Analyzing Workers’ Compensation Claims and Payments Made Using Data from a Large Insurance Provider

Background: All states in the USA have established Workers’ Compensation (WC) insurance systems/programs. WC systems address key occupational safety and health concerns. This effort uses data from a large insurance provider for the years 2011–2018 to provide estimates for WC payments, stratified by the claim severity, i.e., medical only, and indemnity. Methods: Besides providing descriptive statistics, we used generalized estimating equations to analyze the association between the key injury characteristics (nature, source, and body part injured) and total WC payments made. We also provide the overall cost burden for the former. Results: Out of the total 151,959 closed claims, 83% were medical only. The mean overall WC payment per claim for the claims that resulted in a payment was $1477 (SD: $7221). Adjusted models showed that mean payments vary by claim severity. For example, among medical only claims, the mean payment was the highest for amputations ($3849; CI: $1396, $10,608), and among disability and death related claims, ruptures cost the most ($14,285; $7772, $26,255). With frequencies taken into account, the overall cost burden was however the highest for strains. Conclusions: Workplace interventions should prioritize both the costs of claims on average and the frequency.


Introduction
Occupational injury surveillance data are crucial to designing effective preventive interventions [1]. Currently, there exists no single, comprehensive national injury surveillance database, especially for nonfatal injuries. However, there are several databases that provide an opportunity for state or federal level injury surveillance. While some of these may only be demographic in nature, other contain information regarding related health outcomes and potential risk factors, such as workers compensation (WC) datasets. The latter provide information that can be used to conduct workplace hazard evaluations and suggest preventive interventions [2].
All the states within the United States (USA) have established WC insurance systems/programs that provide income protection, medical treatment, and rehabilitation to workers who get injured or ill due to their work. WC coverage is provided to all or nearly all workers for whom such coverage is required and covers over 90% of the USA wage and salary workers [3]. Importantly, because of their design and performance, WC systems address a key concern for the National Institute for Occupational Safety and Health given the implications for occupational safety and health and post-injury outcomes [4].
Note that while there is no comprehensive national surveillance system for occupational injuries and illnesses, the WC data, despite their limitations, are recognized as one of the major data sources to address the former [1]. In fact, WC claims have now long been used to conduct occupational health and safety research and identify priority preventive interventions [3]. While the primary purpose of maintaining a record for WC claims is to ensure appropriate payment to the injured worker, these records still contain crucial public health data regarding nature and source of the injury, body part injured, injury severity, lost work time, and occupation [3]. Furthermore, the WC databases are recognized to have substantial utility for the purposes of epidemiological research given their high sensitivity for case ascertainment [5].
In the year 2007, WC programs covered around 132 million workers and paid over $55 billion in benefits across the USA including over $27 billion for medical care and over $28 billion as cash benefits (for lost work time). The latter differ by the duration and severity of the impairment or disability with temporary total disability benefits being paid when the workers cannot temporality perform pre-injury work. While, in case of total temporary disability, the workers receive two-thirds of their pre-injury wages (varies by state); in case of temporary partial disability where workers return to work before maximum medical improvement and perform reduced work tasks with lower pay, they receive temporary partial disability benefits. In other cases, if the disability is permanent in nature but the worker is not completely limited in their ability to perform tasks, they receive permanent partial disability benefits. On the other hand, permanent disability benefits are paid if the disability is considered permanent following maximum medical improvement [6].
There is still however a dearth of literature with regards to the severity of such injuries across specific injury characteristics and their relation with direct WC payments [7]. Further, existing literature that has used WC data to estimate associated payments has been limited in scope with regards to sample size, states included, limited injury characteristics analyzed individually, etc. [8][9][10][11]. This research effort uses data from a large insurance program covering several years and all fifty states to provide estimates for WC payments across specific injury characteristics related categories, stratified by the claim severity.

Study Sample
Data for this study were obtained from a large, national third-party provider for WC claims. Overall there were over 6 million claims including both work-related injuries and illnesses. Out of these, this effort focuses on 151,959 closed claims corresponding to work-related injuries from the years 2011-2018.

Variables
Claim type: This research effort focusses only on closed claims to ensure that our measures of cost are not truncated. Therefore, all other types of claims including those open or incomplete were excluded from this analysis. Key injury characteristics: These include National Council on Compensation Insurance (NCCI) based classifications for the nature of injury, source of injury, and body part injured. Further information on each of these individual categories is available elsewhere [12]. These are also provided in the tables ahead. Lastly, severity of injury was classified as medical only, and indemnity (disability including permanent disability (total and partial), temporary disability (total and partial), and death) related. Other variables: gender (male, female), marital status (married, never married, divorced, separated), age at injury (calculated as date of injury-date of birth), years of work experience until the date of injury (calculated as date of injury-date of hire), employment status (full time, part time including seasonal employees), and state in which the injury occurred. Total WC payment: These were calculated as a sum of total indemnity and medical and expense payments made on a claim.
Please note that while job titles were available, these were not coded using a standard tool and were available as free text. Therefore, these could not be used for these analyses. Additionally, given the large proportion (>50%) of missing information for lost workday counts, the latter were not included either.

Analyses
We provide descriptive statistics (frequencies, and percentages) for each of the key injury characteristics stratified by the claim's severity, i.e., medical, and indemnity (death and disability) claims. We used generalized estimating equations (GEE) [13] with a log link and gamma distribution and accounted for correlation within claim IDs nested within the insurance company's clients to analyze the association between the key injury characteristics and WC payments made. To clarify, the standard equation for generalized linear models (GLMs) can be expressed as: Y = g (β0 + β1X1 + β2X2 + ... + βkXk + e), where "g" is the link function. In case of correlated outcomes, GEE can build upon for the GLMs to account for the former [13]. All these models were adjusted for age at injury, gender, years of work experience, employment status, and state (because of states having differing medical fee schedules and regulations about compensation for lost wages). These covariates were selected based on Directed Acyclic Graphs [14]. We also noted that the model fit improved on addition of these covariates as compared to the crude model. All these models were stratified by the severity of claim categorized as medical only and death and disability-related claims. Death and disability-related claims were analyzed together as indemnity related claims due to low cell counts for death-related claims (n = 24) across several categories within the three key injury characteristics. In addition to estimating mean payments using GEE models, as a supplement, we also provide estimates of the "burden" posed by each injury characteristic by multiplying its frequency in the data by its mean cost. The Supplementary Tables provide the top five most burdensome claims for each category [15].

Results
There were 272,476 total closed claims in the study period (2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018). Out of these, for 13,945 claims, the state in which the injury occurred was not provided. Furthermore, nature and source of injuries that either corresponded to an illness or included no physical injury were also excluded from these analyses (n = 33,575). Injury severity codes that were labelled as "accidents only" were also excluded (n = 41,667). The analyses were further restricted to workers with age at injury being up to 85 years and those older were excluded (n = 213). Next, those with unknown employment status were also excluded (n = 29,870). Lastly, since almost all of these claims were reported in the year 2011 or later, the analyses were further limited to only these years (n = 151,959).
Out of these, 126,614 were medical only, and 25,345 were disability and death-related claims. Among the latter, 88% were temporary, and 11% were permanent disabilities. There were 51% males (n = 77,236) in this study. Around 33% were in the age group of 30-45 and 45-60 years old, while around 23% were in the 18-30 years old group. Around 76% of the claims involved full-time workers; 30% of these had less than one year of work experience, and 24% had more than 10.
Further descriptive statistics showed that strains, followed by contusions and lacerations, were the most common nature of injuries (Table 1) accounting for respectively 30%, 21%, and 12% of the total claims. Moreover, strains, contusions, and lacerations combined accounted for 61% of death and disability and 63% of medical only claims. Table 2 presents frequencies and percentages for the source of injury by the claim type. Overall, falls/slips were the most common source of injury, followed by strains and being struck by something or someone. Table 3 shows that the most common body parts injured included wrist and hand (n = 33,187), accounting for 23% of the medical only and 14% of the disability and death related claims. These were followed by injuries involving multiple body parts (n = 16,685) and the low back area (n = 15,610).
Our data include claims payments totaling $190 million approximately resulting from 128,688 WC payable claims over the study period. Of these, 81.4% are associated with medical only claims while the rest are associated with disability and death claims. The mean overall WC payment per claim for the claims that resulted in a payment was $1477 (Standard Deviation (SD): $7221). The mean payment varied by the type of claim with claims related to death (n: 24) resulting in a mean payment of $25,389 (SD: $56,790). This was followed by permanent disability (n: 2642, mean payment: $15,493, SD: $34,183), temporary disability (n: 21,235, Mean payment: $3255, SD: $10,772), and medical only claims (n: 104,757, mean payment: $754, SD: $1930). Note that, a total of 1414 disability and death related claims and 21,857 medical only claims did not incur a WC payment. The estimates provided ahead are further provide multivariable adjusted mean payment estimates and claim type stratified GEE models for each key injury categories, i.e., nature and source of injuries and body part injured. The estimates are from GEE models accounting for correlations within claims nested within each client and adjusted for age at injury, gender, years of work experience, employment status, and state. Overall burden (Tables S1-S3) for each of the former stated categories was estimated using mean payment estimates and frequency of each claim. These estimates also highlight the key injury characteristics that are overall associated with the highest cost burden. We additionally identify five of the most common medical and indemnity-related claims.    Table 4 shows that adjusting for covariates mentioned earlier, among medical only claims, with concussions as the reference, the mean payment was the highest for amputations (mean payment: $3849; Confidence Interval (CI): $1396, $10,608). Among disability and death related claims, ruptures cost the most on average with a mean payment of $14,285 ($7772, $26,255). Overall, with reference to Concussions, the five most burdensome (calculated as a product of frequency of respective claims and mean payment) medical-related claims included Strains, Contusions, Lacerations, Sprains, and Punctures. Similarly, the five most burdensome disability and death-related claims included Strains, Sprains, Contusions, Fractures, and Multiple injuries (Table S1).
As can be seen in Table 5, with reference to burns and scalds from chemicals, the most expensive medical claims on average were motor vehicle collisions that resulted in a mean per claim payment amounting to $1089 (CI: $798, $1486). Among indemnity related, i.e., disability and death related claims, the mean payment was the highest for falls or slips from a ladder or scaffolding ($7136; $4950, $10,289). The respective burden, taking into account the frequency of each category of source of injury, is provided in Table S2. Table S2 shows that with reference to chemicals related burn or scalds, the five most burdensome medical related claims were Fall/Slip-On Ice or Snow, Strain/Injury By-Reaching, Motor Vehicle-Collision with other Vehicle, Cut/Puncture/Scrape-Object Being Lifted or Handled, and Fall/Slip-From Liquid or Grease Spills. The same for indemnity, i.e., disability and death related claims were Strain/Injury By-Lifting, Fall/Slip-On Same Level, Strain/Injury Byrepetitive Motion-Carpal Tunnel Syndrome, Struck/Injured By-Another Person, and Strain/Injury By-Pushing or Pulling.
With reference to injuries to the lower log, injuries to body parts that were associated with high WC payments on average among medical only claims, besides those involving multiple body parts and other multiple injuries, including injuries to the spinal cord (mean payment: $911; $722, $1148). Next, among disability and death related claims, spinal cord injuries were the most expensive injuries with a mean payment of $11,115 (2144, 57,637) ( Table 6). As far as overall burden of these injuries is concerned, as can be identified from the Table S3, the five most burdensome medical claims related injuries with respect to the reference were injuries to the wrist and hand, multiple body parts being injured, injuries to the low back area, ankle and feet, and knee. The same for disability and death related claims include multiple body parts being injured, injuries to low back area, wrist and hand, knee, and shoulder(s). Table 4. Log-gamma regression models estimating the mean payments and burden associated with specific nature of injury categories for those claims that resulted in a payment, stratified by medical and death and disability claims.  Table 5. Log-gamma regression models estimating the mean payments and burden associated with specific source of injury categories for those claims that resulted in a payment, stratified by medical and death and disability claims.  Table 6. Log-gamma regression models estimating the mean payments and burden associated with specific body part injured for those claims that resulted in a payment, stratified by medical and death and disability claims.

Body Part Injured Medical Claims Disability and Death Related Claims
Mean Payment ($) 95% CI ($) Mean Payment ($) 95% CI ($) of the total cash benefit payments by the WC program [7]. In the current study, among the claims for which a WC payment was made, 81.4% were related to medical only claims, and the rest were death and disability related claims. While the former resulted in an overall WC payment of $111,129,376, the latter amounted to $78,988,043 in WC payments over the study period. Next, we found that the most common body parts injured were wrist and hand, multiple body parts, and low back. Subsequently, with over $900 in mean WC payments, multiple injuries, including multiple head injuries were the most expensive claims on average. Spinal cord injuries, followed by shoulder injuries, were the most expensive death and disability claims based on the mean payments. Several previous research efforts focusing on particular states and specific occupations have used WC systems data to estimate the prevalence of different types of workplace injuries and the characteristics of affected workers. For example, a 10 year-long study [19] conducted among workers in the private ambulance services industry in Ohio found that around 54% of the claimants were males, and the majority (40%) were in the 25-34-year-old age-group. Around 60% of the claims were attributed to sprains and strains combined. Another previous effort [20] that analyzed data from over 232,000 unique claims from a large Maryland WC provider from 1998-2008 reported that the most common body parts injured were back, knee, and hand. The mean compensation was $6785 across all injury types. Multiple injuries accounted for around 64% of the total WC payments [20]. Another effort [21] that focused on agricultural injuries among workers in Colorado from 2000-2004 found an average cost of $7488 and $12,299 for vehicle-related medical and indemnity claims, respectively. As far as body parts injured are concerned, the most expensive medical and indemnity claims were respectively associated with leg and spine/back. Finally, dislocations were in general the most expensive claims.

Strengths
This study analyzed payments for both medical and indemnity claims across key injury characteristics across all fifty states, controlling for important confounding variables that may not be always available to researchers using these data, e.g., employment status, years of work experience, etc. Thus, the results of this study, given its wide scope, may be more representative of the USA workforce [18]. This study is also among the limited number of studies that have been able to analyze the payments across very specific injury characteristics. Our results provide information that could be useful in designing injury preventive interventions [20].

Limitations
The occupations among which the claimants were employed would have been an important variable to adjust for. Another limitation of using WC data is that nonfatal injuries and illnesses may be underreported in WC claims. However, this is also a limitation of other sources of data for work-related injuries and illnesses. In particular, the widely used Bureau of Labor Statistics (BLS) data have been reported to substantially underestimate nonfatal injuries and illnesses [22,23]. In fact, previous estimates reported by a study that linked individual cases between BLS and WC data in six states found that while WC systems missed over 180,000 lost-time injuries, BLS missed almost 340,000. Further, around 69,000 injuries were unreported by either of the aforementioned surveillance systems [22]. Another limitation relates to the true prevalence of strain-related injuries. These are one of the most common nature of injury codes reported, and yet, it is well known that this injury category is subject to potential misuse/classification error. While this limits the conclusions that injury epidemiologists can draw from our data, our results are nonetheless important to corporate risk managers seeking to understand costs imposed by employees reporting these injuries. Besides these, other limitations of the WC data itself include that no WC dataset covers the universe of employment injuries, nor does any WC dataset necessarily provide a representative sample. There are several reasons for this, including, underreporting of injuries to WC programs, WC data only covering claims from a particular WC provider company. In addition, costs derived from WC claims may underestimate the true extent of lost wages since workers are only compensated for a portion of their wages and because workplace injuries are underreported to WC companies. Finally, WC data can be challenging to access because they are protected for proprietary and privacy purposes by the companies that manage those claims. Besides, there exists no central repository for WC claims within the USA [1].

Conclusions
This effort uses data from a large third-party insurance provider and provides estimates for WC payments across specific categories of injury characteristics. Overall, the estimates show that injuries pose a huge economic burden to the WC system. Priority intervention may target specific injury characteristics that are associated with higher payments. However, it is also important to recognize that certain types of injuries like strains, which are associated with lower mean payments, are also the most frequent types of injuries. It is therefore also crucial to consider the frequencies of various injury causing mechanisms and body parts injured while designing workplace interventions.
Supplementary Materials: The following are available online at http://www.mdpi.com/1660-4601/17/19/7157/s1. Table S1: Log-gamma regression models estimating the mean payments and burden associated with specific nature of injury categories for those claims that resulted in a payment, stratified by medical and death and disability claims, Table S2: Log-gamma regression models estimating the mean payments and burden associated with specific source of injury categories for those claims that resulted in a payment, stratified by medical and death and disability claims, Table S3: Log-gamma regression models estimating the mean payments and burden associated with specific body part injured for those claims that resulted in a payment, stratified by medical and death and disability claims.