Showing papers by "Cancer Epidemiology Unit published in 2002"

Validity of self-reported height and weight in 4808 EPIC-Oxford participants.

Abstract: Objective: To assess the validity of self-reported height and weight by comparison with measured height and weight in a sample of middle-aged men and women, and to determine the extent of misclassification of body mass index (BMI) arising from differences between self-reported and measured values. Design: Analysis of self-reported and measured height and weight data from participants in the Oxford cohort of the European Prospective Investigation into Cancer and Nutrition (EPIC-Oxford). Subjects: Four thousand eight hundred and eight British men and women aged 35-76 years. Results: Spearman rank correlations between self-reported and measured height, weight and BMI were high (r > 0.9, P < 0.0001). Height was overestimated by a mean of 1.23 (95% confidence interval (CI) 1.11-1.34) cm in men and 0.60 (0.51-0.70) cm in women; the extent of overestimation was greater in older men and women, shorter men and heavier women. Weight was underestimated by a mean of 1.85 (1.72-1.99) kg in men and 1.40 (1.31-1.49) kg in women; the extent of underestimation was greater in heavier men and women, but did not vary with age or height. Using standard categories of BMI, 22.4% of men and 18.0% of women were classified incorrectly based on self-reported height and weight. After correcting the self-reported values using predictive equations derived from a 10% sample of subjects, misclassification decreased to 15.2% in men and 13.8% in women. Conclusions: Self-reported height and weight data are valid for identifying relationships in epidemiological studies. In analyses where anthropometric factors are the primary variables of interest, measurements in a representative sample of the study population can be used to improve the accuracy of estimates of height, weight and BMI.

von Willebrand factor and coronary heart disease: prospective study and meta-analysis.

Abstract: Aims To determine whether circulating von Willebrand factor concentrations are prospectively related to risk of coronary heart disease in the general population. Methods and Results We measured baseline von Willebrand factor values in the stored serum samples of 625 men with major coronary events and in 1266 controls ‘nested’ in a prospective study of 5661 men aged 40–59 years, recruited from general practices in 18 British towns in 1978–1980 and followed up for 16 years for fatal coronary heart disease and non-fatal myocardial infarction. We conducted a meta-analysis of previous relevant studies to place our results in context. Men in the top third of baseline von Willebrand factor values (tertile cutoff >126 IU . dl 1 ) had an odds ratio for coronary heart disease of 1·83 (95% confidence interval 1·43–2·35; 2P<0·0001) compared with those in the bottom third (tertile cutoff <90 IU . dl 1 ), after adjustments for age and town. The odds ratio was little changed after further adjustment for risk factors (1·82, 95% CI 1·37–2·41), and was not significantly different in an analysis restricted to the 404 cases and 1007 controls without baseline evidence of coronary heart disease (odds ratio 1·53, 95% CI 1·10–2·12). A meta-analysis of all relevant population-based prospective studies (including the present study) yielded a combined odds ratio of 1·5 (95% CI 1·1–2·0). von Willebrand factor values were strongly correlated with Helicobacter pylori seropositivity and circulating concentrations of C-reactive protein (2P<0·0001 for each), but not with smoking, blood lipids, or most other measured risk factors. Conclusion Though circulating von Willebrand factor concentrations may be associated with incident coronary heart disease, further studies are needed to determine the extent to which this is causal.

Consumption of vegetables, fruit and other plant foods in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohorts from 10 European countries.

Abstract: Objective: To describe and compare the consumption of the main groups and subgroups of vegetables and fruits (V&F) in men and women from the centres participating in the European Prospective Investigation into Cancer and Nutrition (EPIC). Design: Cross-sectional analysis. Dietary intake was assessed by means of a 24-hour dietary recall using computerised interview software and standardised procedures. Crude and adjusted means were computed for the main groups and sub-groups of V&F by centre, separately for men and women. Adjusted means by season, day of the week and age were estimated using weights and covariance analysis. Setting: Twenty-seven centres in 10 European countries participating in the EPIC project. Subjects: In total, 35 955 subjects (13 031 men and 22 924 women), aged 35–74 years, randomly selected from each EPIC cohort. Results: The centres from southern countries had the highest consumption of V&F, while the lowest intake was seen in The Netherlands and Scandinavia for both genders. These differences were more evident for fruits, particularly citrus. However, slightly different patterns arose for some sub-groups of vegetables, such as root vegetables and cabbage. Adjustment for body mass index, physical activity, smoking habits and education did not substantially modify the mean intakes of vegetables and fruits. Conclusions: Total vegetable and fruit intake follows a south–north gradient in both genders, whereas for several sub-groups of vegetables a different geographic distribution exists. Differences in mean intake of V&F by centre were not explained by lifestyle factors associated with V&F intake.

Radiation-induced meningioma: a descriptive study of 253 cases.

Abstract: Object. Ionizing radiation is the only established risk factor recognized today in the causation of meningioma. The aim of the present report is to describe the demographic and clinical characteristics of a large series of 253 patients with radiation-induced meningiomas (RIMs). These parameters were compared with those of 41 patients with meningiomas in whom there was no previous history of irradiation (non-RIM group) and with other series of patients presented in the literature. Methods. The cases of RIM were recruited from a cohort of appproximately 11,000 individuals who had been treated with ionizing radiation during childhood for tinea capitis and from a group of individuals who, as adults, applied for compensation because of that treatment. The non-RIM group was identified through the Israeli Cancer Registry. Exposure to radiation was carefully validated among all cases of RIM and absence of previous irradiation was verified for all patients in the non-RIM group. Significantly, a lower patient age a...

Long-term effects of oral contraceptives on ovarian cancer risk.

Abstract: Several epidemiologic studies have reported a protective effect of oral contraceptives (OCs) on ovarian cancer. However, there remain open issues, including better quantification of time-related factors such as time since last use, age at first use and time since first use. We performed a collaborative reanalysis of 6 case-control studies conducted between 1978 and 1999 in the United Kingdom, Greece and Italy, including a total of 2,768 incident, histologically confirmed cases of epithelial ovarian cancer and 6,274 hospital controls under age 70 years. A reduced risk of ovarian cancer was found for ever- compared to never-users [odds ratio (OR) = 0.66, 95% confidence interval (CI) 0.56-0.79], and a stronger reduction was observed for women who had used OCs for > or =5 years (OR = 0.50, 95% CI 0.33-0.76) compared to those who had used them for or =20 years before compared to <10 years; likewise, no significant modification of risk reduction was observed for age at first OC use and time since first OC use. The present analysis indicates that, after taking into account duration of OC use, the OC protection from ovarian cancer persists for a long time after stopping use.

Vasectomy and risk of prostate cancer.

Abstract: ContextVasectomy is a common method of contraception, but concern exists about a reported association with risk of prostate cancer.ObjectiveTo examine whether vasectomy increases risk of prostate cancer.Design, Setting, and ParticipantsNational population-based case-control study of 923 new cases of prostate cancer among men aged 40 to 74 years from the New Zealand Cancer Registry who were on the general electoral roll. Controls (n = 1224) were randomly selected from the general electoral roll, with frequency matching to cases in 5-year age groups. Cases (3-15 months after diagnosis) and controls were interviewed by telephone between January 1997 and November 1999.Main Outcome MeasuresRelative risk (RR) of prostate cancer for men who had had a vasectomy vs those who had not.ResultsThere was no association between prostate cancer and vasectomy (RR, 0.92; 95% confidence interval [CI], 0.75-1.14) nor with time since vasectomy (RR, 0.92; 95% CI, 0.68-1.23 for ≥25 years since vasectomy). Adjustment for social class, geographic region, religious affiliation, and a family history of prostate cancer did not affect these RRs.ConclusionsVasectomy does not increase the risk of prostate cancer, even after 25 years or more.

Cancer mortality in the European Union, 1988-1997: the fall may approach 80,000 deaths a year.

Abstract: After the peak rate of cancer mortality reached in 1988 in the European Union, steady declines were observed: 9.1% for both sexes combined over the period 1988-1997 (from 147.0 to 133.6/100,000, world standard), corresponding to the avoidance of about 80,000 deaths in 1997 (approximately 39,000 below age 65 and 41,000 above). In 1997, the total number of cancer deaths also declined, for the first time. The major determining cancers for these favorable trends were stomach (-30%), lung (-10%), intestines (-15%), breast (-10%), uterus (mainly cervix; -22%), leukemias (-10%) and, after 1995, prostate (-3%).

Mortality in British vegetarians.

Abstract: Objective: To compare the mortality of British vegetarians and non-vegetarians. Design: Analysis of original data from two prospective studies each including a large proportion of vegetarians ‐ the Oxford Vegetarian Study and the Health Food Shoppers Study. Standardised mortality ratios (SMRs) compared with the population of England and Wales were calculated from deaths before age 90 for vegetarians and nonvegetarians in each study. Death rate ratios (DRRs) for vegetarians compared with nonvegetarians within each study were calculated for each of 14 major causes of death. Setting: UK. Subjects: Twenty-one thousand men and women aged 16‐89 years at recruitment, including more than 8000 vegetarians. Results: SMRs for all causes of death were significantly below the reference level of 100 in both studies: 52 (95% confidence interval (CI) 49‐56) based on 1131 deaths in the Oxford Vegetarian Study and 59 (57‐61) based on 2346 deaths in the Health Food Shoppers Study. For all causes of death, the DRR for vegetarians compared with nonvegetarians was close to one in both studies: 1.01 (95% CI 0.89‐1.14) in the Oxford Vegetarian Study, 1.03 (0.95‐1.13) in the Health Food Shoppers Study. Conclusions: British vegetarians have low mortality compared with the general population. Their death rates are similar to those of comparable non-vegetarians, suggesting that much of this benefit may be attributed to non-dietary lifestyle factors such as a low prevalence of smoking and a generally high socio-economic status, or to aspects of the diet other than the avoidance of meat and fish.

Cooking of meat and fish in Europe: Results from the European Prospective Investigation into Cancer and Nutrition (EPIC)

Abstract: Objectives: There is epidemiologic evidence that the consumption of fried, grilled or barbecued meat and fish that are well-done or browned may be associated with an increased cancer risk. These high-temperature cooking methods are thought to be surrogates for mutagens and carcinogens produced in meat and fish, eg heterocyclic amines or polycyclic hydrocarbons. Since data on food cooking methods are scarce, the aim of this study was to describe the variation in meat and fish cooking methods in different parts of Europe. Design: Using a standardized 24 h recall from a sub-sample of the EPIC cohort (35 644 persons, 35-75 y old), mean daily intake of meat and fish prepared by different cooking methods and the relative contribution of the cooking methods to the overall cooking of meat and fish was calculated. Results: Whereas frying was more often noted in northern Europe, roasting and stir frying were more often used in the south. Concerning high-temperature cooking methods, their frequency of application varies between 15% in the EPIC cohort of North-Italy and 49% in the cohort of The Netherlands. Average consumption of fried, grilled and barbecued meat and fish ranges from a low of 12 g/day in the centres in southern Spain to a high of 91 g/day in northern Spain. Conclusion: High variation in both the kind of meat/fish consumed as well as its cooking methods is observed within EPIC. In order to use this variation for the evaluation of the impact of cooking methods on cancer risk, a questionnaire on meat and fish cooking methods is being developed and could be applied in the whole EPIC cohort.

Vitamin K and childhood cancer: analysis of individual patient data from six case-control studies

Abstract: To investigate the hypothesis that neonates who receive intramuscular vitamin K are at an increased risk of developing cancer, particularly leukaemia, a pooled analysis of individual patient data from six case–control studies conducted in Great Britain and Germany has been undertaken. Subjects comprised 2431 case children diagnosed with cancer before 15 years of age and 6338 control children. The retrospective assessment of whether or not an individual baby received vitamin K is not straightforward. In many cases no record was found in stored medical notes and two types of analysis were therefore conducted; in the first it was assumed that where no written record of vitamin K was found it had not been given, and in the second, where no written record of administration was found, information on hospital policy and perinatal morbidity was used to ‘impute’ whether or not vitamin K had been given. In the first analysis, no association was found between neonatal administration of intramuscular. vitamin K and childhood cancer: odds ratios adjusted for mode of delivery, admission to special care baby unit and low birth weight were 1.09 (95% confidence interval 0.92–1.28) for leukaemia and 1.05 (0.92–1.20) for other cancers. In the second analysis, the adjusted odds ratios increased to 1.21 (1.02–1.44) for leukaemia and 1.10 (0.95–1.26) for other cancers. This shift did not occur in all studies, and when data from the hypothesis generating Bristol study were excluded, the adjusted odds ratios for leukaemia became 1.06 (0.89–1.25) in the first analysis and 1.16 (0.97–1.39) when data on prophylaxis imputed from hospital policy and perinatal morbidity were used. We conclude that whilst the broad nature of the diagnostic groups and the poor quality of some of the vitamin K data mean that small effects cannot be entirely ruled out, our analysis provides no convincing evidence that intramuscular vitamin K is associated with childhood leukaemia. British Journal of Cancer (2002) 86, 63–69. DOI: 10.1038/sj/bjc/6600007 www.bjcancer.com © 2002 The Cancer Research Campaign

Nutrient intake and ovarian cancer: an Italian case-control study.

Abstract: Objective: The role of selected macronutrients, cholesterol, and fatty acids in the etiology of epithelial ovarian cancer was analyzed using data from a case–control study carried out in five Italian areas between January 1992 and December 1999. Methods: Cases comprised 1031 women with incident, histologically confirmed epithelial ovarian cancer, admitted to the major teaching and general hospitals of the study areas. Controls comprised 2411 women admitted for acute, non-neoplastic conditions to the same network of hospitals. Information on dietary habits was elicited using a validated food-frequency questionnaire including 78 food groups and recipes. Odds ratios (OR) and their corresponding 95% confidence intervals (CI) were computed by subsequent quintiles of nutrient intake. Results: Direct associations with ovarian cancer emerged for starch intake (OR = 1.4 in the highest vs the lowest quintile of intake; 95% CI 1.1–1.8), while inverse associations emerged for monounsaturated (OR = 0.7; 95% CI 0.5–0.9), and polyunsaturated (OR = 0.7; 95% CI 0.5–0.9) fatty acids. Among fatty acids, oleic (OR = 0.7; 95% CI 0.5–0.9), linoleic (OR = 0.7; 95% CI 0.5–0.9), and linolenic (OR = 0.8; 95% CI 0.6–1.0) acids were inversely related to ovarian cancer. When, however, six macronutrients were included in the same model, only the adverse effect of high starch intake remained significant. Results were consistent in separate strata of menopausal status, parity, and energy intake. Conclusions: Starch was directly associated, and unsaturated fatty acids were inversely associated, with ovarian cancer risk.

A prospective study of serum bile acid concentrations and colorectal cancer risk in post-menopausal women on the island of Guernsey.

Abstract: Secondary bile acids produced by the action of the colonic microflora may increase risk of colorectal cancer. Serum bile acid concentrations reflect the faecal bile acid profile and may be of value as biomarkers of risk of colorectal cancer. In a pilot investigation we examined: (i) the reproducibility of measurements of serum bile acids in two blood samples collected several years apart; and (ii) the hypothesis that relatively high levels of secondary bile acids, particularly deoxycholic acid, would be positively associated with an increased risk of colorectal cancer in a prospective study of 3680 women in Guernsey. There was poor reproducibility between repeat measurements of absolute serum concentrations of bile acids, but there was moderately good reproducibility for the ratios of serum concentrations of deoxycholic/cholic acid, lithocholic/chenodeoxycholic and secondary/primary bile acid concentrations (duplicate blood samples were available for 30 women). There were no significant differences in ratios of serum secondary to primary bile acids or in absolute concentrations of bile acids between the 46 women who developed colorectal cancer and their matched controls, although there was a suggestion that an increased risk was associated with a high ratio of deoxycholic/cholic acid (relative risk in top third compared to lower third=3.92 (95% CI 0.91-17.0, P for trend=0.096). These findings suggest that the ratios of serum bile acid concentrations are sufficiently reproducible for epidemiological studies, but that a larger study than our own is needed to adequately test the hypothesis of their relation to cancer risk.

BRCA2 mutations in a population-based series of patients with ocular melanoma.

Abstract: We studied the BRCA2 gene for germline mutations in 71 of 99 patients (72%) with ocular melanoma who were diagnosed consecutively in Australia in 1997 and 1998. Patients considered for our study fulfilled one of the following critiera: (i) were 50 years of age or less at diagnosis; (ii) had bilateral disease (2 patients); (iii) reported a family history of ocular melanoma (4 patients). Mutation detection was performed using the protein truncation test and denaturing high-performance liquid chromatography with primers designed to include intron-exon boundaries. Six DNA changes were found of which 2 were exonic, in exons 14 (A>C in nucleotide 7244 leading to His>Arg) and 27 (base pair substitution in nucleotide 9976 leading to a stop codon). One exonic change has been reported previously. None of the intronic mutations were deemed to affect splicing efficiency. Neither exonic mutation was in a person with bilateral ocular melanoma or a family history of cutaneous melanoma. We estimated the prevalence of possible loss of function changes in BRCA2 in patients with ocular melanoma at 3% (95% CI 0–10%). This figure was similar to previous estimates of 2.8% and 2% in nonrepresentative samples of patients with ocular melanoma and 2.1% in a representative sample of young women with breast cancer. © 2002 Wiley-Liss, Inc.

Trends in mortality from Hodgkin's disease in western and eastern Europe

Abstract: Hodgkin's disease mortality rates steadily declined by about 75% between the late 1960's and the late 1990's in the current European Union countries and the USA, and Japan. Eastern European countries, however, showed only an approximately 40% decline between the late 1960's and the early 1990's, and no further fall thereafter.

The attribution of lung cancers to asbestos exposure: a pathologic study of 924 unselected cases.

Abstract: We studied a series of 924 nonselected surgical cases of lung carcinoma (without occupational history in clinical records) by histologic examination and light microscopic determination of asbestos body (AB) concentration to determine cancers attributable to asbestos exposure. Lower lobes showed higher concentrations, but no significant associations were recorded between concentrations and histologic type of the lung carcinomas. Histologic asbestosis was demonstrated in 56 cases considered definitely asbestos-related. In 12 of them, the demonstration of asbestosis was attained only after repeated examination of additional sections, suggested by thefinding of more than 1,000 ABs per gram of dry weight (gdw), an indicator of occupational asbestos exposure. In the 56 cases, the media

The limitations of using hospital controls in cancer etiology – one more example for Berkson's bias

Abstract: The aim of this report was to present an example in which Berkson's bias, most probably, affected the results of a study by overriding the influence of a well-established risk factor (smoking) in the etiology of bladder cancer. The results of a study of 140 male patients with bladder cancer and 280 matched hospital controls confirmed the etiological role of industrial occupation in bladder cancer but failed to confirm the role of smoking. We reanalyzed the proportion of chronic related morbidity as well as the rate of smoking in patients with lung disease in cases and controls. A similar distribution of some chronic diseases known to be highly associated with smoking was found among cases and controls. Highest smoking rates (91%) were found among patients with bladder cancer who also reported a concomitant lung disease, and the lowest rate (67%) was noted among controls without lung disease (p = 0.009). Using the prevalence of smoking in the general Israeli male population (50%), significant odds ratio for bladder cancer among ever smokers compared to never smokers was observed. Our conclusion is that a possibility of Berkson's bias should be considered whenever hospital controls are used. Information on diseases related to the risk factor under consideration and on the prevalence of the risk factor in the general population, may demonstrate the existence of such a bias.

Survival after pleural malignant mesothelioma: a population-based study in Italy.

Abstract: Aims and background: The study analyzed survival after malignant mesothelioma in the population-based Registry of Malignant Mesothelioma of Piedmont (NW Italy, 4.5 million total population). It focused on possible differences related to period of diagnosis a proxy of changes in diagnostic or therapeutic procedures. Methods: Cases were actively searched in pathology units and files of hospital admissions and discharges. In 1990-1998, 693 incident cases were diagnosed in residents in the region: 590 of them had a histologic diagnosis of pleural mesothelioma in life and were included in the study. Vital status was ascertained at the municipality of residence as of January 1, 2000. Results: Fifty-eight cases were alive (9.8%) and 20 were lost (3.6%) at the end of the follow-up. Median survival was 0.71 years (95% CI, 0.64-0.78). Cumulative survival was 35.9% at 1 year (95% CI, 32.0‐39.8) and 14.2% at 2 years (95% CI, 11.2‐17.1). Survival was associated to age (longer survival for younger subjects at diagnosis; P <0.0001) and to histology (longer survival for epithelial mesothelioma, shorter for fibrous and intermediate for mixed or unspecified types; P <0.0001). There was no difference in survival for period of diagnosis. The results were confirmed in multivariate analyses. Analyses according to type of hospital (with vs without thoracic surgery) did not show any statistically significant difference. Discussion: The study on survival after malignant mesothelioma is the second largest of the three population-based studies in the world, which showed results similar to ours. Survival measured in published clinical series ranged between 18.4% and 57.6% at 1 year for pleural and 24.1% and 33.8% for peritoneal mesothelioma. The most striking effect of the present study was the absence of improvement in survival with period of diagnosis. Either there was no change in treatment efficacy or the effect was limited to small subgroups and could not be noticed when the analysis included larger categories.

Squamous cell carcinoma of the hypopharynx treated with surgery and radiotherapy.

Abstract: A series of squamous cell carcinomas (SCC) of the hypopharynx treated with combined surgery and radiotherapy is presented to highlight the results of treatment at an early stage of disease. A retrospective mono-institutional analysis was performed on 153 previously untreated patients with SCC of the hypopharynx, seen between 1980 and 1995 at our institution. Univariate and multivariate analyses were performed using the Cox proportional hazard model. The overall five-year specific, and non-specific, disease survival rates were 68 per cent (95 per cent confidence interval, CI: 60-77) and 47 per cent (95 per cent CI: 39-56), respectively. Compared with other series, this study is characterized by treatment at an earlier stage, better prognosis, and a higher number of multiple malignancies. Twenty-two per cent of hypopharyngeal SCCs were diagnosed during the staging procedures for a different head and neck SCC and 14 per cent during the follow-up for a previous tumour. Multivariate survival analysis of clinical and pathological factors confirmed the clinical class of tumour (T) and node (N) and the nodal capsular rupture as prognosticators of disease.

A possible association between ionizing radiation and pituitary adenoma: a descriptive study.

Abstract: BACKGROUND Despite the recognition of ionizing radiation as a causal risk factor for a variety of solid tumors (including brain tumors), to date, such an association with pituitary adenoma (PA) has not been demonstrated. METHODS To evaluate a possible association between past exposure to radiation and the occurrence of PA, the authors reviewed about 4900 medical records of patients who had been irradiated in childhood for tinea capitis. An additional search for patients was performed using the Israel Cancer Registry. The average radiation dose to the pituitary gland was estimated as 0.56 grays, and, for all patients, a meticulous validation of the irradiation was performed. RESULTS A group of 16 patients who developed symptomatic PA after childhood exposure to radiotherapy were identified. Overall, the clinical and demographic characteristics of these patients were similar to other series reported in the literature. There was an apparently high rate of second primary tumors (25%), all of them in the irradiated area, diagnosed among this group. The methodologic issues that limit the demonstration of a possible association between radiation and PA and the epidemiologic and experimental findings in the literature are discussed. CONCLUSIONS In view of the ample amount of evidence identifying low-dose ionizing radiation as a risk factor for a number of intracranial tumors as well as for tumors arising in endocrine organs, a radiation immunity of the pituitary gland is difficult to accept. Hence, the authors suggest that this series should be considered as preliminary observation that supports the role of ionizing radiation in the development of this tumor. Cancer 2002;95:397–403. © 2002 American Cancer Society. DOI 10.1002/cncr.10667

Comments on paper by C. E. Frangakis, D. B. Rubin and X.‐H. Zhao

Abstract: This wide-ranging paper addresses two complications that arise quite often in applications, randomization on a grouped rather than an individual basis and non-compliance with the study protocol. Rather than comment directly on the theoretical points involved we have chosen to look at the relevance of the approach for the illustrative example, using the data kindly supplied to us by the authors. We have for simplicity ignored patient age throughout, although only minor modifications are needed to account for this clearly relevant variable. We have adopted a semi-exploratory approach partly out of caution concerning apparently general models developed without regard to the specifics of the individual application in question, despite the undoubted importance of generality and flexibility. Some care is first needed over terminology. Randomization is applied to physicians, and so the initial unit of study and comparison must be physician-based not patient-based (Cornfield, 1978). Thus we found puzzling the terminology (Section 2.1) of patients complying or not complying with the physician’s assignment. Surely, at least in this example, the patient’s action is a response and any non-compliance arises from the physicians’ not implementing the proposed discussion. In the other examples that were mentioned, such as encouragement to receive a vaccination or vitamin supplement, it would be possible for both the physician to comply or not comply with the encouragement and subsequently for the patient to comply (or not) with the physician’s advice. Therefore, in cluster encouragement designs it is necessary for compliance to be defined precisely according to the particular situation and also according to the particular effect that is being estimated. Because of the randomization the analysis should respect that the unit of study is the physician. The analysis sketched below achieves this. Physicians are randomized to receive or not to receive encouragement (reminders) to discuss the advance directive forms. Out of the 25 physicians who were randomized to receive reminders only two of the physicians with an appreciable number of patients discuss the forms with nearly all of them and so the first variable for analysis is the proportion of patients with whom discussion takes place, regarded as a simple measure of physician-compliance. That is, in this context, compliance can be measured on a numerical scale. The proportion might depend on the number of patients with that physician, either directly or because the number is a surrogate for some physician or patient features. There is, however, no evidence of such dependence.