Document YjdQb4K3bYJKpj9NRmg4V6Y60

Du Pont Response to Points of Discrepancy Between Analyses of Dr. Schwartz and Du Pont on Relating Blood Pressure to Blood Lead Using NHANES II Data April 11, 1986 Charles G. Pfeifer E. I. du Pont de Nemours & Co., Inc. As you requested in your letter of April 4, 1986, I am providing explanations to the six points you raised concerning possible reasons for differences in the estimated BP-ln(BPb) coefficient between analyses performed by Du Pont and Dr. Schwartz. My reply is organized in the order of the points you raised. 1. A number of significant variables other than those listed in Table 2 of the Pirkle et al (1985) paper were considered in the analyses of white males aged 40-59 done by Schwartz et al (including family history of hypertension, taking hypertension medication, SES, height and coffee), whereas Du Pont did not use them. o Du Pont's analysis first reported in our May 9, 1985 comments to EPA and CASAC on Corrigenda and later reported in our March 7, 1986 comments to EPA and CASAC on Addendum focussed on a methodological flow in the papers of Harlan et al (1985) and Pirkle et al (1985). The identified flaw was the nonadjustment for site effects of the BP-ln(BPb) regression coefficient. Our analysis was intended to address the effect on both the magnitude and statistical significance of the coefficient when attempting to replicate the published age-race-sex groups and regression models. Recognizing that some differences in our analysis and those of Harlan et al and Pirkle et al would inevitably remain, we performed our analyses without site adjustment as well as with site adjustment to enable a fair comparison on the effect of site adjustment on the BP-ln(BPb) coefficient. The other terms noted in the heading were not incorporated not because we believe they are unimportant but rather because they were not considered in the.two published papers. Attention was restricted to the published papers because of EPA rules for citing in Criteria Document. o The analyses of Dr. Schwartz which included additional predictors such as family history of hypertension, coffee, etc, were a part of his memos of June 26, 1985 (Response to Du Pont Critique of Blood Pressure) and of October 2, 1985 (Resonse to Richard Royall's questions on the blood lead-blood pressure relationship in NHANES II). These memos were in response to Du Pont's critique. Dr. Schwartz has complicated the fair comparison between Du Pont9s results, the results in Harlan et al/Pirkle et al and his results. Remember it is the credibility of the published TEH 0412637 2- - results that Du Pont has questioned as a result of the identified methodological flaw. We support continued improvements in analysis to best assess the magnitude and significance of the blood pressure-blood lead relationship (see Objective 2 in cover letter). However, the first step was to get consensus agreement on the invalidity of the results in the published papers. 2. Schwartz and colleagues, in their published paper, used log transform of dietary variables, whereas Du Pont did not. o This observation is correct and is due to an unintentional oversight on my part. The footnote in Table 2 of Pirkle et al was missed. Dietary variables in the Harlan et al models apparently were not log transformed. o One would hardly expect this difference to account in any substantive way for the differences in the magnitude of the BP-ln(BPb) coefficients between the Du Pont and Schwartz analyses. If it did then very serious questions regarding both the data and appropriateness of the models used by both Du Pont and Schwartz would be raised. At best, use of nonlog transformed variables might result in some minor loss of explanatory power. Contrary to Dr. Schwartz' comment in his March 3 rpemo to you, all but Oleic acid maintained significance (p_.05) in our Systolic BP analysis when not transforming whereas none did for the Diastolic BP analysis. 3. Schwartz et al used weights in their analyses taking site into account, so that the results could be mare directly compared to "non-site" analyses using SURREGR, whereas Du Pont did not o While the details of the Du Pont analyses reported in our May 9, 1985 comments and March 7, 1986 comments are unweighted regressions, we noted (Section III.C.3, item 1 of May 9, 1985 comments) that we ran corresponding weighted regressions which provided very comparable results. Summaries from these analyses are provided in Tables 1 and 2 attached, along with the unweighted results previously reported. These results demonstrate the effect of weighting has little effect on either the magnitude or significance of the results. In fact, the one adjusted coefficient that was significant in the unweighted analysis (Males, 12-74) was not significant in the weighted analysis. o The complex survey design was not used in our analyses. The coefficient estimates would be the same as those obtained by the weighted analysis, shown in Tables 1 and 2. An effect of using the survey design is that it tends to produce a larger error term, thereby decreasing TEH 0412638 DUP050453692 -3- likelihood that a term is flagged as statistically significant. The NHANES II Statistical Review Panel, CDC and Du Pont have previously all acknowledged that incorporation of the survey design structure has little impact on the qualitative results based on their experiences with NHANES II analyses. o In fact, use of the complex survey design in analyses affected by within study temporal trends is an area of disagreement among statisticians. It may be argued that only when the design accounts far time-related representativeness (which NHANES II did not) is it valid to use the design structure in such analyses. 4. Schwartz argues that one must pick a reference PSU as an intercept when adjusting for site, rather than having no intercept, as in the Du Pont analyses. o This point is without basis. There will be rm effect on the blood lead coefficient whether a reference PSU is picked or an intercept suppressed as Du Pont did. We suppressed the intercept so that the site effect coefficients have the convenient interpretation of denoting the site mean blood pressure. Suppressing the intercept imposes a linear constraint on the model parameters. Using an intercept term with a "restrict" statement (as Dr. Schwartz claimed he did at the March 11-12 1986 CASAC meeting) also imposes a linear constraint but a different one than we imposed. 5. It is difficult to understand the Du Pont variable Age_YY, as it yields a root mean square age of 5.8. o Dr. Schwartz' comments here are perplexing and apparently reflect an inability to understand the "Descriptive Statistics" table (Appendix IV, first page, Du Pont's March 7, 1986 Comments). There is j t o coding error and these results are accurate. Age_YY variable does denote age in years centered about the overall age mean (see comment below). Dr. Schwartz equates root mean square of 5.8 with mean age of white males aged 40-59. This is incorrect. Root mean square is a standard deviation of the age data (whether age is centered about its mean or not). An age standard deviation of 5.8 is quite consistent with the expected value of 5.5 assuming ages are uniformly distributed over the white male age range, 40-59. o All of our analyses centered the predictor variables (except site terms) about their means which is clearly shown on page 18 of our March 7, 1986 comments. One may interpret Dr. Schwartz' comments on page 2 of his March 3, 1986 memo as implying centering may account for differences in BP-ln(BPb) coefficient values. Centering TEH 0412639 DUP050453693 -4- covariates as Du Pont did has no effect on the estimated coefficients such as BP-ln(BPbT7 It is accepted and widespread statistical practice to center predictor variables in order to remove nonessential correlation among predictor variables. This results in smaller standard errors and increased sensitivity to detect significant effects. Numerous statistical references can be provided to support use of this procedure. (As an example, if one were to plot age and age-squared data against each other an obvious increasing trend relationship would be seen. By entering these two variables as such in the regression equation the effects of age and age-squared would be highly confounded due to nonessential correlation. By centering the age data about its mean and then squaring the centered age data to form age-squared the nonessential correlation is removed. A plot of the centered age data against age-squared formed in this way will demonstrate no linear trend.) o Centering the predictor variables, in fact, is very important in the analyses including site effects because of the temporal trends inherent in NHANES II. If this were not done it would be expected that the sensitivity of the analysis to detect a BP-ln(BPb) relationship would be diminished. It is apparent from the results shown by Harlan et al in Table 2 that they standardized their predictor variables. Standardizing signifies that the predictor variables are centered about their mean and divided by their standard deviation. Division by a standard deviation enables a fair comparison among all coefficients in a given model. Standardization is proper and widely used and also has the benefits of centering but expresses the coefficients in terms of standard deviation units rather than the physical measurements. 6. Analyses performed by Schwartz et al, subsequent to the Pirkle et al (1985) paper, have used age group of all men >20 and variables not included in the Harlan et al analyses, whereas Du Pont used all men aged 12-74 for its analyses and did not include all variables found to be significant by Schwartz. This further complicates comparison of results. o Our response here is the same as in item 1. Du Pont's analyses attempted to mimic the published analyses and not deliberately introduce other considerations that might confuse the effect on the BP-ln(BPb) coefficient by adjusting for site. It was Dr. Schwartz in response to Du Pont's analyses who introduced these other considerations. The age group men 20-74 and the other variables alluded to in (1.) above were not considered in the published papers of Harlan et al and Pirkle et al. TEH 0412640 DUP050453694 -5- Summary of Items 1-6 The concern you expressed in your letter to me of April 4, 1986 about the inconsistency in the magnitude of the BP-ln(BPb) coefficients estimated by Dr. Schwartz and Du Pont is certainly warranted. Based on the foregoing remarks it may be due in part to differences in o explanatory variables considered o male age groups (but not for 40-59 group) o dietary variable metric o sample size for male 40-59 group (Du Pont excluded fingerprick BPb's also) o other potential reasons not apparent It is not due to o data centering o age_YY data coding errors o suppression of the intercept o weighting or lack thereof (see Tables 1 and 2) o use or nonuse of SURREGR Recommendations The question of the consistency of the coefficient values could be addressed very directly. Since Dr. Schwartz has not yet mimicked the published analyses while adjusting properly for site, I suggest he do this. Thus, I would like to see his results for all the age-race-sex groups, both adjusted and unadjusted for sites, considered in Tables 1 and 2 using the model terms reported in the published papers. Then, redo those analyses but with the additional covariates he added to his recent analyses that were not considered in the published papers. The totality of these analyses will provide some useful comparisons to help better assess the total picture. CGP 5.19 TEH 0412641 DUP050453695 TEH 0412642 NOTE; R sum m aries in SA.S s t a t i s t i c a l c o m p u tin g p a cka g e a re m is le a d in g when in t e r c e p t i s s u p p re s s e d in s it e - a d ju s t e d a n a lyse s \ 0) H u (0 8 B d CO d o rH CO CO 03 u tQ 03 Pi 0) 1-1 d d d CO CO CO <u CO }-i XJ cu d <u g a H U0 pi r-4 4H co UH o o y-s XcJo wCpaU d d I 0pLa| nn o a o CO **04 CU 8 8 a. * Cf> 00 O On co CO CM in LO Pv in On rs On <t co . m ON rs <t ON ON oo 00 ro CO oo ps CO CM ON ON CM l-H rs. CM O ON o XJ in P*> vO a O pH CM 0o00 e s 1-4 O u rs ON m ON in m ps u ps o 00 in O m in w co ON rH CM NO CO p H pH pH XJ CO H CM p -4 Mf 4H O ON <NI in p H o CM US \D CM 00 o m o rs st <u CO O CM co o O -H CM o . o 11 erf 1 i i i 1 1 ON vO m co pH 00 CM 00 d CO rs CM p H CO o rs cn H CO ON CO o CM p H rH CM i-H H m NO ON 00 CO o in 00 CO pH in CM ON pH 1 in rs m NO ON m ON o c un 00 CO m vO ON m ps 00 VO ON 00 sf vO st f- ON pH CM p H ps m ON co 00 m PS rs m p H IS St o a p H vO Ml* l*s CM 11l 00 ON On <T p H ON IS. vO r-- st is OO CO CO >t Ps in oo CO p H CM m oo o pH pH rs o CM On CO pH fS <t m o m CM CO rs o o p H 00 m St cs 00 1 1 pH ! 1I 1 111 ' ON NO m to p H 00 CM oo 00 o is p H CM p H CO o rs CO pH ON co ON CO CM p H pH CM p H p H CO CM rs CM H pH rs co pH in rs H 1 m fS m cu co u a> XJ rl u CO u u u w *0 HO 4J CU CO 4J HH MH <U M p4 m oo c m ON <j* CM O co ro CO ro CO ON <r in CM pH 00 o o o pH pH <r r-- ei O oo o n st vO CO rs c m m p H CM CM 00 ON St VO O' O St CM m vO CO CM CM CM <-* CM pH o oi i a m ON U1 -d- oo m st oo N oo o pH iO pH st in tpos CM pH pH pH r-- co c m oo m CM in ON pH pH CO CM CM o c m in CM CM CM 1^- CM CM CM CM p H p H p H <t co <r rs oo rs vO in CM CM CM o CM CM p H St pH ON pH CM CO vO St CM pH CM pH CO pH co CM On OO p H p H rs o\ St 00 CM pH pH St o CM o pH H pH ON in On CM VO CO pH CM Ht- rs VO CM CM CM co m CO V0 m st CO CM CM CM p H p H pH H 1 r- p H On rs n 00 m CS st CO CO co rs o in 00 co CO in rs o pH pH pH GO m 00 co o co p H NO rs CM rs rs pH r*p ON pH m CM 00 ^4 st p H p H CM CM CM pH pH i 1^ CO <f o pH rs m CO CM CM CM r-H pH CM CM p H CM CM CO 00 pH CM pH CM co CO CM in CO p H 00 CM oo WON -H CM p H CO o n- a IS (f| p| ON 00 ON CO o m CM pH pH CM p H pH ON in ro pH oo CM OO oo rs pH CM pH oo CO PH 00 CO CM pH p H CM fH H st rs m OuoCcdOu ouoCdu uCCodOU uodP- CO M CO CO a W 4 3OCH3al us 3W "V m00 VpHp * pdH aU> aqj p3udH2 ymm-s /S Ps .p~\ sHCr-gM/ 0C3O pSdHJ PnnS HCM CCUO pH wpCHM p<0H03 U03 pH sPst smy 0C3O pH s wm p*dC0HO3 up04) M /<IsSt 'Cp_MH/ ptC<d0oBHO3u >inn wpCMH pCdC0B0Hu3O3 yv<mf-_s|Nyt. pffC<dH6HlOu /mopsHo pdH X03J pXU<HLU CHU ys s<mol/t rC<HUO s .X(0HJ3 M COiCcwoO5--2oOOi5i Si HMC3r33 ypm0--0H, pdH X<Jy dd pudH ymm-s /<Isst* nsrs1f spCHMy CCUO rSHdJ ymms. pH sCMy CCOU dpH X wCHM pC<HOU s XCJU p H yMp-sis win dCO rH vm/ pC0HO3 *4paHJ) /s <pst SCrHM' C0O3 pd<aHu PimnS SpCHMH CO dpH C0Sx)i /S v<mfs1yt* 0CO) drH a sy*m0-H0y\ pdH X<Ju <u rH H PH h yvsion-1syt pC0HO3 X0pHJ) PFIEFE: 13 DUP050453696 N 31008.01 \ Comparison o f B P -ln(B P b) C o e ffic ie n ts and R egression Summaries S y s to lic Blood P ressure Ph * * XJ Li U L> w 'd .u CO 4H CM do CJ 4o *O--i *4* Cfr*n>O- 4 o <00r o pH CNOO vCCOMM CO 4 iCnM oo pH 0O0N 0CmO0 oCO oCO 4 CpCOHM rC*M r"H pH pH pH 1 in CM oph 4 ON CM CM or-- i--i m r- 4 o cn eo pH <d CM cn pH pH CM CM p H CM orH in 00 in CM CM o o o*4* 00 oo 4 O oOn CM ON ON in in cn cni*h pH rH i pH pH CCMM pH pH in in pH pH CO CM CM pH NO -4* CD pOHN CM 00 CD NO CCDD CD 4o 0 40* m mp H CCCMDD !O *N*. CoCMD CM ON *NO4ON* in CM CrD- 44CM** OONn r***. CD pH CM H pH oNO CD ON CM ON v0 NO NO n O oco CM CD CD ON CM CD pH pH CM CM p H CM co NO NO ON r-* ON vO CD *n 00 CO r-. 4* 00 pH *H CM pH ON CD CM pH p H p H -d- CD p H pH 1i *Q9J XJ CO m ppcHoH *9 <ID CD CM 5 o HU X ON dpi HI H pH CM *C09O) a> d CL m 00 ON C030L * CD CO H M Pd 1 1 <1* r>>. in n on H in r-p a o oON n- pH CM CM CO or- c d p H ON 00 ON eo c m pH CM pH p H I CD in m CM CD ON CD pH 00 in 0O o 4` CD m pH CM in pH fL p H CD 4* m O o oo m CD o 11 i l11 4 n* m m CD pH m r- ON r> p H CM CM co CD p H ON CO ON CD CM p H p H CM p H p H pH m CD CD 4 m CM m 4 CD m m in CD in 4 4 CD a i9 U w CD 4J md m d s 60 d CM O c9o o 0) in CD pH CD ON 4 00 4 4 CD pH CM 4J 00 ON 4** CM ON CD CM CM CM CM 1 4 CD 4 CO pH vO -t 4 in CO CM N co f- r*- r-. CD CM CM CM CM oo rv **" CD h S Q> Ud Li O Li Li $4 W no *9 CO d XJ 4H 3 <4H *r-j d *Sdd93 O CD in r-* O in CM CD CD m ON CM CD <* in ON CM CM ON pd ON pH p H p H pH CM p H CM 4* pH ON in On 4* rCD CM rH ON ON ON <f in CO CM pH pH CO cD CD CD m in i pH pH pH i-* CM O'st m pH CM CM CM CD in O 4 -it CO CD in oo CM CM o co pH p H p H p H CM p H CM o r- co m CD in m <r CM CM m r**. 4 CO CO co CM CD CD in in pH CD Qd6>0 CD o pH <C0L 60 d H CM XJ pH 3 CD CL P-* o M CD 00 ON co 4 PS CM p H pH CD CM p H 00 pH 4 r*^ co n* d- CM pH p H o CD CM pH d CM a XHJ 4* r- m in CD pH in r-*- On pH CM CM 00 c CD p H ON oo ON CD CM p H p H CM pH p H CD m m 4 r- m m cD pH m r- fN. p H CM CM 00 r- CD p H 00 CD PM p H p H CM p H pH CD m m CO a9 Ou o eo OOLa3t OOC9LiL Cu O3OLi 03 M CO CO w ftS &J ft5 Q W 6- as s s nn m WCpHO pH d X0J) pEddudHd mm 4n* pH CM P-N winl /4H PN imn /**N 4r*- 4 'pCw1HM' pCadHQ> 2: mm wpCHM pW<HD d CO p<HU Sd 4p0J)i r4- m pCaHO> W pQdHJ <D j*Mj C1M WpH p<dSHU rHl cm co p(HV pdgQU) Vin1-/ co p<Hu pdaQH> /*N icno pH pH d XJ <1> p0H) PuHn /-v m 1 v4-x CO p<HU Sd Xp0J)i w JM0Z CO PEeS3 2 WQ Hsd MtOr3 /H WmpH pH d XdJ pddH PLdS( n rs mm 4r-s pH PS /H sCMy m /H 4r- / s mm r-N 4 4I"CrH1M pCOHOJ d mm pH CM pC0HO5 s to pSdQH> -XPdJi S 4r*s win co padH paCdH>O S XOHJI i VpCHHM CO p0H d ad pH CM pdCHO fdasa-) sm1/ CO pdH d a0) p\ icno pH pH d XdJ pdH PHa m 01 4 CO rHd Sd udH Xi d XtoJ CA 4 Hd co pdi Li dgS dCO 3CO > I--I CMOS dd EHOZe3 CD WE&* &3 Ocuh DUP050453697 N 31008.02