- Email: [email protected]
Does hospital discharge policy influence sick-leave patterns in the case of female breast cancer?
Health Policy 72 (2005) 65–71
Does hospital discharge policy influence sick-leave patterns in the case of female breast cancer? Rikard Lindqvist∗ , Magnus Stenbeck, Finn Diderichsen Centre for Epidemiology, National Board of Health and Welfare, SE-106 30 Stockholm, Sweden
Abstract The objective was to investigate how differences among hospitals in the shift from in-patient care to day surgery and a reduced hospital length of stay affect the sick-leave period for female patients surgically treated for breast cancer. All women aged 18–64 who were diagnosed with breast cancer in 2000 were selected from the National Cancer Register and combined with data from the sick-leave database of the National Social Insurance Board and the National Hospital Discharge Register (N = 1834). A multi-factorial model was fitted to the data to investigate how differences in hospital care practice affected the length of sick-leave. The main output measure was the number of sick-leave days after discharge during the year following surgery. The confounders used included age, type of primary surgical treatment, whether or not lymph node dissection was performed, labour-market status, county, and readmission. Women treated with breast-conserving surgery had a 54.7-day (−71.9 ≤ CI95% ≤ −37.5) shorter sick-leave period than those with more invasive surgery. The day-surgery cases had 24.3 (−47.5 ≤ CI95% ≤ −1.1) days shorter sick-leave than those who received overnight care. The effect of the hospital median length of stay (LOS) was U-shaped, suggesting that hospitals with a median LOS that is either short or long are associated with longer sick-leave. In the intermediate range, women treated in hospitals with a median LOS of 2 days had 22 days longer sick-leave than those treated in hospitals with a mean LOS of 3 days. This is possibly a sign of sub-optimising. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Sick absence; Hospital length of stay; Breast cancer
1. Background The cost of the national sick-leave insurance system in Sweden has increased. Between 1997 and 2000 the number of paid sickness benefit days almost doubled [1]. Although changes in insurance policy have been introduced in order to reduce the cost and num∗
Corresponding author. E-mail address: [email protected] (R. Lindqvist).
ber of cases on long-term sick absence from work, the effects of these changes have been small [2]. The underlying causes of sickness absence have been investigated from widely ranging scientific perspectives such as medicine, sociology, psychology, economics and management [3]. Alexandersson lists more than 40 possible factors at three structural levels which, besides illness, can influence sick absence. At a societal level, for example, the design and practical applications of the sickness
0168-8510/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.healthpol.2004.06.003
66
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
insurance system, the composition of the labour force, the level of unemployment, etc., make a difference. At workplace/community level, unemployment and labour force participation, local social insurance practice, etc., play their part, and at individual level the factors include heredity, age, gender, personality and socioeconomic status [3]. The design of the health care system, the physician’s role and work practice, and the interaction between patient and physician, are other factors that influence the prevalence of sick-leave from work. The gender and speciality of the physician also influence sick-listing practice [4]. A study by Englund and Svärdsudd [5] concludes that patient’s self-presentation and expectations on sick listing have a strong influence on sick-listing practice. A parallel trend in health care is that the average length of hospital stay is decreasing and in-patient care is increasingly being replaced by day surgery. This trend is quite obvious in Sweden as in most OECD countries [6]. A pertinent question arises from the observation of these parallel trends. Does a shorter length of stay (LOS) in hospital eventually lead to a cost shift from health care to the insurance system? Can patients be discharged “too early” and therefore be absent from work for longer? The question arises as to whether there is an optimal balance between LOS and sick-leave from a cost-effectiveness and public health point of view. When selecting an application area for the study, we used the following criteria: • The study should focus on a relatively common disease affecting people of working ages (18–64). • The study should contain patients with a manifest indication for treatment, i.e. the bias of case selection and shift in treatment indications should be minimised. • An ongoing change towards out-patient treatment. Based on the above criteria, surgical treatment of female breast cancer was chosen. The mean length of hospital stay (MLOS) for surgical breast cancer treatment is decreasing [7]. Important changes in surgical technique during the past decade explain no more than one-third of this decrease (in some studies less [7,8]). Changes in tumour size and lymph node dissemination at discovery, due to, for
example, earlier discovery because of the introduction of mammography screening programs, do not explain a large portion of the decrease [9]. The present objective was to investigate the roles of the shift from in-patient care to day surgery and reduced hospital length of stay on length of sickleave for female patients treated surgically for breast cancer.
2. Material and method We selected all females from the National Cancer Register (which is comprised of all primary malignant tumours diagnosed in the Swedish resident population) aged between 18 and 64 years who had been diagnosed with breast cancer in the calendar year 2000 (N = 3812). Data for the selected cases sick-leave and labour status were obtained from the sick-leave database of the National Social Insurance Board. These data were combined with data from the National Hospital Discharge Register concerning the length and frequency of episodes (measured as dates of admission and discharge) in public hospitals. The data, included age, date of admission, hospital, type of patient care, length of stay, reasons for hospital treatment (main and contributing diagnoses) and type of surgery. The selected database consisted of the first surgical care event (discharge or day-surgery operation) during the calendar year 2000 (e.g. mastectomy, or breast-conserving surgery) for all the women between 16 and 64 years who were either employed or self-employed (N = 2063) and hence risked sick-leave connection with the treatment. Care events at hospitals with fewer than 10 cases during 2000 were omitted (approximately 5% of the events). Three counties were excluded due to lack of out-patient data. In all, some 11% of the recorded breast cancer care events in Sweden were excluded from the analyses for these reasons, leaving 1834 cases in the database. The length of stay (LOS) was calculated as the difference between the discharge date and the admission date plus one day. Day surgery was counted as 0 days. The number of sick-leave days was calculated as the total number of sick-leave days for any reason during 1 year after the discharge date. Hence,
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
the maximum number of sick-leave days in this study is 365. A multifactorial ordinary least squares model was fitted to the data. In addition, a multistage (mixed) model was tried with the hospital mean LOS treated as an aggregate level effect. The simple hierarchical structure of the multistage model is theoretically problematic when several overlapping aggregate effect are present simultaneously. In such cases the simple OLS model which treats all class level effects equally is preferable. In the present case both the point estimates and the confidence intervals are very similar in both models (the mixed model estimates are presented in a separate column of Table 4). Because of this, we present the results from the single stage OLS model, including the class level effects as properties of the individuals. The main output measure was the number of sickleave days in the national compulsory sick-leave system after discharge following breast cancer surgery, during 1 year after the first surgical care event. In Sweden, for sick-leave shorter than 7 days, the employee gets her sick-leave allowance directly from her employer without a doctor’s sick-leave accreditation certificate. Sick-leave between 7 and 14 days is paid by the employer, but must be formally certified by a doctor. After 14 days there has to be a doctor’s certificate and the government insurance systems pay the allowance [1]. In the present study, we have been unable to retrieve data on the shorter sick-leave periods not covered by the national insurance; hence, to be counted, the sick-leave had to encompass more that 14 days. The main exposure variables were: 1. LOS for the single event and 2. median LOS of surgery for breast cancer of that hospital. The median LOS for the hospital is of interest since the hospital’s discharge policy (formally written or implicit) affects the total number of sick-leave days. The confounders identified and controlled for were: • Age at discharge; • Primary surgical treatment (grouped into mastectomy or breast-conserving surgery (BCS)); • Whether lymph node extraction was performed;
67
• Type of employment (employed versus selfemployed); • County (controlling for variability in sick-leave insurance practice and hospital practice in after care); • Readmissions for re-operation during 1 year after discharge. This serves as a proxy for the level of seriousness of the cancer; • If the surgery was performed in a university/regional hospital or not. Due to referral practices the cases in university hospitals are likely to be exposed to more intense care. Hence, this variable controls for possible differences in case selection in the different hospital types.
3. Results 3.1. Description of basic data There were 1834 care events distributed among 38 hospitals. Some 11% of the operations were day-surgery. The mean LOS of the patients was 2.93 days (S.D. = 1.8) and the mean patient age was 51.8 years. The majority of the operations were performed using breast-conserving technique (72%). The median LOS varied across hospitals between 0 and 5 days, with over 90% of the hospitals ranging between 2 and 4 days (Table 1). Table 1 Descriptive statistics Cases
Percent
Out-patient operations In-patient discharges
208 1626
11 89
Type of operation Breast-conserving Mastectomy
1323 511
72 28
973 861
53 47
Lymph nod dissection No Yes
Hospital median length of stay (LOS) 0 Days 49 3 2 Days 533 29 3 Days 865 47 3.5 Days 40 2 4 Days 258 14 5 Days 89 5
Number of hospitals
1 10 17 1 7 2
68
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
Table 2 Number of sick-leave days Sick-leave days
Cases
Percent
None 1–14 15–31 32–62 63–182 193–364 365 or more
69 13 140 160 423 434 595
4 1 8 9 23 24 32
Some 4% of the cases had no sick-leave days at all during the first year after discharge whereas 32% had 365 days of sick-leave (Table 2). 3.2. Effect of out-patient surgery on sick-leave days Table 3 shows results from the first multi-factorial model that was fitted to the material. The model estimates the effects on the number of sick-leave days of age (centralised on mean age), surgical technique, readmission and day-surgery. The intercept (representing the average age, in-patient, mastectomy with lymph node dissemination) was 287 sick-leave days (Table 3). Every year of age above the median reduced the sick-leave period by 3.5 days. Women undergoing BCS had a 54.7-day-shorter sick-leave period (−71.9 ≤ CI95% ≤ −37.5), and if no lymph node extraction was performed it was 33 days shorter. Readmission increased the sick-leave period by 46.7. Table 3 The effect of out-patient care Number of sick-leave days Intercepta Age (per year) Breast-conserving surgery (yes) No lymph node dissemination surgery Readmission (yes) Out-patient care (yes)
95% confidence limits Low
High
287.2 −3.5 −54.7
244.4 −4.4 −71.9
329.9 −2.6 −37.5
−33.2
−48.7
−17.6
46.7 −24.3
28.9 −47.5
64.6 −1.1
Variability among counties was taken into account in the models but the estimates are not presented in this table. a Represents in-patient, with mastectomy, median age with lymph node dissemination, no readmission.
Finally, day-surgery effect was to reduce sick-leave by 24.3 days (−47.5 ≤ CI95% ≤ −1.1). The effect of being employed versus self-employed was tested as well, but omitted from the final model. The model also controlled for variability among counties, due to known regional differences in sick absence patterns. 3.3. The effect of LOS and median LOS on sick-leave days The second multi-factorial model was fitted to the in-patient part of the material (N = 1626) estimating the effects of age (centralised on mean age in the material), surgical technique, readmission length of stay and hospital median LOS (first- and second-degree terms) on number of sick-leave days. The intercept, describing a mean-age mastectomy with lymph node extraction, was 243 days (Table 4). As seen in the earlier model, there was a clear and significant effect of both BCS and lymph node extraction. The effect of LOS was also apparent. For every day the woman stayed in the hospital the sick-leave period increased by 6.8 days (1.7 ≤ CI95% ≤ 11.9). The effect of the hospital median LOS, which was used as an indicator of hospital discharge policy, had to be evaluated with a second-degree polynomial, suggesting a U-formed relationship (see Fig. 1). This suggests an interesting pattern: hospitals with either short or long median LOS have patients with longer sick-leave periods. A difference of 1 day in median LOS (between medians of 3 and 2 bed days) means a difference in sick-leave of 22 days. On the other hand, a difference between 4 and 5 median LOS prolongs the sick-leave by 33 days. Whether the person was employed or self-employed did not make a difference, nor did it make a difference whether the patient was treated in a university or regional hospital Regional differences in sick absence patterns were accounted for in the final model, but did not affect the extent to which sick-leave varied with mean LOS.
4. Discussion Our results show that patients treated in an out-patient setting has a shorter sick absence than patients treated as inpatients. The type of surgery, both primary surgery (e.g. mastectomy or breast
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
69
Table 4 The effect of median hospital length of stay Number of sick-leave days Intercepta
95% confidence limits Low
High
Estimates produced by a mixed model
Age (per year) Length of stay (per day) Breast-conserving surgery (yes) No lymph node dissemination surgery Readmission (yes)
243.6 −3.6 6.8 −45.0 −26.9 42.6
178.9 −4.5 1.7 −63.7 −43.3 22.0
308.3 −2.7 11.9 −26.2 −10.5 63.2
243.6 −3.6 6.6 −45.0 −27.3 43.4
Median hospital LOS (per day) Linear term Quadratic term
−90.2 13.6
−148.2 4.8
−32.2 22.5
−93.3 13.9
Variability among counties was taken into account in the models but the estimates are not presented in this table. a Represents in-patient, with mastectomy, median age with lymph node dissemination, no readmission.
conserving surgery) and lymph node dissection had a strong effect on the number of sick-leave days. As expected, the less invasive the surgery was, the less sick absence was observed. The main question of the study pertained to the effect of median LOS. The median LOS of the hospital showed a U-shaped relationship with sick-leave, suggesting that hospitals with a median LOS that is either short or long are associated with a longer sick-leave period.
Due to lack of data there is one major confounding factor that we have not been able to control for directly. This is the type of adjuvant therapy (loco-regional or systemic) after surgery. There is reason to believe that a severe radio-therapy or cytostatic therapy has an impact on sick-leave. There are, however, several differences in the way this may affect the results. If there is variability in use of adjuvant therapy across individual patients, but
Fig. 1. Number of sick-leave days per patient in relation to hospital median length of stay, surgical treatment for breast cancer 20.
70
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
these patients are randomly distributed across hospitals, the effects on the LOS effect is small variability is in that case to some extent handled by the variables in the study. For example, in controlling for the effects of “lymph node surgery” we also control for other adjuvant therapies following lymph node dissection which is not administered when no “lymph node surgery” is performed. If there is variability in the normal use of adjuvant therapy across hospitals, due to differences in treatment programmes/strategies this means a greater confounding impact on the LOS effect. The county effects handle most of this problem. In many counties there is just one hospital, and in many other counties the treatment practices are reasonably constant across care units. Another source of bias is the possibility of variability in the use of adjuvant therapy among hospitals due to differences in the case-mix (depending of, for example, case selections mechanisms, for example, university/regional hospitals treat the severe cases). To control for this we have included a flag for university/regional hospital or not. The estimate showed a small impact on sick-leave (5 days) which was neither significant, nor changed the magnitude or direction of other estimates. Our data also contain a quality flaw concerning data on lymph node dissection. As noted in Table 1 it seems that only 47% of the women underwent lymph node surgery. This is, however, almost certainly incorrect. Other studies of breast cancer surgery by Lindqvist et al. have reported much higher occurrences [7]. The quality deficiency is most likely due to the introduction of a new classification system for surgical procedures in 1997, from the Swedish classification of surgical procedures version 8 to the common Nordic NCSP (Nordic Classification of Surgical Procedures). The effect of under-reporting lymph node dissection was tested by omitting the variable in the model. This test did not affect our main results, i.e. the relation between hospital median duration of stay and sick-leave days. We could not include the self-prescribed short-term sick absence periods (<14 days), since no registry records pertain to short sick-leave. In the case of breast cancer this is probably a minor problem, since the majority of women have long sick-leave periods and those treated for breast cancer are closely followed during the first year after surgical treatment.
Health care systems such as the Swedish one have altered rapidly during the past few decades. One dramatic change is that from long in-patient care to shorter stays and day-surgery. The present system can handle more patients, and breast cancer care studies show good results both in-patient’s satisfaction and in care quality [10,11]. There are, however, reasons to discuss possible limits to improved cost-effectiveness in health care achieved by reducing the length of stay. When can we expect unwanted effects? In this study we have indications of one possible unwanted effect, namely increases in sick-leave days due to excessive savings in hospital days. With respect to the shift from in-patient care to out-patient care we see no adverse effect. The daysurgery cases have a shorter sick-leave period. These results should be interpreted with caution. We can expect that when moving patients from in-patent care to a day-surgery setting the physician selects the cases with the lowest risk, e.g. patients that are generally healthy, young and with early-stage cancer. This group is expected to have shorter sick-leave. The lack of information on clinical stage in this study makes it hard to establish whether the calculated difference in sick-leave days between in-patients and out-patients is as big as the real difference. In-patient care shows interesting indications of possible sub-optimising. When hospitals are obliged to, or choose to, cut LOS there seems to be a breakpoint where the adverse effects emerge in the form of increased sick-leave. Whether this is a direct effect of the cutting of LOS or depends on lack of hospital funding, workplace deterioration from lack of funding, or increased workloads, has yet to be investigated. A more surprising result is that not only short, but also long, median LOS is associated with long sick-leave periods. One – at least hypothetical – explanation could be that hospitals with longer median LOS may have had older patients or more advanced cancer cases. However, in the analysis we controlled for age, readmission and the patient’s individual LOS. Another hypothetical interpretation could be that hospital with a focus on economical effectiveness as well as high quality do better than hospitals with either low ambitions or severe economic restraints. The relationship between median LOS and sick-leave days in this study is restricted to surgical breast cancer care. One question for further research
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
is whether this relationship is valid for other patient groups and other areas of care. The results of this study suggests that, when working with changes and rationalization in health care – it is not sufficient to look for quality deficient and cost shifts merely inside the hospital, effectiveness of the entire health care system has to be monitored.
References [1] Swedish Monograph to the 27th General Assembly of the ISSA. Stockholm: National Social Insurance Board, 2001. [2] Arrelov B. The influence of change of legislation concerning sickness absence on physicians’ performance as certifiers. A population-based study. Health Policy 2003;63(3):259–68. [3] Alexandersson K. Sickness absence in a Swedish county with reference to gender, occupation, pregnancy and parenthood. Edsbruk: Akademitryck AB, 1995. [4] Englund L, Tibblin G, Svärdsudd K. Variations in sick-listing practice among male and female physicians of different specialties based on case vignettes. Scandinavian Journal of Primary Health Care 2000;18(1):48–52.
71
[5] Englund L, Svärdsudd K. Sick-listing habits among general practitioners in a Swedish county. Scandinavian Journal of Primary Health Care 2000;18(2):81–6. [6] Organisation for Economic Co-operation and Development. OECD Health Data 98. Paris: OECD, 1998. [7] Lindqvist R, Moller TR, Stenbeck M, Diderichsen F. Do changes in surgical procedures for breast cancer have consequences for hospital mean length of stay? A study of women operated on for breast cancer in Sweden, 1980–1995. International Journal of Technology Assessment in Health Care 2002;18(3):566–75. [8] Kleinman JC, Machlin SR, Madans J, Makuc D, Feldman JJ. Changing practice in the surgical treatment of breast cancer. The national perspective. Medical Care 1983;21(12):1232–42. [9] Lindqvist RL, Alvegard TA, Jönsson PE, Stenbeck M. Hospital stay related to TNM-stage and the surgical procedure in primary breast cancer, 2003, accepted for publication. [10] Bonnema J, van Wersch AM, van Geel AN, Pruyn JF, Schmitz PI, Paul MA, et al. Medical and psychosocial effects of early discharge after surgery for breast cancer: randomised trial. British Medical Journal 1998;316(7140):1267–71. [11] Bundred N, Maguire P, Reynolds J, Grimshaw J, Morris J, Thomson L, et al. Randomised controlled trial of effects of early discharge after surgery for breast cancer. British Medical Journal 1998;317(7168):1275–9.
Does hospital discharge policy influence sick-leave patterns in the case of female breast cancer? Rikard Lindqvist∗ , Magnus Stenbeck, Finn Diderichsen Centre for Epidemiology, National Board of Health and Welfare, SE-106 30 Stockholm, Sweden
Abstract The objective was to investigate how differences among hospitals in the shift from in-patient care to day surgery and a reduced hospital length of stay affect the sick-leave period for female patients surgically treated for breast cancer. All women aged 18–64 who were diagnosed with breast cancer in 2000 were selected from the National Cancer Register and combined with data from the sick-leave database of the National Social Insurance Board and the National Hospital Discharge Register (N = 1834). A multi-factorial model was fitted to the data to investigate how differences in hospital care practice affected the length of sick-leave. The main output measure was the number of sick-leave days after discharge during the year following surgery. The confounders used included age, type of primary surgical treatment, whether or not lymph node dissection was performed, labour-market status, county, and readmission. Women treated with breast-conserving surgery had a 54.7-day (−71.9 ≤ CI95% ≤ −37.5) shorter sick-leave period than those with more invasive surgery. The day-surgery cases had 24.3 (−47.5 ≤ CI95% ≤ −1.1) days shorter sick-leave than those who received overnight care. The effect of the hospital median length of stay (LOS) was U-shaped, suggesting that hospitals with a median LOS that is either short or long are associated with longer sick-leave. In the intermediate range, women treated in hospitals with a median LOS of 2 days had 22 days longer sick-leave than those treated in hospitals with a mean LOS of 3 days. This is possibly a sign of sub-optimising. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Sick absence; Hospital length of stay; Breast cancer
1. Background The cost of the national sick-leave insurance system in Sweden has increased. Between 1997 and 2000 the number of paid sickness benefit days almost doubled [1]. Although changes in insurance policy have been introduced in order to reduce the cost and num∗
Corresponding author. E-mail address: [email protected] (R. Lindqvist).
ber of cases on long-term sick absence from work, the effects of these changes have been small [2]. The underlying causes of sickness absence have been investigated from widely ranging scientific perspectives such as medicine, sociology, psychology, economics and management [3]. Alexandersson lists more than 40 possible factors at three structural levels which, besides illness, can influence sick absence. At a societal level, for example, the design and practical applications of the sickness
0168-8510/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.healthpol.2004.06.003
66
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
insurance system, the composition of the labour force, the level of unemployment, etc., make a difference. At workplace/community level, unemployment and labour force participation, local social insurance practice, etc., play their part, and at individual level the factors include heredity, age, gender, personality and socioeconomic status [3]. The design of the health care system, the physician’s role and work practice, and the interaction between patient and physician, are other factors that influence the prevalence of sick-leave from work. The gender and speciality of the physician also influence sick-listing practice [4]. A study by Englund and Svärdsudd [5] concludes that patient’s self-presentation and expectations on sick listing have a strong influence on sick-listing practice. A parallel trend in health care is that the average length of hospital stay is decreasing and in-patient care is increasingly being replaced by day surgery. This trend is quite obvious in Sweden as in most OECD countries [6]. A pertinent question arises from the observation of these parallel trends. Does a shorter length of stay (LOS) in hospital eventually lead to a cost shift from health care to the insurance system? Can patients be discharged “too early” and therefore be absent from work for longer? The question arises as to whether there is an optimal balance between LOS and sick-leave from a cost-effectiveness and public health point of view. When selecting an application area for the study, we used the following criteria: • The study should focus on a relatively common disease affecting people of working ages (18–64). • The study should contain patients with a manifest indication for treatment, i.e. the bias of case selection and shift in treatment indications should be minimised. • An ongoing change towards out-patient treatment. Based on the above criteria, surgical treatment of female breast cancer was chosen. The mean length of hospital stay (MLOS) for surgical breast cancer treatment is decreasing [7]. Important changes in surgical technique during the past decade explain no more than one-third of this decrease (in some studies less [7,8]). Changes in tumour size and lymph node dissemination at discovery, due to, for
example, earlier discovery because of the introduction of mammography screening programs, do not explain a large portion of the decrease [9]. The present objective was to investigate the roles of the shift from in-patient care to day surgery and reduced hospital length of stay on length of sickleave for female patients treated surgically for breast cancer.
2. Material and method We selected all females from the National Cancer Register (which is comprised of all primary malignant tumours diagnosed in the Swedish resident population) aged between 18 and 64 years who had been diagnosed with breast cancer in the calendar year 2000 (N = 3812). Data for the selected cases sick-leave and labour status were obtained from the sick-leave database of the National Social Insurance Board. These data were combined with data from the National Hospital Discharge Register concerning the length and frequency of episodes (measured as dates of admission and discharge) in public hospitals. The data, included age, date of admission, hospital, type of patient care, length of stay, reasons for hospital treatment (main and contributing diagnoses) and type of surgery. The selected database consisted of the first surgical care event (discharge or day-surgery operation) during the calendar year 2000 (e.g. mastectomy, or breast-conserving surgery) for all the women between 16 and 64 years who were either employed or self-employed (N = 2063) and hence risked sick-leave connection with the treatment. Care events at hospitals with fewer than 10 cases during 2000 were omitted (approximately 5% of the events). Three counties were excluded due to lack of out-patient data. In all, some 11% of the recorded breast cancer care events in Sweden were excluded from the analyses for these reasons, leaving 1834 cases in the database. The length of stay (LOS) was calculated as the difference between the discharge date and the admission date plus one day. Day surgery was counted as 0 days. The number of sick-leave days was calculated as the total number of sick-leave days for any reason during 1 year after the discharge date. Hence,
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
the maximum number of sick-leave days in this study is 365. A multifactorial ordinary least squares model was fitted to the data. In addition, a multistage (mixed) model was tried with the hospital mean LOS treated as an aggregate level effect. The simple hierarchical structure of the multistage model is theoretically problematic when several overlapping aggregate effect are present simultaneously. In such cases the simple OLS model which treats all class level effects equally is preferable. In the present case both the point estimates and the confidence intervals are very similar in both models (the mixed model estimates are presented in a separate column of Table 4). Because of this, we present the results from the single stage OLS model, including the class level effects as properties of the individuals. The main output measure was the number of sickleave days in the national compulsory sick-leave system after discharge following breast cancer surgery, during 1 year after the first surgical care event. In Sweden, for sick-leave shorter than 7 days, the employee gets her sick-leave allowance directly from her employer without a doctor’s sick-leave accreditation certificate. Sick-leave between 7 and 14 days is paid by the employer, but must be formally certified by a doctor. After 14 days there has to be a doctor’s certificate and the government insurance systems pay the allowance [1]. In the present study, we have been unable to retrieve data on the shorter sick-leave periods not covered by the national insurance; hence, to be counted, the sick-leave had to encompass more that 14 days. The main exposure variables were: 1. LOS for the single event and 2. median LOS of surgery for breast cancer of that hospital. The median LOS for the hospital is of interest since the hospital’s discharge policy (formally written or implicit) affects the total number of sick-leave days. The confounders identified and controlled for were: • Age at discharge; • Primary surgical treatment (grouped into mastectomy or breast-conserving surgery (BCS)); • Whether lymph node extraction was performed;
67
• Type of employment (employed versus selfemployed); • County (controlling for variability in sick-leave insurance practice and hospital practice in after care); • Readmissions for re-operation during 1 year after discharge. This serves as a proxy for the level of seriousness of the cancer; • If the surgery was performed in a university/regional hospital or not. Due to referral practices the cases in university hospitals are likely to be exposed to more intense care. Hence, this variable controls for possible differences in case selection in the different hospital types.
3. Results 3.1. Description of basic data There were 1834 care events distributed among 38 hospitals. Some 11% of the operations were day-surgery. The mean LOS of the patients was 2.93 days (S.D. = 1.8) and the mean patient age was 51.8 years. The majority of the operations were performed using breast-conserving technique (72%). The median LOS varied across hospitals between 0 and 5 days, with over 90% of the hospitals ranging between 2 and 4 days (Table 1). Table 1 Descriptive statistics Cases
Percent
Out-patient operations In-patient discharges
208 1626
11 89
Type of operation Breast-conserving Mastectomy
1323 511
72 28
973 861
53 47
Lymph nod dissection No Yes
Hospital median length of stay (LOS) 0 Days 49 3 2 Days 533 29 3 Days 865 47 3.5 Days 40 2 4 Days 258 14 5 Days 89 5
Number of hospitals
1 10 17 1 7 2
68
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
Table 2 Number of sick-leave days Sick-leave days
Cases
Percent
None 1–14 15–31 32–62 63–182 193–364 365 or more
69 13 140 160 423 434 595
4 1 8 9 23 24 32
Some 4% of the cases had no sick-leave days at all during the first year after discharge whereas 32% had 365 days of sick-leave (Table 2). 3.2. Effect of out-patient surgery on sick-leave days Table 3 shows results from the first multi-factorial model that was fitted to the material. The model estimates the effects on the number of sick-leave days of age (centralised on mean age), surgical technique, readmission and day-surgery. The intercept (representing the average age, in-patient, mastectomy with lymph node dissemination) was 287 sick-leave days (Table 3). Every year of age above the median reduced the sick-leave period by 3.5 days. Women undergoing BCS had a 54.7-day-shorter sick-leave period (−71.9 ≤ CI95% ≤ −37.5), and if no lymph node extraction was performed it was 33 days shorter. Readmission increased the sick-leave period by 46.7. Table 3 The effect of out-patient care Number of sick-leave days Intercepta Age (per year) Breast-conserving surgery (yes) No lymph node dissemination surgery Readmission (yes) Out-patient care (yes)
95% confidence limits Low
High
287.2 −3.5 −54.7
244.4 −4.4 −71.9
329.9 −2.6 −37.5
−33.2
−48.7
−17.6
46.7 −24.3
28.9 −47.5
64.6 −1.1
Variability among counties was taken into account in the models but the estimates are not presented in this table. a Represents in-patient, with mastectomy, median age with lymph node dissemination, no readmission.
Finally, day-surgery effect was to reduce sick-leave by 24.3 days (−47.5 ≤ CI95% ≤ −1.1). The effect of being employed versus self-employed was tested as well, but omitted from the final model. The model also controlled for variability among counties, due to known regional differences in sick absence patterns. 3.3. The effect of LOS and median LOS on sick-leave days The second multi-factorial model was fitted to the in-patient part of the material (N = 1626) estimating the effects of age (centralised on mean age in the material), surgical technique, readmission length of stay and hospital median LOS (first- and second-degree terms) on number of sick-leave days. The intercept, describing a mean-age mastectomy with lymph node extraction, was 243 days (Table 4). As seen in the earlier model, there was a clear and significant effect of both BCS and lymph node extraction. The effect of LOS was also apparent. For every day the woman stayed in the hospital the sick-leave period increased by 6.8 days (1.7 ≤ CI95% ≤ 11.9). The effect of the hospital median LOS, which was used as an indicator of hospital discharge policy, had to be evaluated with a second-degree polynomial, suggesting a U-formed relationship (see Fig. 1). This suggests an interesting pattern: hospitals with either short or long median LOS have patients with longer sick-leave periods. A difference of 1 day in median LOS (between medians of 3 and 2 bed days) means a difference in sick-leave of 22 days. On the other hand, a difference between 4 and 5 median LOS prolongs the sick-leave by 33 days. Whether the person was employed or self-employed did not make a difference, nor did it make a difference whether the patient was treated in a university or regional hospital Regional differences in sick absence patterns were accounted for in the final model, but did not affect the extent to which sick-leave varied with mean LOS.
4. Discussion Our results show that patients treated in an out-patient setting has a shorter sick absence than patients treated as inpatients. The type of surgery, both primary surgery (e.g. mastectomy or breast
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
69
Table 4 The effect of median hospital length of stay Number of sick-leave days Intercepta
95% confidence limits Low
High
Estimates produced by a mixed model
Age (per year) Length of stay (per day) Breast-conserving surgery (yes) No lymph node dissemination surgery Readmission (yes)
243.6 −3.6 6.8 −45.0 −26.9 42.6
178.9 −4.5 1.7 −63.7 −43.3 22.0
308.3 −2.7 11.9 −26.2 −10.5 63.2
243.6 −3.6 6.6 −45.0 −27.3 43.4
Median hospital LOS (per day) Linear term Quadratic term
−90.2 13.6
−148.2 4.8
−32.2 22.5
−93.3 13.9
Variability among counties was taken into account in the models but the estimates are not presented in this table. a Represents in-patient, with mastectomy, median age with lymph node dissemination, no readmission.
conserving surgery) and lymph node dissection had a strong effect on the number of sick-leave days. As expected, the less invasive the surgery was, the less sick absence was observed. The main question of the study pertained to the effect of median LOS. The median LOS of the hospital showed a U-shaped relationship with sick-leave, suggesting that hospitals with a median LOS that is either short or long are associated with a longer sick-leave period.
Due to lack of data there is one major confounding factor that we have not been able to control for directly. This is the type of adjuvant therapy (loco-regional or systemic) after surgery. There is reason to believe that a severe radio-therapy or cytostatic therapy has an impact on sick-leave. There are, however, several differences in the way this may affect the results. If there is variability in use of adjuvant therapy across individual patients, but
Fig. 1. Number of sick-leave days per patient in relation to hospital median length of stay, surgical treatment for breast cancer 20.
70
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
these patients are randomly distributed across hospitals, the effects on the LOS effect is small variability is in that case to some extent handled by the variables in the study. For example, in controlling for the effects of “lymph node surgery” we also control for other adjuvant therapies following lymph node dissection which is not administered when no “lymph node surgery” is performed. If there is variability in the normal use of adjuvant therapy across hospitals, due to differences in treatment programmes/strategies this means a greater confounding impact on the LOS effect. The county effects handle most of this problem. In many counties there is just one hospital, and in many other counties the treatment practices are reasonably constant across care units. Another source of bias is the possibility of variability in the use of adjuvant therapy among hospitals due to differences in the case-mix (depending of, for example, case selections mechanisms, for example, university/regional hospitals treat the severe cases). To control for this we have included a flag for university/regional hospital or not. The estimate showed a small impact on sick-leave (5 days) which was neither significant, nor changed the magnitude or direction of other estimates. Our data also contain a quality flaw concerning data on lymph node dissection. As noted in Table 1 it seems that only 47% of the women underwent lymph node surgery. This is, however, almost certainly incorrect. Other studies of breast cancer surgery by Lindqvist et al. have reported much higher occurrences [7]. The quality deficiency is most likely due to the introduction of a new classification system for surgical procedures in 1997, from the Swedish classification of surgical procedures version 8 to the common Nordic NCSP (Nordic Classification of Surgical Procedures). The effect of under-reporting lymph node dissection was tested by omitting the variable in the model. This test did not affect our main results, i.e. the relation between hospital median duration of stay and sick-leave days. We could not include the self-prescribed short-term sick absence periods (<14 days), since no registry records pertain to short sick-leave. In the case of breast cancer this is probably a minor problem, since the majority of women have long sick-leave periods and those treated for breast cancer are closely followed during the first year after surgical treatment.
Health care systems such as the Swedish one have altered rapidly during the past few decades. One dramatic change is that from long in-patient care to shorter stays and day-surgery. The present system can handle more patients, and breast cancer care studies show good results both in-patient’s satisfaction and in care quality [10,11]. There are, however, reasons to discuss possible limits to improved cost-effectiveness in health care achieved by reducing the length of stay. When can we expect unwanted effects? In this study we have indications of one possible unwanted effect, namely increases in sick-leave days due to excessive savings in hospital days. With respect to the shift from in-patient care to out-patient care we see no adverse effect. The daysurgery cases have a shorter sick-leave period. These results should be interpreted with caution. We can expect that when moving patients from in-patent care to a day-surgery setting the physician selects the cases with the lowest risk, e.g. patients that are generally healthy, young and with early-stage cancer. This group is expected to have shorter sick-leave. The lack of information on clinical stage in this study makes it hard to establish whether the calculated difference in sick-leave days between in-patients and out-patients is as big as the real difference. In-patient care shows interesting indications of possible sub-optimising. When hospitals are obliged to, or choose to, cut LOS there seems to be a breakpoint where the adverse effects emerge in the form of increased sick-leave. Whether this is a direct effect of the cutting of LOS or depends on lack of hospital funding, workplace deterioration from lack of funding, or increased workloads, has yet to be investigated. A more surprising result is that not only short, but also long, median LOS is associated with long sick-leave periods. One – at least hypothetical – explanation could be that hospitals with longer median LOS may have had older patients or more advanced cancer cases. However, in the analysis we controlled for age, readmission and the patient’s individual LOS. Another hypothetical interpretation could be that hospital with a focus on economical effectiveness as well as high quality do better than hospitals with either low ambitions or severe economic restraints. The relationship between median LOS and sick-leave days in this study is restricted to surgical breast cancer care. One question for further research
R. Lindqvist et al. / Health Policy 72 (2005) 65–71
is whether this relationship is valid for other patient groups and other areas of care. The results of this study suggests that, when working with changes and rationalization in health care – it is not sufficient to look for quality deficient and cost shifts merely inside the hospital, effectiveness of the entire health care system has to be monitored.
References [1] Swedish Monograph to the 27th General Assembly of the ISSA. Stockholm: National Social Insurance Board, 2001. [2] Arrelov B. The influence of change of legislation concerning sickness absence on physicians’ performance as certifiers. A population-based study. Health Policy 2003;63(3):259–68. [3] Alexandersson K. Sickness absence in a Swedish county with reference to gender, occupation, pregnancy and parenthood. Edsbruk: Akademitryck AB, 1995. [4] Englund L, Tibblin G, Svärdsudd K. Variations in sick-listing practice among male and female physicians of different specialties based on case vignettes. Scandinavian Journal of Primary Health Care 2000;18(1):48–52.
71
[5] Englund L, Svärdsudd K. Sick-listing habits among general practitioners in a Swedish county. Scandinavian Journal of Primary Health Care 2000;18(2):81–6. [6] Organisation for Economic Co-operation and Development. OECD Health Data 98. Paris: OECD, 1998. [7] Lindqvist R, Moller TR, Stenbeck M, Diderichsen F. Do changes in surgical procedures for breast cancer have consequences for hospital mean length of stay? A study of women operated on for breast cancer in Sweden, 1980–1995. International Journal of Technology Assessment in Health Care 2002;18(3):566–75. [8] Kleinman JC, Machlin SR, Madans J, Makuc D, Feldman JJ. Changing practice in the surgical treatment of breast cancer. The national perspective. Medical Care 1983;21(12):1232–42. [9] Lindqvist RL, Alvegard TA, Jönsson PE, Stenbeck M. Hospital stay related to TNM-stage and the surgical procedure in primary breast cancer, 2003, accepted for publication. [10] Bonnema J, van Wersch AM, van Geel AN, Pruyn JF, Schmitz PI, Paul MA, et al. Medical and psychosocial effects of early discharge after surgery for breast cancer: randomised trial. British Medical Journal 1998;316(7140):1267–71. [11] Bundred N, Maguire P, Reynolds J, Grimshaw J, Morris J, Thomson L, et al. Randomised controlled trial of effects of early discharge after surgery for breast cancer. British Medical Journal 1998;317(7168):1275–9.