How Predictive Analytics is Reshaping the Battle Against Serious Injuries & Fatalities
Another Workers’ Memorial Day came and went on April 28. The event, established in 1970, is intended to raise awareness of the lives lost as a result of occupational accidents, and encourage workers and employers to work together to reduce the risk of injuries and fatalities on the job.
But in looking at the latest published statistics, there’s some doubt whether the steps organizations are taking to address workplace fatalities are really having the effect we want.
According to the U.S. Bureau of Labor Statistics (BLS)1, 5,333 workers died on the job in 2019, the highest since 2007. It’s important to note that the rate of workplace fatalities has remained the same over the past 10 years, despite continued reductions in the rate of less-severe occupational injuries.
The lack of progress in reducing workplace deaths is an indication that past tools that enabled us to lower fatality rates aren’t working as they once did, and new ideas are required. But where do we go from here?
In a 2018 whitepaper2, the Campbell Institute argued that to prevent serious injuries and fatalities (SIFs) organizations must focus on identifying the specific attributes or “precursors” that increase the potential for an event to result in a serious or fatal outcome. In other words, while any incident may occur, the conditions that lead to catastrophic failures are unique. Incidents that do not have these precursors – high risk work and the lack of critical controls – are less likely to result in serious injuries and fatalities (SIFs). As a result, to prevent fatalities, organizations would benefit most by focusing their efforts and resources on those situations or tasks where precursors are present and recoverability is limited.
The paper highlighted three situations where uncontrolled precursors may be more likely to develop:
- In organizations in which deviating from accepted standards is tacitly accepted (the “normalization of deviation”).
- In workplaces where risk isn’t consistently perceived and tolerated amongst different workers.
- In organizations where decisions to address safety risk are not based on empirical data are more likely to have latent precursors lying in wait.
How can that last point be true? Organizations are collecting more data than ever before. Global data growth was expected to reach 40,000 exabytes by the end of 20203. That’s 13 million bytes of data created for every person on Earth every single day. How could employers possibly be lacking enough data to make sound, objective decisions on operational risk?
There are a few key reasons:
- Data location: Organizational data is often siloed in separate applications without meaningful connections that prevent quick data assimilation for analysis.
- Data variety: Differences in how data is structured creates challenges in aggregating different datasets to identify relationships.
- Data analysis: Many organizations tend to rely heavily on humans to analyze that data manually4. The extensive time and costs associated with manual data analysis results in much of its potential SIF prevention value being left unexploited.
Removing the data analysis barrier
Natural Language Processing (NLP) is a branch of artificial intelligence5 designed to allow computers read, decipher and understand human language and apply it to different applications to solve problems.
Much of the safety data employers collect is unstructured – organized in no particular manner – and text heavy. That makes the data challenging for computers to analyze. NLP enables organizations to easily decode and interpret massive amounts of unstructured data to help identify where incidents are most likely to occur, or where precursors that could give rise to serious injuries and fatalities are most likely to exist.
In a 2016 study4, researchers from the University of Colorado at Boulder studied whether NLP could be leveraged to analyze historical safety data and identify the presence of specific attributes associated with injuries in the construction sector. By identifying common injury precursors and how workers normally interface with them, this data would help organizations identify where the next injury might occur, enabling them to take more targeted action to prevent the occurrence. By using NLP, the researchers were able to scan 2,200 unstructured injury reports and successfully categorize them for over 100 distinct attributes with over 95% accuracy when compared to manual approaches, while drastically reducing the effort and time involved in data analysis.
Fortunately, enterprise EHS software solutions are making massive investments to expand the breadth and depth of their business intelligence and analytics features, including NLP, to offer organizations a greater toolbox to battle SIFs.
Here are a few things to consider:
- Data Collection: Involvement of front-line workers in the collection of safety data from the field, whether incident reports or observations of identified precursors, is critical. Mobile applications with a simple, intuitive user interface that remove barriers to enable prompt reporting of safety data definitely help in that regard.
- Data Quality: While most natural language processing tools use machine learning AI to overcome barriers in data interpretation, irregularities like spelling and grammatical errors can result in certain data being lost in the uptake. Employers looking to implement a predictive analytics approach must ensure that the solution selected can continuously assess the quality of data, and identify low data quality for immediate action.
- Data Visualization: Organizations should also closely assess what data visualization options are available in commercial software solutions to enable data to be easily presented, understood and consumed in a cadence aligned to the needs of the business.
We know that the things that hurt people aren’t the same things that kill people. And while many organizations are sitting on a mountain of valuable intel to help them understand where SIF precursors may be lying in wait, few have the ability to synthesize that data efficiently to guide their SIF prevention strategy.
By leveraging predictive analytics tools like NLP, organizations can now extract greater value out of their safety data to help them detect SIF precursors with greater speed and accuracy, and hopefully, ensure more people can get home to their loved ones at the end of each day. WMHS