Author for correspondence: Palle Mark Christensen, Clinical Pharmacology, Institute of Public Health, University of Southern Denmark, Winsløwparken 19, 5000 Odense C, Denmark (fax +45 65 91 60 89, e-mail firstname.lastname@example.org).
Abstract: Pharmacological interventions for osteoporosis may reduce morbidity and mortality, but they incur additional health care costs. The aim was to quantify the additional costs and health benefits of prescribing alendronate 10 mg and calcium/vitamin D daily for 71-year-old women with a fracture risk twice that of the population average in stead of calcium/vitamin D alone. A state transition model based primarily on Scandinavian data was developed. Women were followed from age of 71 years until 100. Alendronate was assumed to reduce the fracture risk by 50%. Health benefits from the interventions were expressed in terms of life years, quality adjusted life years, and fractures avoided. Societal costs were estimated using literature estimates and Danish tariffs. All costs were measured in 2002 Danish Kroner (DKK). Future costs and benefits were discounted at 5% per year. The incremental cost per QALY gained was DKK125,000 while the cost per life year gained was DKK 374,000. The use of alendronate was cost-saving when 1) the treatment was extended to five years, 2) the risk of fracture was four times the population average, 3) the effect of alendronate was assumed to persist for three years after discontinuation of treatment, 4) a greater proportion had severe sequelae after a hip fracture, or 5) the start of therapy was delayed until age of 77 years. In conclusion, the use of alendronate compares well with other well established therapies in terms of cost-effectiveness in older women with high risk of fracture.
Osteoporosis is a condition characterised by low bone mass and microarchitectural deterioration of bone tissue with a consequent increase in bone fragility and susceptibility to fractures (Consensus Development Conference 1993). The fractures mainly occur in the hip, the spine and the forearm (Wasnich 1999). Osteoporotic fractures pose a great burden to patients and the health care system because they cause considerable pain, disability, and even premature deaths (Kanis et al. 2001).
Hip and vertebral fractures are associated with excess mortality, but only a fraction of this excess mortality is reversible by an intervention (Kanis & Jonsson 2002), because co-morbidities (e.g. cancer, dementia) explain some of the excess mortality (Browner et al. 1996). Thus, osteoporosis interventions primarily improve quality of life rather than longevity. A number of economic models in the field of osteoporosis have been developed (Zethraeus et al. 2002a; Iglesias et al. 2002; Willis 2002; Johnell et al. 2003), but our model differs from most others in that it allows for simulations of the extent to which an intervention reverses the excess mortality associated with hip fractures.
The aim of this study was, from a broad health-care sector perspective, to asses the cost effectiveness (expressed in terms of cost per quality adjusted life year gained, cost per life year gained, cost per hip fracture avoided, cost per vertebral fracture avoided, cost per forearm fracture avoided) of using alendronate 10 mg per day for three years compared with calcium and vitamin D in 71- year-old Danish women, with a fracture risk twice the population average.
Materials and Methods
General model. The model can follow a cohort of 10,000 women from 50 years of age (71 in this specific analysis) until 100 years of age or until death. The model consists of nine health states, age-dependent transition probabilities determine how simulated patients move from one state to another. The transitions occur in 1-year cycles.
Transitions in the Markov cohort simulation model. Starting in the “well” state a woman can experience a forearm fracture, a vertebral fracture or a hip fracture or remain “well” (unfractured) during the first year (fig. 1). If a woman sustains a forearm fracture, she returns to the “well” state at the end of the cycle. If a woman sustains a vertebral fracture, she will either move to “sequelae vertebral” or “well” states at the end of the cycle. If a woman sustains a hip fracture, she will end up (i) “dead”; (ii) having “mild sequelae” and subsequently return to “well”; (iii) having “moderate sequelae hip” and then remain in this state or become “well”; or (iv) having “severe sequelae hip”. Women not sustaining a fracture can either remain “well” or “die”. In the next cycle, i.e. in the first year after the fracture, a woman can begin in one of the following cycles: “well”; “sequelae vertebral”; “moderate sequelae hip” or “severe sequelae hip”. For those women who are “well”, the possible transitions are those described above (i.e. fracture/not fracture etc.). Patients starting in a “sequelae” health state can either remain in this specific state or “die”.
Transition probabilities. All literature used in this analysis was identified through a systematic search using the Medline database for the period 1980 to December 2002 (table 1). For further details see the technical report (Christensen et al. 2003). We used age and sex specific mortality rates, for Denmark (http://www.dst.dk/dst/665). The risk of sustaining a hip fracture was based on a Norwegian study (Lofthus et al. 2001), because this offered the most updated and validated information. Preliminary Danish data indicate that the incidence of hip fracture is of the same order of magnitude in Denmark (Poulsen et al. 2001). The only identified study on incident clinical vertebral fracture was an American population-based study (Cooper et al. 1992). The risk of forearm fractures was based on a Danish study (Solgaard & Petersen 1985). No Scandinavian studies of the long-term (1 year and beyond) post-fracture effects of a forearm fracture could be identified. In the model, all patients (100%) return to the “well” state in the year after sustaining a forearm fracture. After sustaining a vertebral fracture, 25% of the patients were assumed to have sequelae and the rest to be well after one year (Chrischilles et al. 1994). We were unable to identify Scandinavian studies of the long-term (1 year and beyond) effect of a vertebral fracture. We identified eight Scandinavian studies of mortality during the first year after a hip fracture (Kreutzfeldt et al. 1984; Elmerson and Zetterberg 1989; Eiskjaer et al. 1992; Berglund-Roden et al. 1994; Advocaat and Bautz-Holter 1997; Falch and Meyer 1998; Lund 1998; Meyer et al. 2000). We used the only Danish data available, reporting the mortality in the year after a hip fracture to vary from 10–30% depending on age (Eiskjaer et al. 1992). We found seven different Scandinavian publications (Jensen et al. 1979; Jensen & Tondevold 1979; Holmberg et al. 1987; Ankjær-Jensen et al. 1994; Finsen et al. 1995; Cserhati et al. 2002; Tidemark et al. 2002) describing the probabilities of various hip fracture sequelae. Finsen et al. (1995)found that one year after the fracture, 30% of the patients needed no aids, 48% needed either one or two sticks or a walking frame, while 21% were bedridden. These percentages changed very little during the next two years of follow-up. A recent Swedish study (Tidermark et al. 2002) indicates that 12% of those living independently before the hip fracture end up in institutions. Furthermore, 71% needed no walking aids before the fracture compared to 34% after the fracture and finally 59% needed household help after the fracture compared to 38% before the fracture. Jensen et al. (1979)found that 8% were discharged to a nursing home among 383 Danish patients admitted from their own home. In another Danish study (Ankjær-Jensen et al. 1994) of 180 patients admitted to hospital from their own home 5% were discharged to a nursing home. We assumed that 30% developed “mild sequelae hip”, 60% “moderate sequelae hip” and 10% “severe sequelae hip” in the base case. All patients in “mild sequelae hip” were assumed to be “well” at the end of the cycle. Fifty percent of those with “moderate sequelae hip” ended up being “well” and the rest remained “moderate sequelae hip”. All patients with “severe sequelae hip” were assumed to remain in this health state. Patients starting the next cycle in a sequelae health state were assumed to have 25% increased mortality increase by 25% compared to those without sequelae.
Table 1. Transitions in Markov cohort simulation and source of transitions probabilities.
Source of transition probabilities
Cooper et al. (1992)
Lofthus et al. (2001)
Statistics Demark (2003)
Well (forearm fracture)
Well (vertebral fracture)
Dead (hip fracture)
Eiskjaer et al. (1992)
Mild sequelae hip
Finsen et al. (1995); Jensen et al. (1979)
Moderate sequelae hip
Finsen et al. (1995); Jensen et al. (1979)
Server sequelae hip
Finsen et al. (1995); Jensen et al. (1979)
Mild sequelae hip
Well (mild hip)
Moderate sequelae hip
Well (moderate hip)
Moderate sequelae hip
Moderate sequelae hip
Severe sequelae hip
Severe sequelae hip
Moderate sequelae hip
Health related quality of life (HRQOL) in different health states. We identified seven empirical studies describing HRQOL values, for one or more of the osteoporosis-related conditions (established osteoporosis, hip, vertebral, and forearm fracture) (Dolan et al. 1999; Gabriel et al. 1999; Oleksik et al. 2000; Salkeld et al. 2000; Tosteson et al. 2001a; Brazier et al. 2002; Tidermark et al. 2002). All HRQOL values were based on generic preference based instruments. While the health state is 1.0 by definition when in perfect health, HRQOL is gradually reduced by age even without fractures. We used Danish data for the general population (Pedersen et al. 2003) as norms for pre-fracture HRQOL (table 3) and these values were reduced further by sustaining a fracture and potentially suffering permanent sequelae. According to a systematic review by Brazier et al. (2002), the pre-fracture HRQOL is reduced by a factor 0.797 when suffering a hip fracture, 0.909 when suffering a vertebral fracture, and 0.981 when suffering forearm fracture (all of these values were derived from studies using the multi-attribute instrument, called EQ-5D) (table 2). The impact on health state utility values from a fracture was assumed to be less after the first year. No empirical data on size of this impact for the subsequent years after a fracture have been published. We adjusted the aforementioned multiplies as suggested by Brazier et al. (2002) assuming that hip fractures have half the impact in the subsequent years (i.e. 0.9 for moderate sequelae, and 0.85 for severe sequelae) and for consistency the same assumption was made for vertebral fractures (i.e. 0.955). A forearm fracture was assumed not to have any impact beyond the first year.
Table 3. Danish data for the general population as norms for pre-fracture health related quality of life (Pedersen et al., 2003).
Health state value
Number of persons interviewed
Table 2. Quality of life values in patients with fracture. These values should be applied to values for pre-fracture health related quality of life values.
Value (95% confidence interval)
Brazier et al. (2002)
Oleksik et al. (2000)
Dolan et al. (1999)
Costs. Most of the cost estimates presented in this analysis were based on tariffs and health care utilization data from the Danish and international literature. In principle we include all health care costs including medical costs as well as costs of home help and nursing home. Indirect costs are not included. All costs were expressed in 2002 Danish kroner (DKK).
The annual cost of alendronate 10 mg daily for is DKK 4,535 (http:www.dkma.dkmedicinpriser (accessed January 2003). Women were assumed to see their physician three times during the first year of intervention and have a total of three biochemical tests taken. The cost of a visit to the general practitioner was DKK 103 and DKK 43 for each biochemical test (www.dadlnet.dk/plo/open/4 overenskomser/honrartabe/lok-okt02.htm). Thus, the total first year costs was DKK 6,175. During the following years of intervention the woman will see her general practitioner once a year for a clinical check up at a cost of DKK 103 and have one set of biochemical tests done at a cost of DKK 43. Moreover, a measurement of bone mineral density will be performed every second year during the intervention period. The cost of a bone mineral density scan was DKK 1,200 (Takstkatalog 2002, Odense University Hospital, Danish 2002). Thus, the total costs during the following years were DKK 5,282 per year.
We identified two Scandinavian studies addressing the costs in the first year following a hip fracture. In both studies, the health care costs the last year before the fracture was used as control (Zethraeus et al. 1997; Zethraeus & Gerdtham, 1998). This design accounts for the crucial fact that hip fracture patients would probably consume a substantial amount of health care resources and other services even without the fracture. Beside the two Scandinavian studies we are aware of two other studies using this type of design (Brainsky et al. 1997; De Laet et al. 1999). The costs estimates from the Swedish study (Zethraeus & Gerdtham 1998) were used.
For the costs of hip fractures in the subsequent years, we have not been able to identify any studies using patients as their own control, nor any studies using empirical data. We chose to use estimates based on expert opinions (Christensen et al. 2003). All patients with severe sequelae after a severe hip fracture were assumed to end up in a nursing home. Thus, the total average costs the subsequent years for patients with sequelae after a severe hip fracture was assumed to be DKK 365,000 (Christensen et al. 2003). Based on expert opinion, we assumed that the average costs for the subsequent years for those with moderate sequelae was DKK 2,742 (Christensen et al. 2003).
Regarding costs of forearm fractures, we did not identify any Scandinavian studies comparing the health care costs one year after the forearm fracture with the health care costs one year before the forearm fracture. Based on expert judgements we assumed that the average costs for a forearm fracture was DKK 7,645 (Christensen et al. 2003). This was roughly in accordance with two previous studies (Andersen et al. 1995; Ankjær-Jensen & Johnell 1996). A Swedish study, found that the direct costs were SEK 19,362. This was considerably higher than the studies based primarily on expert opinions (including our own). As a base case assumption we chose a mean cost of DKK 13,000.
Concerning the costs during the first year following a vertebral fracture, we could not identify any Scandinavian studies comparing the health care costs one year after the vertebral fracture with the health care costs one year before the vertebral fracture. Based on expert opinion, we assumed that the average costs the first year after a vertebral fracture was DKK 6,901 (Christensen et al. 2003). These costs were in between data from a Danish study (Ankjær-Jensen & Johnell 1996) (DKK 3,790) and a Norwegian study (Andersen et al. 1995) (NOK 24,784). Both of these studies were based on expert opinions. A Swedish study (Zethraeus et al. 2002b) based on year 2000 empirical data, found the direct costs to be SEK 30,470. This was considerably higher than the studies based primarily on expert opinions (including our own estimates). As a base case assumption we chose a cost of DKK 20,000. Lacking Scandinavian studies we assumed that the average costs of subsequent years for those ending up having sequelae after a vertebral fracture was DKK 1,914 (Christensen et al. 2003).
Effect of intervention. . The effect of alendronate 10 mg daily was based on an American study of 2,027 women with a mean age of 71 years with at least one vertebral fracture and a bone mineral density (BMD) that was 2.1 SD below the BMD of young women (Christensen et al. 2003).
Effect of intervention. The effect of alendronate 10 mg daily was based on an American study of 2,027 women with a mean age of 71 years with at least one vertebral fracture and a bone mineral density that was 2.1 S.D. below the bone mineral density of young women (Black et al. 1996). These women were randomised to receive for three years either alendronate plus calcium and vitamin D or calcium and vitamin D. The effects (and costs) of calcium and vitamin D were omitted from the analysis since both groups received calcium and vitamin D. Twenty-three (2.3%) women in the aldendronate-treated group had a clinical apparent vertebral fracture compared to 50 (5.0%) in the placebo-treated group (relative hazard 0.45 [95% CI 0.27–0.72]). For hip fracture and forearm fracture the corresponding figures were 11 (1.1%) alendronate and 22 (2.2%) placebo (relative hazard 0.49 (0.23–0.99) and 22 alendronate (2.2 %) and 41 placebo (4.1%) (relative hazard of 0.52 [95%CI 0.31–0.87]), respectively. We assumed that intervention was offered to 71- year-old Danish women with a risk of sustaining a fracture twice that of the age- and sex-specific population average. We assumed that the risk of sustaining a hip, forearm, or vertebral fracture to be increased by a factor of 1.9 for each standard deviation decrease in bone mineral density (Schott et al. 1998). An intervention in 71-year-old women with an untreated risk of fracture twice the population average will have a z-score of −1.1 (ln(2)/ln(1.9)) which is approximately equivalent to a t-score of −2.9 in femur (Looker et al. 1998). Previous research indicates that one standard deviation reduction in bone mineral density results in an increase in mortality by a factor 1.19 (Browner et al. 1991) for each standard deviation decrease in bone mineral density. In the base case, we assumed 100% compliance and that alendronate was assumed to reduce the risk of fracture by 50% immediately and maintained for one year after discontinuation. Finally, we assumed that only 14% of the deaths following a hip fracture were caused by or hastened by the hip fracture (Browner et al., 1996) and thus preventable by intervention.
Programming of the model. The model was developed in Microsoft Excel. The total number of fractures was estimated by multiplying the incidence of fractures by the total number of women entering the model each year.
Output from the model. The model estimated the cost per QALY gained, cost per life year gained, the cost per avoided hip fracture, the cost per avoided vertebral fracture; the cost per avoided forearm fracture. Both costs and benefits were discounted by 5% per year in the base case.
Validation. The model was validated according to “Principles of good practice for decision analytic modelling in health-care evaluation” published by Weinstein et al. (2003). The model predicted life expectancy within 2–10% range of the empirical values (Christensen et al. 2003).
Sensitivity analysis. We undertook one-way sensitivity analysis of all relevant the following model parameters by moving the parameters up and down within reasonable boundaries. These boundaries were, to the extent data were available, based on a systematic review of the literature (Christensen et al. 2003).
Three years of treatment with alendronate in 10,000 71-year-old Danish women compared to no treatment would, according to our model estimates, reduce the discounted lifetime number of hip, vertebral, and forearm fractures by 266, 190, and 383, respectively, according to the baseline parameter values (table 4). The discounted number of life years gained would be 73 and the discounted number of quality adjusted life years would be 219 (table 4). The discounted net programme cost would be DKK 2,728 per patient or a undiscounted value of minus DKK180 (table 5). The estimated discounted cost per QALY gained was DKK 125,000, and the discounted cost per life year gained was DKK 374,000 (table 6).
Table 4. Health outcomes for a 71-year-old Danish woman with twice the risk of fracture compared to the background population when using either alendronate for three years or no treatment. Women are followed until 100 years of age or death. Discounted at 5% per year.
Incremental benefit alendronate versus no treatment
Table 5. Osteoporosis related costs (US$ 1=7.7 DKK(Danish kroner) for a 71-year-old Danish woman with twice the risk of fracture compared to the background population when using either alendronate for three years or no treatment. Women are followed until 100 years of age or deaths. Costs were discounted at 5% per year, while undiscounted numbers are indicated in brackets.
Incremental costs (Alendronate versus no treatment)
Cost of alendronate for 3 years
Cost of treating hip fractures
Cost of treating vertebral fractures
Cost of treating forearm fractures
Table 6. The incremental cost-effectiveness ratio per patient using alendronate for three years compared to no treatment in 71-year-old Danish women with twice the risk of fracture compared to the background population. Patients are followed until 100 years of age or deaths. Discounted at 5%.
*DKK=Danish Kroner: 1US$=6.7 DKK.
Cost per life year gained
Cost per QALY gained
Cost per hip fracture avoided
Cost per vertebral fracture avoided
Cost per forearm fracture avoided
According to the sensitivity analysis, treatment with alendronate became a cost-saving strategy if treatment were extended from three to five years, or if the risk of fracture in the intervention group was four-times compared to the background risk, or if the effect of alendronate persisted for three years after discontinuing treatment, or if the proportion having severe sequelae after a hip fracture were increased (table 7). If the starting age of the cohort were 65 years the cost per QALY would be DKK 555,242 and the strategy became cost saving (dominant) if the starting age were 77 years (table 7). Varying the yearly cost of aldendronate, the hip fracture costs and the utility score after a hip fracture had substantial impact on the cost-effectiveness ratio whereas excess mortality due to osteoporosis, proportion dying in the first year after the hip fracture, the proportion of deaths caused by the hip fracture, compliance, and the discount rate had modest effects on the cost-effectiveness ratios (table 7).
Table 7. Sensitivity analysis.
Range of the model parameter
Cost per quality adjusted life year gained (DKK)
No confidence interval used in primary source. The authors of this paper assumed this range.
According to 95% confidence interval as stated by Scott et al. (1998).
According to 95% confidence interval for effect of alendronat as used by Black et al. (1992).
We assumed a risk reduction of 0.75 for three years after alendronate was discontinued.
This strategy was cost saving if we assumed an effect of the intervention in year two.
According to the 95% confidence interval stated by Braizer et al. (2002).
This utility score was also used for the subsequent years.
Linear decrease from 100% to 50% over three years. DKK=Danish Kroner: 1 US$=6.5 DKK.
Relative risk of osteoporotic fracture in intervention group compared to background risk
If all parameters were set at their “best” value, the use of aldendronate would create considerable cost savings. If they were all set at their “worst”, the cost per QALY gained would be astronomical.
Some argue that a cost per QALY lower than US$ 30,000 (approximately DKK 230,000) provides a reasonable yardstick for good value for money (Kanis & Jonsson 2002). According to this threshold, our results indicate that treatment of 71-year-old Danish women with an increased risk of fracture with alendronate for three years is “good value for money“. Moreover, we found that increased duration of alendronate treatment or assuming that the effect of alendronate was maintained after discontinuation would make the strategy cost-saving.
Our model is comparable with the other most recent model for aldendronate published by Johnell et al. (2003), which also has been used in a Danish setting (Jönsson et al. 2003). There are differences however. First, our model allows for analysis of the extent to which osteoporosis interventions reverses the excess mortality attributed to hip fracture. Thus our model allows the intervention to reverse the excess mortality attributed to hip fracture from 100% to 0%, where most other models have set this reversal to 100%. Although this issue is controversial (Kanis & Jonsson 2002), we have not been able to find evidence that an intervention aiming at osteoporosis will fully reverse the excess mortality associated with hip fracture. Surprisingly, variation in this parameter did not have much impact on the results (table 7). Second, we used a 3 year intervention interval as in the FIT study (Black et al. 1996) compared to 5 years in study by Johnell et al. (2003), thus making our estimates more conservative. Third, both studies showed that the persistency of an effect of aldendronate after the therapy is discontinued (the “set-time”) is of major importance. We used a set-time of one year whereas Johnell et al. (2003) used five years, thus our estimates are more conservative. The existence of a set-time is based on data on biochemical bone markers and bone mineral density (Stock et al. 1997; Jonsson et al. 1999; Tonino et al. 2000) since no studies with fracture as endpoint have been published. The risk reduction of 0.75 in the year after discontinuation of aldendronate was arbitrarily chosen. Fourth, the results indicate that the impact of the proportion ending in different post-hip fracture states is of importance. The range of this parameter from 5% to 15% represents pragmatic considerations (i.e.±5% of the base case parameter). We also simulated the effect of varying different cost components (cost of hip fracture, cost of intervention), which surprisingly had fairly high impact on the cost-effectiveness ratio. We were not able to find any of these results in the study by Johnell et al. 2003), thus uncertainty in modelling is dealt with in more details in our study compared to the other. Fifth, we have no conflict of interest while the study by Johnell et al. 2003) received financial support from aldendronate manufacturer. Our study was funded by the University of Southern Denmark. Editorials in both Brit. Med. J. and JAMA have stated that conflict of interest is important: “… Begin to build a solid case that conflict of interest has an impact on the conclusion reached by papers in medical journals” (Krimsky 1999). Thus it is reassuring when two studies, one funded by a pharmaceutical company, and one independent (ours), come to the same conclusion.
Modelling compliance is complex as non-compliance has consequences for both costs and effect, but the consequences can not be inferred directly from the clinical trials. We used simplified assumptions in that we assumed that the effect and cost of aldendronate declined linearly over 3 years and that the persistency after 3 years was 50% (table 7). Preliminary population-based Danish data indicates an adherence to bisphosphonates of approximately 50% three years after starting bisphosphonates (Larsen et al. 2003).
We simulated alendronate intervention in 71-year-old women partly because this was the age of subjects in the randomised controlled clinical trials of aldendronate (Black et al. 1996) and partly because a Danish population-based study found that the average age at start of treatment was 72 years (Larsen et al. 2003).
Our model only allows for pharmacological interventions that specifically aim at reducing the fracture risk (e.g. bisphosphontates, calcium and vitamin D, parathyroid hormone, strontium ranelate), while hormone replacement therapy and selective receptor oestrogen receptor modulators (SERMs) with extra-skeletal effects (Ettinger et al. 1999; Grady 2003) cannot be evaluated in this model.
The effect of alendronate was modelled according to a trial in which the untreated risk of a hip fracture over three years was 2.2% (Black et al. 1996), while it was 4.7% in our model population. The incidence was that high because the incidence of fractures in Scandinavia is higher than in any part of the world. In fact the population risk of hip fracture is about the same as the risk of hip fracture in the FIT study that was supposed to capture a high risk group in the US. Thus, we studied a high risk population, and extrapolating the relative risk reduction of 50% from the FIT study to a Danish setting seems fair. In terms of a clinical application, a 71-old-woman with the risk of fracture twice the population average will have a z-score of −1.1 and a t-score of approximately −2.9 in femur (see methods). In the sensitivity analysis the costs and health consequences of an intervention group of women with a four-times increase in the relative risk of osteoporotic fracture were analysed. Following the same arithmetic principles as presented previously (Christensen et al. 2003) this corresponds to a z-sore of −2.2 and a t-score of −4.1, and therefore represent a very small proportion of women.
We acknowledge that tariffs may not be a good approximation of resource use, and that there may be transferability issues in relation to applying cost data from other countries. Consequently, we performed sensitivity analysis on some of the cost estimates applied in the model.
Our cost estimates for vertebral and forearm fractures were to some extent influenced by a Swedish study (Zethraeus et al. 2002b). This study is preliminary in the sense that data had not been published in an international, peer reviewed journal, and the study had a relatively small sample size (n=50), a relatively low participation rate (42%), and there was no information about use of health care resources prior to the fracture. However, this is the only study published so far presenting empirical cost data on vertebral and forearm fractures in Scandinavia.
We used the term broad health care perspective despite the fact that home help and nursing home costs have been included. However because we did not include the cost of informal care giving or indirect costs (such as production gain/loses) we think the appropriate term is broad health-care perspective.
Modelling has a number of inherent limitations since it is only used when data are incomplete. Thus, a number of assumptions and extrapolations are necessary (Tosteson et al. 2001b; Kanis & Jonsson 2002; Zethraeus et al. 2002a). Some of consequent uncertainties tend to lower the cost-effectiveness ratio. First, in line with common practice we did not include “unrelated” (not related to osteoporotic fractures) costs incurred as a consequence of life-extension. Including unrelated health care costs and consumption costs in added life-years would increase the cost of the intervention. Since the main outcome of this intervention is improvement in quality of life rather than life-extension (table 4) we consider that this choice will have relatively little impact on the cost-effectiveness ratio. In theory omitting these costs will tend to lower the cost-effectiveness ratio. Second, a health utility score for each health state is required to estimate QALY's. The weights we used for the fractures were based on the few empirical available studies published (Brazier et al. 2002) adjusted for age/sex-specific health utility averages for the general population. These weights were all based on data from EQ-5D, which may tend to produce lower weights than for instance 15D (Hawthorne et al. 2001). Thus, using utility scores based EQ-5D rather than those derived from for instance 15D could overestimate the benefit from treatment, thus tending to lower the cost-effectiveness ratio. Third, it was assumed that taking alendronate did not have a negative impact on HRQOL (because of side-effects, inconvenient dosage regime). The HRQOL impact of taking aldendronate is currently unknown even though there were infrequent side-effects and inconvenient dosage regime. The omission of this aspect from our calculation will tend to overestimate the benefits gained from treatment thus tending to decrease the cost-effectiveness ratio. Fourth, the total number of fractures that occurred each year is presumably correctly estimated in our model, but the fractures are distributed across too many patients in that our model does not incorporate the fact that having sustained one fracture increases the risk of a new fracture (Klotzbuecher et al. 2000). This simplification will tend to increase the gain in benefits from treatment because the benefit from avoiding the first fracture is presumably higher than the second in the same patient, thus tending to decrease the cost-effectiveness ratio. Fifth, we used the same health care costs for patients dying and patients surviving in the year after a hip fracture although Swedish data may indicate that the costs of those dying is slightly lower (Zethraeus & Gerdtham 1998). This simplification could overestimate the cost savings measured by using aldendronate, and thus tend to decrease the cost-effectiveness ratio.
Some of the limitations in our study may tend to increase the cost-effectiveness ratio. First, since the target population is over the age of 71, most will not be in the work force and production gains due to improved health are likely to be negligible. Reductions in need for informal care giving may however incur resource gains. This cost component was not included in the cost estimate due to lack of data. Omitting indirect costs in our calculations tends to increase the cost-effectiveness ratio as the resource gains associated with treatment is underestimated. Second, the model only simulated the consequences on hip, forearm, and spinal fractures. It has been shown that fractures of the rib, ankle, and distal femur are associated with considerable costs (Ray et al. 1997). Omission of these fractures from our model may lead to an underestimation of the clinical and economic value of treatment, thus tending to bias the cost-effectiveness ratio up. Third, empirical studies have shown an excess mortality after a vertebral fracture (Cooper et al. 1993; Ensrud et al. 2000). This was not incorporated in our model. The degree to which this excess mortality is reduced by an osteoporosis intervention is currently unknown. The omission of this aspect will bias the benefit estimates down and may in theory bias the cost-effectiveness ratio up.
In conclusion, despite the inherent problems in modelling, this study has shown that treatment with alendronate for three years in 71- year-old Danish women with high risk of fracture compares well with other well established therapies in terms of cost-effectiveness.
Financial support for the development of this model was received by the University of Southern Denmark.