Everything Except The Diagnosis
What predicts whether an injured person works again is already in their file, three pages behind the diagnosis.
IN THIS ISSUE
• A German study of 685,890 people who completed a rehabilitation program found that the six strongest predictors of still being at work a year later were all about employment, income and age. None of the six was a medical variable.
• French researchers followed liver transplant patients and found the operation did not impact the employment rate at all. Thirty-one per cent of patients were working before the transplant and 31 per cent were working a year after it. The researchers concluded that disability insurance itself was one of the obstacles.
• Australian researchers compared eight state and territory compensation schemes and found that two workers with the same injury spend different amounts of time off work depending on which scheme they claim in.
• A Sydney systematic review of 29 prospective studies of compensation and recovery found that not one of them reported a compensation-related factor associated with better health.
A South Korean panel study found that whether an injured worker got their old job back came down to their employment contract. In Australia, neither SIRA’s published return to work rate nor the 2025 National Return to Work Survey records whether a worker went back to their own employer or to a stranger.
Every referral I’ve received in more than thirty years has led with the diagnosis. It’s the first line on the form, it sets the expectation of how long this ought to take, it decides which treatments get funded, and it’s usually how the file gets named. Everything downstream is organised around it.
Two people arrive with the same diagnosis and end up in completely different places, and for most of my career I couldn’t tell you why with any precision. I’ve sat with hundreds of injured workers. Some go back and stay back. Others get stuck, stop leaving the house, become frightened of being seen outside, and don’t recover. Whatever separated them wasn’t on the front page of the file.
A run of studies published since November 2025 has started to answer it, and the answer is an awkward one, because what predicts whether somebody works again is already in the file. It’s three pages back, in the employment history, and nobody reads that part first.
What belongs at the top of the referral
Five things about a person’s working life predict their return to work better than their diagnosis does, and every one of them could sit where the condition sits now:
• how much of the last twelve months they actually worked
• what they earn
• what contract they’re employed on
• whether they were still working when this started
• how long they’ve already been off
None of that needs a new assessment tool, a new program or one extra appointment, because most of it is already in the file and the rest takes a single conversation. The case for moving it to the front of the form is an evidence case, and it comes out of a German pension database, a French transplant unit, a Korean panel survey and eight Australian compensation schemes.
Not one of the six strongest predictors was medical
None of the six strongest predictors of a lasting return to work was medical, across 685,890 Germans who completed a rehabilitation program. Mathis Elling and colleagues used the German Pension Insurance’s rehabilitation statistics database, which records everyone the insurer funds through multimodal medical rehabilitation, described in the paper as “typically provided as a 3-week inpatient program in specialised rehabilitation clinics” delivered by an interdisciplinary team. Everyone in the dataset had a musculoskeletal disorder and had been through that kind of program between January 2018 and December 2021.
The model put employment history ahead of everything else:
• how many days the person had worked in the year before the rehabilitation, counting only regular contribution-paying jobs
• their income
• whether they were certified unfit for work when they were admitted
• their occupational status
• how long they’d already been off work
• their age
Sixty-five per cent achieved what the study calls a stable return to work, defined as being back in “employment subject to social insurance contributions in the ninth to twelfth month after the end of the rehabilitation measure”. That has nothing to do with sick leave, and it’s the German category for a regular job carrying compulsory health, pension, long-term care and unemployment contributions, which excludes the self-employed, civil servants and marginal work. The closest Australian marker is a job that attracts the superannuation guarantee and shows up in payroll reporting, so what Elling measured was an administratively verified job nine to twelve months after the program finished, rather than a worker or an employer saying the return had gone well.
Everybody in this cohort had the same class of problem, so the study isn’t telling us that one diagnosis predicts better than another. It’s answering the question we face in an assessment: take a large group with the same kind of injury, put them through the same program, and what separates the ones who get back to work is how much of the previous year they’d been working and what they earn.
The study behind this
STUDY Elling, Hetzel, Streibelt, Sänger, Schwarz and Seifert, Germany (German Pension Insurance). Published online 6 January 2026
DESIGN “This retrospective cohort study utilized data from the Rehabilitation Statistics Database of the German Pension Insurance. A total of 685,890 individuals receiving multimodal medical rehabilitation for musculoskeletal disorders between January 2018 and December 2021 were included. A Light Gradient Boosting Machine (LightGBM) model was trained on 75% of the sample and tested on the remaining 25%. Model predictions were explained using Shapley values to determine feature contributions and interactions.” Every person in the cohort had a musculoskeletal disorder, so the analysis separates people who share a condition rather than comparing one condition with another
FINDING “Overall, 65.0% of individuals achieved stable RTW. On the test dataset (threshold 0.5), the model achieved an accuracy of 0.815, precision of 0.830, recall of 0.901, and an AUC-ROC of 0.867. The most important predictors were: days in employment, income, incapacity for work at admission, occupational status, duration of incapacity for work, and age.” Stable return to work is defined in the paper as “employment subject to social insurance contributions in the ninth to twelfth month after the end of the rehabilitation measure” (Introduction), and days in employment as the “number of days in employment subject to social insurance contributions in the calendar year prior to the rehabilitation” (Methods, Features)
SOURCE Journal of Occupational Rehabilitation, advance online publication. doi.org/10.1007/s10926-025-10359-3. Quotations are from the published abstract, Methods and Results sections, and from the Introduction and Methods of the full text where marked. Advance online, so no page numbers have been assigned
Accepted, rejected and no claim didn’t differ from one another at all, and that’s the comparison worth sitting with. Being told no was no worse than being told yes and being told nothing was worse than both. The authors are careful to say the finding is exploratory and can’t establish causality, and they’re right to be careful about it, but it matches the Collie data, and it matches what I remember of waiting.
Whether they were working when it started outranked a liver transplant
Thirty-one per cent of liver transplant candidates in Rennes were working when they were assessed for surgery, and 31 per cent were working a year after they’d had it. Dorian Quelard and colleagues followed 141 working-age patients through assessment between June 2022 and December 2023, and 86 of them went on to be transplanted. The authors report that return to work “at one year was 31%, with no increase from baseline” (abstract, Results).
Replacing a failing liver is about as decisive as medicine gets, and it made no difference at all to whether these patients worked. What did make a difference was whether they’d been working beforehand, because patients still in a job at assessment were far more likely to be working a year after surgery (aOR 9.67, 95% CI 3.06 to 35.09). Education predicted who was still working at that first assessment (aOR 2.99, 95% CI 1.18 to 7.63), so those two numbers describe one chain rather than two separate effects. People with more education were more likely to have held onto a job while they were seriously ill, and holding onto the job was what carried them through to the other side of the operation.
The study behind this
STUDY Quelard, Delhaye, Houssel-Debry, Giguet, Uguen, Jezequel, Bardou-Jacquet, Paris, Artru and Saade, Rennes University Hospital, France. Published online 13 August 2026
DESIGN Retrospective cohort of working-age (18 to 64) liver transplant candidates evaluated between June 2022 and December 2023. 141 candidates, of whom 86 were transplanted. Socio-demographic, medical and occupational data collected at baseline and one year post-transplant. Single centre. The two employment percentages rest on different groups, 141 candidates and 86 recipients
FINDING “Among 141 LT candidates (65% male, median age 56), 31% were working at evaluation.” “In multivariable analysis, higher education independently predicted pre-transplant employment (aOR 2.99, 95%CI 1.18–7.63; p = 0.02). Among 86 patients who underwent transplantation, RTW prevalence at one year was 31%, with no increase from baseline. Return-to-work was more likely among those working before LT (aOR 9.67, 95%CI 3.06–35.09; P < 0.001) and inversely associated with post-LT vascular complications.” Conclusion: “Pre-transplant working status and education level were key determinants, while disability insurance hindered reintegration.” The published abstract reports odds ratios for working status, education and vascular complications; the statement about disability insurance is given as a conclusion
SOURCE Journal of Occupational Rehabilitation, advance online publication. doi.org/10.1007/s10926-026-10442-3. Quotations are from the published abstract, Results and Conclusion sections. Advance online, so no page numbers have been assigned
The strongest predictor of all isn't on the form and isn't ours to change
Transplant clinicians writing up a hepatology cohort concluded that disability insurance was getting in the way of people going back to work. They conclude that “pre-transplant working status and education level were key determinants, while disability insurance hindered reintegration”. This is a French surgical unit with no particular reason to be looking at insurance design, and they found it anyway.
Australian researchers have been measuring the same thing for a decade, and they’ve measured it far more precisely. Alex Collie’s group at Monash compared eight Australian state and territory workers’ compensation schemes, covering more than 90 per cent of the labour force, using 95,976 accepted claims from 2010. They controlled for the worker, the injury, the employer and the demographics, and time off work still varied by which scheme a person happened to claim in. Set against New South Wales, workers in Victoria (HR 0.75, 99% CI 0.73 to 0.77) and South Australia (HR 0.84, 99% CI 0.81 to 0.88) took significantly longer to get back to work, while workers in Queensland (HR 1.32, 99% CI 1.29 to 1.36) and Tasmania (HR 1.31, 99% CI 1.24 to 1.39) got back faster. They conclude that “the jurisdiction in which an injured worker makes a compensation claim has a significant and independent impact on duration of time loss”.
Same injury, same sort of worker, different state, different recovery. A systematic review from Ian Cameron and Ian Harris’s group in Sydney puts the harder version of it, grading 29 prospective studies with at least six months of follow-up and finding strong evidence that compensation status is linked to poorer psychological function and legal representation to poorer physical function. Their summary of all 29 is that “no studies reported an association between compensation related factors and improved health outcomes”. How much of what we routinely write up as a poor recovery is actually the scheme?
The study behind this
STUDY Collie, Lane, Hassani-Mahmooei, Thompson and McLeod, Australia (Monash University). Published online 5 May 2016, BMJ Open 6(5), May 2016 issue
DESIGN “Eight Australian state and territory workers’ compensation systems, providing coverage for more than 90% of the Australian labour force. Administrative claims data from these systems were provided by government regulatory authorities for the study.” “95 976 Australian workers with workers’ compensation claims accepted in 2010 and with at least 2 weeks of compensated time off work.” Primary outcome: “Duration of time lost from work in weeks, censored at 104 weeks.” Cox regression. Claims are from 2010, so scheme rules have changed since
FINDING “After controlling for demographic, worker, injury and employer factors in a Cox regression model, significant differences in duration of time loss between state and territory of claim were observed. Compared with New South Wales, workers in Victoria, South Australia and Comcare had significantly longer durations of time off work and were more likely to be receiving income benefits at 104 weeks postinjury, while workers in Tasmania and Queensland had significantly shorter durations of time off work.” Victoria HR 0.75 (99% CI 0.73 to 0.77); South Australia HR 0.84 (99% CI 0.81 to 0.88); Comcare HR 0.91 (99% CI 0.85 to 0.96); Queensland HR 1.32 (99% CI 1.29 to 1.36); Tasmania HR 1.31 (99% CI 1.24 to 1.39)
SOURCE BMJ Open, 6(5), Article e010910, open access. doi.org/10.1136/bmjopen-2015-010910. Quotations are from the published abstract, Setting, Participants, Primary outcome measure and Results sections; the hazard ratios are from the Results section of the full text. Article-numbered, so no page numbers exist
Contract type decides which return you get, and our published rate doesn't record it
Two Korean workers with the same injury end up in different places depending on the contract they were employed on. Non-regular workers were more likely to finish up unemployed or working for a different employer, and regular workers were more likely to go back to the job they’d been injured in. Min-Kyu Kim and Ji-Bum Chung followed 2,643 injured workers across four annual waves of Korea’s workers’ compensation panel study, modelling each employment type separately. Non-regular work is the Korean category for casual, fixed-term, agency and contract employment, covering around 37 per cent of that workforce, and it maps onto casual employment as the Fair Work Act defines it, together with fixed-term contracts and labour hire.
Australian practice takes that distinction seriously. WorkSafe Victoria’s claims manual has agents and providers work down a hierarchy that runs from the same employer and the same job through to a different employer and a different job, and NSW funds three separate programs, JobCover, Transition to Work and Work Trials, that a worker becomes eligible for only once a return to their own employer is off the table. The duty to provide suitable work sits with “the employer liable to pay compensation to the worker under this Act in respect of the injury” (Workplace Injury Management and Workers Compensation Act 1998 (NSW), s 49(1)), which means the pre-injury employer and nobody else.
None of that distinction survives into the numbers we publish. SIRA’s return to work rate is “the percentage of workers who have been off work as a result of their employment-related injury/disease and have returned to work at different points in time from the date of injury” (Open data explanatory notes, 20 November 2024), measured at 4, 13, 26, 52 and 104 weeks, and it doesn’t mention the employer at all. Safe Work Australia’s National Return to Work Survey doesn’t separate them either, and the 2025 report, covering claims made between July 2022 and June 2024, doesn’t contain the phrase same employer anywhere in it.
"A worker who walked back into their own job and a worker placed with a stranger on a wage subsidy arrive in our statistics as the same result."
REHAB REIMAGINED | DR. CAROLINE HOWE
When a casual’s shifts simply stop, or a fixed-term contract reaches its end date, that duty has nobody obvious left to attach to. So the workers Kim and Chung found were least likely to get their own job back are the same workers our framework has least to offer, and the same workers our published rate is least able to describe. If the number can’t tell those two returns apart, what exactly is improving when it goes up?
Diagnosis tells you roughly how long, and almost nothing about who
The condition does tell you roughly how long somebody will be off work, and Line Mouton’s team at KU Leuven has now measured how well. They pulled together the whole published evidence base on sickness absence duration across twenty of the commonest conditions in work disability, screening 4,826 scientific articles and 770 grey-literature documents to find 69 studies, and reported time off ran from 1.04 days to 11.4 years with a median of 24 days. Three of the twenty conditions had a formal guideline, and epilepsy, obsessive compulsive disorder and Ehlers-Danlos syndrome had no reported data at all. The authors conclude that the evidence “is insufficient to support a clear consensus or the development of guideline recommendations”, and duration benchmarks by condition are exactly what gets quoted at a worker taking longer than the file says they should.
When something did sort out who benefited from a treatment, it wasn’t the condition either. Maria Rosenblom’s group in Norway went back to a four-arm trial of 414 people on long-term sick leave with chronic low back pain, and inside the group who received cognitive behavioural therapy, those scoring highest on precontemplation at the start had a 63 per cent lower chance of being back at work twelve months later. Same diagnosis, same treatment, same trial arm, and what separated them was how ready they were. Would a stage-of-change score have changed which treatment we recommended?
Money is the one nobody asks about twice
Financial hardship separated the Americans with long COVID who kept their jobs from those who lost them (p = 0.03), in Han Su’s follow-up of 79 people who had already gone back to work, though that group was self-selected through a clinic, a research register and a peer support group. We take a person’s financial position at intake, write it into the social history and never return to it, and it sits underneath Elling’s income predictor and underneath Quelard’s finding that education predicted who was still working in the first place. A worker whose finances have collapsed during a claim is in a different position from one whose haven’t, and most of the time we don’t know which we’re looking at.
Two other things predict well and take minutes to ask about. How bad somebody’s anxiety and depression symptoms still are on the day they go back to work predicts whether they’ll be off again inside the year (HR 1.176, 95% CI 1.027 to 1.346), from Anke Deprez’s follow-up of 148 Belgian employees, 27.9 per cent of whom needed a certificate again within twelve months. What a person believes about working while unwell goes on predicting sick leave a year after treatment ends, on Marianne Bjørndal’s Oslo data for 236 patients, while the support they report at work stops predicting once treatment does.
What I couldn't say for thirty years
Two people with the same injury end up in different places, and for thirty years the honest answer to why was that I didn’t know. I could see it coming sometimes and never point to anything. What’s changed is that a transplant unit in France, a pension database in Germany and eight Australian compensation schemes have each answered it from a completely different direction, and they agree with one another.
The diagnosis stays at the top of the referral, because we do need to know what we’re dealing with. It just doesn’t tell us who we’re dealing with, and on this evidence the second question is the one that decides how it ends.
Sources, with the date each paper first appeared online
Elling, M., Hetzel, C., Streibelt, M., Sänger, N., Schwarz, B., & Seifert, N. (2026). Machine learning to predict return to work after medical rehabilitation for musculoskeletal disorders: A retrospective cohort study. Journal of Occupational Rehabilitation. Advance online publication, 6 January 2026. https://doi.org/10.1007/s10926-025-10359-3
Quelard, D., Delhaye, H., Houssel-Debry, P., Giguet, B., Uguen, T., Jezequel, C., Bardou-Jacquet, E., Paris, C., Artru, F., & Saade, A. (2026). Socio-occupational aspects in liver transplant candidates and recipients: A retrospective cohort study. Journal of Occupational Rehabilitation. Advance online publication, 13 August 2026. https://doi.org/10.1007/s10926-026-10442-3
Collie, A., Lane, T. J., Hassani-Mahmooei, B., Thompson, J., & McLeod, C. (2016). Does time off work after injury vary by jurisdiction? A comparative study of eight Australian workers’ compensation systems. BMJ Open, 6(5), Article e010910. Online 5 May 2016. https://doi.org/10.1136/bmjopen-2015-010910
Murgatroyd, D. F., Casey, P. P., Cameron, I. D., & Harris, I. A. (2015). The effect of financial compensation on health outcomes following musculoskeletal injury: Systematic review. PLOS ONE, 10(2), Article e0117597. Online 13 February 2015. https://doi.org/10.1371/journal.pone.0117597
Kim, M.-K., & Chung, J.-B. (2026). Return-to-work challenges faced by non-regular workers: Insights from a four-year panel study in South Korea. WORK: A Journal of Prevention, Assessment and Rehabilitation. Advance online publication, 12 June 2026. https://doi.org/10.1177/10519815261458766
Mouton, L., Lambreghts, C., & Godderis, L. (2026). Sickness absence duration across twenty different pathologies: A systematic review. Journal of Occupational Rehabilitation. Advance online publication, 24 April 2026. https://doi.org/10.1007/s10926-026-10370-2
Rosenblom, M. F., Jacobsen, H. B., & Reme, S. E. (2026). Ready or not? Applying the transtheoretical model of change on return-to-work following CBT for chronic low back pain. Journal of Occupational Rehabilitation. Advance online publication, 17 July 2026. https://doi.org/10.1007/s10926-026-10432-5
Su, H., Keller, B., Danesh, V., McPeake, J., Boehm, L. M., Eaton, T. L., Mart, M. F., Patel, M. B., & Ely, E. W. (2026). Long COVID and the challenge of long-term employment: An ecological, sequential explanatory mixed-methods approach. Journal of Occupational Rehabilitation. Advance online publication, 12 March 2026. https://doi.org/10.1007/s10926-026-10376-w
Deprez, A., Boets, I., Vandenbroeck, S., & Godderis, L. (2026). Determinants of sustainable return to work after burnout or depression: A longitudinal cohort study. Journal of Occupational Rehabilitation. Advance online publication, 13 May 2026. https://doi.org/10.1007/s10926-026-10407-6
Bjørndal, M. T., Gjengedal, R. G. H., Frederiksen, K. P., Lending, H. D., Osnes, K., Hannisdal, M., & Hjemdal, O. (2026). Beliefs about working with health problems and support at work as predictors of sick leave at post-treatment and at follow-up: Secondary analyses of a randomised controlled trial with registry data. Journal of Occupational Rehabilitation. Advance online publication, 6 August 2026. https://doi.org/10.1007/s10926-026-10438-z
State Insurance Regulatory Authority. (2024, November 20). Return to work rates: Explanatory notes. SIRA open data. https://www.sira.nsw.gov.au/open-data/return-to-work-rates
Safe Work Australia. (2025, November). National Return to Work Survey 2025: Analytical report. https://data.safeworkaustralia.gov.au/sites/default/files/2025-11/NRTW_Survey_Analysis_Report_Nov2025.pdf
Workplace Injury Management and Workers Compensation Act 1998 (NSW) s 49. https://legislation.nsw.gov.au/view/whole/html/inforce/current/act-1998-086
Join the conversation
Get real-time psychosocial risk strategies, legislative compliance breakdowns, and mature workforce insights delivered straight to your professional feed by Dr. Caroline Howe.