Bayesian decision support for coding occupational injury data.

INTRODUCTION: Studies on autocoding injury data have found that machine learning algorithms perform well for categories that occur frequently but often struggle with rare categories. Therefore, manual coding, although resource-intensive, cannot be eliminated. We propose a Bayesian decision support system to autocode a large portion of the data, filter cases for manual review, and assist human coders by presenting them top k prediction choices and a confusion matrix of predictions from Bayesian models. METHOD: We studied the prediction performance of Single-Word (SW) and Two-Word-Sequence (TW) Naïve Bayes models on a sample of data from the 2011 Survey of Occupational Injury and Illness (SOII). We used the agreement in prediction results of SW and TW models, and various prediction strength thresholds for autocoding and filtering cases for manual review. We also studied the sensitivity of the top k predictions of the SW model, TW model, and SW-TW combination, and then compared the accuracy of the manually assigned codes to SOII data with that of the proposed system. RESULTS: The accuracy of the proposed system, assuming well-trained coders reviewing a subset of only 26% of cases flagged for review, was estimated to be comparable (86.5%) to the accuracy of the original coding of the data set (range: 73%-86.8%). Overall, the TW model had higher sensitivity than the SW model, and the accuracy of the prediction results increased when the two models agreed, and for higher prediction strength thresholds. The sensitivity of the top five predictions was 93%. CONCLUSIONS: The proposed system seems promising for coding injury data as it offers comparable accuracy and less manual coding. PRACTICAL APPLICATIONS: Accurate and timely coded occupational injury data is useful for surveillance as well as prevention activities that aim to make workplaces safer.

Use of multiple data sources for surveillance of work-related amputations in Massachusetts, comparison with official estimates and implications for national surveillance.

BACKGROUND: Accurate surveillance of work-related injuries is needed at national and state levels. We used multiple sources for surveillance of work-related amputations, compared findings with Survey of Occupational Injuries and Illnesses (SOII) estimates, and assessed generalizability to national surveillance. METHODS: Three data sources were used to enumerate work-related amputations in Massachusetts, 2007-2008. SOII eligible amputations were compared with SOII estimates. RESULTS: 787 amputations were enumerated, 52% ascertained through hospital records only, exceeding the SOII estimate (n = 210). The estimated SOII undercount was 48% (95% CI: 36-61%). Additional amputations were reported in SOII as other injuries, accounting for about half the undercount. Proportionately more SOII estimated than multisource cases were in manufacturing and fewer in smaller establishments. CONCLUSION: Multisource surveillance enhanced our ability to document work-related amputations in Massachusetts. While not feasible to implement for work-related conditions nationwide, it is useful in states. Better understanding of potential biases in SOII is needed.

Gender differences in work-related asthma: surveillance data from California, Massachusetts, Michigan, and New Jersey, 1993-2008.

OBJECTIVE: To characterize work-related asthma by gender. METHODS: We analyzed state-based sentinel surveillance data on confirmed work-related asthma cases collected from California, Massachusetts, Michigan, and New Jersey during 1993-2008. We used Chi-square and Fisher's Exact Test statistics to compare select characteristics between females and males. RESULTS: Of the 8239 confirmed work-related asthma cases, 60% were female. When compared to males with work-related asthma, females with work-related asthma were more likely to be identified through workers' compensation (14.8% versus 10.6%) and less likely to be identified through hospital data (14.2% versus 16.9%). Moreover, when compared to males, females were more likely to have work-aggravated asthma (24.4% versus 13.5%) and less likely to have new-onset asthma (48.0% versus 56.5%). Females were also more likely than males with work-related asthma to work in healthcare and social assistance (28.7% versus 5.2%), educational services (11.8% versus 4.2%), and retail trade (5.0% versus 3.9%) industries and in office and administrative support (20.0% versus 4.0%), healthcare practitioners and technical (13.4% versus 1.6%), and education training and library (6.2% versus 1.3%) occupations. Agent groups most frequently associated with work-related asthma were miscellaneous chemicals (20.3%), cleaning materials (15.3%), and indoor air pollutants (14.9%) in females and miscellaneous chemicals (15.7%), mineral and inorganic dusts (13.2%), and pyrolysis products (12.7%) in males. CONCLUSIONS: Among adults with work-related asthma, males and females differ in terms of workplace exposures, occupations, and industries. Physicians should consider these gender differences when diagnosing and treating asthma in working adults.

Promoting integrated approaches to reducing health inequities among low-income workers: applying a social ecological framework.

BACKGROUND: Nearly one of every three workers in the United States is low-income. Low-income populations have a lower life expectancy and greater rates of chronic diseases compared to those with higher incomes. Low- income workers face hazards in their workplaces as well as in their communities. Developing integrated public health programs that address these combined health hazards, especially the interaction of occupational and non-occupational risk factors, can promote greater health equity. METHODS: We apply a social-ecological perspective in considering ways to improve the health of the low-income working population through integrated health protection and health promotion programs initiated in four different settings: the worksite, state and local health departments, community health centers, and community-based organizations. RESULTS: Examples of successful approaches to developing integrated programs are presented in each of these settings. These examples illustrate several complementary venues for public health programs that consider the complex interplay between work-related and non work-related factors, that integrate health protection with health promotion and that are delivered at multiple levels to improve health for low-income workers. CONCLUSIONS: Whether at the workplace or in the community, employers, workers, labor and community advocates, in partnership with public health practitioners, can deliver comprehensive and integrated health protection and health promotion programs. Recommendations for improved research, training, and coordination among health departments, health practitioners, worksites and community organizations are proposed.

Mesothelioma and employment in massachusetts: analysis of cancer registry data 1988-2003.

BACKGROUND: Cancer registries can be used to monitor mesothelioma cases and to identify occupations and industries previously and newly associated with mesothelioma-causing asbestos exposure by using standard registry data on the "usual" occupation and industry of the case. METHODS: We used the National Institute for Occupational Safety and Health's Standardized Occupational Industry Coding Software to code 564 mesothelioma cases for occupation and 543 for industry of the 1,424 incident mesothelioma in the Massachusetts Cancer Registry from 1988 to 2003. Additionally, we coded the occupation and industry of 80,184 comparison cancer cases (35% of comparison cases in our database). These were used to compute Standardized Morbidity Odds Ratios (SMORs). RESULTS: Seventeen occupations and 11 industries had statistically significant elevated SMORs for mesothelioma. Occupations and industries historically associated with mesothelioma remained elevated in these results. However, we also found statistically significant elevated SMORs for several occupations and industries for which there was previously weak or no association such as chemical engineers, machine operators, and automobile mechanics and machine manufacturing, railroads, and the U.S. Postal Service. CONCLUSIONS: Incident cases of mesothelioma do not appear to be declining in Massachusetts, as legacy exposures to asbestos continue to produce cases in individuals involved in shipbuilding and construction. Exposures in occupations and industries not previously associated with mesothelioma also contribute cases. Cancer registries, with improved data collection, should continue to be monitored for mesothelioma cases and asbestos exposures.

Barriers to use of workers' compensation for patient care at Massachusetts community health centers.

OBJECTIVES: To examine barriers community health centers (CHCs) face in using workers' compensation insurance (WC). DATA SOURCES/STUDY SETTING: Leadership of CHCs in Massachusetts. STUDY DESIGN: We used purposeful snowball sampling of CHC leaders for in-depth exploration of reimbursement policies and practices, experiences with WC, and decisions about using WC. We quantified the prevalence of perceived barriers to using WC through a mail survey of all CHCs in Massachusetts. DATA COLLECTION/EXTRACTION METHODS: Emergent coding was used to elaborate themes and processes related to use of WC. Numbers and percentages of survey responses were calculated. PRINCIPAL FINDINGS: Few CHCs formally discourage use of WC, but underutilization emerged as a major issue: "We see an awful lot of work-related injury, and I would say that most of it doesn't go through workers' comp." Barriers include lack of familiarity with WC, uncertainty about work-relatedness, and reliance on patients to identify work-relatedness of their conditions. Reimbursement delays and denials lead patients and CHCs to absorb costs of services. CONCLUSION: Follow-up studies should fully characterize barriers to CHC use of WC and experiences in other states to guide system changes in CHCs and WC agencies. Education should target CHC staff and workers about WC.

Use of statewide electronic emergency department data for occupational injury surveillance: a feasibility study in Massachusetts.

BACKGROUND: Statewide datasets of emergency department (ED) visits may be useful for occupational injury surveillance. Using data from 12 hospitals, we evaluated two indicators of work-relatedness in reported ED data and the availability of employment information in medical charts. METHODS: Workers' compensation as payer and/or "yes" in an injury-at-work field were used to define probable work-related (PWR) injury visits in the Massachusetts ED dataset. Charts were reviewed for a sample of 1,002 PWR and 250 probable nonwork-related (PNWR) cases. RESULTS: Using chart information as the gold standard, indicators of work-relatedness had a sensitivity of 82%, specificity of 97%, and predictive value positive of 86%. Employer name was in charts for 89% of PWR and 42% of PNWR cases. Occupation was available for 34% of PWR cases. CONCLUSION: Electronic ED data are useful for state surveillance of occupational injuries. Improvements in attribution of work-relatedness and collection of available employer identifiers and occupational information would enhance its usefulness. The performance of indicators of work-relatedness in ED datasets should be examined in different states.

Cancer incidence among male Massachusetts firefighters, 1987-2003.

BACKGROUND: Firefighters are known to be exposed to recognized or probable carcinogens. Previous studies have found elevated risks of several types of cancers in firefighters. METHODS: Standardized morbidity odds ratio (SMORs) were used to evaluate the cancer risk in white, male firefighters compared to police and all other occupations in the Massachusetts Cancer Registry from 1986 to 2003. Firefighters and police were identified by text search of the usual occupation field. All other occupations included cases with identifiable usual occupations not police or firefighter. Control cancers were those not associated with firefighters in previous studies. RESULTS: Risks were moderately elevated among firefighters for colon cancer (SMOR = 1.36, 95% CI: 1.04-1.79), and brain cancer (SMOR = 1.90, 95% CI: 1.10-3.26). Weaker evidence of increased risk was observed for bladder cancer (SMOR = 1.22, 95% CI: 0.89-1.69), kidney cancer (SMOR = 1.34, 95% CI: 0.90-2.01), and Hodgkin's lymphoma (SMOR = 1.81, 95% CI: 0.72-4.53). CONCLUSIONS: These findings are compatible with previous reports, adding to the evidence that firefighters are at increased risk of a number of types of cancer.

Work-related asthma among health care workers: surveillance data from California, Massachusetts, Michigan, and New Jersey, 1993-1997.

BACKGROUND: Asthma morbidity has increased, posing a public health burden. Work-related asthma (WRA) accounts for a significant proportion of adult asthma that causes serious personal and economic consequences. METHODS: Cases were identified using physician reports and hospital discharge data, as part of four state-based surveillance systems. We used structured interviews to confirm cases and identify occupations and exposures associated with WRA. RESULTS: Health care workers (HCWs) accounted for 16% (n = 305) of the 1,879 confirmed WRA cases, but only 8% of the states' workforce. Cases primarily were employed in hospitals and were nurses. The most commonly reported exposures were cleaning products, latex, and poor air quality. CONCLUSIONS: Health care workers are at risk for work-related asthma. Health care providers need to recognize this risk of WRA, as early diagnosis will decrease the morbidity associated with WRA. Careful product purchasing and facility maintenance by health care institutions will decrease the risk.