Osteoporosis International

, Volume 24, Issue 3, pp 835–847

The health burden and costs of incident fractures attributable to osteoporosis from 2010 to 2050 in Germany—a demographic simulation model


    • Department for Medical Sociology and Health Economics, Hamburg Center for Health EconomicsUniversity Medical Center Hamburg-Eppendorf
  • A. Konnopka
    • Department for Medical Sociology and Health Economics, Hamburg Center for Health EconomicsUniversity Medical Center Hamburg-Eppendorf
  • P. Benzinger
    • Department of Clinical GerontologyRobert-Bosch-Hospital
  • K. Rapp
    • Department of Clinical GerontologyRobert-Bosch-Hospital
  • H.-H. König
    • Department for Medical Sociology and Health Economics, Hamburg Center for Health EconomicsUniversity Medical Center Hamburg-Eppendorf
Original Article

DOI: 10.1007/s00198-012-2020-z

Cite this article as:
Bleibler, F., Konnopka, A., Benzinger, P. et al. Osteoporos Int (2013) 24: 835. doi:10.1007/s00198-012-2020-z



To predict the burden of incident osteoporosis attributable fractures (OAF) in Germany, an economic simulation model was built. The burden of OAF will sharply increase until 2050. Future demand for hospital and long-term care can be expected to substantially rise and should be considered in future healthcare planning.


The aim of this study was to develop an innovative simulation model to predict the burden of incident OAF occurring in the German population, aged >50, in the time period of 2010 to 2050.


A Markov state transition model based on five fracture states was developed to estimate costs and loss of quality adjusted life years (QALYs). Demographic change was modelled using individual generation life tables. Direct (inpatient, outpatient, long-term care) and indirect fracture costs attributable to osteoporosis were estimated by comparing Markov cohorts with and without osteoporosis.


The number of OAF will rise from 115,248 in 2010 to 273,794 in 2050, cumulating to approximately 8.1 million fractures (78 % women, 22 % men) during the period between 2010 and 2050. Total undiscounted incident OAF costs will increase from around 1.0 billion Euros in 2010 to 6.1 billion Euros in 2050. Discounted (3 %) cumulated costs from 2010 to 2050 will amount to 88.5 billion Euros (168.5 undiscounted), with 76 % being direct and 24 % indirect costs. The discounted (undiscounted) cumulated loss of QALYs will amount to 2.5 (4.9) million.


We found that incident OAF costs will sharply increase until the year 2050. As a consequence, a growing demand for long-term care as well as hospital care can be expected and should be considered in future healthcare planning. To support decision makers in managing the future burden of OAF, our model allows to economically evaluate population- and risk group-based interventions for fracture prevention in Germany.


Cost-of-illnessDemographic changeGermanyHealthcare planningMarkov modelOsteoporosis attributable fractures


In the time period from 2009 to 2050, the proportion of those aged 50 years and older will increase from around 39 % to more than 50 % of the total German population, respectively [1]. This shift in age structure is expected to have a strong financial impact on Germany’s social security system, especially because of increasing numbers of persons suffering from age-related diseases. Osteoporosis is a typical age-related disease, since its prevalence increases strongly in persons over 50 years of age [2]. The disease is characterized “by low bone mass and microarchitectural deterioration of bone tissue, leading to enhanced bone fragility and consequent increase in fracture risk” [3]. Osteoporosis leads to fractures at different locations of the human body like the vertebra, hip, humerus, or pelvis [46]. Fractures, especially of the hip, cause high levels of disability [7] and loss of quality of life (QOL) [8]. Osteoporotic fractures have been shown to increase mortality [9].

As a consequence of the association of osteoporosis-related fractures with many negative health outcomes, affected individuals utilise more medical services and cause additional healthcare costs (excess costs) [10]. Although several cost analyses can be found in the international literature [10], only little information is available on the potentially preventable future costs of incident fractures attributable to osteoporosis in Germany. A prevalence-based top down cost-of-illness study for Germany estimated that in 2002, direct and indirect costs of around 3 billion Euros were caused by hip fractures attributable to osteoporosis and osteopenia [11]. Similarly, Häussler et al. [12], who conducted an empirical cost-of-illness study based on German administrative data, found that direct fracture costs of 3.3 billion Euros were attributable to osteoporosis in 2003.

In Germany, with an increasing number of older persons, total healthcare costs for osteoporosis attributable fractures (OAF) are expected to increase over the next decades. For other countries, simulation models have already investigated the impact of demographic change on the future direct healthcare costs of fractures attributable to osteoporosis [1315]. However, no comparable simulation models have been constructed for Germany.

Therefore, our aim was to develop an innovative Markov simulation model in order to predict the direct and indirect costs of osteoporosis-related incident fractures occurring in the German population, aged 50 years and older during the time period of 2010 to 2050, and to estimate the associated loss of quality adjusted life years (QALYs). The estimated costs will reveal the potential savings of a hypothetical elimination of osteoporosis as a risk factor for the modelled fracture types.



In order to estimate costs and loss of QALYs due to incident fractures attributable to osteoporosis, we used a Markov simulation model. The model tracks yearly fracture events and related direct and indirect costs as well as loss of QALYs. Osteoporosis attributable excess costs and loss of QALYs were calculated by taking the difference of two Markov simulation runs. In the first run, the cohort was assumed to be free of osteoporosis; in the second run, the cohort suffered from osteoporosis according to the applied osteoporosis prevalence rates. To model the whole German population aged 50 years and older during the time period of 2010 to 2050, we simulated all male and female birth cohorts born between 1910 and 2000 twice. Annual total costs and loss of QALYs for the whole German population were calculated by taking year-specific excess costs from each modelled cohort and summarising these over all cohorts for a specific year (2010, 2030, and 2050). Costs and loss of QALYs due to fractures before 2010 were not considered, because of the incident design of the study. This approach made it possible to estimate the future economic burden caused by incident osteoporosis attributable fractures from 2010 to 2050 which reveals the potential savings of a hypothetical elimination of osteoporosis as a risk factor for the modelled fracture types.

Model approach

Using Microsoft Excel 2003 (Microsoft Corporation, Redmond, WA, USA) with Visual Basic for Application, we enhanced an existing Markov cohort state-transition model [16], which has been applied in earlier published cost-of-illness studies [14, 15]. The original model consists of eight mutually exclusive health states [16]. The main fracture states are “hip”, “second hip”, “clinical vertebral”, and “wrist”. Post-fracture states for “hip”, “second hip”, and “clinical vertebral” fractures are included to track long-term fracture consequences in costs and QOL [16]. We extended the model by states for osteoporotic fractures of the proximal humerus and pelvis [6]. However, no post-fracture states were modelled for proximal humerus and pelvic fractures because no data was available. To take the increasing prevalence of osteoporosis with increasing age [2] into account, two additional health states called “Osteoporotic” and “Well”, were added. Persons in the health state “Osteoporotic” suffer from osteoporosis, which is defined as a bone mineral density (BMD) of 2.5 standard deviations or lower (T score ≤ −2.5) as the average BMD of healthy young adults (20–30 years) [3], whereas persons in the “Well” state are not having osteoporosis. All possible transitions between the states are shown in Fig. 1, a transit to the absorbing state death is possible from each health state in the model (not shown in Fig. 1). Detailed information about input values can be found in the Electronic Supplementary Material.
Fig. 1

Markov model structure

As can be derived from Fig. 1, our model allows only two hip fractures over the lifetime of a person [16]. Model cycles were set to 1 year. The selected model population represents the total German male and female population aged 51–100 years for the time frame 2010 to 2050, which is the population at highest risk for getting osteoporosis [2]. The cohort size for each birth cohort was obtained from official statistics of the German Federal Office of Statistics [1]. To model the demographic development in Germany, we applied individual generation life tables for each birth cohort [17]. In total, we simulated 180 birth cohorts (90 men and 90 female) and summarised all incident fracture events per year from 2010 to 2050 [14].

The population of the model started from the “Osteoporotic” or the “Well” state, depending on the age- and gender-specific prevalence rate of osteoporosis extracted from the third National Health and Nutrition Examination Survey (NHANES III) [18, 19]. To estimate the excess costs, number of fractures, and loss of QALYs attributable to osteoporosis, each birth cohort was simulated twice. In the first model run, we assumed that the persons in the cohort are getting osteoporosis according to the disease prevalence rates. Therefore, these have higher fracture rates compared to persons without osteoporosis. In the second run, we assumed that the birth cohort is free of osteoporosis and therefore has lower fracture rates compared to persons with osteoporosis. By keeping the number of persons the same in both runs, it became possible to calculate excess costs, number of fractures, and loss of QALYs due to osteoporosis by subtracting the results of the two model runs.

Transition probabilities

The transitions between the states “Well” and “Osteoporotic” were dependent on osteoporosis prevalence rates [18, 19], which were transformed to age- and gender-dependent yearly transition probabilities [20]. Transition probabilities between the osteoporotic state and fracture states as well as between fracture and post-fracture states were dependent on age- and gender-specific fracture incidence rates, the relative risk of obtaining a fracture after suffering from a prior fracture and the relative fracture risk based on the presence of osteoporosis. The way of calculating transition probabilities differed between the cohort getting osteoporosis (first run) and those that do not (second run).

For the cohort getting osteoporosis (T score ≤ −2.5), the age- and gender-specific relative fracture risks due to osteoporosis RRBMD (compared to the general population) were calculated using a method proposed by Kanis et al. [2],
$$ \mathop{\text{RR}}\nolimits_{\text{BMD}} = \frac{{\Phi \left( {\left( {x - \mu } \right)/\sigma + \log \left( {{\text{R}}{{\text{R}}_{\text{SD}}}} \right)} \right)}}{{\Phi \left( {\left( {x - \mu } \right)/\sigma } \right)}} $$
where x is the BMD value at femoral neck which corresponds to a T score ≤ −2.5 (male BMD, 0.585 g/cm2; female BMD, 0.577 g/cm2), Φ is the normal distribution function, σ is the standard deviation (SD), and μ the mean of BMD in the actual age and gender group, RRSD represents the increase in fracture risk for each 1 SD decrease in BMD at femoral neck, which differs for each fracture type. The age-dependent RRSD for hip fractures were taken from a study by Johnell et al. [4], non age-dependent RRSD for wrist and vertebral fractures available from a meta-analyses by Marshall et al. [5], and the RRSD for proximal humerus and pelvis fractures from a study by Stone et al. [6]. Since there were only limited data available on BMD for the German general population [21], we used US BMD reference values stratified by age and gender from NHANES III [22].

To obtain the 1-year transition probabilities, we multiplied the age-specific fracture incidence rates with the calculated fracture-dependent relative risks RRBMD. Incidence rates for the five fracture types were calculated based on the “deep stratified diagnosis data from female and male inpatients 2009”, which is published yearly by the German Federal Office of Statistics [23]. The database contains all hospital cases in 2009 listed by four digit ICD-10 codes, gender, and age. To calculate yearly incidence rates for persons over the age of 50, we identified all relevant fracture cases with the following ICD-10 codes: hip (S72.0–S72.2), clinical vertebral (S22.0; S22.1; S32.0), pelvis (S32.1–S32.8), wrist (S52.5–S52.6), and proximal humerus (S42.0). Based on these data, we calculated age- and gender-specific incidence rates by dividing the identified fracture cases through the total number of persons in the corresponding age and gender classes, living in Germany in 2009 [1]. To handle the problem of double counting (rehospitalisation) in the hospital discharge statistic, we applied correction factors to each fracture incidence rate [24, 25].

These incidence rates were based only on fractures treated in a hospital. Because a certain proportion of fractures are treated only in an outpatient setting, these incidence rates underestimated the true incidence. To estimate the true incidence rate, we divided the age- and gender-specific hospital-based incidences rate through a constant fracture-specific hospitalisation rate. The following hospitalisation rates were assumed: 100 % for hip fractures [13], 47 % of clinical vertebral fractures [26, 27], 58 % for distal radius fractures [28], 87 % for proximal humerus fractures [28], and 75 % for pelvis fractures [29].

To determine the transition probabilities for the cohort not getting osteoporosis (healthy cohort (HC); second run) we used following formula, which estimates the fracture incidence rate of the population not having osteoporosis:
$$ \mathop{\text{TP}}\nolimits_{\text{HC}} = \frac{{\left( {{C_{\text{size}}} \times I} \right) - \left( {{C_{\text{size}}} \times p \times I \times {\text{R}}{{\text{R}}_{\text{BMD}}}} \right)}}{{{C_{\text{size}}} - \left( {{C_{\text{size}}} \times p} \right)}} = \frac{{I \times \left( {1 - p \times {\text{R}}{{\text{R}}_{\text{BMD}}}} \right)}}{{\left( {1 - p} \right)}} $$
where Csize represents age- and gender-dependent cohort size, I the age- and gender-specific fracture incidence rate, p the age- and gender-specific osteoporosis prevalence rate, and RRBMD the age- and gender-specific relative fracture risk due to osteoporosis.

Having sustained a fracture increases the risk of sustaining further fractures [30, 31]. To include this in our model, we used data from Van Staa et al. [30] which determined age- and gender-specific relative risks to get a radius fracture after hip, vertebral fracture after hip, any fracture after hip, hip after vertebral, radius after vertebral, and any fracture after vertebral for three age classes (65–74, 75–84, 85+). The authors did not examine the relative risk to get a hip fracture after hip and vertebral fracture after vertebral fracture. Hence, we took the age-adjusted relative risk for these combinations from a meta-analysis by Klotzbucher et al. [31]. We applied the relative risk for “any fractures” as proxies for proximal humerus and pelvis fractures, because no relative risks for these fracture types were available. Since BMD and the occurrence of a previous fracture are not independent, we reduced each relative risk by 10 % to adjust for BMD [32].


In our model, we applied two types of mortality, age- and gender-specific all-cause mortality and fracture-specific excess mortality. German generation life tables of the birth cohorts 1910–2000 were used as source for all-cause mortality [17]. To include 1-year excess mortality due to osteoporotic fractures, we multiplied age- and gender-specific all-cause mortalities with relative risks of mortality from a large population-based Canadian study, which examined the excess mortality after hip, clinical vertebral, proximal humerus, wrist and other fractures stratified by five age classes and gender. The study calculated relative mortality risks by different types of fractures adjusted for co-morbidities and location of residence [9]. We assumed no excess mortality for wrist fractures [33].

Quality of life

To account for reductions in health-related quality of life (HRQOL) due to fractures [8], we used QALYs as the health outcome in the model. QALYs combine changes in mortality with changes in HRQOL in a single measure. German-specific QOL weights for the general population were taken from a study by König et al. [34]. The authors used the algorithm from Dolan et al. [35] to calculate EQ-5D index values between 0 for death and 1 for perfect health, based on a representative sample of the German population stratified by age and gender [34]. The resulting EQ-5D index values were used as QOL weights for persons in nonfracture health states. QOL weights for persons suffering from osteoporotic fractures were gathered from a systematic review by Hiligsmann et al. [8], who calculated multipliers for the proportionate effect of hip, clinical vertebral, distal forearm, and other fractures on health utility weights in the year of the fracture (first year) and subsequent years. Health utility weights for proximal humerus and pelvis fractures in the year of the fracture were taken from a publication by Kanis et al. [36]. To obtain health utility weights for persons staying in a specific fracture state, we multiplied the age- and gender-specific QOL weights for the general population with the fracture-specific multipliers. Long-term reduction in QOL in persons suffering from hip or clinical vertebral fractures was reflected by multiplying utility multipliers for subsequent years after fracture with the general QOL weights. No long-term effects in QOL were assumed for wrist, pelvis, and proximal humerus fractures [8].

Direct costs

In our study, we adopted a societal cost perspective, where direct and indirect costs for OAF were determined for the German setting. Direct costs were divided into four categories: inpatient care, outpatient care, long-term care, and rehabilitation care. Inpatient costs for the five fracture types were determined on the basis of the German G-DRG Browser V2010 [37]. The Browser provides information on nearly all German hospitals, which reported total diagnosis-related group (DRG) cases in 2009, with the corresponding ICD10 diagnoses, procedures, cost weights, average length of stay, and age-gender structure. We assumed that the corresponding DRG charges are on average reflective of the real costs of healthcare provision of each inpatient case.

To determine inpatient cost per fracture type, firstly we identified all DRGs corresponding to the group of ICD10 diagnoses, which defines a fracture type (e.g., S 72.0–2). Secondly, a weighted mean cost weight was calculated for each fracture type based on the distribution of DRG cases within the group of ICD10-diagnoses.

The resulting cost weights per fracture type were then multiplied with the state adjusted German DRG base rate of 2,887€ [38] to obtain age- and gender-adjusted fracture-specific inpatient costs per case. Hospital capital costs are not considered in the German DRG system. To take these costs into account, we applied capital costs per fracture-related inpatient day [37] from Krauth et al. [39], adjusted to 2009 Euros.

To estimate direct outpatient costs per fracture type, we used an expert survey conducted in Ulm (Germany) in 2002, investigating the resource use of hip, clinical vertebral and wrist fracture in an outpatient setting (medication, outpatient visits, etc.) [40]. We monetarily valued resources with German unit costs adjusted to 2009 Euros [39, 41, 42]. Since no data on resource use for pelvis and proximal humerus fractures were available, we assumed the same resource use for wrist as for proximal humerus fractures and for pelvis as for hip fractures.

In line with another simulation study [13], we assumed that only hip fractures lead to long term-care costs. The age- and gender-specific direct long-term care costs (LTCcij) were calculated based on the following formula:
$$ {\text{LTC}}{{\text{c}}_{\text{ij}}} = N{\text{hi}}{{\text{p}}_{\text{ij}}} \times \left( {1 - {\text{LT}}{\text{crat}}{{\text{e}}_{\text{ij}}}} \right) \times \pi {\text{LT}}{{\text{C}}_i} \times {\text{LTCc}}; $$
where i represents the age between 50 and 99, j the gender, LTCrateij the age- and gender-dependent all-cause German long-term (nursing home) care rates [43], which are used to correct for the fact that persons will be admitted to nursing homes for other reasons than hip fractures, πLTC the age-dependent probability entering a nursing home after hip fracture [7], Nhipij the number of persons in the model with at least one hip fracture, and LTCc the yearly costs of nursing home weighted by care level (I–IV) in prices of 2009 [43]. In the initial year of a hip fracture, we used a half cycle correction to calculate the long-term care costs arising during the first year after fracture. Once a person transits to long-term care due to hip fracture, the person will remain there for the rest of their life, i.e., the model tracks corresponding long-term care costs until the event death occurs.

Direct costs of rehabilitation were calculated by multiplying the fracture-specific average length of stay in a rehabilitation institution [44], the fracture-specific probability of transiting from hospital to a rehabilitation institution after having a fracture [44], and the per diem costs of rehabilitation in 2009 [45].

Indirect costs

Indirect costs were considered as productivity loss due to the inability to perform paid or unpaid work during the time of disability, rehabilitation and long-term care, or through premature death. Fracture-specific disability and rehabilitation days were gathered from official statistics of a large German sickness fund [44]. We monetarily valued time away from paid work with the average gross income of men and women in Germany in 2009 [46], increased by the employer’s share of social insurance contributions and a yearly income growth rate of 2 % (human capital approach). To take only employed individuals into account, we multiplied the described gross income with the German age- and gender-specific employment rate [47].

Time away from unpaid work was valued with the average (net) hourly rate of a housekeeper (generalist method), where the yearly amount of unpaid work for men and women were taken from a German time budget study [48]. For persons in long-term care, we assumed that only 16 % (19 %) of the amount of unpaid work from women (men) are real opportunistic cost [48], because 84 % (81 %) of the valued unpaid work, for example cleaning, preparation of meals, etc., were already included in the long-term care fees. Productivity loss due to premature death was calculated by taking the difference between the gained productivity of simulation run one (cohort with osteoporosis) and two (cohort without osteoporosis) for each birth cohort. Due to the Markov modelling approach, productivity loss due to premature death was calculated for the first time in 2011. Direct and indirect costs were discounted by 3 % in the base case analysis.

Sensitivity analysis

Univariate and probabilistic sensitivity analysis (Monte Carlo simulation) were applied to test the robustness of our results. In total, eight different univariate sensitivity analyses were conducted. In the first and second analyses, we varied the osteoporosis prevalence rate and all fracture incidence rates by ±20 %. In the third analysis, we decreased all-cause mortality for all cohorts by 10 and 40 % to reflect eventual effects of technological progress on life expectancy. In the fourth analysis, we assumed no excess mortality after fractures and that only 30 % of the raw mortalities after fractures are attributable to the fractures itself. In the fifth and sixth analysis, we simulated the model using the upper and lower confidence limit values of the relative fracture risk for each 1 SD decrease in the BMD at femoral neck and of the relative risk of getting a fracture after suffering from a prior fracture. In the seventh analysis, we changed the discount rate from 3 to 5 %, which reflects higher time preferences. Finally, we applied the upper and lower confidence limits for QOL weights in the model.

For the probabilistic sensitivity analysis (Monte Carlo simulation), all model parameters for which statistical measures like standard errors or confidence intervals were available were simultaneously varied (see Electronic Supplementary Material). We used common distribution assumptions for QOL weights (beta distribution), costs (univariate1 and gamma distribution) and relative risks (log normal distribution) [49].



In Germany, approximately 3.5 million women (19.9 %) and 1 million (6.4 %) men aged 50 years or older suffered from osteoporosis in 2010. Until 2050, this number of women and men will increase to 5 million (25.5 %) and 1.9 million (10.5 %), respectively, reflecting an increase in total numbers of 40 % for women and 81 % for men from 2010 to 2050. Table 1 summarises the total number of incident (hip, clinical vertebral, wrist, proximal humerus, and pelvis) fractures attributable to osteoporosis in 2010, 2030, 2050, and aggregated from 2010 to 2050. The number of OAF per year will rise from 115,248 in 2010 to 273,794 in 2050, which is equivalent to an increase of 238 % (+335 % in men and +218 % in women). The largest increase can be expected in clinical vertebral fractures of men, which are projected to increase from 4,514 fractures in 2010 by 442 % to 19,496 fractures in 2050. In general, OAF in men will increase more sharply. In 2010, men account for around 17 % of total OAF, while this number changes to around 24 % in 2050.
Table 1

Total number of incident osteoporosis attributable fractures (OAF) in Germany by fracture type for 2010, 2030, 2050, and aggregated from 2010 to 2050

Fracture types

Number of OAFs 2010

Number of OAFs 2030

Number of OAFs 2050

Number of OAFs 2010–2050

Men mean (UI)

Women mean (UI)

Total mean (UI)

Men mean (UI)

Women Mean (UI)

Total mean (UI)

Men mean (UI)

Women mean (UI)

Total mean (UI)

Men mean (UI)

Women mean (UI)

Total mean (UI)


11,147 (10,431–11,882)

43,979 (41,716–46,035)

55,126 (52,187–57,885)

22,315 (20,462–24,367)

70,147 (63,277–77,272)

92,462 (84,409–100,605)

32,262 (29,405–35,302)

96,989 (87,042–106,960)

129,251 (117,434–141,104)

909,003 (836,017–988,956)

2,891,893 (2,626,158–3,170,860)

3,800,896 (3,484,268–4,126,362)

Clinical. vertebral

4,514 (839–9,323)

17,605 (4,069–29,403)

22,120 (4,908–38,726)

13,922 (2,781–31,095)

29,374 (6,521–48,637)

43,296 (9,494–78,889)

19,946 (3,809–44,520)

39,277 (8,877–64,261)

59,223 (12,791–107,070)

546,784 (107,507–1,218,355)

1,197,186 (270,137–1,980,307)

1,743,970 (386,161–3,147,410)

Proximal humerus

1,898 (1,381–2,467)

13,130 (10,194–16,165)

15,029 (11,570–18,627)

4,515 (3,331–5,837)

21,146 (16,633–25,799)

25,662 (19,988–31,603)

6,353 (4,703–8186)

27,599 (21,824–33,578)

33,951 (26,569–41,643)

178,079 (131,487–230,554)

859,078 (676,390–1,047,770)

1,037,157 (808,550–1,277,620)


1,334 (757–1,969)

9,134 (5,807–12,101)

10,468 (6,563–14,063)

3,547 (2,141–5,144)

16,580 (10,895–21,493)

20,127 (13,058–26,621)

5,392 (3,277–7,774)

23,890 (15,834–30,787)

29,283 (19,131–38,451)

141,180 (85,302–204,053)

673,458 (442,191–874,513)

814,638 (528,639–1,079,907)


771 (551–1,081)

11,735 (8,609–16,170)

12,506 (9,163–17,251)

1,568 (1,163–2,348)

16,577 (12,244–22,671)

18,161 (13,429–24,897)

1,876 (1,373–2,870)

20,209 (14,964–27,592)

22,085 (16,397–30,320)

61,701 (45,348–90,452)

676,368 (499,576–923,109)

738,069 (545,836–1,010,424)


19,665 (15,832–24,576)

95,583 (80,846–108,829)

115,248 (96,710–133,263)

45,884 (33,680–63,415)

153,824 (127,124–178,150)

199,708 (161,736–241,414)

65,828 (48,517–90,323)

207,965 (171,891–239,887)

273,794 (221,506–328,939)

1,836,747 (1,360,723–2,515,754)

6,297,983 (5,224,863–7,279,258)

8,134,730 (6,610,773–9,793,024)

Mean = deterministic value, UI=95 % uncertainty intervals (Monte Carlo simulation 4,000 trials)

In total, approximately 8.1 million fractures are projected to be attributable to osteoporosis between 2010 and 2050, where 6.3 million fractures (78 %) will occur in women and 1.8 million fractures (22 %) in men. Hip fractures in women provide the largest contribution to total number of fractures from 2010 to 2050, with expected 2.9 million fracture events (36 %).

Costs and loss of quality adjusted life years

Table 2 shows total direct and indirect costs of incident (hip, clinical vertebral, wrist, proximal humerus, and pelvis) fractures attributable to osteoporosis in 2010, 2030, and 2050 as well as aggregated costs from 2010 to 2050, undiscounted and discounted to the base year 2010 by 3 %. The total direct costs of around 898 million Euros in 2010 are projected to increase to 4.7 billion Euros in 2050. This increase is mainly driven by long-term care costs. Figure 2 displays the cumulative direct costs for long-term care due to incident osteoporosis-related hip fractures in between 2010 and 2050. Generally, long-term care costs can be expected to sharply increase until 2050. This trend will be more pronounced for women, which can be explained due to higher osteoporosis prevalence rates, fracture incidence rates, and higher life expectancy in women.
Table 2

Direct and indirect costs of incident osteoporosis attributable fractures (OAF) in million of Euro for 2010, 2030, 2050 and aggregated from 2010 to 2050 (undiscounted and discounted)

Cost categories

Total costs of OAF 2010

Total costs of OAF 2030

Total costs of OAF 2050

Total costs of OAF 2010–2050 undiscounted

Total costs of OAF 2010–2050 discounted with 3 % to 2010


Men mean (UI)

Women mean (UI)

Total mean (UI)

Men Mean (UI)

Women mean (UI)

Total mean (UI)

Men mean (UI)

Women mean (UI)

Total mean (UI)

Men mean (UI)

Women Mean (UI)

Total Mean (UI)

Men mean (UI)

Women mean (UI)

Total mean (UI)

Direct Costs
















Hospital care

109.5 (98.4–122.7)

484.6 (443.3–521.3)

594.1 (542.2–644.2)

236.9 (202.4–283.5)

779.5 (687.7–863.7)

1,016.4 (894.8–1,138.0)

341.4 (292.3–406.5)

1,065.5 (939.5–1,178.8)

1,406.9 (1,236.3–1,575.9)

9,561.0 (8,208.6–11,369.0)

32,000.2 (28,350.9–35,333.0)

41,561.3 (36,762.7–46,411.2)

5,091.8 (4,384.2–6,031.9)

17,474.0 (15,518.0–19,254.1)

22,565.8 (20,041.1–25,144.4)

Rehabilitation care

8.7 (8.6–11.5)

37.4 (37.3–48.3)

46.1 (59.8–46.3)

18.3 (17.8–24.9)

60.3 (59.0–79.9)

78.7 (77.1–104.4)

26.5 (25.7–35.9)

83.2 (81.1–110.3)

109.7 (107.4–145.2)

742.0 (722.8–1,006.6)

2,480.5 (2,433.6–3,274.0)

3,222.6 (3,166.7–4,265.6)

395.6 (386.1–536.0)

1,353.6 (1,331.6–1.783.2)

1,749.2 (1,721.0–2,311.1)

Ambulatory care

20.7 (17.5–32.2)

97.3 (87.6–134.6)

118.1 (105.0–166.8)

50.0 (37.6–86.8)

157.4 (138.3–220.0)

207.5 (177.0–304.9)

71.8 (54.3–123.7)

213.0 (187.3–296.4)

284.8 (243.0–416.9)

1,998.8 (1,519.8–3,423.6)

6,443.6 (5,678.2–8,987.4)

8,442.4 (7,238.7–12,344.2)

1,058.7 (808.9–1,802,7)

3,518.1 (3,107.7–4,906.0)

4,576.8 (3,938.2–6,680.1)

Long-term care

25.9 (21.0–31.6)

113.7 (91.5–138.6)

139.6 (112.4–170.2)

432.8 (345.4–533.3)

1,519.1 (1,226.8–1,832.6)

1,951.9 (1,580.3–2,356.3)

692.9 (555.4–849.7)

2,265.2 (1,833.8–2,731.9)

2,958.1 (2,399.2–3,577.2)

16,643.6 (13,299.6–20,392.3)

58,483.8 (47,321.7–70,611.8)

75,127.4 (60,924.3–90,846.2)

8,367.8 (6,698.5–10,238.7)

30,178.5 (24,436.2–36.440,1)

38,546.3 (32,227.7–46,560.9)

Total direct costs

164.8 (150.0–192.2)

733.1 (681.6–819.6)

897.9 (831.9–1,010.8)

738.1 (636.8–885.7)

2,516.3 (2,194.9–2,900.8)

3,254.4 (2,842.0–3,772.5)

1,132.7 (976.6–1,354.1)

3,626.8 (3,154.1–4,187.1)

4,759.5 (4,140.1–5,523.9)

28,945.5 (25,091.0–34,613.2)

99,408.2 (87,163.1–114,205.5)

128,353.7 (112,526.3–148,499.1)

14,913.8 (12,957.3–17,789.7)

52,524.2 (46,173.8–60,258.1)

67,438.1 (59,289.7–77,864.7)

















Indirect costs (paid and unpaid work) lost productivity due to…

Disability days

21.4 (18.4–25.4)

129.3 (113.8–143.6)

150.7 (132.3–168.8)

39.7 (31.8–51.8)

180.3 (153.9–204.4)

220.0 (186.3–255.9)

49.7 (39.8–63.8)

233.0 (198.9–263.5)

282.7 (239.1–327.0)

1,577.7 (1,270.9–2,032.8)

7,431.8 (6,363.4–8,402.5)

9,009.5 (7,663.6–10,426.5)

866.2 (701.0–1,112.2)

4,121.1 (3,539.3–4,652.0)

4,987.3 (4,251.9–5,762.5)

Rehabilitation days

2.0 (1.9–2.7)

9.9 (9.9–12.8)

11.9 (11.8–15.5)

3.5 (3.3–4.8)

13.1 (12.9–17.2)

16.6 (16.2–22.0)

4.4 (4.2–6.1)

17.2 (16.9–22.5)

21.6 (21.1–28.6)

139.6 (133.8–194.4)

545.3 (537.4–713.9)

684.8 (671.3–905.5)

76.9 (73.8–107.0)

303.1 (299.3–396.4)

380.0 (373.4–502.0)

Long-term care

2.0 (1.5–2.8)

7.4 (6.0–8.9)

9.4 (7.5 –11.6)

33.2 (23.2–47.2)

98.2 (78.9–119.3)

131.3 (103.3–164.5)

42.2 (31.8–55.5)

140.8 (114.1–169.8)

183.0 (147.2–222.9)

1,201.3 (854,4–1,680.4)

3,744.0 (3,021.5–4,543.1)

4,945.3 (3,924.2–6,133.2)

626.5 (441.7–881.8)

1,943.4 (1,569.9–2,357.2)

2,569.9 (2,036.8–3,193.0)

Premature death

0 (0.0–0.0)

0 (0.0–0.0)

0 (0.0–0.0)

201.2 (146.2–302.2)

526.7 (395.5–733.5)

728.0 (565.0–993.2)

258.9 (186.3–391.5)

619.3 (459.8–865.0)

878.2 (672.6–1,202.1)

7,034.5 (5,118.7–10,557.0)

18,414.5 (13,772.5–25,633.8)

25,448.9 (19,704.5–34,735.8)

3,584.3 (2,614.1–5,350.0)

9,586.7 (7,177.9–13,317.7)

13,171.0 (10,234.2–17,930.9)

Total indirect costs

25.4 (22.4–30.1)

146.5 (131.8–163.3)

171.9 (154.5–193.0)

277.7 (214.4–389.8)

818.2 (671.3–1,042.4)

1,095.9 (901.9–1,390.6)

355.1 (273.1–501.6)

1,010.4 (826.7–1,272.9)

1,365.4 (1,122.2–1,726.3)

9,953.0 (7,735.2–13,892.6)

30,135.6 (24,804.5–38,067.0)

40,088.6 (33,164.5–50,521.0)

5,153.8 (4,026.3–7,140.2)

15,954.3 (13,177.2–20,076.7)

21,108.2 (17,515.8–26,486.3)

















Direct and Indirect costs

190.2 (172.7–222.0)

879.6 (814.6–980.9)

1,069.8 (987.3–1,201.8)

1,015.8 (873.5–1.238.8)

3,334.6 (2,944.0–3,832.0)

4,350.4 (3,823.4–5,059.2)

1,487.8 (1,280.6–1,793.3)

4,637.2 (4,080.0–5,327.2)

6,125.0 (5.364.5–7,117.4)

38,898.5 (33,594.5–47,151.4)

129,543.8 (114,73.9–148,524.4)

168,442.3 (148,362.4–195,120.2)

20,067.3 (17,362.8–24,270.9)

68,478,6 (60,807.9–78,390.0)

88,546.2 (78,158.8–102,400.2)

Mean = deterministic value, UI=95 % uncertainty intervals (Monte Carlo simulation 4,000 trials)

Fig. 2

Direct long-term care costs due to incident osteoporosis attributable hip fractures by gender within the years 2010 and 2050

Indirect costs are expected to increase from 172 million Euros in 2010 to 1.4 billions Euros in 2050. The finding was mainly driven by the loss of productivity due to the inability to engage in unpaid work because of premature death.

In total, OAF causes approximately 88.5 billion Euros aggregated direct (76 %) and indirect (24 %) costs from 2010 to 2050. Women account for about 77 % of these costs. The main cost drivers for direct costs are long-term care costs after hip fractures with 38.5 billion Euros (43.5 %) and hospital care costs with 22.6 billion Euros (25.5 %). Productivity losses of 13.2 billion Euros (14.8 %) due to osteoporosis-related premature deaths are the largest share in indirect costs.

Table 3 shows the percentage distribution of aggregated and discounted total costs from 2010 to 2050 over the five fracture types, excluding long-term care costs due to hip fractures. Hip fractures account for 66.1 % of all direct and for 55.8 % of all indirect costs. The fracture type with the second largest impact on overall costs is clinical vertebral fracture with a percentage share of 15.4 %. The lowest impact on overall costs is wrist fracture with a share of 3.6 %.
Table 3

Percentage share of different fracture types on incident osteoporosis attributable fracture costs (aggregated total costs (except long term care costs) from 2010 to 2050 discounted by 3 % (see Table 2))

Cost categories/fracture type


Clinical vertebral

Proximal humerus



Men (%)

Women (%)

Total (%)

Men (%)

Women (%)

Total (%)

Men (%)

Women (%)

Total (%)

Men (%)

Women (%)

Total (%)

Men (%)

Women (%)

Total (%)

Direct costs

Hospital care
















Rehabilitation care
















Ambulatory care
















Tot. direct costs (exc. LTC)
































Indirect costs (paid and unpaid work) lost productivity due to…

Disability days
















Rehabilitation days
















Premature death
















Tot. ind. costs (exc. LTC)
















Tot. in- and direct costs
















From a health outcome perspective, there is a total loss of around 2.5 million QALYs within 2010–2050 (discounted to 2010 by 3 %) due to OAF. Men account for around 0.78 million QALYs lost, whereas women account for 1.76 million. On average, any OAF leads to a loss of 0.83 QALYs.

Sensitivity analysis

The univariate sensitivity analyses are presented in Table 4. The base case is defined as the cumulated discounted costs and lost QALYs from 2010 to 2050. The largest variation in univariate sensitivity was found by increasing age- and gender-specific prevalence rates by 20 %, which led to 45.4 % higher total costs and a 40.1 % increase in loss of QALYs. Decreasing all-cause mortality by 40 %, which can be considered as an extreme scenario, has the second largest effect on total costs with a percentage reduction of 36.4 %. Assuming no excess mortality after fractures led to a reduction of total costs by 6.9 %, whereas the assumption that only 30 % of excess mortality is caused by the fracture itself, resulted in a decrease of 4.7 % in total costs. In comparison to total costs, we found a large impact of excess mortality on loss of QALYs. The assumption of no excess mortality led to a reduction of 38.3 % in loss of QALYs; assuming that 30 % of mortality is fracture-related decreased the loss of QALYs by 26.1 %.
Table 4

Different univariate sensitivity analyses on predicted costs (base case scenario “Aggregated total cost from 2010 to 2050 discounted by 3 %” (see Table 2))



Hospital care (%BC)

Reha. care (%BC)

Ambulatory care (%BC)

Long term care (%BC)

Total cost direct (%BC)

Disability days (%BC)

Rehabilitation days (%BC)

Long term care (%BC)

Premature death (%BC)

Total indirect costs (%BC)

TOTAL costs (%BC)

QALYS loss (%BC)

Base case (BC)*

Total costs 2010–2050 (disc)













Excess mortality

No excess mortality













30 % Excess mortality













All cause mortality

Decreased by 10 %













Decreased by 40 %













Osteoporosis prevalence

Increased by 20 %













Decreased by 20 %













Incidence rate of all fracture types

Increased by 20 %













Decreased by 20 %













Relative risk of fracture for 1-SD decrease in BMD

Upper confidence limits













Lower confidence limits













Relative risk of subsequent fractures

Upper confidence limits













Lower confidence limits













Discount rate

5 %













Quality of life weights

Upper confidence limits













Lower confidence limits













A small variation in the range of ±2.5 % in loss of total QALYs were observed after applying the lower and upper confidence limits of QOL weights. Changing the discount rate from 3 to 5 % led to a decrease of around 31 % in total costs and loss of QALYs. The results of the probabilistic sensitivity analyses of fracture events and costs are reported as 95 % uncertainty intervals in Tables 1 and 2, respectively.


The aim of this study was to develop a Markov disease model to estimate the health and economic burden of incident factures attributable to osteoporosis in the German population aged 50 years and older from 2010 to 2050. Our main findings are that total yearly costs of incident OAF will rise from about 1 billion Euros in 2010 to about 6.1 billion by the year 2050 (undiscounted). The aggregated total costs from 2010 to 2050 are projected to be around 88.5 billion Euros (discounted by 3 % to 2010), which corresponds to the potential savings of a hypothetical elimination of osteoporosis as a risk factor for the modelled fracture types. Direct costs of incident osteoporosis attributable fractures in 2010 equal to around 0.5 % of total German statutory health insurance expenditures [50].

To our knowledge, this is the first simulation-based cost study of incident osteoporotic fractures for a German setting. Similar simulation-based cost studies have been published for Switzerland [13] and the USA [14]. However, findings from these modelling studies cannot be directly compared to our results, mainly because of countries’ specific epidemiology and future demographics, as well as varying cost structures. Unlike these previous studies, we neglected to use expert opinions on osteoporosis attribution probabilities to estimate the share of fracture costs attributable to osteoporosis. Instead, we directly implemented the risk factor osteoporosis as a health state in the model. The advantage of this approach is that osteoporosis attribution probabilities were directly estimated through the model itself, on the basis of empirical data, instead of applying model exogenous expert opinions on osteoporosis attribution probabilities.

Using a direct approximation based on available data, we estimated around 50 % of hip fractures in women over 50 are attributable to the risk factor osteoporosis. Empirical studies in this field found similar osteoporosis attributable fractions between 28 % [6] and 51 % [51] for hip fractures in women over 65 and 70, respectively. In contrast, an expert panel concluded that 90 % of hip fractures in women between 64 and 85 years are attributable to osteoporosis [52]. The use of empirical data, instead of data based on expert opinions, marks a clear improvement to previously applied modelling approaches in osteoporosis-related fracture costs analysis. These marked differences make clear that the use of expert-based osteoporosis attribution probabilities in simulation-based cost studies may lead to an overestimation of costs [53].

Two previous German cost studies [11, 12] with related research questions reported substantially higher attributable costs. However, for a variety of reasons, the findings from these studies are not comparable to ours. Most importantly, both studies applied no simulation model and did not follow an incident (fractures)-based approach. Similar to the present study, Konnopka et al. [11] estimated the health burden and costs of osteoporotic fractures in Germany for the year 2002, 2020, and 2050. In contrast to our study, these authors used a top-down approach and included only hip fractures. In addition, a wider disease definition was applied, i.e., they considered osteopenia and osteoporosis (defined as T < −1).

Häussler et al. [12] empirically estimated the healthcare costs attributable to osteoporosis based on German administrative data in 2003 and calculated osteoporosis attributable fracture costs of 11 fracture types as a sub-analysis. The main differences to the present study are that more fracture types were included in the analysis and that expert opinion was used to determine osteoporosis attributable fracture costs.


Our study has a few limitations that should be considered when interpreting the results. Due to the model structure, we restricted the consideration of long-term consequences to costs and QALYs of hip fractures and QALYs of clinical vertebral fractures. Therefore, it was not possible to track long-term consequences in costs and QOL of other fracture types. The estimated long-term care costs may nonetheless have been overestimated for the following reason: It was assumed that residents of nursing homes will be institutionalised for the remainder of their life, while in reality there are situations where residents change to another residential status, e.g., when they get informal care from relatives after a short time in nursing homes. Fracture incidence rates in our model were based on German hospital discharge statistics and internationally published hospitalisation rates per fracture type, due to the lack of empirical German data on total (inpatient and outpatient) fracture incidence rates. The use of non-age-dependent hospitalisation rates may not hold in a real-world clinical setting because rates are likely to increase with age, which could have biased our total fracture incidence rates.

Especially for vertebral fractures, it should be noted that only a certain percentage comes to clinical attention. These nonclinical vertebral fractures are not included in the modelled fracture incidence rates. Yet, they can be assumed to decrease QOL and increase the risk of a further fracture [36]. Not including these fractures leads to a potential underestimation of the burden due to incident osteoporosis-attributable vertebral fractures in Germany.

Our model assumed age- and gender-specific fracture rates to remain unchanged during the period of observation. In the literature, this assumption is discussed controversially. A recent review reported the secular trends in incident hip and other fractures for different geographical regions and time periods. In the time period 1988–2008, the authors found inconsistent annual rates of change in age- and sex-adjusted hip fracture incidence for Europe. Finland, France, and Switzerland showed an annual decrease in incidence of around 1.3–2.4 %. In Germany [25], Hungary, and Austria, a stabilisation in incidence rates of hip fractures was reported, whereas Spain showed a yearly increasing hip fracture incidence rate [54]. Furthermore, it should be noted that a future increase of osteoporosis treatment may decrease age- and gender-specific fracture rates.

Since there was a lack of German outpatient cost data for fractures, it was necessary to use published expert opinions to estimate outpatient costs for fracture treatment. Generally, the use of expert opinion has the lowest level of evidence and should therefore only be used when no other data is available [55]. Furthermore, costs for ambulatory long-term care, transportation, and early retirement were not included into the model since no data were available. Finally, our cost predictions until 2050 are heavily dependent on the reliability of the estimated generation life tables of the German Federal Office of Statistics since changes in the expected demographic structure will have a large impact on our model results.

Policy implications and conclusion

A major finding of the present study is that inpatient and long-term care costs contribute by far the largest share (around 70 %) to overall incident OAF costs. Therefore, an increasing demand for services in these sectors should be expected in Germany, which must be taken into consideration in future healthcare planning. In light of the high projected future health burden due to incident OAF, a reduction of the prevalence of osteoporosis using preventive measures would be desirable. Research has shown that in Germany, a screen-and-treat strategy generally offered a favourable cost effectiveness for women of the general population aged 50–90 years [56]. Similar interventions without the screening component have been found to be cost-saving for women aged 80 years and older in the UK and the USA [57].

An alternative approach to curtail OAF costs lies in the prevention of fractures itself. Based on a sub-analysis, we found that in the period from 2010 to 2050, around 41 % of aggregated direct discounted fracture costs in the German general population aged 50 years and older were attributable to osteoporosis (results not shown). It has been stated that about 90 % of hip fractures, including those attributable to osteoporosis, are caused by falls [58]. It therefore seems prudent to also prevent falls in older age, in order to prevent fractures. A recent systematic literature review found that the prescription of vitamin D and the re-assessment of medication use are effective measures to reduce the rate of falls in nursing home residents. Furthermore, multifactorial interventions in nursing homes showed a significant reduction in hip fracture rates [59, 60]. For community dwelling elderly, there is evidence that the increase of physical activity, in terms of group or individual exercises like balance or functional training as well as Tai Chi, can reduce the rate of falls (and fallers) and decrease the risk of sustaining a fracture [61].

In terms of reducing the future burden of fall-related fractures in Germany, policy makers should primarily implement interventions that are effective and reduce costs, i.e., are cost saving. A literature review which analysed fall-preventing home-based strength and balance programmes in people aged ≥80 years found three of nine programmes to be cost saving [62]. An Australian study analysed the cost effectiveness of a fall prevention programme for community dwelling elderly and nursing home residents-based on three recently conducted meta-analyses. For fall prevention aimed at community dwelling elders, the authors found cost effectiveness ratios between 44,879 and 172,009 AUS$/QALY, whereas inventions for nursing home residents led to cost-effectiveness ratios from “cost-saving” to 56,752 AUS$/QALY (the cost saving strategy was a re-assessment of medication use) [63]. For Germany, no cost-effectiveness analysis for fall prevention was found, except of a cost-effectiveness analysis for hip protectors in nursing home residents over 80 years of age. The authors found that hip protectors led to cost savings in between 257 and 315 Euros per person over lifetime [64].

In summary, our study showed that OAF will sharply increase until 2050. The corresponding costs pose a substantial economic burden on the German healthcare system. Therefore, interventions capable to reduce the future healthcare costs attributable to OAF should be developed. Effective interventions can be economically evaluated using the model described in this study.


Referring to cost ranges (± percent from mean costs) univariate distribution were used.



This work was funded by the German Federal Ministry of Education and Research (BMBF), Germany, FKZ:01EC1007C. Petra Benzinger and Kilian Rapp were supported by a grant from the “Forschungskolleg Geriatrie” of the Robert Bosch Foundation. The BMBF and the Robert Bosch Foundation had no further role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Conflicts of interest


Supplementary material

198_2012_2020_MOESM1_ESM.doc (310 kb)
ESM 1(DOC 309 kb)

Copyright information

© International Osteoporosis Foundation and National Osteoporosis Foundation 2012