Skip Navigation


Occupational Medicine Advance Access originally published online on January 22, 2008
Occupational Medicine 2008 58(2):99-106; doi:10.1093/occmed/kqm141
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
58/2/99    most recent
kqm141v2
kqm141v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Wilford, J.
Right arrow Articles by Macdonald, E. B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wilford, J.
Right arrow Articles by Macdonald, E. B.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The Author 2008. Published by Oxford University Press on behalf of the Society of Occupational Medicine. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org

Predicting job loss in those off sick

Jane Wilford1, Alex D. McMahon2, Jean Peters3, Simon Pickvance4, Alison Jackson1, Lindsay Blank3, David Craig5, Alan O'Rourke3 and Ewan B. Macdonald1

1 Division of Community Based Sciences, Faculty of Medicine, Public Health and Health Policy Section, University of Glasgow, Glasgow, UK
2 Robertson Centre for Biostatistics, University of Glasgow, Glasgow, UK
3 Section of Public Health, School of Health and Related Research, University of Sheffield, Sheffield, UK
4 School of Health and Related Research, Regent Court, 30 Regent Street, Sheffield S1 4DA, UK
5 Physiotherapy Services, Sheffield PCT, Fulwood House, Fulwood Road, Sheffield S10 3TH, UK

Correspondence to: Ewan B. Macdonald, Division of Community Based Sciences, Faculty of Medicine, Public Health and Health Policy Section, University of Glasgow, Glasgow, UK. Tel: +141 330 3719; fax: +141 330 4038; e-mail: e.b.macdonald{at}clinmed.gla.ac.uk.


   Abstract
Top
 Abstract
Introduction
 Methods
 Results
 Discussion
 Funding
 Conflicts of interest
 References
 
Background Evidence shows incapacity benefit claimants (those off sick >26 weeks) are at greatest risk of long-term job loss.

Aim To develop a screening tool to select those at risk of job loss, defined as failure to return to work among those off sick. The screening tool was for use in the Job Retention and Rehabilitation Pilot of the Department for Work and Pensions.

Methods A literature review identified risks for long-term incapacity and job loss as multifactorial [1]. Potential predictors for return to work were then assembled into a set of questions and tested by a prospective study in general practice surgeries and a retrospective study of occupational health records of local authority employees referred for sickness absence management, using univariate and multivariate logistic regression analysis.

Results Univariate logistic regression analysis of the retrospective study produced odds ratios with 95% confidence intervals for each question (where P ≤ 0.05) and a C-index was then constructed for their predictive power. Five questions holding the greatest predictive power were subjected to multivariate analysis and in the final model had a high C-index of 0.90 (0.5 = no predictive power, 1.0 = perfect prediction). They formed the screening tool. The questions cover self-assessment of ability to return to work after current sick leave, of ability to do current job in 6 months' time, sick leave in past year, current age and whether awaiting a consultation or treatment.

Conclusion A screening tool identifying those most at risk of job loss has been produced.

Keywords      Barriers to return to work; risk factors for job loss; sickness absence >6 weeks


   Introduction
Top
 Abstract
 Introduction
Methods
 Results
 Discussion
 Funding
 Conflicts of interest
 References
 
Most people taking sickness absence are away from work for a short time. UK data from 1995, the last reliable statistics on sickness absence at the time of our study, showed that each week ~17 000 people reached their 6th week of sickness absence and of these 5 out of 6 returned to work [1]. However, ~3000 people a week moved onto incapacity benefit (IB) after 28 weeks. Current data on IB show that ~50% of IB claimants have a claim that lasts for 5 years or more [Department for Work and Pensions (DWP) in-house analysis].

The joint group of the Advisory Committee on Disability Employment and Training suggested in 1999 that the Government should consider how to help such employees through rehabilitation programmes and other forms of assistance [2]. The response was a Job Retention and Rehabilitation Pilot (JRRP) focused on people in employment but at real risk of not working again [3]. However, those at such real risk were not identifiable from existing statistical and administrative sources.

This paper reports research commissioned by the DWP to develop a screening tool for the JRRP identifying those off sick for 6 weeks or more at risk of job loss because of absence from work for a health condition, injury or impairment [4].

Neither occupational health studies nor clinical rehabilitation studies in fields like cardiology and back pain have produced a set of questions or single questionnaire to estimate risk factors for job loss from a multifactorial causative base [1]. Yet the literature review led to the conclusion that job loss risks are multifactorial. The challenge was how to cover a range of factors through a set of questions that would include sensitive and specific factors, be non-intrusive, acceptable to patients and practicable in a telephone interview or short consultation.


   Methods
Top
 Abstract
 Introduction
 Methods
Results
 Discussion
 Funding
 Conflicts of interest
 References
 
The screening tool was developed in two phases: first, a literature review identified concepts reflecting possible risk factors for prolonged sickness absence and therefore job loss, failing to identify any existing screening tool selecting those at such risk. The second phase developed questions for the screening tool and tested and validated these in a small general practice population of those off sick and through examination of occupational health records of a population of local authority employees.

The literature review was undertaken searching MEDLINE, EMBASE, PsycINFO, Cinahl, Health Information Management Consortium, British Nursing Index, English National Board Healthcare database, International Bibliography of Social Science, Web of Science (Science Citation Index, Social Science Citation Index) and Business Source Premier (incorporating Business Source Elite). Citation sources around the use of known instruments, e.g. Finnish Work Ability Index [5] and New Zealand Yellow Flags [6] were searched. Key journals were also hand searched. All references from papers identified in the original search strategy were searched for additional evidence. An author search was conducted. Inclusion criteria were the period 1985–2001, adults aged 16–65 (working age); exclusion criteria were information in editorials, abstracts and individual case studies and absence from work for more than 6 months as well as papers not recording adequate information on return to work.

A set of concepts was identified, a series of terms for each assembled and a combination of free text and thesaurus terms then identified and used in the search (Table 1). Each concept and its block of terms were searched alone and combined into groups of three concepts and searched again to eliminate overlap and duplication. Research teams in Glasgow and Sheffield each had two reviewers read and critically appraise all papers, relevant information being extracted into a summary evidence/extraction table. Meta-analysis was considered but rejected as data were too heterogeneous.


View this table:
[in this window]
[in a new window]

  		Table 1. Concepts and Search terms used in the literature review.

 
From the literature review, a large number of assessment measures, scales and questionnaires were identified. But none predicted job loss as a result of sickness absence of 6 weeks duration or more, nor did any address the multifactorial nature of risk. However, devised to deal with specific problems, they did inform the selection of concepts reflecting possible risk factors for prolonged sickness absence and therefore job loss. The concepts retained for testing included physical health relating also to physical demands of the job and health behaviour (smoking, alcohol consumption, exercise); mental health and family history of mental ill health as well as psychological and mental demands of the job; trauma and injury (whether work related, severity, whether litigation underway); pain and beliefs about illness and pain (health locus of control, illness beliefs, catastrophizing); coping abilities, social support; job, work and workplace characteristics; job satisfaction and commitment; job strain, stress and the demand–control model and contractual conditions, sickness absence and self-reported work ability. Demographics, socio-economic status, education and qualifications were also included.

Questions designed around these concepts formed the screening tool questionnaire. For some of the questions, validated questions were used from other questionnaires such as the Labour Force Survey (Office of National Statistics) [7]. Altogether, 30 concepts were addressed in 117 questions. MultiCentre and Local Regional Ethics Committee approval was sought and obtained.

Using a computer-assisted telephone inquiry, this questionnaire was piloted prospectively on a sample of workers off sick referred by general practitioners. The study had planned to contact 1200 participants off work between 6 and 18 weeks through general practices in Glasgow and Sheffield. However, due to the MultiCentre Regional Ethics Committee requiring formal written consent from participants obtained in person by a general practitioner, recruitment was disappointing. Numbers were too small for statistical analysis but sufficient to test the acceptability and face validity of the questions (n = 141).

To pilot the questionnaire on an adequate number of cases, 1350 consecutive occupational health records of first patient consultations of a local authority were scrutinized. Jobs ranged from manual (gardeners, home support workers, cleaners) through clerical, administrative and executive to professional (teachers and social workers). All had been referred to a contracted occupational health service for sickness absence over the preceding 5 years. The records included physician consultations general practitioners' or treating specialists' reports.

The questions were straightforward requests for facts (How many weeks off work did you have in the past year?), where answers were available from both employee and employers' records or they were asking the employee's opinion (Is pain a reason why you are off work?), where the answer was recorded by the occupational physician as a statement (e.g. ‘back pain too severe to return to work at present’). The data gathered were then put forward for statistical analysis.

The outcome of interest in this study was ‘failure to return to work’. Each variable forming a question in the questionnaire used in the retrospective study was examined first of all by frequency tabulations. Some variables were continuous and were pooled into categories that tried to represent natural groupings, while also attempting to contain reasonable numbers of subjects in each category. The individual predictive power of each variable was assessed by univariate logistic regression models [8]. The results for each question were then expressed in the form of odds ratios with 95% confidence intervals. The predictive power of an individual question and a comparison of sensitivity and specificity were assessed using the ‘C-index’ [9, 10]. The C-index is equal to the area under the curve of a receiver operating characteristic (ROC) plot, a term that comes from the operational research field and is used in the analysis of diagnostic tests in medicine [11]. Note that a variable with no predictive ability has a C-index of 0.5. And a variable with perfect predictive power has a C-index of 1. The C-index was used to reduce the dimension of the questionnaire, which contained hundreds of category options in order to produce a model to predict the probability of ‘returning to work’. All analyses were carried out using the SAS statistical package, Version 8.02.

Only those questions that were univariately significant, using an inclusion criterion of P ≤ 0.05, were considered for further examination. It is possible that two or more questions, although univariately powerful, may actually capture the same concept. In this case, the best of a ‘group’ can be taken to represent the concept without too much loss of predictive power. To this end, the natural approach was to use the small groups of questions with a common stem in the questionnaire. These groups were then analysed using forward stepwise logistic regression. The variables within the group that were independently significant of their peers were then considered for further examination. Next, the shortlisted questions were analysed using a forward stepwise logistic regression procedure.

The sample of local authority employees off sick and seen at occupational health clinics totalled 1350. The original statistical analysis was run on this full sample. However, to produce the risk score algorithm for use in the tool, it was necessary to rerun the analysis on a sample that fitted with the JRRP population parameters (people of working age off sick for between 6 and 26 weeks) eliminating all employees who had been seen after >6 weeks sickness absence. The reduced sample totalled 741. (Most of the reduction in numbers was due to exclusions through length of sickness absence. Management referrals to occupational health were commonly only made after 6 months off work, although this is now changing.) In the new model with reduced numbers, age category odds ratios were changed because those working on beyond retirement age had now been excluded. However, it was decided to include age as probably significant. The same variables were used in the new model, thus producing slightly different odds ratios.

The results of the final model were presented in the form of odds ratios and confidence intervals, as before, except that these were now adjusted for each other (multivariate logistic regression). The C-index and a ROC plot were presented [11].


   Results
Top
 Abstract
 Introduction
 Methods
 Results
Discussion
 Funding
 Conflicts of interest
 References
 
We analysed both the full dataset of 1350 employees and the reduced set of 741 employees. The results for both were similar and we present the results for the reduced set of 741 as requested by the DWP. As the range of ages in the reduced set did not include older people working past retirement age, the significance of age was reduced; however, we were required to include age in the final model for operational reasons.

The distribution of the number of weeks of sickness absence is shown in Table 2. Overall, 266 of 741 subjects returned to work (36%). Univariate logistic regressions for each question are summarized in Table 3. They are shown in decreasing order of the C-index (predictive ability). The C-index varied from 0.85 (Do you think that you will be able to return to work after your current sick leave?) to 0.50 (Do you get sick pay on this job?).


View this table:
[in this window]
[in a new window]

  		Table 2. Frequency of return to work by number of weeks sickness absence between 6 and 26 weeksa

 

View this table:
[in this window]
[in a new window]

  		Table 3. Predictors of the outcome return to work (univariate analysis)

 
The candidates were then shortlisted in their small groups as described previously. Some single-variable groups progressed straight through to the next round (for example, age, weeks off sick, whether work made the symptoms worse, a family history of depression, currently waiting for treatment, a question on social support and a question on qualifications). Also significant from the multivariable groups were ill health involving the digestive system or kidney disease, pain, what is the most difficult problem that will make it harder for you to go back to work, type of job (manual), hours worked and whether on sick pay from the employer, numbers of weeks off sick in the past year, the patient's prediction of likelihood of return to work after current sick leave, ability to do own job in 6 months' time, whether the subject lives with a spouse or partner, consumption of units of alcohol per week and undertaking of an exercise programme.

These variables were then entered into the final selection algorithm (forcing in age as described). The final model is shown in Table 4 and had an extremely high predictive ability with a C-index of 0.90. The questions that were included in the final model were as follows:

  1. Patient's prediction of likelihood of return to work after current sick leave (question: Do you think you will be able to return to work after your current sick leave?).
  2. Patient's prediction of ability to do current job in 6 months' time (question: Do you believe that from the standpoint of your health you will be able to do your current job in 6 months' time?).
  3. The number of weeks off sick in the past year.
  4. Whether they were waiting for a consultation or treatment for their health condition.
  5. Age (more likely to return to work if younger in univariate model).


View this table:
[in this window]
[in a new window]

  		Table 4. Univariate-unadjusted and multivariate-adjusted results for the chosen predictors of return to work: final model

 
Finally, the ROC plot that corresponds to this model is shown in Figure 1.


Figure 1
View larger version (10K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Figure 1. An ROC curve is a plot of the true positive rate against the false positive rate for the different possible cutpoints of a diagnostic test. An ROC curve demonstrates several things. It shows the trade-off between sensitivity and specificity, it demonstrates the closer the curve following the left hand border and then the top border of the ROC space, the more accurate the test; the closer the curve comes to the 45 degree diagonal of the ROC space, the less accurate the test.

 

   Discussion
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
Funding
 Conflicts of interest
 References
 
The literature review demonstrated a lack of robust research data into questions regarding sickness absence and job loss, such as effects of chronic illness (now potentially qualifying under the Disability Discrimination Act for reasonable adaptations to enable employees to remain at work) and the impact of neuroses on attendance at work (most research on mental health and work concerned unemployment and the severely mentally ill, often seen at specialist clinics).

The questions emerging as strongest predictors of job loss were patients' own assessments of their ability to work, length of sickness absence in the previous 12 months, age, whether the patient was waiting for treatment or consultation and patients' perceptions of barriers to their return to work. The predictive importance of perceptions of such barriers (see Table 3) supports research in Finland showing job tasks rather than assessment of own health problems to be important in determining patients' views of their own work ability.

This is the first study to have looked at multifactorial risks for long-term sickness absence and job loss to have measured the risks and devise a questionnaire to select the most vulnerable members of the ill work force. Previous work centred on risks in specific medical conditions. We have approached the subject using the biopsychosocial model of disability and rehabilitation.

The sample used to test the risk factors was representative of a population of public sector employees in Scotland. They were not volunteers but management-selected employees being referred for sickness absence management. There is both a possible bias and a likely confounding factor built into this sample as it represents management referrals by a local authority. The bias is that management would refer workers to OH where it was predicted that workers were unlikely to return to their jobs. The further confounding factor is that local authority workers' generous sick pay entitlement leads to a return to work shortly before 6 months sickness absence when sick pay is reduced to half pay. All those who had been off work for over 6 months were eliminated in the final model when the population sample had been reduced to 741. This reduction affected the predictive power of the age variable.

The questions were tested not during interviews with the patients themselves, but retrospectively by reference to written case records. All patients were seen by an occupational physician, data extracted from case histories, consultation notes and physical examination, correspondence with medical practitioners and management, culminating in the occupational physicians' diagnosis and advice on fitness for work. Fitness for work decisions are arrived at through the weighing of evidence regarding health and the demands of the job, including the patient's own views, often recorded verbatim. Case note data extraction was quality peer reviewed by three specialist occupational physicians. Some questions could not be answered through examination of the records and thus there was missing data which affected the predictive power of a number of questions. For example, smoking and alcohol histories were not recorded nor was social history relating to family members and dependents.

Due to the short lead time available to test the questions in real time (the Job Retention Pilots were scheduled to commence nationally in April 2003), it was decided to run the final model as the operational screening tool to detect those at risk of job loss who would be eligible for entry into the JRRPs. (JRRP insisted that only those off work on medical certificates with a real risk of job loss were eligible for support using public funds.) However, in order to run a prospective trial of all variables identified as relevant through the original literature review, the remaining best variables were incorporated into the live JRRP screening process. The final results of the JRRP were published in 2006 [12]. However, there is no published analysis addressing whether additional variables collected during the JRRP have emerged as being of greater predictive value for job loss than the original set. (The intellectual property rights of this screening tool reside with DWP.)

The study contributes evidence suggesting that listening to patients assess their own work ability is the best diagnostic tool for determining risk of job loss. By enquiring into the patients' perceptions of barriers to their returning to work, occupational health and human resources practitioners may operate a more focused rehabilitation intervention involving adaptations by the employer and additional health or social support to the patient.

Clinicians, in primary care or occupational health, might use the tool to identify patients who require more active management to restore functional capacity. Occupational health services and employers could target those most needing reasonable adaptations under the Disability Discrimination Act or propose redeployment and retraining. The tool provides evidence of the importance of occupational health advice on such adaptations.


Key points
  • Employees' own assessment of their ability to do their job is the strongest indicator of risk of job loss.
  • Risk factors for Job loss are multifactorial and follow the biopsychosocial mode.
  • Age is associated with longer term sickness absence.

 


   Funding
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Funding
Conflicts of interest
 References
 
Department for Work and Pensions, Welfare to Work Division, Job Centre Plus.


   Conflicts of interest
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Funding
 Conflicts of interest
References
 
None declared.


   References
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Funding
 Conflicts of interest
 References
 

  1. DfEE DSS, DoH. Job Retention and Rehabilitation Generic Pilot: Bidding Pack and Guidance for Applicants (2001) Sheffield, UK.

  2. Advisory Committee for Disabled People in Employment and Training. In: Report of the Sub Group of ACDET on Job Retention (1999) London: ACDET Secretariat, DfEE.

  3. DWP Media Centre/Press Releases/ 2003 /March/JRRP. Pioneering Project Could Stop Sick Workers Losing Their Jobs. http://www.dwp.gov.uk. (13 March 2003, date last accessed).

  4. Peters J, Wilford J, Macdonald E, et al. Literature Review of Risk Factors for Job Loss Following Sickness Absence (2003) Sheffield, UK: Department for Work and Pensions Social Research Division.

  5. Ilmarinen J, Tuomi K, Klockars M. Ageing and Work Ability Index: A 10 Year Follow-Up Study of Municipal Employees (1990) Helsinki, Finland: Finnish Institute of Occupational Health.

  6. National Advisory Committee on Health and Disability and Accident and Rehabilitation and Compensation Insurance Corporation. In: Guide to Assessing Psychosocial Yellow Flags in Acute Low Back Pain: Risk Factors for Long Term Disability and Work Loss (1997) New Zealand.

  7. Office for National Statistics. In: Labour Force Survey (2002) London: ONS.

  8. Hosmer DW, Lemeshow S. Applied Logistic Regression (1989) Chichester, UK: John Wiley.

  9. Harrell FE, Lee KL, Califf RM, Pryor DB, Rosati RA. Regression modelling strategies for improved prognostic prediction. Stat Med. (1984) 3:143–152.[Web of Science][Medline]

  10. Harrell FE, Lee KL, Mark DB. Multivariate prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. (1996) 15:361–387.[CrossRef][Web of Science][Medline]

  11. Altman DG, Bland JM. Diagnostic tests 3: receiver operating characteristic plots. Br Med J. (1994) 309:188.[Free Full Text]

  12. Purdon S, Stratford N, Taylor R, Natarajan L, Bell S, Whittenburg D. Impacts of the Job Retention and Rehabilitation Pilot (2006) Sheffield, UK: Department for Work and Pensions Research Report No342.


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?



This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
58/2/99    most recent
kqm141v2
kqm141v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Wilford, J.
Right arrow Articles by Macdonald, E. B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wilford, J.
Right arrow Articles by Macdonald, E. B.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?