The study of the association between environmental exposure
The study of the association between environmental exposure and preterm birth: design and methodological aspects Patrizia Schifano and Xavier Basagana “Challenges for epidemiology in the context of the National Health Service” Rome 15 -16 October 2012
Preterm Birth v. Preterm birth is reported to be the main cause of perinatal mortality in the US and in Europe ü 5 -7 % of total birth in Europe ü 10 -12% of total birth in the US v. The etiology of preterm birth is not well known: ü probably a mix of genetic, environmental and socioeconomic factors v. Also a small reduction in the duration of pregnancy has resulted to be associated with adverse health outcomes in neonatal and later life
Who is preterm? v. Preterm births are classified in early and late preterm, according to gestational age v. Singleton late preterm births (34 -36 weeks) comprise the largest proportion (74%) of singleton preterm births and the fastest-growing segment.
Preterm birth and health outcomes v. The severity of heath conditions decrease with gestational age, but also late preterm infants have a higher incidence of morbidity and mortality when compared with term infants v. Brain injury, neuro-developmental disorders, broncopulmonary displasia and growth impairment are the main group of causes associated with preterm birth
Preterm births and environmental exposure v. Several specific mechanisms are suspected to underlie adverse birth outcomes among mothers exposed to environmental hazards v. Recently there is a growing interest in studying the potential effect of air temperature and air pollutants. 2011, Environmental Research v. Few studies have examined the relationship between ambient temperature and birth outcomes directly v. Very heterogeneus studies by study design, definition of birth outcomes, exposure windows and statistical methods Further research is needed
Methodological critical aspects ØSample Size o. Small expected effect ØMeasure of gestational age ØWindow of exposure o. It is not clear which is the most vulnerable period during pregnancy; this changes according to the exposure and the outcome considered ØAdjustment for seasonality o. Births have their on seasonality; this has to be taken into account in the study of the effect of temperature ØConfounders and effect modifiers o. Week of gestation o. Maternal characteristic o. Other meteorological exposure of the association
Confounding by gestational age in time series üA Time series approach is well suited for the study of most environmental exposures and preterm birth v. Individual-level risk factors that do not vary across time cannot act as confounder üPotential for confounding by gestational age and individual risk factors in time series studies of pregnancy outcomes Darrow et al, Epidemiology, 2009
Confounding by gestational age in time series v. Gestational age is a strong predictor of imminent birth v. Gestational age distribution might vary seasonally in the pregnancy risk set, acting as a confounder of the preterm-temperature relationship v. Risk factors invariant at the individual level may vary seasonally when aggregated into pregnancy risk-set, and act as confounders of the studied relationship v. Methods that account for gestational age are reccomanded
What is relevant to know for public health 1. Is there an effect of heat on preterm probability? Rome study 2. By how much does heat shorten average gestational age? Barcelona study 3. During which gestational weeks can heat induce labor? Rome study Is the heat-preterm birth relationship relevant for public health?
The Study of Rome ØObjective: to evaluate the effect of maternal short-term exposure to heat on the probability of pre-term delivery ØStudy population: cohort of births occurred in Rome between 2001 and 2010, from April to October Inclusion Criteria • Single birth between the 22° and 36° week of gestation • Natural delivery • Other delivery tipology with spontaneous onset of labour • Mother resident in Rome and who gave birth in city hospitals ØGestational age: date of birth - date of last menstrual period 78633 births 4314 preterm births (5. 5%)
Exposure: Daily maximum apparent temperature Statistical analysis • Seasonality of births • Lag structure • Shape Percent change in the mean daily number of preterm birth per 1°C increase in maximum apparent temperature Model: Poisson generalized additive model Outcome variable: daily number of preterm births Offeset: daily number of ongoing gestation in utero at risk for preterm birth Adjusted for: seasonal trend, long term trend, day of the week, holidays, air polluttants: PM 10; NO 2, O 3
Results: Lag distribution of maximum apparent temperature effect on preterm births lag 0 -3 lag 0 -7
Results: Whole Year April - October Linear relationship
Results: %Change in pre-term births between 22° and 36° week per 1°C increase in MAT (best model according to AIC) lag 0 -3 1. 75 IC 95% 0. 73 – 2. 79 lag 0 -7 1. 50 IC 95% 0. 32 – 2. 69 ØNo confounding by air pollutants ØEffect of Ozone in an indipendent model
Results: Sensitivity analysis of births between 22 nd and 32 th week (484 cases) Any significant effect Ø no effect Ø lack of power Sensitivity analysis on cesarean deliveries (elective and emergency) Any significant effect
The study of Barcelona Dadvand et al. Climate extremes and the length of gestation. EHP 2011; 119: 1449 -53. v Cohort of 7, 585 pregnant women. v Similar inclusion criteria than in Rome. v Gestational age: Objective measure (ultrasound exam). v Statistical analysis: v Linear regression of gestational age. v Exposure: binary indicator for extreme heat v based on 99 th percentile of heat index. v lags 0 -6 examined. v 2 -Stage model to control for time trends.
The study of Barcelona v Rationale for the analysis: v Extreme heat could induce labor. v This could happen at any gestational age. v After a extreme heat event we should observe lower lengths of gestation.
The study of Barcelona v Main result: The day after a extreme heat event there is a 5 day reduction in average length of gestation. Concerns: v This provides an answer to Question 2. v It does not directly address effect on pre-term births. v More powerful than binary response. v. Average change may not be the appropriate metric. v Changes can be more noticeable in the left tail. v Gestational age not normally distributed.
Recap v 3 relevant scientific questions. v 2 different types of analysis, each one addressing a different question. v Both techniques have some advantages and some drawbacks. v Power & efficiency v Interpretation, clinical relevance v Distributional properties v Choice of cut-off v…
Way forward Flexible Bayes regression v Flexible method for density regression, which allow the entire density to change flexibly with predictors. From: Dunson DB. Flexible Bayes regression of epidemiologic data. In: The Oxford Handbook of Applied Bayesian Analysis
Flexible Bayes Regression v E. g. Approximate the distribution by a mixture of four Normals (4 latent classes): full term births, late preterm births, early preterm births, and late term births. v. Mixture model:
Flexible Bayes Regression Example: distribution of gestational length as a function of DDE. Thicker left tail at high values of DDE.
Flexible Bayes Regression From this analysis we can get other outputs. v E. g. to answer question 2:
Flexible Bayes Regression v E. g. to answer questions 1 and 3: v use various thresholds for defining preterm birth
Flexible Bayes Regression Advantages: v Single analysis to answer all questions (e. g. multiple thresholds for preterm). v It does not discard information → Efficiency gains. v Especially important since one may have limited data in the tails, where the main interest is. v Bayesian approach of dealing with sparsity → borrowing information. ØAcross values of gestational age ØAcross values of the exposure. v Leads to smooth and realistic estimates.
Future challenges and collaborations ØCollaboration between DEP and CREAL v. To design a more powerful and efficient study: • Multicentric • Italian cohort (multicity cohort, 6 cities) • Spanish cohort (multicity cohort, 2 cities) • Applying the proposed methodology v. Answer to question 1 to 3 v. Study potential effect modifiers of the analysed association v. Effect of ozone and other polluttant on preterm birth v. Interaction between temperature and ozone
- Slides: 27