Family Practice Vol. 21, No. 2, 192-198
Family Practice Vol. 21, No. 2 © Oxford University Press 2004, all rights reserved.
Article |
Doctors' prediction of certified sickness absence
a Department of General Practice and Community Medicine, Section for Occupational Health and Social Insurance Medicine, University of Oslo, b Aust-Agder County Office of the National Insurance Service, Arendal, c HELTEF: Centre for Health Services Research, PO Box 55 No-1474 Nordbyhagen, Norway and d National Insurance Administration, Oslo, Norway
E-mail: harald.reiso{at}samfunnsmed.uio.no
Received 7 April 2003; Revised 15 October 2003; Accepted 3 November 2003.
Reiso H, Gulbrandsen P and Brage S. Doctors' prediction of certified sickness absence. Family Practice 2004; 21: 192–198.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Background. Doctors' ability to predict the duration of their patients' certified sickness absence is important for follow-up efforts aimed at patients with increased probability of long-term absence.
Objectives. The aim of this study was to examine the accuracy of doctors' predictions of their patients' sickness absence status 4 weeks ahead, and which factors were associated with it.
Methods. A questionnaire survey was carried out in primary health care concerning 796 patients certified sick within 140 days after the start of absence. The episodes of absence were labelled short-standing (up to 2 weeks) and long-standing (from 3 to 20 weeks), at the time of consultation. The doctors' prediction of the patients' absence status 4 weeks ahead, diagnoses, work ability, clinical information sources used and the presence of non-medical factors that could have influenced the doctors' work ability assessments were collected. The predictions were compared with the patients' absence status 4 weeks later by positive predictive values (PPVs) for the statements ‘returned to work’ and ‘still certified sick’. Factors associated with the accuracy of the predictions were analysed by multiple logistic regression analyses.
Results. The doctors accurately predicted return to work in 84% [95% confidence interval (CI) 79–87] of the cases in short-standing episodes, and in 53% (43–62) in long-standing episodes. The corresponding PPVs for still certified sick were 72% (62–80) and 91% (85–94). In short-standing episodes, the doctors' probability of making accurate predictions was higher for respiratory disorders [odds ratio (OR) 2.84; 95% CI 1.36–5.90], than for the reference category ‘all other disorders’, and lower for mental disorders (0.46; 0.24–0.89). In long-standing episodes, the probability was lower for musculoskeletal disorders (0.33; 0.12–0.86) and injuries (0.12; 0.03–0.48). Neither the age nor gender of patients or doctors, nor the degree of work ability reduction, nor other factors were associated with the accuracy of the predictions.
Conclusions. The doctors' predictions were highly accurate for return to work in short-standing episodes, and for still certified sick in long-standing episodes. Diagnoses were associated with the accuracy; other factors, including the doctors' work ability assessments, were not.
Keywords. Prediction, sick listing, sickness absence, sickness certification, work ability.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Doctors act as gatekeepers for sickness benefits, in accordance with sickness benefit legislation.1–3 Twenty per cent of consultations in Norwegian general practice involve certified sickness absence.4–6 In these consultations, the doctors assess the importance of medical problems and the consequences these problems have for the patients' work ability. In Norway, employees can certify themselves sick 12 days per year, but usually not for more than three consecutive days, without a sickness certificate issued by a doctor. Employers pay for the first 16 calendar days of certified sickness absence (14 days was the rule until April 1998), and the National Sickness Benefit Scheme covers further absence, to a maximum of 365 calendar days. Sickness benefits can be graded between 100 and 20%. The benefits equal ordinary salary, up to a maximum of NOK 341 166 per year (
£28 600, 
41 400) (at 1 May 2003). Doctors' ability to predict the length of sickness absence is important in enabling them to initiate therapy and rehabilitation for patients with increased probability of long-term sickness absence.1,7–9 Norwegian doctors are requested by the National Insurance to estimate how many weeks their patients will be certified sick. We do not know how accurate doctors' predictive assessments are, or how medical factors are associated with them.10
Primary health care doctors have experience in assessing their patients' need for certified sickness absence.5,11 A recent study in Norway has indicated that GPs are good at identifying their patients' work-related problems,12 and at evaluating the patients' work disability.13 Doctors thus probably have a good idea of how long their patients will be certified sick, based on knowledge about the patients and their diseases. Factors associated with predictions may be the patients' age and gender, the patients' work ability, the doctors' diagnostic and medical assessments, and the doctors' age and gender.8–10
This study aimed to examine the accuracy of doctors' predictions of their patients' certified sickness absence status 4 weeks ahead, and which factors were associated with the accuracy of the predictions.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Data collection
Data collection took place from January to April 1996 in the county of Aust-Agder, southern Norway.3,8 The county had 100 211 inhabitants in 1996. All GPs and company doctors in the county were invited to participate in the study. The participating doctors recorded episodes of certified sickness absence made during office hours.
The study had a two-stage design,14 the inclusion of doctors representing the first stage of data collection, and the consultations between doctors and patients the second. Patients and doctors were assured that participation in the study would not interfere with absence status. For episodes recorded more than once, only the first was included in the analysis.
Variables
The doctors predicted their patients' absence status 4 weeks ahead by the question: ‘What do you believe the patient's situation concerning certified sickness absence will be in 4 weeks?’, with the options: ‘probably full-time certification’, ‘probably part-time certification’ and ‘probably returned to work’. Probably full-time and part-time certifications were recoded as ‘probably still certified sick’.
The doctors classified the main sickness certification diagnoses according to ICPC (International Classification of Primary Care).15 Musculoskeletal, mental, respiratory disorders and injuries are important diagnostic groups in sickness absence.10,16 The diagnoses were therefore recoded into those groups, and ‘all other disorders’.
The doctors assessed the patients' work ability by the 5-point question: ‘To what degree is the patient's ability to perform his/her ordinary, remunerative work reduced today?’ The answer categories were: ‘very much reduced’, ‘much reduced’, ‘moderately reduced’, ‘not much reduced’ and ‘hardly reduced at all’. Due to few observations in some of the answer categories, work ability assessed as moderately, not much or hardly reduced at all was recoded as ‘moderately reduced’, while much and very much reduced were kept unaltered.
The doctors recorded the clinical information sources they used when assessing work ability as: ‘the patient's self-reported work ability only’, ‘mainly self-report, supplemented by clinical findings’, ‘mainly clinical findings, supplemented by self-report’ and ‘clinical findings alone’. The first two were recoded into ‘statements by patients’ and the last two into ‘clinical findings’.
The doctors also evaluated whether non-medical factors could have influenced their work ability assessments, by the question: ‘Do you consider that non-medical factors are responsible for the patient's reduced work ability?’ The answer categories were: ‘yes, certainly’, ‘yes, probably’, ‘no, but uncertain’ and ‘absolutely not’, dichotomized into ‘yes’ and ‘no’.
The duration of the episodes of absence was collected from the National Sickness Benefit Register.17 No episodes were lost to follow-up.
The sample
In a sample of 913 episodes of certified sickness absence,8 909 concerned episodes for which doctors had answered the predictive question. This study focused on prediction of duration early in the course of absence. Episodes with duration of >20 weeks at the time of the assessments were therefore excluded (113 episodes).
The study sample included 796 questionnaires concerning the absence episodes of 796 patients. Fifty-two of 91 invited doctors (57%) participated, 49 GPs and three company doctors. The doctors recorded an average of 15 episodes each (minimum nine, maximum 23). The doctors' age ranged from 28 to 78 years (mean 44 years), and 13 (25%) were female.
The episodes of absence were labelled short-standing (up to 2 weeks, n = 486) and long-standing (from 3 to 20 weeks, n = 310), depending on their duration at the time when the consultations occurred, i.e. from the start of absence to the time of the predictive assessments.
The study was approved by the Regional Ethics Committee for Medical Research, the Norwegian Data Inspectorate and the Legal Affairs Division in the National Insurance Administration.
Analyses
Positive predictive values (PPVs), with 95% confidence intervals, for the doctors' predictive assessments versus the patients' actual absence status 4 weeks ahead were estimated. Design-based analyses, with the doctors as the primary sampling units, were performed using STATA® software. The analysis adjusts for the fact that the assessments of different patients made by a single doctor are not independent.
Multiple, design-based, logistic regression analyses were performed in a backward, stepwise procedure, with the accuracy of the prediction (right/wrong) as outcome variable. The variables with the highest P-values were excluded one by one, keeping the patients' age and gender, and diagnoses, in all steps. Unadjusted results for each variable, and adjusted, reduced models are presented.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The patients' age ranged from 16 to 66 years (mean 40 years), and 55% were women (Table 1).
|
Musculoskeletal (33%) and respiratory disorders (27%) were most common in short-standing episodes, musculoskeletal (48%) and mental disorders (22%) in long-standing episodes.
A total of 125 (41%) of the 308 patients with musculoskeletal disorders had back problems; 76 (25%) had neck/shoulder problems. Eighty-three (62%) of the 134 patients with mental disorders had depressive problems. Thirty-nine (27%) of the 147 patients with respiratory disorders had influenza, and 32 (22%) had upper respiratory tract infections, whereof 38 of 39 and 30 of 32 were limited to short-standing episodes. Twelve (75%) of the 16 patients with respiratory disorders in long episodes had pulmonary problems, such as asthma. Twenty-two (39%) of the 56 injuries concerned sprain and bruises, and 13 (23%) fractures, whereof 14 of 22 and four of 13 were limited to short-standing episodes.
The doctor's work ability assessments were based on clinical findings in 33% of the cases in short-standing episodes, and in 24% in long-standing episodes.
The doctors expected that 82% of the patients would return to work within 4 weeks in short-standing episodes, and 32% in long-standing episodes. Seventy-four per cent of the patients had actually returned to work 4 weeks after the predictive assessments in short-standing episodes, and 23% in long-standing episodes (Table 2).
|
The PPVs for the doctors' prediction of returned to work were 84% in short-standing episodes, and 53% in long-standing episodes. The PPVs for still certified sick were 72 and 91%. Analyses omitting the three company doctors yielded equal PPVs.
The results of the multiple logistic regression analyses for short- and long-standing episodes are shown in Table 3.
|
In short-standing episodes, the probability of the doctors making accurate predictions was higher for respiratory disorders, and lower for mental disorders, both compared with the category ‘all other disorders’. In long-standing episodes, the probability was lower for musculoskeletal disorders and injuries. Neither the age nor the gender of patients or doctors, nor the degree of work ability reduction, nor other factors were associated with the accuracy of the predictions.
Analyses with accurate prediction of returned to work and still certified sick as outcomes were also performed. The probability that doctors made accurate predictions of returned to work was lower for mental disorders, lower with increasing patient age in short-standing episodes, and lower for musculoskeletal disorders in long-standing episodes (not shown in the table). The probability that doctors made accurate predictions of still certified sick was not significantly associated with any variables in short-standing episodes, and was lower for injuries in long-standing episodes (not shown in the table).
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Diseases have typical courses. Most respiratory infections are in remission 2 weeks after their onset; depression lasts for several weeks. Although doctors know this, it is not always easy to predict the course of an illness for individual patients. On the other hand, given the purpose of identifying sickness absence of expected long duration, the doctors' statements were accurate in 91% of the cases in long-standing episodes. Our findings indicate that doctors accurately identify patients with a poor prognosis of returning to work already after 2 weeks of sickness absence. However, the ability to predict the further course of the illness is more difficult when the episodes have already lasted >2 weeks. While the PPVs in short-standing episodes are satisfactory (Table 2), this is not the case for the ability to predict return to work in long-standing episodes. In fact, a PPV of 53% means that the doctors might as well have tossed a coin. Given their knowledge and experience, what can explain this finding?
The overall result is strongly influenced by the large number of musculoskeletal disorders, which obviously cause the doctors problems in predicting the patients' future work ability. However, the doctors' problems are not confined to this diagnostic group. All specified disorders in this study gave much lower estimates of predictive probability than the reference diagnosis in long-standing episodes, even if the estimates were not significant for diagnostic groups other than musculoskeletal disorders and injuries (Table 3). In mental disorders, the doctors may make positive statements about return to work as part of a motivation exercise. In predicting the course in injuries, there was a striking difference in the probability for the doctors to make accurate predictions in short-standing episodes [odds ratio (OR) 3.75; 95% confidence interval (CI) 0.88–16.1] and long-standing episodes (OR 0.12; 0.03–0.48). Patients with injuries returned to work more quickly than the doctors expected in long-standing episodes, as the probability for the doctors to make accurate predictions of still certified sick was also low. Our findings indicate that diagnoses alone do not give the doctors enough information to assess prognoses in long-standing episodes of sickness absence. Good prognoses, here return to work, are hard to identify after 2 weeks of certified sickness absence.
Prior duration may be a predictor of outcome, because recovery (here return to work) may be easy to predict in short-standing respiratory disorders, and conversely that pessimistic predictions (here still certified sick) will be more accurate in prolonged illnesses, such as depression. It may thus be argued that our findings are obvious, more like self-fulfilling prophecies, due to the high number of respiratory infections in short-standing episodes. However, doctors do not necessarily know how self-limiting illnesses are at the time they assess their patients' future duration of absence. Not all respiratory disorders in this study were of short duration, and musculoskeletal and mental disorders were evenly distributed in short- and long-standing episodes (Table 1).
Had the time span for the prediction of the patients' absence status been 1 week ahead, instead of 4 weeks, that would of course reduce the probability of the doctors making accurate predictions of return to work.
Being familiar with the presence of non-medical factors does not seem to help the doctors either. We anticipated that the ability to give accurate predictions would improve if no non-medical factors were present, or if the doctors' work ability assessments were based on clinical findings rather than statements by the patients. We could not confirm these hypotheses, even if non-medical factors apparently could make a difference in short episodes if not adjusted for diagnoses. Another a priori hypothesis, that the doctors' assessments of the patients' work ability would be associated with accurate predictions, was also not confirmed. Temporal consid-erations made by the doctor could explain this. Assessment of the patients' work ability today is different from assessing his/her future function.
The doctors' answers are not just predictions, but also plans of action concerning the future absence of the patients. The doctors had the authority to determine whether the predictions would be accurate or not. The substantial differences between PPVs for return to work in short- and long-standing episodes suggest that the doctors have been overruled by the course of the illness, or the opinion of the patients, and particularly so in long-standing episodes. One may question whether this mutual understanding between doctors and patients for longer duration of absence than the doctors had expected expresses realistic work ability assessments made by the doctors, motivation exercises towards their patients, or reflects that the doctors' predictions contain elements of the patients' motivation for work resumption.
The predictive question may not only reflect an assumed duration of sickness absence. The ability of doctors to predict still certified sick in long-standing episodes (PPV of 91%) could reflect the dialogue and negotiations that take place between patients and doctors in consultations concerning sickness absence. Patients have certain expectations when they decide to consult their doctors about sickness absence. It is rare that doctors do not meet the patient's demands for a certification. Ninety-five per cent of patients receive certifications by their doctors when the initiative is taken by the patients.5 Even where GPs do not recommend sickness certification, certificates are issued in 87% of the cases.11 The patients influence the doctors' decisions.18
The last 30 years have gradually brought a shift towards patient-centred consultations.19 This way of thinking probably has influenced gatekeeper functions, such as the authorization of sick leaves. We found that the doctors based their work ability assessments on statements by the patients in 71% of cases (Table 1). There is reason to question this practice. Patients perceiving themselves as work disabled more frequently have health-affecting psychosocial problems that are not work related.20 With a mixture of work-related and other problems that the doctors cannot cure, the apparent relief given by issuing sickness certificates could be ‘an easy way out’ for both doctor and patient.
Method
Patients in this study were not significantly different from other patients certified sick recorded in the sickness benefit register of Aust-Agder in the corresponding period, with regard to age, gender and diagnoses.3,8 Participating doctors had more female patients certified sick than non-participating doctors, 55% versus 47%. Because the gender of the patients did not affect the main conclusions, we do not think that this difference is of importance. The most common diagnostic groups were musculoskeletal, mental and respiratory disorders, as found by others previously.16
The National Sickness Benefit Register serves as an accounting system for the payment of sickness benefits, and is therefore revised, audited and quality controlled. The register ensures good follow-up of the patients.
Episodes with short and long duration at the time of the predictive assessments differentiate between new episodes of sickness absence and the prolongation of ongoing episodes. This says nothing about how long the duration of the episodes eventually will be. However, one may assume that the doctors have more knowledge about the patients, and thus the probable duration of absence, in prolonged episodes.
Because the PPV for return to work was 84% in short-standing episodes, increased follow-up efforts should not be targeted to that group of patients certified sick.
|
Acknowledgments |
|---|
The Norwegian Ministry of Health and Social Affairs supported the study.
|
References |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
1 Bloch FS, Prins R (eds). Who Returns to Work and Why. A Six-country Study on Work Incapacity and Reintegration. New Brunswick: Transaction Publishers; 2001.
2 Bonner D, Hooker I, White R. Non Means Tested Benefits: The Legislation, Statutory Sick Pay (General) Regulations. London: Sweet and Maxwell; 1998.
3 Reiso H, Ny
rd JF, Brage S, Gulbrandsen P, Tellnes G. Work ability assessed by patients and their GPs in new episodes of sickness certification. Fam Pract 2000; 17: 139–144.
4 Gulbrandsen P, Brage S. Life situation as a reason for sick leave. Tidsskr Nor Lægeforen 1998; 118: 2463–2466 [in Norwegian, English summary].
5 Larsen BA, Førde OH, Tellnes G. Physician's role in certification for sick leave. Tidsskr Nor Lægeforen 1994; 114: 1442–1444 [in Norwegian, English summary].
6 Rutle O. An Analysis of Encounters in General Practice. Oslo: National Institute of Public Health; 1983: Vol 1: 178 [in Norwegian, English summary].
7 http://www.dep.no/sos/norsk/publ/utredninger/NOU/ Governmental report NOU 2000: 27, Sickness absence and disability pension, an including working-life [in Norwegian] (accessed 13 February 2003).
8 Reiso H, Nyrd JF, Brage S, Gulbrandsen P, Tellnes G. Work ability and duration of certified sickness absence. Scand J Public Health 2001; 29: 218–225.[CrossRef][Web of Science][Medline]
9 Tellnes G. Sickness certification in general practice: a review. Fam Pract 1989; 6: 58–65.
10 Söderberg E, Alexanderson K. Literature Study—Published Studies of the Interface Between Medical Practice and Assessments of Social Insurance Legislation. Försäkringsmedicinskt centrum, Rapport 2001: 1. Linköping; 2001 [in Swedish].
11 Englund L, Svärdsudd K. Sick-listing habits among general practitioners in a Swedish county. Scand J Primary Health Care 2000; 18: 81–86.[CrossRef][Web of Science][Medline]
12 Gulbrandsen P, Hjortdahl P, Fugelli P. General practitioners' knowledge of their patients' psychosocial problems: multipractice questionnaire survey. Br Med J 1997; 314: 1014–1018.
13 Gulbrandsen P, Fugelli P, Hjortdahl P. General practitioners' knowledge of their patients' socio-economic data and their ability to identify vulnerable groups. Scand J Primary Health Care 1998; 16: 204–210.[CrossRef][Web of Science][Medline]
14 Lemeshow S, Letenneur L, Dartigues JF, Lafont S, Orgogozo JM, Commenges D. Illustration of analysis taking into account complex survey considerations: the association between wine consumption and dementia in the PAQUID study. Am J Epidemiol 1998; 148: 298–306.
15 Brage S, Bentsen BG, Bjerkedal T, Nygrd JF, Tellnes G. ICPC as a standard classification in Norway. Fam Pract 1996; 13: 391–396.
16 Tellnes G, Svendsen KOB, Bruusgaard D, Bjerkedal T. Incidence of sickness certification. Proposal for use as a health status indicator. Scand J Primary Health Care 1989; 7: 111–117.[Medline]
17 Brage S, Nygrd JF, Tellnes G. The gender gap in musculoskeletal-related long term sickness absence in Norway. Scand J Soc Med 1998; 26: 34–43.[Web of Science][Medline]
18 Ainsworth-Vaughn N. Patients Claiming Power in Doctor–Patient Talk. New York: Oxford University Press; 1998.
19 Stewart M. Patient-centered Medicine: Transforming the Clinical Method. Thousand Oaks (CA): Sage Publications; 1995.
20 Gulbrandsen P, Hjortdahl P, Fugelli P. Work disability and health-affecting psychosocial problems among patients in general practice. Scand J Soc Med 1998; 26: 96–100.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  P. Giri, J. Poole, P. Nightingale, and A. Robertson Perceptions of illness and their impact on sickness absence Occup. Med., December 1, 2009; 59(8): 550 - 555. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. van Dijk, W. Hogervorst, G. t. Riet, and F. van Dijk A protocol improves GP recording of long-term sickness absence risk factors Occup. Med., June 1, 2008; 58(4): 257 - 262. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||