Environmental Epidemiology

11th EPIET Epidemiology Course Menorca, October 2 2006 Environmental Epidemiology (introduction) Dr Georges Salines Institut de Veille Sanitaire Département Santé Environnement I. Objectives To provide a basic knowledge of : ► The definitions of environmental health, environmental epidemiology, environmental risks ► The concept of low-risk and the links between relative risk, prevalence of exposure and attributable risk ► The limits of epidemiology in environmental health ► How to deal with these limits Definitions is the study of the distribution and determinants of health-related states or events in specified populations ► The environment is all the physical, chemical and biological factors external to a person, and all the related behaviours. (WHO) ► The environment is the sum of all external conditions affecting the life, development and survival of an organism (US EPA) ► The environment is everything that is not me (Einstein) ► Epidemiology Traditional exclusions ► Genetics genes/environment) ► Behaviours (except behaviours modifying exposures) ► Social factors (except links between SES and physical environment) ► Infectious diseases (except those transmitted through exposure to media) factors (except interactions Risk measure of the probability that damage to life, health, property, and/or the environment will occur as a result of a given hazard (US EPA) ► Rylander classification  RR > = 10 : people themselves recognize the risk  RR de 9 à 2 : « comfort zone » for epidemiology  RR < 2 : zone where epidemiology reaches its limits... ►A High risks ► occupational environment        aromatic amines and bladder cancer asbestos fibres and mesothelioma cadmium and kidney diseases benzene and leukaemia pesticides and infertility organic solvents and neurological disorders etc ... ► general environment December 1952 - London December 1952 - London 1953 - Minamata December 1984 - Bhopal 1986 - Tchernobyl Thyroid cancer in children 120 Incidence Rate per million 100 80 60 40 20 0 1986 1987 1988 1989 1990 1991 1992 1993 1994 Belarus Ukraine Briansk Gomel Years 2003 - Paris Mortality and mean temperature in Paris 1999-2002 versus 2003 Peak: Aug 13th 2005 - Katrina 2006 Abidjan ► Nature of high risks in general environment ► anthropogenic activities  London 1952  Minamata 1953 … ► natural origin  Heat waves  hurricanes… ► mixed origin  UV and melanoma  tremolite and mesothelioma in New Caledonia  erionite and mesothelioma in Turkey ... Characteristics of high risks ► High RR  benzidine / bladder cancer  asbestos / mesothelioma  tobacco (>25g/d) / lung cancer RR = 500 RR = 50 RR = 30 Usually severe and often specific health endpoints ► “well defined” populations ►  in space, in time  socio-demographic characteristics  relatively small populations Low risks ► urban air pollution and short-term respiratory diseases  RR = 1.1 - 1.5 ► chlorinated water supplies and bladder cancer fields and children leukemia  RR = 1.4 ► electromagnetic  RR = 1.3 ... Small relative risks do not mean small health impacts ► Relative ►ratio risk and attributable risk measure : “it is an indicator for  relative risk epidemiologist ”  attributable risk ►FRA “if the relation is causal, it estimates the = p * ( RR -1) / [ 1+ p * ( RR - 1) ] proportion (amount) of diseases that we can attribute to the exposure” Health impact Exposure prevalence 0,01 0,10 Relative Risks 10 5 2 1,5 1,2 1,1 0,08 0,04 0,01 0,47 0,29 0,09 0,05 0,02 0,01 0,90 0,80 0,50 0,33 0,17 0,09 1,00 Health impact Exposure prevalence Relative Risks 10 5 2 1,5 1,2 1,1 0,08 0,04 0,01 0,47 0,29 0,09 0,05 0,02 0,01 0,90 0,80 0,50 0,33 0,17 0,09 0,01 0,10 1,00 May be not that low after all low risks or weak associations ? Theoretical baseline situation (the wonderful world) E0 E1 E2 Prevalence 80% 15% 5% Incidence* 100 300 500 RR** 1.0 3.0 5.0 E0 = non exposed, E1=low exposure, E2=high exposure * Incidence : x /100.000, ** RR : true Relative Risk Heterogeneity in the population’s sensitivity to the exposure E0 E1 E2 Prevalence 80% 15% 5% 50% Incidence (S) Incidence (s) 100 100 100 300 200 250 500 300 400 50% RR 1.0 2.5 4.0 * (S) : high sensitivity. (s) : low sensitivity Non specific definition of the health outcome E0 E1 E2 Prevalence 80% 15% 5% Incidence (D) Incidence (d) 100 50 150 250 50 300 400 50 450 RR 1.0 2.0 3.0 * (D) : disease specifically related to exposure. (d) : disease not related to exposure Errors in the exposure classification E0 E1 E2 Prevalence 50% 35% 15% Incidence 150 214.3 250 RR 1.0 1.43 1.67 20% of non exposed (E0) are categorised E1 and 10% of non-exposed are categorised E2. Inaccuracy in the exposure categories E0 E1 Prevalence 50% 50% Incidence RR 150 1.0 225 1.5 Epidemiology and weak associations ► Improve data quality  exposure  health endpoints  co-factors ► Improve statistical power designs  Meta-analysis & Multi centres ► Ecological Improving assessment of exposure: better use of environmental data  appropriate selection of sources and routes of exposure  taking account: ►critical periods of exposure ►individual history of exposure : behaviour, spacetime activities … Example Exposure = last place of residence, assuming entire life at this residence and consumption of the town’s drinking water Chlorination exposure Cases (286) Controls (658) OR IC 95% 0 1-25 26-50 50+ 84 193 1.0 31 104 0.7 0.4 - 1.1 79 188 1.0 0.7 - 1.4 92 173 1.2 0.9 - 1.8 Lynch et al, Arch Env Health 1989;44(4):252-259 Example (2) Exposure = entire lifetime residential history combined with the town’s drinking water chlorination history Chlorination exposure Cases (286) Controls (658) OR IC 95% 0 1-25 26-50 50+ 65 211 1.0 92 211 1.4 1.1 - 2.1 102 195 1.7* 1.2 - 2.5 27 41 2.1* 1.2 - 3.7 Lynch et al, Arch Env Health 1989;44(4):252-259 Improving assessment of exposure: personal exposure monitoring  technical, logistical and financial limits …  depends on sensibility / specificity of the method Improving assessment of exposure: biomarkers of exposure  cellular, biochemical, molecular alterations ► measurable in biological media (human tissues, cells or fluids)  advantages of a dose (effectively absorbed) ► integration of all the routes of exposure and sources of absorption ► avoids subjects’ lack of knowledge, memory failure, biased recall, deliberate misinformation … ► measurement  limits costs ► Representativity of a single sample taken at a particular time ► In some cases, route of exposure is of the essence ► Improving assessment of health endpoints  outcomes specified “as precisely as possible” ►subgroups of disease  biomarkers of effects clinical events ►predictive value ? ►variability  biological, laboratory-related, logistical issues (bias) ►sub Measuring confounders and effect modifiers  “as much attention” as exposure and disease variables  Biomarkers of susceptibility Example Combined risk of bladder cancer associated with smoking exposure and GSTµ1 genotype among whites Smoking exposure Genotype GTSµ1 (-) 0 (non smokers) 1.3 [0.6 - 2.7] 1-50 (pack-years) 4.3 [2.1 - 3.5] > 50 (pack-years) 5.9 [2.6 - 13.0] GTSµ1 (+) 1.0 [Ref] 2.2 [1.1 - 4.5] 3.5 [1.5 - 8.0] Bell D.A. J Nat Cancer Inst 1993;85(14):1159-64 Improving statistical power ► Increasing Odds ratio sample size Prevalence of the exposure in the population 20% 4,0 3,0 2,0 1,5 20000 15000 10000 5000 0 1,1 1,2 1,3 10% 55 96 279 903 5% 94 168 504 1674 2% 211 387 1186 4004 1% 418 755 2333 7888 38 63 171 532 1,4 1,5 1,6 Odds ratio 1,7 1,8 1,9 2 Number of cases and controls (1/1) for 1- b = 80%, a = 5%, H0: OR=1 Improving statistical power ► “Mammoth” studies  Expansive  Complex ► Pooling data  Meta-analysis (or combined analysis)  Multi centres studies ► heterogeneity ? Ecological studies : principle data ► Statistical unit = « group »  Group exposure ►Mean ► Agregated exposure, environmental proxy of disease in the statistical unit, SIR,  Group effect ►Frequency SMR Avantages of Ecological studies exposure contrasts may be found between populations than between individuals within the same population ► Large number of observations  Statistical power ► Use ► Wider of existing data  rapid  Cost-effective Geographical studies ► Statistical units = geographical areas  Exposure levels : E1, E2, …, Ei  Prevalence or incidence levels: M1, M2, …., Mi ► Resarch of an association between :  Variations of exposure levels  Variation of health indicators Limits: Biases and fallacies ► Classification ► Surveillance ► Selection ►« Ecological fallacy » Classification errors M E E M Often non differential = Risk dilution toward 1 (bias toward false negative) Surveillance bias Vicinity of a Nuclear Plant Leukemia Register « Non exposed » Zone All cancers Register Often differential: bias toward false positive (if better sensitivity) or toward false negative (if better specificity) Selection Bias ► Example ► Example 1: Texas Sharpshooter (Bias toward false positive) 2: Flight of the sick people (Bias toward false negative) Ecological Fallacy in Geographical study Incidence rate Area A Area B Area C Environmental exposure Ecological Fallacy Incidence rate population A    population B    population C       Individual exposure Example leukaemia cluster among children living near the Sellafield nuclear waste reprocessing plant (United Kingdom) ► Other leukaemia clusters have since been identified near other nuclear sites, such as Dounreay in Scotland and Krümmel in Germany ► 1983: But… view of current knowledge about the relation between exposure to radiation and the risk of leukemia, dose levels around nuclear sites are incompatible with the excess risks observed … ► Studies considering several sites (United Kingdom, France, USA, Germany, Canada, Japan, Sweden, Spain) have not detected any global excess ► Leukaemia clusters have been observed in areas far from any nuclear site ► There are alternative hypotheses which may explain the leukaemia clusters located near some nuclear sites ► In Interpretation of geographical studies of geographical associations ► Very difficult to extrapolate at the individual level ► Causality generaly out of reach of those designs ► Useful for generating hypotheses ► Measures Time series power ►Control of confounding factors +++  Non time-dependant: Population is its own control  Time-dependant: modelling techniques ►Statistical Exemple PSAS9 I D day Exposed population Indicator of exposure Indicator of effect All people living in Marseilles SO2 mg/m3 (Daily mean of 3 monitoring stations) Daily number of deaths Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Raw curves Mean levels of air pollution Marseilles, 1990-1995 Daily counts of deaths, Marseilles, 19901995 D io x y d e de s ouf r e ( S O2) 0 20 40 µg/m3 60 80 10 J an an an an an an J 1990 91 92 93 94 95 J 19 19 19 19 19 J J J Fu mé e s n o ir e s ( FN ) Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 60 80 Time-dependant counfounding factors Serial correlation fonction of daily mortality Fonction totale. 0.20 PACF -0.10 0 -0.05 0.0 0.05 0.10 0.15 5 10 15 Jour 20 25 30 Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Time-dependant counfounding factors Risk factor Health outcome Confounding factor Daily variation of air pollution levels Daily variation of number of deaths from respiratory causes Daily variations of weather Filleul et coll., Rev. Mal. Respir., 2001 Non time-dependant counfounding factors Risk factor Health outcome Confounding factor Daily variation of air pollution levels Daily variation of number of deaths from respiratory causes Tobacco Filleul et coll., Rev. Mal. Respir., 2001 Modeling: Strip-tease of the curves into acount long-term trends (ie:decrease of mortality) ► Taking into acount seasonal variations (Higher mortality during winter) ► Taking into acount the day of the week ► Taking into acount co-factors (Meteorological data, Flu epidemics, Pollinic data...) ► Taking Long-term trends Predicted value of total mortality by trend-modeling. Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Seasonal variations Predicted value of mortality by modelization of seasonal variations Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Meteorological data Naperian Logarithm of Relative Risk of the interaction temperature-humidity on total mortality la t i f L n d u r is q u e r e 0 .0 5 0 - 0 .0 5 20 15 10 hu m 5 id i te 0 -5 -1 0 -5 0 5 a tu pe r te m 1 0 inim a r e .m le 15 20 25 Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Day of the week Predicted value of total mortality by modelization of a « day of the week & holidays» effect Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Full Monty Residual values of total mortality after modelization of trend, seasonal variatipons, Flu epidemics, temperature, humidity, day of the week & holidays -0.10 -0.05 Serial correlation fonction of daily mortality after modelization of trend, seasonal variatipons, Flu epidemics, temperature, humidity, day of the week & holidays 0 5 10 15 Jour 20 25 30 P A C F Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 0.0 0.05 0.10 0.15 0.20 Result “dose-response curve” of total mortality in relation to SO2 levels 1. 10 Ri sq ue Re lati f 1. 08 1. 06 1. 04 non-paramétrique 1. 02 linéaire 1. 00 0 20 40 SO2 moyenne 0-1 jours 60 80 Source : Surveillance épidémiologique air et santé, rapport InVS, mars 1999 Interpretation of time-series studies ► Establishing causation is possible after a careful discussion of Hill criteria 1. 2. 3. 4. 5. 6. 7. 8. 9. Strength. Consistency. Specificity. Temporality. Biological gradient (dose-response). Plausibility. Coherence. Experiment. Analogy. V. Conclusion of the study design that involves measurements of variables are critical, especially in environmental epidemiology where risks from exposure are likely to be small, difficult to detect, and perhaps not clinically significant, yet maybe of public health importance ► Epidemiology is not always the only answer of even the more relevant one to questions submitted to environmental epidemiologists: Risk analysis for example, is a very useful and costeffective method ► ...but this is another story. ► Aspects