Parathyroid surgery is associated with a decreased risk of hip and upper arm fractures in primary hyperparathyroidism: a controlled cohort study


Peter Vestergaard, The Osteoporosis Clinic, Aarhus Amtssygehus, Tage Hansens Gade 2, DK-8000 Aarhus C, Denmark (fax: +45 89 49 76 84; e-mail:


Objectives.  To study if parathyroid surgery reduced fracture risk in patients with primary hyperparathyroidism.

Design.  Controlled cohort study.

Setting.  All subjects diagnosed with primary hyperparathyroidism (ICD8 codes: 252.00, 252.01, 252.03, 252.04, 252.05, 252.08, 252.09, and ICD10 codes: E21.0, E21.2, E21.3) between 1980 and 1999 in Denmark were retrieved from the National Hospital Discharge System.

Subjects.  A total of 3213 subjects (mean age at diagnosis 60.9 ± 16.3 years) were identified of whom 1934 (60%) had parathyroid surgery whilst 1279 (40%) did not. Total follow-up time after diagnosis was 19.565 (median 6.1) years.

Main outcome measure.  Fracture occurrence after diagnosis was identified from the National Hospital Discharge System.

Results.  Conservatively treated patients were older than surgically treated at diagnosis (64.2 ± 17.4 years vs. 58.3 ± 15.2 years, P < 0.01). The mean serum calcium was 2.95 mmol L−1 in surgically treated and 2.74 in conservatively treated in a subgroup from one centre, n = 687). At the time of diagnosis, 172 (9%) of the subsequently surgically treated had had any fracture whilst the same was the case for 176 (14%) of the conservatively treated (P < 0.05). Parathyroid surgery reduced the risk of hip (HR = 0.44, 95% CI: 0.32–0.62) and upper arm (HR = 0.44, 95% CI: 0.27–0.72) fractures after diagnosis whilst the occurrence of other fracture types remained unchanged.

Conclusions.  Parathyroid surgery reduces the occurrence of hip and upper arm fractures with approximately 50% in patients with primary hyperparathyroidism. This may advocate surgery in patients with primary hyperparathyroidism with serum calcium levels in the range presented here. Further studies in patients with mild disease are warranted.


Primary hyperparathyroidism (PHPT) causes skeletal demineralization [1] through an increased skeletal turnover [2]. The decreased bone mineral density (BMD) leads to an increased fracture risk [3]. A decrease in BMD has been reported in the spine [1], the hip [1] and the forearm [1]. Upon surgical treatment for primary hyperparathyroidism, an increase in spine [4, 5], hip [4, 5] and forearm [4–6] BMD has been reported. Only one small randomized study (n = 53) has evaluated the effects of surgery on BMD [7], and found an increase in BMD in surgically, compared with conservatively, treated patients. However, the absolute changes in BMD were small compared with the preoperative period in the above-mentioned studies [4–7], and it remains unclear whether surgery actually would decrease the risk of fractures.

Four studies have been reported on fracture risk in patients with primary hyperparathyroidism. Khosla et al. [8] studied a group consisting mainly of conservatively treated patients (77% of all subjects) and reported a relative risk of hip fractures of 1.4 (95% CI: 1.0–2.0) in patients compared with controls. The same trend was seen in a study by Peacock et al. [9] where the relative risk of hip fractures was 6.7 (95% CI: 4.1–10.9) in a group where 60% were treated conservatively. Larsson et al. [10] reported no increase in hip fracture risk (RR = 0.98, 95% CI: 0.78–1.22) in a group where only 31% were managed conservatively. None of these studies compared surgically and conservatively treated patients. We have previously studied 674 patients who underwent surgery and found a decrease in forearm fractures after surgery and a trend towards a decrease in spine fractures whilst no change in hip fractures were observed [3].

As the disease is relatively rare, it is difficult to accumulate large cohorts for prospective studies of patients with PHPT and to compare patients treated surgically and nonsurgically. Furthermore, it is difficult to perform a randomized controlled study of sufficient size and duration and with fractures as end-point. Data history was collected on a large cohort of patients with PHPT and compared the long-term occurrence of fractures in surgically treated patients with nonsurgically treated patients.

Materials and methods

Denmark offers good possibilities for long-term follow-up of patients due to extensive registration of contacts to the National health system [11]. All contacts to Danish hospitals since 1 January 1977 are registered in a central computer database (the National Hospital Discharge Database –‘Landspatientregistret’) under the National Board of Health using the International Classification of Diseases (ICD) [12]. From the start not all hospitals registered outpatient visits, but gradually more and more did, and from 1 January 1995 all outpatient contacts were included in the register [12]. In 1995, 21% of all fractures were treated on an inpatient basis. Fractures treated on an outpatient basis (e.g. forearm fractures) may not have been included before 1995. For fracture types such as hip fractures who are always treated on an inpatient basis, all incidents are included. A separate analysis on the proportion being hospitalized showed no effect for forearm or upper arm fractures in this cohort. It is thus possible to retrieve data on both diagnosis and hospital admissions for concurrent illnesses in patients with primary hyperparathyroidism. The index date was the date of diagnosis of PHPT.

Each inhabitant is assigned a unique registration code [the CPR (central person register) number] based on his or her date of birth and gender (the system to some extent matches the American Social Security Number). Upon admission to a hospital, the date of admission and discharge is registered along with the ICD code for the disease that led to the contact, and this information is linked to the CPR number [12]. If surgical procedures are performed, the codes for these are also entered into the database [12]. The information is used for administrative purposes and to reimburse the hospitals for the treatments carried out. The hospital system is tax-financed, i.e. it is free of charge for the citizens to use it, and the national hospital system thus covers almost 100% of the hospital based health care in Denmark [12].

The National Hospital Discharge Database was used to retrieve information on patients with a first time diagnosis of primary hyperparathyroidism (ICD8 codes: 252.00, 252.01, 252.03, 252.04, 252.05, 252.08, 252.09, and ICD10 codes: E21.0, E21.2, E21.3) in the period 1 January 1980 through 31 December 1999 (The ICD9 code system was never used in Denmark). This led to the identification of 3225 patients. Twelve (0.4%) were excluded leaving 3213 for analysis. Ten subjects were not Danish citizens and follow-up was thus not possible, in one subject the information on the CPR number was incorrect, and one subject was a Danish national who had emigrated and was operated on a short return to Denmark.

Information on parathyroid surgery performed from 1980 to 1999 from the National Hospital Discharge Database was obtained. We identified 1934 patients (60%) who had undergone surgery and 1279 (40%) who had not. Subsequently, from the same source we obtained information on all hospital contacts that these 3213 patients had had under any fracture diagnosis from 1 January 1977 to 31 December 1999. Our primary end-point of interest was fractures (ICD codes 80099, 80100–80109, 80210–80299, 80399, 80500–80599, 81099, 81199, 81200–81299, 81300–81399, 81400–81409, 81500–81509, 81600–81604, 81899, 82000–82009, 82109–82199, 82300–82399, 82400–82409, 82500–82509, 82600–82609, 82799, S020–S029, S120–S129, S220–S221, S320–S328, S420–S429, S520–S529, S620–S628, S720–S729, S821–S829, S920–S929). We assessed whether any such contact had taken place before or after the date where PHPT was diagnosed. Furthermore, we obtained information on whether the patients had died or emigrated during the period from the Ministry of the Interior using the CPR numbers. Total follow-up time after diagnosis was 19.565 (median 6.1) years. Before diagnosis the observation time was 47.063 (median 15.9) years.

A screening of a random sample of patients (n = 33) revealed that the register-based diagnosis of PHPT was correct in 95% of cases (n = 19) in the surgically treated, whilst 85% of PHPT diagnoses (n = 14) amongst the nonsurgically treated were correct. A screening of a random sample of fracture diagnoses (n = 35) revealed that 97% of register diagnoses were correct compared with hospital records.

Information on plasma calcium and weight of removed pathological parathyroid tissue was available from one of the centres (Aarhus Amtssygehus). From the same centre, data on bone mineral measurements [dual energy X-ray absorptiometry (DEXA) scanning] were available in a subgroup of patients and these data are presented for the convenience of the reader.


Crude odds ratios (OR) were calculated for all outcomes. We used a Cox proportional Hazard model to analyse differences between surgically and nonsurgically treated patients with respect to the specified end-points. We entered age at diagnosis of PHPT (in the categories ≥50 years vs. <50 years), gender (female versus male) and presence of any fracture or not before the time of diagnosis as potential risk factors in the model. Both age and gender are risk factors for fractures and this also applies for a prior fracture, which increases the risk of sustaining an incident fracture [13]. Because a fracture – especially a low-energy fracture – may be regarded as an indication for surgery as a low-energy fracture may signal osteoporosis, we presented occurrence of a fracture or not. To see if age was responsible for any difference in fracture prevalence we did a separate analysis with age as a confounder. We used logistic regression (likelihood ratio method) to evaluate difference in the prevalence of concurrent diseases at the time of diagnosis of PHPT adjusted for age (in the categories ≥50 years vs. <50 years) and gender (female versus male). We computed the models with all variables entered using SPSS 10.0 for Windows (SPSS Inc., Chicago, IL, USA).

Number needed to treat (NNT) was calculated for those fracture sites where the change in fracture risk was significant. This number denotes the number of subjects who will need to undergo surgery to prevent one fracture within a mean observation time of 6.1 years.


Table 1 shows baseline characteristics of the patients. Those who underwent surgery were older and tended to have higher serum calcium than those who were treated conservatively. The deficit in spine BMD was a little higher in surgically treated than the conservatively treated, and the surgically treated tended to be scanned later after diagnosis than the conservatively treated. The median time from diagnosis to surgery was 0.09 year (31 days).

Table 1.  Characteristics of patients at the time of diagnosis stratified by surgery or not. Values are mean and SD
ParameterSurgery (n = 1934)No surgery (n = 1279)Comparison
  1. at-Test for two samples (2p). bχ2 test (p). cData from Aarhus Amtssygehus. dUpper normal limit: 2.56 mmol L−1.

  2. DEXA, dual energy X-ray absorptiometry.

Age (years)58.3 ± 15.264.2 ± 17.4<0.01a
Males500 (26%)293 (23%)0.06b
Females1434 (74%)986 (77%) 
Weight of removed parathyroid tissue (mg)c2349 ± 553 (n = 561)
Plasma calcium (mmol L−1)c,d2.95 ± 0.37 (n = 590)2.74 ± 0.24 (n = 97)<0.01a
Spine Z-scores (DEXA)c−0.47 ± 1.39 (n = 131)−0.15 ± 1.41 (n = 94)0.09a
Mean time from diagnosis to DEXA scan (years)c1.7 ± 3.7 (n = 131)0.8 ± 3.6 (n = 94)0.10a

Table 2 shows fracture occurrence before the time of diagnosis. The surgically treated had a lower prevalence of all types of fractures combined due to a lower prevalence of spine, forearm and femur fractures. Fracture prevalence tended to increase with age. Females had an increased risk of forearm fractures.

Table 2.  Fracture occurrence before diagnosis and differences in fracture occurrence between those who later underwent surgery or not
Fracture site Surgery (n = 1934) No surgery (n = 1279)Crude OR (95% CI) of surgery versus no surgeryaAdjusted OR (95% CI) of surgery versus no surgerybEffect of age (≥50 years vs. <50 years)Effect of gender (female versus male)
  1. aUnadjusted odds ratio. bAdjusted for age (≥50 years vs. <50 years) and gender in a logistic regression.

  2. *P < 0.05.

Skull and face2 (0.1%)5 (0.4%)0.26 (0.06–1.22) –
Spine13 (1%)21 (2%)0.41 (0.21–0.81)*0.43 (0.21–0.86)*2.02 (0.70–5.81)1.36 (0.56–3.33)
Upper arm21 (1%)22 (2%)0.62 (0.34–1.13)0.65 (0.36–1.19)1.82 (0.76–4.36)0.99 (0.48–2.03)
Forearm45 (2%)53 (4%)0.56 (0.38–0.84)*0.59 (0.39–0.88)*2.06 (1.09–3.90)*2.99 (1.50–5.99)*
Hands and fingers11 (1%)10 (1%)0.73 (0.31–1.73)0.71 (0.30–1.67)0.76 (0.29–1.99)0.84 (0.32–2.21)
Femur41 (2%)63 (5%)0.42 (0.28–0.62)*0.45 (0.30–0.67)*5.71 (2.31–14.11)*1.08 (0.66–1.77)
Hip32 (2%)55 (4%)0.37 (0.24–0.57)*0.41 (0.26–0.63)*11.96 (2.93–48.82)*1.19 (0.68–2.06)
Lower leg38 (2%)35 (3%)0.72 (0.45–1.15)0.73 (0.46–1.16)1.53 (0.81–2.88)0.94 (0.54–1.61)
Feet and toes15 (1%)6 (1%)1.66 (0.65–4.24)1.62 (0.62–4.20)1.21 (0.44–3.37)0.35 (0.15–0.85)
All fractures172 (9%)176 (14%)0.61 (0.49–0.76)*0.64 (0.51–0.80)*2.07 (1.49–2.87)*1.14 (0.86–1.50)

Table 3 shows the occurrence of fractures after diagnosis stratified by treatment modality. The surgically treated had a lower occurrence of fracture after diagnosis than the conservatively treated because of a lower occurrence of upper arm and hip and femur fractures. Fracture risk tended to increase with age, female gender and presence of a fracture prior to diagnosis. The risk estimates for any fracture with surgery versus no surgery were approximately identical in the age group above 50 years at diagnosis (unadjusted RR = 0.95, 95% CI: 0.77–1.16, adjusted HR = 0.68, 95% CI: 0.55–0.85 from the Cox analysis) and below 50 years (unadjusted RR = 0.88, 95% CI: 0.55–1.40, adjusted HR = 0.81, 95% CI: 0.49–1.32). The change from a crude OR of 0.90, 95% CI: 0.71–1.11 for any fracture in surgically treated versus conservatively treated to 0.69 (95% CI: 0.56–0.84) after adjustment (Table 3) was due to the fact that men had a larger risk reduction than women and that those with a prior fracture had a higher risk reduction than those without a prior fracture.

Table 3.  Occurrence of fractures (number of subjects with at least one fracture) after diagnosis stratified by treatment modality
Fracture site Surgery (n = 1934)No surgery (n = 1279)Crude OR (95% CI) of surgery versus no surgeryHR (95% CI) of surgery versus no surgeryaEffect of previous fracture (HR)aEffect of age (HR: ≥50 years vs. <50 years)aEffect of gender (HR: female versus male)a
  1. aHazard ratio for surgery compared with no surgery in a Cox model adjusted for age (≥50 years vs. <50 years), gender, and any prior fracture.

  2. *P < 0.05.

  3. The estimates for effect of previous fracture, age and gender have been calculated in a Cox model with all variables including surgery included.

Skull and face3 (0.2%)4 (0.3%)0.50 (0.11–2.15)0.32 (0.07–1.45)2.31 (0.26–20.24)0.24 (0.05–1.13)3.11 (0.36–27.05)
Spine22 (1.1%)7 (0.5%)2.09 (0.90–4.82)1.46 (0.62–3.44)0.85 (0.20–3.60)3.68 (1.10–12.28)*3.84 (0.91–16.25)*
Upper arm31 (1.6%)35 (2.7%)0.58 (0.36–0.94)*0.44 (0.27–0.72)*4.41 (2.57–7.60)*2.09 (1.07–4.10)*0.97 (0.54–1.75)
Forearm64 (3.3%)38 (3.0%)1.12 (0.74–1.68)0.85 (0.57–1.28)2.46 (1.48–4.07)*3.01 (1.63–5.56)*2.39 (1.27–4.49)*
Hands and fingers27 (1.4%)8 (0.6%)2.25 (1.04–4.87)*1.56 (0.71–3.45)1.98 (0.69–5.68)0.59 (0.29–1.18)0.69 (0.34–1.41)
Femur79 (4.1%)84 (6.6%)0.61 (0.44–0.83)*0.50 (0.37–0.68)*3.21 (2.23–4.61)*6.32 (3.20–12.50)*2.04 (1.25–3.35)*
Hip64 (3.3%)77 (6.0%)0.53 (0.38–0.75)*0.44 (0.32–0.62)*2.73 (1.83–4.08)*8.14 (3.56–18.60)*2.32 (1.33–4.04)*
Lower leg35 (1.8%)18 (1.4%)1.29 (0.73–2.29)0.96 (0.54–1.71)4.03 (2.13–7.64)*1.17 (0.61–2.25)1.90 (0.88–4.10)
Feet and toes21 (1.1%)11 (0.9%)1.27 (0.61–2.63)0.89 (0.43–1.86)1.30 (0.39–4.34)0.52 (0.25–1.06)0.88 (0.41–1.89)
All fractures235 (12.2%)170 (13.3%)0.90 (0.73–1.11)0.69 (0.56–0.84)*2.76 (2.14–3.55)*1.90 (1.46–2.48)*1.66 (1.26–2.17)*

Figure 1 shows the effect of surgery on hip fracture occurrence stratified by prior fracture or not. Hip fracture risk was statistically significantly lower in surgically treated than conservatively treated irrespective of whether the patients had sustained a fracture prior to diagnosis or not when analysing with Kaplan–Meier method.

Figure 1.

Kaplan–Meier plot of hip fracture risk in patients stratified by surgery or not and previous fracture or no previous fracture. Comparison of surgically and conservatively treated were performed by log-rank tests.

The number needed to treat was 37 for hip fractures, 88 for upper arm fractures, and 88 for any fracture with a mean observation time of 6.1 years.


In a population-based cohort study, we have demonstrated a decrease in fracture risk in surgically treated patients with primary hyperparathyroidism compared with patients managed conservatively. The main advantage of the study is its size and the long-term follow-up (mean 6.1 years). Furthermore, the study group was population-based, limiting selection bias. Information bias was limited because of the high validity of diagnoses in the register used. For certain fracture types often treated on an outpatient basis such as forearm fractures, some fractures may have not been entered into the register, as outpatient data were first available for all hospitals from 1995. This would not have affected the relative risk estimate if the distribution was equal between surgically and conservatively treated. However, the conservatively treated were older at diagnosis than the surgically treated and this may have overestimated the fracture reducing effect as younger subjects may be less likely to be hospitalized than older subjects. In our data no uneven distribution with age was present. It should be noted that the difference between surgically and conservatively treated was not significant for forearm fractures. An effect was only present for hip fractures, all of which are treated on an inpatient basis and thus not liable to be affected by this type of bias. A significant effect of surgery on upper arm fractures was noted and this may be the result of fewer younger subjects being treated on an inpatient basis.

Spine fractures pose a special problem in this context with regard to detection bias. Many spine fractures are asymptomatic and not clinically apparent [14] as other fracture types – they are thus only diagnosed on X-rays of the spine because of back pain and are often treated on an outpatient basis and therefore not being registered in the register. Furthermore, many cases may be treated as ‘simple’ back pain by general practitioners and never come to the attention of the hospital system and never undergo X-rays.

The main limitation is the fact that the study was not randomized. This may have led to bias in the selection of subjects for surgery or conservative management. This is clearly seen from the fact that those undergoing surgery tended to have more advanced disease judged from the higher serum calcium levels at diagnosis (Table 1). However, this should have led to more fractures before and after diagnosis in the group offered surgery, but such a trend was not seen. Even after adjustment for age, gender and a prior fracture, fewer fractures were seen after diagnosis in surgically than in conservatively managed patients. The higher fracture risk before diagnosis in conservatively (14%) than surgically treated (6%; Table 2) may be due to the higher age at diagnosis of the conservatively treated at diagnosis, but the difference remained after adjustment for age (Table 2). However, the conservatively treated had higher spine BMD than the surgically treated (−0.15 vs. −0.47, Table 1), and this deficit remains unexplained and may signal that those who underwent DEXA scans may have been a selected subgroup.

The present study showed a significant decrease in hip fracture risk in surgically treated in contrast to conservatively managed patient. In a previous large cohort study, Larsson et al. [10] did not find an increase in hip fracture risk in women with primary hyperparathyroidism compared with the general population. In the study by Larsson et al. [10] 69% of the patients underwent surgical treatment and no comparison was made between surgically and nonsurgically treated patients. The lack of a difference between patients and controls in the study by Larsson et al. [10] may thus be due to the fact that the larger part of the patients had parathyroid surgery, which may have decreased their risk of hip fractures. In our study, the number needed to treat was 37 for hip fractures, i.e. it would take 37 surgical procedures to prevent one hip fracture with a mean observation time of 6.1 years. This is a relatively low number for this fracture type and may advocate surgery in patients with primary hyperparathyroidism with a disease in the same stage as our patients. In this context, it should be noted that the severity of the disease was rather high in our patients judged from their high levels of serum calcium and heavy parathyroid adenomas (Table 1) compared with other groups. For example, the mean serum calcium 2.7 mmol L−1 in surgically treated and 2.6 mmol L−1 with a comparable calcium assay in the study by Silverberg et al. [1] versus 2.95 and 2.74 mmol L−1 in our study. In a Swedish series of surgically treated patients, the mean weight of removed parathyroid tissue was 1302 mg [15] vs. 2349 mg in our study. Further studies in patients with mild primary hyperparathyroidism are thus warranted.

No decrease in the risk of spine fractures was seen after diagnosis in our study, and this may be due to the lower spine BMD in surgically than conservatively treated (Table 1). Those selected for parathyroid surgery may have had more advanced bone disease and thus a higher a priori risk of spine fractures which was only partly reverted by the operation. The difference in spine Z-scores in Table 1 was 0.32 between surgically and conservatively treated equalling an expected increase in spine fracture risk of 2.30.32 = 1.3 using the risk estimates presented by Marshall et al. [16]. This was actually close to the estimate of 1.46 for spine fractures after surgery from Table 3 and could indicate that the main reason was indeed preoperative differences in BMD. One should be aware that treatment with antiresorptive drugs may have been initiated in conservatively treated and that we did not have information on use of such drugs.

The 31% decrease in risk of being admitted with any fracture in the present study is a little more than what should be expected from the literature [5–7, 17, 18]. A previous study reported an increase in spine BMD of around 0.5 Z-scores (or standard deviations) after surgery [5]. This would equal a decrease of RR = 1.5−0.5 = 0.82 (a reduction of 18%) in overall fracture risk [16]. However, the same study [5] showed a gain of only 0.1 Z-scores in the femoral neck after surgery equalling a decrease in overall fracture risk of 0.95 (a reduction of 5%), which is less than that suggested by our study and much less than seen for the decrease in hip fractures. This could indicate that the change in BMD did not completely mirror the changes in fracture risk much in the same way that a much larger reduction in fracture risk has been seen with bisphosphonates for osteoporosis treatment than what should be expected from BMD changes [14]. This may mean that the increased bone turnover by itself causes some of the increase in fracture risk [19].

The lag-time of 31 days from diagnosis to surgery was too short to affect the results regarding differences in fracture occurrence between surgically and conservatively treated patients.

In conclusion, parathyroid surgery seems to decrease the overall risk of fractures and reduce the risk of hip and upper arm fractures by approximately 50% in subjects with advanced primary hyperparathyroidism.

Conflict of interest statement

No conflict of interest was declared.


Flemming Melsen MD DMSc, Ib Hessov MD DMSc, Peer Christiansen MD DMSc and Jens Peter Garne MD are acknowledged for providing data on weight of removed pathological parathyroid tissue.