1 Technical appendix Cost-effectiveness of home UVB phototherapy for psoriasis: economic. Content: page 1. Clinical study methods 1 2. Costs of home ultraviolet B treatment 1 3. Costs of ultraviolet B treatment in an outpatient setting 2 4. Costs of productivity losses 3 5. Estimating EQ-5D values and SF-6D values using multilevel models 4 6. Resource use crossovers 6 1. Clinical study - methods End of phototherapy: was defined to be the last irradiation. When UVB treatments exceeded 46 irradiations, we defined 46 irradiations as the end of the therapy (cut off point). Outpatient phototherapy: Hospital personnel routinely evaluated the treatments and prepared the equipment for each subsequent irradiation. Home phototherapy: After instruction, the patients received a treatment schedule. The patients evaluated the treatments and set the treatment times by themselves, and could contact the nursing staff of the home care organisations for supervision. At the end of the treatment period, the home care organisations collected the units. The cost for their services, delivery and pick up of the phototherapy unit was included in the rental price. Blinding: Blinding participants for treatment obviously was impossible, and because of the pragmatic design it was undesirable to blind the dermatologist. However, an independent research nurse blinded to treatment arm assessed the extent and severity of the psoriasis. 1;2 2. Costs of home ultraviolet B treatment: The treatment costs of home phototherapy were calculated for each patient individually, based on commercial tariffs (invoice prices, 2003 price level) of the two Dutch home care organisations providing home phototherapy. The invoice prices are presented in the table below: Home care organisation A: Home care organisation B: Minimum fee: 8 weeks 421.20 Minimum fee: 12 weeks 631 Every additional week 42.12 Every additional week 34 13 weeks 589.68 26 weeks 982.80 52 weeks 1432.08
2 3. Costs of ultraviolet B treatment in an outpatient setting Costs of outpatient ultraviolet B treatment were calculated from the societal perspective using the costs of equipment, staff, accommodation and overhead. A unit price for outpatient phototherapy (1 unit = 1 irradiation) was computed from the following figures: Equipment: Purchase costs (incl VAT): Depending on type of equipment: Waldmann UV7001 29.155 Waldmann UV7001K 24.990 Waldmann UV1000L 14.280 Life span: 15 years Value after depreciation: 0 Interest: 5% Maintenance materials: 100 for 1 TL-01 tube replacement of one TL-01 tube for every 137.6 irradiations (estimated from replacement data in University Medical Center Utrecht) Annual use of equipment The following figures are estimated from data derived from participating hospitals: university hospital general hospital Type of equipment (proportion) Waldmann UV7001 37.5% 46.15% Waldmann UV7001K 62.5% 15.38% Waldmann UV1000L - 38.46% Irradiation units per hospital (mean, n) 1.6 1.44 Irradiations per hospital (n, annually) 4735 3000 Staff: Maintenance of equipment: Supervision treatments: 4 hours per unit per year by nurses or physiotherapists, costs depending on deployment, number of irradiations per year and wage classifications of the hospitals. Average time deployed per irradiation (estimated from data derived from participating hospitals): 26.36 minutes in university hospital and 10.2 minutes in general hospital. Overhead: 35% of costs of equipment and personnel Accommodation: 10% of costs of equipment and personnel
3 4. Costs of productivity losses Productivity losses are caused by absence from paid work, absence from unpaid work, and reduced productivity while at work (presenteeism). In the study, costs of productivity losses per hour paid work were calculated while applying mean hourly productivity costs varying with age and gender as published in the manual for costing. 3 Costs of absence at work: To calculate the costs of absence at work, we applied the Health and Labour questionnaire, containing questions to capture time lost at work due to health problems (including psoriasis). Many participants, however, failed to answer the question regarding the number of hours of absence at work. From the participants who did answer the question, the majority reported to have had no absence from work due to health problems (median values indicated 0 hours of absence per week). The few participants who reported to have missed time at work due to psoriasis or other health problems gave answers varying from an absence at work of 30 minutes to 33 hours per week. Because of the short distances to the hospital in the country where the trial was performed (the Netherlands), it is highly unlikely that an absence of 33 hours per week is attributable to 3 short visits to the hospital for office-based phototherapy. These long absences at work therefore must have been caused by other health problems such as for instance the flu. Unfortunately the longer absences described above were few and caused a skewed distribution of absenteeism within both groups, highly influencing mean values. As such, using the data provided by this question from the Health and Labour questionnaire would not be representative of absenteeism due to (phototherapy for) psoriasis. The data did, however, indicate that most patients have flexible arrangements with their employers and are capable to minimise the costs due to absenteeism for their treatment. To test this hypothesis we performed a pilot study among 36 patients receiving office-based phototherapy. The results of this pilot study confirmed that many patients arrange that irradiations are either planned at their day off, or at the beginning or end of the day, or during a (lunch) break. And when this is not possible and they do have to be absent at work for a few hours per week, they reported to compensate for this during normal working hours. Mostly, the tasks were not passed on to colleagues, and most patients reported that according to their opinion there were no productivity losses for their employer due to the officebased phototherapy. These results and our conclusions are in line with previous studies which reported that short-term absence is often compensated for during normal working hours. 4;5 Therefore for this population and this intervention we considered (the costs due to) short term absenteeism from paid and unpaid work negligible.
4 5. Estimating EQ-5D values and SF-6D values using multilevel models Background Since health-related quality of life depends on health, we assumed that in a population suffering from psoriasis health related quality of life and its fluctuations may be predicted from psoriasis severity. The EQ-5D and SF-6D instruments are health related quality of life instruments, which were both not measured during the follow-up of our trial (after the end of phototherapy). Using analysis with linear multilevel models we substantiated our assumption that psoriasis severity can adequately predict health related quality of life, such as the EQ-5D and SF-6D. Methods To be able to estimate EQ-5D and SF-6D values from the measured psoriasis severity and other determinants while taking advantage of our longitudinal design, we used linear multilevel models. 6 The expression multilevel model is a technical terminology, used for modelling clustered data which is the case in longitudinal data and repeated measures. Patients represent the highest level, the repeated measures within each patient represent the lowest level. In this approach, it is assumed that missing data are missing at random (MAR). 7 The interpretation of linear multilevel models is rather similar to that of linear regression models. Both types of models produce regression coefficients with accompanying standard errors. From a (multilevel) regression model, the variable of interest (dependent variable, Y) can subsequently be estimated from several predicting (explanatory) variables (X) using the regression coefficients (β). An example of a model is: Y = intercept + β 1 *X 1 + β 2 *X 2 + β 3 *X 3 + etcetera. As such, the regression coefficients describe the relationship between the dependent variable Y and the predicting variables X in terms of One point increase in variable X 1 gives a β 1 point increase in variable Y, given that all other predicting variables remain unchanged. The intercept is the value of the dependent variable Y when all X variables equal zero. The major advantage of multilevel modelling as opposed to ordinary linear regression analysis is that linear multilevel models take an additional source of variability into account, namely the correlation between repeated measurements within one person. As such, linear multilevel models can be applied on data comprising several repeated measurements per person (longitudinal data), whereas linear regression models assume independent observations. We created separate models for the EQ-5D and SF-6D. In order to get easy-to-read effect estimates we multiplied the EQ-5D and SF-6D utility-scores by 100 (EQ-5Dx100 and SF-6Dx100) before creating the multilevel models. During modelling, fixed and random effects were considered. As starting point we used the model with random intercept and without other explanatory variables, only considering within patient variability and between patient variability. For both dependent variables we sequentially included several demographic determinants, patient specific characteristics and interaction terms in the model. We considered demographic variables such as gender, age, level of education, employment status, living alone, children living at home and age of the youngest child living at home. Patient characteristics comprised the Self Administered Psoriasis Area and Severity Index (SAPASI), age at onset of psoriasis, concomitant use of medication, time passed since UV therapy had started, treatment-arm and previous (home) treatment with ultraviolet B
5 light. Predicting variables and interaction terms of the predicting variables are described when significant. The goodness of fit was evaluated by the change in deviance of the (nested) models. We used the MLwiN software package for fitting multi-level models (version 2.02; Center for multilevel Modelling, Institute of Education, University of London, UK). Results The SAPASI-score, gender and employment status appeared to be significant predictors of the EQ-5D and SF-6D. The results of the fixed effects of the models of the various HRQoL measures are presented in the table below. Table. Regression coefficients for the linear multilevel models. Predictors EQ-5D (x100) 1 SF-6D (x100) 1 Intercept 2 89.843 (2.328) 82.499 (1.667) SAPASI 3 (0 to 28 points) -1.428 (0.162) -0.976 (0.192) Gender Male (0) 0 (ref) 0 (ref) Female (1) -10.339 (2.249) -7.939 (1.494) Employment status Unemployed (0) 0 (ref) 0 (ref) Employed (1) 8.341(2.320) 6.471 (1.785) Employment status x SAPASI 4 Unemployed (0) x SAPASI 0 (ref) Employed (1) x SAPASI -0.488 (0.230) Using linear multilevel models, the SAPASI score 3 and other important predictors were regressed on the EQ-5D 1, and SF-6D 1. Values presented are fixed effects (standard errors) of the final models and include the significant predictors. Each regression coefficient represents the change in mean health related quality of life for 1 point increase in the corresponding predictor, given that all other predictors remain unchanged. An empty cell indicates that the variable was of no predictive value for the dependent variable of interest. 1 EQ-5D and SF-6D utility scores times 100 2 Intercept: y-intercept, point of origin, the coordinate of the point at which the curve intersects the y-axis 3 SAPASI = Self Administered Psoriasis Area and Severity Index 4 Continuous predictor SAPASI 3 as part of an interaction term. Range SAPASI = 0 to 28 points. In both models the following variables had no predictive value: age, level of education, time passed since UV therapy had started, children living at home, age at onset of psoriasis, living alone, age of the youngest child living at home, concomitant use of medication, treatment-arm and previous (home) UVB treatment.
6 When the interaction term (product) of two determinants was a significant predictor, we also presented the main effects for the separate determinants, regardless of their significance. For the sake of simple representation, we did not show the random effects. Using the results of table, the models for both measures can be described using their own separate regression-coefficients. Each regression coefficient represents the change in mean health related quality of life (that is, 100xEQ-5D or 100xSF-6D) for 1 point increase in the corresponding predictor, given that all other predictors remain unchanged. In general, positive regression coefficients indicate an improvement in quality of life, negative regression coefficients stand for a worsening of the quality of life. For a clear comprehension of all models, we will discuss the results of the SF-6Dx100 model. The model can be represented as follows: SF-6Dx100 = 82.499 0.976xSAPASI 7.939 (only for women) + 6.471 (only when employed ) 0.488xSAPASI (only when employed) As shown in the model of the SF-6D, we found a clear association between the SF-6D and the SAPASI, but gender and employment were also of influence. For instance, for each point increase of the SAPASI score the SF-6Dx100 worsened by 0.976 points, given that all other predictors remained unchanged. Similarly, women had a SF-6Dx100 score 7.939 point lower than men, but being employed improved the SF-6Dx100 with 6.471 points. We also found interaction between employment and SAPASI, meaning that for each point increase in the interaction term (product) of the SAPASIscore and employment (employment status x SAPASI) the SF-6Dx100 worsened a further 0.488 point. The model of the EQ-5Dx100 can be represented as: EQ-5Dx100 = 89.843 1.428xSAPASI 10.339 (only for women) + 8.341 (only when employed) 6. Resource use - crossovers Until the end of phototherapy, four patients who were initially randomised to receive outpatient treatment switched to home treatment. One patient switched from home treatment to phototherapy in an outpatient setting. Three other patients randomised in the home group started with outpatient phototherapy during the wait for home phototherapy, and later on continued their treatment at home. During the follow-up period (t=4 to t=9), 14 patients randomised in the group treated at home versus 11 patients randomised to the group treated in the outpatient department started a new treatment with ultraviolet B light. The new phototherapy treatment took place at home for 8 of them (5 versus 3), and in the outpatient department for 17 participants (9 versus 8).
7 Reference List 1 Koek MBG, Buskens E, Steegmans PHA, van Weelden H, Bruijnzeel-Koomen CAFM, Sigurdsson V. UVB phototherapy in an outpatient setting or at home: a pragmatic randomised single-blind trial designed to settle the discussion. The PLUTO study. BMC Med Res Methodol 2006;6:39. 2 Koek MBG, Buskens E, van Weelden H, Steegmans PH, Bruijnzeel-Koomen CAFM, Sigurdsson V. Home versus outpatient ultraviolet B phototherapy for mild to severe psoriasis: pragmatic multicentre randomised controlled non-inferiority trial (PLUTO study). BMJ 2009;338:b1542. 3 Oostenbrink JB, Bouwmans CAM, Koopmanschap MA, Rutten FFH. [Manual for Costing: Methods and Standard Costs for Economic Evaluations in Healthcare]. Diemen, Netherlands: College voor Zorgverzekeringen; 2004. 4 Jacob-Tacken KH, Koopmanschap MA, Meerding WJ, Severens JL. Correcting for compensating mechanisms related to productivity costs in economic evaluations of health care programmes. Health Econ 2005;14(5):435-43. 5 Severens JL, Laheij RJ, Jansen JB, Van der Lisdonk EH, Verbeek AL. Estimating the cost of lost productivity in dyspepsia. Aliment Pharmacol Ther 1998;12(9):919-23. 6 Snijders T, Bosker R. Multilevel analysis: An Introduction to Basic and Advanced Multilevel Modeling. London: Sage publishers; 1999. 7 Rubin DB. Inference and missing data. Biometrika 1976; 63:581-92.