ALS subjects were recruited through the University of Michigan ALS Clinic with the following inclusion criteria: (1) age greater than 18 years, (2) possible, probable, probable lab-supported, or definite ALS by the revised El Escorial criteria [17], (3) able to provide informed consent, and (4) able to understand and communicate in English. Age- and gender-matched controls were recruited through postings and a University of Michigan website (umclinicaltrials.org) that described the study and eligibility criteria, which included: (1) no diagnosis of ALS or other neurodegenerative condition, (2) no family history of ALS in a first or second degree blood relative, (3) ability to understand and communicate in verbal and written English, and (4) no underlying health condition or medication use that made participation risky, as determined by the study team. Age was matched by decade and gender (e.g., males ages 50–59, females ages 50–59). The study was approved by the Institutional Review Board (Protocol # HUM00028826) at the University of Michigan. All participants provided written informed consent.

Participants completed a detailed self-administered written questionnaire that encompassed occupational and residential exposures, residence location, exercise and sports, body weight, tobacco use, military experience, and family history. Questionnaires were mailed to subjects and telephone follow-up was conducted for clarifications, if needed. For patients who had difficulty communicating, the next of kin completed the instrument. Typically, about 90 minutes was needed to complete the fairly lengthy (200 questions, 28 page) questionnaire, which was divided into several sections and used a structured query approach, allowing participants to skip irrelevant sections. Portions of this questionnaire were adapted from those used in the National Health and Nutrition Examination Survey Questionnaire [18].

After requesting standard demographic information, residential history was collected using 54 questions for the current dwelling and 18 questions for each of the three previous dwellings. Questions addressed dates of building construction and occupancy, building type (including detached single family, duplex, multi-family/apartment, mobile home or trailer, other, or unknown) and features (including building materials, floor coverings, presence of basement, presence of outdoor storage, and presence of garage), weatherization, storage of chemicals (e.g., pesticides, solvents, gasoline, and paint among others), and drinking water source (e.g., well or city water). A detailed smoking history was collected, including whether subjects ever smoked and, if applicable, the year smoking started, stopped, the type of tobacco used, and the frequency of smoking. Participants were asked about hobbies, particularly those that might include chemical exposure, such as wood working, metal working, home remodeling, lawn care, automobile repair, small engine repair, and painting. Participants were asked about their physical activity, including the frequency of ten categories of activities (jogging/running, bicycling, swimming, aerobic dancing, recreational dancing, calisthenics, gardening or yard work, weightlifting, playing soccer/football/baseball/field hockey/golf, playing ice hockey/tennis/boxing/wresting) using a five year recall period. In addition, we asked about other activities not mentioned above. Summary or composite measures of physical activity were developed and classified as low, medium, and high based on the number of activities by an individual (defined as 0–3, 4–6, and 7+ activities, respectively).

A detailed occupational history was also requested, and the questionnaire included 24 questions for each of four previous jobs; specifically, their current or most recent job, the previous job, and the two other jobs that were held for the longest period of time. Requested information included job title, industry type, and dates of employment. In addition, subjects were asked to identify occupational exposures, using lists of potential workplace hazards and exposures (e.g., specific chemicals, particles, radiation), the use and presence of personal protective equipment, and hygiene habits (e.g., hand washing).

Following data entry and consolidation, initial analyses included data cleaning and verification. Dubious or missing entries were checked by either reviewing the original questionnaire, telephone follow up, or as a last resort and if practical, imputed manually based on related questions. Next, using the residence and occupational histories, residence and employment durations were calculated and compared as a test of internal validity. If the residence or job information was inconsistent, such as residing in a different state compared to the workplace, then respondents were queried by telephone and the information revised as indicated. If this attempt failed, then the questioned information was set as missing. Job titles were coded using the Dictionary of Occupational Titles, and workplace types were coded using the North American Industry Classification System. Several new composite variables for specific risk factors were also created; for example, a binary (yes/no) variable called “occupational exposure to heavy metals” combined any positive response to separate questions concerning workplace exposure to arsenic, beryllium, cadmium, chromate, lead, mercury, nickel, and welding fumes. Such variables were created for each job, and also across all jobs. Similarly, composite variables were created for risk factors pertaining to residences and hobbies.

The four exposure time frames that were considered were: (1) no exposure (in the past 30 years), (2) exposure in the last 10 years only (not earlier), (3) exposure 10 to 30 years ago and not in the past 10 years, and (4) continuous exposure over the past 30 years. These time frames were referenced to survey completion in the control population and ALS symptom onset for the case population.

Questions not answered by subjects were interpreted as missing. To obtain a full dataset needed in the conditional logistic regression (CLR) models (described below), missing values were imputed with five replacements. The consistency of imputed data was confirmed with manual checks (for example, the imputed number of years of smoking should be smaller than the subject's age). Subsequent sensitivity analyses confirmed that results obtained using original and imputed data sets were similar.

The survey data were used to generate approximately 100 variables as potential risk factors for each of the four exposure time windows, and a subset of variables for further analysis was selected for use in the CLR models. For continuous variables, univariate statistics (range, mean, medium, quartiles) were calculated and differences between cases and controls were tested using Student t-tests (normally distributed variables) and Kruscal-Wallis tests (non-normal distributions). For categorical variables, cross-tabulations and chi-square tests were used to detect differences between cases and controls. Fisher's exact test was used if the expected count was less than five. Variables showing variability and significant differences were retained for further analyses, as were several variables identified in prior research, such as smoking, physical activity, exposures to metals, and exposures to radiation. Related exposure variables that were moderately to highly correlated (r>0.5) were consolidated as a new variable (e.g., occupational exposures to radiation, x-rays, and electromagnetic fields were grouped into a radiation variable). Stepwise regression was then used to generate final models. A sensitivity analysis for variable entry (with p-values from 0.1 to 0.3) and removal (p-values from 0.15 to 0.35 for variables) showed that in most cases the same variables were selected. Then, odds ratios (OR) and p-values (significance level: p-value <0.1) for potential risk factors were estimated using CLR models for case/control pairs matched on age and gender. Each model included covariates to control for demographics, smoking (cigarette packs per day), physical activity status (low, medium, and high physical intensity groups), and educational attainment. Four models with different exposure conditions (model 1: exposure in the last 30 years, model 2: exposure in the last 10 years, model 3: exposure in the period from 30 years ago to 10 years ago, model 4: continuous exposure in the last 30 years) were constructed to test potential risk factors such that (1)where α =  effect of the stratum (the matching variables age and gender), m =  number of matched sets (strata), β =  log-odds ratio of interested risk factors, x =  risk factor, and k =  number of variables. Several interaction effects were tested (e.g., two-way interactions between education level and exposure risk factors). Because interaction effects were not statistically significant, they were not retained in the final models. Several sensitivity analyses were used to investigate the robustness of the model results to the exposure window.

Data was entered and managed using RedCap [19]. Some analyses used Excel (Microsoft, Redmond, WA, USA). Multiple imputation IVEware 2.0 (Survey Research Center, Institute for Social Research, University of Michigan, Ann Arbor, MI, USA) was used to replace missing data. Statistical analyses were performed using SAS 9.3 (SAS Institute Inc., Cary, NC, USA). The following data represent an interim analysis of an ongoing larger study.

The study population was 31% female and averaged (± SD) 61.6±9.0 years of age. Cases and controls had the same age and gender distribution (Table 1), reflecting successful matching. Controls had a higher level of education attainment than cases (p<0.001), with 67% of cases and 95% of controls receiving education beyond high school. Controls were also less likely to be married (p = 0.008, Table 2). When stratified by gender, differences in education were maintained for both males and females, but differences in marital status were found only among males (Table S1). Cases and controls had similar smoking patterns: 55–56% were current or past smokers, 15–18% smoked less than 1 pack per day, and 38–39% smoked greater than1 pack per day (Table 2). Cases had a slightly, but not statistically, longer duration of smoking than controls (mean (SD) years of smoking: 15.7±18.4 for cases and 12.0±14.5 for controls, p = 0.20). Similarly, cases had higher cigarette pack-years (mean  = 20.5±33.6) than controls (mean  = 13.9±20.4), but again, the difference was not statistically significant (p = 0.18). The average cigarette packs per day for cases (0.64±0.88) and controls (0.61±0.69) was similar (p = 0.85). When stratified by gender, smoking status, cigarette packs per day, and the number of years of smoking did not differ between cases and controls; however, female cases had more cigarette pack-years than female controls, due to both slightly higher numbers of heavy smokers (>1 pack/day) and a longer duration of smoking (Table S1). No significant differences were seen for other measures of smoking, including 25th and 75th percentile values of ever-smokers and cigarette packs per day, years of smoking, and cigarette pack-years.

Involvement in individual sports and other physical activities (Table 2) did not differ between cases and controls, with the exception of recreational dancing (p = 0.029). Similarly, no differences were seen for the number of activities engaged in by participants. Among males, cases were more likely than controls to engage in gardening (p = 0.023) and also in a high intensity of activities when other sports were included (p = 0.054). Among females, the only difference seen was that cases were less likely to dance recreationally (p = 0.038; Table S1). Swimming showed marginally significant differences, but in different directions, when stratified by gender.

Results of the final CLR models for the entire sample (matched by age and gender) are given in Table 3. Multiple regression model results, stratified by age (adjusted by gender) and gender (adjusted by age), are in Table S2 and Table S3, respectively. Each table includes models for four exposure windows, and each model includes education, smoking, activity, and environmental variables.

Higher educational attainment was associated with a lower odds of ALS. ALS was not associated with smoking or physical activity. This applied to both categorical (e.g., smoking status) and continuous (e.g., cigarette packs per day) indicators, and to analyses stratified by age or gender. One residential factor associated with ALS was exposure to fertilizer by treating private yards/gardens (for all participants: ORs (95% CI) = 2.97 (0.81–10.9) and 2.97 (1.01–8.76) for exposure windows 1 (n = 40/29, cases/controls) and 3 (n = 31/15, cases/controls), respectively; Table 3). While this relationship (treating private yards/gardens) was marginally significant in cases, significance was maintained for three of the exposure time conditions, in analyses for younger individuals (<60 years of age), and in males. The second residential factor associated with ALS was living near industry/sewage treatment plants/farms for individuals below 60 years of age (n = 18/12, cases/controls; OR (95% CI) = 4.56 (0.75–27.7); Table S2), and for females (n = 23/16, cases/controls; OR (95% CI) = 5.18 (0.98–27.3); Table S3) for exposure window 3. The sample size did not permit further stratification or examination of interactions, such as analysis of both younger and female subjects.

Occupational exposure to pesticides was associated with ALS (n = 38/19, cases/controls; OR (95% CI) = 6.95 (1.23–39.1)) for exposures occurring over the last 30 years but not in other time frames (Table 3). Pesticide exposure was also associated with ALS in age- and gender-stratified analyses, although significance decreased (p = 0.037–0.090; Table S2 and Table S3). Occupational exposure to metals was uncommon among study participants. The most frequent metal exposures were welding fumes (n = 11/8, cases/controls), lead (n = 9/6), nickel (n = 3/5), mercury (n = 4/2), and chromates (n = 1/3). Given these small numbers, we considered exposure to any heavy metal. No association was observed with this composite measure in the present study. Likewise, no association between exposure to dust/fibers/fumes or gas was found in the final models for all participants. An association between dust/fibers/fumes or gas exposure and ALS was found among males for continuous exposures over the past 30 years (OR (95% CI) = 15.6 (1.38–177); Table S3). Women had an increased odds of ALS with exposure to occupational radiation occurring 10 to 30 years earlier (OR (95% CI) = 67.7 (2.50–999); Table S3). This was not found in men or for other exposure windows.

Case-control studies assessing environmental risk factors in ALS are pursued due to low cost, efficiency, disease latency, and tendency to affect older individuals [16],[57]. While a few prospective cohort studies have been conducted [25],[34],[62], such designs are challenging given the often long period between exposure and ALS diagnosis [63]. Limitations of any case-control study, including our own, include small sample size, selection bias, environment exposure misclassification, recall bias, and confounding.

Typical sizes for ALS case-control studies are about 100 to 400 participants [13]. Larger studies benefit from greater power and enhanced ability to evaluate small effects and interactions. Thus, another limitation of this study is the relatively small sample size; however, our current goal is to recruit 600 participants which should address this concern.

While cases and controls were matched on both age and gender, several differences are worth mention. Recruiting patients via University of Michigan resources may attract more educated control compared to case volunteers. As educational attainment may be related to working and living in an environment with fewer exposures and risk factors, our data may have selection bias [64],[65].

Exposure misclassification is also a concern in epidemiology studies. Data was collected using self-reported questionnaires that queried an exhaustive range of chemicals and potential exposures, designed to aid in subject recall. Nonetheless, individuals may be unaware of exposures and potential risks. In addition, two important challenges in estimating exposure are the long latency and recall period and the large number of toxins of interest. This suggests that a blended approach that combines questionnaires, exposure modeling (e.g., using residence information to evaluate past air pollution exposure), biological measurements (e.g., bone or blood lead measurements), and possibly environmental monitoring data (e.g., water quality measurements) could provide greater specificity of exposures and accuracy.

Recall bias is also a limitation when using questionnaires, especially when seeking long term or historical information regarding exposure as these and other factors of possible significance may have occurred many years before ALS diagnosis and study enrollment, and thus participants may have difficulty remembering potential data. We used self-reported exposure, several time periods, and considered exposures up to 30 years earlier. Future studies are needed to confirm exposures using biomarkers.

Risk factors such as educational attainment and smoking status may act as confounders or effect modifiers that alter results of statistical models. As previously noted, controls were more likely to have a higher educational attainment, which may affect exposure. Including education variables in the regression models should help adjust for potential differences, but does not eliminate the potential for confounding; however, we obtained comparable results when analyses were stratified by educational attainment, suggesting this was not an issue (Table S1).