect of age and period on mortality. The method combines ANOVA technique and nonparametric smoothing for estimating the parameters. Dr. Ohtaki applied the method to mortality data of rectal cancer in Japanese males and females during the period 1950-1986 and found interesting patterns of mortality in the age and period in both males and females.
Session 7: Risk Assessment Models
Dr. David G. Hoel discussed a method of combining Japanese data on radiation exposure and cancer mortality with U.S. data on cancer incidence and lethality to estimate the effects of ionizing radiation on cancer incidence. The method was based on the mathematical relationship between the mortality rate and the incidence and lethality rates, as well as on statistical models that relate Japanese incidence rates to U.S. incidence rates and radiation risk factors. In particular, Dr. Hoel focused on the incidence of breast cancer in Japanese women and how its incidence was affected by radiation risk factors. The analysis used Japanese exposure and mortality data from the Radiation Effect Research Foundation study of atomic bomb survivors and U.S. incidence and lethality data from the Surveillance, Epidemiology, and End Results registry. Even without Japanese incidence data, Dr. Hoel obtained reasonable estimates of the incidence of breast cancer in Japanese women and identified the radiation risk factors that affected this incidence: it was demonstrated that the age at exposure was an important risk factor, but that the incidence of breast cancer was not affected by the city of residence (Nagasaki versus Hiroshima) or the time since exposure.
In his closing address, Dr. Kunio Aoki concluded that the discussions in the conference were extensive and highly useful for information exchange between the two countries, and that the quality of scientific presentations was excellent. In particular, he stressed the need for closer and intensive interaction between biostatistics and epidemiology to further develop research on human cancer. It was hoped by all participants that the Japan-US Biostatistics Conference will be held periodically.

(2) Workshop on “Cancer Registries”
Held in conjunction with the Annual Meeting of
the American Association of Central Cancer Registries
Chicago, Illinois, March 9-10, 1989
The use of data from cancer registries is an important component of cancer epidemiology. Determining the distribution of type-specific cancers makes it possible to target high-risk groups for prevention and early detection. Studies of mortality reveal the effectiveness of therapy. Incidence and prevalence data are helpful in planning the resources needed to treat cancer. The cases registered can provide the basis for case-control studies to identify risk factors. International comparisons reveal demographic differences of interest in the study of cancer causation. Registries are thus a foundation for comparing type-specific cancers under the US-Japan Cooperative Cancer Program. An extraordinary opportunity to bring together the specialists in registries from the two countries came through the annual meeting of the newly formed American Association of Central Cancer Registries (AACCR), held in Chicago on March 9-10, 1989, at the Headquarters for the American College of Surgeons. There were eight speakers from Japan and 16 from the United States.
The AACCR is a professional society founded in 1988 by the American Cancer Society, the Commission on Cancer (American College of Surgeons), the National Cancer Institute, and the National Tumor Registrars Association. AACCR's primary goal is to contribute to a national profile of cancer incidence and to identify regional differences for further study. The aim of the meeting with the Japanese was to initiate the exchange of information on 1) ways to train data-entry personnel, 2) the development and coordination of cancer registries, and 3) methods for analyzing registry data.
Dr. Hankey discussed the Surveillance, Epidemiology and End Results (SEER) Program of NCI, which is population-based and collects its information from 11 registries in the U.S., covering 10.6 percent of the population. Its data include modalities of diagnosis and treatment, of value in studies of end-results; 92 percent of diagnoses are histologically confirmed.
Dr. Fujimoto stated that cancer registries were first established in Japan in 1962. In 1975 a research group for population-based cancer registries was organized. Of Japan's 49 prefectures, 17 have collaborating registries for studies of incidence, the outcome of treatment, the value of area-wide screening programs, cases as compared with controls, and information about the health care-provider (hospitals, private physicians or others).
Dr. Maruyama described how the Japanese Research Society for Gastric Cancer was started in 1963 and expanded to cover 26 cancer sites at present. The data, from 200 cooperating hospitals, concern 38,000 lung cancers, 15,000 esophageal carcinomas, 134,000 gastric cancers, 22,000 breast cancers, and 29,000 each of brain cancers and osteosarcomas. Generally the data show that the outcome was improved by early detection and an increased use of surgery. In 1985 a US-Japan collaborative study on the prognosis of stomach cancer showed that differences in survival rate were related to differences in clinical management and in the behavior of the cancer.
Dr. Horm reported that SEER data show an upturn in 1984 in the incidence rates of 1) in situ and 2) localized breast cancer. The trends differ geographically. Noninvasive cancer incidence increased by about 50 percent among women 50 or more years of age, but there has been no change in their incidence of invasive cancer and mortality.
Through the use of SEER data, Dr. Henson has found that breast cancer was more often localized in Whites than in Blacks, but Blacks are now experiencing a decrease in mortality. Mortality from colon cancer is lower in Whites, and, as expected, survival is strongly associated with the stage of the cancer.
Dr. Chu reported on a randomized trial by the Health Insurance Plan of New York in which 30,000 women were screened for breast cancer and an equal number were not screened. The results after 18 years showed a lower mortality for the first six years among screenees, and the difference was maintained thereafter.
Dr. Hanai had evaluated 78,000 cancer cases registered in 1966-1980. Survival rates were moderately improved for cancers of the stomach, uterus, colon, and breast, but not for cancers of the lung, liver, and pancreas. A shift in the stages of localized cases that occurred over time accounted for improved survival over time. Another study concerned declining trends for stomach cancer by age and by histologic classification: metaplasia as compared with the diffuse type, according to Lauren's criteria. The decline in rates was greater for the diffuse type in males, but greater for the metaplasia type in females. Metaplasia, more common than the diffuse type in Japan but not in the U.S., had similar frequencies in native Japanese and Japanese-Americans, although the decline in mortality was much greater among the Japanese-Americans. Distribution by histologic type appears, therefore, to be racially determined.
Dr. Conley spoke on the limitations of clinical trials with respect to cancer control. It cost about $100 million for the breast cancer screening program for distribution by stage over time. Similar funding would be needed to evaluate a screening program for colon cancer.
Dr. Tsukuma studied screening procedures for colon cancer. In a rural town, 5,919 persons were screened for fecal occult blood and followed for two years. In that time, 12 colon cancers were diagnosed; the sensitivity of the test was 83 percent and the specificity was 79 percent. Another study used double contrast barium enemas, which after follow-up for 5-9 years, had a sensitivity of 87 percent and a specificity of 96.5 percent.
Mammography was the best screening procedure to detect breast cancer according to Dr. Garfinkle. This observation came from a study of 280,000 women screened five times over nine years, during which time 4,240 developed breast cancer. The detection rate was 5.6 per thousand on first screening and 2.4 to 2.7 for the second through fifth screenings. Survival was not changed by screening: 89 percent at five years and 79 percent at ten years.
Are second primary cancers related to radiotherapy for carcinoma of the uterine cervix? Dr. Hiyama studied 97 second primary cancers among 2,207 women given radiotherapy and 3,387 who were not so treated. Radiotherapy was unrelated to the frequency of all second primary cancers, or to cancers of the breast, stomach lung, and leukemia. In a record-linkage case control study, heavy drinkers of alcoholic beverages had more than five times as much liver cancer than the control group, after correcting for cigarette-smoking.
Dr. Lerchen had investigated the completeness of cancer treatment in the four months after diagnosis, and found problems in obtaining good information. Remissions occurred, of course, after treatments that spanned more than four months.
From a case control study of 91 stomach cancer deaths as compared with 273 controls in a small town in Japan, Dr. Sobue found that screening within five years was associated with fewer deaths. A similar study of screening for uterine cervical cancer revealed that 78 percent of non-invasive cancer and 59 percent of invasive cancer were preventable.
Dr. Friedell said that his registry of cervical cancer in eastern Kentucky revealed that the incidence of invasive cancer was twice that reported elsewhere in the U.S. by the SEER Program. The peak for in situ tumors was at 25 years and for invasive cancer was at 35. The need for stronger preventive measures is apparent.
Dr. Henson had used SEER data to study cancer grade as it relates to stage and survival. He found that grade is useful for predicting the clinical course. It is strongly influenced by the histologic type and stage of the cancer. Staging seems to be a better indicator of survival than grading of cancers of the bladder, prostate, breast, and corpus uteri.
Hiroshima and Nagasaki have two types of cancer registries: the conventional type and a Tumor Tissue Registry as described by Dr. Mabuchi The Tissue Registry makes it possible to compare histology of the same cancer in different geographic areas. Adenocarcinoma of the lung is more frequent in Japan than in the U.S. Japanese exposed to the atomic bomb develop an excess of small cell or squamous carcinoma, as do uranium miners in the U.S. Thus, racial differences do not alter the type of lung cancer induced by radiation.
Dr. McWhorter, using SEER data, showed a dissimilar incidence of certain cancers among Blacks as compared with Whites with respect to socioeconomic status (SES), as measured in part by family income and level of education. The cancers were related to SES, being substantial for cancer of the cervix and nil for multiple myeloma.
Dr. Aoki described a large 3-year cohort study begun in 1988 involving 120,000 inhabitants of 20 towns throughout Japan. The purpose is to study life-style in relation to trends in cancer rates. One-third of the sample is to be examined physically and have serum drawn repeatedly to be stored for future study. The study participants will be followed for ten years through information obtained from physicians, health insurance, home visits by nurses, and screening information from local health agencies.
Recurrence rates in 20,000 persons with cancer were studied by Dr. England. Among the cancers studied (breast, colon, lung, prostate, and lymphatic), recurrence rates varied by site.
Reevaluation of a U.S. cooperative lung cancer screening program, reported by Dr. Smart, showed a high frequency of detection in early stages when operability was high, but the 10-year survival rate was no better than in the controls. Evaluation of screening programs should take into account the multiple variables that may affect prognosis. The data were from Memorial Sloan-Kettering Cancer Center, the Mayo Clinic, and other medical centers.
Cancer among the elderly in the future was projected by Dr. Yancik. She reported that at present about 50 percent of cancers occur in persons 65 years of age or older. The number in this age-group will swell from 24.5 million at present to 64.3 million in 2020. The patterns of distribution by site will change. Comparison of cancer occurrence in the elderly of Japan as compared with the U.S. should suggest ideas for cancer control programs, such as early late alterations in life-style.
The discussion of US-Japan Exchange Plans for the Future was led by Dr. Li and Dr. Aoki. The Japanese noted the U.S. strength in population-based registries and sought guidance in this area. They were particularly interested in technical details of data abstraction, coding, verification, and follow-up. They lack trained personnel. They would like to visit several U.S. registries, create courses, develop the infrastructure of tumor registries, and strengthen nationwide cancer registration. With regard to training, the U.S. side suggested Dr. Zippin's course for tumor registrars, and several software packages for entry of tumor registry data.
The U.S. side was interested in the knowledge gained by the Japanese in hospitalbased registries, as a possible future development in the U.S. Now that the U.S. population-based registries are on a sound basis, there is interest in increasing their use to address clinical questions.
Subjects for the future include binational exchanges on population-based survival experience, incidence rates, changes over time, and the relation of age, stage, histologic subtypes, and modes of treatment to end results. Updating migrant studies of the past is another possible fruitful area.
Because of the amount of information exchanged at this meeting, periodic US-Japan meetings on cancer registries should be held in the future.



SEMINAR AGENDA AND PARTICIPANTS

(1) SEMINAR ON BIOSTATISTICS IN CANCER
Hiroshima University, November 11-13, 1988

AGENDA