Document J3yJ249bkVjJDyv728xjQroRe

SSSCREmRE tIe ENoRoong RIEOR"G3/6,7 3M PWoFrOkSin(gppMb)emLeovrealsndinuCmurorneDnattaanQduaHliisttoyrAicsaslesHsummeantn (of the Mean Populations) and Range of `This report documents Battelle's evaluation of the statistical quality and defensibility of 12 data sources ofserum perfluorooctanesulfonate (PFOS) levels found in non- occupational human populations. These data were summarized in the 3M white paper, "Perfluorooctane Sulfonate: Current Summary of Human Sera, Health andToxicology Data" provided to the EPA in January, 1999. The purposeofthis initial collectionof human non-occupational serum samples by the 3M Medical Department was to determine the presence or absence of PFOS in human blood from current and historical samples. 3M requested a review of these data by Battelle to determine what additional inferences could be obtained by analyses ofthese data. Battelle statisticians felt that only the pooled serum from the 18 U.S. blood banks was. judged defensible fora statistical analysis. Ofthese data, the geometric mean PFOS blood level was 28.2 ppb (95% confidence interval 23.9-33 3). additional data and analyses, Regional differences could not be explained without 001722 WORKING MEMORANDUM DATA QUALIToYn ASSESSMENT Bruce Buxton, WarrebnyStrauss, Owen Chang BATTELLE September 22, 1998 `1T2hidsarteapsoorutrdcoecsupmreenstesntBeadttbeyll3e'Mssetvaaflfuwathioonstuoggdeastteoefdtthhaet stthaetsiestdicaatlaqsuoaulritcyesanmdigdhetfebnesiubsileidtytoof . peosptuilmaattieonbsl.ooAdclceovreldsionfg pteor3flMu,ortohoecsteadnaetsaulwfeorneatceol(lPeFcOteSd)tionhneolnp-aodcdcruepsastitohnraele hdiusmcaonvery qbuleosotdiofnrsompoasecodrptoortahtee3-bMasMeedd3icMalemDpelpoarytemeepnotp:ul(a1t)iIosn kPnFoOwSndneottecttoahblaeveinwionrdkievidduianl3hM.uman dfeltueocrtoacbhleemiincaglenmearnaulfpaocptuulraitnigonplpaonotsle(dbubtlokondowslaemdpgleeasblferoamboduitfffelruenotrogcehoegmriacpahlisc)a?l l(o2c)aItsioPnFsOS acoclrloescsttehdepUrinoirtetod tShteatienst?roadnudct(i3o)nIosfPPFFOOSS-dbeatsecetdabclheeimnichaulmsainntboltohoedmwahrikcethpwloaucelodfthahveegbeeneenral pionpfeurleantcieosn?couSludbsbeequdeenrtilvye,dBfartotemlltehewsaesdaatsak;etdhiwshreetphoertr aadndyrefusrstehserthsattatqiusetsitciaollny. deIfnetnhsiisbrleeportwe. (d1e)felnisstibilmepoirntfaenretnccheasr,ac(t2e)rjiusdtigcesewahcihchofd3aMta'sset1s2dmautsatshoauvreceisn oargdaeinrsttotshuepspeocrrtitsetraitai,staincdall(y3) statistically analyze the available data, where appropriate. Ofoulrlofwiinndginpgosinftrso:m this initial assessmentofstatistical data quality can be summarized in the sWoeurcseusggpersetse1n2tcerditbeyri3a wMhistcahff,denfoinseoustracteiswtaicsaljluyddgeed sfufeficndiaesnttaisoonbuarllclee1s2. crOiftterhiea, 1a2nddaotnaly > osunfefiscoiuernctef,orthseub1s9e9q8uepnotolsetadtissetricuaml aonamlysUisS.. blood banks (datasource A3), wasjudged * Tlehveel1o9f9288p.o2oplpebd swietrhuma b9l5o%odcboannfkiddeantcaeiinndtiercvaatledofa2n3a.t9iontoal33g.e3ompeptbr.icSimgeniafnicPaFntOrSegbiloonoadl differences are apparent but cannotbe explained without additional data and analysis. nc for Statist moIfnutsohtredsbeeer rmteoeqtmuairrekegemaersndtaittnsigsctdaiecnsalbilgeyndoasefsftoecnhiseaitbseltdeudwcyiotnahcnldiumsipimooprnltseamfnetrnoqtmuaaatnitsioetntaootfiftdvheateaas,dsaustmeavpectroialollnerscetuqinoudnierereflfmoyerintnt.gsEthaech `cmnoiungmsheitrdiewcraaenldtmtothodreecelonsdsiwifdhfeeirrc:ehnt(a1rt)eytpueessmepodofrteaosltmicamhkaaetnigsoetnastpiisrntoiPbcalFleOimSnsfrberelegonaocrdedsli.envgIenlPsFoouOvrSearstbshleoesopsdamseltnevteshleesvrewerha,ilwceh one dwdoeercmaloddgewrsia,dpe2h.i)cTsatparbtelineadls|tisrneunPmdmFsaOriSnizPbeFlsoOo1Sd2blsletvoaetolidsstlaiecmavleolnrsegfqovurairrvieaomrueisnotusssurbfeporgoipdouanltsaatadiceorfnoessnssiintbthihleeiUt.yU,S..Sa,,nadonradss(o3c)iates each with the three typesofestimation problems1listed above. Note in this table that the firs five . 001723 Table1. Statistica Requirements for Defasibe Quantitative Conclusions [en [remem EERE [nT= SD{mstham tlesaddidnooyfpe ecwoefulnomfniaaelsoiuisd.3apejeipctisaniiovoewnnaasnmdStshwiephnpedinrcsasdioSe a2sgoi6cnghm des.03 Tv = 5 riseastsoCtlihneitosirwh:chDenraIskin8Shuar sonGsnOgRoE,h v v L 7 CloompwonhtPgRoOsSaimen,Twohe,ndsacrnebc romeo 6 15sn rsE io roooFor R is yr | E SE EReem Et ||] Sedeamtoonofairino0.3t.2mo,pVnEd, och,aot)4743 aSsp led andor TESGimetc belrLi,wnb1pgchndarriaeonopasndpiasrorcendtoiinionrGaeodstdSongosreses adn soeniso 2g . 33 2 `Tequirements are more general in nature and apply equally to all three typesofestimation . pesrtoibmlaetmiso,nwphriolbeletmh.e last seven requirements apply more specifically to a different type of Asseosf 1s 2DamtaeSon urctes On May 12, 1998, 3Mstafffaxed to Batteatlalblee listing 12data sourcestheybelievemight be used retyp to ed estimate PFOS blood as shown in Figure 1. levels Subse in non-occupational human popula quently, 3M staff sent to Battelle tions. This table has electronic files with t bee he n PFOS blood data they had available, as well as associated written documentation on the data sources which they had. Based on this information, Battelle attempted, and was partially successful, to obtain through literature searches more extensive documentation on these 12 studies. However, in many cases no additional written documentation could be found. Given the informationsentto Battelle by 3M and other documentation Battelle obtained from the literature search, Table 2 summarizes our assessmentofhow well eachofthe 12 data sources `meets each ofthe 12 defensibility requirements listed in Table 1. In Table 2, a "/" indicates that the given data source seems to meet the associated defensibility criterion, while a "XTM indicates that the data sourcedoes notmeet the criterion. This assessment by Battelle is based solely on the written documentation we wereable to obtain, either from 3Morthe literature review. In a large majorityofcases, no written documentation was found andso wecannot judge whether the defensibility criteria are met by a data source, or not. In these cases, a "2" is i. listed in Table 2. In some cases,noted as "v2", documentation mayhavebeen found about the initial study which collected the how particular subsamples were blood samples, but selected tobe sent tnoo3dMo.cuAmlesnot,atinisoonmweascaasveasilwaebljeduedsgcerdibthiantg a given data source would be appropriate for addressing just one typeofestimation problem (i.c., temporal changes, spatial trends, or demographic trends), and so defensibility criteria for other problems would not be applicable. These casesaredenoted bya "~" in Table 2. Table3 lists more detailed notes about eachofthe data sources for those criteria where sufficient information `was found in the written documentation to warrant comment. - : 001725 `Summaryof Mean andRaAnlgoDfatPaFAOnSa(lpypzbe)diFLnoivgFeualbrs1eeiunaCruyrr-eAnptaend1H9i9s8toricalHumanPopulations. B A C21.orNeC omPao cocppoun teiblmosocamdrc roceriioesaloa msonisenarTvtOsesTys-- 1inhi E |-- o 230 3 PoSseodngrniunnLbaoabmror2ryaspateU.S oodbanks oePoe osmersse 233.sa1a0mm0ppeleesstoeechhswwniitehhs311p00t00oddooarnnsswith 810 10 os sani "5 poSiio 5. eTorn to SWuamrsrhoUnS.ciocgyeoniee, 14810.10posedsumswith 10donorsprsang nor no. 222 SSi ueednocsSraao armgupsCs,a11n99n c7517Sty, 1968-1971 5. NRTsoci calraton sais,1976 f1100 mdriindossetmseanepgilen ostsomes ihsroamber of 2 NNnDop.s-2e . ITpoke cubsampals,1n880 2: inasamo nan,nr roi, 1984 dp3oroloadptunratatimaesnpsyith kn mbarof Godsven 5 = ese a o5M.TraCnhdeasdpeelnesSsahini,r10nd85sedvicee,1,99433binidti ssetmoi r nno5o.. noansoe.. s Notdracad Hot Th we wastedfom fnsm Bata on 513138wihtchanginposible isa. 23 83J 0 Table 2. Overall AssessmentofDefensibility of12 Data Sources [nto nemrsmaee aT Ts Tor ToPe Eo oeTm TorTa] o[ m2rmer |L 1 o L[eoLoLm eD e [eelea s el] a]l [ [ ai) m e TCEDCELCLDEe m NEEE EEN Fmes eeeee|L D L D DD E E T Tem re LL T L D LLD L Table 3. Notes Regarding Defensibility of 12 Data Sources T e BEI mE mre e gernee espn t--o]e 7 Detalled documentation = Ths only documentation isinGeary Olsen's 6/15/38 fier 1._Ostalled documentation ~ The only documentation is inGesry Olsen's 6/15/98 later TE 1." Dotalod documentation - Th only documentation i in Geary Olsen's 8/15/98 Terr RE 8. Uniform sampling across spac~e The samples are geographically dispersed. 1." Dotalled documentation The only documentatioins in Geary Olsen's 8/15/98 fetter a se aet: 0017727 `Table 3. Notes Regarding Defensibility of 12 Data Sources(Continued) 7. 2. dDiDineescftleiausnldseeioddngtdaJorAcgeuMtmAep.notpVauoltlai.toi2no3n5-,MNRToF.hIeT@t,a1srFgdbeo.tcpu2o3mp,eunl1at9ce7ido6ni,n wbGuaetsasrmuyabnsias3me0pn6l'0si8g375f1.o55r7/S9wM8itTihsetnmeorut asdnodicuTnrmaemknaetrofsioeheJaorutrnal 3. sUcnrbeioansiendgssampling protocols ~ The sampling protocal selected cartsin charactrstcs, with mutiple 4. iAncfcourrmaatteioenxpalbaonuattotrhye ipnofoolremdatciaobnra~tiIonfnorsmaatmiopn issbakuntowthne rina sample population was recorded, but no ~ T." 2. iDDneocftliaunldeoidndgtdaJorAcgeMutmAep,notpVauotllai.toi2no3n--5,-MNRToF.hIeT8t,airsFgadebot.cp2uo3mg,eanl1ta9et7do6n,inwbGuaetsasrmuyebnOslaasmegpnei'dnsg366/f-1o587/S3wM8itTfhsetnmeortisdnnedecuTFmnaeasnetfafsoiheJarotur 3. dUinvboinasseed sampling screenings protocols ~ The sampling protocol selected certain characteristics, with multiple 4. iAncfcourrmaatteioenxpalbaonuattotrhye ipnofoolremdatciaoinbe~Isntfoonrsmaamtpiloen skbnotwhine origina sample population was recorded, but no 2o . d(DeCebafinilcnioetradtIitnnasgrtT gd,eitVselpao.speu8sl5oa,tiNoop.nri-or18Te,hseoSpethpaatrgge1e5a,lpoo1pr9ue9sl3at,ytoimboaunScthwseaucsbasanaacilmonartlhaienrgdesfidooenSratMs owfenLiansdiooscnuramgeenodte4d0-69awtiotnhaoult 43.. AiUncnfcbouirramasatetedioesnxapmalpbaloniuanttgotrhpyeroiitnnofdcioovrlimsdaut--ailoTnshae~mpsIlanemfopirlsmooatsiwpopnnewarbsoutto tbheeroerprienseantsaatmipvleeofpotphuelattairognetwpeospusleastairodned, but na iDnectlauldeidngdJoAcuMmAe,ntVaotli.oZn35,MRNoF.IT6,8isF.odbo.Rc2uF3m,eTn1t9e7dn6o,iindbGuetasruybOslasmepni'ns7g988/fo15r5/S3M8 iasrnortadnodc7umFenoesu ou-rs 2. dDiasfoisnseed target population The target poguleton was mon aged 35-67 with muse rake so heart 3. Uscnrbeieansiendgssampling protocols ~ The sampling protocol selected crtain characte, with multiple 4. iAncfcourrmaatteioenxpsbloaunattohrey ipnofoolremdatciaoinbr~atInofnorsmaamtipolneasbkontowthne okginal sample population was recorded, but no ~ 1. Detailed documentation --- The only documentation is in Geary Olsen's 8/15/98letter JtTuhhdeegreemiawsihsneitcmhopenlrcyltnuhosetiroeennsuoflurtgoinhmgowdurairttaotveaenrredaloslctauatmsiesstneitscaasltmlieyondnteoofffettnhhseiebsl1t,2udodFaytndaoetss.oiugIrnncaesnsodimssetachmaaptsleiisnnm(gocps.rt,otcMoacRsoFelsIsTt)o scdoaolmclpuelmceteesndt,taobt3uitoMnnfoiosrdaovccahuiemlmaeibncltaealtoiannoatnlhyissoitsrh.ieginIniaavltasihtleaubdltyehueonondreihrgoiwnwhalithcsehamsapunlbiesnnatgmioprrleitsnhegteowsfaubsblsopaoemrdpflsoiranmpgeldweatsoswsaesnd nGbiaialvsyeeznd,itnhtgihsMea'cthskeoddfaadtoaacaounbmdteanpitrnaeetsdieobntnyi,3ntgMhsetwraietliilsstaiccluaslroecbnoetnlcbyliualsstieoldn,s(.iifa,Pnoyso,yljsiutnsegtmiatfthiieccaadtlailtoyantrfoooorsisstgahtsitsuotdfiicteaoslol(y0o.w5).. to look Widely dfiofrfberleonotdstteumdpyoorbajle,cstpiavteisala,nodr tdaergmeotgproappuhliactitornesndasm) oisngnotthreesctuodmimese,nded `b ecau se ofthe Of allthe data sources provided by 3M, probably the most defensibleforstatistical analysis purposes is the data set of 1998 pooled serum fr6om U.S. blood banks (data source A3). Although . 001728 detailed written documentation is lacking, it does appear that the samples may have been selected in an unbiased manner, at least within a, particular blood bank. We are concemed with `possible non-response bias potentially introduced by the large number ofblood banks (approximately ha ofl thof se contacted) which declined to send 3M the samples they requested. c`Tohnecreemairengaltshoepdoetemnotgiraalpbhiiacsseos fretshueltdionngofrrsoofmtthheepsraombpalbelse lwahcikcohfcwoenrteroslenbtytoe3acMh.blHooowdevbearn,k `recognizing these potential serious data deficiencies which limit our `ability to make valid inferences, in the next section we have subjected the: resulting PFOS blood level data to a limited - statistical analysis. `StaAnatlysiisosfBtlooid BcankaDalta `1T7hegeroegcreanpthlyiccoalllleyctdeidspaenrdsedanRaeldyzCerdopsosoblleodobdlboaondkssahmapvleesbefernomcaonmibnidneepdetnodeesnttiMmianteneasnoattaiaonnadlly representative mean PFOS (ppb) level. However, a varietyof assumptions are necessary, including: 1) Epoapcuhlaptoioolnedhasdamanpleequisalrecphraensecnetoaftibveeionftghreep`rreesgeinotnealdipnoptuhleatsiaomnpl(ei.)c.., each memberofthe 2) Bach pooled the sample. sample has a known numberofdonors, and each donor is represented equally in 3) The sampling and analytical methods for `obtaining serum PFOS levels is the same across all samples. ~ A1tlsiso,untlhiekdealtyathhaatvtehtehseamfopllleoswionbgtaaidndeidtiboyna3l Mlimciotmaptlioentse:ly satisfy allofthe above assumptions. 1) The blood banks represent only those that donated bloo--d ie, cities where the blood bank(s) refused to participate hadnochaonf bceie ng representeind 3M's sample. Furthermore, it is pugeenonleiprklaeellwyphtohopautdloathnteaitosenamibnplloeeoasdchotobgtetaohigenreRadepdhfirCcormroespgsai.rotni.ciRpaantthebrl,otohdebseanskasmpalreesreaprreerseepnrteasteinvteaotftivheeof 2) wItaiss usenlkecntoewd.n how many donors are in each pooled sample, nor is it known how each donor same. 3) Most ofthese samples methods arethesame. came from Red The Minnesota Csraomspslbelsporoodbbaabnlkysd,osonoittihsalviekeslyigtnhiofsicesaantmlpyldiinffgerent methods. As allofthe analysis was performedby 3M, theanalyticalmethods shouldbethe Aeflftehcotsugmhodtehelswe adsatuasmedaytoneosttsiamtaitsefy allof the national gaesosmuemtprtiiconmseadentaPiFleOdSabloevveel,:the following random " In(PFOS;) = u +B; +5; where ' 001729 dIsianfm(fpPelFreOenSfcyre)o(rmoentphrteehseietnnhatBtslutrohaoeldnlaBotgaursrkca,all-eel)xopgb(tesrt)awniesseftnohretmhneeadtmiPeoFananOlSPgFeleoOvmSeeltlraesivsceolmcatieaahtneedPwiFtihOthSthleveilt,h Bp,ooilsetdhe tnhaetiionndailvimdeuaanl,saanmdplee;iPFtOhSe rmeesaidsuuarleermreonrtlsefatnduntehxepmlaeiannedPbFyOtShelmevoedlealf(sthceobdrlirofeofsdep-robennadcnieknabgnebdtlwoteoheden s Bblaonokd)B.usI,tiasnadstshuatmed; ftohaltloBw;sfaolnloorwmsaalndoisrtmriabudtiisotnriwbiutthiomnewainthzemreoaanndzevraoraianndcveariW ancei BlohoudiBn raiennk cTionadnbiflvieidd4eunapclreorivenigtdieeorsnvatglheseofmeoesrttitrmhiaectoemvseeraoabnltslaiwgneheoidmcehftrwroeimcrefmtefiaonnug.ndthtiosmfaoldl eoulttsoidtehetdhaet9s.5%Taanbdl9e9l%st the Table 4. Parameter Estimates for Blood Bank Data [totsrewenrosin | 29%[Coontntanen 2 | anon | masa] [oom [emma | ee || Table S. GReeogmieotnrsiWciMtehaGneometric Means Falling Outside Confidence Intervals for Overall fBreemmwm ore--T |5 T[cw oms ewasip iow | avo issiso avon] o o at F s e me | Fond Fomor s 5 a a y y----~ J Pt PPn oommy se A s 1 [ s s 5T T ep ps ys------ --| -- ---- ] ------ PPo P omo so or mMmot 1 [5 3Tje p e --p-- ---- -- 13[gee ------ oe meet 5 he ---- ----| 5 Mees 3 Tas we s s e] 7] ( [omC e o--mT ts eT sa sy e-- 3 7 | ~ n`PTahSesabicoodvneecerntethsreualt9tis5odn%e.mcooTnnhsfitisrdhaeintgcehetnihantteu1r2vamolofcfo1brn8esbgterliuooconrtadelbdmafeonaksntshheafdeaslgtleiiomnmaesttiergdinmciafmtieciaaonnntPgSFeoOseSselosegveeelas etthant the overall mean is aresultof thefact that there is significant variation between the geometric. 001730 " `famcetaonrsPtFhOatSccaonnceexnptlraaitniaonpsorotbisoenrovefdthaitsevaarcihabbilliotoyd. baFnokr.exIatmisplpeo,sstihbelebltohoatdtbhaenrkesarienmeaningful Minidnunsetarpioeslitsh,atMaNp,plLya3faMyeptrteo,duLcAtsa(nsducGhraeseacvairlpee,t mSaCnumfaacytubreerlso)c,atwehdiilnecSlaonsteapBraorxbiamriat,yCtoA, dOimfafherae,ncNesEianntdheBidlolninogrs,poMpuTlamtaioynbaet iesaoclhatbeldorodobmasnukcmhaeyxpboesuarneotshoeurrcpeoss.siHbloeweevxpelra,nation for the observed variability between blood banks. To investigate the reasons for these regional differences, additional data analysis with more `explanatory information would be required. Resathme1p8 Bllooid Bnangks hd3aetMlap,aestvhkaeelrdueaBataretettevhlaelreuinatcthieeorinrtsaooipnfitny2i0ionnpoPpfbFtOohrSembsotlaroteoisdbtielcetavlweelmseernwiitttohhefimnriegsneaoimgmprualmpihnaignctadhlmelsaoecxa1it8imobunlsmo,oPdIFnbOatShneklsceuvtreorlesnt `measuredindifferent pooled samples from the same geographical location. I`fmaodsdtibteinoenfailciianlfdoartmaatwioounlcdanbebdeecmoolglercatpehdifcrionmfotrhmeabtliooondobnantkhse,poBaotlteedllsearmpelceosmmwehnidcsh wtheartethe purnedveirosutsalnydstehnetftaoct3oMrs.inWfeluesnucgignegstthtehaPtFitOiSs mbolsotodimlpevoerltavanrtiaabitlhiitsy.poCionltliencttiinmgenteowbebtleorod . sfarmopmltehsew1i8thlooucatldpeompouglrataipohnisc, biuntfoirtmwaotuiolndwnootulhdelhpelepxqpulaanitnitfhyetshoeurvcaer(isa)boilfitthyaotnvearsihaobuillidtye.xpFeocrt `dbeustimgonrinegiemfpfoirctieannttlfyuttuhreemsaajmoprlienngviprroongmreanmtsa,l,wbeehnaeveidortaol,knnuotwrihtioonwalm,udcehmovgarraiapbhiilcit,y to expect, socioeconomic, etc. factors that affect tha variabisloithtayt wecan try to control these factors r(ee.s8a.,mpsltriantgiftyhbeyseth1e8selofcaacttiooross) itsotohbattatihnemyomraeyaccocnusrtaitteutseamapbliiansgedressaulmtpsl.iAngosfetchonedgceonnercaelrnU.wSi.th population. By taking more samples from thesc cities, we might be simply generating more precise but highly biased estimates. 9 001731 TOTAL P. 10