Where Will Your Next Recordable Come From? The Data You Already Have Can Tell You
National data shows injuries cluster in a handful of mechanisms. A site that finds its own clusters can act before the next claim is filed.
- By Imuetinyan Osarenren
- Nov 20, 2026
In 2024, U.S. private-industry employers reported 2.5 million nonfatal workplace injuries and illnesses, the lowest count since the Bureau of Labor Statistics (BLS) began the series in 2003 [1]. The National Safety Council (NSC) still estimates 4,337 preventable work deaths and $181.4 billion in societal costs for the year [2].
Most of that decline came from illnesses, which fell 26% as respiratory cases dropped. Injury cases held near 2.34 million, about where they stood in 2022 and 2023 (Figure 1) [1]. Most of us can recite our total recordable incident rate (TRIR). Fewer can say where next month's injury is most likely to come from.
Figure 1: U.S. private-industry injury and illness cases, 2015-2024
The Scoreboard Lags
TRIR and lost-time rates are the language of regulators, insurers and clients, and they belong on the dashboard. As steering tools, they arrive late. By the time the rate moves, the injury has happened and the claim is filed. BLS reports a median of eight days away from work for days-away cases and 15 days for job-transfer or restriction cases [1].
Leading indicators help, but raw counts mislead. Ten near-miss reports this month against six last month could mean a healthier reporting culture or more hazards. Two checks help. Divide by hours worked, so a busy month does not look like a dangerous one. Then plot the monthly rate on a control chart (a u-chart suits counts per unit of exposure). A change inside the control limits is likely noise. A change outside them deserves a look at who is reporting, and where.
Injuries Cluster
Start with the national pattern. Across 2023 and 2024 combined, three event categories made up about 85% of the roughly 3 million private-industry cases with days away, restriction or transfer (DART): overexertion, repetitive motion and bodily conditions (32%), contact incidents (29%), and falls, slips and trips (24%) (Figure 2) [1].
Figure 2: DART cases by event, 2023-2024
Rates also differ widely by industry. In 2024 they ran from 0.3 cases per 100 full-time equivalent workers in finance and insurance to 4.4 in transportation and warehousing, against 2.3 for all private industry (Figure 3) [3]. Benchmark your site against your own industry's published rate, not the all-industry figure.
Figure 3: Total recordable case rates by sector, 2024
National clusters describe the national workforce, and the survey behind them is an employer-reported sample that misses some workers, such as the self-employed. Yours may differ, and finding out how is the point of the method below.
A Four-Step Method
1. Structure the data. Incident reports, near misses, inspections and observations often sit in separate systems. Pull them into one table with consistent fields: date, time, location, task, equipment, injury type, body part and the worker's tenure. It is unglamorous, and nothing else works without it.
2. Find your concentration. Run a Pareto analysis on two to three years of history, sorted by task, equipment, shift and condition. In the operations I have worked in, a handful of task types accounts for most recordables, but which tasks differ by site. Two cautions. Small sites have small counts, so a cluster of three cases may be chance. And sort by days lost as well as by count, since one severe case can outweigh several minor ones.
3. Tie leading indicators to the clusters. Test which indicators move ahead of incidents where yours cluster. I watch four: hazard reporting (volume and quality), near-miss reporting rate, training completion, and a supervisor scorecard on how consistently frontline leaders close the actions assigned to them. When one softens in a cluster area, that is the early warning. Whether these predict injuries at your site is an empirical question, so check them against your own history. Watch for gaming too: a target for report counts gets you reports that meet the target.
4. Intervene on the forecast, then measure outcomes. Put controls on the highest-risk tasks first. For ergonomic injuries, score each task with a structured assessment such as the Rapid Entire Body Assessment (REBA) or the National Institute for Occupational Safety and Health (NIOSH) lifting equation, then favor engineering controls over behavior reminders. Two examples are magnetic reach poles, so workers do not overreach into racking, and raised ground-level pallet positions, to improve lifting posture. Keep TRIR and lost-time rate on the dashboard as validation. Track serious injury and fatality potential separately, because a low TRIR says little about it.
Worked Example: Sort Your Cases the Way BLS Does
Take your last 24 months of cases and sort each into the BLS event categories: overexertion, repetitive motion and bodily conditions; contact incidents; falls, slips and trips; harmful substances or environments; transportation incidents; and violent acts. Compare your shares with the national DART split for 2023-2024: about 32%, 29%, 24%, 8%, 4% and 3% in that order (Figure 2).
Suppose, with illustrative numbers, that your sort shows 45% overexertion, 25% contact and 12% falls. Overexertion runs 13 points above the national share, so manual material handling is the first place to look. Pull the hazard and near-miss reports for those tasks and check which leading indicators are slipping there, for example, hazard reporting on the night shift. You now have a targeted intervention before the next recordable, instead of a sitewide campaign that dilutes everyone's attention.
Start Monday
You need a spreadsheet and consistent data entry, not a data science team. Check whether your incident form captures task, equipment and tenure. If it does not, fix that first. Then sort two years of cases by task and event type, and re-run the sort quarterly, because clusters move once you fix them.
Before you look at your next rate, ask where your next recordable is most likely to come from, and whether your data can answer.