Identification of Risk Factors in the Development of Heterotopic Ossification After Primary Total Hip Arthroplasty.

PURPOSE: Heterotopic ossification (HO) is a process by which bone forms abnormally in soft tissues. Known risk factors for developing HO include male sex, spinal cord injury, trauma, and surgery. We investigated additional risk factors in the development of HO after hip arthroplasty.
METHODS: We performed a retrospective review of electronic medical records of 4070 individuals who underwent hip arthroplasty from September 2010 to October 2019 at the University of California, San Francisco Hospital. Demographics, anthropometrics, medications, and comorbid conditions were used in logistic regression analysis to identify factors associated with the development of HO.
RESULTS: A total of 2541 patients underwent primary hip arthroplasty in the analyzed timeframe (46.04% men, mean age at procedure: 62.13 ± 13.29 years). The incidence of postsurgical HO was 3% (n = 80). A larger proportion of individuals who developed HO had underlying osteoporosis (P < 0.001), vitamin D deficiency (P < 0.001), spine disease (P < 0.001), type 1 or 2 diabetes (P < 0.001), amenorrhea (P = 0.037), postmenopausal status (P < 0.001), parathyroid disorders (P = 0.011), and history of pathologic fracture (P = 0.005). Significant predictors for HO development were Black/African American race [odds ratio (OR) 2.97, P = 0.005], preexisting osteoporosis (OR 2.72, P = 0.001), spine disease (OR 2.04, P = 0.036), and low estrogen states (OR 1.99, P = 0.025). In the overall group, 75.64% received perioperative nonsteroidal anti-inflammatory drugs (NSAIDs), which negatively correlated with HO formation (OR 0.39, P = 0.001).
CONCLUSIONS: We identified new factors potentially associated with an increased risk of developing HO after primary hip arthroplasty, including African American race, osteoporosis, and low estrogen states. These patients may benefit from HO prophylaxis, such as perioperative NSAIDs.

Heterotopic ossification (HO) is the formation of bone within normally unmineralized tissues and can occur after traumatic injuries, surgeries, burns, and neurologic damage (1-4). HO has been reported in rheumatic diseases (ie, dermatomyositis, scleroderma, ankylosing spondylitis) and after major trauma such as hip fracture repair or spinal surgeries (3, 5, 6). Although several forms of HO have a genetic predisposition (7-9), nongenetic forms of HO are more common and appear to have similarities in their origin (3, 10). Nongenetic forms of HO also have a wide age distribution, although they typically occur in adults in their second and third decades after a history of trauma or repetitive mechanical stress (11). The HO can range from being clinically insignificant to resulting in chronic pain and significant morbidity (11, 12).

The reported incidence of HO varies depending on the inciting event and patient population. Meta-analyses, retrospective reviews, and randomized, controlled trials show a high degree of variability in HO incidence by mode of injury, with average rates ranging from 10% to 53% in individuals with a history of traumatic brain or spinal cord injury (13-17), 4% to 20% in those with a history of burns (20-60% in severe or third-degree burns) (11, 12, 18-20), up to 65% in high energy extremity trauma or combat injury (11, 12), and 57% to 90% in traumatic amputations (21-23). The incidence of HO after hip arthroplasty has been reported to be from 2% to 40% (11, 24-27). The incidence varies depending on specific factors such as indication for surgery, surgical approach, and threshold for classification of significant HO. Current known risk factors for the development of HO after hip surgery include male sex, prior hip surgery, history of HO, preceding trauma, soft tissue injury, preexisting inflammatory conditions, vascular diseases, spinal cord injury, and bone-forming diseases such as ankylosing spondylitis (10, 28-31). Surgical approach, prosthesis type, and use of trochanteric or femoral osteotomy also appear to affect the risk of HO formation (32, 33). Anterolateral and lateral approaches to hip arthroplasty have been shown to increase the risk of HO formation in retrospective studies (34, 35). HO formation is a common complication following operative repair of acetabular fractures, particularly those treated with posterior, combined, and extensile approaches, and is associated with poor functional recovery (36-38).

Studies evaluating risk factors for HO formation have largely focused on a subset of related or similar independent predictor variables such as inciting events for HO formation or surgical techniques. There are few studies investigating comorbid conditions and HO formation. The aim of this study was to identify and expand the range of potential risk factors potentially contributing to HO development after hip arthroplasty in a large, single-center, retrospective cohort.

Methods

Patient Selection

This study was approved by the University of California, San Francisco (UCSF) Institutional Review Board (IRB 10-03053). A retrospective review of medical records was performed in individuals who underwent any hip surgery, which included primary arthroplasty, revision hemiarthroplasties, or acetabular repair surgeries at UCSF (CPT 27130, 27076, 27137, 27220, 27222, 27226, 27227, 27228, 27254, 29915) from September 1, 2010 to October 31, 2019 (Fig. 1). Data were extracted with the assistance of the Clinical and Translational Science Institute at UCSF. Only patients with complete operative medical records and preoperative and postoperative radiographs (occurring within 1 year of surgery) at UCSF were included. Exclusion criteria included incomplete medical records, postoperative follow-up outside of our electronic medical record (EMR) system, and pediatric cases. The American Academy of Pediatrics age cutoff of 21 years was used for exclusion of pediatric cases. Purely arthroscopic cases were excluded as these cases were predicted to have a higher number of unadjusted confounders since individuals undergoing arthroscopic procedures are generally younger and routinely receive HO prophylaxis with nonsteroidal anti-inflammatory drugs (NSAIDs). Due to the limited number of revision arthroplasties and hemiarthroplasties, these surgeries were also excluded.

Figure 1.

Selection criteria flowchart.*Due to overlap between types of hip surgery in an individual patient, acetabular repair, hemiarthroplasty, revision arthroplasty, arthroscopy, and repeated total hip arthroplasty are included together.

The initial data extraction included 4070 unique patients that underwent ≥1 hip procedures during the specified time period, for a total of 4729 surgical cases. Of these patients, 2 were excluded due to missing data that were not transferred from the old medical records system to the new record database, which occurred in 2012. A further 120 patients were excluded as there were no imaging data available. There were 282 pediatric patients who underwent hip surgeries, and these patients were excluded. A further 1125 patients underwent hemiarthroplasty, revision arthroplasty, acetabular surgery, or arthroscopy or had multiple procedures under the billing code for total hip arthroplasty, and these patients were excluded. Patients who underwent more than one primary total hip arthroplasty were excluded since lateralization was not consistently coded in the EMR, and we could not confirm that the repeat procedures were performed on the contralateral hip rather than being revision surgeries that were incorrectly coded as primary hip arthroplasties. After the previously described exclusion criteria was met, there were 2541 patients who underwent single total hip arthroplasty (Fig. 1).

Hospitalization data that was collected included sex, race and ethnicity, age at time of procedure, and length of hospital stay. Data were missing on length of stay (LOS) in 208 patients. Classification of race by data extraction included 9 categories: American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White or Caucasian, Other, Unknown, Declined, and Unknown or Declined. The categories of Unknown, Declined, and Unknown or Declined were combined into 1 category labeled “Unknown or Declined.” Ethnicity was extracted in the following categories: Hispanic or Latino, Not Hispanic or Latino, Unknown, Declined, and Unknown or Declined. The categories of Unknown, Declined, and Unknown or Declined were combined into 1 group. The following comorbid conditions were included: type 1, type 2, and unspecified diabetes mellitus [International Classification of Diseases, 10th Edition (ICD10) E11.0-11.9, E10.9, E13.9], osteoporosis (M81.0-M81.8), traumatic brain injury (S06.9X9A), spinal cord lesion (G95.9 and G82.2), burns (T30.0-31.0), ankylosing spondylitis (M45.9), fibrodysplasia ossificans progressiva (M61.1), acetabular hip and/or femoral fracture (S32.4-S72.0), amenorrhea (N91.1-91.2), menopause (Z78.0, E28.3, E89.4, Z90.7), vitamin D deficiency (E55.9, R79.8), parathyroid disorders (E21.0-21.3, N25.8, C75.0), pathological fractures (M84.4, M80.0), chronic kidney disease (N18.1-18.9), and fall (W01.0). Comorbid conditions were included in the statistical analysis only if they preceded the surgery date. The primary outcome, HO, was attributed to the surgery if the diagnosis date was listed within 1 year after the procedure date. A history of motor vehicle accident (V87.7-V89.2) occurring up to 1 week prior to the surgery date was included. A separate category of acute fracture was generated to denote a history of fracture necessitating surgery. This was defined as a history of hip or femoral fracture occurring 14 days before or after the surgery date. We included postoperative events to capture fracture coding that may have been noted in electronic medical records during discharge billing. Given the difficulties in accurately coding for HO by electronic record systems, multiple ICD codes were used (728.10-728.13, 728.19, 731.2, 733.99, 781.99).

Inpatient and outpatient medications documented in the electronic medical record 1 year before and after the procedure date were included in the data extraction to avoid missing data. These data were then refined by expected duration of action for each medication and included in the statistical analysis as a categorical variable. The timeframe for inclusion in the statistical analysis varied. Bisphosphonates (alendronate, zoledronic acid, risedronate, ibandronate, and pamidronate), anastrozole, aspirin, calcitriol, calcium supplements, vitamin D supplements (cholecalciferol, ergocalciferol), oral contraceptive pills, teriparatide, denosumab, testosterone, raloxifene, and progesterone were considered in the statistical analysis if they were included as an active medication within 1 year preoperative to 7 days postoperative. NSAIDs (diclofenac, ibuprofen, indomethacin, ketorolac, meloxicam, naproxen, and celecoxib), prednisone, dexamethasone, and antibiotics (azithromycin, cephalexin, doxycycline, levofloxacin, bacitracin IM, cefazolin, ciprofloxacin, clindamycin, sulfamethoxazole-trimethoprim, and vancomycin) were used in statistical analysis if active 0 to 14 days before or after hip arthroplasty. Aspirin was separated from NSAIDs as dosage was not often specified and included 81-mg cardioprotective dosing rather than HO prophylaxis dosing. Data on medication use were missing for 323 patients. Imaging (CPT 72170, 72190, 72191, 72192, 72193, 72194, 72195, 72196, 72197, 72198, 73500, 73501, 73502, 73503, 73510, 73520, 73521, 73522, 73523, 73525, 73530, 73540, 73550, 73551, 73552, 76140, 76498) was used to verify adequate follow-up within our healthcare system, but only cases positive for HO were individually reviewed for confirmation and classification of HO. A board-certified, licensed musculoskeletal radiologist (D.M.) who was blinded to the patients’ clinical histories performed a radiologic review of cases identified as having HO by coding. Imaging modality varied on a case-by-case basis and included plain film X-ray, computed tomography, and magnetic resonance imaging. HO was graded using the Brooker classification system (26, 39, 40).

A history of radiation (CPT 77427, 77431, 77435, 77469, 77470) was extracted and included in the analysis if it was started within 7 days of the procedure date.

Statistical Analysis

Due to limited and varied documentation of medication dosage and duration, medication use was described as a categorical variable. To avoid sparse-data bias predictor, variables with <50 observations were excluded. Continuous variables were described using mean and SD and compared using an unpaired t-test. Categorical variables were described with the χ 2 test. For descriptive statistics, P-value < 0.05 was considered significant. To increase power, a combination group of low estrogen state, which included amenorrhea and postmenopausal females, was created.

Variable selection was considered first by univariate regression with a threshold of P < 0.25 to account for confounding and showed race; ethnicity; spine disease; osteoporosis; amenorrhea; postmenopausal status; type 1, type 2, and unspecified diabetes; chronic kidney disease; history of pathologic fracture; parathyroid disorders; and vitamin D deficiency. Combination groups of aspirin, bisphosphonates, and NSAIDS were significant for further variable selection. Variables were evaluated for interaction and collinearity with each other and comorbid conditions for the final logistic regression model and were found to be negative.

For the multivariable regression model, a stepwise regression using a least absolute shrinkage and selection operator analysis was employed to identify significant variables for the final model. P-value < 0.05 was considered significant. With this, the following variables dropped out of the final model: ethnicity, type 1 diabetes, chronic kidney disease, history of pathologic fracture, and parathyroid disorders. Preoperative (≤30 days prior to surgery) and postoperative (within 14 days after surgery) antibiotic use was evaluated on the outcome of HO formation. Perioperative antibiotic use was significant on univariate analysis for cefazolin.

Calculations were performed using STATA (StataCorp. 2019, Stata Statistical Software: Release 16. College Station, TX, USA; StataCorp LLC and StataCorp. 2021. Stata Statistical Software: Release 17).

Results

After the exclusion criteria outlined in the Methods section was applied, a total of 2541 unique patients who underwent primary hip arthroplasty were identified (Fig. 1). The incidence of HO development was 3% (80/2541 patients). Review of Brooker classifications showed a rating of severe HO (classes 3 and 4) in 18 patients of HO development, moderate HO (class 2) in 11 patients, and class 1 in 51 patients. The overall average age at the time of surgery was 62.14 ± 13.31 years, which was not significantly different between groups: 62.60 years in individuals who developed HO compared to 62.13 years in those who did not. The population was predominantly Caucasian (74.89% in the non-HO group vs 73.75% in the HO group) followed by Other (8.17% in the non-HO vs 6.25% in the HO group), Asian (8.09% in the non-HO vs 5% in the HO group), African American (6.14% in non-HO vs 13.75% in HO group), Unknown or Declined (2.03% in the non-HO vs 0% in the HO group), Native Hawaiian or Other Pacific Islander (0.45% in the non-HO vs 1.25% in the HO group), and American Indian or Alaska Native (0.24% in the non-HO vs 0% in the HO group; overall P-value = 0.082). Most individuals were not of Hispanic or Latino ethnicity (89.88% in the non-HO vs 95% in the HO group, P = 0.19). The average LOS was 2.67 ± 2.61 days in the non-HO group vs 3.02 ± 1.69 days in the HO group (P = 0.30).

A larger proportion of individuals who developed HO had preexisting spinal cord injury or spine disease (20% vs 7.15%, P < 0.001), osteoporosis (48.75% vs 18%, P < 0.001), type 1 diabetes (5.00% vs 0.93%, P < 0.001), type 2 diabetes (36.25% vs 16.98%, P < 0.001), vitamin D deficiency (33.75% vs 10.4%, P < 0.001), or amenorrhea (6.67% vs 2.03%) or were postmenopausal (46.67% vs 17.85%, P < 0.001) (Table 1). Acute fracture potentially necessitating arthroplasty was not significant between the 2 groups [234 (9.51%) of patients without HO vs 10 (12.5%) patients with HO]. Some variables surveyed such as burns (n = 25), preceding motor vehicle accident likely necessitating surgery (n = 15), and ankylosis spondylitis (n = 1) occurred infrequently so were omitted from Table 1. NSAID use and local radiation are sometimes used preoperatively as prophylaxis for HO. In our cohort, we did not detect a difference in HO formation between individuals who received radiation and those who did not (n = 53 in the non-HO vs n = 2 in the HO group, P = 0.834). Aspirin (P = 0.005), NSAID (P = 0.000), and bisphosphonates (P = 0.000) were significantly different between groups (Table 2). Perioperative NSAID use in the overall group was 75.6% but 323 patients had missing data regarding NSAID use.

Table 1.

Baseline characteristics of total/primary hip arthroplasty, by outcome

	No HO (n = 2461)	HO (n = 80, 3%)	P-value
Age at procedure	62.13 (13.29)	62.60 (13.76)	0.76
Race
White/Caucasian	1843 (74.89)	59 (73.75)	0.082
American Indian or Alaska Native	6 (0.24)	0 (0.00)
Asian	199 (8.09)	4 (5.00)
Black/African American	151 (6.14)	11 (13.75)
Native Hawaiian or Other Pacific Islander	11 (0.45)	1 (1.25)
Other	201 (8.17)	5 (6.25)
Unknown or Declined	50 (2.03)	0 (0.00)
Ethnicity
Hispanic or Latino	2212 (89.88)	76 (95.00)	0.19
Not Hispanic or Latino	162 (6.58)	4 (5.00)
Unknown or Declined	87 (3.54)	0 (0.00)
Sex, male (n = 1168)	1133 (46.04)	35 (43.75)	0.69
Length of stay, days	2.67 (2.61)	3.02 (1.69)	0.30
Spine disease	176 (7.15)	16 (20.00)	<0.001
Osteoporosis	443 (18.00)	39 (48.75)	<0.001
Amenorrhea, in females only (n = 1373	27 (2.03)	3 (6.67)	0.037
Postmenopausal state, in females only (n = 1373)	237 (17.85)	21 (46.67)	<0.001
Chronic kidney disease	237 (9.63)	13 (16.25)	0.050
Type 2 diabetes	418 (16.98)	29 (36.25)	<0.001
Type 1 diabetes	23 (0.93)	4 (5.00)	<0.001
Diabetes, Uunspecified	57 (2.32)	8 (10.00)	<0.001
Acute fracturea	234 (9.51)	10 (12.50)	0.37
Parathyroid disorders	50 (2.03)	5 (6.25)	0.011
Pathologic fracture history	75 (3.05)	7 (8.75)	0.005
Vitamin D deficiency	256 (10.40)	27 (33.75)	<0.001

Data are presented as mean (SD) for continuous measures and n (%) for categorical measures.

Abbreviation: HO, heterotopic ossification.

aAcute fracture is defined as hip or femoral fracture coded in the electronic medical record coding 14 days before and after surgery date.

Table 2.

Baseline characteristics of prophylactic treatment

	No HO	HO	p-value
Radiation	53 (2.15)	2 (2.50)	0.834
Aspirina	1,112 (51.65)	22 (33.85)	0.005
NSAIDs	1,881 (87.37)	41 (63.08)	0.000
Bisphosphonatesb	37 (1.72)	5 (7.69)	0.000

Abbreviations: HO, heterotopic ossification; NSAIDs, nonsteroidal anti-inflammatory drugs (diclofenac, ibuprofen, ketorolac, meloxicam, naproxen, and celecoxib).

aAspirin included uses of 81 mg and 325 mg dosing.

bBisphosphonates included alendronate, zoledronic acid, risedronate, ibandronate, and pamidronate.

Significant predictors for HO development identified in the multivariable regression model were African American race (OR 2.97, 95% CI 1.38-6.41, P = 0.005), osteoporosis (OR 2.72, 95% CI 1.52-4.88, P = 0.001), low estrogen state (OR 1.99, 95% CI 1.09-3.65, P = 0.025), and spine disease (OR 2.04, 95% CI 1.05-3.99, P = 0.036). As expected from prior studies, the use of NSAIDs within 14 days of surgery was negatively correlated with HO development (OR 0.39, 95% CI 0.22-0.69, P = 0.001) (Table 3). Aspirin was also found to be negatively correlated with HO (OR 0.53, 95% CI 0.31-0.92, P = 0.024).

Table 3.

Multivariable logistic regression on development of heterotopic ossification in total hip arthroplasties

	Odds ratio	P-value	95% CI
Racea—Black/African American	2.97	0.005	1.38-6.41
Spine disease	2.04	0.036	1.05-3.99
Osteoporosis	2.72	0.001	1.52-4.88
Type 2 diabetes	1.30	0.409	0.70-2.42
Low estrogen stateb	1.99	0.025	1.09-3.65
Unspecified diabetes	1.65	0.326	0.61-4.48
Vitamin D deficiency	1.72	0.084	0.93-3.17
Aspirin	0.53	0.024	0.31-0.92
NSAIDs	0.39	0.001	0.22-0.69
Bisphosphonates	1.45	0.505	0.49-4.33

Abbreviation: NSAIDs, nonsteroidal anti-inflammatory drugs.

aFor race, Caucasian was used as the baseline comparator.

bLow estrogen state is a combination group of cases with amenorrhea and postmenopausal status.

Of the antibiotics used, only postoperative cefazolin approached the significance threshold of P < 0.25 in univariate analysis. However, it was not a significant predictor in multivariable regression.

Discussion

Our analyses on baseline characteristics by outcome (Table 1) showed that the 2 groups, non-HO and HO, were equally matched by age and sex. Our study did not identify male sex as a significant predictor variable as has been seen in prior studies (41, 42); however, this could be explained by the relatively low incidence of HO, at 3%, in our cohort. While our incidence is low compared to the average incidence rate of HO development post–hip arthroplasty or acetabular repair, it remains within the range of published data (41, 42). Our HO incidence rate may also reflect our strict exclusion criteria and the rarity of HO.

Of interest is the predictor of African American race on the development HO. This was also seen in a smaller study by Slone et al (43). Their retrospective case control study of 253 patients showed African-American patients were significantly more likely to develop severe HO following acetabular fracture compared to Caucasian patients (OR 2.24 95% CI 1.22-4.11). A limitation in our study is that the cohort was primarily Caucasian [n = 1902 (74.9%) of the population overall]. The association observed in the African American group was among a fraction of the overall cohort [n = 162 (6.4%) overall]. Additionally, although we had information on the race categories outlined by the Office of Budget and Management standards, we lacked more granular data within these race classifications (44). Further stratification in a larger more diverse dataset would be useful for elucidating this potential link.

The average hospital LOS postsurgery was similar between the groups (2.67 ± 2.61 days in the non-HO group vs 3.02 ± 1.69 days in the HO group, P = 0.30). Studies published with a focus on hospital stay and HO have focused on the time from injury to surgery and include mostly critically ill patients. Hayashi et al showed a greater degree of HO formation, based on the Brooker classification system, in individuals who had a delay in hip arthroplasty after fracture, with patients demonstrating an OR of 1.296 for grade ≥3 HO when surgery was delayed an average 6 days vs 2 days (45). D’Heurle et al showed a similar association between surrogates of injury and HO development in patients treated for acetabular fracture with intensive care unit (ICU) LOS 1 to 2 days and non-ICU LOS >10 days being associated with development of severe (Brooker grade ≥3) HO (46). Generally, severe HO is associated with a poorer functional outcome and the predictor of LOS may be a result of the decreased functional status (47, 48). The association between LOS and HO formation may have been washed out by our broad study population, which included individuals with both ICU and non-ICU stays and varying degrees of HO classifications.

We identified an unexpected positive correlation between preexisting osteoporosis (OR 2.72, 95% CI 1.52-4.88, P = 0.001) and HO development. This correlation was independent of acute fracture or history of femoral, hip, or pathologic fracture, indicating that the association between osteoporosis and HO appears to be independent of trauma. The association between osteoporosis and nongenetic HO has not been clearly reported in the literature; however, a study from Okano et al comparing bone mineral density in a prospective study of females undergoing total hip arthroplasty for osteoarthritis did not show a significant difference in observed bone densities of the lumbar spine, distal radius, or calcaneus between the group that developed HO and the group that did not develop HO (49). Notably, our cohort was younger, with the mean age of our total population at 61.73 years compared to 77.20 years in the Okano et al study population. We were unable to extract data such as bone mineral density measurements to confirm whether the diagnostic criteria for osteoporosis were met; thus, the diagnosis of osteoporosis was based solely on the ICD coding.

Notably, univariate analysis on our cohort suggests a positive association of bisphosphonate use on the development of HO (OR 4.20, 95% CI 1.61-10.93, P = 0.003) despite a relatively small number of osteoporosis patients documented as receiving bisphosphonates. This is likely an underrepresentation, as medications such as zoledronate are given as intermittent doses that may not be captured in medication lists or may be given outside of the EMR system or through infusion centers. In comparison, the Okano et al study population focused only on patients who were not taking medications known to affect bone turnover, likely excluding patients with underlying osteoporosis (49). Thus, the use of bisphosphonates and other bone active drugs is likely underestimated in our cohort, and the potential contribution of bisphosphonates to HO formation remains to be explored.

The prevalence of an acute preoperative fracture did not differ significantly between groups (P = 0.37) (Table 1). The incidence of acute fracture as the cause for primary arthroplasty was noted to be 9.6%, which is higher than would be expected. A large database study of 139 635 patients undergoing total hip arthroplasty showed that osteoarthritis was the indication for surgery in 96.7% of cases (135 013 cases) while femoral neck fracture was the indication for 3.33% of cases (50). Another study evaluating the differences in hospital LOS for total hip arthroplasty cases performed for osteoarthritis or femoral neck fracture in the United States yielded an incidence of acute fracture preceding surgery of 2.2% (953/42 692 patients) (51). Although we restricted the coding time period to 14 days pre- and postoperatively, this may have overestimated the incidence and limited causal associations.

In keeping with multiple prior studies of HO in spinal cord injury or traumatic brain injury patients, our study showed a higher proportion of patients with a history of spine disease (which included ICD codes for spine disease, spine injury, and traumatic brain injury) in the group that developed HO (52-54) (Table 1). The low frequency of spine disorders may also reflect incomplete documentation, particularly if HO developed in an area that was not imaged after surgery. In patients with spinal cord or traumatic brain injury, HO formation can develop distant to the site of injury, and a broader imaging assessment could be helpful to address this question (52, 55).

Conditions that affect bone turnover were noted in higher proportions in the HO group and were selected for multivariable regression analysis. Vitamin D deficiency occurred at a higher frequency within patients who developed HO. This could be an independent association or the result of vitamin D deficiency being common in disorders associated with decreased bone density such as kidney disease and type 2 diabetes (56, 57). Chronic kidney disease and parathyroid disorders were noted to be more prevalent in the HO group, although the number of observations was low, which could explain the lack of significance in regression selection. However, progressive chronic kidney disease can result in metabolic disturbances that affect bone turnover and quality (56), particularly in patients with renal osteodystrophy. Being a tertiary center, the study population could have included patients with a higher frequency of underlying metabolic bone disorders or a genetic predisposition to HO, although only a handful of genes have been associated with genetic HO (7, 10, 58), and how genetic variants contribute to HO remains largely unknown.

Low estrogen state was found to be significantly correlated with HO (OR 1.99, 95% CI 1.09-3.65, P = 0.025). Burn tenotomy murine models support a role for increased Phospho-SMAD (pSMAD) and insulin-like growth factor 1 signaling in HO development in male mice (59). It is not understood whether individuals with amenorrhea would exhibit a similar pathophysiology for HO formation that differs from postmenopausal females. Coding limited further detail regarding the etiology of amenorrhea and postmenopausal status. The age of initial diagnosis and whether the diagnosis was primary or secondary would be difficult to ascertain without manual chart review. Further studies into the role of estrogen on HO and its clinical significance need to be probed to determine whether these variables are associated with HO development.

Of the antibiotics surveyed, cefazolin had enough observations (>50) to be evaluated for associations on HO development. Preoperative cefazolin use was significantly associated in univariate analysis (OR 0.59, P = 0.11). Currently, there is active investigation of the effects of gastrointestinal microbiota on bone remodeling with evidence that insulin-like growth factor 1 plays a role in supporting bone growth (60, 61). Murine models have shown an increase in trabecular bone mass with antibiotic use (60). Conversely, a study evaluating HO in blast injury mice models exposed to methicillin-resistant Staphylococcus aureus showed a higher volume of HO formation in infected mice (62). With continuing evidence that the gut microbiota plays a role in inflammatory response, we added antibiotic use as a potential modifier for HO formation.

We found that NSAIDs were negatively correlated with HO formation (OR 0.39, 95% CI 0.22-0.69, P = 0.001). Due to limitations in the data extraction, we were only able to view information on the tablet dose but not the frequency of administration. The medication descriptions as noted in the EMR data extraction is provided [Supplementary Table 1 (63)]. Aspirin was also negatively correlated with HO (OR 0.53, 95% 0.31-0.92, P = 0.024). As noted, aspirin was not included in the NSAID group because aspirin 81 mg, cardioprotective dosing, was common in the population. Prophylactic options for HO include the use of NSAIDs, selective cyclooxygenase-2 (COX-2) inhibitors, and radiation therapy (64-66). Within our institution, the decision and dosing for NSAID prophylaxis is at the discretion of the individual surgical team. The presence of NSAID use perioperatively in these cases could have been for pain management or for prevention of HO.

The first evaluation of the use of NSAIDs in HO prophylaxis was in 1975 in a study of 39 patients taking indomethacin for analgesia after total hip arthroplasty (67). Since then, numerous studies comparing the use of different NSAID dosages and duration vs controls or radiation have been completed (67). Current studies have mostly been limited to retrospective reviews, cohort studies, and case reports (68). In the past 5 years there has been 1 randomized controlled study on the use of naproxen 500 mg twice daily for 3 weeks following arthroscopic surgery for femoroacetabular impingement (69). The study was stopped prematurely as the efficacy of the naproxen group was dramatic (46% in placebo vs 4% in the naproxen group, P < 0.001). The use of the selective COX-2 inhibitor, etodolac, at a dose of 600 mg once daily for 2 weeks postoperatively was evaluated in a cohort study of 263 patients undergoing hip arthroscopic surgery and showed no occurrences of HO compared to 36% in the control group (P < 0.0001) (70). These studies have shown benefit in reducing HO formation when NSAIDs are used prophylactically. In contrast, a single center study, which accounted for surgical approach and included 423 patients, noted no differences in HO development when routine indomethacin prophylaxis was stopped. Radiographic HO was noted in 47% of patients (71). Thus, the use of NSAIDs in appropriate patients should be considered before hip arthroplasty.

Prophylactic radiation therapy has been used preoperatively to reduce the risk of HO formation. The optimal timing appears to be within 3 days of the procedure (64, 72, 73). Only 2 patients in our cohort who developed HO underwent postoperative radiation. This small sample size likely contributed to the lack of a significant difference in HO in our cohort. Individual chart review showed that these individuals carried an increased risk of HO formation due to underlying ankylosing spondylitis, which may explain the use of radiation prophylaxis. In the patients who did not develop HO but underwent radiation, the indication for radiation included disorders with increased risk for HO formation (such as ankylosing spondylitis) and the need for radiation treatment in the setting of malignancy. Radiation has been shown to be effective in HO prophylaxis, although it has a higher risk profile than NSAIDs and COX-2 inhibitors (74). Most studies on radiation therapy have been limited to patients undergoing hip surgeries, and there are currently no consensus-based guidelines for dosing (75).

There are several limitations to our study. First, it is a single-institution, retrospective study by data extraction, which resulted in some variables (eg, osteomalacia, motor vehicle accident, radiation therapy) having too few observations to gain significance or power to be used in regression analysis. This was seen in variables known to be associated with HO including metabolic bone disorders such as myositis ossificans progressiva and risk factors such as burn history.

Also, in determining the indication for surgery in our patient population, 9.6% of primary hip arthroplasties were performed in the setting of an acute fracture. As noted earlier, fracture rates preceding total hip arthroplasty are generally low. This number may reflect a misclassification of distant fractures as acute fractures or may represent a higher rate of hip arthroplasties performed for the treatment of acute fractures at our institution. Although the presence of an acute fracture was not significantly associated with the outcome of HO, a differential misclassification may decrease the interpretation of our results in relation to individuals with perioperative fracture.

As with other retrospective record reviews, the quality of data is dependent on the accuracy and coding of medical records. As seen in a study by Fury et al, an evaluation in the accuracy of coding for HO in 522 patients showed that 38% of patients with an HO-specific ICD-9 code had no evidence of HO documented in their EMR, and only 48% of patients with traumatic injury preceding HO formation had documentation of HO in their EMR (76). Additionally, due to limitations of the data extraction, nuance can be lost, and there may be areas of missing data. In our initial data extraction, there was missing data for 122 patients. On our EMR system, there was a transition between data storage software in 2012. We speculate that during this EMR storage transition information did not crossover for these 122 patients. With limited granularity on descriptors such as medications, many of the variables were assigned categorical descriptions. For each disease state, nonpositive cases were coded as not having the comorbid condition, which could have been a limitation in EMR coding and underestimated the prevalence of these conditions. It is likely that our HO reporting rate is an underestimate due to lack of documentation in the EMR, especially as HO is often undocumented unless it causes pain or functional limitations.

Given the large number of patients in this study, it was not feasible to perform an independent radiology review on 2541 patients. Although images were reviewed by a radiologist blinded to the patient cases, this was only performed for the identified positive cases, given the large cohort. Independent review identified 30 false-positive cases. The power and significance of associations may have been stronger if some cases were reclassified from the non-HO to the HO group. Limited nuance was also seen in the extraction of medication use where dosages could not be estimated and variables had to be classified as categorical, which underestimates the extent of in-group variation on the outcome. An example would be our classification of zoledronate, which often could not be accounted for since infusion medications are not recorded on a patient’s outpatient medication list. Additionally, there were 323 patient cases that were missing data regarding medication use. Finally, as UCSF is a tertiary care center, it is possible that there is a bias toward a more complex patient population or higher acuity surgical cases.

Despite these limitations, this study is among the largest single-center retrospective analyses and identifies potential areas of further research into the pathophysiology driving the development of HO in patients who have undergone primary hip arthroplasty.

Conclusions

In summary, this large, single-center retrospective cross-sectional analysis identified potential new risk factors for developing HO, including African American race, osteoporosis, and low estrogen state. These risk factors appear to increase the association between primary hip arthroplasty and HO development. Additionally, these select patients may benefit from perioperative prophylaxis with NSAIDs, as our study also confirmed the known reduction of HO formation with NSAID use in hip arthroplasties. These results provide support for further research to understand whether these correlations are surrogate or causative markers of HO development.