Return to Work After Stroke
A Follow-up Study
Background and Purpose Few studies have reported the longitudinal trend of return to work after stroke. The purpose of our study was to evaluate the longitudinal trend of proportion of patients who return to work after stroke and further to examine the predictors of return to work while taking follow-up periods into consideration.
Methods We conducted a retrospective cohort study on the association between characteristics of stroke patients at admission and return to work after first stroke, taking length of follow-up period into consideration (n=183). The patients were all younger than 65 years and were working at the time of their stroke. A follow-up questionnaire evaluated return to work and related information. Data were analyzed using the Kaplan-Meier method for curves of the proportion of return to work and Cox’s proportional hazards model for odds ratios of return to work.
Results The curve of proportion of return to work had two steep slopes, and the proportion was at a maximum at 18 months from patient admission. The adjusted odds ratios of return to work for patients with normal muscle strength versus severe weakness, without apraxia versus with apraxia, and with white-collar versus blue-collar occupations were 5.16 (P<.05), 4.16 (P<.05), and 1.43 (.05<P<.10), respectively.
Conclusions The increase of proportion of return to work after stroke was nonlinear, and this trend was referable to the social security systems available to the patients included in this study. Normal muscle strength and absence of apraxia were significant predictors of return to work after stroke. White-collar occupation showed a tendency to promote return to work.
Return to work is one of the important outcomes for stroke patients and is a manifestation of reintegration into their society. Several studies have identified factors that predict return to work after stroke.1 2 3 4 Our previous study, based on a multiple logistic regression model adjusting for potential confounding factors, indicated the following as significant predictors of return to work after stroke: normal muscle strength, absence of apraxia, and white-collar occupation.4
However, there have been two unsettled problems in these studies. First, few studies have evaluated longitudinal trend of return to work after stroke. Data on vocational outcome of stroke patients are difficult to access. Second, those studies did not take into account censored observations (time-to-failure or withdrawals), which might affect the results. In other words, each subject did not always have the same follow-up period in each study.
The purpose of our study was to evaluate the longitudinal trend of proportion of return to work after stroke according to follow-up periods. We also reexamined the predictors found in our previous study, taking censored data or follow-up periods into consideration.
Subjects and Methods
We conducted a retrospective cohort study on the association between patient characteristics at admission and return to work after first stroke, taking length of follow-up period into consideration (time-to-failure analysis).
The University of Occupational and Environmental Health Hospital serves as one of the central hospitals in the city of Kitakyushu. Stroke patients, most of whom are referred by general physicians, are admitted to the hospital. They are diagnosed by neurological signs and neuroradiological findings and receive conservative, surgical, and/or rehabilitation treatments.
Of 703 stroke patients discharged alive from the above hospital between January 1986 and December 1990, 183 were included in this study based on the following criteria: (1) first stroke, (2) diagnosis of stroke by International Classification of Diseases Ninth Revision (ICD-9) codes (cerebral hemorrhage for 431; cerebral infarction for 434; subarachnoid hemorrhage for 430; or other for 432, 433, 436, and 437), and (3) working age (18 to 64 years) and competitive employment status of the patient at the time of stroke. We excluded patients whose diagnoses were transient ischemic attack (ICD-9 code 435) or late effects of cerebrovascular disease (ICD-9 code 438) or whose occupation was classified as housewife or student.
Subjects consisted of 129 men and 54 women; there were 115 white-collar and 68 blue-collar workers. The Table⇓ shows other characteristics of the subjects.
We examined patients for factors found in our previous study as predictors: normal muscle strength, absence of apraxia, and white-collar occupation. These data were obtained from the medical record of each patient.
Apraxia was neurologically determined by medical doctors. Examination included tests of imitation of learned complex acts, use of actual objects, copy of geometric forms, arrangement of Kohs blocks in a desired design, and dressing. Considering paralysis, sensory loss, or disturbance of coordination, a patient unable to perform one or more of the above tests was regarded as having apraxia. More detailed information on the evaluation of each predictor has been reported elsewhere.4
The end point of this study was return to work after stroke, which was defined as 1 month or more of work in active employment after stroke. A minimal period of 6 months after discharge was allowed to give the subjects time to recover and be reemployed. Follow-up questionnaires were sent between August 1991 and January 1992, which included questions on employment status and related information (the initial date of return to work) after discharge.
The methods of time-to-failure analysis were used: product limit, or Kaplan-Meier curves, and Cox’s proportional hazards regression model.5 We used sas for all data analyses, including lifetest6 and phreg7 procedures.
Each predictor (maximum weakness, apraxia, and occupation) was evaluated by use of the product limit curve. This curve shows the percentage of subjects who have experienced the event in question (successful return to work in this study) during the follow-up period.
Next, we used the length of follow-up as the dependent variable and the specified predictors as independent variables in Cox’s proportional hazards regression model. This multivariate analysis computed odds ratios to measure the strength of the association of each predictor with return to work, adjusting for the effect of confounding by other factors in the model.
The length of follow-up was defined as follows: for 75 of 85 censored case subjects who were followed up but did not return to work, it was the interval between admission and the date of response to the questionnaire; for cases lost to follow-up (the remaining 10), it was the interval between admission and discharge (patients in the hospital had not been returning to work). For 98 noncensored cases, ie, patients who returned to work successfully, we regarded the length of follow-up as the interval between admission and the date of return to work.
The curve of proportion of return to work was nonlinear. This curve had two steep slopes in the next periods: (1) during the first 6 months (0 to 180 days) after admission and (2) from 12 to 18 months (360 to 540 days) after admission (Fig 1⇓). The proportion of return to work at 12 and 18 months of follow-up was 0.4 and a little over 0.5, respectively. These two steep slopes in the curve were also observed for each predictor (Figs 2 through 4⇓⇓⇓).
The proportion of return to work after stroke was significantly higher for patients with less impaired muscle strength (Fig 2⇑). The proportion of return to work of patients without apraxia was significantly higher than that of patients with apraxia (Fig 3⇑). The proportion of return to work of patients with blue-collar occupations was significantly higher than that of those with white-collar occupations in the early follow-up period, but the two curves crossed at approximately 3 months and reversed later (Fig 4⇑). The difference between the two curves was not significant by the log-rank test, which placed more weight on longer periods, but significantly different by Wilcoxon’s test, which placed more weight on the early period.
Cox’s proportional hazards model produced the adjusted odds ratios of each predictor for return to work as follows: 5.16 (P<.05) for patients with normal muscle strength versus severe weakness, 4.16 (P<.05) for those without apraxia versus with apraxia, and 1.43 (.05<P<.10) for those with white-collar versus blue-collar occupations.
We found two steep slopes in the curve of proportion of return to work in this time-to-failure analysis (Fig 1⇑). The vocational end point occurred by 18 months after admission for most patients. The first slope during the first 6 months after admission was due to early discharge from the hospital. One reason for the second slope from 12 months to 18 months after admission seems to be that this period coincides with the expiration of patients’ sickness benefits from the public fund source in Japan. Almost all Japanese workers receive minimum sickness benefits from the public fund, generally limited to a period of 18 months. These results indicate that, unless stroke patients returned to work during the earlier period, they preferred to receive longer sickness benefits, and such benefits may delay them from returning to work. Heinemann et al2 reported that stroke patients who had a private fund source had returned to work more often than those who had a public fund source. On the other hand, such long certified absences required enormous expenditure on the part of society.8 Our study did not take into account private fund sources, but these findings illustrate that the social security systems, particularly those sickness benefits available from the public fund source for an extended time, influenced the length of the interval between admission and return to work.
The differences of the curves by the categories in maximum weakness and apraxia were statistically significant and almost constant after 6 months (Figs 2⇑ and 3⇑). As expected, the worse impairment patients had, the less they returned to work. Blue-collar workers returned to work earlier than white-collar workers in the early follow-up period. Over a longer time, white-collar workers tended to return to work more often than blue-collar workers. This implies that, unless blue-collar workers returned to work earlier, they would have more difficulty in continuing a former job.
However, the above findings, which were obtained using the Kaplan-Meier method, do not take into account confounding by other factors. Adjusting for the confounding effect of the specified predictors, Cox’s proportional hazards model showed normal muscle strength and absence of apraxia to be significant predictors. Though occupation was not a significant predictor by multivariate Cox time-to-failure analysis, white-collar occupation showed a tendency toward return to work. These findings confirm our previous results obtained by the multiple logistic model.4
The following problems may limit the generalization of results obtained in this study. First, the data in this study are based on only one hospital survey and do not represent the general workers with stroke. Second, factors not included in our model may have increased the possibility of confounding. Third, forms of work such as full time or part time were not distinguished in this study.
In consideration of these limitations, the results may be of interest and applicable in prevocational management after stroke, since similar longitudinal analyses using time-to-failure methods have not been performed before.
In conclusion, with censored data and follow-up days taken into consideration, the proportion of patients returning to work after stroke increased nonlinearly and reached a maximum at 18 months after admission. Normal muscle strength and absence of apraxia were significant predictors of return to work after stroke. White-collar occupation was not a significant predictor but showed a tendency to promote return to work.
- Received July 12, 1994.
- Revision received November 28, 1994.
- Accepted November 30, 1994.
- Copyright © 1995 by American Heart Association
Black-Schaffer RM, Osberg JS. Return to work after stroke: development of a predictive model. Arch Phys Med Rehabil. 1990; 71:285-290.
Saeki S, Ogata H, Okubo T, Takahashi K, Hoshuyama T. Factors influencing return to work after stroke in Japan. Stroke. 1993; 24:1182-1185.
Cox DR. Regression models and lifetables. J R Stat Soc Series. 1972;34:187-220.
SAS Institute Inc. SAS/STAT User’s Guide, Version 6. 4th ed. Cary, NC: SAS Institute Inc; 1989;2:1027-1069.
SAS Institute Inc. SAS Technical Report P-229. In: SAS/STAT Software: Change and Enhancements, Release 6.07. Cary, NC: SAS Institute Inc; 1992:433-479.
Mackay A, Nias BC. Strokes in the young and middle-aged: consequences to the family and to society. J R Coll Physicians Lond. 1979;13:106-112.