Use of narrative analysis for comparisons of the causes of fatal accidents in three countries: New Zealand, Australia, and the United States

Testing and Validating Semi-automated Approaches to the Occupational Exposure Assessment of Polycyclic Aromatic Hydrocarbons

INTRODUCTION

When it is not possible to capture direct measures of occupational exposure or conduct biomonitoring, retrospective exposure assessment methods are often used. Among the common retrospective assessment methods, assigning exposure estimates by multiple expert rater review of detailed job descriptions is typically the most valid, but also the most time-consuming and expensive. Development of screening protocols to prioritize a subset of jobs for expert rater review can reduce the exposure assessment cost and time requirement, but there is often little data with which to evaluate different screening approaches. We used existing job-by-job exposure assessment data (assigned by consensus between multiple expert raters) from a large, population-based study of women to create and test screening algorithms for polycyclic aromatic hydrocarbons (PAHs) that would be suitable for use in other population-based studies.

METHODS

We evaluated three approaches to creating a screening algorithm: a machine-learning algorithm, a set of a priori decision rules created by experts based on features (such as keywords) found in the job description, and a hybrid algorithm incorporating both sets of criteria. All coded jobs held by mothers of infants participating in National Birth Defects Prevention Study (NBDPS) (n = 35,424) were used in developing or testing the screening algorithms. The job narrative fields considered for all approaches included job title, type of product made by the company, main activities or duties, and chemicals or substances handled. Each screening approach was evaluated against the consensus rating of two or more expert raters.

RESULTS

The machine-learning algorithm considered over 30,000 keywords and industry/occupation codes (separate and in combination). Overall, the hybrid method had a similar sensitivity (87.1%) as the expert decision rules (85.5%) but was higher than the machine-learning algorithm (67.7%). Specificity was best in the machine-learning algorithm (98.1%), compared to the expert decision rules (89.2%) and hybrid approach (89.1%). Using different probability cutoffs in the hybrid approach resulted in improvements in sensitivity (24-30%), without the loss of much specificity (7-18%).

CONCLUSION

Both expert decision rules and the machine-learning algorithm performed reasonably well in identifying the majority of jobs with potential exposure to PAHs. The hybrid screening approach demonstrated that by reviewing approximately 20% of the total jobs, it could identify 87% of all jobs exposed to PAHs; sensitivity could be further increased, albeit with a decrease in specificity, by adjusting the algorithm. The resulting screening algorithm could be applied to other population-based studies of women. The process of developing the algorithm also provides a useful illustration of the strengths and potential pitfalls of these approaches to developing exposure assessment algorithms.

Understanding vehicle crashes in work zones: Analysis of workplace health and safety data as an alternative to police-reported crash data in Queensland, Australia

Objectives: Vehicle crashes in work zones are significantly underreported in official crash datasets of many countries, including Australia. This leads to underestimations of work zone crash frequencies and limited understanding of crash causation factors. To address this important gap in the literature, this paper examines historical data from two different sources - police-reported crash data and organizational Workplace Health and Safety (WHS) records - to understand work zone crashes and their characteristics in Queensland, Australia.Methods: WHS data including text fields were cleaned and coded to match police-reported crash data variables for comparative descriptive analysis of a 45-month period. involvement of a moving vehicle that collided with another vehicle, pedestrian, object, or overturned, at a work zone accessible to public traffic.Results: There were more work zone crashes in the WHS data (N = 820) than the police-reported data (N = 128) and the WHS data offered a deeper understanding of incident causes due to the greater breadth of information available. The two data sets varied in terms of the patterns of crash type, the mixes of road users and vehicles involved, and the contributing factors that were identified, highlighting dangers of relying on single sources for understanding crash characteristics. The WHS data appear relatively consistent with the overall work zone safety literature, but their use has limitations regarding processing and reliability. Conversely, police-reported crash data can be analyzed efficiently but they suffer from underreporting and selective reporting.Conclusions: The WHS dataset is a valuable alternative to police-reported crash data for understanding vehicle crash characteristics in work zones, particularly where restrictive reporting criteria lead to inability to identify these crashes in police data. Reliability and utility of WHS data could be improved through advanced reporting systems and procedures, potentially including development of an app-based system for use on mobile electronic devices.

Text-document clustering-based cause and effect analysis methodology for steel plant incident data

The purpose of this study is to develop a text clustering-based cause and effect analysis methodology for incident data to unfold the root causes behind the incidents. A cause-effect diagram is usually prepared by using experts' knowledge which may fail to capture all the causes present at a workplace. On the other hand, the description of incidents provided by the workers in the form of incident reports is typically a rich data source and can be utilized to explore the causes and sub-causes of incidents. In this study, data were collected from an integrated steel plant. The text data were analysed using singular value decomposition (SVD) and expectation-maximization (EM) algorithm. Results suggest that text-document clustering can be used as a feasible method for exploring the hidden factors and trends from the description of incidents occurred at workplaces. The study also helped in finding out the anomaly in incident reporting.

Classifying injury narratives of large administrative databases for surveillance-A practical approach combining machine learning ensembles and human review

Injury narratives are now available real time and include useful information for injury surveillance and prevention. However, manual classification of the cause or events leading to injury found in large batches of narratives, such as workers compensation claims databases, can be prohibitive. In this study we compare the utility of four machine learning algorithms (Naïve Bayes, Single word and Bi-gram models, Support Vector Machine and Logistic Regression) for classifying narratives into Bureau of Labor Statistics Occupational Injury and Illness event leading to injury classifications for a large workers compensation database. These algorithms are known to do well classifying narrative text and are fairly easy to implement with off-the-shelf software packages such as Python. We propose human-machine learning ensemble approaches which maximize the power and accuracy of the algorithms for machine-assigned codes and allow for strategic filtering of rare, emerging or ambiguous narratives for manual review. We compare human-machine approaches based on filtering on the prediction strength of the classifier vs. agreement between algorithms. Regularized Logistic Regression (LR) was the best performing algorithm alone. Using this algorithm and filtering out the bottom 30% of predictions for manual review resulted in high accuracy (overall sensitivity/positive predictive value of 0.89) of the final machine-human coded dataset. The best pairings of algorithms included Naïve Bayes with Support Vector Machine whereby the triple ensemble NB_SW=NB_BI-GRAM=SVM had very high performance (0.93 overall sensitivity/positive predictive value and high accuracy (i.e. high sensitivity and positive predictive values)) across both large and small categories leaving 41% of the narratives for manual review. Integrating LR into this ensemble mix improved performance only slightly. For large administrative datasets we propose incorporation of methods based on human-machine pairings such as we have done here, utilizing readily-available off-the-shelf machine learning techniques and resulting in only a fraction of narratives that require manual review. Human-machine ensemble methods are likely to improve performance over total manual coding.

Construction fatality due to electrical contact in Ontario, Canada, 1997-2007

BACKGROUND

Electrical contact is a leading cause of occupational fatality in the construction industry. However, research on the factors that contribute to electricity-related fatality in construction is limited.

OBJECTIVES

To characterize, using an adapted Haddon's Matrix, the factors that contribute to electricity-related occupational fatalities in the construction industry in Ontario, Canada.

METHODS

Coroner's data on occupational electricity-related fatalities between 1997-2007 in the construction industry were acquired from the Ontario Ministry of Labour. Using an adapted Haddon's Matrix, we characterized worker, agent, and environmental characteristics of electricity-related occupational fatalities in the province through a narrative text analysis.

RESULTS

Electrical contact was responsible for 15% of all occupational fatalities among construction workers in Ontario. Factors associated with said occupational fatalities included direct contact with electrical sources, lower voltage sources, and working outdoors.

CONCLUSIONS

This study provides a profile of electricity-related occupational fatalities among construction workers in Ontario, and can be used to inform safety regulations.

A practical tool for public health surveillance: Semi-automated coding of short injury narratives from large administrative databases using Naïve Bayes algorithms

Public health surveillance programs in the U.S. are undergoing landmark changes with the availability of electronic health records and advancements in information technology. Injury narratives gathered from hospital records, workers compensation claims or national surveys can be very useful for identifying antecedents to injury or emerging risks. However, classifying narratives manually can become prohibitive for large datasets. The purpose of this study was to develop a human-machine system that could be relatively easily tailored to routinely and accurately classify injury narratives from large administrative databases such as workers compensation. We used a semi-automated approach based on two Naïve Bayesian algorithms to classify 15,000 workers compensation narratives into two-digit Bureau of Labor Statistics (BLS) event (leading to injury) codes. Narratives were filtered out for manual review if the algorithms disagreed or made weak predictions. This approach resulted in an overall accuracy of 87%, with consistently high positive predictive values across all two-digit BLS event categories including the very small categories (e.g., exposure to noise, needle sticks). The Naïve Bayes algorithms were able to identify and accurately machine code most narratives leaving only 32% (4853) for manual review. This strategy substantially reduces the need for resources compared with manual review alone.

Children at risk: A comparison of child pedestrian traffic collisions in Santiago, Chile, and Seoul, South Korea

OBJECTIVE

We examine and compare pedestrian-vehicle collisions and injury outcomes involving school-age children between 5 and 18 years of age in the capital cities of Santiago, Chile, and Seoul, South Korea.

METHODS

We conduct descriptive analysis of the child pedestrian-vehicle collision (P-VC) data (904 collisions for Santiago and 3,505 for Seoul) reported by the police between 2010 and 2011. We also statistically analyze factors associated with child P-VCs, by both incident severity and age group, using 3 regression models: negative binomial, probit, and spatial lag models.

RESULTS

Descriptive statistics suggest that child pedestrians in Seoul have a higher risk of being involved in traffic crashes than their counterparts in Santiago. However, in Seoul a greater proportion of children are unharmed as a result of these incidents, whereas more child pedestrians are killed in Santiago. Younger children in Seoul suffer more injuries from P-VCs than in Santiago. The majority of P-VCs in both cities tend to occur in the afternoon and evening, at intersections in Santiago and at midblock locations in Seoul. Our model results suggest that the resident population of children is positively associated with P-VCs in both cities, and school concentrations apparently increase P-VC risk among older children in Santiago. Bus stops are associated with higher P-VCs in Seoul, and subway stations relate to higher P-VCs among older children in Santiago. Zone-level land use mix was negatively related to child P-VCs in Seoul but not in Santiago. Arterial roads are associated with fewer P-VCs, especially for younger children in both cities. A share of collector roads is associated with increased P-VCs in Seoul but fewer P-VCs in Santiago. Hilliness is related to fewer P-VCs in both cities. Differences in these model results for Santiago and Seoul warrant additional analysis, as do the differences in results across model type (negative binomial versus spatial lag models).

CONCLUSIONS

To reduce child P-VCs, this study suggests the need to assess subway station and bus stop area conditions in Santiago and Seoul, respectively; areas with high density of schools in Santiago; areas with greater concentrations of children in both cities; and collector roads in Seoul.

Etiology of drug abuse: a narrative analysis

Introduction and Aim. Further gains in the prevention of drug abuse disorders require in-depth and holistic understanding of the risk factors of addiction from different perspectives. Lay persons and experts have different concepts of risk which could complement each other. The purpose of this study was to elaborate drug abuse risk factors through the story of individuals who had become drug dependent. Design and Methods. In this qualitative research, 33 individuals attending treatment centres for drug abuse were interviewed about the story of their addiction in Kerman, Iran. Interview questions were around the story of the participants. Results. All participants were male and in the age range of 18–40 years. Narrative analysis identified five themes as the main risk factors: family factors, peer pressure, the effect of gateway drugs (especially waterpipe), individual characteristics, and the community factors. More emphasis was placed upon the role of family factors, peer influence, and gateway effect. Discussion and Conclusion. This study elicited information from drug dependent subjects regarding the risk factors of drug abuse. According to drug dependent individuals' views, more attention should be devoted to family and peer influences by policy makers, in developing culture-based preventive strategies.

Gaining insight into the prevention of maternal death using narrative analysis: an experience from kerman, iran

Reduction in maternal mortality requires an in-depth knowledge of the causes of death. This study was conducted to explore the circumstances and events leading to maternal mortality through a holistic approach. Using narrative text analysis, all documents related to maternal deaths occurred from 2007 to 2011 in Kerman province/Iran were reviewed thoroughly by an expert panel. A 93-item chart abstraction instrument was developed according to the expert panel and literature. The instrument consisted of demographic and pregnancy related variables, underlying and contributing causes of death, and type of delays regarding public health aspects, medical and system performance issues. A total of 64 maternal deaths were examined. One third of deaths occurred in women less than 18 or higher than 35 years. Nearly 95% of them lived in a low or mid socioeconomic status. In half of the cases, inappropriate or nonuse of contraceptives was seen. Delay in the provision of any adequate treatment after arrival at the health facility was seen in 59% of cases. The most common medical causes of death were preeclampsia/eclampsia (15.6%), postpartum hemorrhage (12.5%) and deep phlebothrombosis (10.9%), respectively. Negligence was accounted for 95% of maternal deaths. To overcome the root causes of maternal death, more emphasis should be devoted to system failures and patient safety rather than the underlying causes of death and medical issues solely.

A comparison of fatal occupational injury event characteristics from the Census of Fatal Occupational Injuries and the Vital Statistics Mortality System

OBJECTIVES

The aim of this study was to examine utility of appending International Classification of Diseases (ICD) codes from Vital Statistics Mortality (VSM) data to Bureau of Labor Statistics (BLS) Census of Fatal Occupational Injuries (CFOI), and compare occupational event characteristics based on ICD external cause and BLS Occupational Injury and Illness Classification System (OIICS) event codes.

METHODS

We linked VSM records with CFOI records for 2003 and 2004.

RESULTS

Ninety-five percent of approximately 11,000 CFOI cases were linked to VSM cases. Linked data suggest that CFOI OIICS event and VSM ICD codes identified similar leading events. However, VSM data were generally less specific.

CONCLUSION

Lack of detail inherent in ICD codes and death narratives limits specificity of injury characteristics in VSM data. Appending ICD codes to CFOI appears to offer little value. Research comparing work- and non-work-related events may be better served by having a defined framework to crosswalk both coding schemes to facilitate comparisons.

IMPACT ON INDUSTRY

Over the last two decades, both ICD and OIICS have been used to characterize occupational injury circumstances; however, this is the first study to use linked case comparisons of the OIICS and ICD codes at a detailed level. This study confirmed that multiple source data systems provide more detail surrounding an incident than a single source data system does. Our results suggest that OIICS-coded CFOI data are a better source for occupational injury research and prevention purposes. For future comparison studies requiring ICD, it would be advantageous to have a defined framework that could easily be used to map both coding schemes (OIICS and ICD).

Correlates of undefined cause of injury coded mortality data in Australia

The objective of this research was to identify the level of detail regarding the external causes of death in Australia and ascertain problematic areas where data quality improvement efforts may be focused. The 2003 national mortality dataset of 12,591 deaths with an external cause of injury as the underlying cause of death (UCOD) or multiple cause of death (MCOD) based on ICD-10 code assignment from death certificate information was obtained. Logistic regression models were used to examine the precision of coded external cause of injury data. It was found that overall, accidents were the most poorly defined of all intent code blocks with over 30% of accidents being undefined, representing 2,314 deaths in 2003. More undefined codes were identified in MCOD data than for UCOD data. Deaths certified by doctors were more likely to use undefined codes than deaths certified by a coroner or government medical office. To improve the quality of external cause of injuries leading to or associated with death, certifiers need to be made aware of the importance of documenting all information pertaining to the cause of the injury and the intent behind the incident, either through education or more explicit instructions on the death certificate and accompanying instructional materials. It is important that researchers are aware of the validity of the data when they make interpretations as to the underlying causes of fatal injuries and causes of injury associated with deaths.

Age differences in fall-related injury hospitalisations and trauma presentations

AIM

To examine fall-related hospitalised morbidity in New South Wales (NSW) and to describe the pattern of fall-related major trauma presentations at a Level 1 Trauma Centre in NSW for younger and older fallers.

METHODS

Fall-related injuries were identified in the NSW Admitted Patients Data Collection during 1 July 1999-30 June 2008 and the trauma registry of the NSW St George Public Hospital during 1 January 2006-6 December 2008.

RESULTS

There were 434 138 hospitalisations and 862 fall-related trauma presentations. Older fallers had a higher incidence of hospitalisation, being more likely to fall on the same level during general activities at home, injuring their hip or thigh. Older fallers were also more likely to have an Injury Severity Score > 9, undergo physiotherapy and stay in hospital for >1 day than younger fallers.

CONCLUSION

Falls, particularly for older individuals, are an important cause of serious injury, representing a considerable burden in terms of hospitalised morbidity.

Identifying work related injuries: comparison of methods for interrogating text fields

Background

Work-related injuries in Australia are estimated to cost around $57.5 billion annually, however there are currently insufficient surveillance data available to support an evidence-based public health response. Emergency departments (ED) in Australia are a potential source of information on work-related injuries though most ED's do not have an 'Activity Code' to identify work-related cases with information about the presenting problem recorded in a short free text field. This study compared methods for interrogating text fields for identifying work-related injuries presenting at emergency departments to inform approaches to surveillance of work-related injury.

Methods

Three approaches were used to interrogate an injury description text field to classify cases as work-related: keyword search, index search, and content analytic text mining. Sensitivity and specificity were examined by comparing cases flagged by each approach to cases coded with an Activity code during triage. Methods to improve the sensitivity and/or specificity of each approach were explored by adjusting the classification techniques within each broad approach.

Results

The basic keyword search detected 58% of cases (Specificity 0.99), an index search detected 62% of cases (Specificity 0.87), and the content analytic text mining (using adjusted probabilities) approach detected 77% of cases (Specificity 0.95).

Conclusions

The findings of this study provide strong support for continued development of text searching methods to obtain information from routine emergency department data, to improve the capacity for comprehensive injury surveillance.

Examination of triage nurse text narratives to identify sports injury cases in emergency department presentations

Narrative text can be a useful means of identifying injury in routine data collections. An analysis of data from a near real-time emergency department surveillance system (NREDSS) in New South Wales (NSW, Australia) was conducted to determine if sports injuries can be identified from routine narrative text recorded in emergency departments. Around one-third of all emergency department (ED) presentations during 1 September 2003 to 15 February 2007 were identified as injury-related. Narrative text searching of triage nursing assessments using keywords identified between 282 (i.e. football) and 26,944 (i.e. play) potential sports injury presentations depending on the selected sports-related keyword used. Routine narrative text descriptions from triage nurse assessments show promise for the identification of sports injury presentations to EDs. Further work is required regarding in-depth assessment of case detection capabilities and the likelihood of improving the quality of narrative text recorded.

The use of narrative text for injury surveillance research: a systematic review

OBJECTIVE

To summarise the extent to which narrative text fields in administrative health data are used to gather information about the event resulting in presentation to a health care provider for treatment of an injury, and to highlight best practise approaches to conducting narrative text interrogation for injury surveillance purposes.

DESIGN

Systematic review.

DATA SOURCES

Electronic databases searched included CINAHL, Google Scholar, Medline, Proquest, PubMed and PubMed Central. Snowballing strategies were employed by searching the bibliographies of retrieved references to identify relevant associated articles.

SELECTION CRITERIA

Papers were selected if the study used a health-related database and if the study objectives were to a) use text field to identify injury cases or use text fields to extract additional information on injury circumstances not available from coded data or b) use text fields to assess accuracy of coded data fields for injury-related cases or c) describe methods/approaches for extracting injury information from text fields.

METHODS

The papers identified through the search were independently screened by two authors for inclusion, resulting in 41 papers selected for review. Due to heterogeneity between studies meta-analysis was not performed.

RESULTS

The majority of papers reviewed focused on describing injury epidemiology trends using coded data and text fields to supplement coded data (28 papers), with these studies demonstrating the value of text data for providing more specific information beyond what had been coded to enable case selection or provide circumstantial information. Caveats were expressed in terms of the consistency and completeness of recording of text information resulting in underestimates when using these data. Four coding validation papers were reviewed with these studies showing the utility of text data for validating and checking the accuracy of coded data. Seven studies (9 papers) described methods for interrogating injury text fields for systematic extraction of information, with a combination of manual and semi-automated methods used to refine and develop algorithms for extraction and classification of coded data from text. Quality assurance approaches to assessing the robustness of the methods for extracting text data was only discussed in 8 of the epidemiology papers, and 1 of the coding validation papers. All of the text interrogation methodology papers described systematic approaches to ensuring the quality of the approach.

CONCLUSIONS

Manual review and coding approaches, text search methods, and statistical tools have been utilised to extract data from narrative text and translate it into useable, detailed injury event information. These techniques can and have been applied to administrative datasets to identify specific injury types and add value to previously coded injury datasets. Only a few studies thoroughly described the methods which were used for text mining and less than half of the studies which were reviewed used/described quality assurance methods for ensuring the robustness of the approach. New techniques utilising semi-automated computerised approaches and Bayesian/clustering statistical methods offer the potential to further develop and standardise the analysis of narrative text for injury surveillance.

Hazard scenarios of truck drivers' occupational accidents on and around trucks during loading and unloading

Recent epidemiological studies have shown that there is a clear need for efforts to prevent non-traffic occupational injuries among truck drivers. The objective of the present study was to establish the hazard scenarios for truck drivers during loading/unloading through analyses of text descriptions of accident processes. Focus was on accidents that were primarily related to movement/operation on and around the truck, which are particular to truck drivers. Special emphasis was placed on falls from heights, as this was shown to be the most frequent type of accident and a major cause of fractures among truck drivers. Analyses of text descriptions of 136 accidents, including 63 cases of fall from height, collected in one company over a period of three years, revealed that: (a) the major triggering factors for falls from heights on and around the truck were stepping off the edge at height (33.3%), wrong footing (27.0%), and loss of balance/control of wagon (15.9%); (b) the major triggering factors for accidents on and around the truck in general were slip/trip (44.1%) and defect/malfunction (14.7%). The present study identified four target areas for improving prevention of occupational accidents of truck drivers in connection with movement/operation on and around trucks during loading/unloading: (1) improvement of the procedures for unloading to reduce the risk of fall from the back-hatch lift, (2) improvements of shoes and housekeeping to reduce the risk of slip/trip, (3) improvement of truck maintenance, and (4) reconciliation of views on causes of accidents between employers and truck drivers as a first step for a dialogue for improving safety in the goods-transport branch.

Completeness and accuracy of International Classification of Disease (ICD) external cause of injury codes in emergency department electronic data

The accuracy of external cause of injury codes (E codes) for work-related and non-work-related injuries in Massachusetts emergency department data were evaluated. Medical records were reviewed and coded by a nosologist with expertise in E coding for a stratified random sample of 1000 probable work-related (PWR) and 250 probable non-work-related (PNWR) cases. Cause of injury E codes were present for 98% of reviewed cases and accurate for 65% of PWR cases and 57% of PNWR cases. Place of occurrence E codes were present in less than 30% of cases. Broad cause of injury categories were accurate for about 85% of cases. Non-specific categories (not elsewhere classified, not specified) accounted for 34% of broad category misclassifications. Among specified causes, machinery injuries were misclassified most often (39/60, 65%), predominantly as cut/pierce or struck by/against. E codes reliably identify the broad mechanism of injury, but inaccuracies and incompleteness suggest areas for training of hospital admissions staff, providers, and coders.

Narrative text analysis of Kentucky tractor fatality reports

Narrative information in fatality investigation reports contains data elements not routinely analyzed with coded occupational injury surveillance data. A narrative text analysis of 69 Kentucky Fatality Assessment and Control Evaluation (FACE) agricultural tractor fatality reports from 1994 to 2004 was performed. The FACE reports were developed using the National Institute for Occupational Safety and Health, Division of Safety Research-recommended FACE report format that incorporates Haddon's matrix. Haddon's matrix separates the fatal incident into three event phases and is used to develop points of intervention based on human, organizational, and environmental factors. A multivariate logistic regression analysis for association between identified exposure variables and the outcomes of interest was undertaken. The operation of a tractor with an attached bucket, muddy terrain, and being thrown from the tractor were independent risk factors for being declared "dead at the scene". A tractor rollover and operation of a tractor on a slope were independent risk factors for being "crushed" by a tractor. Narrative text analysis of FACE fatality investigation reports is a valuable tool for the identification of additional factors contributing to tractor fatalities that can inform farm safety training, identify new areas for agricultural interventions, and support the development of new agricultural engineering strategies.

The quality of cause-of-injury data: where hospital records fall down

OBJECTIVES

This research identifies the level of specificity of cause-of-injury morbidity data in Australia. The research explores reasons for poor-quality data across different causes-of-injury areas, including a lack of clinical documentation and insufficient detail in the classification system.

METHODS

The 2002/03 hospital morbidity dataset of 593,079 injury-related hospital admissions was analysed to examine the specificity of coded external cause-of-injury data.

RESULTS

While overall specificity appeared high, the cause of 47,660 injuries was not specifically defined according to the code assigned. Only 56% of cases for whom injury was the result of an accidental fall were assigned a specific code to identify the causal detail; 19% were assigned an 'Other Specified' fall code, suggesting a lack of specific code availability; and 25% were assigned an 'Unspecified Fall' code, suggesting a lack of clinical documentation to facilitate code selection.

CONCLUSIONS

To improve the quality of injury-related hospital morbidity data, two main areas to focus resources are: 1) the development of more specific cause-of-injury codes; and 2) the provision of more detailed documentation from clinicians.

IMPLICATIONS

Clinicians and clinical coders need to work together to improve the quality of injury-related coded data through the provision of specific codes and improved clinical documentation. Accurate and comprehensive data pertaining to the circumstances surrounding hospitalised injury events will benefit injury prevention and surveillance initiatives, provide justification for resources related to injury hospitalisation, and assist in external cause research in Australia.

Sport- or leisure-related injury hospital admissions: Do we need to get more out of being struck?

The usefulness of New South Wales (NSW) hospitalisation data for the identification of prevention measures for sport- or leisure-related injury hospitalisations for one common injury mechanism, struck by/struck against injuries, is illustrated. Sport- or leisure-related hospitalisations were identified during 1999-2000 to 2003-2004 from the NSW hospitalisation data using activity and place of occurrence information. Struck by/struck against injury hospitalisations were identified using the International Classification of Disease, 10th Revision, Australian Modified (ICD-10-AM) codes W20-W23 and W50-W52. Information regarding the number of hospitalisations for basic demographic descriptors (such as age and sex), the type of injury experienced, the injury mechanism, the activity, and the place of occurrence of the injury event are available from NSW hospitalisation data. Additional information than what is currently available would be required for the identification of targeted injury prevention strategies for sport- or leisure-related struck by/struck against injuries leading to hospitalisation. Assessing the feasibility of collecting information regarding the object or agent of injury, the phase of activity at the time of the injury, the collection of narrative text and the date of injury are all recommended. These recommendations have national and international implications as ICD-10 is widely used to classify hospitalised morbidity data.

Comparison of fatalities from work related motor vehicle traffic incidents in Australia, New Zealand, and the United States

OBJECTIVE

To compare the extent and characteristics of motor vehicle traffic incidents on public roads resulting in fatal occupational injuries in Australia, New Zealand (NZ), and the United States (US).

DESIGN AND SETTING

Information came from separate data sources in Australia (1989--92), NZ (1985--98), and the US (1989--92).

METHODS

Using data systems based on vital records, distributions and rates of fatal injuries resulting from motor vehicle traffic incidents were compared for the three countries. Common inclusion criteria and occupation and industry classifications were used to maximize comparability.

RESULTS

Motor vehicle traffic incident related deaths accounted for 16% (NZ), 22% (US), and 31% (Australia) of all work related deaths during the years covered by the studies. Australia had a considerably higher crude rate (1.69 deaths/100,000 person years; 95% confidence interval (95% CI) 1.54 to 1.83) compared with both NZ (0.99; 95% CI 0.85 to 1.12) and the US (0.92; 95% CI 0.89 to 0.94). Industry distribution differences accounted for only a small proportion of this variation in rates. Case selection issues may have accounted for some of the remainder, particularly in NZ. In all three countries, male workers, older workers, and truck drivers were at higher risk.

CONCLUSIONS

Motor vehicle traffic incidents are an important cause of work related death of workers in Australia, NZ, and the US. The absolute rates appear to differ between the three countries, but most of the incident characteristics were similar. Lack of detailed data and inconsistencies between the data sets limit the extent to which more in-depth comparisons could be made.

Methods for using narrative text from injury reports to identify factors contributing to construction injury

BACKGROUND

Several methods exist for classifying injuries from written text, thereby identifying possible points of intervention. We describe an innovative method for such classification.

METHODS

Using Haddon's matrix as a framework, two independent reviewers coded text from over 4,000 injury reports into a qualitative software package to identify factors contributing to injuries sustained during construction of Denver International Airport (DIA). We compared our classification scheme with three others.

RESULTS

This process created a coded data set, an expanded version of Haddon's matrix adapted for construction injury, and coding rules for interpreting narrative text. Haddon's matrix provides a flexible theoretical framework for coding information about a spectrum of contributing factors.

CONCLUSIONS

Narrative descriptions from injury reports can provide detail on circumstances surrounding injuries and identify factors contributing to injury. Forms guiding investigators to explicitly consider human, organizational, and environmental factors could foster more complete descriptions of factors contributing to construction injury.

Comparison of unintentional fatal occupational injuries in the Republic of Korea and the United States

OBJECTIVES

To compare the profile of unintentional fatal occupational injuries in the Republic of Korea and the United States to help establish prevention strategies for Korea and to understand country specific differences in fatality risks in different industries.

METHODS

Occupational fatal injury data from 1998-2001 were collected from Korea's Occupational Safety and Health Agency's Survey of Causes of Occupational Injuries (identified by the Korea Labor Welfare Corporation) and from the United States Census of Fatal Occupational Injuries. Employment estimates were obtained in both countries. Industry coding and external cause of death coding were standardized. Descriptive analyses of injury rates and Poisson regression models to examine time trends were conducted.

RESULTS

Korea exhibited a significantly higher fatal injury rate, at least two times higher than the United States, after accounting for different employment patterns. The ordering of industries with respect to risk is the same in the two countries, with mining, agriculture/forestry/fishing, and construction being the most dangerous. Fatal injury rates are decreasing in these two countries, although at a faster rate in Korea.

CONCLUSIONS

Understanding industrial practices within different countries is critical for fully understanding country specific occupational injury statistics. However, differences in surveillance systems and employment estimation methods serve as caveats to any transnational comparison, and need to be harmonized to the fullest extent possible.

Computerized coding of injury narrative data from the National Health Interview Survey

OBJECTIVE

To investigate the accuracy of a computerized method for classifying injury narratives into external-cause-of-injury and poisoning (E-code) categories.

METHODS

This study used injury narratives and corresponding E-codes assigned by experts from the 1997 and 1998 US National Health Interview Survey (NHIS). A Fuzzy Bayesian model was used to assign injury descriptions to 13 E-code categories. Sensitivity, specificity and positive predictive value were measured by comparing the computer generated codes with E-code categories assigned by experts.

RESULTS

The computer program correctly classified 4695 (82.7%) of the 5677 injury narratives when multiple words were included as keywords in the model. The use of multiple-word predictors compared with using single words alone improved both the sensitivity and specificity of the computer generated codes. The program is capable of identifying and filtering out cases that would benefit most from manual coding. For example, the program could be used to code the narrative if the maximum probability of a category given the keywords in the narrative was at least 0.9. If the maximum probability was lower than 0.9 (which will be the case for approximately 33% of the narratives) the case would be filtered out for manual review.

CONCLUSIONS

A computer program based on Fuzzy Bayes logic is capable of accurately categorizing cause-of-injury codes from injury narratives. The capacity to filter out certain cases for manual coding improves the utility of this process.