Formaldehyde and Leukemia: Critical Evaluation of Epidemiological Studies. University of Cologne. Germany. Peter Morfeld

Transcription

1 : Critical Evaluation of Epidemiological Studies of Cologne Germany Peter Morfeld Institute for occupational Epidemiology and Risk Assessment (IERA) Institute for Occupational Medicine

2 IARC 2004 Madrid, 19 April 2012 Seite 2

3 IARC Workshop, June 2004 on IARC 2006, Vol 88, Chapter 5.2, p. 276 In summary, there is strong but not sufficient evidence for a causal association between leukemia and occupational exposure to formaldehyde". Increased risks for leukemia has consistently been observed in studies of professional workers and in two out of three of the most relevant studies of industrial workers. These findings fall slightly short of being fully persuasive because of some limitations in the findings from the cohorts of industrial and garment workers in the USA because they conflict with the non-positive findings from the British cohort of industrial workers. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 3

4 NCI study, up to 2004: Formaldehyde Exposure and Leukemia Hauptmann et al (2003) - NCI s formaldehyde industry cohort 10 plants, 25,619 workers, mortality follow-up through 1994, 65 cases (incl 30 deaths from myeloid leukemia); no trend with cumulative exposure and (weak with) duration of exposure, positive trends with average and peak exposure intensity in internal (>0 ppm) comparisons: leukemia peak: RR=2.0 at ppm and RR=2.5 at 4 ppm, relative to >0-2 ppm average: RR=1.5 at ppm and RR=1.7 at 1 ppm, relative to >0-0.5 ppm myeloid leukemia peak: RR=2.4 at ppm and RR=3.5 at 4 ppm, relative to >0-2 ppm average: RR=1.2 at ppm and RR=2.5 at 1 ppm, relative to >0-0.5 ppm Marsh and Youk (2004) - re-analysis external comparisons revealed that the elevated leukemia and myeloid leukemia RRs and associated trends reported by NCI for peak and average intensity occurred because null (or slight) to moderate mortality excesses were compared with statistically significant baseline category deficits in deaths. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 4

5 Other Studies on Formaldehyde Exposure and Leukemia up to 2004 Meta-Analysis (incl. NCI study) Collins and Lineker (2004) 8 studies, industrial workers: RR=0.9 (0.95-CI: ) 7 studies, embalmers: RR=1.6 (0.95-CI: ) 3 studies, pathologists/anatomists: RR=1.4 (0.95-CI: ) all 18 studies: RR=1.1 (0.95-CI: ) Studies by exposure levels of formaldehyde or duration of exposure Coggon et al (2003), UK - MRC study 6 plants, 14,014 workers, 31 cases; no positive trend: SMR=1.0 at 2 ppm, SMR=0.7 at > 2 ppm Pinkerton et al (2004), USA - NIOSH study 3 plants, 11,039 workers, 24 cases; no clear trend: SMR=1.0 at < 3 yrs, SMR=0.7 at 3-9 yrs, SMR=1.5 at > 10 yrs Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 5

6 After IARC 2004 Meta Analyses Madrid, 19 April 2012 Seite 6

7 After IARC 2004: Three Meta-Analyses: Bosetti et al (2008), Bachand et al (2010), Schwilk et al (2010) [superseded Zhang et al (2009)] New NCI epi study on funeral directors and embalmers: Hauptmann et al (2009) included in Bachand et al (2010) and Schwilk et al (2010) Updated NCI epi study on 10 industrial plants: Beane Freeman et al (2009) updating Hauptmann et al (2003) by a longer mortality follow-up through 2004 included in Schwilk et al (2010) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 7

8 Meta-Analyses after IARC 2004: Bosetti et al (2008) The overall RR was (professionals / industry workers): for all lymphatic and hematopoietic cancers 1.31 / 0.85, and 1.39 / 0.90 for leukemia. For lymphohematopoietic neoplasms there were modestly elevated risks in professionals, but not in industry workers. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 8

9 Meta-Analyses after IARC 2004: Bachand et al (2010) For leukemias, the summary relative risk (RR) was 1.05 (95% CI: 0.93, 1.20) for cohort studies, and the summary odds ratio (OR) was 0.99 (95% CI: 0.71, 1.37) for case-control studies. Based on cohort and case-control studies, no significant differences were seen by leukemia subtype, job type, publication period, or region. Previous meta-analyses showed elevated summary estimates for leukemia; however, these analyses included results from proportionate mortality studies and did not explore other factors that could influence or confound results. By limiting analyses to stronger case-control and cohort study designs, considering the effects of smoking, these meta-analyses provided little support for a causal relationship between formaldehyde exposure and leukemia. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 9

10 Meta-Analyses after IARC 2004: In conflict to the meta-analyses presented so far: Meta-Analysis: Zhang et al (2009) [and Zhang et al (2010)] Focusing on occupations known to have high formaldehyde exposure, summary relative risks (RRs) were elevated in 15 studies of leukemia: RR=1.54; 0.95-CI: studies of myeloid leukemia: RR=1.90; 0.95-CI, superseded by Meta-Analysis: Schwilk et al (2010) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 10

11 Meta-Analyses after IARC 2004: Schwilk et al (2010) this meta-analysis includes two studies not used in previous analyses (Zhang et al 2009, 2010) 1) case-control investigation nested in a cohort of US funeral directors and embalmers (Hauptmann et al 2009) replaced Wallrath and Fraumeni (1983, 1984) and Hayes (1990) 2) updated NCI s formaldehyde cohort study (Beane Freeman et al 2009) replaced Hauptmann et al (2003) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 11

12 Meta-Analyses after IARC 2004: Schwilk et al (2010) Meta-analysis focusing on high-exposure groups: increased risks of leukemia: RR = 1.53; 0.95-CI: 1.11 to 2.21; 14 studies increased risks of myeloid leukemia: RR = 2.47; 0.95-CI : 1.42 to 4.27; 4 studies Conclusion of authors These findings provide evidence of increased myeloid leukemia risk with exposure to formaldehyde. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 12

13 Meta-Analyses after IARC 2004: Two studies performed by almost the same group of scientists Leukemia Zhang et al 2009: 15 studies; RR = 1.54; 0.95-CI: 1.18 to 2.00 Schwilk et al 2010: 14 studies; RR = 1.53; 0.95-CI: 1.11 to 2.21 (Harrington and Shannon 1975 split into 2 sub-studies, 3 studies replaced by 1 update) Myeloid Leukemia Zhang et al 2009: 6 studies; RR = 1.90; 0.95-CI: 1.31 to 2.76 Schwilk et al 2010: 4 studies; RR = 2.47; 0.95-CI: 1.42 to 4.27 (4 studies replaced by 2 updates) Observation: updating of studies caused a relevant change in RR estimate for myeloid leukemia only Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 13

14 Meta-Analyses after IARC 2004: Bachand et al 2010 (B): RR=1 for leukemia and RR=1 for myeloid leukemia vs. Zhang et al 2009 / Schwilk et al 2010 (S): RR=1.5 for leukemia and RR=2 for myeloid leukemia What are the reasons for these substantial differences? The updated studies (Hauptmann et al 2009, Beane Freeman et al 2009) had only an impact on the RR for myeloid leukemia. The inclusion of updated studies in S cannot explain the differences between B and S in the leukemia RR estimate The general methodology of computing overall relative risk estimates is similar and cannot explain the differences Does the set of eligible studies differ? Do the eligible data extracted from studies differ? Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 14

15 Meta-Analyses after IARC 2004: Differences in study selection B vs. S Dell and Teta 1995 was excluded in B but used in S for leukemia with RR= 2.65, 0.95-CI: 1.15 to 5.24 However, Dell and Teta 1995, wrote on p. 382: Among the R&D workers, the number of observed deaths (eight) from leukemia exceeded the expected deaths (three) There did not appear to be obvious common projects or exposures, except in general terms, that is, numerous references to solvents such as benzene and toluene. Evaluation: Dell and Teta 1995 should be excluded (in favour of B) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 15

16 Meta-Analyses after IARC 2004: Differences in study selection B vs. S Government Reports (Matonoski 1991, Robinson et al. 1987) used in B for leukemia with RR=1.35, 0.95-CI: 0.92 to 1.92 and RR=0.59, 0.95-CI: 0.01 to 3.28 but both excluded in S Evaluation: both views appear to be acceptable Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 16

17 Meta-Analyses after IARC 2004: Differences in study selection B vs. S Marsh et al (2004) as alternative to NCI analyses (Beane Freeman et al 2009) of the NCI formaldehyde industry cohort additionally included in B for leukemia with RR= 0.79, 0.95-CI: 0.62 to 1.01 but not in S Evaluation: exclusion appears to be reasonable (in favour of S) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 17

18 Meta-Analyses after IARC 2004: Differences in study selection B vs. S Partanen et al (1993) used in B for leukemia with RR=1.40, 0.95-CI: 0.25 to 7.91 but excluded in S because: Partanen et al 1993 was excluded because exposure was assessed much differently in cases (personal interviews and company records) than in controls (only company records). Partanen et al (1993): This asymmetry in exposure data between cases and referents introduces a potential source of bias Evaluation: exclusion appears to be reasonable (in favour of S) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 18

19 Meta-Analyses after IARC 2004: Differences in study selection B vs. S Sensitivity analysis by Schwilk et al (2010) 14 leukemia studies (FE): RR = 1.53, 0.95-CI: studies, incl. the 4 studies additionally listed in Bachand et al (2010): RR = 1.45, 0.95-CI: Evaluation: Difference in overall RR not caused by differences in study selection! What caused the difference? Zhang et al (2009) and Schwilk et al (2010): Meta-analyses focusing on high-exposure groups Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 19

20 Meta-Analyses after IARC 2004: Differences in selecting data from studies B vs. S Example: Beane Freeman et al (2009), NCI formaldehyde industry cohort S: myeloid leukemia at peak > 4 ppm: RR = 1.78, 0.95-CI: 0.87 to 3.64 B: all leukemias among all exposed: RR = 1.02, 0.95-CI: 0.85 to 1.22 B (Beane Freeman, Pinkerton, Stroup): myel leuk among all exp: RR = 1.09, 0.95-CI: 0.84 to 1.40 Beane Freeman et al (2009): myeloid leukemia, all leukemias observed SMR (0.95-CI) unexposed 4, (0.25 to 1.74), 0.48 (0.23 to 1.01) exposed 44, (0.67 to 1.21), 1.02 (0.85 to 1.22) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 20

21 Meta-Analyses after IARC 2004: Main methodological difference between Zhang et al (2009) and Schwilk et al (2010) vs. Bachand et al (2010) and Bosetti et al (2008) Zhang et al (2009) and Schwilk et al (2010) selected the highest exposed subgroups from the eligible studies Comment (Peter Morfeld): This is unusual in meta-analysis because highest cut points vary across studies metrics vary across studies (duration, peak exp, cum exposure, ) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 21

22 Meta-Analyses after IARC 2004: Results of Schwilk et al (2010) depend considerably on highest cut points chosen in each study even if the metric is the same Marsh and Youk 2004 (example): myeloid leukemia in NCI formaldehyde industry cohort average intensity SMR (US rates) NCI highest cat >1 ppm 1.45 UPitt highest cat > 0.74 ppm 1.02 NCI cats: 60th and 80th percentiles of all exposed cancer deaths Upitt cats: tertiles of exposed leukemia deaths Methodological comment and advice: Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Statistics in Medicine 2006; 25: Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 22

23 Meta-Analyses after IARC 2004: Main finding of Schwilk et al (2010) All 14 leukemia studies (FE): RR = 1.53, 0.95-CI: Four myeloid leukemia studies (FE): RR = 2.47, 0.95-CI: Sensitivity analysis Schwilk et al (2010), Table 1, FE n = 18 studies, incl. the 4 studies additionally listed in Bachand et al (2010) RR = 1.45, 0.95-CI: (no substantial diff. to RR=1.53 from 14 studies) Re-calculation of this meta-analysis, FE (Peter Morfeld) n = 18 studies, incl. the 4 studies with data as listed in Bachand et al (2010) RR = 1.13, 0.95-CI: (remarkable diff. to RR=1.53 from 14 studies) Results of Schwilk et al (2010) depend considerably on the specific data extraction process Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 23

24 Meta-Analyses after IARC 2004: Zhang et al (2009) and Schwilk et al (2010) combined RRs of highest categories across studies Comment (Peter Morfeld): this approach depends considerably on definition of metric depends considerably on definition of highest cat does not use the full information available Methodological comment and advice: Greenland and Longnecker: Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992;135: Orsini et al: Meta-Analysis for Linear and Nonlinear Dose-Response Relations: Examples, an Evaluation of Approximations, and Software. Am J Epidemiol 2012;175:66 73 Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 24

25 Meta-Analyses after IARC 2004: Peter Morfeld: Data combined are heterogeneous (no statistical test necessary to be convinced) However, Schwilk et al (2010) performed tests of heterogeneity (Table 2) they found large p-values, usually > 0.2 This appears to rule out heterogeneity Schwilk et al (2010) wrote in the discussion section: Nevertheless, the relatively low heterogeneity statistics (and the heterogeneity P > 0.05) highlight the overall consistency in these data. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 25

26 Meta-Analyses after IARC 2004: Peter Morfeld: Data combined are heterogeneous (no statistical test necessary to be convinced) This obvious heterogeneity cannot be ruled out by global p-values that are large; the tests as presented in Schwilk et al (2010) are insensitive Example given in Erren and Morfeld 2012: considerable influence on the calculated overall OR= 1.26 even though the global test for heterogeneity was not significant (p=0.3). Indeed, when measuring heterogeneity due to one study against all other 16 studies we calculated an I 2 =89.5% (p=0.002). When omitting this study we found a remarkably reduced overall OR. Warning in a leading textbook (Rothman et al. 2008, p. 671): large P-values do not indicate that it [heterogeneity] can be safely ignored Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 26

27 Meta-Analyses after IARC 2004: Calculation of tests and statistics (Peter Morfeld) Test(s) of heterogeneity for all 14 studies analyzed in Schwilk et al (2010): chi 2 df P I 2 Leuk % Myeloid % Overall % Random effects meta-analysis with standard confidence intervals is no solution Al Khalaf MM, Thalib L, Doi SAR. Combining heterogeneous studies using the random-effects model is a mistake and leads to inconclusive meta-analyses. J Clin Epidemiol 2011;64: Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 27

28 Meta-Analyses after IARC 2004: A possible way out Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ 2011;342:d549 predictive intervals should be calculated in random effects meta-analyses if findings are positive and heterogeneity cannot be excluded for sure Graham PL, Moran JL. Robust meta-analytic conclusions mandate the provision of prediction intervals in meta-analysis summaries. Journal of Clinical Epidemiology 2012; 65: Seventy-two meta-analyses from 70 articles were identified, containing between three and 80 studies each, with median nine studies. Substantial heterogeneity results in exceedingly wide PIs. Results imply that meta-analytic practitioners should be even more cautious about the conclusions of a meta-analysis than they have been previously. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 28

29 Meta-Analyses after IARC 2004: Peter Morfeld: Data from 14 studies analyzed by Schwilk et al (2010) L= leukemia M= myeloid leukemia Predictive intervals taking heterogeneity of selected data seriously: RR (0.95-CI) Overall (n=14): 1.58 ( ) Myeloid (n=4): 2.64 ( ) Madrid, 19 April 2012 Seite 29

30 Meta-Analyses after IARC 2004: The apparent conflict between published meta-analyses Bossetti et al (2008) and Bachand et al (2010) vs. Schwilk et al [Zhang et al (2009) and Zhang et al (2010)] can be resolved. The latter studies selected the highest exposed subgroups from the eligible studies (not recommended: Altman, Greenland, ) metrics vary across studies (duration, peak exp, cum exposure ) highest cut points vary across studies even if the metric is the same results are difficult to interpret (even if p is large: Al Khalaf, Rothman, ) no significant excess risk if heterogeneity is accounted for by predictive intervals (robust procedures necessary: Riley, Graham, ) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 30

31 IARC 2009 New Epi Studies Madrid, 19 April 2012 Seite 31

32 IARC s recent decision on s IARC F workshop - Baan et al (2009): The epidemiological evidence has become stronger: a recent study* found that embalming was significantly associated with an increased risk for myeloid leukemia, with significant trends for cumulative years of embalming (p trend =0.020) and for increasing peak formaldehyde exposure (p trend =0.036) The Working Group concluded that, overall, there is sufficient evidence for leukemia, particularly myeloid leukemia * new NCI study on embalmers: Hauptmann et al (2009) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 32

33 New NCI study on embalmers: Design Main hypothesis acc to Hauptmann et al (2003) and Bean Freeman et al (2009) Myeloid leukemia risk is linked to peak exposure in internal analysis Hauptmann et al (2009) Case-control study on professionals employed in the funeral industry 168 lymphohematopoetic deaths (incl 34 myeloid leukemias) [=cases of major interest] + 48 brain tumors 265 controls with death attributed to other causes (excluding respiratory and nervous system), matched to cases by data source, sex, dates of birth and death Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 33

34 New NCI study on embalmers: Exposure assessment Hauptmann et al (2009) Interviews with next of kin and co-workers on work practices, characteristics and tobacco use of study subjects Exposure assessment by a model that links rate of embalming, effect of ventilation, and concentration of formaldehyde solution to concentration in the air Qualitative interview data (guesses) transformed into quantitative data to apply the model [details in about 50% of subjects missing: no data to calculate peak, average or cumulative exposures for these subjects] Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 34

35 New NCI study on embalmers: Results Hauptmann et al (2009) Unconditional logistic regression adjusting for matching variables and smoking Results focus on myeloid leukemia and embalming Ever vs. never: OR = 11.2 (0.95-CI: 1.3 to 95.6) Positive trends observed with duration and number of embalmings, not with cumulative exposure but with average and peak formaldehyde exposure effect of duration of embalming and peak exposure were reported by IARC as important findings, Baan et al (2009) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 35

36 New NCI study on embalmers: Limitations of myeloid leukemia findings (1) Hauptmann et al (2009) No incidence density sampling of controls, deceased controls only [Peter Morfeld: weak design, text book: Rothman et al 2008] Table 2: Mean average and peak exposures are reported to be identical among myeloid leukemias and controls, just duration and number of embalmings and cumulative exposures are higher among cases [Peter Morfeld: this does not support the modelling results reported] Race distribution clearly different between myeloid leukemias and controls (see Table 1) but ignored in analyses and discussion [Peter Morfeld: blacks should be dropped from analyses, note that Beane Freeman et al 2009 adjusted for race] Calendar year of first employed clearly different between myeloid leukemias and controls but not adjusted for in analyses [Peter Morfeld: cases began work more often before WWII] Madrid, 19 April 2012 Seite 36

37 New NCI study on embalmers: Limitations of myeloid leukemia findings (2) Hauptmann et al (2009) Risks were attenuated when missing data were taken into account Trends no longer significant or positive in internal models the model type emphasized by Hauptmann et al (2003) and Beane Freeman et al (2009) Trend in peak exposure depend on only one case among controls the model type emphasized by Hauptmann et al (2003) and Beane Freeman et al (2009). The authors performed an analysis using a larger referent group: again no trend in internal analysis Peter Morfeld: Hauptmann et al (2009) clearly limited Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 37

38 Embalmers = very informative? Industrial Workers Have Much Higher Average Formaldehyde Exposures Job TWA Peak Foundry Source: IARC (1995) Plywood Manufacture Resins Operation Garment Manufacture Monomer Production Pathologists Embalmers Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 38

39 Additional New Data: Update of NCI cohort leukemia study Updating Hauptmann et al (2003) by a longer mortality follow-up through Bean Freeman et al (2009): new leukemia risk estimate is lower than before! Relative Risk 4 3,5 3 2,5 2 1,5 1 0,5 0 Hauptmann et al (2003) Beane Freeman et al (2009) 0 >0-< < Category Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 39

40 Additional New Data: Update on NCI cohort leukemia study An important problem detected in Hauptmann et al (2003) = NCI cohort study on leukemia with mortality follow-up until 31 Dec 1994: Beane Freeman et al (2009) Hauptmann et al (2003) missed 1006 deaths due to an invalid tracing of vital status. Surprisingly, one calculates 995 missing deaths by comparing Table 1 (Hauptmann et al. 2004) and Table S1 (Beane Freeman et al. 2009b). This difference with the published number of 1006 missing deaths creates doubt whether all errors were fixed by the NCI working group. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 40

41 Additional New Data: Revised # of deaths in NCI cohort Observed Deaths (Hauptmann et al., 2003) Revised Deaths (Beane-Freeman et al., 2009b) Change in Observed Deaths Percent Cange All Causes Unexposed % Exposed % Total % All Cancer Deaths Unexposed % Exposed % Total % All Solid Neoplasms Unexposed % Exposed % Total % All LHP Unexposed % Exposed % Total % Leukemia Unexposed % Exposed % Total % Source: Marsh et al (2010) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 41

42 NCI cohort study: differential missing of deaths 1. Differential missing among the unexposed The percent increase in revised deaths among unexposed" persons is twice that of the exposed" for all deaths, all cancer deaths, and all solid neoplasm deaths 2. Differential missing among the exposed? The increase of 7 leukemia deaths is only seen among the exposed. However Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 42

43 Leukemia deaths were missed in the lowest exposed category (>0 to 1.9 ppm) of the highest ever peak exposure Table 2. Correction of NCI 1994 Data and Additional Follow-up Attentuated Exposure-Response Relationship for Highest Ever Peak Formaldehyde Exposure and Leukemia. Highest Ever Peak Category NCI* Original 1994 Follow-up a,b NCI Revised analyses for 1994 Follow-up c NCI 2004 Follow-up d U Pitt Analysis of Original 1994 Follow-up e Deaths, RR* (95% CI*) Deaths, RR (95% CI) Deaths, RR (95% CI) SMR* (95% CI) Unexposed 4, 0.78 (0.25, 2.43) 4, 0.52 (0.17, 1.57) 7, 0.59 (0.25, 1.36) 0.38 (0.10, 0.97) > 0 to 1.9 ppm* (referent) 16, , , (0.28, 0.81) 2.0 to 3.9 ppm 20, 2.04 (1.04, 4.01) 20, 1.36 (0.73, 2.51) 27, 0.98 (0.60, 1.62) 1.04 (0.63, 1.60) 4.0 ppm 29, 2.46 (1.31, 4.62) 29, 1.60 (0.90, 2.82) 48, 1.42 (0.92, 2.18) 1.31 (0.88, 1.89) Trend Test (p-value) All groups Exposed only * NCI, National Cancer Institute, RR, relative risk; CI, confidence interval; SMR, standardized mortality ratio; ppm, partsper-million. a Hauptmann et al., 2003 b Used by IARC in 2004 reclassification c Beane Freeman et al., 2009b d Beane Freeman et al., 2009a e Marsh and Youk, 2004 Source: Marsh et al (2010) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 43

44 NCI cohort study: open questions 1. The percent increase in revised deaths among unexposed" persons is twice that of the exposed" for all deaths, all cancer deaths, and all solid neoplasm deaths (however, no new risk estimates provided by NCI yet) 2. The increase of 7 leukemia deaths is only seen among the low exposed: corrected risk estimates are lower than the estimates published before, and still reflect a deficit of deaths in the baseline category (Marsh and Youk 2004) 3. The difference between 995 missing deaths according to the tables and the published number of 1006 missing deaths creates doubt whether all errors were fixed by the NCI working group. Described and discussed in: Marsh, Youk, Morfeld, Collins, Symons 2010: Incomplete follow-up in the National Cancer Institute's formaldehyde worker study and the impact on subsequent reanalyses and causal evaluations. Regulatory Toxicology and Pharmacology 2010; 58: Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 44

45 NCI cohort study: Hauptmann et al 2003 in perspective Figure. Exposure-Response Relationship for Highest Ever Peak Formaldehyde Exposure and Leukemia Based on Original, Corrected and Updated NCI Data and Reanalysis of Original NCI Data Relative Risk or SMR NCI-1994 Original NCI-1994 Revised NCI-2004 SMRs NCI-1994* * Marsh and Youk (2004). Based on incomplete data (Beane Freeman et al., 2009a,b) Used by IARC in 2004 Reclassification Source: Marsh et al (2010) Unexposed >0 - < Highest Ever Peak Formaldehyde Exposure (ppm) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 45

46 NCI cohort study: Is the reduction of relative risk a reflection of leukemia latency periods? Leukemia Patterns by Duration and Latency not Consistent with Causal Association: Source: Marsh and Youk (2004) Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 46

47 Summary and Conclusions Madrid, 19 April 2012 Seite 47

48 Summary: Epidemiology of Most important medical endpoints discussed nasopharyngeal cancers (NPCs) leukemia(s) Of major relevance: US National Cancer Institute (NCI) studies. Embalmers: Hauptmann et al. 2009, Industry: Beane Freeman et al 2009 on leukemias,. IARC decided (2004, 2009) that formaldehyde exposure is a human carcinogen causing NPCs and leukemia Critical discussions and re-evaluations of the NCI cohort study (Marsh and coworkers 2004, 2007), errors in former studies of NCI (Hauptmann et al 2003, 2004), no support by a large British study (Coggon et al 2003) and clear limitations of the embalmer case-control study (Hauptmann et al 2009) shed doubt on IARC s decision Recent meta-analyses (Bossetti et al 2007, Bachand et al 2010) lend support to this critical point of view. Schwilk et al (2010) is not convincing due to the specific data extraction method applied. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 48

49 Summary: Formaldehyde and Leukemia by ANSES Proposal for Harmonised Categorization and Labelling based on Regulation (EC) No 1272/2008 (CLP Regulation), Annex VI, Part 2 ANSES (on behalf of the French MSCA): CLH REPORT FOR FORMALDEHYDE, Sep 2011, p.170: Conclusion about Formaldehyde and leukemia Overall, some positive observations have emerged in industrial populations but meta-analyses generally show a discrepancy in the results between industrial and professional populations in which several studies indicate an increased risk of leukaemia and especially myeloid leukaemia. Therefore, it is considered that available data do not provide causal evidence for formaldehyde as the aetiological factor as a bias specific to professionals cannot be ruled out. Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 49

50

51 Broader Context: CoI = Conflict of Interest Boffetta P, McLaughlin JK, Veccia CL, Tarone RE, Lipworth L, Blot W.: False-positive results in cancer epidemiology: A Plea For Epistemiological Modesty, JNCI (2008) Letters to the Editor: Cogliano and Straif (2009) Hauptmann and Ronckers (2009) and reply of authors and reply of authors Issues: false positive findings: overstated degree of evidence conflict of interest of experts at IARC workshop: careerism, advancing own research results in discussions, increasing prestige and possibility of future funding One discussed example is formaldehyde epidemiology Please read the paper and the letter exchanges and make up your mind! Madrid, 19 April 2012 peter,morfeld@evonik.com Seite 51