Risk factors for prostate cancer: a case-control study in Greece
A case-control study of prostate cancer (PC) was undertaken in Athens, Greece. Cases were 320 patients with histologically confirmed incident disease, whereas controls were 246 patients without history or symptomatology of benign prostatic hyperplasia, treated in the same hospitals as the cases for minor diseases or conditions. Cases and controls had similar distributions with respect to height, body mass index, sibship size and birth order in the parental family, marital status and number of offspring in the subject's own family and a long series of previous surgical operations and medical diagnoses, including diabetes mellitus, hepatitis and sexually transmitted diseases. There was also no evidence for a positive association between vertex baldness, tobacco smoking and drinking of coffee or alcoholic beverages, on the one hand, and PC, on the other. There was evidence, however, that some aspect of urban life may increase the risk for PC and a suggestion that sexual activity in early adulthood may be inversely associated with this risk. Int. J. Cancer80:699–703, 1999. © 1999 Wiley-Liss, Inc.
Prostate cancer (PC) is one of the most common types of cancer among men worldwide. Together with colorectal cancer, it ranks as the second most common cause of cancer death, after lung cancer, among male citizens of economically developed countries (American Cancer Society, 1998). Clinical PC incidence and mortality vary more than 50-fold internationally, whereas the prevalence of latent cancer at autopsy is fairly similar (Nomura and Kolonel, 1991). Long-term incidence of and mortality from clinical PC have been slowly increasing in most populations, but during the last decade these patterns have been disturbed by the introduction of new diagnostic modalities and treatment policies (Nomura and Kolonel, 1991; American Cancer Society, 1998). Increased blood levels of testosterone (Gann et al.,1996) and insulin-like growth factor 1 (Mantzoroset al.,1997; Chan et al.,1998) have been implicated in the pathogenesis of clinical PC, and the molecular basis of the genetic susceptibility to the disease (Giovannucci et al.,1997b) is beginning to be revealed. The exogenous etiology of PC, however, remains elusive. The incidence of the disease is almost twice as high in African-American men in comparison to white Americans (Nomura and Kolonel, 1991), and there is evidence for early-life influences on future disease risk (Nomura and Kolonel, 1991;Andersson et al.,1995; Giovannucci et al.,1997a), while the role of vasectomy remains undetermined. With respect to diet, promising results have been obtained from 3 distinct areas that are by no means mutually exclusive: one implicating dietary fat or certain categories of it (Giovannucci et al.,1993), another focusing on dairy products with emphasis on a detrimental role for calcium intake (Chan et al.,1998) and a third postulating a beneficial role for dietary lycopene (Giovannucci et al.,1995).
We have undertaken a case-control study in Athens, Greece, to explore general and nutritional aspects of the aetiology of PC. There is no established cancer registry in Greece, and data concerning the epidemiology of PC have not previously been reported from this country. Yet such data could be important. Mortality from PC in Greece is about half of that in economically developed Western countries of Europe and North America and twice as high as that in economically developed countries of Southeast Asia (American Cancer Society, 1998). Moreover, the diet and lifestyle of Greeks and other Mediterraneans are distinctly different from those of North Americans, northern and western Europeans and Southeast Asians, i.e., populations in which the epidemiology of PC has been intensively studied. In this report, demographic, somatometric, lifestyle and medical history aspects of the epidemiology of PC in Greece are presented, while nutritional factors are considered in a companion report (Tzonou et al.,1999).
PATIENTS AND METHODS
From February 1994 to January 1997, 372 patients with histologically confirmed, newly diagnosed PC (incident cases) and residents of the Greater Athens area were contacted in any of 6 large hospitals of this area. It was not possible to interview 52 patients, so 320 were finally included in the study. Of these, 52 were histologically identified following an operation for benign prostatic hyperplasia (BPH), whereas another 70 had overt symptomatology but not diagnostic pathology of BPH at the time of diagnosis of PC.
Control subjects were selected from the same hospitals as PC cases. It was required that they were residents of the Greater Athens area, never had been diagnosed with cancer or BPH and belonged to the same 5-year age strata as the cases. A complete interview was obtained from 246 of 308 eligible controls. Of those eventually included in the study, 209 had an eye ailment, 25 had a minor injury and 12 had a minor ear, nose and throat condition.
Cases and controls were interviewed in the hospital wards, before discharge, by either of 2 specially trained interviewers. The interview lasted about 1 hr and was based on a structured questionnaire. The questionnaire covered demographic, socio-economic, reproductive, biomedical and dietary variables. Additional details about the study design are given in a companion report on dietary factors (Tzonou et al.,1999).
The data were modeled through unconditional logistic regression, using the SAS statistical package. A core model was used, which included age (in 5-year groups), height (in 5-cm groups), body mass index (BMI, in kg/m2), years of schooling as an indicator of socio-economic status (in 3 categories) and total energy intake (in quintiles of kilocalories). These factors were introduced as ordered variables.
Patients with PC were older than control subjects and of higher educational level, whereas there was no statistically significant or substantial difference between the 2 series with respect to height or BMI (Table I). All subsequent analyses adjusted for these 4 variables. Concerning demographic characteristics, there were no statistically significant differences with respect to marital status, number of offspring and sibship size or birth order in the parental family, but there was significant evidence that cases were more likely than controls to have lived longer in an urban (>10,000 inhabitants) than a rural environment before settling as permanent residents in the Greater Athens area (Table II).
Table I. DISTRIBUTION OF 320 PATIENTS WITH PROSTATE CANCER AND 246 CONTROLS BY AGE, HEIGHT, BODY MASS INDEX (BMI) AND YEARS OF SCHOOLING
| <60||16||24||0.97 (0.33)||2.02 (0.04)|
| <165||51||39||−0.20 (0.84)||−0.61 (0.54)|
| <20||12||15||−0.40 (0.69)||0.12 (0.90)|
|Years of (total)|
| <5||68||81||4.83 (<0.001)||5.26 (<0.001)|
Table II. DISTRIBUTION OF 320 PATIENTS WITH PROSTATE CANCER AND 246 CONTROLS BY DEMOGRAPHIC CHARACTERISTICS
|Usual past residence|
| Rural||104||122||4.09 (0.001)||2.66 (0.008)|
| 1||16||6||−1.35 (0.18)||−0.69 (0.49)|
| 1||91||67||−0.32 (0.75)||0.32 (0.75)|
| Never married||12||11||−0.43 (0.67)||−1.00 (0.32)|
| Ever married||308||235|
|Number of offspring|
| 0||37||25||−0.87 (0.38)||0.26 (0.80)|
Table III shows frequency distributions of cases with PC and control subjects with respect to selected personal characteristics or habits. With the exception of tobacco smoking, there is no evidence that any of these characteristics or habits is a risk factor for PC. Heavy smokers appear to be at lower risk than non-smokers, but in the absence of an exposure–response pattern and supporting published evidence, the inverse association is more likely to be a chance outcome in the multiple testing process.
Table III. DISTRIBUTION OF 320 PATIENTS WITH PROSTATE CANCER AND 246 CONTROLS BY PERSONAL CHARACTERISTICS AND HABITS
| Frontal||163||124||−0.27 (0.79)||−0.46 (0.64)|
| Vertex||105||85||−0.50 (0.62)||−0.79 (0.43)|
| None||28||12||0.39 (0.63)||0.57 (0.57)|
| No||123||95||0.95 (0.34)||0.87 (0.39)|
| Gonorrhea||39||22||1.19 (0.24)||0.73 (0.47)|
| Other||9||8||−0.24 (0.80)||−0.31 (0.76)|
| (per week)|
| 1||3||0||0.89 (0.37)||0.94 (0.35)|
| 1–19||76||48||0.13 (0.90)||−0.07 (0.95)|
| 20+||135||127||−1.87 (0.06)||−2.04 (0.04)|
| None||101||58||−2.14 (0.03)||−0.68 (0.49)|
| None||30||15||0.54 (0.59)||0.27 (0.79)|
In Table IV, the frequency of past surgical operations, as a reliable proxy for the underlying conditions, is compared between cases and controls. For none of the 9 indicated operations is there statistically significant evidence for an association with PC. In Table V, comparisons are made with reference to specific diagnoses. With the exception of tuberculosis, no statistically significant differences were noted. The marginally significant positive association with allergy or asthma is incompatible with the inverse association found in several other studies between these conditions, on the one hand, and other forms of cancer, on the other. The statistically significant positive association with tuberculosis, however, is fairly strong and cannot be explained in terms of residual confounding by social class since cases of PC were, in general, of higher educational level.
Table IV. PAST SURGICAL OPERATIONS REPORTED BY 320 PATIENTS WITH PROSTATE CANCER AND 246 CONTROLS
| No||266||215||1.41 (0.16)||1.03 (0.30)|
| No||319||243||−1.18 (0.24)||−0.39 (0.16)|
| No||227||190||0.59 (0.08)||1.53 (0.13)|
| No||305||234||−0.11 (0.92)||−0.03 (0.98)|
|For stomach ulcer|
| No||309||236||−0.39 (0.70)||−0.58 (0.56)|
|For duodenal ulcer|
| No||311||237||−0.57 (0.57)||−0.73 (0.47)|
| No||309||244||1.90 (0.06)||1.73 (0.08)|
|For kidney stones|
| No||313||243||0.85 (0.39)||0.63 (0.53)|
| No||257||200||0.29 (0.77)||0.18 (0.86)|
Table V. DISEASES REPORTED BY 320 PATIENTS WITH PROSTATE CANCER AND 246 CONTROLS
| No||293||229||0.43 (0.67)||0.31 (0.75)|
| No||231||182||0.42 (0.68)||−0.06 (0.95)|
|Allergy or asthma|
| No||247||207||2.05 (0.04)||1.90 (0.06)|
| No||293||239||2.55 (0.01)||2.45 (0.01)|
| No||294||231||0.92 (0.36)||0.71 (0.48)|
| No||313||243||0.85 (0.39)||0.65 (0.51)|
| No||274||211||0.29 (0.77)||−0.26 (0.80)|
| No||283||229||1.85 (0.06)||1.62 (0.10)|
| Gonorrhea||39||22||1.19 (0.24)||0.73 (0.47)|
| Other||9||8||−0.24 (0.80)||−0.31 (0.76)|
Patients with PC and control subjects were asked to report the weekly frequency of ejaculations in different decades of their lives. Many of the study subjects refused to answer this question, and others gave vague answers, which justifies concern for selection or information bias. The reported data were somewhat more reliable when questions focused on their youth, before the age of 30. Of 123 patients with PC who answered these question, 66 (54%) reported 4 or more ejaculations per week, whereas among the 44 control subjects who answered this question, the corresponding number was 30 (68%). The difference is statistically suggestive (p < 0.10), is not affected by adjustment for possible confounders (current age, reported sexually transmitted diseases, marital status) and implies that frequency of ejaculations in early life may reduce the risk of PC. When focus is shifted to older decades, the data are more sparse and there is no overall evidence for an association between sexual activity and risk for PC.
We have repeated all analyses after exclusion of the 52 PC cases who were identified following an operation for BPH, and the patterns of results remained largely unchanged.
This study was of moderate size, but it had the power to document common risk factors (i.e., factors with prevalence exceeding 0.15) associated with relative risk estimates outside the 0.5 to 2.0 range. It was not possible to exclude individuals with sub-clinical PC from the control series, but many epidemiologists, including ourselves, consider this exclusion undesirable as well as unrealistic: controls represent the study base that generates the cases, and sub-clinical PC is the pool from which the clinical cases eventually emerge (MacMahon and Trichopoulos, 1996).
There is also intense debate in epidemiological publications concerning the relative merits of hospital and population controls (MacMahon and Trichopoulos, 1996). If the proportion of eligible controls who either are inaccessible or refuse to cooperate is low (e.g., <30%), population controls are clearly superior because disease-related selection becomes a non-issue. However, in virtually all populations that lack suitable registries, enrolment rarely exceeds 50%, making the choice of appropriate hospital controls an arguably preferable option as well as economically realistic. Information bias is also minimised when cases and controls are interviewed in a similar setting by the same interviewers, as in our study. Moreover, control of confounding was done in this investigation as well as our current understanding of PC permits and statistical procedures allow.
There are very few black Greeks or men who have undergone vasectomy, so these 2 factors could not be considered in our investigation. Moreover, height was reported, rather than recorded, so that the null relevant finding of our study does not represent a genuine aberration from the collective evidence linking tallness to PC risk (Andersson et al.,1996; Giovannucci et al.,1997a). Weight data are mostly recorded, however, and the lack of association of BMI with PC allows the inclusion of our study with the majority of investigations that found no relation between adult life weight or BMI and PC risk (Nomura and Kolonel, 1991;Andersson et al.,1995; Giovannucci et al.,1997a). In this study, as in a few earlier ones (Hiatt et al.,1994), PC was apparently more common among the more educated, but the association was weak and could be explained, at least in part, by detection bias.
Sibship size and birth order in the parental family strongly affect early-life infection patterns, and infectious agents have occasionally been considered as component causes of PC (Nomura and Kolonel, 1991). We found no evidence for an association of PC risk with either of these variables. We also found no indication that marital status or number of children were associated with PC. Few studies have examined these factors (Nomura and Kolonel, 1991; Andersson et al.,1996), and the collective evidence does not support a strong association of PC with any of them. In contrast, both this study and an earlier investigation (Andersson et al.,1995) suggested that some aspect of urban life may increase PC risk. The finding needs to be confirmed in other data before attempts are made to explain the pattern in causal terms.
Androgens are thought to play an important role in the development of baldness, and high levels of bioavailable blood testosterone have been positively associated with PC risk (Gann et al.,1996). On the basis of this evidence, bald men might have been expected to be at increased risk for PC. In contrast, in our data, the association was, if anything, inverse, and null associations have been reported from other studies that have examined hair patterning in relation to PC (Oishi et al.,1989). It may be that local, rather than systemic, steroid availability modulates the expression of baldness, or that systemic effects of androgens on baldness and PC risk have different time windows. There was also no evidence in our study that dense chest hair was a risk factor for PC, though a strong positive association has previously been reported from a study in Japan (Oishi et al.,1989).
Several studies have examined consumption of alcoholic beverages in relation to PC, the expectation being that the inverse association between ethanol intake, on the one hand, and serum testosterone levels and production rates, on the other, would be reflected in an inverse association between alcohol consumption and PC risk. Contrary to expectations, however, most studies found no association (Nomura and Kolonel, 1991; Hiatt et al.,1994) or a weak positive association (Andersson et al.,1996). The weak inverse association found in our study is compatible with the collective evidence on this topic and supports the hypothesis that drinking alcoholic beverages does not affect PC risk. Neither our investigation nor previous ones found any evidence that coffee intake is related to PC risk. Most reviewers of the epidemiological evidence on tobacco smoking in relation to PC are inclined to conclude that there is no association (e.g., Nomura and Kolonel, 1991), and our results are in line with this. Several studies, however, have reported a weak positive association (e.g., Andersson et al.,1996), which at present cannot be dismissed.
Sexual activity and the likelihood of infection by a sexually transmitted disease are positively associated, and mutual adjustment is generally difficult because the relevant information is frequently poor. Both sexual hyperactivity and sexually transmitted agents have been hypothesised to increase PC risk, and most epidemiological findings have been interpreted as compatible with either or both hypotheses (Nomura and Kolonel, 1991; Andersson et al.,1996). Nevertheless, other studies found no association of PC risk with sexual activity and/or history of sexually transmitted disease(s) (e.g., Ewings and Bowie, 1996), and there have even been reports of inverse associations (Oishi et al.,1990). Our results do not indicate that sexually transmitted agents play an important role in the aetiology of PC; in fact, they appear to be more compatible with the hypothesis that sexual activity in early adulthood may reduce the risk for PC in later life.
PC cases and controls were very similar in frequency of past surgical operations and major medical diagnoses. This is reassuring evidence that the 2 series were comparable but provides no aetiological hints. Diabetes mellitus has been reported to be both positively (Steenland et al.,1995) and inversely (Thompson et al.,1989) associated with PC risk. To the extent that diabetes mellitus may reduce testosterone levels, an inverse association would be biologically more credible. The empirical evidence, however, is, as indicated, equivocal, and the results of our study provide no support to the hypothesis of an inverse association between diabetes mellitus and PC. A somewhat similar situation exists with respect to liver diseases, which are occasionally associated with higher levels of estrogens or lower levels of androgens. Notwithstanding biomedical plausibility, however, epidemiological studies, including the present one, have provided limited or no evidence for an association of liver diseases with PC (Wynder et al.,1971). We are also inclined to consider the 2 suggestive associations in our data on medical history, those with tuberculosis and atopy, as chance findings, even though adjustment for a series of possible confounding variables did not affect the strength of the respective associations.
In conclusion, our results provide strong evidence against BMI, reproductive patterns, tobacco smoking, alcohol drinking, coffee consumption, hair patterning and medical history (including history of sexually transmitted diseases) being important risk factors for PC. There was, however, evidence in our data that some aspect of urban life may increase the risk for PC and a suggestion that sexual activity in early adulthood may reduce this risk.
This study was supported by the Europe Against Cancer Programme of the European Union and grant DY5a/1183-F210-KAE2549/1995 from the Greek Ministry of Health. The authors gratefully acknowledge the contribution of the interviewers, public health visitors, Ms. D. Filippa and Ms. S. Lambrokostopoulou.