Skip to main content
SpringerLink
Log in

Using Predictive Analytics for Early Identification of Short-Term Disability Claimants Who Exhaust Their Benefits

  • Published:
Journal of Occupational Rehabilitation Aims and scope Submit manuscript

Abstract

Purpose Early interventions can help short-term disability insurance (STDI) claimants return to work following onset of an off-the-job medical condition. Accurately targeting such interventions involves identifying claimants who would, without intervention, exhaust STDI benefits and transition to longer-term support. We identify factors that predict STDI exhaustion and transfer to long-term disability insurance (LTDI). We also explore whether waiting for some claims to resolve without intervention improves targeting efficiency. Methods We use a large database of STDI claims from private employer-sponsored disability insurance programs in the United States to predict which claims will exhaust STDI or transition to LTDI. We use a split sample approach, conducting logistic regressions on half of our data and generating predictions for the other half. We assess predictive accuracy using ROC curve analysis, repeating on successive subsamples, omitting claims that resolve within 2, 4, and 6 weeks. Results Age, primary diagnosis, and employer industry were associated with the two outcomes. Rapid attrition of short-duration claims from the sample means that waiting can substantially increase the efficiency of targeting efforts. Overall accuracy of classification increases from 63.2% at week 0 to 82.9% at week 6 for exhausting STDI benefits, and from 63.7 to 83.0% for LTDI transfer. Conclusions Waiting even a few weeks can substantially increase the accuracy of early intervention targeting by allowing claims that will resolve without further intervention to do so. Predictive modeling further narrows the target population based on claim characteristics, reducing intervention costs. Before adopting a waiting strategy, however, it is important to consider potential trade-offs involved in delaying the start of any intervention.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2

Similar content being viewed by others

Notes

  1. Five states and Puerto Rico have mandatory STDI benefits; three of those states provide insurance to most workers via public funds: California, New Jersey, and Rhode Island. Hawaii and New York require employers to provide short-term disability benefits through self-insurance or a licensed carrier.

  2. As with STDI, LTDI policies provide wage replacements following a specified duration of disability (often 14, 26, or 52 weeks), up to a maximum benefit duration.

  3. Workers’ compensation covers cash benefits and medical care to workers with work-related (or “on-the-job”) medical conditions. In 46 states and the District of Columbia, employers may either purchase workers’ compensation insurance in a competitive marketplace or self-insure. Four states—North Dakota, Ohio, Washington, and Wyoming—rely exclusively on state workers’ compensation funds.

  4. Ideally, we would also be able to distinguish between claimants whose labor force participation outcome is responsive to early intervention and claimants who cannot continue to work even with intervention. We do not address this complicated challenge in our paper.

  5. With the exception of New York, New Jersey, Rhode Island and Hawaii which require that employers provide STDI benefits to workers, in the U.S., disability insurance is provided at an employers’ discretion as an employee benefit. California requires that all employees maintain disability insurance, but does not compel employers to purchase policies on behalf of workers. Whether or not an employer-based policy is state-mandated, the employer typically pays the STDI policy’s premiums if it fully insures through an insurance carrier. Employers may also self-insure, paying the dollar value of all wage replacements costs from its own cash reserves, with or without contracting for administrative services through a third party.

  6. Claims with benefit duration of 26 weeks made up 73% of the claims in the IBI data. Regression and prediction results were similar for claims with benefit duration of 13 weeks and 52 weeks. We cannot observe reason for claim closure, and therefore cannot distinguish between claims that closed because the underlying condition resolved enough for a return to work and claims that closed for other reasons, such as death or transitions away from the sponsoring employer.

  7. It should not be inferred from the lack of LTDI information on a claim that an employer does not have both STDI and LTDI policies for their employees. It may be that STDI and LTDI policies are provided by different entities, hence the data are held by different data suppliers.

  8. We organize diagnoses into the major categories described in the ICD-9. Pregnancy-related claims are excluded. We selected specific subcategories based on relevance to workplace disability. These include back conditions such as intervertebral disc disorders, mental health conditions such as depression and affective disorders, malignant neoplasms, and sprain injuries. Categories of diagnosis that occur very rarely in STDI (such as diseases of the skin, blood, and blood-forming organs, congenital anomalies, and symptoms without a diagnosis) are grouped together as “other illnesses”. Other illnesses account for almost 8% of diagnoses, about half of which are undiagnosed symptoms.

  9. Winsorizing involves setting extreme values to a specified percentile of the data to limit the effect of outliers, which may be spurious; in this case, we set all data below the 1st percentile to the 1st percentile value and all data above the 99th percentile to the 99th percentile value.

  10. Presumably, some STDI carriers already intervene in some fashion to improve outcomes for claimants on certain employer policies. In that context, “resolve on their own” can be interpreted as “resolve under current practice.”

  11. We also tested a version of the logistic regression model that included interactions between each age group and each diagnosis category, and a version that also included sex-diagnosis interaction terms. We also tested for nonlinear effects of wage and company size by using wage tercile categories and company size categories (0–5 k, 5–10 k, 10 k+). The results were nearly identical, so we present results from the model with no interaction terms, linear wage, and log of company size.

  12. The AUC represents the probability that the predicted value for a randomly chosen positive is greater than the predicted value for a randomly chosen negative.

  13. Usual practice may, in fact, include some efforts on the part of insurers and/or employers to resolve STDI claims. In the context of discussions around early intervention, we assume that proposed interventions would be in addition to business as usual, and that the attrition of claims from the sample over time forms the baseline resolution rate.

References

  1. Monaco K. Disability insurance plans: trends in employee access and employer costs. Beyond the numbers: pay and benefits 2015. https://www.bls.gov/opub/btn/volume-4/disability-insurance-plans.htm. Accessed 25 April 2018.

  2. Goetzel RZ, Long SR, Ozminkowski RJ, Hawkins K, Wang S, Lynch W. Health, absence, disability, and presenteeism cost estimates of certain physical and mental health conditions affecting US employers. J Occup Environ Med. 2004;46(4):398–412.

    Article  Google Scholar 

  3. Arena VC, Padiyar KR, Burton WN, Schwerha JJ. The impact of body mass index on short-term disability in the workplace. J Occup Environ Med. 2006;48(11):1118–1124.

    Article  Google Scholar 

  4. Neuhauser F, Ben-Shalom Y, Stapleton D. Early identification of potential SSDI entrants in California: the predictive value of State Disability Insurance and Workers’ Compensation claims. J Occup Rehabilit. 2018. https://doi.org/10.1007/s10926-018-9791-9.

    Article  Google Scholar 

  5. Stover B, Wickizer TM, Zimmerman F, Fulton-Kehoe D, Franklin G. Prognostic factors of long-term disability in a workers’ compensation system. J Occup Environ Med. 2007;49(1):31–40.

    Article  Google Scholar 

  6. Sears JM, Blanar L, Bowman SM, et al. Predicting work-related disability and medical cost outcomes: estimating injury severity scores from workers’ compensation data. J Occup Rehabilit. 2013;23(1):19–31.

    Article  Google Scholar 

  7. Steenstra IA, Munhall C, Irvin E, et al. Systematic review of prognostic factors for return to work in workers with sub acute and chronic low back pain. J Occup Rehabilit. 2017;27(3):369–381.

    Article  Google Scholar 

  8. Savych B, Neumark D, Lea R. The impact of opioid prescriptions on duration of temporary disability. Cambridge: Workers Compensation Research Institute; 2018.

    Google Scholar 

  9. Autor D, Duggan M, Gruber J. Moral hazard and claims deterrence in private disability insurance. Am Econ J: Appl Econ. 2014;6(4):110–141.

    Google Scholar 

  10. Franklin GM, Turner JA, Wickizer TM, Fulton-Kehoe D, Mootz RD. Disability prevention. In: Gebhart GF, Schmidt RF, editors. Encyclopedia of pain. Berlin: Springer; 2013. p. 1006–1011.

    Chapter  Google Scholar 

  11. Loisel P, Anema JR, editors. Handbook of work disability. New York: Springer; 2013.

    Google Scholar 

  12. Shaw WS, Reme SE, Pransky G, Woiszwillo MJ, Steenstra IA, Linton SJ. The pain recovery inventory of concerns and expectations: a psychosocial screening instrument to identify intervention needs among patients at elevated risk of back disability. J Occup Environ Med. 2013;55(8):885–894.

    Article  Google Scholar 

  13. Bevan S. Back to work: exploring the benefits of early interventions which help people with chronic illness remain in work. Brussels: Fit for Work Europe, the Work Foundation; 2015.

    Google Scholar 

  14. Stapleton D, Anfield R, Burns R, Cashdollar W, Doornink B, Gifford B, Harris M, Ufier K. Targeting early intervention to workers who need help to stay in the labor force. Washington, DC: Mathematica Policy Research; 2015.

    Google Scholar 

  15. Lasko TA, Bhagwat JG, Zou KH, Ohno-Machado L. The use of receiver operating characteristic curves in biomedical informatics. J Biomed Inform. 2005;38(5):404–415.

    Article  Google Scholar 

  16. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3(1):32–35.

    Article  CAS  Google Scholar 

  17. Van der Klink JJL, Blonk RWB, Schene AH, van Dijk FJH. Reducing long term sickness absence by an activating intervention in adjustment disorders: a cluster randomised controlled design. Occup Environ Med. 2004;60(6):429–437.

    Article  Google Scholar 

  18. Reme SE, Grasdal AL, Løvvik C, Lie SA, Øverland S. Work-focused cognitive-behavioural therapy and individual job support to increase work participation in common mental disorders: a randomized controlled multicentre trial. Occup Environ Med. 2015. https://doi.org/10.1136/oemed-2014-102700.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Richmond H, Hall AM, Copsey B, Hansen Z, Williamson E, Hoxey-Thomas N, et al. The effectiveness of cognitive behavioral treatment for non-specific low back pain: a systematic review and meta-analysis. PLoS ONE. 2015;10(8):e0134192. https://doi.org/10.1371/journal.pone.0134192.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Srivastava P. Getting engaged: giving employees a nudge toward better health. Compens Benefits Rev. 2012;44(2):105–109.

    Article  Google Scholar 

  21. Sacarny A, Yokum D, Finkelstein A, Agrawal S. Medicare letters to curb overprescribing of controlled substances had no detectable effect on providers. Health Aff. 2016;35(3):471–479.

    Article  Google Scholar 

  22. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018;319(13):1317–1318.

    Article  Google Scholar 

  23. Shrestha A, Bergquist S, Montz E, Rose S. Mental health risk adjustment with clinical categories and machine learning. Health Serv Res. 2017. https://doi.org/10.1111/1475-6773.12818.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

The authors thank the Integrated Benefits Institute for providing the data, and David Mann and David Stapleton for providing helpful comments on an early draft. The research reported herein was performed pursuant to a grant from the Social Security Administration that was funded as part of the Disability Research Consortium (Grant DRC12000001-04-00). The opinions and conclusions expressed are solely those of the authors and do not represent the opinions or policy of SSA nor of any other agency of the federal government. Neither the U.S. government nor any of its agencies or employees makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of the contents of this paper.

Funding

The research reported herein was performed pursuant to a grant from the Social Security Administration that was funded as part of the Disability Research Consortium (Grant No. DRC12000001-04-00).

Author information

Authors and Affiliations

  1. Mathematica Policy Research, 505 14th Street, Suite 800, Oakland, CA, 94612, USA

    Kara Contreary

  2. Mathematica Policy Research, 1100 1st Street, NE, 12th Floor, Washington, DC, 20002, USA

    Yonatan Ben-Shalom

  3. Integrated Benefits Institute, 595 Market Street, Suite 810, San Francisco, CA, 94105, USA

    Brian Gifford

Authors
  1. Kara Contreary
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Yonatan Ben-Shalom
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Brian Gifford
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Kara Contreary.

Ethics declarations

Conflict of interest

Contreary, Ben-Shalom, and Gifford declare that they have no conflicts of interest.

Ethical Approval

This article does not contain any studies with human participants or animals performed by any of the authors.

Appendix

Appendix

See Table 4.

Table 4 Counts for closed claims, by outcome
Full size table

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Contreary, K., Ben-Shalom, Y. & Gifford, B. Using Predictive Analytics for Early Identification of Short-Term Disability Claimants Who Exhaust Their Benefits. J Occup Rehabil 28, 584–596 (2018). https://doi.org/10.1007/s10926-018-9815-5

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10926-018-9815-5

Keywords

Search

Navigation