Expected Benefits and Budget Impact From a Microsimulation Model Support the Prioritization and Implementation of Fracture Liaison Services

Osteoporotic‐related fractures cause significant patient disability, leading to a growing burden on health care systems. Effective secondary fracture prevention can be delivered by fracture liaison services (FLSs), but these are not available in most countries. A major barrier is insufficient policy prioritization, helped by the lack of economic assessments using national data and providing estimates of patient outcomes alongside health care resource use and cost impacts. The aim of this study was to develop an economic model to estimate the benefits and budget impact of FLSs and support their wider international implementation. Five interconnected stages were undertaken: establishment of a generic patient pathway; model design; identification of model inputs; internal validation and output generation; and scenario analyses. A generic patient pathway including FLS activities was built to underpin the economic model. A state‐based microsimulation model was developed to estimate the impact of FLSs compared with current practice for men and women aged 50 years or older with a fragility fracture. The model provides estimates for health outcomes (subsequent fractures avoided and quality‐adjusted life years [QALYs]), resource use, and health and social care costs, including those necessary for FLSs to operate, over 5 years. The model was run for an exemplar country the size of the United Kingdom. FLSs were estimated to lead to a reduction of 13,149 subsequent fractures and a gain of 11,709 QALYs. Hospital‐bed days would be reduced by 120,989 and surgeries by 6455, while 3556 person‐years of institutional social care would be avoided. Expected costs per QALY gained placed FLSs as highly cost‐effective at £8258 per QALY gained over the first 5 years. Ten different scenarios were modeled using different configurations of FLSs. Further work to develop country‐specific models is underway to delivery crucial national level data to inform the prioritization of FLSs by policy makers. © 2023 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).


Introduction
O steoporotic-related fractures cause significant patient disability and reduce survival, leading to a substantial and growing burden on health care systems globally. (1) Patients with a previous fragility fracture are at higher risk of subsequent fracture. (2) Further, a recent major fragility fracture increases risk more in the imminent risk period as well as in the longer term. (3)(4)(5) The clinical effectiveness of secondary fracture prevention is established. Randomized controlled trials have demonstrated clinically significant reductions in fracture risk in those with recent major fragility fractures (6) and those at very high fracture risk. (7,8) This is supported by real-world evidence studies that have demonstrated clinically significant reductions in major fractures by anti-osteoporosis medication in patients with a recent fracture. (9) This has led to national (10,11) and international (12)(13)(14)(15)(16) initiatives to deliver effective secondary fracture prevention by implementing fracture liaison services (FLSs). FLSs are small groups of health care professionals who identify, assess, recommend treatment, and monitor adults who are recently diagnosed with a fragility fracture to reduce their risk of another fracture. (17) As expected, FLSs reduce the risk of subsequent fragility fracture in clinical studies (18,19) and reviews. (20,21) Despite the expected increase in aging populations leading to significant increases in fragility fractures, (22) the majority of health care settings that manage adults with fragility fractures do not have FLSs in place. In the EU, many countries have no reported FLSs and, where present, 50% of countries reported FLS coverage in less than 10% of hospitals. (23) A major barrier to FLS provision is the lack of policy prioritization, especially in comparison with provision for other long-term conditions with similar secondary prevention strategies. (1) In addition to considerations including local need and local health care capacity, understanding the benefits and budget impact of FLSs remains a key barrier. Policymakers need to prioritize secondary fracture prevention in relation to other global, national, and regional health priorities. Understanding the expected benefits as well as costs are critical to informing this decision making. Economic modeling studies have been widely employed to examine the cost-effectiveness of osteoporotic fracture prevention, but they have been limited not only by the use of less advanced techniques (24) but also by the paucity of national data, the flexibility of the patient pathway to reflect real-world patient journeys in terms of rates of identification, treatment recommendations and adherence, previous fracture and anti-osteoporosis medication history, variable outputs that include clinical events, health care use and costs, estimating the scale and type of FLS resources that would be needed, and, finally, independence from commercial bias.
The aim of this study was to develop an economic model based on an internationally applicable care pathway of individuals presenting with a fragility fracture and incorporating the key activities of FLSs to estimate their benefits and budget impact and hence support their wider international implementation.

Materials and Methods
The development of the model followed five interconnected stages: (i) establishment of a generic patient pathway; (ii) model design; (iii) identification of model inputs; (iv) internal validation and output generation; and (v) scenario analyses.

Patient pathway
Following best practice guidelines for the development of economic models, (25) previously published economic studies were reviewed to contextualize the modeling of costs and effectiveness of FLS programs. Key elements of previous models were discussed with an international group of clinical FLS experts from Japan (n = 4), Spain (n = 4), and the UK (n = 2) to ensure the patient care pathway for individuals presenting with fragility fractures was flexible enough to be adapted to different countries and health care systems.

Model design
With a generic patient pathway described in detail, the most appropriate target population, perspective of costs (ie, costs to the payer as opposed to the patient, hospital, or society as a whole), health outcomes, resources used, costs, and time horizon that best served our specific aim were identified as recommended by best practice guidelines. (25) An economic model incorporating all relevant health states, events, transitions, interdependencies, use of health and social care resources, and costs was designed. The key activities of FLSs, namely (i) the proportion of hip, spine, and other fragility fractures identified, (ii) laboratory and bone density testing, (iii) anti-osteoporotic medication recommendations, and (iv) monitoring to boost adherence were included in the model. This ensured the model was responsive to different FLS configurations. Comparators (ie, the strategies being compared) were defined by our aim of estimating the impact of FLSs vis-à-vis current clinical practice, hence the model was run separately with inputs characterizing "current practice" and then with those reflecting patient experience under an FLS. The difference in patient outcomes, resource use, and costs would then be interpreted as the impact of the FLSs. This allowed us to run the model under different FLS configurations in the scenario analysis and compare how impacts varied. Once the model was conceptually designed, the simulation was coded in the R Programming language, defining all necessary input parameters and mathematical relationships to generate expected patient outcomes, resource use, and costs for each comparator.

Model inputs
Different methods were used to identify model inputs. For values that were considered applicable to any country such as antiosteoporotic medication (AOM) efficacy and time to onset, published reviews of AOM efficacy were used. For inputs that were country-specific, we obtained values from (i) published literature, (ii) government or other regional sources, or (iii) consensus from national key opinion leaders (CC, MKJ) where no evidence was found. Key opinion leaders (KOLs) were asked to provide their expert opinion on the most conservative estimates.
The model was run for an exemplar country by populating it with international data for most inputs: general risk of refracture, health care treatment, social care, and FLS activities, as well as treatment effects from the most-used AOMs. To identify input values, the literature was reviewed and the most recent and reliable evidence on each model parameter identified, regardless of country. Where local data were needed such as the number of fragility fractures for the modeled cohort, evidence from the United Kingdom was used. Values from the literature were adapted to match model requirements and then confirmed with a group of expert KOLs. Where no evidence was found, the group was asked to provide their expert opinion on the most conservative estimates. Inputs on unit costs were identified in British pound sterling (GBP; £) and corresponding to year 2021.

Statistical analysis and internal validation
Once the model was populated with all required input parameters, both face and internal validations were conducted. For the former, input sources and confirmation that results correspond with reality were conducted in meetings with KOLs. For the latter, technical consistency and validity were examined by verifying equations, codes, and data against their sources by varying the values of key inputs one at a time and checking whether resulting outputs moved in the direction expected by clinical experts. For example, holding everything else constant, higher values of AOM effectiveness should lead to more fractures avoided and higher cost of hospitalizations should lead to higher total costs. Where this did not happen, code and formulas were reviewed and corrected to make sure all mathematical relationships were accurately captured.
Model outputs are reported as number of subsequent fractures, quality-adjusted life years (QALYs) gain, health care resources used (both inpatient and outpatient), social (formal home and long-term institutional) care, and associated costs by comparator. Each of these was further stratified by sex and sentinel fracture site. Costs and QALYs were discounted at 3.5% yearly rate to provide an estimate of the incremental costeffectiveness ratio commonly used to inform decision-making in health.

Scenario analyses
Several scenario analyses were conducted to explore the ability of the model to capture the potential impact of implementing various FLS configurations based on clinical expertise. Analyses were run on target identification rates at 100% for all sentinel fracture sites, time to treatment initiation at 1 month, and monitoring rates at 100% at both 4 and 12 months after fracture, reflecting published key performance indicators. (17) Scenarios where AOM would be restricted for FLSs to alendronate only, injectables only, adherence of 100%, and a "perfect" FLS with 100% identification, monitoring, and adherence simultaneously (purposely to be used as benchmark) were also explored. Finally, FLS models where only patients with hip fractures, or patients with either hip or spine fractures would be treated, were also simulated.
The Consolidated Health Economic Evaluation Reporting Standards (CHEERS) 2022 statement (26) was followed to make sure all relevant components of this economic assessment were reported appropriately and in a manner that is useful for decision making. The checklist of items of this statement is included in the Supplemental Material.

Patient pathway
A generic patient pathway specifically designed to include the key activities of FLSs was built to underpin the development of the economic model. No previously published economic model was found explicitly based on a patient pathway, but different aspects of the wider context of fracture prevention considered in those models helped inform the process. Fig. 1 shows the pathway developed after several rounds of discussions with clinical experts and key opinion leaders from the United Kingdom, Japan, Mexico, the Netherlands, and Spain.
Sentinel as well as subsequent fractures were grouped into hip, spine, or major fractures in other sites. This was done to reflect the different patient pathways patients with these fractures can experience. Sentinel fractures could lead to hospital admission, generally via accident and emergency (A&E), although some such as spine fractures could be seen directly in an outpatient trauma clinic. After going through A&E, patients would be admitted or discharged, and if admitted, they would receive either surgical procedures or nonsurgical treatment. After discharge from hospital or the outpatient trauma clinic, some patients in certain settings might receive additional residential temporary rehabilitation, but they would all eventually be discharged either back to their own homes, with or without support from a caregiver, to the home of a relative, or to a residential care institution, more commonly after a hip fracture. At any stage of this pathway, patients face the risk of a subsequent fracture at any site as well as a risk of dying. The identification of patients to be included into an FLS could happen at any point, before or after clinical discharge, and in a proportion to be set in the respective model parameters as identification rates would vary by setting, fracture site, and potentially even sex.

Model design
The patient pathway described above underpinned the design of the economic model. As the aim of the model is to estimate the impact of FLSs, the target population was set to be men and women aged 50 years or older, as that is the age at which the likelihood of experiencing a fragility fracture starts to increase rapidly. (27) To be able to inform public health care policy decisions, the base case analysis followed the perspective of the public health care payer; considering the significant impact of fragility fractures to hospitals as well as patients and their families, the model was designed so that it could also take the perspective of a hospital, another payer, or the broader society.
Model outputs were divided into three categories: health outcomes, resource use, and costs. Under health outcomes, the model reports the number of subsequent fractures by site and QALYs. Resources of interest were length of stay at hospital, number of surgical procedures, clinic appointments, number of temporary rehabilitation days, and time in institutional care. For the implementation of the FLS program, the number of hours of staff needed, laboratory tests, and dual-energy X-ray absorptiometry (DXA) scans were also included. Costs are reported in the model separately for those linked to hospital care (procedures, length of stay, and clinics), temporary rehabilitation, community care (for spine fractures or monitoring after discharge), social care (home support and institutional care), and those associated with the FLS program (staff, tests, scans, and AOM). All of the above are tracked in the model over a time horizon of 5 years, chosen based on the trade-off between length of the time horizon and the certainty of key inputs such as site-and sex-specific subsequent fracture rates. Five years provides a timeframe that allows us to use good-quality evidence on risk of subsequent fracture while at the same time providing policymakers with outputs within a relevant time period for their political decision making.

Model structure
The model was designed as a state-based microsimulation (schema shown in Supplemental Fig. S1), with monthly cycles. As people experiencing a fragility fracture enter the model, in each cycle they can either die, suffer a subsequent fracture (in any site), or spend the cycle without any further fracture. The choice was guided by evidence on the time-varying association between risk of subsequent fracture and recency of the sentinel one. (5,(28)(29)(30) This imminent fracture risk has been shown to lead to potentially significant impacts on the benefits of FLSs. (31) The choice of a microsimulation design is justified by the impact of a patient's prior history of fractures on their risk of a subsequent fracture, mortality, and quality of life.

Model logic
Simulated individuals of a given sex (male, female) and fracture site (hip, spine, other) enter the model immediately after their sentinel fracture and can move into the fracture-free state or experience a subsequent fracture in the same or another site.
Each time a simulated individual moves into a subsequent fracture state, events from the patient pathway in Fig. 1 are randomly assigned based on inputs. That is, attending hospital or an outpatient trauma clinic (for patients with spine fracture), hospital admission, having a surgical procedure, discharge to temporary rehabilitation, and ultimate discharge destination are all randomly assigned based on probability inputs. This also applies to FLS activities, with different values applied to "current practice" or the "FLS" strategy. A flow diagram for FLS identification is shown in Supplemental Fig. S2.
Each subsequent fracture is followed by the individual facing a probability of being identified and potentially recommended an AOM. If a change of prescription occurs, it would happen only for one of equal or superior strength. Some AOMs have limits to how long patients can take them: zoledronate can only be taken for up to 3 years continuously, abaloparatide and teriparatide for 2 years, and romosozumab for up to 1 year. In each case, patients could be allocated to no treatment or to any other eligible AOM. More details about the logic behind treatment assignment and a corresponding flow diagram are shown in the Supplemental Material (Supplemental Fig. S3 for the latter).
For treatment adherence, an individual is assigned to being an adherer or non-adherer to their AOM at time of prescription based on probability inputs. Those that are adherent can then become non-adherent at 4, 12, or 24 months, when adherence is reassigned based on probabilities by AOM and sex, if relevant. Supplemental Figs. S4-S6 illustrate the logic for primary and 4-month adherence, as well as the flow diagram for 4-month monitoring.
Each AOM is associated with a specific relative risk reduction for subsequent fracture and to a time lag, expressed in months, denoting the period between treatment initiation and when the relative fracture risk reduction is applied. An individual must be adhering to the medication for any associated relative reduction to be applied. Supplemental Figs. S7 and S8 illustrate the logic behind the application of treatment effect after sentinel fracture and after a further fracture, respectively.
Average age at time of sentinel fracture is an input of the model and is used for all-cause mortality. Fracture-related risks of mortality depend on site of fracture. After a hip or spine fracture, risks of mortality are drawn from inputs. After a fracture in another site, risks of mortality are based on population life tables specific to the country of interest. After a subsequent fracture, the risk of mortality is taken from the highest of either (i) the continuation of the risk given their latest fracture or (ii) a new risk based on their subsequent fracture.
QALYs are estimated by applying a health utility decrement every time a fracture occurs and a progressive improvement afterwards. Starting health utility levels immediately after the sentinel fracture as well as decrements and improvements were taken from an international study reporting on 18-month followup of patients after hip, spine, and forearm fractures. (32) As the model allows for more than one fracture and recovery does not reinstate individuals to their original health utility levels, we applied the proportion of potential change (33) implicit in the Journal of Bone and Mineral Research original study using the lowest health utility as a reference for decrements associated with a subsequent fracture.
Further details about the logic of the model are included in the Supplemental Material.

Assumptions
In addition to the assumptions associated with the model structure and logic as described above, it is assumed that individuals are treatment-naive when entering the model; hence, they can be assigned to any of the AOM in the first instance. Also, no more than one fracture was allowed in a given month; non-hip, nonspine fractures are assumed not to lead to excess mortality beyond baseline all-cause risk of death, and health utility remains constant after 18 months post-fracture, unless another fracture occurs.

Model inputs
The model requires inputs that characterize the trajectory of patients over the pathway shown in Fig. 1 under current practice and under the assumption FLSs would be widely implemented. Each of these strategies had their corresponding mean transition probabilities, use of resources, and costs. One of the key inputs of the model is the risk of a subsequent fracture as experienced by people not under treatment, so that a relative risk can then be applied when the patient is under treatment and adherent to it, as explained in the model logic. These fracture rates are obtained at 5 or 10 years, depending on availability, and then monthly probabilities estimated based on previously reported 10-year non-linear progression of subsequent fractures in men and women following individual sentinel fractures. (5) Table 1 lists several key inputs feeding the model for both current practice and FLS for the exemplar country used for this analysis. International values were obtained from the literature, estimated, or identified by KOLs. These, together with their respective sources, are shown in the Supplemental Material for all inputs.

Internal validation and results
To estimate population-level outcomes and budget impact, the model was run for a given number of simulations and results scaled to the size of each of the six cohorts (3 fracture sites Â 2 sexes). The model was run for a country of the size of the United Kingdom, with the following number of expected sentinel fragility fractures in a given year: 16,826 spine, 22,434 hip, and 72,911 other major fractures for men, and 33,651, 44,868, and 145,812, respectively, for women. Health outcomes, resource use, and costs were generated scaling the results from the number of simulations to the specified cohort sizes. The number of simulated individuals considered sufficient to run the model was taken to be the number after which the expected number of refractures for each comparator remained stable. This was reached at 75,000 simulated individuals per cohort, totaling 450,000 simulated individuals per comparator (900,000 in total), before scaling to produce outputs for the specified cohorts. Health outcomes, resource use, and costs were generated by cohort for each comparator, with the differences representing the impact of FLSs and reported by sentinel fracture site, sex, and overall.
Internal validation was conducted by running the base case model multiple times, first with the set of input parameters (reported in the Supplemental Material) and subsequently by increasing and decreasing the values for average age, refracture rate, treatment rate, case identification, time to treatment initiation, treatment effect, adherence rate, monitoring rate, mortality rate, % of patients receiving a procedure, time spent by stage of secondary fracture prevention, unit costs of AOM, and discharge destination, one at a time. Expected impacts from each change in input parameters were contrasted with corresponding model outputs. Coding errors or shortfalls were identified and corrected until all changes in parameters led to expected changes in outcomes. Final results were discussed with KOLs and face validity was confirmed.
A summary of the health outcomes under current practice compared with FLS is shown in Table 2. Based on model input parameters and model assumptions, the implementation of FLSs would avoid 13,149 subsequent fractures over the first 5 years of FLS implementation. Avoided non-hip, non-spine fractures accounted for 48% of the total, with those of the hip and spine accounting for 36% and 16%, respectively. Avoiding these subsequent fractures would lead to a gain of 11,838 QALYs over this period.
Health and social care resource use and costs for current practice and FLS are also shown in Table 2. As reported in the table, FLSs would lead to a reduced demand for health care provided at the hospital and community, reflected in the lower numbers of surgeries, hospital bed days, clinic appointments, and rehabilitation. This would be due to the reduction in subsequent fractures as a direct result of the services provided by FLSs, which would require increased resources in the form of FLS staff, DXA scans, and laboratory tests, as shown in the table. FLSs would lead to a shift in patient-facing time from doctors toward FLS administrators and coordinators, often provided by nurses and other health-related roles.
The model shows that fractures avoided also means the number of people requiring institutional care due to their diminished independence would decrease. Both the expected number of patient years in institutional care and the number of patients ever to require moving into them would drop if FLSs are implemented (by 3556 and 1910, respectively), as shown in Table 2.
In terms of costs, as with resource use, FLSs would lead to savings of health and social care funding associated with the treatment of and care for people after fractures, with an expected increase in the costs of FLS prevention services. As shown in Table 2, there would be an expected cost savings of £69.2 million from health care services, mainly from the number of surgical interventions avoided (accounting for 50% of total health care savings) and temporary rehabilitation (43%). Social care services also report expected savings of an additional £36.6 million. Although savings from long-term institutional care would be expected to reach £130 million, many people avoiding a subsequent fracture and institutional care as a result would still require formal care at home, leading to an estimated cost increase of £93.3 million, yet leading to a savings in social care overall. The cost impact of providing the FLS services responsible for the above savings are also reported in Table 2. AOMs make up 82% of the costs of running the modeled FLS. Overall, the implementation of FLS programs lead to a 5-year additional investment of £96.7 million (discounted over time). This represents 0.4% of the total costs estimated to be incurred under current practice, which combined with the expected QALY gains would lead to an incremental cost-effectiveness ratio of £8258 per QALY over the first 5 years of FLS implementation. Fig. 2 shows the yearly extra (undiscounted) costs and QALYs gained by implementing FLS compared with current practice over the 5-year time horizon of the model. For this cohort  Table 3 reports the main results of the model for the 10 different scenarios examined, reflecting various configurations of FLSs. In all cases, specific changes were made as reflected in scenario titles while keeping all other FLS inputs constant to examine the impact of the specific change being investigated. FLS identifying all fragility fractures would lead to more fractures avoided and more QALYs gained at a slightly higher cost compared with the FLS base case, hence producing a better (lower) ratio of extra cost per QALY gained. Starting treatment 1 month after sentinel fracture would also represent an improvement compared with the FLS base case, though only slightly. FLS costs would be £3 million higher as patients would be on treatment for longer. Compared with the base case, FLSs monitoring all patients would lead to more fractures avoided and more QALYs gained at similar levels to initiating treatment at 1 month but at overall costs lower than in the base case, which already has a monitoring rate of 80%.

Scenario analyses
An FLS strategy of interest would be to provide patients with injectable AOMs only, given their higher adherence and generally higher effectiveness. We explored two such scenarios: one with clinical criteria that considered both effectiveness and costs ("standard") and a second one where the most effective injectables were chosen regardless of costs ("maximum reduction"). Both were significantly more costly (FLS costs and overall), but they also improved health outcomes compared with FLS base case. As expected, the max reduction scenario led to higher number of fractures avoided vis-à-vis current practice but at significantly higher FLS operation costs than standard injectables only. This was reflected in a nearly double incremental costeffectiveness ratio of £49,201 per QALY gained under the injectables-only max reduction scenario compared with £27,452 under standard injectables only. In contrast, if FLSs recommended only the use of alendronate, they would reduce fractures by only 10,993 but at such low costs that, considering overall health and social care cost impacts, FLSs would have dominated current practice, ie, resulted in gains in QALYs while being cost-saving at the same time.
Keeping pharmacologic prescriptions as used in the FLS base case but rising adherence to 100% would increase fractures avoided and QALYs gained over the base case and lower costs, leading to a more favorable cost per QALY of £7055. Considering FLSs that simultaneously achieved 100% identification, treatment initiation at 1 month, 100% monitoring, and 100% adherence (perfect FLS scenario) would produce the best results of all in terms of health outcomes (fractures avoided and QALYs) at a cost per QALY of £5785. FLS costs would be slightly higher than the base case, but overall costs would be lower. Finally, we explored two more scenarios where only hip, or only hip and spine fragility fractures would be treated in FLSs. They both expectedly led to significantly lower numbers of factures avoided, although treating only hip and spine fractures would reduce subsequent fractures by 13% of those estimated under a current practice that treated hip and spine fractures only as well, the highest expected change reported by all scenarios examined. Both scenarios reported lower overall costs, hence becoming dominant over their respective current practice comparator.

Discussion
We established a patient pathway for people having fragility fractures that summarizes the main events and contacts with the health and social care systems and has the flexibility to be adapted and used in different countries. A scope of the literature for its development found no previous models reporting an explicit patient pathway underpinning the economic model. The pathway developed for this model furthermore highlights the importance of social care after a fragility fracture given the The model developed in this study uses microsimulation methods to estimate the incidence of subsequent fractures and their QALY impact on men and women who have had a previous fragility fracture, all relatively common features of previously published osteoporosis models. (34)(35)(36)(37) However, the model presented here is novel in that it uses a 1-month cycle, which allows the incorporation of imminent risk of subsequent fracture; it accounts for time to treatment initiation, time for treatment to take effect depending on the AOM prescribed, and adherence to the specific drug, all critical to accurately estimate the incidence of subsequent fractures and the impact of programs that prevent them. The model is also unique in that it is centered on assessing the benefit and budget impact of FLSs by including as part of the pathway and microsimulation the key activities of FLSs, ie, patient identification, assessment, treatment, and monitoring. The model reports key data for decision makers, including the number of staff, laboratory and bone density testing, and medication costs required for FLS implementation. Finally, the model permits an FLS to enter its current performance data to understand the expected impact on patient, health care resource, and economic outcomes, and where to focus service improvement.
Many inputs are needed to run the model, and these are expected to be obtained from various sources, with emphasis on local data if available, followed by evidence from neighboring countries if found acceptable to local experts, international evidence for those inputs generalizable across countries, and local expert opinion otherwise. For this study and for all future application of this model, iterative discussion and validation of inputs with local experts is a priority, not only for the face validity of  results but also because of its directed aim to support local decision making for the implementation of FLSs. The identification of inputs to run this model for any setting would place a significant burden on analysts, given its complexity and large number of elements it considers. Although other calculators offer simpler structures that require considerably fewer inputs, (38) the model described here offers benefits the simpler models cannot, which are relevant to policymakers such as disaggregation of results by sex and fracture site, proportion of cohort benefiting from the application of relative risk reduction from medication intake over time, impact of adherence, and detailed estimated resource use and costs of health and social care, among others. By generating estimates of fractures avoided and QALYs gained, as well as savings in health care resource use, extra resources needed to provide the prevention services (health care staff, tests, scans), and cost implications, this model can comprehensively inform decisions around FLS implementation. The ability of the model to assess the benefit and budget impact of different configurations of FLSs, known to vary substantially among different settings, will particularly facilitate the decisionmaking process.
For the base case examined in this study, model results using international data as inputs for a country the size of the United Kingdom showed that an FLS program would lead to a reduction of 13,149 subsequent fractures and a gain of 11,709 QALYs over its first 5 years of operation. This would also free important levels of health and social care services and funding that would otherwise have been needed for the care of individuals having subsequent fractures. Expected extra costs per QALY gained placed FLSs as a highly cost-effective intervention at £8258 per QALY gained over the first 5 years, which compares highly favorably with the cost-effectiveness threshold of £20,000 to £30,000 per QALY gained used in England or the US $50,000 commonly used in many other countries. Furthermore, model results make explicit that while extra costs are concentrated on the first year for a static cohort, gains in QALYs would continue to grow over time. The former is explained by the prevention investment required at the start of implementation and later dropping for two reasons: lower demand because of decreasing subsequent fracture incidence, and savings from resources freed due to the refractures avoided. The increasing year-on-year QALY gain is explained by the sustained QALY benefit of a fracture avoided as its positive impact in terms of health utility gain would remain over the individual's lifespan.
The model proved uniquely useful to investigate the impact that specific configurations of FLSs would have on health outcomes as well as health and social care resources and costs. A scenario termed perfect FLS, which assumed that FLSs would identify all fractured patients, treat them within 1 month, monitor them all, and achieve 100% adherence, served as a reference benchmark against which the other effectively plausible configurations could be matched to. Although it may not represent a realistic scenario for most countries, this scenario does illustrate what countries considering FLS implementation could potentially achieve. We found that the extra costs per QALY gained under this configuration (£5785) is in line with several publications reporting on the cost-effectiveness of FLSs, (39) suggesting that they implicitly assume that these services can operate at that level. Otherwise, maximum fracture prevention was achieved if FLSs identified 100% of fractured patients, if they prescribed only injectable AOMs or if they managed to lead to 100% adherence. In some cases, there were relevant trade-offs between these results and costs, such that prescribing only injectable AOMs would lead to the highest levels of additional costs. Although the lowest extra cost per QALY gained was achieved by FLSs if they only recommended the use of alendronate, they would reduce fractures by 27% less than FLSs prescribing injectables only. Treating only patients with hip/spine or only with hip fractures would be expected to prevent more fractures and be cost saving compared with current practice but at the cost of many fewer people benefiting from avoiding a fracture.
We observed that FLSs increase formal care at home in favor of reducing institutional stays, which could potentially place a higher burden on individuals or their families, including higher informal costs of social care and even productivity losses. Keeping people away from institutionalized care is, however, a sign that FLSs can safeguard the independence of people who avoid further fractures, which has a significant impact on their lives and that of their families.
However, this work is not without limitations. First, we modeled subsequent fractures only over a period of 5 years, which is not a realistic reflection of the time horizon of the impact of fractures on patients or FLSs to prevent subsequent ones. Although this was done mainly to prioritize accurate and available data on site-and sex-specific refracture rates, which rarely extends beyond a couple of years, it is also aligned with the time window of interest for decision makers in many settings. The model could be strengthened by covering the full expected lifetime of patients; notwithstanding this, examining the benefits and budget impact over the first 5 years of implementation can provide valuable insight for decision makers assessing the value of initiating FLS operations. Also, the model assumes an average refracture rate for all adults, which on average might lead to accurate results, but it is known that patients identified after a fracture are recommended treatment based on their risk of a further fracture and this varies depending on several patient-related variables. As a result, our findings likely underestimate the benefits of FLSs as these effectively operate by identifying patients at higher risk and treat them, which the model could simulate if a risk stratification were to be introduced. Another limitation adding to its conservative estimations of FLS benefit is the assumption that all simulated patients are treatment naïve when they enter the model. This is not the case in many developed countries, where especially people suffering from a hip fracture would have had some AOM treatment before. In those cases, model findings would underestimate fracture prevention and QALY benefit; however, in most other settings where secondary fracture prevention is weak, this assumption would be fitting. Further, many of the data required for the model were challenging to identify at the country level. Siteand sex-specific refracture rates were especially hard to find, especially in men, as was mortality by sex after non-hip fractures, discharge destinations for individuals with non-hip fractures especially after temporary rehabilitation, and adherence to therapy at 0, 4, and 12 months and 5 years across the range of AOMs. The costs of informal care and of productivity loss were not included in this study but would be relevant to incorporate in a future version. Evidence about the use of informal care and the impact on productivity of fragility fractures is scarce, but the model can incorporate these if they were available.
In conclusion, we have developed a flexible model that estimates the expected benefits and budget impact from FLS implementation, including the examination of different FLS configurations. Further work to develop country-specific models is underway to deliver crucial national-level data to inform the prioritization of FLSs by policymakers.