The association of proximal femoral shape and incident radiographic hip OA in elderly women

Summary

Purpose

Variations in femoral head shape are reported to predict incident hip osteoarthritis (OA). This study evaluated if proximal femur shape at baseline was a risk factor for incident radiographic hip OA (RHOA) after 8.3 years of follow-up in a cohort of elderly Caucasian women.

Methods

Supine pelvic radiographs were obtained as part of the Study of Osteoporotic Fractures (SOF) at baseline and Visit 5 (8.3 years later), and were scored for RHOA. A nested case–control study was performed: hips were eligible for inclusion if they had no prevalent RHOA in either hip at baseline. Cases of incident RHOA were defined as no RHOA at baseline and RHOA in their right hip present at Visit 5 [or right total hip replacement (THR) for OA between baseline for follow-up] and a random selection of one half of all incident RHOA cases plus right THR cases (n = 102) were chosen. A random selection (n = 249) of control subjects who had no RHOA in their right hip at both baseline and follow-up visit were included for comparison.

The shape of the right proximal femur was outlined on a digitized baseline radiograph and a statistical image analysis technique, Active Shape Modeling (ASM), was used to generate 10 unique and independent “modes” or variations in shape, which explained 95% of the variance in the shape of the proximal femurs studied. Any hip shape was therefore described as the average shape plus a linear combination of these 10 independent modes of variation. The values for each of these 10 modes for each hip analyzed were entered into a logistic regression model as independent predictors of incident RHOA adjusting for covariates.

Results

The incident RHOA cases were slightly taller, heavier and had higher total hip bone mineral density (BMD) than control subjects (P < 0.05), but were otherwise similar demographically. Results of ASM showed that Modes 1, 2 and 3 together explained 81% of the variance in proximal femur shape among all subjects analyzed. Modes 3, 5, 9 which accounted for 8.9%, 3.3% and 0.8% of the variance respectively, were significant predictors of incident RHOA with adjusted odds-ratios ranging from 1.61 to 1.99 (P < 0.001) for every 1 standard deviation (SD) increase in the mode score.

Conclusion

These results suggest that variations in the relative sizes of the femoral head and neck at baseline are modest determinants of incident RHOA in elderly Caucasian women.

Background

Osteoarthritis (OA) of the hip is a health problem that affects a large number of aging adults, and is associated with significant disability¹. Currently, the treatment is directed at pain relief and improvement of function once the disease is established. Less is known about specific interventions that could be tested to prevent OA.

Radiographic assessment of the disease has focused on scoring systems which grade cartilage changes and bony changes as OA worsens^2,3. While cartilage protection has been a focus of potential therapy^4,5 changes in the structure, content and shape of bone also are major features of the disease. A number of investigators have observed that there appear to be definite patterns of hip geometry or shape, that are associated with an increased risk of hip OA development, such as measures of the center-edge angle of Wiberg in acetabular hip dysplasia^6,7, and reduction in the femoral head–neck–shaft angle with hip OA⁸. Examination of bone composition in hips with OA has shown that the proportion and distribution of trabecular and cancellous bone change with the presence of OA^9,10. Similarly, bone mineral density has also been found to be increased locally in hips with OA^11,12.

Evaluation of the gross features of hips with OA was reported recently by Gregory et al.¹³ using the method of active shape modeling (ASM) to identify and analyze different femoral head and neck shapes associated with incident radiographic hip OA (RHOA). The authors found that distinct shape patterns in the femoral head and neck were associated with a higher risk of incident RHOA [odds ratio (OR) = 1.62; 95% confidence interval (CI): 1.08–2.45], prior to any radiographic evidence of RHOA. These findings are interesting, since alterations in bone shape or geometry early in the course of the disease could lead to abnormal joint loading which may accelerate the localized degenerative process.

ASM is a method of analyzing patterns in the shapes of objects that have inherent variability. In addition to the above report, the ASM method was used by Gregory et al. to analyze hip shape features in association with hip fracture risk¹⁴, and has been used by others to analyze different patterns in vertebral body shape¹⁵. The ASM method analyzes patterns in hip shape in a cohort of patients by building an average shape of the hip in that cohort. The ASM program then builds a statistical model that describes the variation in the shape of the hips in the cohort and generates distinct shapes, or “modes,” of variation in the hip shape across this cohort. Each mode effectively describes a distinct change in the hip shape a number of standard deviations (SDs) away from the average shape of the entire cohort, that is unrelated to and independent of the other modes.

Based on earlier studies that reported distinct proximal femur geometry that predicted hip OA risk, we hypothesized that there would be distinct hip shapes that would be associated with a group of elderly Caucasian women with incident RHOA. The specific aim of this study was to use the ASM method to examine the association of hip shape and incident RHOA. This study builds on the work done by Gregory et al. with several important differences from the previous study¹³. The population studied previously was a heterogeneous population of older men and women, and looked at the association of the femoral head and neck with incident RHOA, but not at the shape of the whole proximal femur.

Since men and women may have inherent differences in bony features of OA¹⁶, the study presented here examined whether the shape of the proximal femur assessed by ASM, was a predictor of incident RHOA in a population comprised solely of elderly Caucasian women.

Methods

STUDY SUBJECTS AND POPULATION

Patients were Caucasian women ≥65 years of age at baseline who were enrolled in the Study of Osteoporotic Fractures (SOF), a cohort whose characteristics have been described previously^17,18. Participants were recruited between September 1986 and October 1988 from population-based listings in four areas of the United States¹⁷. Nonwhite women were excluded from the original cohort because of their low incidence of hip fracture, as were women who were non-ambulatory or who had undergone bilateral hip replacement (THR)¹⁷. In addition, women with radiographically confirmed rheumatoid arthritis, Paget’s disease, hip fracture, or bilateral total hip joint replacement at baseline were excluded from this analysis.

RADIOGRAPHY AND INTERPRETATION

At the baseline and follow-up visits (average 8.3 years follow-up time), supine anteroposterior radiographs of the pelvis were obtained using a standard protocol¹⁷, and digitized using a VIDAR digitizer (VIDAR Systems Corp, Herndon, VA) at a resolution of 0.169 mm (150 dpi) and stored at 16-bit DICOM images for further analysis.

Radiographs had previously been read as described by Lane et al.¹⁷ for five individual radiographic features (IRFs) of hip OA [joint space narrowing (JSN), osteophyte formation, subchondral sclerosis, cysts, and deformity] using atlas photographs to improve the reliability of the readings^3,19. Minimum joint space (MJS) was measured using digitized calipers as has been previously published²⁰. The methods for radiographic interpretation have also been previously published^3,17.

The baseline and follow-up radiograph pairs were initially read and measured by one primary reader (NEL) side-by-side with the reader blinded to the order by masking identifying information and randomly assigning the order of films. Radiographs with either definite osteophytes or definite narrowing (severity score ≥ 2) in any location on the initial reading were jointly evaluated by two readers to reach consensus scoring. A total of 21% of the film pairs underwent a consensus reading. Inter-rater reliability for the radiographic readings evaluated from a random sample of 178 pairs was good to excellent: for MJS < 1.5, kappa = 0.95; for definite JSN, kappa = 0.91; and for definite osteophytes, kappa = 0.71^17,18.

DEFINITION OF INCIDENT RHOA

The definition of RHOA has been used previously^18,21. A summary score of 0–4, modified from Croft³, was assigned to each hip based on IRFs. Grade 1 hips were defined as hips with possible osteophytes or narrowing, Grade 2 hips required the presence of either definite (severity grade ≥ 2) JSN or osteophytes plus at least one other feature (cysts or subchondral sclerosis). Grade 3 hips required the presence of three of the following features: (1) definite osteophytes, or (2) JSN (grade ≥ 2) plus (3) either cysts or sclerosis. Grade 4 hips met the criteria for Grade 3, and also had femoral head deformity present.

Radiographs were assessed for the following IRFs and summary scores: hips were considered to have radiographic findings of hip OA if a summary grade ≥ 2 was present. A hip was defined as having developed RHOA (incident disease) if the right hip at baseline radiograph had a summary grade < 2 at baseline and had grade ≥ 2 on the 8.3 year follow-up film radiograph. Controls were subjects with right hips that did not have RHOA (grade ≥ 2) at the baseline or follow-up radiograph. Hips which underwent a THR who had RHOA = 0–1 at baseline were also included in the analysis.

SELECTION OF CASES AND CONTROLS FOR ASM

From the SOF cohort, 5928 subjects with supine pelvic X-rays at baseline and 8.3 years apart were identified, and the readings of RHOA on these subjects were retrospectively examined to design a nested case–control study, similar to the methods reported previously¹⁸.

Cases of incident right RHOA

Subjects were eligible for ASM analysis if they had RHOA score 0–1 at baseline in both right and left hips. A total of 206 subjects meeting the eligibility criteria went on to develop incident RHOA (RHOA score ≥ 2) in their right hip at follow-up and a random selection of one half of them were chosen for as cases of incident RHOA for ASM analysis, along with 18 subjects who had undergone right THR for OA between baseline and follow-up. In order to be included, the radiographic film had to have the complete anatomy of the right proximal femur, including the greater trochanter, visible to be included in the ASM analysis. A total of 19 cases (16%) had images unsuitable for analysis, leaving 102 cases in the ASM analysis.

Control subject selection

To be eligible for selection as a control subject, required the same baseline RHOA status as the case subjects, but control subjects had to have RHOA score 0–1 in their right hip at follow-up. Randomly selected controls (N = 300) from eligible subjects were chosen, but 51 (17%) were dropped due to unsuitable radiographs, leaving 249 controls in the ASM analysis.

ASM

ASM was performed using the method of Cootes et al. from the University of Manchester, UK²². Subjects’ digitized X-rays were evaluated by a reader (RKC) who outlined the shape of the femoral head and neck by placing a series of 60 evenly spaced points, from the lesser trochanter to an opposite point on the femoral shaft (Fig. 1). The algorithms used are the same as those of Gregory et al.¹³ apart from the fact that rather than using 16 points placed around only the femoral head and neck, we placed 60 evenly spaced points around the whole proximal femur to the level of the lesser trochanter.

Fig. 1.

Outline of proximal femur. 60-point outline of proximal femur from the lesser trochanter to the opposite shaft. These point coordinates were inputted into the ASM program to generate the 10 modes of shape variation.

Three hundred and fifty-one baseline right hip radiographs of both cases and controls were used by the ASM program to generate the composite average right proximal femur shape of this sample, which formed the point of reference for comparison of variations from this average shape. The ASM program calculated the different modes of variation in hip shape using the method of Principal Components Analysis (PCA). Each mode of variation described a percentage of the overall variance in hip shape, and each mode was independent of every other mode of variation. Modes of variation in shape which is between them explained 95% of the total variance in shape were then subsequently entered into a logistic regression analysis as independent predictors of incident RHOA. In addition, the overall size of the proximal femur was removed from the model, so the different modes did not reflect absolute differences in hip size, only differences in shape and relative size of different proximal femur features. The program generated 10 “modes” or variations for each hip shape in this group of subjects.

POINT PLACEMENT AND RELIABILITY

Building on the methods of outlining the femoral head published previously¹³, we manually placed 30 points around the femoral head and proximal neck. Then, to accurately characterize the shape of the entire proximal femur, we expanded the number of points to outline the entire proximal femur, adding 30 additional points to outline the femoral neck, trochanter and femoral shaft, resulting in a total of 60 evenly spaced points for the proximal femur. Reliability of point placement was assessed in two ways. Initially the principal reader (RKC) placed points to outline the femoral head in 20 radiographs; a second reader (JAL) also outlined the femoral head in the radiograph. Inter-reader agreement on the point placement within 2 mm was 92%. In addition, intra-reader reliability for point placement within 2 mm on 10% of the sample was 95%.

ASSESSMENT OF POTENTIAL CONFOUNDERS

All participants completed a self-administered questionnaire at baseline that assessed age, self-reported health status, hours sedentary each day, education level and current medication use. Physical activity was assessed using with a modified Paffenberger survey²³. Height was measured using a wall-mounted Harpenden stadiometer (Holtan, Dyfed, UK) and weight was measured with a balance beam scale²⁴.

STATISTICAL ANALYSIS

Baseline characteristics between patients with and without incident RHOA were analyzed through Student’s t test and chi-squared methods. The association of modes with incident hip OA was tested using logistic regression, with OA case or control status as the outcome, adjusting for age, height, weight and BMD (total hip). Each mode was included as an independent variable, and since each mode is by definition independent of the other modes, there was no necessity to enter them in a stepwise manner. In addition, we applied a Bonferroni correction to our multiple comparisons between modes, resulting a level of significance at P-value of ≤ 0.005.

Results

BASELINE CHARACTERISTICS

Baseline characteristics were not different between cases and controls, except that cases were slightly taller, heavier and had a higher total hip BMD compared to controls (P < 0.05) (Table I).

Table I.

Baseline characteristics of the study subjects: mean ± SD for continuous variables, frequency (%) for dichotomous of categorical variables

Variables	Cases (n = 102)	Controls (n = 249)
Continuous: mean ± SD
Age (years)	70.7 ±4.3	70.7 ±4.5
Height (cm)	160.7 ±6.2*	159.2 ±5.7
Weight (kg)	70.5 ±15.1*	67.0 ±10.9
BMI (kg/m2)	27.2 ±5.3	26.4 ±4.2
Total hip BMD (g/cm2)	0.80 ±0.13*	0.76 ±0.12
Dichotomous: frequency of Yes (%)
Estrogen use	45 (44.12)	99 (39.76)
Vitamin D use	55 (56.7)	138 (56.56)
Walks for exercise	60.0 (58.8)	124 (49.8%)
Categorical: frequency (%)
Health status
Excellent	31 (30.4%)	100 (40.2%)
Good	58 (56.9%)	119 (47.8%)
Fair	13 (12.8%)	27 (10.8%)
Poor	–	3 (1.2%)

^*Significant difference cases vs controls at P < 0.05.

ACTIVE SHAPE MODEL

ASM showed that a total of 10 modes of variation explained 95% of the variance in hip shape (Table II). Modes 3, 5 and 9 were significant predictors of incident RHOA (P < 0.001).

Table II.

Percentage of variance in the proximal femur shape explained by each mode of variation, and association with incident RHOA

Mode	% of variance explained (N = 351)	Controls (N = 249)		Cases (N = 102)		P-value
		Adjusted OR	95% CI	Adjusted OR	95% CI
Mode 1	54.4	1.0	–	1.24	0.92–1.68	0.164
Mode 2+	18.2	1.0	–	0.58	0.39–0.85	0.006
Mode 3*	8.9	1.0	–	1.73	1.25–2.39	<0.001
Mode 4	5.3	1.0	–	0.99	0.74–1.34	0.961
Mode 5*	3.3	1.0	–	2.31	1.63–3.28	<0.001
Mode 6+	1.5	1.0	–	0.67	0.50–0.91	0.010
Mode 7+	1.1	1.0	–	0.74	0.55–0.99	0.043
Mode 8+	0.9	1.0	–	0.67	0.51–0.90	0.007
Mode 9*	0.8	1.0	–	1.81	1.32–2.49	<0.001
Mode 10	0.7	1.0	–	0.99	0.72–1.37	0.96

^*Significant at P < 0.005;

⁺not significant at P < 0.005 using Bonferroni correction. Adjusted for age, height and total hip BMD.

Mode 3

Mode 3 was associated with changes in size of the femoral head and neck in relation to the trochanters and shaft, as can be seen in Fig. 2(A), where the outlines for positive and negative variations in this mode mainly differ around the femoral head and neck. Mode 3 increased the odds of incident RHOA compared to the controls [OR = 1.61; 95%CI: 1.22–2.13] per +1 SD increase in Mode 3. Such positive values of Mode 3 are related to larger femoral heads, with longer necks and slightly thinner necks relative to the size of the trochanters and shaft [Fig. 2(A)].

Fig. 2.

Figures showing the main effects of proximal femur shape variation in the Modes 3, 5 and 9. Each figure shows the average (—), +2 SD (- - - -) and −2 SD (..) shapes in each of these modes. In each mode the +2 SD outline is associated with a higher risk of incident RHOA. (A) Mode 3: +2 SD shape associated with variations in the relative size of femoral head and neck length. (B) Mode 5: +2 SD shape associated with variations in relative size and width of femoral neck and greater trochanter. (C) +2 SD shape associated with variations in relative size of both femoral head and greater trochanter (see text for details).

Mode 5

Mode 5 reflects the relative sizes of the greater trochanter and femoral neck compared to the femoral head and shaft. High values for this mode appear to be related to larger than average greater trochanter size and smaller femoral neck size relative to the average size of the femoral head and shaft. This can be seen in Fig. 2(B) where high or low values for this mode deviate away from the average shape mainly around the greater trochanter and femoral neck. Mode 5, explained 3.3% of the variance in the proximal femur shape, and Mode 5 increased the odds of incident RHOA [OR = 1.99; 95%CI: 1.46–2.71] per +1 SD increase in Mode 5.

Mode 9

Mode 9, like Mode 3, was related to the size of the femoral head relative to the femoral shaft, and like Mode 3, high values of Mode 9 occur when the femoral head is large compared to the femoral neck [see Fig. 2(C)]. The difference between Mode 3 and Mode 9 is that in Mode 9, there is a concomitant change in the size of the greater trochanter which becomes more pronounced for high values of Mode 9, whereas in Mode 3, the change in the size of the femoral head was unrelated to the size of the greater trochanter. Mode 9 explained 0.8% of the variation in proximal femur shape, and also increased the odds of RHOA compared to the control group [OR = 1.73; 95%CI: 1.29–2.33] per + 1 SD increase in Mode 9.

For Modes 3 and 9, the relative curvatures at the superior and medial aspects of the femoral head remained unchanged, but in Mode 5, high values of this mode (associated with incident RHOA), alter these relative curvatures. For the average hip shape in our model, the radius of curvature at the superior aspect of the femoral head (point #28) is about 4.5% larger than at the medial aspect (point #20). Alterations in Modes 3 and 9 leave this ratio unchanged, but a −2 SD alteration in Mode 5 reduced this ratio to 2.0%, indicating that low values for Mode 5 are associated with more circular femoral heads. A +2 SD alteration in Mode 5 increased the ratio to 6.8%, indicating that high values of this mode occur in hips where the femoral head is flatter than average (increased radius of curvature) on the superior compared to the curvature on the medial aspect.

Eleven subjects with no RHOA at baseline, developed incident severe RHOA and had a THR after 8.3 years of follow-up. Analysis of the THR group separately found that Modes 3, 5, and 9 were associated with an increased risk of RHOA defined by THR, compared to control subjects, in a similar manner to these cases with incident RHOA on their follow-up radiographs, but for this small sample of cases with THR, the effects did not reach statistical significance.

Discussion

We evaluated use of the ASM method to generate variations in shape of the proximal femur from radiographic images to examine association of hip shape with incident RHOA and identified three distinct hip shapes, or “modes” or hip shapes that were significantly associated with an increased risk of incident RHOA: Modes 3, 5 and 9. The proximal femur shape that was characterized by positive values of Mode 3 was a larger femoral head and longer, slightly thinner femoral neck. Mode 3 was strongly associated [OR = 1.61; 95%CI: 1.22–2.13] with the risk of incident RHOA. Mode 9 was also related to the size of the femoral head, but also with a concomitant change in the size of the greater trochanter. Mode 5 (with the highest OR for predicting incident RHOA) was related not only to the relative sizes of femoral head, neck and trochanters, but also the relative curvatures of the superior and medial aspects of the femoral head. High values for Mode 5 predicted incident RHOA and were related to relatively flatter superior aspects compared to more curved medial aspects of the femoral head.

The use of ASM to assess hip shapes was first described by Gregory et al., who operationalized the use of ASM to examine associations between hip shape and OA. Additionally, Gregory et al. used ASM to evaluate the risk of osteoporotic hip fracture and found an increased risk of osteoporotic fracture associated with a combination of four modes generated by the ASM program [OR = 1.83; 95%CI: 1.08–3.11]¹⁴.

Gregory et al. found that the high-risk OA mode was characterized by a decreased curvature of the superior femoral neck and sharper transition from the femoral head to the lower part of the neck [OR = 1.62; 95%CI: 1.08–2.45]¹³, but unlike the data presented in this paper, Gregory et al. only looked at associations between the shape of the femoral head and hip OA. Due to limitations in their X-ray acquisition, they could not examine the shape of the whole proximal femur.

Differences between the Gregory study and ours may also reflect in the populations studied, since the cohort used by Gregory et al. was drawn from the Rotterdam study, and included both men and women, whereas we only examined Caucasian women. There were also several other significant differences between our study and the study by Gregory et al. First, our study analyzed only baseline films, prior to the development of moderate or severe RHOA, while the Gregory study looked at both baseline and follow-up films. Also, the radiographic case definitions of incident RHOA differed; we used modified Croft while Gregory et al. used Kellgren–Lawrence (K–L) grades, and our respective case definition may have differed, since the K–L system requires the presence of both osteophytes and JSN for a K–L grade = 2, while the modified Croft does not necessarily require both.

We did not identify the nonspherical, or “pistol grip,” deformity which has been reported as being associated with RHOA²⁵. Such a deformity would have involved a mode of proximal femur shape in which the concavity of the superior segment of the femoral neck was reduced, but none of our modes of variation represented this shape. Our population comprised women only, who have been described as having a significantly lower prevalence of pistol grip deformity than male subjects with symptomatic hip OA²⁶. In addition, women with femoralacetabular impingement the proximal femoral head is more anterior such that the AP view may actually underestimate the impingement area²⁷.

The association of proximal femur shape and risk of hip OA has been studied using other geometric measurements. While measurements of the head–neck–shaft angle to assess varus and valgus alignment⁸, and measurement of the center-edge angle of Wiberg to determine the presence and the severity of hip dysplasia⁷ have provided quantifiable measures of hip geometry differences, the ASM imaging method, by taking the entire shape of the proximal femur into account, incorporates these geometric differences in its assessment of the modes of variation. However, formal comparisons of the correlation between ASM modes associated with RHOA and measures of hip geometry in the same population are needed to understand the strengths and weaknesses of each measurement modality.

Our study had several strengths, including the use of a well-defined, large cohort of elderly women with longitudinal hip radiographic data. However, there were several weaknesses. Our study was limited to Caucasian women, and the results are therefore not generalizable to other groups. Secondly, this study only examined the association with incident RHOA so no conclusions can be drawn regarding progression. Third, ASM was only performed on the proximal femur, and acetabular dysplasia is a known increased risk for incident RHOA. Additional studies will need to be performed that combine femoral and acetabular shape to assess hip OA and the risk of hip OA. We did not examine the shape of the left hip in this study, but since none of the subjects in our study had prevalent left RHOA, and because equal numbers of case and control subjects developed left RHOA during the study (18/102 cases – 18% and 47/249 controls – 19%), we do not feel that the status of the contralateral hip at the final study visit has affected our results. Also, we were unable to control for rotation or anteversion during the acquisition of the pelvic radiograph, which may have influenced the prevalence of the modes. We will need to determine if rotation or anteversion changes the proximal femur shapes when it is controlled for during the radiograph acquisition, possibly with the Osteoarthritis Initiative Project.

In summary, quantitative assessment of the shape of the proximal femur through use of the ASM program found that three distinct shapes were associated with an increased risk of incident RHOA. These shapes were characterized physically by variations in the relative sizes of the femoral head, neck and trochanteric region. ASM, in combination with other potential biomarkers may be useful predictors of incident RHOA.