The EPILYMPH study is a multicentre case–control study on lymphoma aetiology conducted in 22 centres of six European countries (six centres in Germany, two in Italy, four in Spain, six in Ireland, three in France, and one in the Czech Republic). Case recruitment began in 1998 and ended in 2004. The methods are described in more detail in previous papers
Controls were matched to cases in each centre by age (+/− 5 years) and gender. A random sample of the general population was selected to serve as a control in Germany and Italy. Meanwhile, centres in Ireland, France, Spain and the Czech Republic selected hospital controls, using uniform criteria for eligibility: patients were excluded as potential controls if the reason for hospitalization at the time of recruitment was cancer, organ transplantation, and/or a systemic infection.
Trained personnel conducted personal interviews for cases and controls alike using the same standardized questionnaire translated into the local language. The questionnaire addressed socio-demographic characteristics, including smoking status, weight and height, and -included a detailed lifelong occupational history.
Smoking status was categorized as ever smoker vs. never smoker, with those who smoked further analysed based on the number of packs per year. Weight and height were used to construct the Body Mass Index categorized using 30 as the cut-off point. Level of education was broken down into three categories: low level (primary school), medium level (high school) and high level (tertiary education).
The occupational history interview compiled a detailed lifelong job history, including job title, task description and duration. Based on the job title indicated, the participant was asked to provide additional detail using 15 special job modules which contained specific question on certain exposures—pesticides, organic solvents, living animals, working with children, radiation, dust—of prior interest to the researchers.
Each job was coded by a trained local industrial hygienist (IH) in participating countries. All job entries of one year or more listed under work histories were coded using the 1968 Standard Classification of Occupations (ISCO), and the 1990 Statistical Classification of Economic Activities in the European Community (NACE).
The ISCO coding system classifies jobs in a set of groups according to the tasks and duties undertaken in the job. ISCO codes can have a maximum of five digits. The first two digits describe broader occupational categories (e.g. 62 = general farmers). The third digit specifies the occupation in more detail (e.g. 621= ‘general farm workers’). The 4th and 5th digits further specify the task within the occupation group (e.g. 62200= ‘field crop and vegetable farm workers’). For the purpose of this study we used the more general first two digits of the codes, except for farming categories and exposure to organic solvents, for which the more detailed five-digit codes were used
Duration of employment was categorized as <= 9 years and ≥ 10 years in reference to participants who never held that particular job title.
The EPILYMPH study investigated specifically the effect of several exposure groups: inorganic and organic pesticides, contact with live animals, dust, contact with meat, organic solvents, working with children and ionizing radiation.
The IH were trained to consistently assess exposure according to the job and task description. Three measures were used: frequency of exposure, intensity of exposure and confidence in the exposure assessment. Frequency of exposure was categorized as low (1) if the exposure occurred 1% to 5% of the working time; medium (2) if it occurred during 5% to 30% of working time; and high (3) if the exposure occurred more than 30% of the working time. Intensity of the exposure was categorized as low (1), medium (2) or high (3), with cut points for each agent based on the US Occupational Safety and Health Administration’s Threshold Limit Value (TLV), when available. Low intensity was defined as less than 50% of the TLV; medium was 50%-150% of the TLV, and high was more than 150% of the TLV. Intensity levels were based on a common standardized intensity scale with local and subject-specific estimates. When no TLV was available for a specific agent, benchmark occupations or tasks were used.
In terms of the third component, the IH expressed his/her own confidence in the exposure assessment using a three point scale, representing the degree of certainty that exposure to a specific substance had occurred (three being the highest). Duration of exposure was also recorded.
The study was approved by the Ethics Committee of each participating centre. All study subjects provided informed consent.
A power calculation for this analysis was carried out to ascertain the minimum odds ratio (OR) to be detected with a 5% probability of α error and 80% of statistical power (ß=80), assuming a 20% prevalence of exposed subjects in the control group, a calculation based on the proportion of the control group who were farmers, resulting in a maximum matching ratio of controls to cases of 4:1.
Under such conditions, an OR of 1.7 would be detectable with 89% power. Consequently, four controls for each MM case were randomly selected from the total EPILYMPH controls, matching by age (+/− 5 years), gender and centre.
The OR and the respective 95% confidence interval were calculated using univariate analysis for level of education, smoking status, body mass index.
An unconditional regression model was used to deduct OR and 95% CI for each of the two-digit ISCO code categories adjusting for statistical significant co-variables.
The exposures were dichotomized into ever exposed /never exposed. For pesticide exposure and organic solvents, three additional variables were constructed: duration (no exposure, less than 10 years, 10 years or more); weighted cumulative exposure (duration x frequency), categorized in tertiles; and intensity categorized based on the IH classification as low, medium and high. A sub-group analysis by level of confidence was performed to test confidence-related changes in the risk estimates. The log-likelihood ratio was used to calculate the risk trend.
Sensitivity analysis was done to determine whether the risk varied between countries that selected population controls versus countries that selected hospital controls.
All the analyses were carried out using the Stata 9® software.