KTL Sarcoma study

• Authors: Jouni T. TUOMISTO, Juha PEKKANEN, Hannu KIVIRANTA, Erkki TUKIAINEN, Terttu VARTIAINEN and Jouko TUOMISTO
• Reference: Int. J. Cancer: 108, 893–900 (2004) Chemosphere (2005) 60: 78: 854-869
• Urn: NBN:fi-fe200901121020
• Ethics: National Public Health Institute Ethics Committee
• Journal: Int J Cancer}}

Question

Because it is obvious that there is a great need for improved exposure assessment in studying cancer risk of dioxins, we decided to undertake the major effort of conducting a large case-control study on soft-tissue sarcoma and measure dioxin concentrations individually in both patients and controls. Because this can be done accurately only from very large blood samples or from fat samples taken during an operation, we studied STS patients coming to surgery because of their tumor and selected appendicitis patients as controls. In the general population, the exposure to dioxins is almost totally from dietary sources — in Finland mostly from fish — and it varies widely among the population. Because of the extremely long half-life of dioxins, measured levels of dioxin at the time of operation can be used to estimate the lifetime cumu- lative exposure accurately. There is a priori no simultaneous exposure to chlorophenols or phenoxy acid herbicides, which behave completely differently in the environment, have relatively short half-lives in humans and are excreted in a few days. This enables us to estimate the association of STS with clean dioxin exposure without concomitant exposure to the main chemical, in contrast to occupational studies.

Answer

There is simulated data available about the study. For details, see #Simulated data.

Main fish consumption and PCDD/F variables

Some plots about dioxin congeners.

What congener do you want to plot on X axis?: TCDF2378 TCDD2378 PF12378 PF23478 PD12378 HF123478 HF123678 HF234678 HF123789 HD123478 HD123678 HD123789 F1234678 F1234789 D1234678 OCDF OCDD toxsum WHOTEQ What congener do you want to plot on X axis?: TCDF2378 TCDD2378 PF12378 PF23478 PD12378 HF123478 HF123678 HF234678 HF123789 HD123478 HD123678 HD123789 F1234678 F1234789 D1234678 OCDF OCDD toxsum WHOTEQ + Show code- Hide code library(OpasnetUtils) a <- opbase.data("Op_en2721") oprint(head(a)) plot(a[a$Congener == first, "Result"], a[a$Congener == second, "Result"], xlab = first, ylab = second) plot(a[a$Congener == "OCDD", "Result"], a[a$Congener == "WHOTEQ", "Result"]) plot(a[a$Congener == "WHOTEQ", "Result"], a[a$Congener == "OCDD", "Result"]) plot(a[a$Congener == "OCDF", "Result"], a[a$Congener == "OCDD", "Result"])

Rationale

Methods

Study population

The majority of sarcoma patients in southern Finland are treated by the multidisciplinary sarcoma group of Helsinki University Central Hospital, with the remaining cases in the University Hospitals of Kuopio, Turku, or Tampere. All patients referred to these hospitals for operative treatment of STS between June 1997 (August 1996 in Helsinki) and December 1999 and more than 15 years of age were eligible as cases. The diagnoses were verified histologically for all except 7 patients. Sarcomas connected with known familial or genetic conditions, as well as sarcomas arising in visceral organs and bone, were excluded. Also other malignancies than STS, as well as nonmalignant tumors, were rejected. Some patients were operated twice during the study period; the second sample was not processed.

All patients who were operated due to an appendicitis diagnosis in a study hospital and who were more than 15 years of age were eligible as controls. They were collected from the same catchment area as the STS patients by dividing it into 15 areas (mainly according to former Finnish health care districts). One hospital performing appendectomy operations was recruited to the study from each area (in Helsinki, 2 hospitals). These were the university, central, or district hospitals of Helsinki, Hyvinka¨a¨, Ha¨meenlinna, Joensuu, Jyva¨skyla¨, Kotka, Kuopio, Lahti, Lappeenranta, Pori, Seina¨joki, Tampere, Turku and Vaasa, and the municipality hospitals in Espoo (Jorvi Hospital) and Helsinki (Maria Hospital). Informed consent was obtained from all patients in writing before the operation. The study was approved by the ethics committees of the National Public Health Institute and the hospitals involved.

The total number of patients recruited during the fieldwork was 972. One case was deleted due to missing address information, 1 case and 2 controls due to missing age information, and 3 cases and 11 controls since their fat samples were too small for dioxin analysis. As a result, we had 954 patients (148 cases and 806 controls) available for matching. The age range was 17.0 –91.1 years for cases and 15.0–88.7 years for controls. Based on National Cancer Registry data, we caught 70%, 9%, 17% and 26% of STS patients in Helsinki, Turku, Tampere and Kuopio University hospital regions, respectively, during the study period (calendar years 1997–1999). In Helsinki, all patients treated surgically with correct diagnosis were caught and agreed to participate; those not caught were either treated nonsurgically or misdiagnosed. Based on hospital discharge registry data, we estimate that about onefourth of appendicitis patients were caught in average during the most active collection period, but differences between hospitals were large.

The cases and controls were individually matched for area and age at the end of the fieldwork. This was done to ensure that there are enough controls from small areas and old age groups in the final data set, as it was not possible to analyze all recruited patients for dioxin. Area was defined based on the area of residence using the 15 areas described above. The age was determined at the day of operation. Maximum allowed difference in age between cases and controls was ± 3 years if case was < 38.0 years old, and ± 6 years if case was >= 38.0 years old. The control closest by age was matched to the case. Cases with fewer controls had a priority over cases with more controls. The number of controls per case was limited to 3. For 110 cases, 227 matching controls could be found in the pool. Thirty-nine cases had 1 control, 25 cases had 2 and 46 cases had 3 controls; for 38 cases, no control matching both age and area could be found.

Exposure assessment

From the matched 337 patients, concentrations of the 17 toxic polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/Fs) were measured from a subcutaneous fat sample obtained during an appendectomy or sarcoma operation. Measurements were done by gas chromatography-mass spectrometry30 at the Laboratory of Chemistry, which is an accredited testing laboratory (T077) for the analysis of dioxins in human samples (current standard: EN ISO/ IEC 17025) and has successfully participated in WHO/Euro intercalibrations. The concentrations were summed up after the value of each congener was multiplied by its relative toxic potency (toxic equivalency factor, TEF). The TEF values according to WHO31 were used, resulting in toxic equivalent concentrations (WHOTEq). Fat samples were analyzed during and after the collection period. Samples from STS patients were always analyzed in a batch containing also samples from appendicitis patients. All analytical work was performed blind so that the chemistry laboratory did not know the diagnosis of the patient. Quality assurance of analysis was performed with 2 separate means: 2 preformulated pools of human fat with different concentrations of dioxins [10.6 (n = 35) and 40.2 (n = 33) ng/kg (WHO-TEq in fat)] were always run with each lot of samples, and 36 individual fat samples with WHO-TEqs ranging from 6.9 to 116 ng/kg fat were analyzed as duplicates. The coefficients of variation for WHO-TEq in preformulated pools were 5.1% and 5.7%, respectively, and in duplicate analysis, 6.2%.

A detailed questionnaire about socioeconomic and lifestyle factors and chemical exposures was given to the patients in the hospital. If the patient was found not to have received the questionnaire in the hospital or if the patient did not return it, a new copy was sent to the patient’s home address. Of the matched subjects, 84 cases (76%) and 185 controls (81%) have also questionnaire information.

Detailed exposure assessment

The concentrations of the 17 toxic PCDD/F congeners and of the 36 PCB congeners were measured from fat of a subcutaneous tissue sample (0.3–1.5 g of fat) which was obtained during an appendectomy or sarcoma operation. The toxic equivalents (WHOPCDD/F-TEQ and WHOPCB-TEQ) were calculated with the sets of toxic equivalency factors (TEF), recommended by WHO in 1998 (Van den Berg et al., 1998).

Fat from tissue sample was extracted with toluene for 18–24 h using the Soxhlet apparatus. The fat content was determined gravimetrically after changing the solvent to hexane using nonane as a keeper. Fat sample was spiked with a set of 13C-labeled internal standards: sixteen 2,3,7,8-chlorinated PCDD/F congeners, three non-ortho PCBs (PCB 77, 126, 169), and nine other PCBs (PCB 30 [12C-labeled], 80, 101, 105, 138, 153, 156, 180, 194).

The sample was defatted in a silica gel column containing acidic and neutral layers of silica, and all analytes were eluted with dichloromethane (DCM):cyclohexane (c-hexane) (1:1). PCDD/Fs were separated from PCBs on activated carbon column (Carbopack C, 60/80 mesh) containing Celite (Merck 2693). The first fraction including PCBs was eluted with DCM:c-hexane (1:1) following a back elution of the second fraction (PCDD/Fs) with toluene. Eluents from both of the fractions were evaporated using nonane as a keeper and then fractions in n-hexane were further cleaned by passing them through an activated alumina column (Merck 1097). The PCDD/F fraction was eluted from the alumina column with 20% DCM in n-hexane and recovery standards (13C 1,2,3,4-TCDD and 13C 1,2,3,7,8,9-HxCDD) were added to the fraction before DCM and n-hexane were replaced by 10-15 μl of nonane. The PCB fraction was eluted from the alumina column with 2% DCM in n-hexane, and the fraction, after changing the eluent to n-hexane, was transferred to another activated carbon column (without Celite) in order to separate the non-ortho PCBs from other PCBs. DCM (50%) in n-hexane was used to elute other PCBs while non-ortho PCBs were back eluted with toluene. Recovery standards, PCB 159 for other PCBs and 13C PCB 60 for non-ortho PCBs were added prior to analysis; the solvent for other PCBs (DCM:n-hexane, 1:1) was replaced by 300 μl of n-hexane, for non-ortho PCBs toluene was replaced by 10–15 μl of nonane. The quantitation was performed by selective ion recording mode using a VG 70–250 SE (VG Analytical, UK) mass spectrometer (resolution 10,000) equipped with a HP 6890 gas chromatograph with a fused silica capillary column (DB-DIOXIN, 60 m, 0.25 mm, 0.15 μm). Two μl were injected into a split-splitless injector at 270 °C. The temperature programs for PCDD/Fs, non-ortho-PCBs, and other PCBs were:

• start, 140 °C (4 min), rate 20 °C min−1 to 180 °C (0 min), rate 2 °C min−1 to 270 °C (36 min);
• start, 140 °C (4 min), rate 20 °C min−1 to 200 °C (0 min), rate 10 °C min−1 to 270 °C (12 min);
• start, 60 °C (3 min), rate 20 °C min−1 to 200 °C (0 min), rate 4 °C min−1 to 270 °C (14 min); respectively.

Limits of quantitation (LOQ) for PCDD/Fs and non-ortho PCBs varied between 0.1–5 and 1–5 pg g−1 fat, respectively, and for other PCBs between 0.02 and 0.1 ng g−1 fat, depending on each individual congener. Recoveries for internal standards were more than 50% for all congeners. Concentrations were calculated with lower bound method in which the results of congeners with concentrations below the LOQ were designated as nil.

This code was used to upload the data to Opasnet Base:

+ Show code- Hide code library(OpasnetBaseUtils) library(reshape) a <- tidy(op_baseGetData("opasnet_base", "Op_en2721")) colnames(a)[colnames(a) == "Result"] <- "WHOTEQ" for(i in 1:ncol(a)) {a[, i] <- as.numeric(a[, i])} a <- a[!is.na(a$public_id), ] #colnames(a) #class(a) #class(a$Result) #a$Result #as.numeric(a$Result) #a$Result <- as.numeric(a$Result) a <- melt(a, id.vars = "public_id") colnames(a) <- c("Obs", "Congener", "Result") a[a$Congener == "WHOTEQ", ] plot(a[a$Congener == "OCDD", "Result"], a[a$Congener == "WHOTEQ", "Result"]) plot(a[a$Congener == "WHOTEQ", "Result"], a[a$Congener == "OCDD", "Result"]) plot(a[a$Congener == "OCDF", "Result"], a[a$Congener == "OCDD", "Result"]) a #op_baseWrite("opasnet_base", a, ident = "Op_en2721", unit = "pg /g_fat", who = "Jouni", acttype = 4)

Quality control and assurance

Fat samples were analyzed during and after the collection period 1997–1999. All analytical work was performed blind such that the chemistry laboratory knew only the code of the sample. The laboratory reagent and equipment blank samples were treated and analyzed with the same method as the actual samples, one blank for every eight to ten samples. Quality assurance of analysis was performed in two separate ways: (a) two preformulated pools of human fat with different concentrations of PCDD/Fs [10.6 (n = 35) and 40.2 (n = 33) pg g−1 (WHOPCDD/F-TEQ in fat)] and PCBs [4.72 and 24.2 pg g−1 (WHOPCB-TEQ), respectively] were always run with each lot of samples and (b) 36 individual fat samples with WHOPCDD/F-TEQs ranging from 6.9 to 116 pg g−1 and WHOPCB-TEQs from 4.6 to 95 pg g−1 were analyzed in duplicate. The coefficients of variation (CV) for WHOPCDD/F-TEQ in preformulated pools were 5.1% and 5.7%, respectively, and for WHOPCB-TEQ 12 and 9.0%, respectively. In duplicate analysis the CV was 6.2% for WHOPCDD/F-TEQ and 18% for WHOPCB-TEQ.

The laboratory has successfully participated in several international quality control studies for the analysis of PCDD/Fs, and PCBs. Matrices in these studies have included cow milk, human milk and human serum. (Yrjänheikki, 1991, Rymen, 1994, WHO, 1996 and Lindström et al., 2000). The laboratory of chemistry in the National Public Health Institute is an accredited testing laboratory (No T077) in Finland (EN ISO/IEC 17025). The scope of accreditation includes PCDD/Fs, non-ortho PCBs, and other PCBs from human tissue samples.

Statistical analyses

Conditional logistic regression analysis was performed with SAS PHREG procedure. Odds ratios were estimated for each quintile of WHO-TEq, the sum of the toxic congeners and the most relevant individual congeners, i.e., 2378-TCDD, 2378-TCDF, 12378-PeCDD, 23478-PeCDF and 123678-HxCDD (abbreviations: T, tetra; Pe, penta; Hx, hexa; Hp, hepta; O, octa; CDD, chlorinated dibenzo-p-dioxin; CDF, chlorinated dibenzofuran). In the other congener-specific analyses, exposures were treated as continuous variables and odds ratios were calculated for an increase of an interquartile range of the exposure.

All analyses were adjusted for sex. Several variables collected with the questionnaire were used as confounders in the analysis one by one. Nonbinary variables were analyzed as quartiles. Radiation therapy given to an STS patient was considered as diseaserelated and ignored in the analyses if the link to the disease was stated in the questionnaire or if the therapy had been given within 1 year before the operation. The analysis with the largest number of missing values was that with education years with 63 cases and 112 controls, but otherwise there were at least 70 cases and 125 controls in the analyses.

Fish consumption was studied in detail. Specific questions about the frequency of fish consumption were asked: 1 about total fish consumption, and 10 about specific types of fish or fish species. Four fish types contributed most to the total fish consumption. They were assumed to have high (Baltic herring, Baltic salmon) or low (predatory fish from lakes, rainbow trout) dioxin concentration based on previous results. The consumption frequencies (times per month) were calculated for high- and low-dioxin fish separately based on these 4 fish types. Exposure to the following chemicals was asked as a binary variable: solvents, solvent-based paints, formaldehyde, insecticides, fungicides/herbicides, wood preservatives, strong detergents, heavy metals, other chemicals.

Data

• The original data, in csv file
• KTL Sarcoma study statistical analyses.txt
• Log file about the statistical analyses: Part 1, Part2
• Compilation of the results of statistical analyses

The code below runs the main fish consumption and PCDD/F variables, but because this is personal-level data, you need a password to run it. However, you can see ready-made results [1].

For variable descriptions, see D↷

Password: + Show code- Hide code library(OpasnetUtils) objects.get("isqT7nvhd0ViUR7d") data <- objects.decode(etable, password) colnames(data) <- t(data[1, ]) data <- data[2:nrow(data), 2:ncol(data)] for(i in 1:ncol(data)) { data[[i]] <- as.numeric(as.character(data[[i]])) } colnames(data)[colnames(data) == "aluenro"] <- "BMI" # Poistetaan aluenro-sarake ja korvataan se BMI:llä. data$BMI <- data$Paino / (data$Pituus / 100)^2 # oprint(head(data)) cat("Data from P:\\huippuyksikko\\Tutkimus\\R16_sarkooma\\Data\\Panulle20031216\\Analyysi020712_typistetty.xls.", nrow(data), "observations.\n") oprint(cor(x = data, use = "pairwise.complete.obs", method = "pearson")) # Basic Scatterplot Matrix pairs(~ika+BMI+Kalaa+Silakkaa+PF23478+WHOTEQ, data = data, main="Simple Scatterplot Matrix")

Variable information

The variable information was originally documented in Log file about the statistical analyses: Part 1, but unfortunately mostly in Finnish.

Show details

Tähän on syytä kirjata myös muuttujaluettelo, koska sitä ei ole missään muualla kunnolla tehty. LEIKKAUS leikkauspvm, SASin oma formaatti (päivää jostakin kiintopisteestä?) IKA ikä vuosina leikkauspäivänä ALUENRO aluenro tutkimusalue (potilaan kotiosoitteen postinumeron perusteella) 1 - 2 Espoo 3 Helsinki 4 Hyvinkää 5 Hämeenlinna 6 Joensuu 7 Jyväskylä 8 Kotka 9 Kuopio 10 Lahti 11 Lappeenranta 12 Pori 13 Seinäjoki 14 Tampere 15 Turku 16 Vaasa PNRO postinumero K3 (K#=kyselylomakkeen kysymys nro #) STRATUM stratum eli tapauksen numero ilman S-etuliitettä STRATASE strataset eli tässä analyysissä 9 DGLUOKKA diagnoosiluokka tapauksilla DG_id DGnimi 1 MFH 2 Liposarcoma 3 Leiomyosarcoma 4 Angiosarcoma 5 Chondrosarcoma 6 Sarcoma synoviale 7 Sarcoma Ewing 8 Dermatofibrosarcoma 9 Sarcoma alia 10 Sarcoma NUD 11 Ei tietoa 12 Osteosarcoma extrasceletale 21 Lipoma 22 Tumor Desmoides 23 Myxoma 24 Muu benigni tuumori 25 Melanoma 26 Muu kuin tuumori 27 Tuplanäyte SP sukupuoli 1: mies, 2: nainen K6 KOULUV kouluvuodet K8 PITUUS pituus, cm K15 PAINO paino nyt, kg K16 KALAA kalansyönti K21 1 Harvemmin kuin kerran kuukaudessa tai en lainkaan 2 Kerran tai pari kuukaudessa 3 Kerran viikossa 4 Pari kertaa viikossa 5 Lähes joka päivä 6 Kerran päivässä tai useammin PETOKALA K22 (muuttujaan _YRI_ISI asti) 0 En lainkaan 1 Harvemmin kuin kerran kuukaudessa 2 Kerran tai pari kuukaudessa 3 Kerran viikossa 4 Pari kertaa viikossa 5 Lähes joka päivä MUIKKUA SIS_VESI KIRJOLOH SILAKKAA IT_MEREN MUUTA_IT PAKASTEK KALAS_IL VALTAMER _YRI_ISI SADE sädehoito K47 VASTATTU onko kysely palautettu? YES/NO SADEAIK aikaisempi sädehoito (yhdistelty kyselyn ja sairaalan tiedoista) TCDF2378 kongeneerispesifinen pitoisuus ng/kg (absoluuttiyksiköt rasvassa) TCDD2378 PF12378 PF23478 PD12378 HF123478 HF123678 HF234678 HF123789 HD123478 HD123678 HD123789 F1234678 F1234789 D1234678 OCDF OCDD TOXSUM edellä mainittujen 17 kongeneerin summa WHOTEQ edellä mainittujen 17 kongeneerin WHO-TEF-kertoimella painotettu summa TUPAK oletko koskaan tupakoinut 1=ei, 2=kyllä K30 TSTATUS tupakointi: 1= < 6 kk sitten, 2= > 6 kk sitten, 3=ei koskaan K30+K32 PACKYEAR askivuodet: poltetut vuodet K31*( eri valmisteiden summa kpl/pvä K33)/20 NULL, jos K30=2 mutta K32=NULL KEMIK monelleko kemikaaliryhmälle on altistunut? K42 Onko altistunut 1=en, 2=kyllä (K42 muuttujaan MUU2 asti) LIUOTTI liuottimet MAALIT liuotinpohjaiset maalit FORMALDE formaldehydi HYONTEIS hyönteismyrkyt KASVINSU kasvinsuojeluaineet KYLLASTE puunkyllästeet PESUAINE voimakkaat pesuaineet METALLIT raskasmetallit MUU1 muu MUU2 muu K42 kuvauksen perusteella 0=ei altistusta 4=pahin altistus HYONLUOK hyönteismyrkyt KASVLUOK kasvinsuojeluaineet KYLLLUOK puunkyllästeet ALKO kuinka usein alkoholia K35 1 päivittäin 2 muutaman kerran viikossa 3 noin kerran viikossa 4 pari kertaa kuukaudessa 5 noin kerran kuukaudessa 6 noin kerran parissa kuukaudessa 7 3-4 kertaa vuodessa 8 pari kertaa vuodessa 9 kerran vuodessa tai harvemmin 10 en koskaan ALKOUS K35 ilmoitettuna kertaa/viikko (7, 3, 1, 0.5, 0.2, 0.1, 0.05, 0.03, 0.01, 0) ALKOKULU alkoholin kulutus annosta/viikko. Lasketaan K36=pal ja K35 avulla alkokulu: IIf([pal] Is Null And [alko]>8,0,[alkous]*[alkomaar]) alkomaar K36 ilmoitettuna annosta/kerta: 0,1,2,3,5,8,12 annosta ASUKKAIT potilaan kotikunnan asukasluku (vuonna 1996? Pitäisikö tarkistaa mistä tilasto on?) bmi paino / pituus / pituus * 10000 mkala Nämä kolme kalamuuttujaa on saatu regressiomallilla sovittamalla luokittelu- mkaladio muuttujat suoraan regressiomalliin. Yksiköt ovat siis epämääräisiä. Ehkä pitäisi mkalamuu muuttaa ensin kalamuuttujat muotoon annosta/kk niin olisi mielekäs tulkinta? muu if muu1 = 1 or muu2 = 1 then muu = 1 (muuten muu=0) HUOM! Tämä muu on laskettu päin mäntyä: 1 tarkoittaa ei altistusta eli jos ei ole kahta altistusta, merkitään altistuneeksi ja jos ei ole tietoa tai on kaksi altistusta, merkitään altistumattomaksi. ÄLÄ KÄYTÄ SIIS TÄTÄ MUUTTUJAA MISSÄÄN! smoking 3-tstatus; 0=ei ikinä tupakoinut, 1= > 6 kk sitten, 2= < 6 kk sitten never smoking jaetaan 0->never=1, 1->former=1, 2->current=1 former current mkala = 1.180840432 + PETOKALA * 0.285437176 + MUIKKUA * 0.120796063 + SIS_VESI * 0.081192631 + KIRJOLOH * 0.432524410 + SILAKKAA * 0.201481292 + IT_MEREN * 0.186820358 + MUUTA_IT *-0.096115539 + PAKASTEK * 0.105839909 + KALAS_IL *-0.059836406 + VALTAMER * 0.095303307 + _YRI_ISI * 0.033074683; mkaladio = 1.180840432 + SILAKKAA * 0.201481292 + IT_MEREN * 0.186820358; mkalamuu = 1.180840432 + PETOKALA * 0.285437176 + MUIKKUA * 0.120796063 + SIS_VESI * 0.081192631 + KIRJOLOH * 0.432524410 + MUUTA_IT *-0.096115539 + PAKASTEK * 0.105839909 + KALAS_IL *-0.059836406 + VALTAMER * 0.095303307 + _YRI_ISI * 0.033074683;

Data management

Code to manage the data. It takes the original data files and merges them. Works only if files are available.

Show details

# Sarcoma epidemiological data # Data was obtained from this data file (saved 23.9.2002, version 12.7.2002) # It contains all data that was used in the publication but not e.g. all questionnaire data. # U:\arkisto_kuopio\huippuyksikko\Tutkimus\R16_sarkooma\Analyysit\Analyysi020712\Analyysi020712.xls library(OpasnetUtils) sarc <- read.table( "//cesium/yhteiset/YMAL/Projects/Silakan riskiarvio/Data/Salaiset/Sarkooma/Analyysi020712_copy.csv", sep = ",", header = TRUE ) colnames(sarc)[c(24, 27, 28, 30, 32)] <- c( "Sisävesikalaa", "Itämeren.lohta", "Muuta.Itämerestä", "Kalasäilykkeitä", "Äyriäisiä" ) ########################################### Questionnaire # The questions come from the questionnaire form # U:\arkisto_kuopio\huippuyksikko\Tutkimus\R16_sarkooma\KTL_sarcoma_study_questionnaire.odt # The questionnaire data comes from # U:\arkisto_kuopio\huippuyksikko\Tutkimus\R80_Sarkooma2\Data\Kyselyt.xls (Kyselyt.csv does not contain åäö) # Kyselyt.xls was saved to N:\YMAL\Projects\Silakan riskiarvio\Data\Salaiset\Kyselyt.csv ques <- read.table( "//cesium/yhteiset/YMAL/Projects/Silakan riskiarvio/Data/Salaiset/Kyselyt.csv", sep = ",", header = TRUE ) poista <- c( "Pituus", "Paino", "Kalaa", "Petokalaa", "Muikkua", "Sisävesikalaa", "Kirjolohta", "Silakkaa", "Itämeren.lohta", "Muuta.itämerestä", "Pakastekalaa", "Kalasäilykkeitä", "Valtamerikalaa", "Äyriäisiä", "Vastattu", "AlkoKuinka.usein", "Onko1", "Onko2", "Onko3", "Onko4", "Onko5", "Onko6", "Onko7", "Onko8", "Onko9", "Onko10" ) # The tests below shows that the questionnaire columns in sarc and ques are actually identical. Therefore # we remove them from ques and keep those in sarc, the data file that was used in the publication. #for(i in testaa) { # x <- dat[[paste(i, ".x", sep = "")]] # if(is.factor(x)) x <- as.numeric(x) # levels(x)[x] # y <- dat[[paste(i, ".y", sep = "")]] # if(is.factor(y)) y <- as.numeric(y) # levels(y)[y] # print(paste(i, sum( x != y, na.rm = TRUE))) #} #sum(as.numeric(dat$alko) != as.numeric(dat$AlkoKuinka.usein), na.rm = TRUE) #sum(as.numeric(dat$liuotti) != as.numeric(dat$Onko1), na.rm = TRUE) #sum(as.numeric(dat$maalit) != as.numeric(dat$Onko2), na.rm = TRUE) #sum(as.numeric(dat$formalde) != as.numeric(dat$Onko3), na.rm = TRUE) #sum(as.numeric(dat$hyonteis) != as.numeric(dat$Onko4), na.rm = TRUE) #sum(as.numeric(dat$Kasvinsu) != as.numeric(dat$Onko5), na.rm = TRUE) #sum(as.numeric(dat$Kyllate) != as.numeric(dat$Onko6), na.rm = TRUE) #sum(as.numeric(dat$Pesuaine) != as.numeric(dat$Onko7), na.rm = TRUE) #sum(as.numeric(dat$Metallit) != as.numeric(dat$Onko8), na.rm = TRUE) #sum(as.numeric(dat$Muu1) != as.numeric(dat$Onko9), na.rm = TRUE) #sum(as.numeric(dat$Muu2) != as.numeric(dat$Onko10), na.rm = TRUE) ques <- ques[!colnames(ques) %in% poista] dat <- merge(sarc, ques, by = "kysely_id", all.x = TRUE) ruoat <- list(c( "Harvemmin kuin kerran kuukaudessa tai en lainkaan", "Kerran tai pari kuukaudessa", "Kerran viikossa", "Pari kertaa viikossa", "Lähes joka päivä", "Kerran päivässä tai useammin" ), NA, TRUE ) kalat <- list( c( "En lainkaan", "Harvemmin kuin kerran kuukaudessa", "Kerran tai pari kuukaudessa", "Kerran viikossa", "Pari kertaa viikossa", "Lähes joka päivä" ), 0:5, TRUE ) yn <- list(c("No", "Yes"), NA, TRUE) locs <- list( aluenro = c( "-", "Espoo", "Helsinki", "Hyvinkää", "Hämeenlinna", "Joensuu", "Jyväskylä", "Kotka", "Kuopio", "Lahti", "Lappeenranta", "Pori", "Seinäjoki", "Tampere", "Turku", "Vaasa" ), dgluokka = list( c( "MFH", "Liposarcoma", "Leiomyosarcoma", "Angiosarcoma", "Chondrosarcoma", "Sarcoma synoviale", "Sarcoma Ewing", "Dermatofibrosarcoma", "Sarcoma alia", "Sarcoma NUD", "Ei tietoa", "Osteosarcoma extrasceletale", "Lipoma", "Tumor Desmoides", "Myxoma", "Muu benigni tuumori", "Melanoma", "Muu kuin tuumori", "Tuplanäyte" ), c(1:12, 21:27) ), sp = c("Male", "Female"), Kalaa = ruoat, Petokalaa = kalat, Muikkua = kalat, Sisävesikalaa = kalat, Kirjolohta = kalat, Silakkaa = kalat, Itämeren.lohta = kalat, Muuta.Itämerestä = kalat, Pakastekalaa = kalat, Kalasäilykkeitä = kalat, Valtamerikalaa = kalat, Äyriäisiä = kalat, sade = yn, tupak = yn, tstatus = c("< 6 mo ago", "> 6 mo ago", "Never"), liuotti = yn, maalit = yn, formalde = yn, hyonteis = yn, Kasvinsu = yn, Kyllaste = yn, Pesuaine = yn, Metallit = yn, Muu1 = yn, Muu2 = yn, Hyonluok = list(c("No", "Very mild", "Mild", "Moderate", "High"), 0:4, TRUE), Kasvluok = list(c("No", "Very mild", "Mild", "Moderate", "High"), 0:4, TRUE), Kyllluok = list(c("No", "Very mild", "Mild", "Moderate", "High"), 0:4, TRUE), alko = list(c( "en koskaan", "kerran vuodessa tai harvemmin", "pari kertaa vuodessa", "3-4 kertaa vuodessa", "noin kerran parissa kuukaudessa", "noin kerran kuukaudessa", "pari kertaa kuukaudessa", "noin kerran viikossa", "muutaman kerran viikossa", "päivittäin" ), 10:1, TRUE ), Koulutus = list(c( "Kansakoulu tai peruskoulu", "Keskikoulu", "Ammattikoulu tai vastaava", "Opistotutkinto ja/tai lukio", "Akateeminen tutkinto" ), NA, TRUE ), Työntekijäryhmä = c( "Ylempi toimihenkilö", "Alempi toimihenkilö", "Työntekijä", "Maanviljelijä", "Yrittäjä", "Opiskelija", "Eläkeläinen", "Kotirouva", "Työtön" ), Painonmuutos = list(c( "Olen laihtunut", "Painoni ei ole juuri muuttunut", "Olen lihonut ja laihtunut", "Olen lihonut" ), c(3, 2, 13, 1), TRUE ), Ruokavalio = c( "ei erityisruokavaliota", "kasvisruoka sekä maito- ja munatuotteet", "ainoastaan kasvisruokavalio", "gluteeniton", "maidoton (ei edes hyla-tuotteita)", "muu, tarkempi kuvaus" ), Leipää = ruoat, Puuroja = ruoat, Makaronia = ruoat, Muutaviljaa = ruoat, Viiliä = ruoat, Juustoja = ruoat, Rasvaisia.juustoja = ruoat, Jäätelöä = ruoat, Liharuokaa = ruoat, Mitä.maitoa = list(c( "en juo maitoa enkä piimää", "rasvatonta maitoa", "rasvatonta piimää tai kirnupiimää", "muuta piimää", "ykkösmaitoa", "kevytmaitoa", "täysmaitoa" ), c(7, 4, 5, 6, 3, 2, 1), TRUE ), # Tästä välistä puuttuu Mitä.piimää. Vai puuttuuko? Onko yhdistetty maitoon? Mitä.leivälle = list(c( "En mitään", "Kasvimargariinia", "Voi-kasvirasvaseosta", "Voita" ), NA, TRUE ), Paljonko.rasvaa = list(c( "En lainkaan", "Voinapilla voitelen kolme viipaletta tai enemmän", "Voinapilla voitelen 1-2 viipaletta", "Käytän enemmän kuin yhden voinapin viipaletta kohti" ), NA, TRUE ), Mitä.rasvaa = list(c( "Ei mitään rasvaa", "Kasviöljyä", "Kasvimargariinia", "Talousmargariinia", "Voi-kasviöljyseosta", "Voita" ), c(6, 1, 2, 3, 4, 5), TRUE ), Ruokamuutos = c( "Ei", "Kyllä" ), Vesilähde = c( "kunnan vesijohtovettä", "oman kaivon vettä", "kaupan pullotettua vettä", "muuta" ), Tupakointi = c( "En", "Kyllä" ), TupakSäännöllisyys = c( "en ole koskaan tupakoinut säännöllisesti", "olen tupakoinut säännöllisesti" ), Tupakviimeksi = list(c( "yli 10 vuotta sitten", "6 - 10 vuotta sitten", "1 - 5 vuotta sitten", "puoli vuotta - vuosi sitten", "1 kk - puoli vuotta sitten", "2 pv - 1 kk sitten", "eilen tai tänään" ), 7:1, TRUE ), Alkoholi = list(c( "en ole koskaan käyttänyt alkoholijuomia", "en, olen lopettanut alkoholin käytön kokonaan", "kyllä, harvemmin kuin kerran kuussa", "kyllä, vähintään kerran kuussa" ), 4:1, TRUE ), AlkoKuinka.paljon = list(c( "vähemmän kuin yhden annoksen", "1 annoksen", "2 annosta", "3 annosta", "4-5 annosta", "6-10 annosta", "yli 10 annosta" ), NA, TRUE ), Asuntotyyppi = c( "omakotitalossa", "rivitalossa", "kerrostalossa" ), Lämmitystyyppi = c( "kaukolämpö", "öljylämmitys", "sähkölämmitys", "puulämmitys", "muu" ), Paikkaus = c( "Ei ole", "Kyllä" ), PaikkaPoisto = yn, Montako.paikkaa = list(c( "ei yhtään", "1 - 2", "3 - 6", "7 - 15", "yli 15" ), NA, TRUE ), Sairaus = yn, Elinsiirto = yn, Tekonivel = yn, Muu = yn, Sädehoito = yn, Vierasesine = yn, AIDS = yn, Neurofibr = yn, vonHippel = yn ) for(i in names(locs)) { if(!is.list(locs[[i]])) locs[[i]] <- list(locs[[i]], NA) if(!is.numeric(locs[[i]][[2]])) locs[[i]][[2]] <- 1:length(locs[[i]][[1]]) if(length(locs[[i]]) < 3) locs[[i]][[3]] <- FALSE if(i == "Kalaa") {print(locs[[i]]); print(dat[[i]][1:100])} dat[[i]] <- factor(dat[[i]], levels = locs[[i]][[2]], labels = locs[[i]][[1]], ordered = locs[[i]][[3]]) if(i == "Kalaa") {print(locs[[i]]); print(dat[[i]][1:100])} } levels(dat$ikäAlin)[levels(dat$ikäAlin) == "20.5"] <- "20" dat$ikaA1 <- as.numeric(substr(dat$ikäAlin, 1, 2)) dat$ikaA2 <- as.character(dat$ikäAlin) dat$ikaA2 <- as.numeric(substr(dat$ikaA2, nchar(dat$ikaA2)-1, nchar(dat$ikaA2))) levels(dat$ikäYlin)[levels(dat$ikäYlin) == "20.5"] <- "20" levels(dat$ikäYlin)[levels(dat$ikäYlin) == "21.5"] <- "21" levels(dat$ikäYlin)[levels(dat$ikäYlin) == "28.5"] <- "28" levels(dat$ikäYlin)[levels(dat$ikäYlin) == "121"] <- "21" dat$ikaY1 <- as.numeric(substr(dat$ikäYlin, 1, 2)) dat$ikaY2 <- as.character(dat$ikäYlin) dat$ikaY2 <- as.numeric(gsub("[- ,]", "", substr(dat$ikaY2, nchar(dat$ikaY2)-2, nchar(dat$ikaY2)))) dat$Pnetto <- as.numeric(sapply(as.character(dat$Muutoskg), FUN = function(x) {strsplit(x, " ")[[1]][1]})) dat$Ppoikk <- -as.numeric(sapply(as.character(dat$Muutoskg), FUN = function(x) {strsplit(x, " ")[[1]][2]})) dat$Ppoikk[is.na(dat$Ppoikk)] <- 0 dat$Pnetto <- ifelse(dat$Painonmuutos %in% c("Olen laihtunut"), -dat$Pnetto, dat$Pnetto) dat$Pnetto[dat$Painonmuutos == "Painoni ei ole juuri muuttunut"] <- 0 dat$Pnetto <- dat$Pnetto + dat$Ppoikk # Then we'll input values for NA based on the averages of the respective subgroups dat$Pnetto[is.na(dat$Pnetto) & dat$Painonmuutos == "Olen lihonut"] <- 8 dat$Pnetto[is.na(dat$Pnetto) & dat$Painonmuutos == "Olen laihtunut"] <- -8 dat$Pnetto[is.na(dat$Pnetto) & dat$Painonmuutos == "Olen lihonut ja laihtunut"] <- 1 dat$Rintamaitoa1[is.na(dat$Rintamaitoa1)] <- 0 dat$Rintamaitoa2[is.na(dat$Rintamaitoa2)] <- 0 dat$Rintamaitoa3[is.na(dat$Rintamaitoa3)] <- 0 dat$Rintamaitoa4[is.na(dat$Rintamaitoa4)] <- 0 dat$Rintamaitoa5[is.na(dat$Rintamaitoa5)] <- 0 dat$Rintamaitoa6[is.na(dat$Rintamaitoa6)] <- 0 dat$Rintamaitoa7[is.na(dat$Rintamaitoa7)] <- 0 dat$Rintamaitoa8[is.na(dat$Rintamaitoa8)] <- 0 dat$Rintamaitoa <- dat$Rintamaitoa1 + dat$Rintamaitoa2 + dat$Rintamaitoa3 + dat$Rintamaitoa4 + dat$Rintamaitoa5 + dat$Rintamaitoa6 + dat$Rintamaitoa7 + dat$Rintamaitoa8 ### Graphs about WHOTEQ as a function of age and BMI. if(FALSE) { ggplot(dat) + geom_point(aes(x = ika, y = WHOTEQ, colour = Paino / (Pituus/100)^2)) + scale_colour_gradientn(colours = rainbow(3)) ggplot(dat) + geom_point(aes(x = ika, y = WHOTEQ, colour = Paino / (Pituus/100)^2>25)) ggplot(dat, aes(x = ika, y = WHOTEQ, colour = Paino / (Pituus/100)^2>25)) + geom_point() + geom_smooth() for(i in colnames(dat)[c(4, 6:8, 14:16, 18:35, 37:76, 82:84, 101:138, 144:152, 154:162)]) {#182, 184:199, 201:205, 238, 247, 267:268)]) { print(ggplot(dat, aes_string(x = i)) + geom_bar() + labs(title = i)) # par(ask = interactive()) # This makes R to wait for enter before continuing } } temp <- as.character(dat$leikkaus_dt) temp <- as.numeric(substr(temp, nchar(temp) - 1, nchar(temp))) temp <- temp - pmax( dat$Vuosi1, dat$Vuosi2, dat$Vuosi3, dat$Vuosi4, dat$Vuosi5, dat$Vuosi6, dat$Vuosi7, dat$Vuosi8, na.rm = TRUE ) temp[is.na(temp)] <- 0 dat$Syntymasta <- temp # Montako vuotta on viimeisen lapsen syntymästä leikkausvuoteen? dat$ikaA1[is.na(dat$ikaA1)] <- 20 dat$ikaA2[is.na(dat$ikaA2)] <- 20 dat$ikaY1 <- ifelse(is.na(dat$ikaY1), dat$ika, dat$ikaY1) dat$ikaY2 <- ifelse(is.na(dat$ikaY2), dat$ika, dat$ikaY2) dat <- within(dat, lihomisvauhti <- (kgYlin - kgAlin) / ((ikaY2 + ikaY2) / 2 - (ikaA1 + ikaA2) / 2)) ## From here on, dat contains also other information than that produced in the KTL sarcoma study. ffq <- opbase.data("Op_en2721", subset = "Portions per month") ffq$Result <- ffq$Result ffq$Obs <- NULL foods <- opbase.data("Op_en2721", subset = "Food energy and dioxin") # Add two columns (energy "e" and dioxin "d") for each food item in the ffq questionnaire dat$Energy <- 0 dat$Dioxin <- 0 for(i in unique(foods$Food)) { # Merge classified ffq questionnaire data with quantitative interpretation of ffq. cole <- paste("e", i, sep = "") colnames(ffq) <- c(i, cole) # Convert words to numbers of portions per month dat <- merge(dat, ffq, all.x = TRUE) dat[[cole]] <- ifelse(is.na(dat[[cole]]), 0, dat[[cole]] / 30) # Replace NA with 0, otherwise /mo -> /d. dat[[paste("d", i, sep = "")]] <- dat[[cole]] * as.numeric(as.character(foods[foods$Food == i & foods$Observation == "Dioxin", "Result"])) dat[[paste("m", i, sep = "")]] <- dat[[cole]] * as.numeric(as.character(foods[foods$Food == i & foods$Observation == "Mass", "Result"])) dat[[cole]] <- dat[[cole]] * as.numeric(as.character(foods[foods$Food == i & foods$Observation == "Energy", "Result"])) dat$Energy <- dat$Energy + dat[[cole]] dat$Dioxin <- dat$Dioxin + dat[[paste("d", i, sep = "")]] } objects.get("x89WlgJlDLA02vnL") # PCDD/F-data table1 pcdd <- table1[2:nrow(table1), c(2,3)] # Aika tiputetaan pois toistaiseksi tarpeettomana (2002, 2009 dataa) colnames(pcdd) <- c("PCDDF", "PCB") pcdd$PCDDF <- as.numeric(as.character(pcdd$PCDDF)) pcdd$PCB <- as.numeric(as.character(pcdd$PCB)) pcdd$TEQ <- pcdd$PCDDF + pcdd$PCB dat$Silteq <- dat$mSilakkaa * mean(pcdd$TEQ) dat$Kalteq <- dat$Silteq + dat$mKalaa * 1 # Estimated mean 1 pg/g fw

Assumptions

The following assumptions are used to interpret survey answers:

Assumptions for calculations(-)

Obs	Variable	Value	Unit	Result	Description	Kysymys suomeksi
1	Q23		dl /glass	2	Desilitres of milk or sourmilk per glass
2	Q24	1	times /a	0	Never
3	Q24	2	times /a	0.5 - 0.9	less than once a year
4	Q24	3	times /a	2 - 5	A few times a year
5	Q24	4	times /a	12 - 36	1 - 3 times per month
6	Q24	5	times /a	52	once a week
7	Q24	6	times /a	104 - 208	2 - 4 times per week
8	Q24	7	times /a	260 - 364	5 or more times per week

How much mass, energy, and dioxin does one portion contain? Data are guesswork of from Fineli.

Food energy and dioxin(g,kJ,pg/portion)

Obs	Food	Mass	Energy	Dioxin
1	Kalaa	100	600	7
2	Silakkaa	100	792	470
3	Petokalaa	100	301	25
4	Muikkua	100	750	28
5	Sisävesikalaa	100	668	23
6	Kirjolohta	100	1067	74
7	Itämeren.lohta	100	1067	770
8	Muuta.Itämerestä	100	668	15
9	Pakastekalaa	100	324	7
10	Kalasäilykkeitä	60	600	7
11	Valtamerikalaa	100	600	7
12	Äyriäisiä	60	200	7
13	Leipää	50	406	0.01
14	Puuroja	200	642	0.02
15	Makaronia	200	846	0.02
16	Muutaviljaa	150	600	0.02
17	Viiliä	200	334	0.008
18	Juustoja	40	300	0.012
19	Rasvaisia.juustoja	40	600	0.03
20	Jäätelöä	150	1200	0.03
21	Liharuokaa	150	1400	1.5
22	Maitoa	200	358	0.004
23	Piimää	200	358	0.004

Portions per month(portions/mo)

Obs	Answer	Interpretation
1	En lainkaan	0.003
2	Harvemmin kuin kerran kuukaudessa tai en lainkaan	0.1
3	Harvemmin kuin kerran kuukaudessa	0.5
4	Kerran tai pari kuukaudessa	1.5
5	Kerran viikossa	4
6	Pari kertaa viikossa	8
7	Lähes joka päivä	20
8	Kerran päivässä tai useammin	40

Analyses

Simulated data

This code was used to create a csv file that contains a simulated data from this study. When compared with the original data, the simulated data

• has the same number of observations,
• has the same range of values in each variable,
• has approximately the same correlation structure between all variables.

+ Show code- Hide code library(OpasnetUtils) library(MASS) library(mc2d) library(reshape2) library(ggplot2) objects.get("isqT7nvhd0ViUR7d") data <- objects.decode(etable, password) colnames(data) <- t(data[1, ]) data <- data[2:nrow(data), 2:ncol(data)] data2 <- data fun <- c(rep("normal", 5), rep("poisson", 12), rep("lognormal", 19)) params <- list() for(i in 1:ncol(data2)) { data2[[i]] <- as.numeric(as.character(data2[[i]])) if(i > 17) data2[[i]] <- ifelse(data2[[i]] == 0, 0.01, data2[[i]]) params[i] <- fitdistr(data2[[i]][!is.na(data2[[i]])], fun[i]) } simu <- data.frame(temp = rep(NA, 968)) for(i in 1:5) { simu[[i]] <- rnorm(968, params[[i]][1], params[[i]][2]) } for(i in 6:17) { simu[[i]] <- rpois(968, params[[i]]) } for(i in 18:36) { simu[[i]] <- rlnorm(968, params[[i]][1], params[[i]][2]) } simu[[3]] <- rbern(968, 0.5) + 1 colnames(simu) <- colnames(data) korre <- cor(x = data2, use = "pairwise.complete.obs", method = "spearman") simu <- as.data.frame(cornode(as.matrix(simu), target = korre)) korre2 <- cor(x = simu, use = "pairwise.complete.obs", method = "spearman") qplot(melt(korre)$value, melt(korre2)$value) for(i in 1:ncol(simu)) { simu[[i]] <- ifelse( simu[[i]] > max(data[[i]], na.rm = TRUE) | simu[[i]] < min(data[[i]], na.rm = TRUE), NA, simu[[i]] ) } for(i in 1:ncol(data2)) {print(paste( min(data2[[i]], na.rm = TRUE), max(data2[[i]], na.rm = TRUE), min(simu[[i]], na.rm = TRUE), max(simu[[i]], na.rm = TRUE) ))}

POPs and obesity

Dioxins and PCBs have been assosiated to type 2 diabetes. Do dioxins cause diabetes, or do diabetes decrease dioxin elimination, or does obesity increase diabetes and decrease dioxin elimination, or something else? We tried to make sense of this by looking at sarcoma study data.

+ Show code- Hide code library(ggplot2) dat <- re.ad.csv("V:/TUSO/Projects/POPit ja lihavuus/Sarkoomakyselydata/Copy of sarkooma_kysely_ja_dioksiinit_korjattu.csv") dat$Diet <- dat$Rasvaa.maitotuotteista + dat$k21liha + dat$k21kala * 8 hist(dat$Diet) dat$Diet3 <- cut(dat$Diet, 3) ggplot(dat, aes(x = ika, y = IntakePCDDFTEQ, colour = Diet3)) + geom_point() + geom_smooth() ggplot(dat, aes(x = ika, y = PCDDFWHO05TEQ, colour = Diet3)) + geom_point() + geom_smooth()

Self-reported chemical exposure

We looked at self-reported chemical exposure, especially pesticides and wood preservatives.

+ Show code- Hide code library(lme4) # Data from //helfs01.thl.fi/groups2/TUSO/Projects/POPit ja lihavuus/Dioksiinit vs sarkooma/Data.xlsx dat <- read.csv("V:/TUSO/Projects/Sarkooma/Analyysit/Data_2.12.2016.csv", encoding = "UTF-8") names(dat) dat$PCDDFWHO05TEQ <- dat$PCDDFWHO05TEQ / 20 # Scale to a nominal interquartile range (ca. 19.5 pg/g fat, depending on subgroup) #Pekan malli clogit-funktiolla library("survival") #exact estimation. Tuottaa saman tuloksen kuin Riikalla. # Several different models were run. All included Sex as a confounder. # Four pairs of models looked at each chemical risk separately (Analysis: Separate), # and dioxin risk in the respective population. # Four models looked at each chemical + dioxin in a combined model, # adjusting for each other (Analysis: Combined). # Finally, one model contained all three chemicals and dioxin in a single model, # naturally not containing the combined chemical exposure this time. models <- list() models[[1]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + Exposure.woodpreservatives + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.woodpr == 1 , ] ) models[[2]] <- clogit(Sarcoma.matched ~ Sex + Exposure.woodpreservatives + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.woodpr == 1 , ] ) models[[3]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.woodpr == 1 , ] ) models[[4]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + Exposure.fungicidesherbicides + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.funher == 1 , ] ) models[[5]] <- clogit(Sarcoma.matched ~ Sex + Exposure.fungicidesherbicides + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.funher == 1 , ] ) models[[6]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.funher == 1 , ] ) models[[7]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + Exposure.insecticides + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.insect == 1 , ] ) models[[8]] <- clogit(Sarcoma.matched ~ Sex + Exposure.insecticides + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.insect == 1 , ] ) models[[9]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.insect == 1 , ] ) models[[10]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + Exposure.any + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.any == 1 , ] ) models[[11]] <- clogit(Sarcoma.matched ~ Sex + Exposure.any + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.any == 1 , ] ) models[[12]] <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + strata(Sarcoma.matched.pair), method="exact", data = dat[dat$Inclusion.criteria.any == 1 , ] ) out <- data.frame() for(i in 1:length(models)) { out <- rbind( out, cbind( as.data.frame(summary(models[[i]])$coefficients), as.data.frame(summary(models[[i]])$conf.int) ) ) } out$Subpop <- rep(c( "Wood preservatives", "Fungicides, herbicides", "Insecticides", "Any of above"), each = 7 ) print(out, digits = 3) out <- out[c(2,3,5,7,9,10,12,14,16,17,19,21,23,24,26,28) , c(2, 8, 9, 5, 10)] out$Analysis <- rep(c("Combined","Separate"), each = 2, times = 4) out <- out[order(out$Subpop, rownames(out), out$Analysis) , ] print(out, digits = 3) #### Analysis where all chemicals are in a single model. mo <- clogit(Sarcoma.matched ~ Sex + PCDDFWHO05TEQ + Exposure.insecticides + Exposure.fungicidesherbicides + Exposure.woodpreservatives + strata(Sarcoma.matched.pair), method="exact", data = dat#[dat$Inclusion.criteria.any == 1 , ] ) summary(mo) # Fungicides/herbicides clearly elevate the risk and is statistically significant # Woodpreservatives also shows a high risk but is only marginally significant. # Insecticides and dioxin are not associated with higher risk. # Chemicals are moderately correlated with each other and somewhat with dioxin # as shown by the correlation table below. cor(dat[ dat$Inclusion.criteria.any == 1 , c("Exposure.any", "Exposure.fungicidesherbicides", "Exposure.insecticides", "Exposure.woodpreservatives", "PCDDFWHO05TEQ" )], use = "pairwise.complete.obs" ) # Is Baltic herring an independent risk factor for sarcoma? # Well, the risk is increased but clearly non-significant (OR 1.388, 95 % CI 0.8063 - 2.389) dat$Silakka <- as.numeric(dat$Silakkaa) > 3 table(dat$Silakkaa, dat$Silakka) fit <- clogit(Sarcoma.matched ~ Sex + Silakka + strata(Sarcoma.matched.pair), method="exact", data = dat ) summary(fit) # A question was raised why the PCDDWHO05TEQ estimates for wood-preservative group and any-exposure group were identical. # The reason can be seen from here: table(dat[c( "Inclusion.criteria.woodpr", "Inclusion.criteria.any", "Sarcoma.matched")], exclude = NULL ) # The two groups are practically identical with only two additional controls in the any-exposure group. # These two controls do not much change the PCDDFWHO05TEQ impact on sarcoma, and therefore the estimates are the same # with precision of three decimals. table(dat[c( "Inclusion.criteria.insect", "Inclusion.criteria.any", "Sarcoma.matched")], exclude = NULL ) # However, with insecticides, there are three controls and TWO CASES more in any-exposure group and that does change estimates. # With herbicides and fungicides, there are one control and two cases more, also enough to change estimates.

Correlation of dioxin and fish

How do individual dioxin congeners correlate with individual fish parametres in the questionnaire?

+ Show code- Hide code # This is code Op_en2721/ on page [[KTL Sarcoma study]] library(lme4) library(Hmisc) library(MASS) library(ggplot2) library(rjags) # Data from //helfs01.thl.fi/groups2/TUSO/Projects/POPit ja lihavuus/Dioksiinit vs sarkooma/Data.xlsx dat <- read.csv("V:/TUSO/Projects/Sarkooma/Analyysit/Data_2.12.2016.csv", encoding = "UTF-8") names(dat) kalat <- colnames(dat)[c(29, 27, 79:89)] # All: 27, 29, 79:89 dioksiinit <- colnames(dat)[c(310:326, 365)] # All: 310:368) dat[1:20,kalat] dat[1:20,dioksiinit] colnames(dat) unique(unlist(lapply(dat[kalat], FUN = levels))) inn <- c( "", "En lainkaan", "Harvemmin kuin kerran kuukaudessa tai en lainkaan", "Harvemmin kuin kerran kuukaudessa", "Kerran tai pari kuukaudessa", "Kerran viikossa", "Pari kertaa viikossa", "Lähes joka päivä", "Kerran päivässä tai useammin" ) # Datat peräisin V:\TUSO\Projects\POPit ja lihavuus\Excel-mallit\Concentration modeling.xlsx # paitsi doses on näppituntuma # Meals per week doses <- c(NA, 0, 0.1, 0.15, 0.3, 1, 2, 5, 9) # Congener half-life in years t1.2 <- c(7.2, 11.2, 9.8, 13.1, 5.1, 4.9, 6.7, 2.1, 3.5, 7.0, 6.4, 7.2, 7.2, 2.8, 3.1, 4.6, 1.4, 7) # WHO2005 TEF TEF <- c(1, 1, 0.1, 0.1, 0.1, 0.01, 0.0003, 0.1, 0.03, 0.3, 0.1, 0.1, 0.1, 0.1, 0.01, 0.01, 0.0003, 1) dat2 <- dat[c(kalat, dioksiinit, "Age")] # Convert fish intake answers to units meals/week dat2[kalat][-1] <- lapply(dat2[kalat][-1], FUN = function(x) doses[match(x, inn)]) # Convert dioxin concentrations to TEQs dat2[dioksiinit] <- lapply(as.list(1:length(TEF)), FUN = function(x) TEF[x] * dat2[dioksiinit][[x]]) cor(x = dat2[dioksiinit], y = dat2[kalat], use = "pairwise.complete.obs") dat3 <- dat2[!is.na(rowSums(dat2[c(kalat, dioksiinit)])) , ] dat3.diox <- resid(lm(cbind( PCDDFWHO05TEQ, X2378.TCDD, X12378.PD, X123478.HD, X123678.HD, X123789.HD, X1234678.D, OCDD, X2378.TCDF, X12378.PF, X23478.PF, X123478.HF, X123678.HF, X123789.HF, X234678.HF, X1234678.F, X1234789.F, OCDF ) ~ Age, dat3)) correlations <- round(cor(dat3.diox, dat3[kalat]), 3) pvalues <- round(rcorr(dat3.diox, as.matrix(dat3[kalat]))$P, 3)[1:18, -(1:18)] fit <- lm( paste("cbind(", paste(dioksiinit, collapse = ","), ") ~ ", paste(c(kalat, "Age"), collapse = " + "), collapse = ""), data = dat3 ) summary(fit) out <- data.frame() ################# Explain each congener with all fish variables + Age for(i in 1:length(dioksiinit)) { fit <- lm(paste(dioksiinit[[i]], "~", paste(kalat, collapse = " + "), "+ Age"), dat3) fit <- summary(stepAIC(fit, direction="both")) out <- rbind( out, data.frame( fit[[4]], Var = rownames(fit[[4]]), adj.r.squared = fit[[9]], Congener = dioksiinit[[i]], Halflife = t1.2[[i]], Test = "With age" ) ) } ############# Age is removed from the models to see the explanatory power of fish variables alone for(i in 1:length(dioksiinit)) { fit <- lm(paste(dioksiinit[[i]], "~", paste(kalat, collapse = " + ")), dat3) fit <- summary(stepAIC(fit, direction="both")) out <- rbind( out, data.frame( fit[[4]], Var = rownames(fit[[4]]), adj.r.squared = fit[[9]], Congener = dioksiinit[[i]], Halflife = t1.2[[i]], Test = "Without age" ) ) } ################# Explain each congener with SINGLE fish variables + Age for(i in 1:length(dioksiinit)) { fit <- lm(paste(dioksiinit[[i]], "~", paste(kalat[c(-1, -2)], collapse = " + "), "+ Age"), dat3) fit <- summary(stepAIC(fit, direction="both")) out <- rbind( out, data.frame( fit[[4]], Var = rownames(fit[[4]]), adj.r.squared = fit[[9]], Congener = dioksiinit[[i]], Halflife = t1.2[[i]], Test = "Without generic fish variables, with age" ) ) } oprint(out) write.csv(out, "V:/TUSO/Projects/Sarkooma/lineaariregressiot.csv") colnames(out) head(out) temp <- out[out$Var != "(Intercept)", ] ggplot(temp, aes(x = Var, y = Estimate, colour = Congener, size = temp$Pr...t. < 0.05)) + geom_point()+ theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust = 0.5))+ facet_wrap(~ Test) ################# Bayesian approach kal <- kalat[-1] datb <- dat[c(kal)]#, dioksiinit, "Age")] #x <- datb[[1]] datb[kal] <- lapply( datb[kal], FUN = function(x) { factor( x, levels = if(inn[3] %in% x) inn[c(3,5:9)] else inn[c(2,4:8)], ordered = TRUE ) } ) test <- as.data.frame(lapply(datb, FUN = is.na)) datb <- datb[rowSums(test) < 8 , ] datb <- as.data.frame(lapply(datb, FUN = function(x) as.numeric(x) -1)) #table(datb[1:2]) datblong <- melt(datb, measure.vars = 1:12) ggplot(datblong, aes(x = value, weight = 1))+geom_bar()+facet_wrap(~ variable) correlations <- cor(datb, use = "pairwise.complete.obs") melt(correlations, measure.vars = 1:12) pvalues <- round(rcorr(as.matrix(datb))$P, 3) mod <- textConnection(" model{ for(j in 1:12) { for(i in 1:N) { datbb[i , j] ~ dbin(p[j], 6) #Six alternatives in each question } p[j] ~ dunif(0,1) } } ") # A binomial distribution is assumed for bins of answer choices. jags <- jags.model( mod, data = list( N = length(datb), datbb = datb ), n.chains = 4, n.adapt = 100 ) update(jags, 1000) samps <- jags.samples(jags, 'p', 1000) samps.coda <- coda.samples(jags, 'p', 1000) plot(samps.coda[[1]]) head(samps.coda) summary(samps.coda) hist(cor(samps.coda[[1]])[cor(samps.coda[[1]]) != 1]) ### In practice, all correlations between -0.05 and 0.05 -> meaningless # Important corralations can be (and have been above) calculated directly from data. probs <- colSums(samps.coda[[1]]) / nrow(samps.coda[[1]]) gramm <- data.frame( Fish = kal, P = probs, Answer = rep(0:6, each = length(kal)), Freq = dbinom(rep(0:6, each = length(kal)), 6, rep(probs, 7)) ) ggplot(gramm, aes(x = Answer, weight = Freq))+geom_bar()+facet_wrap(~ Fish)

These estimates are based on the code above.

Binomial distribution parameter(probability)

Obs	Fish	Parameter
1	Kalaa	0.23179078
2	Petokalaa	0.17746642
3	Muikkua	0.14939457
4	Sisävesikalaa	0.09493785
5	Kirjolohta	0.29775961
6	Silakkaa	0.2152247
7	Itämeren.lohta	0.08175282
8	Muuta.Itämerestä	0.0282421
9	Pakastekalaa	0.21510166
10	Kalasäilykkeitä	0.21598007
11	Valtamerikalaa	0.1078198
12	Äyriäisiä	0.16021416

Correlation coefficients between fish dishes

Show details

Correlation coefficients between fishes(correlation) Obs Food1 Food2 Coefficient 1 Kalaa Kalaa 1 2 Petokalaa Kalaa 0.37831574 3 Muikkua Kalaa 0.286657251 4 Sisävesikalaa Kalaa 0.365112639 5 Kirjolohta Kalaa 0.530184225 6 Silakkaa Kalaa 0.425923276 7 Itämeren.lohta Kalaa 0.305152183 8 Muuta.Itämerestä Kalaa 0.2111844 9 Pakastekalaa Kalaa 0.207057983 10 Kalasäilykkeitä Kalaa 0.21677219 11 Valtamerikalaa Kalaa 0.273916098 12 Äyriäisiä Kalaa 0.143764271 13 Kalaa Petokalaa 0.37831574 14 Petokalaa Petokalaa 1 15 Muikkua Petokalaa 0.422402864 16 Sisävesikalaa Petokalaa 0.499609578 17 Kirjolohta Petokalaa 0.205863144 18 Silakkaa Petokalaa 0.202035751 19 Itämeren.lohta Petokalaa 0.079332036 20 Muuta.Itämerestä Petokalaa 0.10229554 21 Pakastekalaa Petokalaa -0.055693699 22 Kalasäilykkeitä Petokalaa 0.049204262 23 Valtamerikalaa Petokalaa 0.080707577 24 Äyriäisiä Petokalaa 0.082444392 25 Kalaa Muikkua 0.286657251 26 Petokalaa Muikkua 0.422402864 27 Muikkua Muikkua 1 28 Sisävesikalaa Muikkua 0.394067217 29 Kirjolohta Muikkua 0.227258746 30 Silakkaa Muikkua 0.24510588 31 Itämeren.lohta Muikkua 0.104823821 32 Muuta.Itämerestä Muikkua 0.127881897 33 Pakastekalaa Muikkua -0.006076891 34 Kalasäilykkeitä Muikkua 0.065331013 35 Valtamerikalaa Muikkua 0.157485773 36 Äyriäisiä Muikkua 0.072256144 37 Kalaa Sisävesikalaa 0.365112639 38 Petokalaa Sisävesikalaa 0.499609578 39 Muikkua Sisävesikalaa 0.394067217 40 Sisävesikalaa Sisävesikalaa 1 41 Kirjolohta Sisävesikalaa 0.231100033 42 Silakkaa Sisävesikalaa 0.219228101 43 Itämeren.lohta Sisävesikalaa 0.136672786 44 Muuta.Itämerestä Sisävesikalaa 0.105376927 45 Pakastekalaa Sisävesikalaa -0.013742777 46 Kalasäilykkeitä Sisävesikalaa 0.106690375 47 Valtamerikalaa Sisävesikalaa 0.188327229 48 Äyriäisiä Sisävesikalaa 0.114591507 49 Kalaa Kirjolohta 0.530184225 50 Petokalaa Kirjolohta 0.205863144 51 Muikkua Kirjolohta 0.227258746 52 Sisävesikalaa Kirjolohta 0.231100033 53 Kirjolohta Kirjolohta 1 54 Silakkaa Kirjolohta 0.38967992 55 Itämeren.lohta Kirjolohta 0.282808883 56 Muuta.Itämerestä Kirjolohta 0.111980079 57 Pakastekalaa Kirjolohta 0.166699904 58 Kalasäilykkeitä Kirjolohta 0.267376587 59 Valtamerikalaa Kirjolohta 0.304243078 60 Äyriäisiä Kirjolohta 0.154695247 61 Kalaa Silakkaa 0.425923276 62 Petokalaa Silakkaa 0.202035751 63 Muikkua Silakkaa 0.24510588 64 Sisävesikalaa Silakkaa 0.219228101 65 Kirjolohta Silakkaa 0.38967992 66 Silakkaa Silakkaa 1 67 Itämeren.lohta Silakkaa 0.303378534 68 Muuta.Itämerestä Silakkaa 0.317929228 69 Pakastekalaa Silakkaa 0.106271923 70 Kalasäilykkeitä Silakkaa 0.202491349 71 Valtamerikalaa Silakkaa 0.27027233 72 Äyriäisiä Silakkaa 0.17446606 73 Kalaa Itämeren.lohta 0.305152183 74 Petokalaa Itämeren.lohta 0.079332036 75 Muikkua Itämeren.lohta 0.104823821 76 Sisävesikalaa Itämeren.lohta 0.136672786 77 Kirjolohta Itämeren.lohta 0.282808883 78 Silakkaa Itämeren.lohta 0.303378534 79 Itämeren.lohta Itämeren.lohta 1 80 Muuta.Itämerestä Itämeren.lohta 0.589928968 81 Pakastekalaa Itämeren.lohta 0.083123081 82 Kalasäilykkeitä Itämeren.lohta 0.167824298 83 Valtamerikalaa Itämeren.lohta 0.415932209 84 Äyriäisiä Itämeren.lohta 0.312466845 85 Kalaa Muuta.Itämerestä 0.2111844 86 Petokalaa Muuta.Itämerestä 0.10229554 87 Muikkua Muuta.Itämerestä 0.127881897 88 Sisävesikalaa Muuta.Itämerestä 0.105376927 89 Kirjolohta Muuta.Itämerestä 0.111980079 90 Silakkaa Muuta.Itämerestä 0.317929228 91 Itämeren.lohta Muuta.Itämerestä 0.589928968 92 Muuta.Itämerestä Muuta.Itämerestä 1 93 Pakastekalaa Muuta.Itämerestä 0.061080508 94 Kalasäilykkeitä Muuta.Itämerestä 0.164729369 95 Valtamerikalaa Muuta.Itämerestä 0.381394106 96 Äyriäisiä Muuta.Itämerestä 0.295389062 97 Kalaa Pakastekalaa 0.207057983 98 Petokalaa Pakastekalaa -0.055693699 99 Muikkua Pakastekalaa -0.006076891 100 Sisävesikalaa Pakastekalaa -0.013742777 101 Kirjolohta Pakastekalaa 0.166699904 102 Silakkaa Pakastekalaa 0.106271923 103 Itämeren.lohta Pakastekalaa 0.083123081 104 Muuta.Itämerestä Pakastekalaa 0.061080508 105 Pakastekalaa Pakastekalaa 1 106 Kalasäilykkeitä Pakastekalaa 0.345409216 107 Valtamerikalaa Pakastekalaa 0.129614914 108 Äyriäisiä Pakastekalaa 0.125383855 109 Kalaa Kalasäilykkeitä 0.21677219 110 Petokalaa Kalasäilykkeitä 0.049204262 111 Muikkua Kalasäilykkeitä 0.065331013 112 Sisävesikalaa Kalasäilykkeitä 0.106690375 113 Kirjolohta Kalasäilykkeitä 0.267376587 114 Silakkaa Kalasäilykkeitä 0.202491349 115 Itämeren.lohta Kalasäilykkeitä 0.167824298 116 Muuta.Itämerestä Kalasäilykkeitä 0.164729369 117 Pakastekalaa Kalasäilykkeitä 0.345409216 118 Kalasäilykkeitä Kalasäilykkeitä 1 119 Valtamerikalaa Kalasäilykkeitä 0.212919187 120 Äyriäisiä Kalasäilykkeitä 0.296174683 121 Kalaa Valtamerikalaa 0.273916098 122 Petokalaa Valtamerikalaa 0.080707577 123 Muikkua Valtamerikalaa 0.157485773 124 Sisävesikalaa Valtamerikalaa 0.188327229 125 Kirjolohta Valtamerikalaa 0.304243078 126 Silakkaa Valtamerikalaa 0.27027233 127 Itämeren.lohta Valtamerikalaa 0.415932209 128 Muuta.Itämerestä Valtamerikalaa 0.381394106 129 Pakastekalaa Valtamerikalaa 0.129614914 130 Kalasäilykkeitä Valtamerikalaa 0.212919187 131 Valtamerikalaa Valtamerikalaa 1 132 Äyriäisiä Valtamerikalaa 0.272081755 133 Kalaa Äyriäisiä 0.143764271 134 Petokalaa Äyriäisiä 0.082444392 135 Muikkua Äyriäisiä 0.072256144 136 Sisävesikalaa Äyriäisiä 0.114591507 137 Kirjolohta Äyriäisiä 0.154695247 138 Silakkaa Äyriäisiä 0.17446606 139 Itämeren.lohta Äyriäisiä 0.312466845 140 Muuta.Itämerestä Äyriäisiä 0.295389062 141 Pakastekalaa Äyriäisiä 0.125383855 142 Kalasäilykkeitä Äyriäisiä 0.296174683 143 Valtamerikalaa Äyriäisiä 0.272081755 144 Äyriäisiä Äyriäisiä 1

EU kalat

• The code that used to be here was moved to EU-kalat#Calculations.
• What updates should be done:
  • Plot iterations to see that the model results do not drift.
  • Take modelled parameters and develop a MC model to produce predicted concentrations.
    • TCDD concentration should be added to the hierearchical Bayes model for this?
  • KTL Sarcoma study, EU-kalat and Goherr: Fish consumption study should all be combined into one model. ----#: . Can models be combined as text with paste()? This could work if all submodels had unique parameter names like N.eu and N.goh rather than just N. And data lists are merged simply with c(). --Jouni (talk) 16:22, 22 January 2017 (UTC) (type: truth; paradigms: science: comment)
  • A causal diagram should be drawn to show the model structure.
• JAGS user manual [2] (with e.g. distribution names and other guidance)
• How to generate predictions in JAGS [3]
• Using rjags, a simple guidance [4]

Related:

• Easily generate correlated variables from any distribution (without copulas) [5]

See also

• Clinicaltrials: Another clinical study on sarcoma and pollutants (incl. dioxin)

Related files

• KTL_sarcoma_questionnaire_finnish
• KTL_sarcoma_questionnaire_swedish