Androgenetic Alopecia and Prostate Cancer: Findings from an Australian Case-Control Study¹

One important characteristic of prostate cancer is its rapid increase with age. Male pattern baldness, AA,³ is also strongly age dependent and, similar to prostate cancer, is considered to be androgen dependent (1, 2).

Androgens exert their effects by binding to a single cytoplasmic AR, and their potency is determined by the binding affinity to the AR, with DHT binding five times more strongly than T (2). The enzyme 5αR converts T to its active form, DHT. DHT is implicated not only in the development of benign prostatic hypertrophy but also in the pathogenesis of prostate cancer (3, 4). Isozymes of 5αR are differentially expressed in tissues; 5αR-1 is expressed in the skin, sebaceous glands, liver, adrenal, and kidney, whereas 5αR-2 is expressed in the prostate, testes, seminal vesicles, liver, and hair follicles (5, 6). Inherited deficiency of 5αR-2 leads to absence of AA and a small prostate (7). Finasteride, a 5αR-2 inhibitor with little 5αR-1 activity, has been useful in the treatment of AA and benign prostatic hypertrophy (8, 9). Finasteride down-regulates expression and secretion of PSA (10), but its short-term use in the chemoprevention of prostate cancer, benign prostatic hypertrophy, and elevated PSA has not been successful (11), and long-term use is still subject to trial (12). Studies that have specifically addressed the question of whether AA is associated with prostate cancer are few and have produced inconsistent findings (13, 14, 15, 16, 17). We examined associations of AA with early-onset, moderate- to high-grade prostate cancer in a large case-control study (18). The main thrust of the case-control study was to examine lifestyle associations with the diagnosis of “clinically important” prostate cancer. To this end, we excluded tumors that were well differentiated (low grade or Gleason score <5). We also focused on early-onset cancers because we were interested in finding factors relevant to the prevention of prostate cancer in men before the age of 70.

We carried out an age-stratified, population-based case-control study of prostate cancer in Melbourne, Sydney, and Perth, Australia (18). The subjects were residents of the three cities’ metropolitan areas. Prior approval of the study protocol was obtained from all relevant hospital and cancer registry human research ethics committees in Victoria, New South Wales, and Western Australia.

Eligible cases comprised all male residents of Melbourne, Sydney, and Perth diagnosed from 1994 to 1997 and recorded in the population-based cancer registries with a histopathology-confirmed diagnosis of adenocarcinoma of the prostate (International Classification of Diseases, 9th revision, rubric 185), excluding well-differentiated tumors (defined as low grade, i.e., those with Gleason scores <5). Cases had to be <70 years of age at diagnosis and also had to be registered to vote on the state electoral rolls (adult registration to vote is compulsory in Australia). Meeting this criteria, all cases diagnosed before the age of 60 years were included, and random samples of 50% of cases diagnosed at 60–64 years of age and 25% of cases diagnosed at 65–69 years of age were selected, with the proportions varying over time to fit interview quotas.

Controls were randomly selected from men on the current state electoral rolls and were frequency matched to the predicted age distribution of the cases in a ratio of one control per case. Potential controls were matched against the cancer registries at the time of recruitment to exclude men with a known history of prostate cancer. Controls were identified and interviewed contemporaneously with the cases over the period 1994 to 1997. During the course of this study, 3 controls were subsequently diagnosed with prostate cancer and were selected as eligible cases. These subjects are included as cases and as controls.

After seeking advice from the case subjects’ urologists, from whom some clinical details were sought, we wrote to each subject inviting him to participate. We wrote to the control subjects directly. Face-to-face interviews were arranged, usually at the man’s home. A structured interview schedule was used to obtain information on lifestyle exposures and personal attributes. While the subject was completing a sexual history questionnaire in private, the interviewer scored the subject’s AA according to a set of four pictures (Fig. 1) adapted from the Hamilton-Norwood scale (19). The first picture shows essentially no balding (Hamilton-Norwood stages I and II), the second shows frontal balding (Hamilton-Norwood stages II, III, IIIa, and IVa), the third shows vertex balding (Hamilton-Norwood stage III vertex-V), and the fourth shows frontal baldness concurrent with vertex baldness (Hamilton-Norwood stage IV, V, Va, VI, and VII).

Case-control analyses were conducted using unconditional logistic regression, adjusting for reference age (at diagnosis for cases and at date of selection from electoral roll for controls), study center (Sydney, Melbourne, and Perth), calendar year, family history (none versus any first-degree relative diagnosed with prostate cancer), and country of birth (Australia versus other). Study center and calendar year were included because the zeal for prostate cancer screening using PSA tests varied among centers and over time (20). The ORs and CIs of each category of AA relative to no balding were estimated. The effect of including AA in the model was assessed by a likelihood ratio test with 3 degrees of freedom.

Given that smoking and educational status may be predictors of response, we assessed the association of these factors with AA (no balding versus other) by logistic regression. To investigate whether effects differed by potential disease aggressiveness, we analyzed men with moderate-grade (Gleason scores 5–7) and high-grade (Gleason scores 8–10) tumors separately. We also examined the effects in two age groups (i.e., reference age <60 versus 60–69). Tests for heterogeneity in the ORs between high- and moderate-grade prostate cancer were performed using polytomous logistic regression models (21).

Of the 2328 cases and 3125 controls that were considered eligible at the time of selection, 1497 cases (65%) and 1434 controls (46%) were interviewed. Doctors could not be identified for, or refused access to, 16% of cases. Controls were more likely to refuse (38% versus 17%), to be unable to speak sufficient English to be interviewed (7% versus 1%), or to have moved (8% versus 2%).

Because the electoral roll is never completely up to date, we considered that a more appropriate response rate in controls excluded those who had moved or died, which was 50%. After excluding subjects with missing data on AA and the variables to be controlled for in the analysis, there were 1446 cases (1088 with moderate-grade and 358 with high-grade tumors) and 1390 controls available.

The distributions of demographic variables are shown in Table 1. The cases were more likely than the controls to be Australian born (OR, 1.38; 95% CI, 1.17–1.63) and to have at least one first-degree relative affected with prostate cancer (OR, 3.21; 95% CI, 2.45–4.23). There was no association between AA and educational status, smoking status, or migrant status in either cases or controls (all P > 0.1). Models that excluded those born overseas produced very similar findings (data not shown).

Table 2 shows that, compared with no balding, vertex balding was associated with a 50% increase in risk of prostate cancer in all subjects. No increased risk was seen for frontal balding or frontal concurrent with vertex balding. The effect of vertex balding appeared stronger in the subgroup of high-grade tumors (2-fold increased risk), but the differences in ORs between moderate- and high-grade tumors were not significant (P = 0.6).

The ORs for all cases in each of the two age groups were similar to those for the whole group (Table 2). In the younger subjects, the ORs were also similar for moderate- and high-grade disease (Table 2). In the older subjects, the ORs for moderate-grade disease were again similar. For high-grade disease, the ORs for all types of AA were elevated, nearly 3-fold for vertex balding and close to 2-fold for frontal balding and frontal concurrent with vertex balding, and all CIs excluded unity. However, as in the analysis in which the age groups were combined, the difference between the ORs for high-grade and moderate-grade tumors was not significant (P = 0.31).

Our analysis suggests a positive association between prostate cancer and vertex baldness that appeared to be more evident for high-grade prostate cancer, especially when diagnosed in men 60–69 years of age. We have considered the extent to which this finding might be attributable to bias or confounding, given the response rates, and the fact that subjects were ascertained during a period of intense PSA testing in the population (20). With respect to response, in neither cases nor controls could we find an association between either educational status or smoking status (as surrogates for response) with AA (data not shown). The association between prostate cancer and AA was at least as strong for high-grade prostate cancer as for moderate-grade prostate cancer, suggesting that PSA testing, which identifies large numbers of moderate-grade tumors (20), cannot explain the difference. Furthermore, we believe it is implausible that vertex balding would be associated with PSA testing, and although the interviewers were often not blind to the case-control status of subjects, they were not informed of any hypothesis concerning AA. Having controlled for the strongest established risk factors (age and family history) and given the lack of other known risk factors, we consider that confounding is unlikely to have influenced our findings.

Our observation of an increased risk of prostate cancer associated with vertex baldness is consistent with two small case-control studies (13) and one cohort study (14) but not with another cohort study (15) and two hospital-based case-control studies (16, 17). Of those studies not in agreement with ours, the cohort study of Harvard alumni (15) assessed men’s baldness from photographs taken from college yearbooks published 25 years after graduation when former students were in their mid-forties and compared the men who had died from prostate cancer with those who had not. No attempt was made to assess vertex baldness. A hospital-based case-control study (17) conducted in Athens, Greece, on 320 prostate cancer patients and a mix of 246 patients without prostate cancer excluded men with benign prostatic hypertrophy from the controls. This could be problematic, given that AA may be associated with benign prostatic hypertrophy (22) and both are related to 5αR activity (23, 24).

Of the case-control studies in agreement with ours (13), one was a hospital-based study of 78 cases and 71 controls and the other was a community-based study of 56 cases and 74 controls (13). The Hamilton-Norwood scale was used to measure AA (19); subjects were asked to choose pictures that best described their hair pattern at 30 and 40 years of age. The analysis produced consistent but statistically not significant results: an OR for vertex baldness of 2.44 (95% CI, 0.57–10.46); an OR for baldness by age 30 of 2.11 (95% CI, 0.66–6.73); and an OR for baldness by age 40 of 1.37 (95% CI, 0.47–4.06). A follow-up of the first National Health and Nutrition Examination Survey, comparing 214 prostate cancer cases with the remaining 4421 men examined for AA at baseline when 25–75 years of age, found an OR of 1.5 (95% CI, 1.12–2.00) for any form of baldness (14). This estimate has been considered to be possibly attenuated, not only by the lack of specificity of the baldness measurement (they were not able to distinguish vertex baldness) but also by the wide range of ages at which baldness was assessed (25). It is possible that any association with baldness may be stronger in men who become bald at an early age.

A mechanism for the putative relationship between AA and prostate cancer risk is yet to be established. Because both AA and prostate cancer are androgen dependent, differences in androgen metabolism, coactivators of the AR, AR gene mutations and polymorphisms of the AR, and 5αR genes are all obvious candidates for investigation (26, 27, 28). Other candidates for investigation include physiological pathways important to prostate cell differentiation and proliferation, e.g., IGF-1 and the VDR. IGF-1 can lead to aberrant activation of the AR and mediates the perpetuating effects of growth hormone on AA (29). Vitamin D (as 1α,25-hydroxyvitamin D) inhibits prostate cell growth (30), and polymorphic variation in the VDR has been linked to prostate cancer risk (31). 1α,25-Hydroxyvitamin D resistance has been linked to alopecia in humans (32), and VDR knockout mice also develop alopecia (33).

A case-control study of 159 cases and 156 controls found a positive association between free T levels in serum from men with frontal or vertex baldness, compared with men who had only minimal hair loss (16). The association between T (and IGF-1) and AA was also found in a cross-sectional study (34). Associations between prostate cancer and elevated T have been reported in a case-control (35) and a prospective (36) study. In the latter, elevated T levels in blood sampled before diagnosis were associated with increased risk of prostate cancer, especially advanced disease. Other analyses of this cohort study have shown positive associations between IGF-1 and prostate cancer (37) and also between IGF-1 and vertex baldness (38).

It is considered that premature AA is related to high levels of androgens generally and to high DHT levels specifically in the frontal scalp (39), with 5αR-2 playing a central role in the intrafollicular conversion of T to DHT (2). This is supported by the immunohistochemical localization, in cryosections of scalp from men with AA, of 5αR-1 staining within sebaceous glands but not in hair follicles and 5αR-2 staining in the root sheath and the infundibular region of the follicle but not within the dermal papilla or sebaceous glands (40). Others have shown that the outer root sheaths of frontal hair follicles have higher levels of AR, 5αR-1, and 5αR-2 and less aromatase than in occipital follicles (41), and a higher level of AR has been demonstrated in hair follicles from balding skin compared with nonbalding skin (42). Aberrant activation of the AR has been demonstrated in vitro with IGF-1, keratinocyte growth factor, and epidermal growth factor. These agents can directly activate the AR in the absence of androgens and may contribute to the progression of prostate cancer and AA (43, 44). Some consider that prostate cancer risk might be associated with the CAG repeat polymorphism in the AR (11, 27, 28), although we have been unable to detect such an association in our study (data not shown). However, shorter CAG repeat lengths in the ARmay affect androgen-mediated gene expression in hair follicles and sebaceous glands (45). Platz et al. (38) and Ellis et al. (46), in comparing men with early-onset AA and older nonbald men, failed to detect any variation in allele, genotype, or haplotype frequencies in the genes encoding 5αR-1, 5αR-2, and insulin, suggesting that these were not associated with early-onset AA. This is not altogether surprising because whole follicle transplantation experiments have demonstrated that each hair follicle is genetically programmed not only to respond, or not respond, to androgens but also in what manner to respond (47). Although the geographical patterning of the hair loss in AA is associated with quantitative differences in androgens and numbers of ARs, these are likely to be secondary phenomena because the hair follicle is able to regulate its own response to androgens by enhancing expression of 5αR and ARs in vitro (48). Genetic control of AA may reside with differentiation/morphogen genes, e.g., genes that code for developmental regulator proteins implicated in the sonic hedgehog signaling pathway or its cognate receptor patched (49). Notably, these genes also play an important role in oncogenic transformation (50).

The extent to which different androgens interact with each other, and with the molecules that produce, transport, activate, receive, and remove them from circulation, is far from completely understood. It is possible that common polymorphisms in genes that encode steroid hormones and their reductases and other relevant molecules, e.g., the genes for IGF (and its receptor), the AR, and aromatase, might influence the etiology of all these conditions. To better understand this complexity, there is an obvious need for larger studies of prostate cancer and AA that include measures of polymorphic variation in an increasing number of candidate genes.

We thank study coordinator Margaret Staples and the research team, Bernadette McCudden, John Connal, Richard Thorowgood, Chris Costa, Melodie Kevan, and Sue Palmer. We thank the many urologists who kindly assisted us by providing information and access to their patients. We also thank the many men who participated.