Background: The long-term objective of this research is to provide comprehensive data on serious occupational injuries, in order to be able to identify priority areas for reduction of death and disability in the workplace. While much is known about fatal work-related injuries, there is little information regarding the incidence of injuries serious enough to require treatment in an emergency department (ED) or admission to a hospital. The goal of this research was to develop a model for comprehensive statewide surveillance of severe occupational injuries at the state level. To our knowledge, this is the first state to attempt such an endeavor that involves linking different databases. Methods: Severe injuries were defined as deaths, hospitalizations, or emergency department visits. Available sources of data in Maryland were linked together to create an incident-specific database. These data included emergency department and hospital discharge records, trauma registry records, ambulance run sheets (EMS), police crash reports, death certificates and medical examiners' autopsy reports. We established a definition of what constituted a work-related injury for each database. These definitions ranged from Workers' Compensation as expected payor source in hospital data, to industrial as injury type in ambulance run reports and crash injuries occurring in commercial vehicles. If any source used in the linkage indicated an injury as work-related we considered it to be an occupational injury case. We also conducted validation studies of each variable for its usefulness as an indicator of work-relatedness. There were two aspects to the study: (1) a passive, retrospective linkage of the previously mentioned data sets for the years 2001-2004, and (2) an active prospective study of injured trauma patients, including screening and documentation of work-relatedness, and in-depth interviews of a sample of patients who were injured while working. A chart review of a sample of ED patients, inpatients, and deceased workers with sentinel injuries was used to validate reporting of work-relatedness by the variables used from linked data. This was conducted for a sample of community hospitals and for the major trauma centers in the state. Sentinel injuries were selected based on their strong likelihood of being work-related, including falls from elevation and machinery or electrical injuries. About 43-64% of cases involving machinery, electric current and "caught between objects" were identified as occupational in the hospital data, but many fewer cases involving "struck by object or overexertion were classified as work-related. The sensitivity and specificity of each proxy work-related variable in each database were evaluated with regard to whether they identified these sentinel injuries as occupational in nature. Finally, we also sought to determine the type of industry/occupation and circumstances of the injury and the availability of this data in medical records. Results: Data linkage and sentinel identifiers increased detection of occupational injuries. Based on the linkage of the data for 2001-2004, 7.6% of all ED visits for injury were work-related versus 7.5% before linkage. For hospital admissions 5.3% were work-related compared to 4.5% before linkage and 8.0% vs. 5.1 % for Trauma registry cases. Percentage increases in counts and rates for work injuries from data linkage were as follows: 2% in ED patients, 19% in hospital discharges and 58% in trauma registry cases. Over the four year period there were an estimated 163,739 ED visits and 9,847 hospitalizations for occupational injuries. Among the 9,847 cases of work-related hospitalizations, payor source of Workers' Compensation was the identifying factor for 77%, with a further 6% identified as having an industrial place of occurrence. Using farm occurrence, only 56 additional cases were identified. Falls from 15 or more feet, identified from the EMS data, brought in an additional 5% of the cases. Finally, linkage with the trauma registry added 6%, representing the most severe of the injuries requiring hospitalization. Agreement varied by data source. While agreement of work-relatedness for each data source with the ED data occurred approximately75% of the time, much lower agreement rates were found with death certificate data, police reports, and ambulance run sheets. As expected, identification of work-related injuries due to motor vehicle crashes was relatively low. Using police reported data on drivers of commercial vehicles as the "gold" standard, 52% and 65% of the ED and inpatient data, respectively, gave an indication that the injury was work-related. For the trauma registry, there was only a 49% agreement. For death certificates, however, 89% were identified as work-related. The largest discrepancy between data sources was with the EMS variable injury mechanism fall over 15 feet as a work-related proxy, especially outside the working age group 16-64 years of age. Falls from 15 feet can capture both occupational and non-occupational injuries. Chart review of sentinel injuries. Chart review of sentinel injuries suggest that even with our data linkage, there is still considerable under-reporting of work-relatedness using available data. Of the 225 hospital charts for sentinel injuries classified as non-occupational based on linked data, chart reviews indicated that 4.3% actually were work-related (15.6% for ED injuries and 10.9% for trauma center cases). Further examination of the fall reporting by EMS seems to indicate that it may not be specific enough to indicate work-related injuries, especially in the extremes of age. Injury characteristics Age and gender had strong effects on injury risk and hospitalization risk. The highest proportion of occupational injuries occurred among those 29-32 years of age. By age 60 the proportion of work-related injuries started to decline, falling to a low of 2% by age 70. Only 887 (0.5% of total) ED visits and 85 (0.8%) hospitalized injuries occurred to persons under 16. Further analyses are ongoing to examine these injuries in younger persons. Occupational injuries tended to be more severe than other injuries resulting in hospital admission: 59% were admitted to a trauma hospital, compared with 39% among all hospitalized injuries. Inpatients were significantly older than those treated in EDs (42 vs. 36 years), and included more men (82% vs. 68%). The major injury mechanisms identified among the total work-related population were: falls (20%), motor vehicle-related incidents (6%), and hot objects or fire (3%). The other mechanisms included injuries from cutting and piercing instruments (20%), struck by an object (18%) and overexertion (14%). Workers injured in falls or motor vehicle crashes were significantly more likely to be hospitalized. Upper and lower extremity injuries accounted for more than half of all occupational injuries. Inpatients, as compared with workers treated in the ED, were more likely to have injuries to the head, thorax, and abdomen. Although the most common injuries were to the extremities, for upper extremities, those treated in the ED primarily had lacerations, while those admitted had forearm fractures. For lower extremities, ankle/foot sprains were common among those treated in the ED, while those admitted had primarily ankle/foot fractures. Interviews of injured workers Interviews conducted with injured workers hospitalized in a trauma center provided more insights into injury causes and further analyses are ongoing. Linkage of interviews with workers' compensation records found that only 53% of injured workers interviewed had a claim on file for the current injury in the state workers' compensation files, underscoring the importance of using ancillary data sources as adjuncts to Workers' Compensation files when conducting occupational injury surveillance. Discussion. We were successful in linking together a disparate group of databases, and creating an incident specific database to conduct a surveillance of serious occupational injury. Expected payor type as Workers' Compensation is commonly used as a proxy for work-related injuries in other studies but would have missed 23% of work injuries identified by data linkage. Despite our ability to link multiple data sources, chart reviews suggest our estimates of work-related injuries likely were still undercounts. In the course of this research, we have identified those datasets that provide the most accurate information, and can provide the basis for an ongoing surveillance system for occupational injury. Our planned publications will serve as a manual for Maryland and other states to enhance methods for occupational injury surveillance and to monitor trends in occupational injury. We are working with the different state agencies to determine how reporting of occupational injuries can be improved in available databases in order to provide more useful information on injury source and work-relatedness. The addition of an "injury at work" variable (as exists on death certificates) to the hospital discharge and ED databases for example would greatly improve reporting of work-related injuries. We are also exploring collaboration with the Maryland Injured Workers' Insurance Fund regarding linkage with workers' compensation data. Findings from this research are being shared with the Maryland Department of Health and Mental Hygiene, Maryland Occupational Safety and Health Administration, as well as the leaders of the state trauma care system and Office of the Chief Medical Examiner. The knowledge we have gained in this process will be useful to other states who wish to improve identification of occupational injuries.