Tobacco Documents Online

Analysis of "Multicenter Case-Control Study of Exposure to Environmental Tobacco Smoke and Lung Cancer in Europe," Boffetta, P., Agudo, A., Ahrens, W., et al., Journal of the National Cancer Institute 90 (19): 1440-1450;1998 A. Introduction The long-awaited publication of the IARC multicenter epidemiological study, designed to investigate the possible association of reported environmental tobacco smoke (ETS) exposure with lung cancer, has been recently published in the Journal of the National Cancer Institute. This study is not only one of the largest case-control epidemiological studies which has been carried out to investigate this possible association, but also one of the best conducted. The results of this study suggest an extremely weak, not statistically significant numerical association between reported spousal or workplace ETS exposure and lung cancer (OR = 1.16, 95%; Cl, 0.93-1.44 and OR = 1.17; 95% Cl, 0.94-1.45, respectively). The authors report no associations between reported ETS exposure in social situations or during childhood and lung cancer in non-smokers. This analysis will begin with a summary of the study, followed by a brief discussion of the reported results. Following this will be a detailed analysis of the results in order to determine if, as suggested by an accompanying editorial (Blot, W. J., and McLaughlin, J. K., 1998), "...the inescapable scientific conclusion is that ETS is a low-level lung carcinogen." This analysis will address two factors; namely, the possible contribution of systematic biases to the reported odds ratios, and the general ability to draw conclusions regarding associations between ETS exposure and lung cancer based on the existing published epidemiological data. B. Description of the Study 1. Study Design The study was a multicenter, case-control study with 12 centers from Europe contributing. To a large extent each center used the same design; however, there were some differences noted among the centers. The most important differences are summarized below. They are addressed in detail in the body of our analysis. a. Selection of controls Control subjects were hospital based in the centers from France, Portugal, Spain, and one of the Italian centers (Italy 3). Control subjects were both hospital and community based (percentages not given) in the center from the UK. All other control subjects were community based. It should be noted that all hospital controls with smoking related diseases were excluded from the control groups in all centers. In addition, there was an inconsistency among centers with respect to the matching of cases and controls. Individual matching was done for some centers but others used frequency matching. 1

b. Diagnostic criteria for case eligibility The report states that "in selected centers, case subjects without a histologic or a cytologic diagnosis were also included." There is no information given as to the number of centers which included cases without histologic or cytologic confirmation of diagnosis, nor is there any information as to the overall percentages of such cases without this confirmation. On the other hand, the report points out that 96.5% of the cases had microscopically confirmed disease. c. Inclusion of smokers Smokers were studied in all but the Portuguese centers. The term "smokers" in this case refers to individuals who smoked not more than 400 cigarettes lifetime. According to the publication, the study was designed to have a power of 80% to detect a relative risk of 1.3 significant at the 95% confidence level. 2. Subjects The study included 650 non-smoker lung cancer cases where non-smoker, as indicated above was defined as having smoked less than 400 cigarettes in a lifetime. As was also noted above, lung cancer was confirmed by either cytology or histology for 96.5% of all cases. A total of 1542 control subjects took part in the study. Response rate, particularly for controls, appeared to vary quite substantially among the centers. It was less than 50% for two of the German centers and one of the Portuguese centers. These three centers contributed more than 30% of the total cases. With respect to the remaining centers, the publication simply states that the response rates were 55% to more than 95%. Clearly these low response rates for at least 3 of the centers could have produced significant bias in the results, but there is no way to evaluate the possible extent of such bias. The distribution of cases among the study centers is uneven. Most notable is the major contribution made by the three German centers, greater than 35% of the total cases, whereas all other countries contributed only around 10%. 3. Measurement of ETS exposure A common questionnaire was used to estimate ETS exposures. Methodology for estimating amounts of exposure differed according to what exposure metric was being examined. a. Childhood exposure Quantitative variables used in the estimation of childhood ETS exposure (exposures up to age 18) included the number of smokers in the household and the cumulative exposure. The latter was expressed as the number of years of exposure weighted for the type of smoker - mother, weight of 1; father, weight of 0.75; and other adults, weight of 0.25. These weights were reportedly based on urinary cotinine concentrations in children. 2

b. Spousal exposure Quantitative variables for estimating exposure to ETS from the spouse within marriage, as well as from other cohabitants, included the following: 1) the total number of years of exposure; 2) the product of the number of years and the number of hours per day of exposure; 3) the average number of cigarettes smoked per day by the spouse in the presence of the index subject; and 4) the cumulative exposure expressed as pack- years and derived from the product of variables 1 and 3 listed above. For reported exposure to ETS from sources other than cigarettes the following conversion factors were used. The number of cigarillos smoked was multiplied by 2, and for cigars and pipes, the number was multiplied by 3. The authors state that a relatively small percentage of ETS exposure came from sources other than cigarettes, and exclusion of these data would not have in any way changed the results. C. Workplace Exposure Quantitative variables for reported workplace ETS exposure were the total number of years of exposure and the total number of years of exposure weighted for the number of hours of exposure per day and for a "subjective index of smokiness in the workplace." No information was provided as to how the "subjective index of smokiness in the workplace" referred to above was obtained. Data were also collected relating to time since either reported spousal or workplace exposure ceased. For each source of reported exposure, investigation of possible dose-response trends was carried out as follows. Cases and controls who reported no exposure to ETS were considered as the reference group. Exposed individuals were grouped into three categories defined as below the 75th percentile, between the 75t'' and 90th percentile or above the 90t'' percentile. Two-tailed tests for linear trends were carried out by testing the significance of the regression parameter of a trend variable which also included the reference category. 4. Other factors considered The common questionnaire apparently also collected information on demographic variables, residential history (including cooking and heating arrangements), and exposure to known or suspected occupational lung carcinogens. Eight of the centers also collected some data on dietary intake from which were derived indicators of intake of vegetables, fruits, P-carotene, total carotenoids, and retinol. Potential confounders that were analyzed by the regression analysis were educational level (as a variable with three categories based on center-specific cut points), proportion of life spent in urban areas, occupational exposure to lung carcinogens, and intake of vegetables, P-carotene, total carotenoids, and retinol. C. Results 1. Childhood exposure Based on 389 cases and 1021 control subject reporting exposure during childhood, an overall odds ratio of 0.78 (95% Cl, 0.64-0.96) was reported. The authors claim that there was a 3

significant decreasing trend as a function of increasing exposure with a test for trend of p = 0.02 (1.00, 0.83, 0.68, 0.90 with increasing level of cumulative exposure). It should also be noted that these figures include all subjects exposed during childhood whether they were exposed later in life or not. The authors do report that analysis of the data excluding all individuals reporting adult exposure gave results that were "similar... although more unstable because of the small numbers." 2. Spousal Exposure: ,, The study reports OR's for spousal exposure using two different indexes. The index chosen by the authors as being most representative of actual exposure is "self-reported exposure to spousal smoke." For this index the reported OR for all subjects was 1.16 (95% Cl, 0.93-1.44) based on 344 case subjects and 700 control subjects. The other index used was "ever married to a smoker." The OR's reported for this index were 1.27 (95% Cl, 1.00-1.62) for all subjects, 1.20 (95% Cl, 0.92-1.55) for women, and 1.65 (95% Cl, 0.85-3.18) for men. A summary of the results presented in the publication are displayed in the table below. Sub rou anal sis OR (95% CI Comments Overall population - "ever married to a smoker" 1.27 (1.00-1.62) It would appear that "ever-married" also includes co-habitants that were not married. Women 1.20 0.92-1.55 Men 1.65 (0.85-3.18) "self-reported exposure to spousal smoke" 1.16 0.93-1.44 344 cases vs. 700 controls Excluding "never married" 1.18 0.92-1.51 Ever exposure to spousal smoke, stratified by sex Women 1.11 (0.88-1.39) 321 cases vs. 623 controls Men 1.47 0.81-2.66 23 cases vs. 68 controls Duration (years of exposure) P for trend = 0.10 0 1.00 305 cases vs. 838 controls 1-34 1.05 (0.83-1.33) 223 cases vs. 498 controls 35-42 0.63 (0.12-2.37) 65 cases vs. 103 controls >43 1.07 0.68-1.68 38 cases vs. 80 controls Duration (hrs/day x yrs.) P for trend = 0.02 0 1,00 297 cases vs. 778 controls 1-135 0.90 (0.70-1.16) 165 cases vs. 396 controls 136-223 1.20 (0.78-1.85) 44 cases vs. 81 controls ?224 1.80 1.12-2.90 41 cases vs. 53 controls Average exposure (cigs./day) P for trend = 0.88 0 1,00 297 cases vs. 778 controls 0.1-10 1.10 (0.86-1.40) 206 cases vs. 411 controls 10.1-18.0 0.58 (0.35-0.90) 25 cases vs. 83 controls >18• 1 1.37 0.85-2.20 35 cases vs. 55 controls Cumulative exposure(pack yrs) P for trend = 0.09 0 1.00 297 cases vs. 778 controls 0.1-13.0 1.00 (0.78-1.28) 188 cases vs. 411 controls 13.1-23.0 0.89 (0.57-1.39) 36 cases vs. 83 controls >23•1 1.64 1.04-2.59 42 cases vs. 55 controls Adjusted for exposure to suspected or known 1.18 (0.94-1.46) occu ationaf carcino ens 4

Adj. for urban vs. rural residence 1.15 0.91-1.45 Adj. for consumption of vegetables above or 1.14 (0.89-1.45) below median level Stratified by age <55Y 0.99 (0.64-1.52) 55-64y 1.19 (0.80-1.76) 65-74Y 1.25 (0.89-1.75) ~ ~ Stratified by histology Not statistically significant. "For all Small cell (11.3% of cases) 1.39 (0.79-2.45) major histologic types, a dose- Squamous cell (17.2% of cases) 1.21 (0.77-1.91) response relationship was suggested with cumulative exposure and duration Adenocarcinoma(5o.6% of cases) 1.08 (0.82-1.42) (in hours/day x years) of exposure to s ousal smoke." Exp. from cohabitant other than spouse Based on only 44 cases unexposed to Ever Exposed 1.10 (0.88-1.36) spousal smoke. 0.1-13.0 pk yrs 0.96 (0.74-1.23) 13.1-25 pk yrs 1.02 (0.66-1.59) ?25.1 pk yrs 1.37 0.85-2.20 The authors claim that there is no significant heterogeneity of reported effect as shown by tests for heterogeneity (P = 0.42). They do note, however, that there is considerable inter- center variability with point estimates ranging from < 0.7 in one center to >1.5 in four centers. The authors also claim that there is little effect of adjusting for the potential confounders that they included in their multiple regression analysis. As is often the case for such studies, however, they fail to address that they may have missed some potential confounders, that they may not be able to accurately measure confounding effects, or that the effect of all confounders combined could have had a greater effect on the relative risks. 3. Workplace exposure The authors write: "A total of 374 case subjects and 855 control subjects reported ever exposure of ETS at the workplace, yielding an OR of 1.17 (95% Cl, 0.94-1.45)." Results reported for workplace exposures are displayed in the table below. Sub rou anal sis OR (95% Ci Comments Overall 1.17 0.94-1.45 374 cases vs. 855 controls Women 1.19 0.94-1.51 269 cases vs. 476 controls Men 1.13 0.68-1.86 105 cases vs. 379 controls Duration (years) P for trend = 0.21 0 1.00 276 cases vs. 687 controls 1-29 1.15 (0.91-1.44) 278 cases vs. 634 controls 30-38 1.26 (0.85-1.85) 55 cases vs. 129 controls ?39 1.19 0.76-1.86 39 cases vs. 91 controls Duration (level x hrs./day x yrs.) P for trend <0.01 0 1.00 276 cases vs. 687 controls 0.1-46.1 0.97 (0.76-1.25) 196 cases vs. 525 controls 46.2-88.9 1.41 (0.93-2.12) 47 cases vs. 105 controls _89.0 2.07 1.33-3.21 48 cases vs. 71 controls By histology No statistical significant differences Small cell 1.17 (0.67-2.04) among types Squamous cell 1.27 (0.82-1.97) Adenocarcinoma 1.06 0.81-1.40 5

The authors again note that there was no significant heterogeneity of reported effect among the centers. However, they do point out that eight of the centers reported OR's greater than 1.00, which presumably implies that four centers reported OR's equal to or less than 1.00. Unlike reported spousal exposure, no pattern was reported for age at interview. , , No data are provided for the effects of those potential confounders investigated. It was merely stated that educational level, occupational exposures, urban vs. rural living, and intake of vegetables/fruits, etc. had no appreciable effect on the OR's. It should be noted that the effect of educational level was not reported in the multiple logistic regression for the spousal data. 4. Combined spousal and workplace exposure The OR for ever exposure either from the spouse or from the workplace was 1.14 (95% Cl, 0.88-1.47). This result was based on a total of 527 case subjects and 1201 control subjects. All of the results are presented in the table below. Sub rou anal sis OR (95% CI Comments Ever exposure to either spousal or 1.14 (0.88-1.47) 527 cases vs. 1201 controls work lace ETS Men 1.13 0,6g-1,gg 97 cases vs. 390 controls Women 1.15 0.86-1.55 420 cases vs. 811 controls Duration (in years) P for trend = 0.13 0 1.00 115 cases vs. 331 controls 1-36 1.11 (0.85-1.46) 362 cases vs. 876 controls 37-43 1.26 (0.87-1.81) 82 cases vs. 185 controls >_44 1.29 0.87-1.g2 70 cases vs. 125 controls Duration (hrs/day x years) P for trend = 0.01 0 1,00 122 cases vs. 339 controls 0-165 0.91 (0.69-1.20) 289 cases vs. 749 controls 166-253 1.31 (0.88-1.94) 63 cases vs. 151 controls ?254 1.46 0.96-2.22 57 cases vs. 101 controls Time since exposure 0 1.00 122 cases vs. 339 controls _16yrs 0.92 (0.67-1.26) 121 cases vs. 327 controls 3-15yrs 1.20 (0.89-1.62) 175 cases vs. 394 controls 0-2yrs 1.18 0,8g-1,5g 211 cases vs. 459 controls As before, the authors claim that there was no significant heterogeneity of effect among centers (P = 0.82). However, there is considerable variability. Results ranged from a point estimate of 0.72 in France to 2.29 in Sweden. Three centers had point estimates of less than 1.00, while 9 centers had point estimates greater than 1.00. The authors state that a "weak increase in lung cancer risk was present for increasing duration of exposure [in years]," and that the trend was stronger for duration measured in hours/ day x years. A dose-response was also claimed for duration of exposure with squamous cell and small cell carcinoma, but not adenocarcinoma, which represents -50% of the cases included in this study. 6

5. Exposure in vehicles and public indoor settings The results for variables representing reported exposure in vehicles and other public indoor settings are presented below. Sub rou ana! sis OR (95% CI Comments Overall vehicles 1.14 0.88-1.48 125 cases vs. 310 controls Overall public indoor settings 1.03 0.82-1.29 174 cases vs. 454 controls The OR's for these reported exposures were not consistent among centers. For exposure in vehicles the point estimates ranged from 0 to 2.85, while for exposure in other public indoor settings, the range of the point estimates was 0.24 to 2.32. The authors stated that analysis by duration of exposure suggested no consistent pattern for these two sources of exposure. D. Discussion The results reported in the IARC study may be interpreted by some to suggest that there is a weak, numerical association between reported spousal and workplace ETS exposure and lung cancer. The question, however, is whether any reported association is likely to be meaningful, or is simply a result of a combination of various types of errors and systematic biases. Some of the potential systematic biases that could have resulted in an elevated relative risk are discussed in some detail within the publication. However, as will be seen below, IARC's treatment of these biases may not be adequate. In evaluating the meaning of the reported numerical association, the authors devote a large part of their discussion to a comparison of their results with results published in other studies. However, often such comparisons are inappropriate, since these differences can best be explained by either random variation or clear systematic biases. Several of these points will be discussed at length below. 1. Systematic Biases a. Methodological Weaknesses As with any epidemiological study, there are methodological weaknesses in the IARC study as well. However, the authors of this publication discuss potential methodological weaknesses quite frankly, and they have made an effort to estimate how such weaknesses have contributed to the reported results. As was pointed out in the study design section above, some of the centers used hospital-based controls, whereas other centers selected community-based controls. The publication points out that there are possibilities for potential biases using either type of control. Hospital-based controls are more likely to suffer from selection bias; however, community-based controls are more likely to be subject to differential recall bias. Because of the fact that both methodologies were used, the authors compared results from subsets of centers defined according to their criteria for selection of control subjects. The OR for ever spousal or workplace exposure was 1.12 (95% Cl, 0.75-1.66) for centers with hospital-based controls and 1.13 (95% Cl, 0.80-1.61) for centers with community-based controls. The authors judge this difference to be small, and this review concurs with that judgment. However, the fact that there is agreement 7

between the two methods of control selection does not suggest that no bias was present, since each method is possibly subject to a different type of bias. There were large differences in response rates among the centers, particularly with controls, with a number of centers having quite low response rates. Both the differences among study centers and the low response rate can result in selection bias. In order to determine if there may'I'iave been selection bias resulting from low response rates for some of the centers the authors carried out the following calculation. For each exposure index stated (childhood, workplace, and spousal) two regressions were performed - one of the log OR against non-response rates in cases, the other of the log OR against non-response rates in controls. Had non-response been a problem, it is likely that one would have seen a relationship between OR and non-response. However, no relationship was found suggesting that non-response was not a problem. Although simple, this analysis would initially seem a reasonable way of detecting major bias. However, non-response may be correlated with other features of the study, e.g., region, and such simple correlations may be misleading. A preferred method would have been to have included response rate in cases and controls in a multivariate analysis which looked at various features differing by center that may have affected the study-specific OR estimate. Cytologic or histologic verification of lung cancer was not required as a criterion for inclusion of cases in the study. The authors restricted their analysis to histologically verified cases resulting in an OR for either spousal or workplace exposure of 1.11 (95% Cl, 0.86-1.43). The difference between this OR and the OR for the total sample was also designated by the authors as being "minor." However, this correction represents a 21 % decrease in excess risk. Such a difference cannot be regarded as minor. Moreover, the fact that the confidence interval was slightly more narrow than the confidence interval for the total sample suggests that the OR for those cases that lacked histological confirmation was both noticeably higher and contained much greater random fluctuation than the OR for the confirmed cases. Consequently, it would appear that restricting the analysis to those cases which were histologically confirmed would yield a better estimate of the true case population, and that an OR of 1.11 is a more reasonable estimate than is an OR of 1.14. There was an inconsistency among study centers with regard to matching or not matching of controls. As a consequence, either all of the data could have been treated using unconditional multivariate regression, or conditional multivariate regression could have been used for centers with matched controls and unconditional multivariate regression for those centers without matched controls. Technically, the mixed conditional/unconditional methodology is regarded as being somewhat more accurate. IARC tested both methods, stating that "the results were similar for most of the variables analyzed (Fig. 1)." As a consequence of this similarity, only the unconditional OR's were reported. Again, as was the case for the difference between the OR's for cases with cytologic confirmation and all cases, what is described as minor differences are far from minor. The difference between the two procedures for reported spousal exposure is a decrease of only 0.01. However, the difference for either reported spousal or workplace exposure is 0.03, a decrease of 21 % in excess risk, and the difference for reported workplace exposure is 0.07, a decrease of 41 % in excess risk. 8

b. Misclassification of Smoking Status The authors devote a considerable portion of their discussion to the issue of misclassification of smoking status, which they agree is "a potential source of bias in studies of lung cancer and ETS." Their conclusion is that misclassification has a very small effect on the OR's, and they propose three explanations to support this conclusion. ' ' The first explanation was based on their analysis of the data with exclusion of those individuals whom they classified as occasional smokers. As noted above, the criterion with respect to smoking for inclusion in the study was smokers of less than 400 cigarettes lifetime. The authors make the assertion that misclassified smokers are more likely to be found in this group of acknowledged occasional smokers, and that when this group was excluded, there was no change in OR. The OR restricted to self- reported never smokers for spousal exposure was 1.15 (95% Cl, 0.86-1.54) compared to 1.16 (95% Cl, 0.93 - 1.44) when the entire data set was used. There is no question that these numbers are not substantially different. However, the assertion that misclassified smokers are more likely to be represented in self-reported occasional smokers (less than 400 cigarettes lifetime) than in self-reported never smokers is not adequately supported. The authors' second explanation uses an assumption that rests on very slim evidence. The publication summarizes the conclusions reached by Lee and Forey (Lee, P. N., and Forey, B. A., 1996) stating that: If there is no true risk related to ETS exposure, a relative risk of the magnitude of that found in our study (i.e. 1.15) can be obtained assuming a misclassifcation rate of 2% (14), a proportion of smoking spouses of the order of 30-50%, a proportion of smokers in the underlying population of 20-40%, a concordance ratio of 3, and a relative risk of smoking in the order of 10-20. The authors concede that all of these assumptions are reasonable with the exception of the last one. Instead they state that "[A] more reasonable relative risk of smoking of 2 [Nyberg, F. et al., 1997] would result in a relative risk due to misclassification of 1.01- 1.02." A misclassification rate of 2% is clearly a reasonable assumption, since in an IARC supported study by Riboli, et al., (Riboli, E., et al., 1990, reference 14 in the first IARC citation above), a misclassification rate of 1.9% was reported for a sample of 1,369 non-smoking women. Riboli, et al., used a cut-off point of 100 ng cotinine/mg urinary creatinine to arrive at their misclassification rate of 1.9% (26 of the 1,369). If 1.9% of this sample had cotinine levels of at least this value, one can evaluate whether the relative risk of smoking of 2 used by Boffetta et al. is truly reasonable. Levels of cotinine in active smokers average about 1000 ng cotinine /mg creatinine. Therefore, each of the self-reported never-smokers who were true smokers as suggested by their cotinine levels had levels of at least 10% that of active smokers (Riboli, et al., do not give the actual level of cotinine found in these 26 subjects). It is extraordinarily unlikely that all of these subjects had levels of cotinine of exactly 100 ng cotinine/mg creatinine. If we make the very conservative estimate that the levels ranged from 100 to 500, with an average of 250 ng cotinine/mg creatinine, the estimated relative risk of lung cancer for the misclassified never smokers would be between 2.25 and 4. This 9

would result in a correction for misclassification of 0.03. This correction, which has to be regarded as highly conservative, would not appear to be very different from the correction suggested by the authors; however, once again, what appears to be a small correction generates a reduction of almost 20% in the excess risk attributed to spousal smoking. The authors' justification for the value of'2 as the relative risk for misclassified never smokers can be found in a publication by Nyberg, et al. (Nyberg, F., et al., 1997, reference 24 in the second IARC citation above). Nyberg, et al., used a test and retest method for a Swedish twin cohort to identify misclassified smokers. They found a misclassification rate of 4.9% among men and 4.5% among women. They then calculated the relative risk for lung cancer for the group of misclassified smokers. This relative risk was 1.9 for men (there were 0 cases of lung cancer for women). However, this value of 1.9 was based on only two cases out of 285 subjects, and the 95% confidence limits were 0.4-9.1. Clearly, to determine a relative risk of 2 for misclassified never smokers based upon a single cohort study with such a small sample is hardly justifiable. It is also important to point out that the authors' rejection of Lee and Forey's (1996) assumption of a relative risk of misclassified smokers of between 10 and 20 as being much too high seems to rest on a misinterpretation of the Lee and Forey publication. There is clearly a gradation of misclassified smokers ranging from occasional or former smokers all the way to current regular smokers. In order to attempt to use a single set of numbers, one can either take a weighted estimate of misclassified smokers, or a weighted estimated of relative risk. Lee and Forey assume a weighted estimate of misclassified smokers as being 2%, although their total estimate is considerably higher than that. Consequently, when the IARC publication assumes a misclassification rate of 2% and then also adjusts the relative risk downwards to 2, the correction for the fact that there is a range of misclassified smokers may have been made twice. The IARC's publication's third explanation rests on a validation study that was carried out by Nyberg et al. (Nyberg, F., et al., 1998). This study, wherein next-of-kin were interviewed in order to confirm never-smoking status for 175 case subjects and 233 control subjects in the Swedish study group, found a misclassification rate of only 1.2%. However, the next-of-kin interview methodology is not particularly sensitive for the detection of misclassified smokers. The total sample of 408 individuals in Nyberg, et al., included 26 individuals who acknowledged that they were occasional or former smokers. Inspection of Table 2, page 175 of this publication, shows that 15 of these 26 individuals were classified by next-of-kin interviews as being never-smokers, that is smokers of less than 400 cigarettes lifetime. Therefore, a total of 57.7% of next-of-kin incorrectly classified self-reported former or occasional smokers as never-smokers. If next-of-kin interviews were so strikingly inaccurate in confirming self-reported smoking, it is not surprising that this technique would be too imprecise to detect more than a fraction of ever-smokers who did not acknowledge their smoking. c. Confounders The method used to treat confounders was unusual. Although information was collected on a number of potential confounding factors, data were presented for only 10