Background
Osteoporosis is a major public health concern in Mainland China. It was estimated that there were up to 70 million osteoporotic patients, and over 200 million patients with osteopenia in China in 2006 [
1]. Osteoporotic fracture is the most severe complication of osteoporosis and results in hospitalization, disability, and mortality. The prevalence of vertebral fracture in postmenopausal Chinese women increased from 13.4% at ages 50–59 years, to 58.1% at age 80 years or older [
2]. The overall major osteoporotic fracture in Chinese population was estimated to be 2.69 million cases in 2015 [
3], indicating an enormous disease burden of osteoporotic fractures in China.
Osteoporosis treatment beforehand exhibits efficacy in fracture risk reduction. Of all treatment regimens, bisphosphonate is the most commonly used drug and was predicted to be cost-effective in other models [
4]. In one of our previous studies (unpublished), alendronate was dominated by zoledronate in cost-effectiveness analysis, suggesting zoledronate a better option for treating Chinese postmenopausal women. In Mainland China, there is no universally accepted policy for treating patients at high risk, and only a minority of osteoporotic people receive proper treatment due to lack of evidence-based guidelines and recommendations [
5]. Fracture risk assessment tool (FRAX) is a popular tool worldwide, used to predict 10-year probability of osteoporotic fracture based on clinical risk factors and femoral neck bone mineral density (BMD) [
6]. Several cost-effectiveness analyses based on Western settings suggest that osteoporotic treatment should be initiated when the 10-year major fracture probability, based on FRAX, is over 7-15% [
7‐
10]. Previous studies on estimating the FRAX threshold for the Chinese population were only limited to clinical observations or expert consensus [
11‐
14].
We introduced real-world data from the Peking Vertebral Fracture (PK-VF) study into modeling to improve the robustness of cost-effectiveness analysis. The PK-VF study is currently the largest prospective cohort study of postmenopausal osteoporotic women in Mainland China, which involved 1100 women and follow-up from 2008 to 2013, to capture baseline epidemiological data of osteoporosis and osteoporotic vertebra fracture [
2,
15‐
17]. The epidemiological data of the cross-sectional survey in 2008 has been published, including prevalence of osteoporosis and osteoporotic vertebra fracture and the distribution of risk factors of osteoporosis in the population [
2].
In this study, we built a micro-simulation Markov model based on the PK-VF study. We aim to estimate the intervention threshold of FRAX in Mainland China, at which treatment with zoledronate could be cost-effective.
Discussion
In this study, we built an individual-based Markov model of osteoporosis simulating postmenopausal women in Mainland China, based on real-world data from the PK-VF study. We predicted the intervention threshold of FRAX in Mainland China at which treatment with zoledronate could be cost-effective and calculated the variation of the threshold in different age groups.
In Mainland China, there is no accepted criterion for initiating the treatment of osteoporosis. Current expert consensus in China is that anti-osteoporosis treatment should begin in any of the following conditions: (1) the occurrence of vertebral fragile fracture or hip fragile fracture; (2) BMD of femoral neck, whole hip, lumbar vertebrae, or distal radius < − 2.5; (3) − 2.5 < BMD < − 1 plus (1) occurrence of the proximal humerus, distal forearm, or pelvis fragile fracture or (2) 10-year hip fracture probability > 3% or 10-year major fracture probability > 20% predicted by FRAX [
41]. However, there has been no cost-effectiveness analysis to determine a Chinese-specific threshold to initiate osteoporosis treatment based on FRAX. Thus, building a cost-effectiveness model that predicts FRAX thresholds for the Chinese population is both important and urgent.
Several cost-effectiveness analyses have been performed on predicting the disease burden of osteoporosis in postmenopausal women in Mainland China [
3,
42‐
45], but none of them were based on real-world data of osteoporotic cohorts in the Chinese population. This study, however, was based on our previous Peking Vertebral Fracture (PK-VF) [
2,
15‐
17], in which 1100 Chinese postmenopausal women were followed up from 2008 to 2013, and data were collected on the incidence and prevalence of osteoporosis and osteoporotic vertebra fracture, and the distribution of risk factors of osteoporosis in the population [
2]. Introducing real-world data into modeling would significantly improve the simulation.
The determination of cost-effectiveness FRAX intervention threshold should be country-specific, considering the large variation of epidemiology and economic conditions across countries. For example, osteoporosis treatment was predicted to be cost-effective when 10-year major fracture exceeded 7% in UK [
7], 15% in Switzerland [
8], 10% in Greece [
9], 8.8% in Portugal [
10], and when 10-year hip fracture exceeded 3% in USA [
46]. There have been a few studies estimating the intervening threshold of osteoporosis for Chinese population based on FRAX, but they were only limited to clinical observations or expert consensus without the evaluation of the economic and disease burden of osteoporosis [
11‐
14].
No cost-effectiveness analysis has been done on estimating FRAX threshold in China so far. One of the reasons was that real-world data of the distribution of clinical risk factors were not available in China before [
45]. There were eight clinical risk factors in FRAX model to determine a 10-year probability of major fracture and hip fracture of an individual, including age, sex, BMI, previous fracture, parent fracture hip, current smoking, glucocorticoid use, rheumatoid arthritis, secondary osteoporosis, and alcohol use (with or without the addition of femoral neck BMD). As risk factors of osteoporosis were always not independent of each other, it would lead to potential bias if we were to bring each risk factor separately into the model. For example, current smokers are more likely to be a regular alcohol drinker. To solve this issue, we generated a sampling table of risk factors based on our PK-VF cohort, and randomly sampled a group of risk factors as a whole from the sampling table for each subject simulated in the model.
There are various anti-osteoporosis drugs among which bisphosphonate is the most widely used one. Bisphosphonate functions in inhibiting the bone reabsorption process of osteoclasts. Alendronate, an oral bisphosphonate, is dosed once per week, and zoledronate is dosed once a year intravenously. Alendronate and zoledronate reduced the risk of vertebral fracture by 48% and 70% respectively in clinical trials [
26,
47], and in one of our previous studies (unpublished), alendronate was dominated by zoledronate in cost-effectiveness analysis, suggesting zoledronate a better option for treating Chinese postmenopausal women.
As indicated in Fig.
2, FRAX threshold for the whole Chinese population was predicted to be 7%. FRAX thresholds differed in different age groups (Fig.
3 and Supplementary Table
4), as the number first increased in age groups younger than 65 years, and decreased in elder age groups. The reason behind this phenomenon is the bidirectional effect of aging on the resultant gain of QALYs, and thus on the ICERs, in a treatment-as-prevention mode, (1) early intervention with low FRAX threshold among younger patients, who have longer expected lifetime, resulted in more cumulative QALYs gained. (2) Intensive intervention with low FRAX threshold among older patients, who have higher fracture risk, resulted in a marked reduction of osteoporotic fractures and related deaths and thus in more QALYs gained. Therefore, we suggested additional resources and attentions shift to the middle aged or advanced aged.
In the external validation table (Supplementary Table
3), a gap was found between the projected life expectancy (16.49) and the reference life expectancy (16.9) in the age group of 65–69 years. We think the reason for the gap was that we used the overall death rate by age from a national survey [
33], and applied osteoporosis, fractures, and deaths due to fractures to the simulated population additionally. Therefore, the simulated population presented a slightly higher death rate and thus a shorter life expectancy, especially among the elders. However, such an overestimated death rate would result in an underestimated benefit of anti-osteoporosis treatment, because it relies on long-term follow-up to observe the benefit of treatment of osteoporosis in preventing fractures, and overestimation of the death rate tends to underestimate the preventive gain of QALYs and the cost-effectiveness of treatment. Currently, treatment of osteoporosis in clinical practice was rather insufficient. With such a conserved estimation, we put forward a FRAX-based intervention threshold as 7%.
There are certain limitations to this study. First, not all data needed for modeling were available for the Chinese population. For example, the incidence of wrist fracture was adopted from the Norwegian population, and multiplied by 0.72 to adjust to the Chinese population [
20]. And we generally stratified transition-probabilities to osteoporotic status from healthy status by age, regardless of those listed risk factors such as BMI, parent fracture, current smoking. Since only 67 new cases of osteoporosis were observed in the 5-year follow-up of PK-VF study, the impact of these covariates on the incidences was not available. And for the same reason, we took only age, previous fractures, treatment, and osteoporotic status as impact factors on transition probabilities of fractures but did not considered the impact of all listed risk factors above. Second, our real-world data were collected in Beijing population, which may not represent the complete picture of the Chinese population. Our group is now performing an ongoing national-wide epidemiological study on osteoporosis with over 20,000 people recruited, and will attempt to incorporate these data into modeling in the future. Third, side effects of zoledronate, such as osteonecrosis of the jaw and atypical femoral fracture, were not considered in the model. The reason is that the health state utility of these side effects was unclear for Chinese population, and these effects occur rarely at the dose of treating osteoporosis. Fourth, one-way sensitivity analysis was performed in the study, which may not be as robust as probabilistic sensitivity analysis. However, we found ICERs tested on all variables in the current sensitivity analysis fell below $23,000, suggesting the results were not sensitive to the uncertainty of those input data. Fifth, we used visual analogue scale (VAS)-based QALY by fracture types as the input data of the model. Though a trade time-off (TTO)-based QALY model was available for Chinese population [
48], we were unable to use it to calculate QALY by fracture types, as we did not collect the information in 5 dimensions (i.e., mobility, self-care, usual activities, pain/discomfort, and anxiety/depression) and 3 levels (i.e., 1 = no problems, 2 = some/moderate problems, and 3 = extreme problems) for each individual in PK-VF study. As an alternative, we used the VAS-based QALY data by fracture types in the model [
25], which has also been used in other previous cost-effectiveness analysis [
43,
45].
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.