N. Sutcliffe, A. E. Clarke 1 , R. Taylor 2 , C. Frost 2 and D. A. Isenberg Centre for RheumatologyuBloomsbury Rheumatology Unit, Department of Medicine, University College London, UK, 1 Divisions of Clinical ImmunologyuAllergy and Clinical Epidemiology, Montreal General Hospital, McGill University, Montreal, Quebec, Canada and 2 Medical Statistics Unit, London School of Hygiene and Tropical Medicine, London University, London, UK Abstract Objective. To determine the annual direct, indirect and total costs and predictors of costs in patients with systemic lupus erythematosus (SLE). Methods. One hundred and ve patients with SLE completed questionnaires on health-care utilization and employment history. Predictors of costs were determined by multiple regression analyses using direct, indirect and total costs as outcome variables. Demographics, health status, disease activity, end-organ damage, social support and satisfaction with care were used as predictor variables. Results. The mean annual total cost per patient was 7913. Direct costs were a third and indirect costs two-thirds of the total cost. Higher education level, greater disease activity and lower physical functioning were associated with higher direct, indirect and total costs. Higher direct costs were also associated with greater damage and younger age. Conclusion. SLE has a considerable impact on the health-care system and society. Improvement in disease activity and physical health and prevention of end-organ damage may reduce costs in SLE. KEY WORDS: SLE, Direct costs, Indirect costs, Total costs, Predictors of costs. In the current political atmosphere of limited health- care budgets, it is increasingly important to determine the economic consequences of disease. The economic impact of rheumatic conditions is immense and is likely to increase with the ageing of the population w1x. It has been shown that health-care costs and wages lost as a result of rheumatic disease equal 2.5% of the gross national product in the USA w2x. Thus it is not surprising that there has been increasing interest in the cost of caring for patients with these diseases. There have been a number of studies evaluating the costs of rheumatic diseases, particularly rheumatoid arthritis (RA) w3, 4x. However, very few studies have examined the nancial consequences of systemic lupus erythematosus (SLE) w5, 6x. The present paper reports part of a trinational study which looked at the health status, health services utiliza- tion and the direct, indirect and total costs for patients with SLE in the UK, the USA and Canada w711x. The trinational study showed that annual resource utilization did not vary signicantly between the three countries. However, differences were observed within each resource category. Canadians saw more specialists than the British, the British more generalists. Canadians and Americans were more frequent users of the emer- gency room, Americans of laboratoryuimaging proce- dures. Canadians had greater hospital utilization than Americans. However, the trinational studydidnot express health costs using British prices and did not investigate predictors of costs. SLEcan serve as a model for studying the management of many chronic rheumatic diseases. It is a prototypical autoimmune condition that primarily affects women, involves multiple systems and is treated with poten- tially toxic therapies, mainly by specialists. Therefore the methods used and the conclusions derived from this study can be applied to other SLE populations and, to a certain extent, to populations affected by other multisystem autoimmune diseases w12x. Cost analysis is an important step in performing an economic evaluation. Although cost studies do not answer the question of whether the costs are worth paying for, they are still valuable for their ability to determine the magnitude of a problem and to identify areas for further investigation w13x. Correspondence to: N. Sutcliffe, Centre for Rheumatology, Bloomsbury Rheumatology Unit, Arthur Stanley House, 4th Floor, 4050 Tottenham Street, London W1P 9PG, UK. Submitted 31 January 2000; revised version accepted 24 July 2000. Rheumatology 2001;40:3747 2001 British Society for Rheumatology 37 The cost of illness is often expressed through three components: direct costs, indirect costs and intangible costs. Direct costs refer to all the resources consumed in delivering care to the patient and can be categorized into direct health-care costs and direct non-health-care costs w14x. Direct health-care costs include the costs for in-patient, out-patient and emergency care, while direct non-health-care costs include, for example, transport to receive health care and the time caregivers spend in delivering or aiding the patient in receiving health care. Indirect costs result from loss of productivity due to disease. These may include values for paid or unpaid labour. Intangible costs are attributable to pain and psychological suffering. These are the most difcult to express in monetary terms and are often conveyed as clinical effects or quality-of-life scores w5, 15x. The indirect costs of illness result from output loss because of cessation or reduction of productivity due to morbidity and mortality. Output loss may arise from diminished activity in both the labour market and the non-labour market or household, and can represent a signicant portion of the overall cost of illness. For diseases affecting predominantly women, much of the output loss will result from a reduction in non-labour market activities, such as housekeeping and childcare. The method of valuing these non-labour activities is therefore critical if the cost estimates of women's diseases are to be comprehensive w10x. In this study we assessed costs from a societal per- spective, which incorporates all costs regardless of who incurs them. We studied only a single centre in the UK. The costs examined also included those attributable to comorbidity. Since SLE can cause a broad spectrum of health problems, we considered it to be impossible for either the patient or the caring physician to ascertain if problems were SLE-related. The aim of our study was to investigate all costs incurred by these patients. We also identied the predictors of costs. Once the contributors and determinants of costs have been identied, it may be possible to assess whether investment in these areas can reduce costs and improve patient outcomes. Methods One hundred and seven consecutive out-patients with SLE attending a specialized SLE clinic were enrolled in a trinational study of health services utilization and outcomes in SLE. Costs incurred by these patients had already been estimated using Canadian prices for services w7x. We estimated the costs to patients of the University College London Hospitals (UCLH) using local prices. All patients were assessed between June and September 1995. Information about sociodemographics, disease activity, as assessed by the revised Systemic Lupus Activity Measure (SLAM-R), and end-organ damage, as assessed by the Systemic Lupus Inter- national Collaborating ClinicsuAmerican College of Rheumatology Damage Index (SLICCuACR DI), were obtained during an out-patient clinic visit. Disease duration was calculated from the time the patient rst fullled four of the revised American College of Rheumatology (ACR) criteria for the classication of SLE w16x. Patients were also asked to ll in self-report questionnaires about health status using the Medical Outcomes Study Short Form 36 (SF-36), social support using the Interpersonal Support Evaluation List (ISEL), their satisfaction with medical care using the Patient Satisfaction Questionnaire (PSQ) and an economic ques- tionnaire. The economic questionnaire was repeated after 6 months. Measures The SLAM was developed by Liang et al. w18x in Boston on the basis of a consensus of members of the Lupus Council of the American College of Rheumatology. It includes 32 items, divided into 11 organ systems, and assigns a degree of severity on a scale of 13, with 1 being mild and 3 most severe. The total possible score is 86 w17x. Its validity and reliability have been shown previously w18, 19x. We used the revised SLAM (SLAM-R) in this study, which differs very little from the original SLAM w18x. The SLICCuACR DI is a measure of cumulative end- organ damage in SLE. Its validity and reliability have been shown previously w2022x. Damage is described as non-reversible change, not related to active inamma- tion, occurring since the onset of lupus, ascertained by clinical assessment and present for at least 6 months unless stated otherwise. It is dened for 12 organs or systems: ocular (range 02), neuropsychiatric (06), renal (03), pulmonary (05), cardiovascular (06), peripheral vascular (05), gastrointestinal (06), mus- culoskeletal (07), skin (02), gonadal (01), endocrine (01) and malignancy (02). The maximum possible total score is 46. The SF-36 is a short-form, 36-item health question- naire that measures eight multi-item dimensions: physical function; role limitations due to physical problems; role limitations due to emotional problems; social function; mental health; general health perception; vitality; and pain. Thirty-ve of the items contribute to these scales. There is a further unscaled single item asking respondents about health change over the past year. For each dimension, item scores are coded, summed and transformed onto a scale from 0 (worst possible health) to 100 (best possible health state) w23x. Physical (PCS) and mental component summary (MCS) scores, generated by collapsing the eight subscales, were also calculated w24x. The UK version of the question- naire was used in this study w25x. Its internal consis- tency, validity and reliability have been demonstrated w23, 25, 26x. The ISEL scale was developed to assess social support w27x. It includes scales measuring four social support functions: belonging (availability of people one can do things with); appraisal (availability of someone to talk to); tangible (availability of material aid); and self-esteem (availability of a positive comparison when comparing oneself with others). Each ISEL subscale consists of 10 items with four possible responses (03) 38 N. Sutcliffe et al. with a total subscale score range from 0 (least support) to 30 (most support). The total ISEL score is the sum of the four component scores (0120). The Medical Outcomes Study PSQ was originally developed by Ware et al. w28x. We used version IV in this study. This questionnaire enquires about global level of satisfaction, considering all health-care providers and settings, without specifying a particular provider or hospitalization. It comprises seven dimensions: general satisfaction; technical competence (i.e. diagnosis and management); interpersonal satisfaction (e.g. courtesy and respect); communication satisfaction; time spent with doctor; and access to care. The nancial dimension was omitted because it poses questions of limited appl- icability in the British system of health-care funding. Each scale is scored 0100. Higher scores indicate more satisfaction with care. A summary score has not been developed. The economic questionnaire asked about health service utilization and diminished (work force and non-work force) productivity over the preceding 6 months and was repeated 6 months after study entry to develop annual cost estimates. It enquired about the utilization of all health services without asking the respondent to make attribution to SLE or other conditions. Given that SLE can cause such a broad spectrum of health problems, it was believed that it would be impossible for either the patient or the treating physician to ascertain if problems were SLE-related. It is a modied version of the economic portion of the Stanford Health Assessment Questionnaire (HAQ) w29x. This questionnaire has previously been used and validated for both SLE and other rheumatic diseases w5, 12, 30x. Health services utilization included the number of out-patient visits to doctors and other health-care professionals; out-patient imaging, laboratory and other diagnostic procedures; prescription and non-prescription medications; renal dialysis; assistive devices (e.g. walking stick); visits to the accident and emergency department and mode of transport; use of day-case surgery; and stays in hospitals or nursing homes. Patients were also asked about their employment history over the preceding 6 months. They indicated their employment status, annual income, days of activity limitation with respect to labour-force and household activities, and their need for domestic help. Assessment of costs Direct costs were recorded initially as units of service and then given a sterling value per unit of a service, expressed in 19961997 UK pounds. For each patient, the unit price of a service was multiplied by the number of units of each service and the products were summed. To determine the annual costs, the 6-monthly costs were summed. The prices were assigned to each unit of service as follows. Visits to health-care professionals The prices for out-patient visits to hospital doctors were calculated by subtracting the diagnostic procedure costs from the total out-patient costs for each speciality at UCLH and dividing this result by the number of out-patient attendances. This provided the cost per attendance for each hospital specialist. The cost of general practitioner (GP) visits was based on a report of the Department of Health and Ofce of Population Censuses and Surveys w31x. For visits to psychiatrists, psychologists and district nurses (health professionals not employed directly by our hospital trust), the costs per attendance were obtained from the local com- munity health services trust (personal communication). The cost of a visit to a physiotherapist was obtained from our hospital's physiotherapy department (personal communication). For a social worker visit, the cost per visit was obtained from the local city council (personal communication). All the cost calculations mentioned above were comprehensive and included the overhead allowances. The prices for diagnostic investigations and renal dialysis were either obtained from the relevant hospital department (personal communication with clinical ser- vices directory and business management) or based on our hospital's imaging charges for 199697. These prices included the cost of technicians and reporting physi- cians as well as overhead allowances. Prescription and non-prescription medication prices were calculated as the product of the price per milligram, total daily dose and therapy duration. The price per milligram for each medication was based on the drug tariff w32x. It was assumed that most of our patients had their medications supplied by community pharmacists every 3 months (community pharmacists are reimbursed according to the drug tariff ). A xed price per medication, reecting professional fees and overhead allowances, was added to the cost of drugs. The cost of attendance at the accident and emergency department was obtained from the management accounts department at UCLH and the cost of transport by ambulance was obtained from London Ambulance Service NHS Trust (personal communication). The costs of assistive devices were obtained from the surgical appliances and surgical supplies departments at the UCLH (personal communication). The prices for day-case surgery and hospital stays were assigned according to the UCLH NHS Trust GP fundholders tariff (199697) for patients who had specic procedures or operations. For patients who had medical treatment only, the average cost of a stay at UCLH (according to the number of days) was used. These prices were comprehensive and included the running costs, expenditure on doctors and nurses and diagnostic tests. Indirect costs refer to impaired productivity due to illness and work time lost, either in market work or unpaid work at home, and represent productivity losses to the economy for the length of impairment w10x. We identied two groups of patients: (i) labour-force participants, i.e. those working either full-time or part- time; and (ii) those not in the labour force, i.e. those who were retired, disabled, unemployed or homemakers, and others. 39 Total costs and predictors of costs in SLE For labour-force participants, we used the `human capital' method to calculate indirect costs, applying employment income to value lost time. Specically, indirect costs were determined as follows. Patients indicated their annual personal income in ranges (i.e. -5000, 500010 000, 10 00015 000, 15 000 20 000, 20 00025 000 and )25 000). The midpoint of each range was taken as the income; for the minimum and maximum ranges actual values were taken. Hourly and daily incomes were derived from these values. The patients' hourly income was multiplied by the number of extra hours per week they stated they would be working if they did not have SLE. This result was then multiplied by 26 (number of weeks that would be worked in a 6-month period). The number of self- reported days of lost work was also multiplied by the daily income. These two products were summed. For non-labour-force participants, we calculated the implicit income losses. These patients were divided into two categories. For those who indicated they would be working if they did not have SLE, we used the opportunity cost method, which is based on the idea that the economic value of non-market work is equal to its best alternative economic use, i.e. the amount the person could have earned in market work. The income loss was therefore calculated by multiplying the number of hours per week they would be working by the hourly income of a sex- and occupation-matched person w33x. The result was then multiplied by 26 (number of weeks that would be worked in a 6-month period). For the remaining non-labour-force participants, we used the replacement cost approach, which imputes market values to perform equivalent duties at home, i.e. income loss was calculated by multiplying the number of self-reported days the individual was unable to attend to activities of daily living by the daily income of a sex-matched domestic worker w33x. The cost of employing a paid helper was also added to the indirect cost for the patients who worked. To determine the annual indirect cost, costs for the rst and second 6-month periods were added. All cost data were expressed in 199697 UK pounds. The total cost was the sum of the direct and indirect costs. Statistical analysis Multiple regression analyses were used to determine the predictors of direct, indirect and total costs. Predictive models were tted for direct, indirect and total costs separately. In regression models for cost data, it is desirable to analyse untransformed costs so that the models predict actual costs in pounds. Because of the skewed nature of cost data, it is recommended w34x that non-parametric bootstrap methods w35x be used to calculate condence intervals for regression coefcients. Because no automated bootstrap methods are available to select important variables for inclusion in regression models, standard parametric analyses were used to identify a number of candidate regression models, and condence intervals for each of these candidate models were calculated using bootstrap methods. The can- didate models were obtained by tting models with every combination of the predictor variables and using Mallows' Cp criteria w36x to determine how many variables were required in the model and the best-tting model with this number of predictors. The next four best-tting models of this dimension and the four best-tting models of dimension one greater than this best-tting model were also identied. Bias-adjusted bootstrap condence intervals were constructed from 1000 replicate samples for each of the parameters in each of these models. The candidate model selected as the nal best-tting model was that in which all of the bootstrap 95% condence intervals excluded a null effect. The all-subsets regression analysis was carried out in MINITAB. The bootstrap condence intervals were computed using STATA version 5. All other analyses were carried out in SAS. Results Of the 107 patients initially enrolled at our centre, two were lost to follow-up. The remaining 105 patients were included in the analyses. Ninety-four per cent were female. Seventy-ve per cent were Caucasian. Fifty-eight per cent were married. Mean age was 39.9"11.8 yr (median 38.5 yr, range 18.875.3 yr). Mean disease dura- tion was 10.5"6.6 yr (median 10.8 yr, range 133.5 yr). Mean duration of education was 13"2.7 yr (median 12 yr, range 620 yr). The disease characteristics of the patients are shown in Table 1. The number of visits to health-care professionals over 1 yr varied between 1 and 131, with a median of 12 visits (mean 17.4"17.8). Ninety per cent or more of the patients visited a rheumatologist in each 6-month period. More than 70% of patients visited their GP. Only a small number of patients visited allied health professionals. Details of visits to other medical special- ists in the rst and second 6-month periods are shown in Table 2A and B. The number of diagnostic procedures per year varied between 0 and 66, with a median of 12 (mean 14.7"11.5). Although a majority of patients had blood tests (90 and 74% in the two 6-month periods respectively) and urine tests (77 and 68%), the more expensive tests were used for relatively small numbers of patients (-20% of patients had at least one of these expensive investigations in each 6-month period). Ninety-ve per cent of patients were on medications. The most commonly used medications, in order of frequency of use, were oral prednisolone, hydroxy- chloroquine, paracetamol, azathioprine and aspirin. Only two patients in the rst 6 months and one patient in the second 6 months had peritoneal dialysis. About a third of patients used aids. Twenty-two per cent of patients had emergency room visits over 1 yr. Fourteen patients made 19 visits during the rst 6-month period and 15 visits during the second 6-month period. Forty- three per cent of patients had day-case surgery or in-patient treatment. Six patients underwent six out- patient surgeries in the rst 6-month period and ve 40 N. Sutcliffe et al. patients underwent ve out-patient surgeries in the second 6-month period. Twenty-eight patients had 40 hospital stays in the rst 6-month period, and 18 patients had 30 hospital stays in the second 6-month period. The average length of stay for hospitalized patients was 8 and 9 days in the two 6-month periods respectively. None of the patients used a nursing home. Employment characteristics of patients at study entry and at 6 months are shown in Table 3A. Almost half of the patients did not work. Of the patients who worked, TABLE 2A. Mean number of visits and cost of health professionals (rst 6 months) No. of patients with at least one visit Mean no. of visits Cost per visit Six-monthly cost per patient n (%) in 6 months () () Rheumatologist 102 97 3.02 61 184 Physician 5 5 0.12 48 6 GP 80 77 2.86 15.5 44 Nephrologist 19 18 0.60 52 32 Dermatologist 13 13 0.32 62 20 Ophthalmologist 23 21 0.31 73 23 Miscellaneous a 36 35 0.82 58.4 48 Surgical specialist b 5 5 0.09 28.5 2 Physiotherapist 9 9 0.94 20 19 Social worker 3 3 0.11 15.6 2 District nurse 5 5 0.07 26 2 a Miscellaneous medical specialist (cardiologist, chest physician, etc.). Cost per visit refers to the weighted average cost. b Miscellaneous surgical specialist (orthopaedic surgeon etc.). Cost per visit refers to the weighted average cost. TABLE 2B. Mean number of visits and cost of health professionals (second 6 months) No. of patients with at least one visit Mean no. of visits Cost per visit Six-monthly cost per patient n (%) in 6 months () () Rheumatologist 94 90 2.15 61 131 Physician 6 6 0.12 48 6 GP 77 73 2.53 15.5 39 Nephrologist 14 13 0.4 52 21 Dermatologist 8 8 0.14 62 9 Ophthalmologist 21 20 0.29 73 21 Miscellaneous a 30 29 0.74 52.8 39 Surgical specialist b 6 6 0.78 57.5 45 Physiotherapist 6 6 0.66 20 13 Social worker 3 3 0.2 15.6 3 District nurse 7 7 0.19 26 5 a Miscellaneous medical specialist (cardiologist, chest physician, etc.). Cost per visit refers to the weighted average cost. b Miscellaneous surgical specialist (orthopaedic surgeon etc.). Cost per visit refers to the weighted average cost. TABLE 1. Disease characteristics of patients Mean (S.D.) Median Range SLAM (081; higher sgreater activity) 6.3 (3.9) 5 022 SLICC (046; higher sgreater damage) 1.1 (1.3) 1 05 ISEL (0120; higher smore support) 89.7 (19.9) 94 32119 SF-36 (higher sbetter health) PCS 36.4 (12.5) 33.9 13.266.8 MCS 42.9 (12.8) 44.6 10.365.5 PSQ (higher smore satisfaction) General satisfaction 62.6 (16.8) 62.5 16.7100 Technical quality 63.8 (14.7) 63.9 25100 Interpersonal satisfaction 72.3 (15.7) 75 25100 Communication satisfaction 67.5 (17.6) 70 20100 Time spent with doctor 61.8 (22.9) 75 12.5100 Access to care 65 (14.3) 66.7 20.897.9 41 Total costs and predictors of costs in SLE about a third (18.1% of all patients at study entry) to a quarter (13.5% at 6 months) worked part-time. Of the patients who were disabled, the majority indicated that they were disabled because of SLE. Similarly, of the patients who were retired, the majority indicated that they were retired because of their SLE. Overall, 22% of the patients were either disabled or retired because of their SLE. A signicant portion of working patients missed some days of work (Table 3B). The number of work days lost was variable (5 and 2% of working patients declared work loss of more than 1 month at 0 and 6 months respectively). A similar proportion of unemployed patients declared disability days (days they were unable to carry out their usual activities). Again, the number of disability days was very variable (23 and 16% of unemployed patients declared more than 1 month of disability days at 0 and 6 months respectively). Components of costs are shown in Table 4. Direct costs were a third of the total cost. Indirect costs were twice the direct costs. Nineteen patients incurred no indirect costs. Four patients had extremely high indirect costs over 1 yr ()20 000). Six patients had very high annual direct costs ()10 000). The biggest components of direct costs were the costs of in-patient care and day- case surgery. These were followed by the cost of visits to health-care professionals; visits to the rheumatologists accounted for between 48% (at study entry) and 39% (at 6 months) of this component (Table 2A and B). The costs of medications, including dialysis and diagnostic procedures, were also important contributors to direct costs. The costs of aids and emergency room visits were small. The results of multiple regression analyses for annual direct, indirect and total costs are shown in Tables 5, 6 and 7 respectively. The tables show unadjusted and adjusted regression coefcients for each of the potential predictors of costs. These coefcients represent the costs associated with changes in each of these factors. The parametric analysis of direct costs suggested that disease activity and end-organ damage were positively asso- ciated with direct costs and that age and technical competence were negatively associated with direct costs. Calculation of the more reliable bootstrap 95% con- dence intervals for the model with these four factors and for other models identied as having similarly good ts conrmed that the associations between direct costs and disease activity, end-organ damage and age were TABLE 3A. Employment characteristics of patients at study entry and at 6 months No. of patients (%) in different categories At study entry At 6 months (n s105) (n s104) a Employed 57 (54.3) 54 (52) Disabled 16 (15.2) 15 (14.4) Disabled due to SLE 15 (14.3) 13 (12.5) Retired 11 (10.5) 14 (13.4) Retired due to SLE 8 (7.6) 10 (9.6) Homemaker 10 (9.5) 9 (8.7) Other 11 (10.5) 12 (11.5) a One patient did not declare employment status at 6 months. TABLE 3B. Employment characteristics of patients with SLE at study entry and at 6 months At study entry At 6 months Patients reporting lost work days a 63.1% 55.5% Patients reporting disabled days b 58.3% 60% Mean"S.D. Median (range) Mean"S.D. Median (range) No. of lost work days a 10.5"20.4 4 (0120) 8.2"21 3 (0150) No. of disabled days b 32.8"50.9 8 (0180) 28.3"47.7 9 (0180) a Of the patients who worked. b Of the patients who did not work. TABLE 4. Annual costs for patients with SLE (1996 UK pounds) Mean cost (S.D.) Percentage Median Range Visits to health professionals 714 (687) 27 a 534 154796 Diagnostic procedures 388 (309) 15 a 296 01448 Medications and dialysis 451 (1883) 17 a 157 015291 Aids 20 (44) 0.7 a 0 0225 Emergency room visits 37 (85) 1 a 0 0484 In-patient careuday-case surgery 1003 (1949) 38 a 0 011915 Total direct cost 2613 (3163) 33 b 1392 2015724 Total indirect cost 5299 (6981) 67 b 2187 043423 Total cost (directqindirect) 7913 (8668) 4677 2048160 a Of total direct costs. b Of total costs. 42 N. Sutcliffe et al. statistically signicant. However, bootstrap 95% con- dence intervals for the effect of technical quality on direct costs included zero in all of these models. In addition, bootstrap 95% condence intervals for the effects of education level and physical component summary scores (narrowly) excluded zero when these terms were included in the model with disease activity, end-organ damage and age. This provides some evidence (P-0.05) that higher education level and lower physical component summary scores are associated with higher costs. Similar analysis of the indirect and total costs showed that higher education level, greater disease activity and lower physical component scores were statistically signicantly associated with higher indirect and total costs. No other predictor variables were signicantly related after adjusting for the effects of these three variables. Discussion Our study is the rst prospective cost-identication study in SLE in the UK. We identied both direct and indirect costs, thus determining the burden of this disease on the health-care system and society. Although we assessed costs from a societal perspective, we were not able to evaluate all components of the direct costs. However, we included non-physician health-care pro- viders, such as psychologists and social workers, as well as some transport costs, which are often omitted in studies of the cost of illness. TABLE 5. Direct costs () associated with changes in potential predictors of costs Change in cost () (95% condence interval) Variable Unadjusted Adjusted Age (1 yr increase) 47.7 (90.9 to 15.3)* 38.9 (75.8 to 11.1)* Sex (male s1) 501.2 (2465.7 to 5344.9) Ethnic origin (non-Caucasians1) 827.4 (808.8 to 2914.8) Marital status (unmarried s1) 339.8 (872.9 to 1641.2) Education level (1 yr increase) 260.9 (68.4 to 550.9)* 204.6 (61.7 to 403.5)* Disease duration (1 yr increase) 2.2 (83.6 to 82.5) SLAM (1 point increase) 356.2 (199.9 to 595.7)* 239.4 (115.4 to 364.6)* SLICC (1 point increase) 1146.8 (583.6 to 1720.1)* 875.7 (411.7 to 1503.4)* ISEL (1 point increase) 5.6 (32.8 to 23.3) PCS (1 point increase) 92.5 (144.9 to 50.2)* 38.0 (81.0 to 2.1)* MCS (1 point increase) 20.5 (24.8 to 71.6) General satisfaction (1 point increase) 40.1 (84.2 to 3.8)* Technical (1 point increase) 46.0 (107.7 to 1.4)* Interpersonal (1 point increase) 19.6 (84.5 to 23.1) Communication (1 point increase) 30.4 (85.3 to 18.5) Time spent with doctor (1 point increase) 37.9 (66.7 to 9.1) Access to care (1 point increase) 25.3 (69.4 to 11.2) *P-0.05. TABLE 6. Indirect costs () associated with changes in potential predictors of costs Change in cost () (95% condence interval) Variable Unadjusted Adjusted Age (1 yr increase) 20.6 (128.9 to 102.9) Sex (male s1) 1933.2 (6039.1 to 7593.6) Ethnic origin (non-Caucasians1) 1.5 (279.5 to 3148.9) Marital status (unmarried s1) 508.5 (2007.6 to 3891.3) Education level (1 yr increase) 428.1 (8.0 to 997.0)* 407.5 (14.9 to 875.4)* Disease duration (1 yr increase) 20.0 (156.5 to 226.5) SLAM (1 point increase) 577.0 (230.0 to 939.4)* 317.4 (14.2 to 669.1)* SLICC (1 point increase) 560.0 (393.0 to 1432.8) ISEL (1 point increase) 9.8 (63.4 to 41.5) PCS (1 point increase) 223.6 (360.8 to 137.1)* 188.5 (313.6 to 93.1)* MCS (1 point increase) 25.3 (124.6 to 73.2) General satisfaction (1 point increase) 48.7 (156.7 to 27.1) Technical (1 point increase) 58.5 (206.4 to 38.0) Interpersonal (1 point increase) 37.7 (130.6 to 35.6) Communication (1 point increase) 13.0 (72.8 to 60.1) Time spent with doctor (1 point increase) 55.4 (135.1 to 10.0) Access to care (1 point increase) 3.3 (84.5 to 70.8) *P-0.05. 43 Total costs and predictors of costs in SLE The assignment of hospital costs was mainly based on our hospital's previously determined charges for other providers and GP fundholders. For this reason they reect the direct costs for a patient treated at UCLH. In other parts of the country, hospital costs may be lower because of lower running costs and cheaper land. Many other economic evaluations conducted for other diseases in the UK used local cost data, sometimes in conjunction with national data w3739x. We would not be able to conduct such a detailed cost analysis based entirely on national cost data because of a lack of such detailed data. We determined the mean annual total cost of caring for patients with SLE as 7913 per patient. The direct cost constituted a third of the total cost (2613). The costs of in-patient careuday-case surgery and visits to health-care professionals contributed almost two-thirds of the direct cost. Although a relatively small number of patients had in-patient care, this high cost was due to the high prices of hospital care. Patients made many visits to health-care professionals, especially their rheumato- logists and GPs. The cost of medications, including dialysis and diagnostic procedures, contributed almost a third of the direct costs. We may have underestimated the cost of medications since we only included the drugs SLE patients most commonly used. Further analysis, as was done in the trinational study, indicates that other drugs may contribute to almost half of the overall cost of medications w7x. The indirect cost was two-thirds of the total cost. The mean annual indirect cost was 5299 per patient. Almost half of the patients did not work. Most disability and retirement was due to SLE. Signicant numbers of patients either lost work days or had disability days. Our indirect cost analysis was comprehensive and valued patients who did not work (patients who were unem- ployed but claimed they would be employed if they did not have SLE, and others who did not work, such as homemakers and retirees). This is particularly important for a condition like SLE, since the majority of patients are young women of reproductive age who may not be in the labour force. By considering the days they were unable to perform their household chores because of illness, we partially eliminated the potential bias against this group. However, by using the replacement cost method to value the lost time of those non-labour-force participants who would not be working if they were not ill, we may have understated their indirect costs. Since the wage information used in the replacement cost method is based on the wages of low-paid workers, this method might underestimate the indirect cost in patients with SLE. This is because our SLE patients tend to have an educational attainment near the average, which is higher than that of low-wage workers. Educational attainment is known to be related to the value of time w40, 41x. Therefore, replacement costs probably underestimate the true value of the average SLE patient. Although we attempted to be as comprehensive as possible in estimating both direct and indirect costs, we did not assess intangible costs. An alternative approach, used in costbenet studies estimating the social cost of disease, the willingness-to-pay approach, values not just the impact on productivity but also improved health per se. There are certain challenges in this method, and it was not used in this study. We think the conclusions of our study can be gener- alized. Although this study was performed at a spe- cialized SLE centre, the patients represented a spectrum from mild to severe cases, and there is no reason to suspect that our practice pattern should be different from that of others in the UK who treat patients with this condition. Furthermore, as was shown in the tri- national study, the characteristics of our patients were similar to those of other SLE populations w7x. TABLE 7. Total costs () associated with changes in potential predictors of costs Change in cost () (95% condence interval) Variable Unadjusted Adjusted Age (1 yr increase) 69.8 (210.8 to 62.8) Sex (male s1) 1432.0 (8495.8 to 122) Ethnic origin (non-Caucasians1) 828.9 (3605.5 to 4534.6) Marital status (unmarrieds1) 848.3 (2155.0 to 4948.4) Education level (1 yr increase) 689.0 (121.4 to 1487.6)* 584.1 (36.4 to 1158.0)* Disease duration (1 yr increase) 17.8 (204.3 to 275.4) SLAM (1 point increase) 933.2 (476.3 to 1343.5)* 557.0 (157.9 to 973.9)* SLICC (1 point increase) 1706.7 (629.4 to 2961.8)* ISEL (1 point increase) 29.1 (107.3 to 37.9) PCS (1 point increase) 316.0 (468.5 to 204.3)* 232.2 (398.0 to 99.1)* MCS (1 point increase) 4.7 (121.0 to 107.8) General satisfaction (1 point increase) 88.7 (196.1 to 1.4) Technical (1 point increase) 104.5 (234.9 to 32.1) Interpersonal (1 point increase) 57.2 (158.7 to 38.4) Communication (1 point increase) 43.4 (134.8 to 47.6) Time spent with doctor (1 point increase) 93.4 (176.7 to 11.7)* Access to care (1 point increase) 24.0 (70.3 to 122.9) *P-0.05. 44 N. Sutcliffe et al. There have been no previous studies examining the cost of care for patients with SLE in the UK. Comparison of our ndings with those of other cost-of-illness studies of SLE in other countries or other diseases in the UK would be of limited value because of the differences in the categories of cost, the methods used, the pattern of health services utilization and the health-care systems. In Canada, Clarke et al. w5x studied the total medical costs for patients with SLE using similar methods to assess health services utiliza- tion and lost productivity, although their assignment of prices was somewhat different, reecting the Canadian health-care system. They determined the mean total annual cost as $13 094 in 1990 Canadian dollars. Indirect costs constituted 54% of this, in comparison with 67% in our study. The total cost of in-patient care and day-case surgery was the biggest contributor to direct costs, as in our study (38%). This was followed by visits to health professionals in our study but diagnostic procedures in their study. In their study, predictors of higher direct costs were higher values of creatinine and a poorer level of physical functioning. Predictors of higher indirect costs were a poorer SLE well-being score, a combination of education and employment status (higher levels), and a weaker level of social support. Their study did not include standard measures of disease activity and damage. In the same cohort they also used regression trees to characterize patients with SLE who were expected to experience the highest direct and indirect costs and greatest diminished productivity. They found that patients with poor physical or poor psychological functioning incurred the highest direct costs, those with the poorest psychological functioning incurred the high- est indirect costs, and those with the most intense pain experienced the greatest impairment in productivity w6x. In the USA, Finn et al. w42x determined the direct annual medical cost for SLE patients in 198892. They found that in-patient costs were the major compon- ent of total costs. Blacks, females, hypertensives and patients with higher activity or damage scores had higher costs w42x. Health-care costs vary between countries. Gironimi et al. w12x showed that direct health-care costs for patients with SLE in the USA were almost twice those in Canada as a result of the higher price of health services in the USA and the more severe disease mix w12x. Although Canada has a national health-care system similar to the UK, the pattern of health services utilization is somewhat different and the applicability of Canadian studies to the UK is limited w7x. We also determined the predictors of cost. Patients with higher levels of education incurred higher costs. An increase of 1 yr in education was associated with increases of 205, 408 and 584 in direct, indirect and total costs respectively. This effect of education level on direct costs may be due to higher levels of com- pliance with treatment, higher attendance rates for clinic appointments and investigations, and perhaps the more demanding nature of the more highly educated. The effect on indirect costs is likely to be due to the higher earning capacity of this group, and thus any time lost from work would be more highly valued. Higher disease activity was also associated with higher costs. An increase of 1 score point in disease activity was associated with increases of 239, 317 and 557 in direct, indirect and total costs respectively. This is not surprising, since it is expected that patients with more active disease will make more use of the health services and will also take more time off work. Clarke et al. w5x also showed the effects of SLE activity (as represented by renal and physical functioning) on direct costs and the effects of global well-being on indirect costs. In our study, damage also had an impact on costs, although only on direct costs. An increase of 1 score point in total damage was associated with an increase of 876 in direct cost. It is to be expected that patients with more damage use more health services. It is also interesting that patients with better physical functioning incurred lower direct, indirect and total costs. The effect of physical functioning appears to be less than the effects of education level, disease activity and damage (an increase of 1 point in physical component summary scores was associated with reduc- tions of 38, 189 and 232 in direct, indirect and total costs respectively). However, it is likely that a 1-unit change in physical functioning is not as clinically important as a 1-unit change in disease activity or damage. The association of physical functioning with direct costs was also shown in the studies by Clarke et al. w5, 6x. Patients with better physical health may use health services less and lose less time from work. Unlike the studies of Clarke et al. mentioned above, our study showed that psychological functioning and social support did not inuence costs. Age also had an effect on direct costs in our study, older patients incurring lower costs (an increase of 1 yr in age was associated with a reduction of 39 in the direct cost). This may have been due to the reduction in disease activity as patients age, especially after the menopause. Our study had a different and more appropriate approach to statistical analysis compared with the studies of SLE costs mentioned above. Our approach did not require normative assumptions and was there- fore more appropriate for this sort of data. We used non-logged models, which are easier to interpret than log models. Log models are also a problem in studies which have large numbers of zero cost values (about 20% for indirect costs in this study). There have been a number of costs-of-illness (COI) studies for other rheumatic diseases in Europe, particu- larly RA. McIntosh w3x studied the cost of RA in England. The total economic effect of RA was estimated to be 1.256 billion in 1992, of which 52% was a result of production loss caused by RA disability. Direct costs amounted to 604.5 million, with hospitalization by far the largest expenditure, accounting for almost 30% of the direct cost. Nurse visits and home help visits combined made up 24% of the direct cost, with drugs, including toxicity management, accounting for 15% of the total. Van Jaarsveld et al. w4x studied the direct 45 Total costs and predictors of costs in SLE cost of RA during the rst 6 yr of disease in The Netherlands. The annual direct cost of RA averaged 3680 per patient. A high total direct cost in the rst 6 yr of disease was related to severe functional disability and lower age. COI studies have been criticized because of limitations in their comparability. Apart from the usual problems of measurement errors and biases, such as recall bias, the comparability of indirect cost of illness estimates may be compromised as a result of the different methods used w13x. McIntosh w3x also pointed out the limitations of COI studies. However, as she indicated, the COI framework, if used correctly and imaginatively, can be a valuable tool in cost-effectiveness analysis. By documenting the disaggregated costs of illness, as we have done in this paper, COI estimates can help policy-makers examine resource utilization, cost patterns, the effects of different treatment packages, the effects of changing patterns of service utilization and the results of health service cost- cutting procedures. These patterns can be monitored within the COI framework and provide invaluable information on shifting resource utilization and the effects on costs. In summary, we have shown that SLE has consider- able impact on the patient, the health-care system and society in general. Improvements in treatment to reduce disease activity and prevent damage, as well as improve- ments in physical health, are likely to reduce the costs incurred by these patients. Acknowledgements We gratefully acknowledge Jean Heath and Dimitria Panaritis for their technical assistance, Peter Large for his help in the assessment of costs at UCLH, Panos Kanavos at the London School of Economics for his advice and Bruce Bovill at the SAS Institute, Maidenhead, UK, for his support and advice. References 1. Lubeck DP. Health care economics and policy. Curr Opin Rheumatol 1991;3:28690. 2. Yelin E, Callahan LF. The economic cost and social and psychological impact of musculoskeletal conditions. Arthritis Rheum 1995;38:135162. 3. McIntosh E. Clinical audit, the cost of rheumatoid arthritis. Br J Rheumatol 1996;35:78190. 4. Van Jaarsveld CHM, Jacobs JWG, Schrijvers AJP, Heurkens AHM, Haanen HCM, Bijlsma JWJ. Direct cost of rheumatoid arthritis during the rst six years: a cost of illness study. Br J Rheumatol 1998;37:83747. 5. Clarke AE, Esdaile JM, Bloch DA, Lacaille D, Danoff DS, Fries JF. A Canadian study of the total medical costs for patients with systemic lupus erythematosus and the predictors of costs. Arthritis Rheum 1993;36:154859. 6. Clarke AE, Bloch DA, Danoff DS, Esdaile JM. Decreasing costs and improving outcomes in systemic lupus erythematosus: Using regression trees to develop health policy. J Rheumatol 1994;21:224653. 7. Clarke AE, Petri MA, Manzi S, Isenberg DA, Gordon C, Senecal JL et al. An international perspective on the well-being and health care costs for patients with systemic lupus erythematosus. J Rheumatol 1999;26:150011. 8. Sutcliffe N, Clarke AE, Levinton C, Frost C, Gordon C, Isenberg DA. Associates of health status in patients with systemic lupus erythematosus. J Rheumatol 1999;26:23526. 9. Sutcliffe N, Clarke AE, Gordon C, Farewell V, Isenberg DA. The association of socio-economic status, race, psychosocial factors and outcome in patients with systemic lupus erythematosus. Rheumatology 1999;38:11307. 10. Clarke AE, Penrod J, St Pierre YS, Petri MA, Manzi S, Isenberg DA et al. Underestimating the value of women: assessing the indirect costs of women with systemic lupus erythematosus. J Rheumatol 2000;27:2597604. 11. Moore AD, Petri MA, Manzi S, Isenberg DA, Gordon C, Senecal J-L et al. The use of alternative medical therapies in patients with systemic lupus erythematosus. Arthritis Rheum 2000;43:14108. 12. Gironimi G, Clarke AE, Hamilton VH, Danoff DS, Fries JF, Esdaile JM. Why health care costs more in the US: Comparing health care expenditures between SLE patients in Stanford and Montreal. Arthritis Rheum 1996; 39:97987. 13. Prashker MJ. Health care research and policy studies. Curr Opin Rheumatol 1994;6:1406. 14. Gold MR, Siegel JE, Russell LB, Weinstein MC. Cost-effectiveness in health and medicine. New York: Oxford University Press, 1996. 15. Maetzel A. Costs of illness and the burden of disease. J Rheumatol 1997;24:35. 16. Tan EM, Cohen AS, Fries JF et al. The 1982 revised criteria for the classication of systemic lupus erythema- tosus. Arthritis Rheum 1982;25:12717. 17. Gladman DD, Goldsmith CH, Urowitz MB, Bacon P, Bombardier C, Isenberg D et al. Sensitivity to change of 3 systemic lupus erythematosus disease activity indices: Internal validation. J Rheumatol 1994;21:146871. 18. Liang MH, Socher SA, Larson MG, Schur PH. Reliability and validity of six systems for the clinical assessment of disease activity in systemic lupus erythematosus. Arthritis Rheum 1989;32:110718. 19. Gladman DD, Goldsmith CH, Urowitz MB, Bacon P, Bombardier C, Isenberg D et al. Cross cultural validation and reliability of 3 disease activity indices in systemic lupus erythematosus. J Rheumatol 1992;19:60811. 20. Gladman DD, Ginzler E, Goldsmith C, Fortin P, Liang M, Urowitz MB et al. The development and initial validation of the Systemic Lupus International Collaborating Clinicsu American College of Rheumatology Damage Index for Systemic Lupus Erythematosus. Arthritis Rheum 1996; 39:3639. 21. Stoll T, Seifert B, Isenberg DA. SLICCuACR damage index is valid, and renal and pulmonary organ scores are predictors of severe outcome in patients with systemic lupus erythematosus. Br J Rheumatol 1996; 35:24854. 22. Gladman DD, Urowitz MB, Goldsmith CH, Fortin P, Ginzler E, Gordon C et al. The reliability of the Systemic Lupus International Collaborating ClinicsuAmerican College of Rheumatology Damage Index in patients with systemic lupus erythematosus. Arthritis Rheum 1997;40:80913. 23. Garratt AM, Ruta DA, Abdalla MI, Buckingham JK, Russell IT. The SF-36 health survey questionnaire: an 46 N. Sutcliffe et al. outcome measure suitable for routine use within the NHS. Br Med J 1993;306:14404. 24. Ware JE Jr, Kosinski M, Keller SD. SF-36 Physical and Mental Summary Scales: A user's manual. Boston: The Health Institute, New England Medical Center, 1994. 25. Jenkinson C, Wright L, Coulter A. Criterion validity and reliability of the SF-36 in a population sample. Qual Life Res 1994;3:712. 26. Stoll T, Gordon C, Siefert B et al. Consistency and validity of patient administered assessment of quality of life by the MOS SF-36: its association with disease activity and damage in patients with systemic lupus erythematosus. J Rheumatol 1997;24:160814. 27. Cohen S, Mermelstein R, Kamack T, Hoberman HM. Measuring the functional components of social support. In: Sarason IG, ed. Social support: theory, research and applications. The Hague: Martinus Nijhoff, 1985:7394. 28. Ware JE Jr, Snyder MK, Wright R, Davies AR. Dening and measuring patient satisfaction with medical care. Eval Program Plann 1983;6:24763. 29. Fries JF, Spitz P, Kraines RG, Holman HR. Measurement of patient outcome in arthritis. Arthritis Rheum 1980; 23:12734. 30. Clarke AE, Zowall H, Levinton C, Assimakopoulos H, Sibley JT, Haga M et al. The direct and indirect medical costs incurred by Canadian patients with rheumatoid arthritis: A twelve year study. J Rheumatol 1997; 24:105160. 31. Department of Health and Ofce of Population Censuses and Surveys. The government's expenditure plans 199697 to 199899. London: Her Majesty's Stationery Ofce, 1996. 32. Drug tariff, October 1996. London: Department of Health, Welsh Ofce, 1996. 33. New earnings survey 1996: Part A. Streamlined and summary analyses: Description of the survey, September 1996. London: Ofce for National Statistics, 1996. 34. Barber JA, Thompson SG. Analysis and interpretation of cost data in randomised controlled trials: review of published studies. Br Med J 1998;317:1195200. 35. Efron B, Tibshirani R. Introduction to the bootstrap. London: Chapman and Hall, 1993. 36. Mallows CL. Some comments on Cp. Technometrics 1973;15:66175. 37. Coast J, Richards SH, Peters TJ, Gunnell DJ, Darlow MA, Pounsford J. Hospital at home or acute hospital care? A cost minimisation analysis. Br Med J 1998; 316:18026. 38. Shepperd S, Harwood D, Gray A, Vessey M, Morgan P. Randomised controlled trial comparing hospital at home care with inpatient hospital care. II: cost minimisation analysis. Br Med J 1998;316:17916. 39. UK Prospective Diabetes Study Group. Cost effectiveness analysis of improved blood pressure control in hyperten- sive patients with type 2 diabetes: UKPDS 40. Br Med J 1998;17:7206. 40. Gronau R. Home production. A survey. In: Ashenfelter OC, Layard R, eds. Handbook of labor economics, Vol. 1. Amsterdam: Elsevier Science, 1986:273304. 41. Willis RJ. Wage determinants: A survey and reinterpreta- tion of human capital earnings functions. In: Willis RJ, Ashenfelter OC, Layard R, eds. Handbook of labor economics. Amsterdam: Elsevier Science, 1986:525602. 42. Finn D, Allen P, Hoffstein, Goldman D, Petri M. Economic costs of systemic lupus erythematosus: A ve year prospective study wabstractx. Arthritis Rheum 1993;36(Suppl. 9):192. 47 Total costs and predictors of costs in SLE