Data on acute pesticide poisoning cases were obtained from the National Institute for Occupational Safety and Health (NIOSH)’s Sentinel Event Notification System for Occupational Risks (SENSOR)-Pesticides program and CDPR’s Pesticide Illness Surveillance Program (PISP). The SENSOR-Pesticides program has collected pesticide poisoning surveillance data from 12 states using standardized definitions and variables available since 1998 (Calvert et al. 2010). This study included data from 11 states for the following years: Arizona, 1998–2000; California, 1998–2006; Florida, 1998–2006; Iowa, 2006; Louisiana, 2000–2006; Michigan, 2000–2006; New Mexico, 2005–2006; New York, 1998–2006; Oregon, 1998–2006; Texas, 1998–2006; and Washington, 2001–2006. North Carolina, which joined SENSOR-Pesticides in 2007, was not included. Because each state removes personal identifiers from the data before submission to the Centers for Disease Control and Prevention (CDC), this study was exempt from consideration by the federal Human Subjects Review Board.

Participating surveillance programs identify cases from multiple sources, including health care providers, poison control centers, workers’ compensation claims, and state or local government agencies. They collect information on the pesticide exposure incident through investigation, interview, and medical record review. In California, on some occasions, such as large drift events, active surveillance is undertaken for further case finding by interviewing individuals living or working within the vicinity affected by the off-target drift (Barry et al. 2010). Although the SENSOR-Pesticides program focuses primarily on occupational pesticide poisoning surveillance, all of the SENSOR-Pesticides state programs except California collect data on both occupational and nonoccupational cases. In California, PISP captures both occupational and nonoccupational cases. SENSOR-Pesticides and PISP classify cases based on the strength of evidence for pesticide exposure, health effects, and the known toxicology of the pesticide and use slightly different criteria for case classification categories (Calvert et al. 2010). This study restricted the analyses to cases classified as definite, probable, possible, or suspicious by SENSOR-Pesticides and definite, probable, or possible by PISP. We also performed analyses restricted to definite and probable cases only. Because the findings from these restricted analyses were similar to those that included all four classification categories (i.e., definite, probable, possible, or suspicious), only the findings that used the four classification categories are reported here.

In this study, a drift case was defined as acute health effects in a person exposed to pesticide drift from an outdoor agricultural application. Drift exposure included any of the following pesticide exposures outside their intended area of application: a) spray, mist, fumes, or odor during application; b) volatilization, odor from a previously treated field, or migration of contaminated dust; and c) residue left by offsite movement. Our drift definition is broader than U.S. EPA’s “spray or dust drift” definition, which excludes postapplication drift caused by erosion, migration, volatility, or windblown soil particles (U.S. EPA 2001). A drift event was defined as an incident where one or more drift cases experienced drift exposure from a particular source. Both occupational and nonoccupational cases were included. An occupational case was defined as an individual exposed while at work. Among occupational cases, agricultural workers were identified using 1990 and 2002 Census Industry Codes (CICs): 1990 CICs, 010, 011, 030; 2002 CICs, 0170, 0180, 0290 (U.S. Census Bureau 1992, 2005).

Figure 1 presents the process of case selection. We selected cases if exposed to pesticides applied for agricultural use including farm, nursery, or animal production, and excluded cases exposed by ingestion, direct spray, spill, or other direct exposure. We then manually reviewed all case reports and excluded persons exposed to pesticides used for indoor applications (e.g., greenhouses, produce packing facilities), persons exposed within a treated area (e.g., pesticide applicators exposed by pesticides blown back by wind, workers working within or passing through the field being treated), and persons exposed to pesticides being mixed, loaded, or transported. Drift cases therefore represented the remaining 9% and 27% of all pesticide illness cases identified by the SENSOR-Pesticides and PISP, respectively. We also searched for duplicates from the two programs identifying California cases. Because personal identifiers were unavailable, date of exposure, age, sex, active ingredients, and county were used for comparison. A total of 60 events and 171 cases were identified by both California programs. These were counted only once and were included only in the PISP total.

Drift events and cases were analyzed by the following variables: state, year, and month of exposure, age, sex, location of exposure, health effects, illness severity, pesticide functional and chemical class, active ingredient, target of application, application equipment, detection of violations, and factors contributing to the drift incident. U.S. EPA toxicity categories ranging from toxicity I (the most toxic) to IV (the least toxic) were assigned to each product (U.S. EPA 2007). Cases exposed to multiple products were assigned to the toxicity category of the most toxic pesticide they were exposed to. Illness severity was categorized into low, moderate, and high using criteria developed by the SENSOR-Pesticides program (Calvert et al. 2010). Low severity refers to mild illnesses that generally resolve without treatment. Moderate severity refers to illnesses that are usually systemic and require medical treatment. High severity refers to life-threatening or serious health effects that may result in permanent impairment or disability. Contributing factors were retrospectively coded with available narrative descriptions. One NIOSH researcher (S.J.L.) initially coded contributing factors for all cases. Next, for SENSOR-Pesticides cases, state health department staff reviewed the codes and edited them as necessary. Any discrepancies were resolved by a second NIOSH researcher (G.M.C.). For PISP cases, relatively detailed narrative descriptions were available for all incidents. These narratives summarize investigation reports provided by county agriculture commissioners, who investigate all suspected pesticide poisoning cases reported in their county. After initial coding, the two NIOSH researchers discussed those narratives that lacked clarity to reach consensus.

Data analysis. Data analysis was performed with SAS software (version 9.1; SAS Institute Inc., Cary, NC). Descriptive statistics were used to characterize drift events and cases. Incidence rates were calculated by geographic region, year, sex, and age group. The numerator represented the total number of respective cases in 1998–2006. Denominators were generated using the Current Population Survey microdata files for the relevant years (U.S. Census Bureau 2009). For total and nonoccupational rates, the denominators were calculated by summing the annual average population estimates. A nonoccupational rate for agriculture-intensive areas was calculated by selecting the five counties in California where the largest amounts of pesticides were applied in 2008 (Fresno, Kern, Madera, Monterey, and Tulare) (CDPR 2010). For occupational rates, the denominators were calculated by summing the annual employment estimates including both “employed at work” and “employed but absent.” The denominator for agricultural workers was obtained using the same 1990 and 2002 CICs used to define agricultural worker cases (U.S. Census Bureau 1992, 2005). Moreover, in California, where data on pesticide use are available, incidence was calculated per number of agricultural applications and amount of pesticide active ingredient applied (CDPR 2009). Incidence trend over time was examined by fitting a Poisson regression model of rate on year and deriving the regression coefficient and its 95% confidence interval (CI).

Drift events were dichotomized by the size of events into small events involving < 5 cases and large events involving ≥ 5 cases. This cut-point was based on one of the criteria used by the CDPR to prioritize event investigations (CDPR 2001). Illness severity was dichotomized as low and moderate/high. Simple and multivariable logistic regressions were performed. Odds ratios (ORs) and 95% CIs were calculated.

Number and incidence of drift events and cases. From 1998 through 2006, we identified 643 events and 2,945 illness cases associated with pesticide drift from agricultural applications (Figure 1). Of these, 382 events (59%) and 791 cases (27%) were identified by SENSOR-Pesticides (excluding 60 events and 171 cases also identified by PISP), and 261 events (41%) and 2,154 cases (73%) were identified by PISP. Drift cases consisted of 53 definite (1.8%), 2,019 probable (68.6%), 823 possible (27.9%), and 50 suspicious (1.7%) cases. Among drift cases, 1,565 (53%) were nonoccupational and 1,380 (47%) were occupational. Agricultural workers accounted for 73% (n = 1,010) of the occupational cases. A total of 340 events (53%) occurred between May and August, and these involved 1,407 cases (48%).

The overall incidence rate of drift-related pesticide poisoning was 2.93 per million person-years (Table 1). The rates of nonoccupational and occupational drift-related pesticide poisoning were 1.56 and 2.89 per million persons-years, respectively. Among occupational cases, the rate was 114.3 for agricultural workers and 0.79 for all other workers. Among nonoccupational cases identified in California, the rate was 42.2 for residents in the five agriculture-intensive counties and 0.61 for residents of all other California counties (data not shown). The rate was highest in the western states for both nonoccupational and occupational cases (Table 1). In California, per 100,000 agricultural applications, 1.6 drift events and 11.8 cases were identified; per 10 million pounds applied, 1.9 events and 14.4 cases were identified (data not shown).

The total annual incidence rate ranged from 1.39 to 5.32 per million persons over the 9-year time period (Table 1). Over time, the rate of drift cases involved in large events showed the same pattern as the rate of all drift cases, showing a spike every 3 years (Figure 2). The rate of drift cases involved in small events varied within a narrow range from 0.49 to 1.11, and we found no significant rate change over this time period; however, for the five states that provided data for all 9 years, we found a significant decrease in the rate (i.e., an estimated 9% decrease per year; 95% CI, 3–15%; p = 0.004).

Men comprised 53% of all cases (Table 1). The rate by sex was similar among nonoccupational cases. For occupational cases, the rate was 1.25 times higher in male workers than in female workers but 2.89 times higher in female agricultural workers than in male agricultural workers. Among nonoccupational cases, children < 15 years of age accounted for 33% of cases with known age and showed the highest rate (1.88/million person-years; Table 1).

Responsible pesticides, application targets, and application equipment. In 430 (67%) of 643 drift events, exposure was to pesticides from a single functional class (Table 2). Insecticides were the most commonly identified (31% of events), accounting for 23% (n = 678) of all cases. Fumigants were involved in only 8% of drift events but accounted for 45% (n = 1,330) of all cases. Organophosphorus compounds were the most common pesticide chemical class involved in drift events (28%). Most cases (66%) were exposed to toxicity I (high toxicity) pesticides.

For the intended application targets, 71% of events involved applications to fruit, grain/fiber/grass, or vegetable crops (Table 2). Soil applications accounted for 9% of drift events and 45% of all cases. For application equipment, aerial applications (e.g., by airplane) were responsible for 39% of drift events, accounting for 24% of all cases. Chemigation (i.e., application via an irrigation system) or soil injectors were used in 7% of drift events and accounted for 44% of cases. All soil injector events and 95% of chemigation events involved the use of fumigants applied to soil (data not shown).

Location of exposure and health effects. Common exposure locations were private residences (44%) and farms/nurseries (37%; Table 3). More than half of cases experienced ocular (58%) or neurological (53%) symptoms or signs, and illness severity was low for most cases (92%; Table 3). Moderate/high severity illness was significantly associated with females, older age groups, and exposure to multiple active ingredients, before and after controlling for other case and pesticide characteristics (p < 0.05; Table 4). Compared with fumigants, exposures to herbicides, insecticides, or multiple classes were significantly associated with moderate/high illness. Table 5 lists 15 active ingredients most commonly found among drift cases and their distribution according to illness severity.

Size of drift events. Most drift events involved a single case (n = 387, 60%). For multiperson events, 168 events (26% of the total) involved 2–4 cases, 78 events (12%) involved 5–29 cases, and 10 events (1.5%) involved ≥ 30 cases. Table 6 provides details on the 10 largest events. Detailed investigation reports of some of these events are available elsewhere (Barry et al. 2010; CDC 2004; O’Malley et al. 2005). The occurrence of large versus small events (events with ≥ 5 vs. < 5 cases) was significantly associated with the use of fumigants (compared with insecticides) and applications to soil, small fruit crops, or leafy vegetable crops (compared with other targets; p < 0.05; Table 7).

Contributing factors to drift incidents. Of 299 drift events with information on violations of pesticide regulations, 220 (74%) had one or more violations and accounted for 2,093 cases (89% of cases with violation information; Table 8). However, not all of the observed violations may have directly contributed to the drift exposure. Factors contributing to the drift exposure were identified in 164 events, accounting for 1,544 (52%) cases. Common contributing factors identified for drift events included applicators’ carelessness near or over nontarget sites (e.g., flew over a house, did not turn off a nozzle at the end of the row), unfavorable weather conditions (e.g., high wind speed, temperature inversion), and poor communication between applicators or growers and others. Improper seal of the fumigation site (e.g., tarp tear, early removal of seal), which were identified in nine events, accounted for the largest proportion (60%) of cases with contributing factors identified.

The distance between the application and exposure site was identified in 1,428 (48%) cases (Table 8). Occupational cases accounted for 68% of cases exposed within 0.25 miles of the application site, and nonoccupational cases accounted for 73% of cases exposed > 0.25 miles away.

To our knowledge, this is the first comprehensive report of drift-related pesticide poisoning in the United States. We identified 643 events involving 2,945 illness cases associated with pesticide drift from outdoor agricultural applications during 1998–2006. Pesticide drift included pesticide spray, mist, fume, contaminated dust, volatiles, and odor that moved away from the application site during or after the application. Although the incidence for cases involved in small drift events (< 5 cases) tended to decrease over time, the overall incidence maintained a consistent pattern chiefly driven by large drift events. Large drift events were commonly associated with soil fumigations.

Occupational exposure. Occupational pesticide poisoning is estimated at 12–21 per million U.S. workers per year (Calvert et al. 2004; Council of State and Territorial Epidemiologists 2010). Compared with those estimates, our estimated incidence of 2.89 per million worker-years suggests that 14–24% of occupational pesticide poisoning may be attributed to off-target drift from agricultural applications. Our study included pesticide drift from outdoor applications only and excluded workers exposed within the application area. Our findings show that the risk of illness resulting from drift exposure is largely borne by agricultural workers, and the incidence (114.3/million worker-years) was 145 times greater than that for all other workers. Current regulations require agricultural employers to protect workers from exposure to agricultural pesticides, and pesticide product labels instruct applicators to avoid allowing contact with humans directly or through drift (U.S. EPA 2009).

Our study found that the incidence of drift-related pesticide poisoning was higher among female and younger agricultural workers and in western states. These groups were previously found to have a higher incidence of pesticide poisoning (Calvert et al. 2008). It is not known why the incidence is higher among female and younger agricultural workers, but hypotheses include that these groups are at greater risk of exposure, that they are more susceptible to pesticide toxicity, or that they are more likely to report exposure and illness or seek medical attention. However, we did not observe consistent patterns among workers in other occupations. This finding requires further research to identify the explanation. The higher incidence in the western states may suggest that workers in this region are at higher risk of drift exposure; however, it may also have resulted from better case identification in California and Washington states through their higher-staffed surveillance programs, extensive use of workers’ compensation reports in these states, and use of active surveillance for some large drift events in California.

Nonoccupational exposure. This study found that more than half of drift-related pesticide poisoning cases resulted from nonoccupational exposures and that 61% of these nonoccupational cases were exposed to fumigants. California data suggest that residents in agriculture-intensive regions have a 69 times higher risk of pesticide poisoning from drift exposure compared with other regions. This may reflect California’s use of active surveillance for some large drift events. Children had the greatest risk among nonoccupational cases. The reasons for this are not known but may be because children have higher pesticide exposures, greater susceptibility to pesticide toxicity, or because concerned parents are more likely to seek medical attention. Recently several organizations submitted a petition to the U.S. EPA asking the agency to evaluate children’s exposure to pesticide drift and adopt interim prohibitions on the use of drift-prone pesticides near homes, schools, and parks (Goldman et al. 2009).

Contributing factors. Soil fumigation was a major cause of large drift events, accounting for the largest proportion of cases. Because of the high volatility of fumigants, specific measures are required to prevent emissions after completion of the application. Given the unique drift risks posed by fumigants, U.S. EPA regulates the drift of fumigants separately from nonfumigant pesticides. The U.S. EPA recently adopted new safety requirements for soil fumigants, which took effect in early 2011 and include comprehensive measures designed to reduce the potential for direct fumigant exposures; reduce fumigant emissions; improve planning, training, and communications; and promote early detection and appropriate responses to possible future incidents (U.S. EPA 2010). Requirements for buffer zones are also strengthened. For example, fumigants that generally require a > 300 foot buffer zone are prohibited within 0.25 miles (1,320 feet) of “difficult-to-evacuate” sites (e.g., schools, daycare centers, hospitals). We found that, of the 738 fumigant-related cases with information on distance, 606 (82%) occurred > 0.25 miles from the application site, which suggests that the new buffer zone requirements, independent of other measures to increase safety, may not be sufficient to prevent drift exposure.

This study also shows the need to reinforce compliance with weather-related requirements and drift monitoring activities. Moreover, applicators should be alert and careful, especially when close to nontarget areas such as adjacent fields, houses, and roads. Applicator carelessness contributed to 79 events (48% of 164 events where contributing factors were identified), of which 56 events involved aerial applicators. Aerial application was the most frequent application method found in drift events, accounting for 249 events (39%). Drift hazards from aerial applications have been well documented (CDC 2008; Weppner et al. 2006). Applicators should use all available drift management measures and equipment to reduce drift exposure, including new validated drift reduction technologies as they become available.

Limitations. This study requires cautious interpretation especially for variables with missing data on many cases (e.g., age, violation, contributing factors, distance). This study also has several limitations. First, our findings likely underestimate the actual magnitude of drift events and cases because case identification principally relies on passive surveillance systems. Such underreporting might have allowed the totals to be appreciably influenced by a handful of California episodes in which active case finding located relatively large numbers of affected people. Pesticide-related illnesses are underreported because of individuals not seeking medical attention (because of limited access to health care or mild illness), misdiagnosis, and health care provider failure to report cases to public health authorities (Calvert et al. 2008). Data from the National Agricultural Workers Survey suggests that the pesticide poisoning rates for agricultural workers may be an order of magnitude higher than those identified by the SENSOR-Pesticides and PISP programs (Calvert et al. 2008). Second, the incidence of drift cases from agricultural applications may have been underestimated by using crude denominators of total population and employment estimates, which may also include those who are not at risk. On the other hand, the incidence for agricultural workers may have been overestimated if the denominator data undercounted undocumented workers. Third, the data may include false-positive cases because clinical findings of pesticide poisoning are nonspecific and diagnostic tests are not available or rarely performed. Fourth, when we combined data from SENSOR-Pesticides and PISP, some duplication of cases and misclassification of variables may have occurred, although we took steps to identify and resolve discrepancies. Also, SENSOR-Pesticides and PISP may differ in case detection sensitivity because the two programs use slightly different case definitions. Lastly, contributing factor information was not available for 48% of cases, either because an in-depth investigation did not occur or insufficient details were entered into the database. We often based the retrospective coding of contributing factors on limited data, which may have produced some misclassification.

These study findings suggest that the incidence of acute illness from off-target pesticide drift exposure was relatively low during 1998–2006 and that most cases presented with low-severity illness. However, the rate of poisoning from pesticide drift was 69 times higher for residents in five agriculture-intensive California counties compared with other counties, and the rate of occupationally exposed cases was 145 times greater in agricultural workers than in nonagricultural workers. These poisonings may largely be preventable through proper prevention measures and compliance with pesticide regulations. Aerial applications were the most frequent method associated with drift events, and soil fumigations were a major cause of large drift events. These findings highlight areas where interventions to reduce pesticide drift could be focused.