Archives of Osteoporosis (2018) 13: 45

https://doi.org/10.1007/s11657-018-0456-2

ORIGINAL ARTICLE

Treatment costs and cost drivers among osteoporotic fracture

patients in Japan: a retrospective database analysis
Yurie Taguchi 1 & Yuta Inoue 2 & Taichi Kido 2 & Nobuhiro Arai 2

Received: 28 November 2017 / Accepted: 28 March 2018 / Published online: 25 April 2018
# The Author(s) 2018

Abstract
Summary This study estimated the direct medical costs of osteoporotic fractures from a large claim database in Japan. We further
identified several comorbidities which drove the treatment costs. The results would contribute to health economic analysis as well
as understanding of individual financial burden in Japan.
Introduction The purposes of this study were to estimate treatment costs of osteoporotic fractures and to investigate the cost
drivers.
Methods Male and female patients aged 50 years and older with a hip, vertebral, or non-hip/non-vert (NHNV) fracture between
April 2008 and December 2016 were analyzed from claim database. Two types of costs were estimated. The incremental yearly
costs of fractures and comorbidity treatments (total medical costs) were calculated by subtracting pre-fracture costs from post-
fracture costs. The costs exclusive for fracture treatments (fracture treatment costs) were estimated by summing up the costs of
fracture treatments within 1 year after fracture. The associations between comorbidities and costs were examined with a gener-
alized linear model.
Results Total 12,898 patients were identified (83% was female). The total medical costs of fractures were $14,592 for male-hip,
$15,691 for female-hip, $4268 for male-vertebral, $3819 for female-vertebral, $3790 for male-NHNV, and $4259 for female-
NHNV. The fracture treatment costs were $4506 for male-hip, $5427 for female-hip, $1022 for male-vertebral, $1044 for female-
vertebral, $1035 for male-NHNV, and $1408 for female NHNV. Three comorbidities were associated with increasing fracture
treatment costs whereas four comorbidities were associated with decreasing fracture treatment costs. Five comorbidities were
associated with increasing total medical costs whereas one comorbidity was associated with decreasing total medical costs.
Conclusions Yearly treatment costs were increased considerably after fracture. Several comorbidities were considered to be cost
drivers for osteoporotic fracture treatment. The cost estimates with different patient profile would support conducting health
economic analysis in the future.

Keywords Osteoporosis . Fracture . Cost . Japan . Comorbidity . Database

Introduction quality of life [1, 2]. It is established that a risk of osteoporosis

and osteoporotic fracture increases as age increases [3], and
Osteoporotic fractures often have serious health impacts. For Japanese population is going through the rapidly increasing
example, in a previous research, those with osteoporotic frac- elderly population. In 2015, people aged 65 years and older
tures had a higher mortality rate and decreased individual consisted 26.6% of total population, and it was estimated that
this proportion would increase to 37.7% by 2050 [4].
Furthermore, particularly in Japan, the financial burden of
* Yuta Inoue
osteoporotic fracture is immense. Moriwaki and Kamae esti-
yutinoue@tohmatsu.co.jp mated that the financial burden due to osteoporotic hip and
vertebral fracture was 27.5 billion JPY (250 million USD
1 when 1 USD = 110 JPY) in 2009 [5].
Amgen Astellas BioPharma K.K, 1-7-12 Marunouchi, Chiyoda-ku,
Tokyo, Japan Under this situation, the health economic analysis in osteo-
2 porosis and osteoporotic fracture is featured more than ever. In
Deloitte Tohmatsu Consulting LLC, 3-3-1 Marunouchi, Chiyoda-ku,
Tokyo, Japan recent years, new drugs, such as teriparatide and denosumab,
45 Page 2 of 12 Arch Osteoporos (2018) 13: 45

have been launched in Japan with higher drug prices than con- Study population
ventional bisphosphonates. In line with the increasing varieties
of osteoporosis drugs, attention by researchers has been paid to Patients aged 50 years and older with diagnosis of osteopo-
cost-effectiveness of these drugs. Ding et al. (2008) analyzed rotic fracture, dating from April 2008 to January 2017, were
the cost-effectiveness of risedronate sodium compared with no- identified from the database using International Classification
treatment [6]. Moriwaki and his colleagues examined cost- of Diseases (10th revision) (ICD-10). Patients with osteopo-
effectiveness of alendronate versus no-treatment in 2013 as rotic fractures were defined as those with M80 (osteoporosis
well as cost-effectiveness of zoledronic acid versus alendronate with pathological fracture) or those with both M81 (osteopo-
in 2017 [7, 8]. Mori et al. (2017) reported cost-effectiveness of rosis without fracture) and ICD-10 code for fractures which
denosumab versus alendronate in 2017 [9]. In terms of treat- can be assumed as osteoporotic fractures. These fractures were
ment cost parameters such as the yearly treatment costs for decided based on the guidelines and expert opinion, and cat-
fractures, most studies used the cost input retrieved from the egorized into three groups: hip, vertebral, and non-hip and
Japanese medical fee schedule with a set of treatment assump- non-vertebral (NHNV) fractures [3]. The list of fractures is
tions or estimates based on small-sized patient population. This as follows: hip (S72.0, S72.1, S72.2, and S32.4); vertebral
is because there has been no study which estimated the treat- (S12, S22, and S32 excluding S32.3-5); and NHNV (S22.3-
ment costs of osteoporosis and osteoporotic fractures with 5, S32.3, S32.5, S42.2, S42.4, S52.0-1, S52.3, S52.5-6,
large-sized real-world data (RWD). Given that a considerable S72.3-4, S72.7, S72.9, S82.0-1, and S92.0).
size of claim database has recently become commercially avail- The index date was defined as the date when the fracture
able, estimating osteoporotic fracture treatment costs using was first diagnosed since they were enrolled in the database. If
them is considered valuable. a patient develops a fracture of different ICD-10 code from the
Therefore, the primary objective of this study is to es- first one within 1 year, the fracture was defined as the second
timate the current yearly treatment costs due to osteopo- fracture. Patients who had a record on January 2017 were
rotic fractures as well as treatment costs of osteoporosis excluded as they may still continue their treatments thereafter.
patients in Japan using large-sized RWD. It will not only As a result, we used the patient data from April 2008 to
help estimate the financial burden of osteoporotic fracture December 2016. In addition, patients whose initial records
but also help conduct various health economic analyses. or final records in the database were within 1 year from the
The secondary objective of this study is to identify cost index fracture were excluded to avoid including those who
drivers among osteoporotic fracture patients. We defined changed hospitals or died within 1 year.
cost drivers as the factors associated with increased or
decreased treatment costs of osteoporotic fracture hereaf- Treatment costs
ter and explored them via the retrospective database
analysis. Two types of treatment costs were estimated in this study. One
is the cost incurred by any medical resource use due to fracture
development, and this was named as Btotal medical cost.^ The
Methods other is the cost incurred by the medical resource use for
osteoporosis and fracture treatments, and this was named as
Data source Bfracture treatment cost.^ Medical costs in Japan can be cate-
gorized into four: doctor consultation costs, drug costs, non-
In this retrospective observational study, the medical claim drug treatment costs (including diagnosis), and inpatient spe-
database provided by Medical Data Vision Co. Ltd. (MDV; cific costs (such as room charges). The Btotal medical costs^
Tokyo, Japan) was used. MDV accumulated anonymous includes all of these whereas the Bfracture treatment cost^
patient-level medical claim data from more than 17.23 million includes drugs and non-drug treatment costs related to fracture
unique patients in more than 240 acute-care hospitals, which and general orthopedic treatments. The total medical cost was
is almost one-eighth of the total Japanese population [10]. estimated by subtracting the treatment costs within 1 year be-
The MDV database includes patient demographics (age fore index fracture from the treatment costs within 1 year after
and gender), resource use, and diagnosis history including index fracture. The fracture treatment cost was estimated by
the date when it was first diagnosed at enrolled hospitals. aggregating the costs for the specific medical resources used
The resource use history covers amounts used and prices for for fracture treatments within 1 year after fracture develop-
each resource (drug/medical procedure/tools). However, it ment. These resources were defined by using two reference
does not contain socio-economic status (SES) records, patient books which describe the resources by disease indication [11,
behavioral records such as smoking history for outpatients, 12]. The resources with indication of fracture and general
and treatment outcome records (i.e., cured/dead/transferred) orthopedic treatments were extracted from these books.
in outpatient service. When estimating yearly costs for the second fracture, 1 year
Arch Osteoporos (2018) 13: 45 Page 3 of 12 45

since the date of second fracture development was defined as a total medical cost, the associations between patient profiles
period for estimation. (demographic factors, fracture type, treatment period, treat-
Costs were calculated by multiplying the amount used by ment costs in the year before fracture, and comorbidities)
the unit price for each resource use record from public health and change in the total medical costs were examined.
care payer’s perspective. The prices used in this study were Drivers for the treatment costs of NHNV fracture were not
adjusted to the latest unit prices issued in 2017 by the govern- examined because NHNV fractures consist of fractures at dif-
ment, which are applied to every hospital/clinics in Japan. All ferent body parts (e.g., radius/ulna or chest), and its result
the costs were transformed into USD (assuming 1 USD = 110 would be difficult to interpret as we would not be able to
JPY). identify which NHNV fracture was driven by a particular
factor. All statistical analyses were performed using R version
Data analysis 3.3.3 (R foundation for statistical computing, Vienna,
Austria).
Both the total medical costs and the fracture treatment costs
were estimated at patient level and reported as means with
95% confidence interval (CI). Before reporting the descriptive Results
statistics, data with extreme values in the total medical costs
were excluded as outliers. The data were stratified by gender Participant characteristics
and fracture category, then a box-and-whisker plot was made
for each subgroup. The outliers were defined as those outside From April 2008 to December 2016, a total of 14,001 patients
the whisker. When comparing the costs of those patients with were identified as patients with a single osteoporotic fracture
the second fracture to that of single fracture patients, t test was for the analysis. Of those patients, the fracture treatment costs
used for the total medical costs and Wilcoxon ranked sum test were incurred among 12,898 patients. These 12,898 patients
was used for the fracture treatment costs, as it was expected were included for the later analysis because the remaining
that the fracture treatment costs were skewed compared with 1103 patients were assumed to be those who did not receive
the total medical costs. fracture treatments at the enrolled hospitals but may have re-
Comorbidity profiles of the patients were also gathered to ceived at different hospitals. While there was little difference
use as explanatory variables. Both highly frequent comorbid- in gender proportion, age group, and fracture type between
ities and diseases associated with osteoporosis risk were in- these two datasets, total costs were generally higher among
cluded. To extract highly frequent comorbidities, prevalence dataset of those 12,898 patients. Their characteristics are
rates of comorbidity within 3 months before and after index shown in Table 1. It was found that 10,755 patients (83%)
fracture were calculated for each of three cost groups (A: top were female patients, and the patients aged 70–89 consisted
30% tiles; B: middle 40% tiles; and C: bottom 30% tiles ac- large proportion (male 80%, female 76%). Although the treat-
cording to the total medical costs). The following two criterion ment costs incurred were more expensive in male than in
were used to determine the comorbidities included in the later female regardless of pre- or post-fracture, the total medical
analysis: (1) If prevalence rates were more than 30% for all the costs were higher in female than in male.
three groups or (2) a prevalence rate in the group A was more Only female patients with the second vertebral fractures
than 20%. Medical conditions with M45–49 in ICD-10 code were reported in this study, since male patients with the sec-
(spondylopathies) were excluded from the list, as it was as- ond fractures and female patients with the second fractures
sumed to be highly correlated with vertebral fractures. To that were not vertebral fractures were very few. For this rea-
extract the comorbidities associated with osteoporosis risk, son, in order to analyze the impact of the second fractures on
those described as major risk factors in the guidelines for individual treatment costs, 100 female patients with the sec-
prevention and treatment of osteoporosis were also included. ond vertebral fractures alone were included in the comparative
To identify the cost drivers for fracture treatment costs analysis of fracture costs between patients with single fracture
based on patient profiles (demographic factors, fracture type, and those with the second fracture (in the next section). Their
treatment period reflecting both inpatient stay and outpatient mean age was 78.9, which was slightly higher than the mean
visit, and comorbidities), a generalized linear model (GLM) age of female patients with single vertebral fracture (77.6 [not
with gamma family with log link function was applied for reported in the table]).
patients with single fracture. The increased or decreased asso-
ciation was further examined by conducting a subgroup anal- Cost profiles by fracture site
ysis of the patients with one type of fracture (hip or vertebral).
In this subgroup analysis, the driving factors for the total med- Treatment costs incurred were stratified by gender and fracture
ical costs were also tested by a GLM with Gaussian family site as shown in Table 2. Significantly higher total medical costs
and identity link function. In this regression analysis with the were observed in hip fracture patients of both genders (male
45 Page 4 of 12 Arch Osteoporos (2018) 13: 45

Table 1 Patient characteristics at

index fracture Total population with single fracture Multiple vertebral
and fracture treatment cost > 0 fracture patients

Male Female Female

n = 2143 n = 10,755 n = 100

Age
Mean (SD) 77.7 (8.4) 77.6 (8.7) 78.9 (8.6)
50–59, n (%) 73 (3%) 311 (3%) 3 (3%)
60–69, n (%) 261 (12%) 1589 (15%) 8 (8%)
70–79, n (%) 830 (39%) 4050 (38%) 36 (36%)
80–89, n (%) 877 (41%) 4038 (38%) 46 (46%)
90-, n (%) 102 (5%) 767 (7%) 7 (7%)
Fracture type, n (%)
Hip 341 (16%) 2074 (19%) –
Vertebral 1407 (66%) 5304 (49%) –
NHNV 395 (18%) 3377 (31%) –
Treatment period, mean (SD) 9.1 (4.1) 9.3 (4.0)
Medical cost (USD), mean (SD)
1 year before (total fracture) 8200 (13,971) 4310 (7867) –
1 year after (total fracture) 14,023 (15,197) 10,557 (10,852) –
Total medical cost (difference) 5823 (8898) 6246 (8522) –
Fracture treatment cost (USD), mean (SD)
1 year after (total fracture) 1579 (2577) 2004 (2968) –

$14,592 [95% CI $13,356 to $15,829], and female $15,691 Similar result was observed in the fracture treatment costs.
[95% CI $15,250 to $16,132]) compared to other two types The significantly higher costs were observed in hip (male
of fractures. Cost estimates for vertebral and NHNV fractures $4506 [95%CI $4093 to $4920]; female $5427 [95%CI
were not largely different (male-vertebral: $4268 [95% CI $5253 to $5601]) and no large difference was observed be-
$3891 to $4645]; female-vertebral: $3819 [95% CI $3661 to tween the vertebral and NHNV fracture patients’ costs.
$3977]; male-NHNV: $3790 [95% CI $3094 to $4486]; fe- Among hip fracture patients, the fracture treatment costs were
male-NHNV: $4259 [95% CI, $4033 to $4484]). higher in female.

Table 2 Cost profiles among analyzed population by fracture site

Total population with Multiple vertebral

fracture treatment cost > 0 fracture patients

Male Female Female

n = 2143 n = 10,755 n = 100
Mean (95% CI) Mean (95% CI) Mean (95% CI)

Total medical cost (USD)

Hip 14,592 (13,356 to 15,829) 15,691 (15,250 to 16,132) –
Vertebral 4268 (3891 to 4645) 3819 (3661 to 3977) –
NHNV 3790 (3094 to 4486) 4259 (4033 to 4484) –
Multiple vertebral – – 7651 (3939 to 11,364)
Fracture treatment cost (USD)
Hip 4506 5427 (5253 to 5601) –
(4093 to 4920)
Vertebral 1022 (929 to 1114) 1044 (996 to 1092) –
NHNV 1035 (862 to 1208) 1408 (1340 to 1476) –
Multiple vertebral – – 2458 (1763 to 3153)
Arch Osteoporos (2018) 13: 45 Page 5 of 12 45

When comparing the costs of the second vertebral fracture kidney disease, thyrotoxicosis, hypofunction and other disor-
female patients with the costs of the single vertebral fracture ders of the pituitary gland, primary hyperparathyroidism, and
female patients, the second fracture costs were significantly Cushing’s syndrome) were included. COPD was most preva-
higher in both the total medical costs and the fracture treat- lent among those 6 medical conditions but less common com-
ment costs (for the total medical costs, mean was $7651 in the pared to the 12 medical conditions. Prevalence rates of 4 out
second vertebral fracture and $3819 in the single vertebral of the 6 medical conditions (thyrotoxicosis, hypofunction and
fracture; p < 0.05. For the fracture treatment costs, mean was other disorders of the pituitary gland, primary hyperparathy-
$2458 in the second vertebral fracture and $1044 in the single roidism, and Cushing’s syndrome) were very low among stud-
vertebral fracture; p < 0.01). ied patients (less than 2%).

Comorbidity profiles for osteoporotic fracture Cost drivers for the fracture treatment cost
patients
The result of regression analysis seeking drivers for the frac-
Comorbidity profiles of the studied population is shown in ture treatment cost is presented in Table 4. Among demo-
Table 3. Twelve comorbidities (GI diseases, hypertension, graphic factors, both gender and age were significantly asso-
other functional intestinal disorder, dorsalgia, disorder of lipo- ciated with the fracture treatment costs. Female had 1.08 times
protein metabolism and other lipidemia, sleep disorders, dia- higher fracture treatment costs (95% CI 1.00 to 1.16) com-
betes, osteoarthritis of knee, angina pectoris, heart failure, dis- pared with male, and those aged 60–69, 70–79, and 80–89 had
order of refraction and accommodation, and iron deficiency higher fracture treatment costs compared with those aged 50–
anemia) met with the eligible criteria. The lists of comorbidi- 59 (60–69: 1.20 [95% CI 1.01 to 1.42], 70–79: 1.26 [95% CI
ties selected based on the criteria were not different between 1.07 to 1.48], and 80–89: 1.26 [95% CI 1.07 to 1.48]) after
male and female. adjusting for other covariates (treatment duration [in month],
In addition, from the guidelines for prevention and treat- fracture category [hip, vertebral, and NHNV], and comorbid-
ment of osteoporosis, 6 medical conditions (COPD, chronic ities). In terms of treatment status among patients, a longer

Table 3 Comorbidity profiles

among analyzed population Comorbidity ICD10 Total population with fracture
treatment cost > 0

Male Female

n = 2143 (%) n = 10,755 (%)

GI diseasesa K21,K25, K29 1498 (70%) 7165 (67%)

Hypertension I10-I15 1200 (56%) 5567 (52%)
Other functional intestinal disorder K59 1018 (47%) 4389 (41%)
Dorsalgia M54 834 (39%) 3846 (36%)
Disorders of lipoprotein metabolism E78 585 (27%) 3649 (34%)
and other lipidemia
Sleep disorders G47 649 (30%) 3406 (32%)
Diabetes E10-E14 743 (35%) 2853 (27%)
Osteoarthritis of knee M17 260 (12%) 2689 (25%)
Angina pectoris I20 486 (23%) 1855 (17%)
Heart failure I50 496 (23%) 1801 (17%)
Disorder of refraction and H52 362 (17%) 1776 (17%)
accommodation
Iron deficiency anemia D50 359 (17%) 1554 (14%)
COPD J40–44, J47 402 (19%) 1127 (10%)
Chronic kidney disease N18 241 (11%) 396 (4%)
Thyrotoxicosis E05 28 (1%) 172 (2%)
Hypofunction and other disorders E23 9 (0.42%) 24 (0.22%)
of the pituitary gland
Primary hyperparathyroidism E210 2 (0.09%) 7 (0.07%)
Cushing’s syndrome E24 0 (0%) 4 (0.04%)
a
Gastrointestinal (GI) diseases includes gastro-esophageal reflux disease, gastric ulcer, gastritis, and duodenitis
45 Page 6 of 12 Arch Osteoporos (2018) 13: 45

Table 4 Cost drivers for the

fracture treatment cost by Coefficients 95% CI 95% CI P value
multivariable regression analysis a Lower Upper

Gender (base = male) 1.08 1.00 1.16 < 0.05

Age (base = 50–59)
60–69 1.20 1.01 1.42 < 0.05
70–79 1.26 1.07 1.48 < 0.01
80–89 1.26 1.07 1.48 < 0.01
90- 1.12 0.93 1.35
Treatment period (month) 1.05 1.05 1.06 < 0.01
Fracture category (base = hip)
Vertebral 0.20 0.19 0.22 < 0.01
NHNV 0.26 0.24 0.28 < 0.01
Comorbidities
GI diseases 1.01 0.95 1.08 0.66
Hypertension 1.13 1.06 1.20 < 0.01
Other functional intestinal disorder 1.03 0.97 1.09 0.40
Dorsalgia 0.90 0.85 0.96 < 0.01
Disorders of lipoprotein metabolism 1.00 0.94 1.06 0.98
and other lipidemia
Sleep disorders 1.03 0.97 1.10 0.29
Diabetes 0.97 0.91 1.04 0.38
Osteoarthritis of knee 1.01 0.94 1.07 0.83
Angina pectoris 1.05 0.98 1.13 0.20
Heart failure 0.95 0.88 1.03 0.19
Disorder of refraction and 0.86 0.80 0.93 < 0.01
accommodation
Iron deficiency anemia 1.15 1.07 1.25 < 0.01
COPD 0.91 0.84 1.00 < 0.05
Chronic kidney disease 0.93 0.82 1.05 0.24
Thyrotoxicosis 1.11 0.91 1.39 0.32
Hypofunction and other disorders 1.09 0.68 1.94 0.73
of the pituitary gland
Primary hyperparathyroidism 2.08 0.88 6.81 0.15
Cushing’s syndrome 0.46 0.14 3.27 0.30
a
n = 12,898

treatment period was associated with 1.05 times higher frac- An association was further investigated after limiting the
ture treatment costs (95% CI 1.05 to 1.06). Non-hip fracture type of patients to hip fracture and vertebral fracture patients
patients had significantly lower fracture treatment costs than (Tables 5). With regard to demographic factors, only the gen-
hip fracture patients (vertebral 0.20 [95% CI 0.19 to 0.22] and der and the age of 70–79 years old retained association with
NHNV 0.26 [95% CI 0.24 to 0.28]). the increased fracture treatment costs among hip fracture pa-
Among comorbidities we identified, associations with the tients. An association with the treatment period was observed
increased fracture treatment costs were suggested for hyper- in both hip and vertebral fractures. Among comorbidities driv-
tension and iron deficiency anemia (hypertension 1.13 [95% ing the fracture treatment cost in Table 4, only iron deficiency
CI 1.06 to 1.20] and iron deficiency anemia 1.15 [95% CI 1.07 anemia was associated with increased fracture treatment costs
to 1.25]). On the contrary, associations with the decreased in both hip and vertebral fractures. Associations between co-
fracture treatment costs were observed for dorsalgia (0.90 morbidities and change in the fracture treatment costs other
[95% CI 0.85 to 0.96]), disorder of refraction and accommo- than disorder of refraction and accommodation were observed
dation (0.86 [95% CI 0.80 to 0.93]), and COPD (0.91 [95% CI only in the hip fracture patient group. In addition, associations
0.84 to 1.00]). between the fracture treatment costs and two comorbidities
Arch Osteoporos (2018) 13: 45 Page 7 of 12 45

Table 5 Cost drivers for the fracture treatment cost in hip and vertebral fracture population by multivariable regression analysis

Hip fracture patients (n = 2415) Vertebral fracture patients (n = 6711)

Coefficient 95% CI 95% CI P value Coefficient 95% CI 95% CI P value

Lower Upper Lower Upper

Gender (base = male) 1.20 1.09 1.31 < 0.01 0.97 0.87 1.07 0.55
Age (base = 50–59)
60–69 1.24 0.99 1.54 0.05 0.98 0.71 1.33 0.90
70–79 1.30 1.05 1.58 < 0.05 1.09 0.80 1.45 0.56
80–89 1.17 0.95 1.43 0.12 1.13 0.83 1.51 0.41
90- 0.99 0.79 1.22 0.91 1.09 0.77 1.53 0.62
Treatment period (month) 1.01 1.01 1.02 < 0.01 1.07 1.06 1.08 < 0.01
Comorbidities
GI diseases 1.00 0.93 1.07 0.93 1.02 0.93 1.13 0.64
Hypertension 1.11 1.04 1.20 <0.01 1.04 0.95 1.15 0.39
Other functional intestinal disorder 1.02 0.95 1.09 0.60 1.03 0.94 1.13 0.49
Dorsalgia 0.93 0.86 1.00 <0.05 0.93 0.85 1.01 0.08
Disorders of lipoprotein metabolism 1.07 0.99 1.15 0.07 1.05 0.95 1.16 0.34
and other lipidemia
Sleep disorders 1.11 1.04 1.19 < 0.01 1.01 0.91 1.11 0.89
Diabetes 0.92 0.86 0.99 < 0.05 0.95 0.86 1.06 0.33
Osteoarthritis of knee 1.05 0.97 1.14 0.24 1.11 1.00 1.23 0.06
Angina pectoris 1.08 1.00 1.17 0.06 1.06 0.94 1.20 0.33
Heart failure 0.96 0.89 1.05 0.37 0.94 0.83 1.07 0.35
Disorder of refraction and 0.99 0.90 1.09 0.83 0.82 0.74 0.92 < 0.01
accommodation
Iron deficiency anemia 1.11 1.03 1.20 < 0.01 1.19 1.04 1.36 < 0.05
COPD 0.89 0.80 0.98 < 0.05 0.93 0.82 1.06 0.26
Chronic kidney disease 0.95 0.84 1.08 0.44 0.95 0.77 1.20 0.68
Thyrotoxicosis 1.03 0.80 1.37 0.82 1.24 0.89 1.79 0.23
Hypofunction and other disorders 1.18 0.67 2.35 0.59 0.95 0.46 2.41 0.90
of the pituitary gland
Primary hyperparathyroidism 1.84 0.74 6.67 0.26 0.84 0.09 690.89 0.92
Cushing’s syndrome 0.45 0.13 3.24 0.29 0.48 0.08 22.56 0.55

were newly suggested among the hip fracture patient group. In terms of comorbidities, associations between the in-
Sleep disorders were associated with the increased fracture creased total medical costs and three medical conditions (hy-
treatment costs (1.11 [95% CI 1.04 to 1.19]), whereas diabetes pertension, sleep disorders, and iron deficiency anemia) were
was associated with the decreased fracture treatment costs observed as well (hypertension: $1972 [95% CI $1065 to
(0.92 [95% CI 0.86 to 0.99]). $2879] for hip and $1306 [95% CI: $963 to $1649] for verte-
bral fracture, sleep disorders: $1444 [95% CI: $585 to $2302]
Cost drivers for the total medical cost in hip for hip and $474 [95% CI: $133 to $815] for vertebral frac-
and vertebral fractures ture, and iron deficiency anemia: $2126 [95% CI: $1153 to
$3099] for hip). In contrast to the previous analysis, an asso-
Drivers for the total medical costs were also explored in the ciation between dorsalgia and the decreased total medical
hip and vertebral fracture patient groups (Table 6). It was costs was observed only among the vertebral fracture patient
found that the total medical costs of female vertebral fracture group (−$426 [95% CI: −$724 to −$127]). In this analysis,
were less than male vertebral fracture patients (−$655 [95% associations between two new comorbidities (other functional
CI −$1018 to − 292]). The association between age and the intestinal disorder and chronic kidney disease) and the in-
increased total medical costs was observed only in vertebral creased total medical costs were observed (other functional
fracture patients (aged 80–89: $1084 [95% CI $47 to $2120] intestinal disease: $1644 [95% CI: $798 to $2490] for hip,
and aged 90-: $1460 [95% CI $267 to $2652]). $1436 [95% CI: $1114 to $1758] for vertebral fracture, and
45 Page 8 of 12 Arch Osteoporos (2018) 13: 45

Table 6 Cost drivers for the total medical cost in hip and vertebral fracture population by multivariable regression analysis

Hip fracture patients (n = 2415) Vertebral fracture patients (n = 6711)

Coefficients 95% CI 95% CI P value Coefficients 95% CI 95% CI P value

Lower Upper Lower Upper

Gender (base = male) 15 − 1146 1176 0.98 − 655 − 1018 − 292 < 0.01
Age (base = 50–59)
60–69 42 − 2770 2853 0.98 332 − 748 1412 0.55
70–79 49 − 2551 2649 0.97 542 − 490 1575 0.30
80–89 746 − 1832 3324 0.57 1084 47 2120 < 0.05
90- 153 − 2593 2898 0.91 1460 267 2652 < 0.05
Treatment period (month) 194 95 293 < 0.01 −2 − 38 33 0.90
1 year cost before fracture − 0.003 − 0.004 − 0.003 < 0.01 − 0.001 − 0.002 − 0.001 < 0.01
Comorbidities
GI diseases 174 − 732 1081 0.71 49 − 289 386 0.78
Hypertension 1972 1065 2879 < 0.01 1306 963 1649 < 0.01
Other functional intestinal disorder 1644 798 2490 < 0.01 1436 1114 1758 < 0.01
Dorsalgia 169 − 777 1115 0.73 − 426 − 724 − 127 < 0.01
Disorders of lipoprotein metabolism 581 − 340 1501 0.22 − 34 − 387 318 0.85
and other lipidemia
Sleep disorders 1444 585 2302 < 0.01 474 133 815 < 0.01
Diabetes 557 − 368 1482 0.24 −5 − 366 356 0.98
Osteoarthritis of knee 551 − 478 1580 0.29 − 278 − 640 85 0.13
Angina pectoris 358 − 670 1387 0.49 − 157 − 582 268 0.47
Heart failure − 131 − 1171 909 0.81 386 − 43 815 0.08
Disorder of refraction and 773 − 431 1977 0.21 − 320 − 708 68 0.11
accommodation
Iron deficiency anemia 2126 1153 3099 < 0.01 − 12 − 497 473 0.96
COPD − 898 − 2210 415 0.18 58 − 389 505 0.80
Chronic kidney disease 4196 2441 5950 < 0.01 1285 483 2086 < 0.01
Thyrotoxicosis 1081 − 2400 4562 0.54 745 − 455 1944 0.22
Hypofunction and other disorders 2426 − 5459 10,312 0.55 − 171 − 2989 2646 0.91
of the pituitary gland
Primary hyperparathyroidism 5341 − 8285 18,966 0.44 − 3127 − 14,867 8612 0.60
Cushing’s syndrome − 9072 − 28,286 10,141 0.35 5036 − 3419 13,490 0.24

chronic kidney disease: $4196 [95% CI $2441 to $5950] for inpatients and outpatients, thus considered to be adequately
hip, $1285 [95% CI: $483 to $2086] for vertebral fracture). representing Japanese population despite limitation of the da-
tabase itself. Kondo et al. (2009) investigated the treatment
costs of 41 hip fracture patients in three hospitals in Japan
Discussion [13]. They reported that the costs were $19,034 on average.
Although their result was slightly more expensive than what
This study estimated the total medical costs and the fracture our study has found, it would be mainly due to the different
treatment costs in osteoporotic patients in Japan. We also ex- payment system used by the hospital. The hospitals studied in
plored the common comorbidities among osteoporotic frac- this study were using only DPC payment system (similar to
ture patients, and examined the potential cost drivers for the Diagnostic Related Group [DRG] in US) whereas the study
treatment costs. by Kondo et al. included the hospitals with non-DPC payment
Although a few studies have reported the treatment costs system. DPC payment hospitals tend to manage their resource
incurred for osteoporotic fracture in Japan, the number of pa- more effectively, and therefore it may lead to the lower treat-
tients was limited and the hospitals from which the patients ment cost estimates. In contrast, Hagiwara et al. (2000) found
were recruited was underrepresented. Our cost estimation was that the mean treatment costs of hip fractures with 101 patients
based on the latest claim data of 12,898 patients including both who were 60 years and older were 1,470,000 JPY ($13,364
Arch Osteoporos (2018) 13: 45 Page 9 of 12 45

when 1 USD = 110 JPY) [14]. This is less than what we have (hypertension, iron deficiency anemia, and diabetes) were
estimated, probably because their estimate was based on the consistent with previous research. Firstly, in our study, it was
inpatient stay in one hospital and did not include outpatient or found that hypertension was associated with increased fracture
rehabilitation care after discharge. Furthermore, it was reported treatment costs in hip fracture subgroups. This positive asso-
that an average treatment costs for vertebral fracture was ciation was persistent when investigated in the total popula-
780,000 JPY ($7091 when 1 USD = 110 JPY) with 272 hospi- tion. The association between hypertension and bone remod-
talized patients aged 60 years and older [15]. This can also be eling has been suggested [3], and one previous study showed
explained by the different study populations. Our study in- that the BMD level of female patients with hypertension was
cludes patients who were cared solely by outpatient visits; lower than the BMD level of female patients without hyper-
therefore, our research produced less expensive estimate than tension [21]. Another study further suggested that hyperten-
the research that only recruited hospitalized patients in terms of sion was a predictor for fracture nonunion [22]. Considering
vertebral fracture patient costs. these empirical evidence, our result could additionally indicate
The treatment costs borne by the patients with the second that patients with hypertension have worse bone turnover, and
fracture were significantly higher compared with single frac- it takes additional care to heal, leading to higher fracture treat-
ture patient, which is consistent with previous studies in ment costs. This association was consistent with the total med-
USA and UK [16–18]. In this study, those with the second ical costs among hip and vertebral fracture patients.
vertebral fracture bore more than twice of total medical costs Secondly, the result showed a strong positive association
and fracture treatment costs when comparing their point es- between iron deficiency anemia and treatment costs (both in
timates with those with a single vertebral fracture. This does the fracture treatment costs and the total medical costs among
not only suggest the importance of preventing the second hip fracture patients). It was reported that several potential
fracture within 1 year from the index fracture but also indi- interactions between iron deficiency and bone loss (i.e., the
cate the great benefit of preventing the second fracture. role of iron in collagen synthesis or in vitamin D metabolism)
From this perspective, prescription of osteoporosis drugs [23]. Thus, it might be claimed that the patient with iron defi-
after the osteoporotic fracture can contribute in reducing ciency anemia bore more treatment costs due to its downside
financial burden of osteoporotic fracture. effect on bone metabolism. However, it should be noted that
The results of cost estimation indicate that total medi- anemia can be an intermediate variable between other comor-
cal costs would be preferable in economic evaluation be- bidities and fractures as well. Anemia can be observed among
cause the fracture treatment costs were almost only one the patients with chronic medical conditions such as CKD (the
third of total medical costs. Other two third would consist detail association between CKD and treatment costs will be
from non-fracture treatment costs such as doctor consul- explained later) [24]. Therefore, the estimated impact of iron
tation costs, comorbidity treatment costs, and inpatient deficiency anemia in this study may reflect the impact of such
specific costs. The amount of these costs are not negligi- comorbidities on treatment costs.
ble, and hence they should be considered when evaluating Lastly, it is interesting to find that diabetes had an associa-
the economic value of fracture treatment. To our best tion with the decreased fracture treatment costs among hip
knowledge, there has been no study in Japan comparing fracture patients. One possible explanation for this negative
the total medical costs and the fracture treatment costs, association is the influence of some antidiabetic drugs.
and the results would support the official guideline for Although it has not been concluded yet, drugs such as insulin
the economic evaluation of drugs/medical devices and metformin have had positive effects such as improved
recommending the inclusion of related comorbidity costs bone bridging or osteogenic effect in some experimental stud-
for cost parameters [19]. ies [25]. Therefore, patients with these antidiabetic drugs can
Interestingly, it was found that more fracture treatment have better bone quality and need less care. However, this
costs were incurred among female hip fracture patients where- should be further examined carefully as mixed results exist
as less total medical cost was incurred female vertebral frac- in this research area. For example, one recent study showed
ture patients, suggesting that more non-fracture (comorbidity) that the insulin use was negatively associated with bone min-
treatment costs were incurred among male than female. Male eral density increase [26]. In addition, this could be explained
could be more likely to worsen their comorbidities. In fact, by behavioral aspect. One previous study in Japan which com-
one study reported higher relative hazards of all-cause mortal- pared the healthcare resource use between osteoporosis dia-
ity after hip fracture in male during 0–1 year age interval [20]. betic patients and non-diabetic patients found that BMD test-
Male may be more vulnerable after fractures, which lead to ing was less frequently used among the patients with diabetes
increasing non-fracture treatment cost. [27]. Although the reason behind this is unclear, our study
Associations between comorbidities and treatment costs results may indicate similar phenomenon.
were explored, and nine comorbidities were detected as cost Unpredicted positive associations were also found with
drivers. The associations with three medical conditions three medical conditions (sleep disorders, CKD, and other
45 Page 10 of 12 Arch Osteoporos (2018) 13: 45

functional intestinal diseases). Sleep disorders were positively patients and total patient population, even after adjusting
associated with both the fracture treatment costs among hip the demographic factors. One possible reason is that this
fracture patients and the total medical costs among hip and variable may have worked as intermediate variable and
vertebral fracture patients. Sleep disorders are common claims reflected severity of other comorbidities such as diabetes,
among the elderly population. In a previous study, it had po- which is significant risk factor for retinopathy. The third
tential associations not only with increasing risks of falls but one is COPD. It was demonstrated that COPD was asso-
also with osteoporosis [28], suggesting that sleep disorders ciated with decrease in the fracture treatment costs among
may have some impact on bone metabolism. For example, hip fracture patients. It was preserved even after the pop-
Amstrap et al. (2015) showed that melatonin, the hormone ulation was extended to total population. Given that the
regulating the sleep, improved BMD among women with os- COPD is a well-known risk factor of osteoporosis and low
teoporosis [29]. Given this, our study may indicate that sleep BMD [3], this result looks counterintuitive. One potential
disorders cause malignant bone metabolism among osteopo- association is that COPD patients often develop vertebral
rotic fracture patients, and consequently patients bear more fractures, and therefore they could have already had pre-
total medical costs to recover from the fracture. Moreover, ventive treatments before hip fracture, which would result
patients with CKD may also bear more total medical costs, in improving fracture recovery. However, further research
although it was not observed in our analysis of fracture treat- is necessary.
ment costs. This may be because some osteoporosis treatment The following limitations were recognized in this study.
drugs such as bisphosphonates and painkillers can induce side Firstly, a selection bias due to the nature of the database was
effects such as hypocalcemia. Therefore, monitoring kidney possibly considered. The claim data used in this study were
conditions or treatments for such side effects may increase the collected from acute care hospitals, allowing us to collect se-
total medical costs among CKD patients. verer patients in Japan. On the other hand, these acute care
Lastly, other functional intestinal diseases had a positive hospitals may spend less treatment costs due to DRG-like pay-
association only with the total medical costs. Fracture patients ment system in Japan (DPC). In fact, it was shown that less
would increase the risk of other functional intestinal disorder treatment costs were incurred in DPC hospitals [13], which
because of the physical inactivity during the hospitalization could underestimate total treatment costs. Furthermore, it could
period. Vertebral compression fracture itself can cause consti- create the issue related to loss to follow-up. As there is no
pation as complication [30]. Thus, it could be considered that information regarding the clinical outcome of the patients, it
those with other functional intestinal disorder represented the is impossible to identify whether the patient was recovered,
patient cohort with severe fractures and longer inpatient stay, transferred to another care institution, or died. The Japanese
which made their total medical costs higher than the costs of government has introduced the policy that promotes speciali-
other patient group. Therefore, there may be no causal asso- zation of medical institution into four categories (advanced
ciation between other functional intestinal disorder and treat- acute-care, acute-care, rehabilitation-care, and long-term care)
ment costs. and collaboration between these hospitals in 2015 [31]. Hence,
Our study also showed unexpected negative associa- it is likely that the total medical costs of patients who were
tions between treatment costs and the other three medical transferred to rehabilitation-care or long-term care institutions
conditions (dorsalgia, disorder of refraction and accom- were not fully captured. In addition, accuracy of diagnosis re-
modation, and COPD). Although the causal association cord could be questioned. One study revealed that the valida-
between these medical conditions and treatment costs tion indicators such as sensitivity or positive predictive value of
needs to be carefully interpreted with further research, diagnosis were varied depending on the medical conditions
potential causal pathways are hypothesized for each co- [32]. In this sense, generalizability of this study might be
morbidity as follows. The first one is dorsalgia. It had a questioned as it is uncertain that the codes used in our study
negative association with the fracture treatment costs in fully captured the osteoporotic fracture population. Finally, our
hip fracture patients and the total medical costs in verte- study could not adjust patient behavioral and SES factors which
bral fracture patients, and the association between could confound the results. Behavioral factors such as smoking
dorsalgia and fracture treatment costs remained in total and SES factor would not only affect patients’ health but also
population. The guidelines for prevention and treatment influence their treatment choice. These limitations should be
and osteoporosis suggest that dorsalgia is one of the fac- considered carefully when interpreting the study results.
tors for diagnosing osteoporosis [3]. General patients with In conclusion, our study reported the treatment costs in-
dorsalgia during pre-fracture period might receive osteo- curred to osteoporotic fracture patients in Japan and examined
porosis treatments earlier and therefore may not experi- the cost drivers by analyzing the associations with comorbid-
ence severe fractures. The second one is disorder of re- ities. To our knowledge, this is the first pragmatic study esti-
fraction and accommodation. It was negatively associated mating the treatment costs of osteoporotic fracture patients
with the fracture treatment costs in vertebral fracture through the analysis of large-sized RWD, and it showed
Arch Osteoporos (2018) 13: 45 Page 11 of 12 45

financial burden of osteoporotic fracture in Japan including 8. Moriwaki K, Mouri M, Hagino H (2017) Cost-effectiveness anal-
ysis of once-yearly injection of zoledronic acid for the treatment of
patients with the second fracture. This study also detected
osteoporosis in Japan. Osteoporos Int 28(6):1939–1950
the novel associations between some comorbidities and the 9. Mori T, Crandall JC, Ganz AD (2017) Cost-effectiveness of
increase or decrease in treatment costs. The findings from this denosumab versus oral alendronate for elderly osteoporotic women
study would help clinicians, researchers, and policy makers in Japan. Osteoporos Int 28(5):1733–1744
understand the potential financial burden of osteoporotic frac- 10. Medical Data Vision Co. (2017) Press release-overview of data-
base-. https://www.mdv.co.jp/press/2017/detail_728.html.
ture and would also support various health economic analyses Accessed 4 September 2017 (in Japanese)
in the future. 11. Terashima H (2016) Guidebook of treatments for surgery. Igaku-
TsushinSha, Tokyo (in Japanese)
Acknowledgements This study was sponsored by Amgen Astellas 12. Terashima H (2016) Guidebook of treatments for clinical practice.
BioPharma K.K. and Astellas Pharma Inc. Igaku-TsushinSha, Tokyo (in Japanese)
13. Kondo A, Zierler KB, Isokawa Y, Hagino H, Ito Y (2009)
Comparison of outcomes and costs after hip fracture surgery in
Compliance with ethical standards three hospitals that have different care systems in Japan. Health
Policy 91(2):204–210
This study was conducted in the form of retrospective database study, and 14. Hagino H, Katagiri H, Okano T, Toyoshima R (2000) Cost survey
for this type of study formal consent is not required. of hip fracture. J Central Jpn Orthop Surg Traumatology 43:881–
882 (in Japanese)
Conflicts of interest Y.T. is a full-time employee of Amgen Astellas 15. Hagino H, Kondo A, Okieki M (2009) Medical costs for the treat-
BioPharma K.K., and a councilor of ISPOR Japan. ment of osteoporotic fractures. The Bone 23(2):47–51 (in Japanese)
Y.I., T.K., and N.A. are full-time employees of Deloitte Tohmatsu 16. Pike TC, Birnbaum GH, Schiller M, Swallow E, Burge TR, Edgell
Consulting LLC. TE (2011) Prevalence and costs of osteoporotic patients with sub-
Deloitte Tohmatsu Consulting LLC receives consulting fees from sequent non-vertebral fractures in the US. Osteoporos Int 22(10):
Amgen Astellas BioPharma K.K. 2611–2621
17. Leal J, Gray MA, Prieto-Alhambra D, Arden KN, Cooper C, Javaid
Open Access This article is distributed under the terms of the Creative KM et al (2016) Impact of hip fracture on hospital care costs: a
Commons Attribution 4.0 International License (http:// population-based study. Osteoporos Int 27(2):549–558
creativecommons.org/licenses/by/4.0/), which permits unrestricted use,
18. Weaver J, Sajjan S, Lewiecki ME, Harris TS, Marvos P (2017)
distribution, and reproduction in any medium, provided you give appro-
Prevalence and cost of subsequent fractures among U.S. patients
priate credit to the original author(s) and the source, provide a link to the
with an incident fracture. J Manag Care Spec Pharm 23(4):461–471
Creative Commons license, and indicate if changes were made.
19. Ministry of Health, Labor, and Welfare. (2016) Guideline for the
Economic Evaluation of Drugs/Medical Devices at Central Social
Insurance Medical Council (Chuikyo). http://www.mhlw.go.jp/file/
05-Shingikai-12404000-Hokenkyoku-Iryouka/0000109789.pdf.
Accessed 20 February 2018 (in Japanese)
References 20. Haentjens P, Magaziner J, Colón-Emeric CS, Vanderschueren D,
Milisen K, Velkeniers B, Boonen S (2010) Meta-analysis: excess
mortality after hip fracture among older women and men. Ann
1. Nishimura A, Akeda K, Kato K, Asanuma K, Yamada T, Uchida A, Intern Med 152(6):380–390
Sudo A (2014) Osteoporosis, vertebral fractures and mortality in a 21. Yang S, Nguyen DN, Center RJ, Eisman AJ, Nguyen VT (2014)
Japanese rural community. Mod Rheumatol 24(5):840–843 Association between hypertension and fragility fracture: a longitu-
2. Hagino H, Nakamura T, Fujiwara S, Oeki M, Okano T, Teshima R dinal study. Osteoporos Int 25(1):97–103
(2009) Sequential change in quality of life for patients with incident 22. Zura R, Braid-Forbes JM, Jeray K, Mehta S, Einhorn AT, Watson
clinical fractures: a prospective study. Osteoporos Int 20(5):695–702 TJ et al (2017) Bone fracture nonunion rate decreases with increas-
3. Committiee for Development Guidelines for Prevention and ing age: a prospective inception cohort study. Bone 95:26–32
Treatment of Osteoporosis. (2015) Guidelines for prevention and 23. Toxqui L, Vanguero PM (2015) Chronic iron deficiency as an
treatment and osteoporosis 2015. Japan Osteoporosis Society, emerging risk factor for osteoporosis: a hypothesis. Nutrients
Japanese Society for Bone and Mineral Research, and Japan 7(4):2324–2344
Osteoporosis Foundation, Tokyo (in Japanese) 24. Yamamoto H (2015) Kidney diseases and metabolic disorders- basics
4. National Institute of Population and Social Security Reserach. and applications required for general physicians. Topics: VIII. Anemia
(2017) Population projection in Japan (2017). http://www.ipss.go. and iron metabolism. J Jpn Soc Int Med 104:960–966 (in Japanese)
jp/pp-zenkoku/j/zenkoku2017/pp29_gaiyou.pdf. Accessed 2 25. Simpson MC, Calori MG, Giannoudis VP (2012) Diabetes and
September 2017 (in Japanese) fracture healing: the skeletal effects of diabetic drugs. Expert Opin
5. Moriwaki K, Kamae I (2009) Socioeconomic aspect of osteoporo- Drug Saf 11(2):215–220
sis: burden of illness research for osteoporosis. Jpn J Geriatr 46: 26. de Waard E, Driessen J, de Jong J, van Geel T, Henry R, van
286–298 (in Japanese) Onzenoort H et al (2017) The association between insulin use and
6. Ding H, Koinuma N, Stevenson M, Ito M, Monma Y (2008) The volumetric bone mineral density, bone micro-architecture and bone
cost-effectiveness of risedronate treatment in Japanese women with strength of the distal radius in patients with type 2 diabetes—the
osteoporosis. J Bone Miner Metab 26(1):34–41 Maastricht study. Bone 101:156–161
7. Moriwaki K, Komaba H, Noto S, Yanagisawa S, Takiguchi T, 27. Sato M, Ye W, Sugihara T, Isaka Y (2016) Fracture risk and
Inoue H, Toujo T, Fukagawa M, Takahashi HE (2013) Cost- healthcare resource utilization and costs among osteoporosis pa-
effectiveness of alendronate for the treatment of osteopenic post- tients with type 2 diabetes mellitus and without diabetes mellitus
menopausal women in Japan. J Bone Miner Res 28(2):395–403 in Japan: retrospective analysis of a hospital claims database. BMC
45 Page 12 of 12 Arch Osteoporos (2018) 13: 45

Musculoskelet Disord. 17(489). https://doi.org/10.1186/s12891- 30. Alexandru D, William S (2012) Evaluation and management of
016-1344-9 vertebral compression fractures. Perm J 16(4):46–51
28. Yen MC, Kuo LC, Lin CM, Lee FC, Lin YK, Lin LC et al (2014) 31. Ministry of Health, Labor, and Welfare. (2016) Meeting material for
Sleep disorders increase the risk of osteoporosis: a nationwide all divisions meeting. http://www.mhlw.go.jp/topics/2016/01/dl/
population-based cohort study. Sleep Med 15(11):1339–1344 tp0115-1-03-01p.pdf. Accessed 5 September 2017 (in Japanese)
29. Amstrup KA, Sikjaer T, Heickendorff L, Mosekilde L, Rejnmark L 32. Yamana H, Horiguchi H, Fushimi K, Yasunaga H (2016)
(2015) Melatonin improves bone mineral density at the femoral Comparison of procedure-based and diagnosis-based identifica-
neck in postmenopausal women with osteopenia: a randomized tions of severe sepsis and disseminated intravascular coagulation
controlled trial. J Pineal Res 59(2):221–229 in administrative data. J Epidemiology 26(10):530–537