Notes Washed Up in a Bottle

Don. W. MacCorquodale M.D. M.S.P.H.

Periodic Notes From the Field on Philosophy and Science.

POSTMENOPAUSAL HORMONES AND CARDIOVASCULAR DISEASE

          A long, complex article on postmenopausal hormones and cardiovascular disease in the American Journal of Epidemiology (R. L. Prentice et al) has been ably summarized by Diana B. Petitti and David A. Freedman (Invited commentary:   How far can epidemiologists get with statistical adjustment.   Am J Epidemiology 2005; 162:   415-418).   A report on the Women's Health Initiative (WHI) clinical trial pointed out that combined estrogen-progestin therapy 1) did not prevent coronary heart disease, 2) increased the risk of stroke by a factor of 1.4 on average, and 3) doubled the risk of venous thrombosis.

          Earlier studies suggested that the relative risk of coronary heart disease was 0, 50-0.60 in hormone users as compared to non-users.   Seven randomized trials, however, reported no benefit on arterial disease among hormone users, and a 1997 meta-analysis of small randomized trials found that hormone therapy did not prevent cardiovascular disease (emphasis – DWM).   The estrogen-only arm of the WHI clinical trial was terminated early after it failed to demonstrate prevention of coronary heart disease.  

          The Prentice article reported an analysis of the data from the WHI clinical trial and the WHI observational study.   Both were conducted by the same researchers during the same time frame.   Prentice and his colleagues examined the reasons for the differences in estimates of coronary heart disease, stroke, and venous thrombosis between the clinical trial and the observational study by controlling for various confounders.  

          Petitti and Freedman pointed out that “given the differences in age between hormone users and nonusers in the WHI observational study, it is not surprising that age adjustment has a large effect on hazard ratios for all three endpoints.”   They added that adjustment for such confounders as known cardiovascular risk founders and education, the hazard ratios for coronary heart disease, stroke, and thromboemoblism for the WHI observational trial all moved more closely to those of the WHI clinical trial.   Petitti and Freedman suggested that “inclusion of a measure of socioeconomic status (SES) in the analysis of observational data for virtually any exposure/disease relations seems fundamental.”   A summary estimate of the relative risk of coronary heart disease on the part of current hormone users based solely on studies that adjusted for a measure of SES, the relative risk was 0.97, 95% CI:   0.82-1.16.   Reports on coronary heart disease in the most influential observational study of hormone therapy, the Nurses Health Study, did not adjust for SES.  

          The authors of this commentary concluded that “the analysis by Prentice et al confirms that there is no overall benefit of hormone therapy in preventing cardiovascular disease.”   They argued that earlier conclusions that hormone therapy prevented coronary heart disease were simply wrong.

          Petitti and Freedman wrote that Prentice and his colleagues highlighted some ways that observational studies could be better designed and better analyzed, and they added that observational studies are not a substitute for clinical trials.  

          COMMENT:   I can not recall any earlier epidemiologic study that created quite such a stir as the report of the WHI clinical trial that showed no benefit from hormone therapy as far as preventing coronary heart disease was concerned.   It certainly shook me.   Somehow, I had failed to read the articles reporting observational studies that showed no benefit, and while I did read the Nurses Health Study, I did not notice that it did not control for SES.   I suppose the authors assumed that because all of the subjects were nurses, they all came from the same socioeconomic background.  

          Of course, observational studies are no substitute for clinical trials, but clinical trials for some reputed risk factors cannot be conducted for ethical reasons.   No one could possibly propose a trial in which non-smoking, pregnant women were randomized to continue not to smoke or to start to smoke and smoke throughout their pregnancies for some monetary reward.   There are times when the best we can hope for are carefully designed and analyzed observational studies.  

STATINS AND PROSTATE CANCER

          One of every six men 60 years of age or older will be diagnosed with cancer of the prostate, the second leading cause of cancer death among American men (Shannon J et al.   Statins and prostate cancer risk:   A case-control study.   Am J Epidemiology 2005:   162:   318-325).     

          Statins have been shown to inhibit the proliferation of prostate and breast cancer cells, and the induction of tumor-specific apoptosis in in vitro studies.   A recent meta-analysis of five trials of stains and heart disease with sufficient power to assess a cancer-reducing effect, reported no association between statin use and the risk of developing incident cancers.   Studies that examined statin use and incident cancer risk have been largely observational and have yielded conflicting results.   The authors conducted a case-control study of diet and prostate cancer risk.   The study population consists of biopsy-confirmed prostate cancer and clinic controls with a normal (<4 ng/ml) prostate-specific antigen test within the past 12 months.  

          All veterans referred to the Portland Veterans Affairs Medical Center (PVAMC) for a biopsy of the prostate between Dec. 2001 and Aug. 2004 were eligible.   Slightly more than 780 men were referred, and almost 700 met the eligibility criteria.   About 550 were contacted and 349 veterans agreed to participate in the study.   Almost 300 (85%) underwent biopsy.   Of these, 100 (34%) were diagnosed with prostate cancer, and almost 180 (60%) had a negative biopsy.   Six percent, n = 19, had prostatic intraepithelial neoplasia.   Data regarding use of statins was obtained from the records of a pharmacy database at the PVAMC.        

          There were 100 9incident prostate cancer case and 202 clinic controls for analysis.   The median age was 65.5 for cases and 65.0 for controls.   Among prostate cancer cases, the majority (92%) had localized disease.  

          Clinic controls had a significantly higher total fat (n = 0.05) and calorie (p = 0.02) intake as well as a greater body mass index (p = 0.04).   Prostate cancer cases were significantly more likely to be African-American.   The prevalent use of any statin and specifically lovastatin and simvostatin was significantly greater among clinic controls (p – 0.01, 0.01 and 0.04 respectively).  

          The use of NSAIDs did not differ significantly among the subject groups.   There was no significant difference in the number of subjects with diabetes, a family history of cancer, or a family history of prostate cancer.  

          After adjustment for age, men with any recorded statin use had a 50 percent lower risk of prostate cancer. OR = 0.5, 95% CI:   0.30, 0.80.   After adjustment for age, race, BMI, NSAID use, diabetes, total calorie, and use of other lipid-lowering drugs, men with any recorded statin use had a 65 percent reduction in risk of prostate cancer, OR = 0.35, 95% CI:   020, 0.64.   The reduction in risk was greatest for the most aggressive tumors, i. e., those with a Gleason score of 7 or more, OR = 0.24, 95% CI:   0.11, 0.53.  

          With regard to limitations of the study, the authors pointed out that it is possible that some of their subjects may have obtained statin prescriptions outside of the VA system, and they would not have been able to detect that.   Selection bias is a possibility since 42% of the eligible controls refused to participate or failed to show up for appointments.   Misclassification seems unlikely.   All cases were confirmed by biopsy, and all controls had normal PSAs.     

          COMMENT:   I find the evidence presented here that statins reduce the risk of prostate cancer very compelling.   If I were thirty years younger, I would seriously consider taking statins.   I already take an aspirin every other day in hope of preventing a coronary thrombosis.   I wonder where this will end.   Will healthy middle-aged men and women find themselves taking a dozen or so medications daily to ward off various diseases?   I'm sure my fear isn't entirely rational, but somehow, I find the prospect frightening.      

POLYCHLORINATED BIPHENYLS AND PREGNANCY OUTCOME

          Polychlorinated biphenyls (PCBS) are widespread contaminants of the environment, including food.   Low level exposures to PCBS have been reported as associated with low birth weight in several studies.   Many studies in humans have failed to show the association however.

The authors of this study measured PCBS in 1,168 maternal serum samples and examined their association with birth weight (Longnecker M P et al.   Maternal levels of polychlorinated biphenyls in relations to preterm and small-for-gestational-age (SGA) birth.   Epidemiology 2005; 16:   641-647).   The authors measured PCBS in slightly more than 1,100 pregnant women to examine their relationship with birth weights.

          Serum levels of PCBS, DDT, DDE, and 7 other oregano-chlorines were measured at the Centers for Disease Control and Prevention.   Length of gestation was the date of delivery minus the date of the last menstrual period, and delivery was classified as preterm if it occurred before 37 completed weeks of pregnancy.   SGA was defined as birth weight less than the 10 th percentile at each week of gestation using all live births in a larger study from which the 1,100 participants were selected.

Logistic regression was employed to estimate odds ratios and 95% confidence limits.   Of the more than 1,100 participants, almost 130 were excluded because of missing covariate data.  

          Almost half the subjects were black, nearly as many were white, and Puerto Ricans comprised the balance.   One third of the subjects had no previous pregnancies.   Most of the women, 59%, had a socioeconomic index that was below the U. S. median of 5.   The overall prevalence of preterm birth in this population was 13%, greater than that current for the entire U. S.   The prevalence of preterm birth was greater among boys, blacks, those with lower SES index, and smokers, which is consistent with known risk factors.   The greater prevalence of preterm birth among younger mothers, mothers with short stature, lower prepregnancy BMI, or lower rate of pregnancy gain has been reported previously but less consistently.  

          Black mothers, mothers who were shorter, slimmer, had lower rates of pregnancy weight gain, or who were nulliparous or smokers had a greater proportion of SGA infants, which is consistent with previous reports.  

          For preterm birth, the OR (odds ratio) for the highest compared with the lowest exposure category was 1.1 with a wide confidence interval.   PCB levels were associated with odds of SGA after adjustment for child sex, race, maternal age, smoking, serum cholesterol level, and other variables, but the estimates were imprecise.   For those in the highest exposure category compared with those in the lowest, the adjusted OR for boys was 6.6, 95% CI:   1.4-30.0, and for girls, it was 0.9, 95% CI:   0.3-2.6.  

          For SGA the association with PCBS was stronger among blacks.   The adjusted OR for those in the highest exposure category compared with the lowest was 3.7, 95% CI:   1.1-12.3, for blacks.   It was 0.6, 95% CI:   0.2-2.4, for whites.   In a model of birth weight adjusted for the same variables, an association with PCB levels was not seen.   Similarly, length of gestation showed little relation with levels of PCB exposure.

          The authors concluded that their research provided little evidence to support an association of low-level PCB exposure with preterm birth, birth weight, or length of gestation.   For SGA, the data were inconclusive but suggested a direct relation that was stronger among boys and blacks.  

          COMMENT:   Many occupational and environmental health studies are flawed by lack of precise estimates of exposure.   Typically, one reads workers were considered exposed if they worked one or more months in a given plant that produced Compound X.   This study has the advantage of quite precise estimates of exposure based on serum levels of PCBS.

          EPA has classified PCBS as probably human carcinogens, B2.   The studies of carcinogenicity are all based on samples of fewer than 4,000 workers, usually about half that number.   Interestingly, one study found an excess rate of skin cancers, all melanomas, in a group of workers exposure to a given PCB.