**5.1 Safety data collection**

According to Section 3, safety incident data contains leading indicators (e.g. near misses) and lagging indicators (injures, illnesses, fatalities). Near-miss and incident reporting programs have been promoted and developed across high-risk industries [49]. OSHA requires employers to report all work-related fatalities and severe injuries according to OSHA Regulations (Standards—29 CFR 1904):


Besides the OSHA recordkeeping, many of the current companies in the highhazard industries use daily reporting applications (apps) to collect the safety data from their project sites. However, those data are mainly used internally for company future development. **Figure 2** shows examples of safety data collected by the Bureau of Labor Statistics (BLS) and safety apps.

**73**

*Industrial Safety Management Using Innovative and Proactive Strategies*

According to OSHA, the incidence rates represent the number of injuries and

*Safety data collection methods (image source: https://www.bls.gov/iif/oshsum.htm and https://conappguru.*

(#of cases or days year ×200,000 ) Incident rate =

Where *N* = number of injuries and illnesses; EH = total hours worked by all employees during the calendar year; 200,000 = base for 100 equivalent full-time

The US BLS Injuries, Illnesses, and Fatalities (IIF) program produces a wide range of information about workplace injuries and illnesses. These data are collected and reported annually through the Survey of Occupational Injuries and Illnesses (SOII) and the Census of Fatal Occupational Injuries (CFOI). **Table 2** is the latest

An accident cost usually includes direct and indirect costs. The biggest difference is if the costs can/cannot be directly attributed to the incident. The National Council on Compensation Insurance (NCCI)'s EMR is a metric to calculate workers'

Claims in \$ Loss ratio =

( )

( )

Premiums payments in \$ (2)

workers (working 40 hours per week, 50 weeks per year).

industry incidence rates (OSHA recordable case rates) from BLS.

Total employee hours year *per*

*per*

(1)

illnesses per 100 full-time workers and are calculated as (*N*/EH) × 200,000:

*DOI: http://dx.doi.org/10.5772/intechopen.93797*

**5.2 Safety data analysis**

*com/apps/apps-for-construction-safety-2016/).*

**Figure 2.**

*5.2.1 EMR (cost of accidents)*

compensation insurance premiums.

EMR is calculated by trends in the loss ratio:

*Industrial Safety Management Using Innovative and Proactive Strategies DOI: http://dx.doi.org/10.5772/intechopen.93797*

#### **Figure 2.**

*Concepts, Applications and Emerging Opportunities in Industrial Engineering*

application of hazard energy within occupational safety [48].

**5. Safety data collection, analysis, and sharing**

then investigate the accident.

**5.1 Safety data collection**

24 hours.

Because of the importance of hazard recognition, employers adopt several methods to improve hazard recognition levels. One of these methods is the retrospective hazard recognition method which is based on deducing or extrapolating knowledge gained and lessons learned from past safety incidents (i.e. accidents data) to new situations and projects [46, 47]. Despite these significant advancements, there is still a dearth of research that investigates the scientific extension and practical

A strong system of safety data collection, analysis, and sharing will assist the industry to understand the root causes of an event, explore existing and potential hazards, and continuously improving existing safety programs. Different countries and industries have conducted multiple reporting systems to collect, analyze, and share information with the public. For example, HSE has collected data on fatal injuries, nonfatal injuries, and ill health through the Labour Force Survey (LFS). The nonfatal injury and ill health estimates from the LFS are based on averages over 3 years. The fatal injuries data are collected based on RIDDOR (the Reporting of Injuries, Diseases, and Dangerous Occurrences Regulations) reports. In the United States, the Occupational Safety and Health Administration (OSHA) inspects the workplace to ensure compliance with minimum safety standards. If OSHA compliance officers find any violations on a site, they may issue a citation and a penalty. A company that had more than 10 employees at any time during the last calendar year must keep OSHA injury and illness records. Even if an employer is not required to keep injury and illness records, they are still required to report to OSHA within 8 hours any workplace incidents that result in death or the hospitalization of three or more employees. If there is a serious accident at a job site in which three or more workers are hospitalized or someone is killed, OSHA must be notified. OSHA will

According to Section 3, safety incident data contains leading indicators (e.g. near misses) and lagging indicators (injures, illnesses, fatalities). Near-miss and incident reporting programs have been promoted and developed across high-risk industries [49]. OSHA requires employers to report all work-related fatalities and severe

• All employers are required to notify OSHA when an employee is killed on the job or suffers a work-related hospitalization, amputation, or loss of an eye.

• An in-patient hospitalization, amputation, or eye loss must be reported within

Besides the OSHA recordkeeping, many of the current companies in the highhazard industries use daily reporting applications (apps) to collect the safety data from their project sites. However, those data are mainly used internally for company future development. **Figure 2** shows examples of safety data collected by the Bureau

injuries according to OSHA Regulations (Standards—29 CFR 1904):

• A fatality must be reported within 8 hours.

of Labor Statistics (BLS) and safety apps.

**72**

*Safety data collection methods (image source: https://www.bls.gov/iif/oshsum.htm and https://conappguru. com/apps/apps-for-construction-safety-2016/).*

#### **5.2 Safety data analysis**

According to OSHA, the incidence rates represent the number of injuries and illnesses per 100 full-time workers and are calculated as (*N*/EH) × 200,000:

$$\text{Incident rate} = \frac{\text{(\text{\textbullet of cases or days per year})} \times 200,000}{\text{Total employees hours per year}} \tag{1}$$

Where *N* = number of injuries and illnesses; EH = total hours worked by all employees during the calendar year; 200,000 = base for 100 equivalent full-time workers (working 40 hours per week, 50 weeks per year).

The US BLS Injuries, Illnesses, and Fatalities (IIF) program produces a wide range of information about workplace injuries and illnesses. These data are collected and reported annually through the Survey of Occupational Injuries and Illnesses (SOII) and the Census of Fatal Occupational Injuries (CFOI). **Table 2** is the latest industry incidence rates (OSHA recordable case rates) from BLS.

#### *5.2.1 EMR (cost of accidents)*

An accident cost usually includes direct and indirect costs. The biggest difference is if the costs can/cannot be directly attributed to the incident. The National Council on Compensation Insurance (NCCI)'s EMR is a metric to calculate workers' compensation insurance premiums.

EMR is calculated by trends in the loss ratio:

$$\text{Loss ratio} = \frac{\text{Claims (in s)}}{\text{Premiums payments} \left(\text{in s}\right)} \tag{2}$$


#### *Concepts, Applications and Emerging Opportunities in Industrial Engineering*

#### **Table 2.**

*Incidence rates of nonfatal occupational injuries and illnesses by selected industry and case types, private industry, 2017-2018 [50].*

The average EMR is 1.0. If a company's EMR is above 1.0, the company is considered riskier than most. For example, if a company has an EMR of 1.3, that means the insurance premiums could be up to 30% higher than a company with an EMR of 1.0. On the other hand, if a company has an EMR below 1.0, the company is considered safer than most which could receive a lower premium.

To better understand the collected incident data, many statistical modeling methods were used to identify the impact factors of incidents. The following two case studies introduced how statistical modeling helps with analyzing safety leading indicators and lagging indicators*.*

#### *5.2.1.1 Case study #1 (using binary logit regression)*

In this case study, approximately 2300 reported incidents at an active steel manufacturing facility in the U.S. between January of 2010 and August of 2016 were input into statistical predictive models [44]. The objective of this research is

**75**

**Figure 3.**

*Industrial Safety Management Using Innovative and Proactive Strategies*

to identify specific variables using statistical models that increase the probability of an unsafe event or condition within a steel manufacturing facility [44]. Due to the organization and metrics recorded for the steel manufacturing safety incident database analyzed in this research, a statistical prediction model - Binary Logit Model was selected for data analysis. The probability denoted Pr(Y), is assumed to be determined by a set of independent variables (X1, X2, …, Xj), and a corresponding set of parameters (β0, β1, β2, …, βj). The dependent (Response) variables include OSHA Recordable, Lost time, First Aid, Property Damage, Environmental Incident, and Fire. The independent (predictor) variables describe the incident occurred situations [44]. Variables can be divided into six categories: summer indicator (June, July, or August), task performed indicators (operating or driving), moving equipment indicators (crane, truck, forklift, or trailer), mobile equipment indicator, location indicators (roll shop, coil yard/ disposition, cut to length shop, west gate, water system, or melt shop), and preliminary cause indicators (defective equipment or personal responsibility) [44]. Findings from the regression analysis suggest that a positive correlation exists between incidents and summer months. One possible explanation is that employees have higher possibility to be distracted and fatigued due to high temperature. Results also suggest that injuries have a positive correlation with pedestrian employees near pieces of moving equipment [44]. Mobile equipment including trucks, forklifts, and truck and trailer combinations have a positive correlation

This study analyzes OSHA inspected fatalities data in the past 5 years from June

2014 to Aug 2018 with a total of 4769 accident records. Text mining techniques were deployed in this study for hazard report extraction [36]. **Figure 3** shows the

The incident description variables were processed using the R package 'openNLP' [51]. This package allows users to clean text data and perform machinelearning-based entity extractions. An energy source recognition method was then used to categorize the data into 10 energy source groups. Several corresponding key words were identified based on the energy source groups (**Table 3**). For example, "Worker died in fall from ladder" should be classified to Gravity due to the presence of the key term "fall". The findings show that gravity, motion, mechanical, and electrical related incidents have the largest percentage rate (**Figure 4**). This presented data analysis method can help with predicting future events, preventing reoccurrence of similar accidents, making scientific risk control plans, and incorporating

*Research framework of incident analysis using text mining and geospatial mapping [36].*

*DOI: http://dx.doi.org/10.5772/intechopen.93797*

with an incident [44].

research framework.

*5.2.1.2 Case study #2 (using text mining)*

hazard control measures into work tasks.

*Industrial Safety Management Using Innovative and Proactive Strategies DOI: http://dx.doi.org/10.5772/intechopen.93797*

*Concepts, Applications and Emerging Opportunities in Industrial Engineering*

**Industry Total recordable cases Cases with days away from** 

**Private industry** 2.8 2.8 0.9 0.9 Agriculture, forestry, fishing, and hunting 5.0 5.3 1.7 1.7 Mining, quarrying, and oil and gas extraction 1.5 1.4 0.7 0.6 Construction 3.1 3.0 1.2 1.2 Manufacturing 2.5 3.4 0.9 0.9 Wholesale trade 2.8 2.9 1.0 1.0 Retail trade 3.3 3.5 1.0 1.1 Transportation and warehousing 4.6 4.5 2.0 2.1 Utilities 2.0 1.9 0.7 0.7 Information 1.3 1.3 0.6 0.6 Finance and insurance 0.5 0.5 0.1 0.1 Real estate and rental and leasing 2.4 2.3 1.0 0.8

Management of companies and enterprises 0.9 0.8 0.2 0.2

Educational services 1.9 1.9 0.5 0.6 Health care and social assistance 4.1 3.9 1.1 1.1 Arts, entertainment, and recreation 4.2 4.1 1.2 1.1 Accommodation and food services 3.2 3.1 0.9 0.9 Other services (except public administration) 2.1 2.2 0.7 0.8

**work**

**2017 2018 2017 2018**

0.8 0.8 0.2 0.2

2.2 2.3 0.9 0.9

The average EMR is 1.0. If a company's EMR is above 1.0, the company is considered riskier than most. For example, if a company has an EMR of 1.3, that means the insurance premiums could be up to 30% higher than a company with an EMR of 1.0. On the other hand, if a company has an EMR below 1.0, the company is

*Incidence rates of nonfatal occupational injuries and illnesses by selected industry and case types, private* 

To better understand the collected incident data, many statistical modeling methods were used to identify the impact factors of incidents. The following two case studies introduced how statistical modeling helps with analyzing safety leading

In this case study, approximately 2300 reported incidents at an active steel manufacturing facility in the U.S. between January of 2010 and August of 2016 were input into statistical predictive models [44]. The objective of this research is

considered safer than most which could receive a lower premium.

indicators and lagging indicators*.*

Professional, scientific, and technical

Administrative and support and waste management and remediation services

services

**Table 2.**

*industry, 2017-2018 [50].*

*5.2.1.1 Case study #1 (using binary logit regression)*

**74**

to identify specific variables using statistical models that increase the probability of an unsafe event or condition within a steel manufacturing facility [44]. Due to the organization and metrics recorded for the steel manufacturing safety incident database analyzed in this research, a statistical prediction model - Binary Logit Model was selected for data analysis. The probability denoted Pr(Y), is assumed to be determined by a set of independent variables (X1, X2, …, Xj), and a corresponding set of parameters (β0, β1, β2, …, βj). The dependent (Response) variables include OSHA Recordable, Lost time, First Aid, Property Damage, Environmental Incident, and Fire. The independent (predictor) variables describe the incident occurred situations [44]. Variables can be divided into six categories: summer indicator (June, July, or August), task performed indicators (operating or driving), moving equipment indicators (crane, truck, forklift, or trailer), mobile equipment indicator, location indicators (roll shop, coil yard/ disposition, cut to length shop, west gate, water system, or melt shop), and preliminary cause indicators (defective equipment or personal responsibility) [44]. Findings from the regression analysis suggest that a positive correlation exists between incidents and summer months. One possible explanation is that employees have higher possibility to be distracted and fatigued due to high temperature. Results also suggest that injuries have a positive correlation with pedestrian employees near pieces of moving equipment [44]. Mobile equipment including trucks, forklifts, and truck and trailer combinations have a positive correlation with an incident [44].

### *5.2.1.2 Case study #2 (using text mining)*

This study analyzes OSHA inspected fatalities data in the past 5 years from June 2014 to Aug 2018 with a total of 4769 accident records. Text mining techniques were deployed in this study for hazard report extraction [36]. **Figure 3** shows the research framework.

The incident description variables were processed using the R package 'openNLP' [51]. This package allows users to clean text data and perform machinelearning-based entity extractions. An energy source recognition method was then used to categorize the data into 10 energy source groups. Several corresponding key words were identified based on the energy source groups (**Table 3**). For example, "Worker died in fall from ladder" should be classified to Gravity due to the presence of the key term "fall". The findings show that gravity, motion, mechanical, and electrical related incidents have the largest percentage rate (**Figure 4**). This presented data analysis method can help with predicting future events, preventing reoccurrence of similar accidents, making scientific risk control plans, and incorporating hazard control measures into work tasks.

#### **Figure 3.**

*Research framework of incident analysis using text mining and geospatial mapping [36].*


#### **Table 3.**

*Categorization of data into 10 energy source groups.*

**Figure 4.** *Distribution of the energy sources [36].*
