Using Partnerships and Multiple Data Sources to Surveil Agricultural Injuries: Considerations and Recommendations

Abstract

This article describes an interprofessional collaboration between Sanford Health and North Dakota State University that strengthens agricultural injury surveillance in the upper Midwest by using multiple sources of health data and geographic information systems (GIS) technology. We provide methodological insights and considerations for using and combining facility-level trauma registry (FLTR) data, national data sets, and GIS to identify areas with disproportionate agricultural injury prevalence. Additionally, we discuss the benefits of FLTR data, how and why it is collected, the data it contains, and how it can be combined with national datasets to fill-in surveillance gaps. Lastly, we offer recommendations for building cross-institutional and interprofessional partnerships.

Introduction

Agriculture continues to rank among the most hazardous industries worldwide, and high rates of occupational fatalities, injuries, and illnesses are observed throughout the agricultural injury literature and U.S. government reports [1–5]. However, these reported rates underrepresent agricultural injury among agricultural populations. BLS surveillance of non-fatal injuries through the Survey of Occupational Injuries and Illnesses (SOII) excludes self-employed farmers and family members, and workers on farms with fewer than 11 employees. Additionally, the SOII undercounts occupational agricultural injuries and illnesses by an estimated 78% [6].

A significant barrier to the prevention of agricultural injury is the lack of a single surveillance system with comprehensive and detailed information about these incidents. The large number of agricultural operations, smaller sized operations, the dispersion of farms across the United States, and the diversity of farm and ranching populations present challenges for researchers to collect data, identify risk factors and develop tailored interventions based on the needs of agricultural communities [7]. Additionally, agricultural injury surveillance lacks federal-level support, funding, and dissemination [7,8].

This lack of surveillance is detrimental to the examination of agricultural injury in the upper Midwest, a region that includes North Dakota (ND) and South Dakota (SD). One reason for this critical lack of knowledge is that ND and SD do not participate in the SOII conducted by the BLS due to insufficient sample sizes [9]. Most data about agricultural injury in ND come from media reports and self-report surveys sent to self-employed agricultural workers by The Central States Center for Agricultural Safety and Health (CS-CASH). The most comprehensive examination of agricultural injury in North Dakota to date was conducted by North Dakota State University (NDSU) Extension in 1982 who conducted The North Dakota Farm Accident Survey that examined agricultural injuries between October 1979 and September 1980 [10]. Therefore, current, actionable and detailed information about injury characteristic, mechanisms of injury and risk factors in the upper Midwest is scarce.

In 2015, the National Institute for Occupational Safety and Health (NIOSH) developed a model to gather quality agricultural injury data that that relies upon cross-organizational collaboration [7]. Among their recommendations, NIOSH suggested exploring data managed by other agencies, including electronic health records, maximizing regional data, develop and evaluate methodologies for using multiple data systems and using data for prevention efforts [7]. Building upon these directives, an injury Surveillance Working Group (SWG) concluded that the success of obtaining and presenting more comprehensive information on the extent, distribution, and characteristics of injuries and exposures is dependent upon “the actions of researchers and stakeholders from across disciplines” [7]. They emphasize the necessity of partnerships and the development of agricultural injury databases that are regionally-tailored to uncover risk factors and injury characteristics that can ultimately contribute to agricultural injury interventions [7].

This article discusses an interprofessional collaboration between Sanford Research, Sanford Health, NDSU, and NDSU Extension. We combined data from multiple sources, including facility-level trauma registry (FLTR) data and Census of Agriculture (COA) data, and geographic information systems (GIS) technology to fill surveillance gaps that uncover common sources and mechanisms of agricultural injury, risk factors, and geographic hotspots of injuries in the upper Midwest. Our partnership combines public health, data sciences, geography, adult and community education, research methodology, medical knowledge, and agricultural safety and practices. We provide first-hand knowledge and considerations regarding interprofessional collaborations, using multiple datasets, and the benefits of GIS to fill surveillance gaps. Our insights are relevant for others seeking to learn more about agricultural injury data, statistical and GIS methodologies and/or developing partnerships to improve agricultural health and safety in their communities.

Situating our collaboration

In 2020, Sanford Research, Sanford Health, NDSU, and NDSU Extension partnered to advance knowledge and research by using transdisciplinary and epidemiological approaches to examine multiple sources of health and agricultural injury data collected by Sanford Health, Sanford AirMed and USDA NASS COA. Our goals were to: 1) identify geographic areas within the upper Midwest region where agricultural injuries disproportionately occur, 2) describe injury characteristics, 3) describe the situational context in which the agricultural injuries happen, and 4) suggest ways to prevent and/or mitigate the occurrence and severity of agricultural injuries .

Sanford Health

Headquartered in Sioux Falls, SD, Sanford Health serves more than one million patients across its 250,000 square-mile footprint. They operate 47 medical centers and employ 2,800 Sanford physicians and advanced practice providers and 170 clinical investigators and research scientists [11]. Sanford Health’s largest trauma centers are located in Fargo, ND; Sioux Falls, SD; Bismarck, ND; and Bemidji, MN. Sanford Medical Center Fargo (SMCF) is the only American College of Surgeons (ACS) verified Adult Level I trauma center between Minneapolis, Seattle, Denver and Omaha, and it provides trauma care to ND, SD, western Minnesota, and Montana. Also, SMCF is the only Pediatric Level II trauma center in ND, with the nearest pediatric trauma centers located in Minneapolis, Sioux Falls, Denver and Seattle.

Sanford USD Medical Center in Sioux Falls (SUMC) is a Level II adult and pediatric trauma center with a similar annual trauma case load. In addition to providing services to patients from SD, SMC cares for patients from Iowa, Nebraska, and Minnesota. It is the largest hospital in SD and a teaching hospital for the University of South Dakota’s Sanford School of Medicine. Sanford Medical Center – Bismarck (SMCB) is located in western ND, a region that is home to the majority of ND’s ranches. SMCB is a Level II trauma and stroke center accredited by the ACS. Patients sometimes travel from as far as Montana and Wyoming to receive care at this facility. Lastly, Sanford Bemidji Medical Center is a Level III trauma center with a 24-hour emergency department. This location has the lowest patient count of the trauma facilities, yet it cares for many of the agricultural injuries that occur on Minnesota’s western farmlands.

As the largest rural health system in the United States, Sanford Health’s providers are familiar with agricultural injury . SMCF sees about 2,000 trauma cases per year, many of which result from agricultural injuries . As part of its verification as a Level I trauma center, the center provides outreach services to the communities where it operates, services that include injury prevention and education. The trauma center physicians also consult with rural providers in instances where it may benefit the patient to stay in the facility close to the patient’s home. Due to their ongoing involvement in their community, SMCF trauma physicians observe the burden agricultural injury imposes on the injured patient, their family, and the community. Therefore, SMCF has a vested interest in helping prevent and reduce agricultural injuries .

Origins of our collaboration

Hilla Sang, PhD, the director of Sanford Research’s Research Design and Biostatistics Core and the in-house biostatistician for the SMCF trauma center, has been working with the physicians and residents on multiple studies using data from the SMCF trauma registry. Concurrently, Elizabeth Gilblom, PhD, an assistant professor of education and research methodology at NDSU, and a long-time collaborator with Sang, learned that one of her graduate students, Angela Johnson, had started working as the Farm and Ranch Safety specialist for NDSU Extension. Gilblom reached out to Sang to discuss the potential of using Sanford Health’s data sources to facilitate a collaboration. Sang approached Sheryl Sahr, MD, a trauma surgeon from SMCF with decades of experience working in rural communities, to join the collaboration. Each individual agreed to share their expertise, experience, and perspectives to help identify, describe, track, and evaluate risks and causes of agricultural injury in the upper Midwest, with the goal of mitigating and preventing its occurrence.

Using facility-level trauma registry data to investigate agricultural injuries

Facility-level trauma registry (FLTR) data are unique, highly descriptive and proprietary. Consequently, researchers are increasingly using FLTR data to investigate varied domains in trauma research. For example, FLTR data has been used to characterize the incidence, injury characteristics, and outcomes of patients admitted to Omaha, Nebraska trauma centers with farm machinery injuries [12]; evaluate the impacts of comorbidities on the prognosis for patients with blunt trauma injuries [13]; and assess how quarantine restrictions during COVID-19 impacted trauma admissions [14].

Variables from FLTRs include patient demographics, clinical data about the patient (comorbidities and vitals) and outcomes (length of stay, mortality, and discharge disposition), information about the injury (date and time of injury, geographical location, ICD codes for location of injury, mechanism, and type of injury), injury severity codes, ICD diagnosis codes, and ICD procedure codes. FLTRs also include information about patient care quality benchmarks that are important to the facility and for national quality improvement (QI) and process improvement (PI) programs. Each of the four Sanford trauma facilities has a FLTR with detailed information on trauma cases that includes information about the patient, the injury, clinical outcomes, and an open-text field detailing the cause and context of the injury.

While many of the aforementioned variables are available through the state and national trauma databases, including the ACS National Trauma Data Bank (NTDB), the detail in the FLTR is superior for several reasons. First, FLTR databases are more inclusive than the NTDB and state databases. FLTR have broader inclusion criteria. The NTDB guidelines indicate that only patients with injuries included in specific ICD10 codes [15] will be included in the data bank. FLTRs includes all trauma patients seen at the facility, meaning that there are more patients in the FLTRs than the NTDB. While FLTRs also have the same inclusion criteria as the state trauma databases, they collect more variables, including facility-specific QI/PI initiatives and other miscellaneous information.

Second, FLTR databases include geographic locations of injuries smaller than a state and dates of injury or care. Data in the state and national trauma databases are de-identified according to HIPAA rules, meaning that these variables are not available in the databases. Geographic locations, including the city, county, zip code, and state in which the injury occurred and dates of medical care, are available and permitted for research with institutional approval.

Third, FLTR include a free-text or open-text field where registrars can enter information about the injury, including where the injury occurred (e.g., barn, field, fence) mechanism of injury (e.g., tractor, fall, animal, gates) and other details that are not captured by ICD codes. This free-text field is beneficial for research purposes because registrars enter detailed narrative about the injury, details that help to categorize the mechanism of injury. This information is provided by the patient, family members, first responders or individuals who were at the scene of the injury. This free-text field is not available through NTDB or state or national trauma databases.

Considerations for using FLTR data in agricultural injury research

FLTR data are collected for purposes of QI and participation in state and national programs. National and state registries have deadlines for data entry and upload, which may result in data that is sufficient for QI and external programs, but not for research. In our studies, we found that some data were missing or insufficient for research. For example, ICD codes for location, diagnosis, and procedures were available and complete, but the open-text field did not have clear information, hampering the categorization of injuries.

However, QI standards can be helpful in identifying research-related trends. One of the current QI initiatives at SMCF is identifying drug- and alcohol-related injuries. This alcohol and drug proved beneficial for our agricultural safety projects because we were able to identify that alcohol and drug use was not a common occurrence among agricultural injury patients. Subsequently, we were able to focus our interventional efforts on machine safety or pediatric agricultural injury instead of spending resources on investigating an issue that is rarely the cause of agricultural injury in our communities.

Another issue we encountered was combining FLTR data from the four trauma centers into a single dataset. Sanford Health’s footprint extends over several states and each state uses a different software for data collection. Each software uses different variable names and variable types (for example, numbers vs. text – one software uses 1 and 2 to categorize female and male, while another uses F and M), which had to be standardized when creating a single dataset for analysis. Even within the same state, data were entered differently into the same state-used software.

For example, Abbreviated Injury Scores (AIS) are codes used to classify the body region, type, and severity of injuries. AIS is structured as a 6-digist code classifying the body region and type of injury, and one digit following a decimal point that classifies the injury severity. In our study, we saw that facilities within the same state that use the same software enter the AIS in different forms. Specifically, one registry entered the whole AIS code into a single field in one registry, while another registry split the AIS code into two variables where the main, six-digit code was typed in one variable and the post-dot severity code in another. When combining data from various sources, closely examine the datasets and create a data dictionary that includes a crosswalk of the variables from the various data sources.

Additionally, like any database, FLTR are subject to the judgement of the individuals entering the data. While registrars use data dictionaries and decision-making algorithms when coding, some cases are subjective. For example, some machinery can be used for both agricultural and home maintenance activities. If clear and sufficient data is not available at the time of data entry to differentiate between agricultural work and home maintenance, registrars make judgment calls on how to code the activity and mechanism.

However, most of the trauma registrars in the Sanford footprint are local to the region and many were raised on farms or still living on farms. Due to their background and experience, the registrars essentially ‘speak the language of agriculture’ and are able to provide detailed information about agricultural injury incidents. Therefore, they are able to differentiate between a tractor and skid steer and identify the correct machinery attachment that was involved in an injury, valuable information that helps us correctly categorize injury mechanisms.

In addition to FLTR data, we used data from Sanford AirMed. AirMed, the Sanford Health air medical transport team, provides emergency air ambulance services in the upper Midwest. Operating from five air bases in the upper Midwest, Sanford AirMed provides life-saving medical care while transporting patients between facilities and from the scene of an emergency to a trauma center. Sanford AirMed data contain notes written by the AirMed team that have insights about how an individual was injured, details that can inform safety interventions. These additional details enhanced the FLTR data, providing deeper insight about the context where agricultural injuries occur.

The FLTR data contains all patients who were transported to a Sanford Trauma Center to receive medical care. If the patients survived or succumbed to their injuries while receiving care, they are listed in the FLTR. However, if an individual is found deceased at the scene of an accident by medical personnel, the individual is not entered into the FLTR because they did not receive medical care.

The last point for consideration is identifying agricultural injury in FLTR. Agricultural injury identification depends on the data collected by the FLTR and level of details available. In our collaboration, we filtered the patients from the complete FLTR data using a combination of ICD9 and ICD10 codes along with text mining the open-text field to identify potential agricultural injury patients. Once the list of potential patients was identified, we reviewed each case manually to ensure the patient experienced an agricultural injury and then we combined data from AirMed with data from the four trauma facilities, making a single dataset for analysis.

Geographic Information Systems in AI research

A GIS is a database system that can assemble, store, manipulate, and display geographically referenced data, including census tracts, shape files, geocoded injury cases, farm parcels, linking this data to points, lines and areas on a map or in a table. Geographically referenced information is also known as geospatial information, data that references a place and includes location and an attribute for the location. Attribute is the information associated with a place. For example, attributes of a point on a map that represents a hospital will include the hospital name, number of beds, and its non-profit status. GIS mapping software, like ArcGIS developed by Environmental Systems Research Institute (ESRI), provides an advantageous method to rapidly visualize information that exposes latent associations and reveals patterns that would be difficult to recognize without mapping technology. This approach allows researchers to explore the various characteristics of AI and connect the patient’s injury location and environment with their injury characteristics and outcomes, relationships that can lead to evidence-based recommendations and preventative strategies that meet the specific agricultural safety training needs of local communities.

GIS has been used in a variety of agricultural injury studies, including examining road characteristics in farm vehicle crashes [15], developing a database of agriculture, forestry, and fishing (AgFF)-related motor-vehicle crashes [17], and mapping trauma registry data in Northeast Texas to identify geographic areas of high agricultural injury risk [15]. The potential benefits of GIS analysis for agricultural injury are vast. GIS has the functionality to help researchers interrogate a variety of relationships that affect agricultural communities.

Methodologic considerations for using GIS and Multiple Data Sets

In this section, we will describe the methods we used to combine data from different sources to achieve the first goal of our collaboration, “to identify geographic areas within the upper Midwest region where agricultural injuries disproportionately occur.” To accomplish this, we combined the proprietary FLTR data from January 2010 through August 2022 from Sanford Health’s four trauma centers and Sanford AirMed with data from the 2017 USDA NASS COA. Individuals with traumatic agricultural injuries who were treated outside of Sanford Health’s four trauma centers were not included in our dataset. The total number of patients included in our database is 1,206.

We calculated the prevalence of agricultural injury in each county in R [19] and mapped the data using ArcGIS [20]. We identified areas with disproportionate agricultural injuries to uncover counties that require critical interventions. The GIS maps revealed the age groups in each county who are more frequently injured so interventions can be tailored to at risk populations.

The COA is conducted every five years and provides counts of U.S. farms, ranches, and agricultural operators and producers. The current census year for data collection is 2022, so we used the most recently available data from 2017. Data from the COA is available in county-level geography, enabling us to calculate prevalence rates of agricultural injury by county. The variables we used are: number of producers; percent of producers 65 years old and over; and number of farms with young producers, aged 35 years or less, as percent of number of farms. We focused on these two age groups throughout our study because these are the age groups identified by the COA.

Combining two datasets requires finding a common variable or variables. Both the COA and FLTR data have variables identifying the county and state, so we used these variables to combine the two datasets. Once the data were combined, we able calculated the prevalence of agricultural injury for each county.

Calculating prevalence

Prevalence (or prevalence rate) is a calculation commonly used in epidemiology to identify “the proportion of a population who have a specific characteristic in a given time period” [21]. It is calculated by dividing the “number of people in the sample with the characteristic of interest [by] the total number of people in the sample” [21]. The benefit of using prevalence, rather than absolute numbers, is in its ease of comparisons. Prevalence provides a standardized value that makes comparisons over time, or over geographies, easier to understand. Comparison matters because it shows where disproportionate amounts of injuries occur. The denominator in the prevalence formula is a specific and well-defined population, like number of agricultural producers of a certain age group. Defining the population in the denominator ensures “apples to apples” comparisons. Conversely, providing absolute counts or even regular frequencies skews the comparisons because the denominator is either too broad or non-existent. For example, a simple “five producers were injured” (absolute number) is not as meaningful as “five out of 1000 people who live in the county” (simple frequency), or as the most meaningful “five out of 15 producers over 65 years were injured in the county” (prevalence).

Mapping prevalence

Mapping in GIS means connecting information, in this case, prevalence rate for each county, with spatial data. The map in figure 1 illustrates agricultural injury prevalence by age group and county.

Figure 1:

Agricultural injury prevalence for patients who receive care for traumatic agricultural injuries between January 2010 through August 2022 by age group and county (n=1,206).

The spatial data we used was state and county shapefiles, which are files readable by GIS software, downloaded from the U.S. Census. We merged the shapefiles with the prevalence data in ArcGIS, enabling us to find geographic areas of disproportionate agricultural injury by age group. We calculated the quantiles of injury prevalence by age group to identify counties in which a disproportionate number of injuries occur. Counties in which the prevalence rate was at or above the 50^th percentile were categorized as “much worse” for the respective age group. Counties with a prevalence rate below the 50^th percentile for each age group had similar risk of injury for both age groups. We also identified counties where agricultural injuries occurred among only one of the age groups. To help with overall orientation, we added a layer illustrating the region’s interstate highways which was also downloaded from the U.S. Census.

In figure 1, purple counties indicate that only farmers 65+ were injured in that area and dark orange indicate counties where only young farmers under 35 years were injured. A few takeaways from this map are that farmers 65+ were disproportionately injured in the northeastern region of the map while young farmers under 35 years were disproportionately injured in western counties. Younger farmers in counties along I-94 in ND and I-90 in SD also experienced more injuries than older farmers.

This map can be used to explore the injury characteristics and the situational context in which agricultural injury happens, allowing us to focus our efforts on counties or areas that need more help, meaning that we can use resources efficiently. We will use maps like figure 1 in conjunction with descriptive statistics to develop a more comprehensive understanding of individuals who are at higher risk of agricultural injury, geographic areas with disproportionate agricultural injuries , patterns of agricultural injury by age and mechanism, all of which can lead to more effective prevention and healthcare delivery strategies.

Final thoughts

Through our interprofessional collaboration, we examined the problem of agricultural injury from multiple perspectives and leveraged GIS maps to assist health practitioners and systems and agricultural safety specialists to interrogate ways to improve access to trauma care and better allocate resources. With many complex factors affecting the challenge of agricultural injury surveillance, innovative and interprofessional collaborations are needed to effectively monitor, evaluate and develop appropriate intervention to lessen the burden of injuries and deaths in agricultural communities. The preliminary achievements of our collaboration [22,23] may suggest that long-term, cross-institution partnerships may be effective in building sustainable systems to help surveil agricultural injuries and develop tailored safety interventions in smaller regions that are often overlooked in national data collection.

We suggest the following ideas for readers who are looking to leverage trauma or hospital data to study agricultural injuries . If you are a safety professional and looking for agricultural injury data, contact a university that has a medical school, if one is available in your region. Finding the physician responsible for trauma or emergency care residents is especially important because many medical professionals and clinicians in training look for collaboration opportunities. They may see agricultural injuries in their own work or want to learn more about rural health, but don’t know how to approach agricultural injury from the perspective of agricultural safety and prevention. The medical school or hospital system may also have a research division and connecting with the research professionals will facilitate materials and methods for data collection and analysis. If you are a clinician or a researcher who wants to partner with an agricultural safety professional and your state has an Extension service, contacting your state’s Extension safety specialist or your county’s Extension agent are great places to start.

It is important to remember that not all medical facilities or states participate in the NTDB and each facility maintains a different set of data based on what is important to them and what is required by their verification body. So, work with the trauma registrars to identify what data and variables are available.

Lastly, remember that the facility’s Institutional Review Board (IRB) oversees the use or patient data, including trauma registry data, for research. Some facilities may require data use agreements or other contracts to allow the use of patient data with collaborators outside the facility. Reach out to your university, medical school, and facility to ensure you have the documents and approvals needed for your study.

The purpose of this paper was to share our experiences and what we learned through our collaboration. We hope that readers find some of our experiences to fit their needs or that they are inspired to create collaborations in their own communities. We also encourage others to share what they learned in their collaborations with the broader community.