 KNEE

Trabecular bone density of the proximal tibia as it relates to failure of a total knee replacement M. A. Ritter, K. E. Davis, S. R. Small, J. G. Merchun, A. Farris From Joint Replacement Surgeons of Indiana, St. Francis Hospital, Mooresville, Indiana, United States

 M. A. Ritter, MD, Chief of Orthopaedic Surgery  K. E. Davis, MS, Research Statistician  J. G. Merchun, BS, Medical Research Student  A. Farris, BA, Medical Research Writer St. Francis Hospital, Joint Replacement Surgeons of Indiana, 1199 Hadley Road, Mooresville, Indiana 46158, USA.  S. R. Small, MS, Director of Engineering Rose-Hulman Institute of Technology, Department of Applied Biology and Biomedical Engineering, Terre Haute, Indiana, USA. Correspondence should be sent to Dr M. A. Ritter; e-mail: [email protected] ©2014 The British Editorial Society of Bone & Joint Surgery doi:10.1302/0301-620X.96B11. 33465 $2.00 Bone Joint J 2014;96-B:1503–9. Received 12 November 2013; Accepted after revision 2 May 2014

The relationship between post-operative bone density and subsequent failure of total knee replacement (TKR) is not known. This retrospective study aimed to determine the relationship between bone density and failure, both overall and according to failure mechanism. All 54 aseptic failures occurring in 50 patients from 7760 consecutive primary cemented TKRs between 1983 and 2004 were matched with non-failing TKRs, and 47 failures in 44 patients involved tibial failures with the matching characteristics of age (65.1 for failed and 69.8 for non-failed), gender (70.2% female), diagnosis (93.6% OA), date of operation, bilaterality, pre-operative alignment (0.4 and 0.3 respectively), and body mass index (30.2 and 30.0 respectively). In each case, the density of bone beneath the tibial component was assessed at each follow-up interval using standardised, calibrated radiographs. Failing knees were compared with controls both overall and, as a subgroup analysis, by failure mechanism. Knees were compared with controls using univariable linear regression. Significant and continuous elevation in tibial density was found in knees that eventually failed by medial collapse (p < 0.001) and progressive radiolucency (p < 0.001) compared with controls, particularly in the medial region of the tibia. Knees failing due to ligamentous instability demonstrated an initial decline in density (p = 0.0152) followed by a nondecreasing density over time (p = 0.034 for equivalence). Non-failing knees reported a decline in density similar to that reported previously using dual-energy x-ray absorptiometry (DEXA). Differences between failing and non-failing knees were observable as early as two months following surgery. This tool may be used to identify patients at risk of failure following TKR, but more validation work is needed. Cite this article: Bone Joint J 2014;96-B:1503–9.

Bone mineral density (BMD) is believed to be an important determinant of success following total knee replacement (TKR).1-3 Although previous studies have investigated changes in BMD following TKR, few report a direct association between bone density and clinical outcome.2-7 There are conflicting views in the published literature whether higher BMD was associated with superior1-2,4-6,8-9 or inferior10 outcomes following TKR, or whether the proper balancing of forces intra-operatively is the most important factor in achieving optimal BMC.3 Patients with a low BMD prior to surgery have been demonstrated to be at higher risk of failure by prosthetic loosening and migration.8 Patients with poor bone quality are at higher risk of failure following revision TKR.1,5,9 However, other investigators have reported higher rates of failure following TKR in patients with high BMD in the medial tibial region.10 They suggest that demographic factors including age, gender, and activity level

VOL. 96-B, No. 11, NOVEMBER 2014

may be more important in explaining this finding,10 or that proper balancing of forces, to a more physiologically natural state, through proper alignment is more important in maintaining optimal conditions for bone density.3 BMD, as measured using dual-energy x-ray absorptiometry (DEXA), consistently declines with age.2-7,9 However, the use of DEXA to determine the effect of bone density on failure rates of TKR would require routine bone scanning of a very large population of patients, over a long period of time, at great expense. It has been suggested that standard digitised radiographs may be sufficiently sensitive to examine density changes in failed and non-failed TKRs over time11-15 with the advantage of being considerably less expensive than DEXA. The purpose of this study was to examine the relationship between post-operative tibial BMD and failure following TKR. We hypothesise that different mechanisms of failure will correspond to different patterns of change in 1503

1504

M. A. RITTER, K. E. DAVIS, S. R. SMALL, J. G. MERCHUN, A. FARRIS

Table I. Demographic information (mean +/- one standard deviation) by joint count including gender, age, overall anatomical alignment, body mass index (BMI), and diagnosis for each cohort of patients Gender

Age (yrs)

BMI (kg/m2)

Diagnosis

Control Failure

M14/F33 M14/F33

69.8 (51 to 85) 64.5 (42 to 80)

30.0 (21.0 to 48.4) 30.2 (20.3 to 48.1)

OA 44/ON 2/RA 1 OA 44/ON2/RA 1

Subgroup demographics Instability Failed Control p-value Medial Collapse Failed Control p-value Radiolucency and loosening Failed Control p-value

Post-op alignment

BMI

Opdate difference*

Age

8.4 (+/-10.9) 8.3(+/-11.2) 0.9797

26.8 (+/-6.6) 26.6(+/-6.4) 0.9312

1.8 (+/-3.1) yrs 0.0397

62.0 (+/-6.8) 67.3(+/-7.1) 0.0392

-4.0 (+/-7.2) -4.0 (+/-7.4) 0.9999

32.4 (+/-7.0) 32.2(+/-6.7) 0.9166

1.2 (+/-5.2) yrs 0.2922

65.0 (+/-7.2) 67.3(+/-8.7) 0.3040

-4.3 (+/-4.3) -4.8 (+/-4.6) 0.8635

30.6 (+/-2.4) 30.4(+/-2.8) 0.9065

4.9 (+/-5.2) yrs 0.0193

60.5 (+/-7.1) 75.0(+/-10.0) 0.0295

* paired student’s t-test M, male; F, female; OA, osteoarthritis; ON, osteonecrosis; RA, rheumatoid arthritis

BMD.16,17 As described in biomechanical studies of medial loading and migration, we expect to observe an increased BMD prior to failure in knees which fail by medial tibial collapse.8,18-25 Conversely, for knees failing by ligamentous instability, we expect to observe decreased tibial density as the prosthesis remains well-fixed, and because the patient is likely to subject the implant to lower loads, resulting in the resorption of bone.2,3,16,17,26 For knees which fail by loosening secondary to migration of the prosthesis with progressive radiolucency, we anticipate that density changes will also increase on standardised radiographs due to increased stress.8,18

Patients and Methods A total of 7760 TKRs were performed at our unit between 1983 and 2004. In all cases, the cemented Anatomic Graduated Component total knee replacement (AGC prosthesis) (Biomet, Warsaw, Indiana) was used. The tibial components were made with Himont 1900 polyethylene compressionmolded onto a Co-Cr component, producing a non-modular or monoblock design, and were implanted without augments on the medial or lateral tibial plateau. Of the 7760 knees, 54 failed in the absence of infection, 13 knees failed by infection, 15 developed loosening of metal-backed patella, 19 had isolated patella failure, 533 TKRs (6.9%) were completely lost to follow-up, and mean follow-up for those TKR remaining was 6.9 years (2 to 22). A total of 1920 TKRs (29%) implanted in 1278 patients died in the duration of the study, but were included based on present follow-up time if demographically matched to a failing TKR. All polyethylene tibias (536 THRs in 405 patients) were not studied in the present analysis. Of the 54 aseptic prosthesis failures, seven involved the femoral component alone, and although these behaved similarly to the medial collapse and radiolucency followed by loosening groups, were excluded from the results. This left 47 involving the

tibia failures: 26 failed due to medial collapse, 16 failed due to ligamentous instability, and five failed due to total radiolucency and eventual loosening.27 Each of the 47 failed TKRs (44 patients) was matched to a corresponding control (i.e., a non-failed TKR) based on gender, whether the operation was bilateral or unilateral, the diagnosis, preoperative alignment, body mass index (BMI), age, and date of operation; since femoral failures were excluded from the findings, we report the results of 47 failures in 44 patients in the current manuscript. As each of the surgeons in the unit used similar surgical techniques, it was not deemed necessary to match on surgeon. All alignments were measured on short film radiographs by either the operating surgeon or another of the six orthopaedic surgeons at our centre; the surgeon used a goniometer and measured the femorotibial anatomic alignment axis, expressed in negative values for anatomic varus alignments and positive values for anatomic valgus alignments. Details of patient demographic details and length of follow-up are given in Table I. The mean time to failure for failing knees involving the tibia was 4.9 years (SD 3.4, 0.6 to 13.1). Mean followup for non-failed knees in surviving patients was 10.4 years (SD 4.8, 2.9 to 20.3). Retrospective analysis of tibial bone density was performed using digital radiographs. For each patient, the radiographs taken at each follow-up interval were analysed using digital image processing software (ImageTool 3.0, University of Texas Health Science Center, San Antonio, Texas). Hard copy radiographs were digitised at 300 dots per inch (dpi) using a medical image scanner (DiagnosticPRO Advantage, VIDAR Systems Corp, Herdon, Virginia), resulting in 8-bit digital images with 256 shades of grey. Radiographs were then calibrated and standardised in order to minimise the variation between images caused by differences in exposure, the age of the radiograph and the soft-tissue mass of the patient. Calibration was achieved by THE BONE & JOINT JOURNAL

TRABECULAR BONE DENSITY OF THE PROXIMAL TIBIA AS IT RELATES TO FAILURE OF A TOTAL KNEE REPLACEMENT

Fig. 1 Radiograph showing the calibration regions A (air 0) and F (255 femoral prosthesis), along with medial measurement regions (M1, M2, M3), lateral measurement regions (L1, L2, L3), and the distal region (D1). Cortical bone, cement, and overlap of tibia and fibula are avoided.

assigning greyscale values within each radiograph of 0 in a region of air adjacent to the knee, and 255 in a region within the femoral prosthesis, using a method described in a previous studies.11-13,15 Measurement regions were defined by dividing the length of the tibial stem into three medial regions (M1, M2, M3), three lateral regions (L1, L2, L3), and a region distal to the stem of the prosthesis (D1) (Fig. 1). Measurement regions were defined in this way to sample the largest possible area of trabecular bone, avoiding peripheral cortical bone, cement, and overlap of the fibula. The greyscale value represents the density in each region; higher greyscale values correspond to higher bone densities. Statistical analysis. Patients were matched using a predefined algorithm. Each failed TKR was matched to a list of control TKRs on the basis of age and bilaterality. Nearestneighbour matching was then performed on each variable in order of importance: alignment was matched, followed by BMI, date of surgery and age in that order. Failing TKRs were matched to controls on a 1:1 basis. A power calculation was performed, taking the mean and standard deviation from within the present data at one year in M1 (mean 192.7, standard deviation 28.9) with power set at 80% and a significance level of 0.05. In order to detect a 9.2 point (4.8%) difference between paired radiographs at one year, 18 patients were needed per subgroup. At six months, the expected variance was higher and as a result, the numbers required in each group were higher: 28 patients were required in each group to detect a 10 point (5.2%) difference with the same power and level of significance. Following the sub-group analyses, a post hoc power calculation was performed. With the number of patients available, the minimum detectible differences for the collapse, instability and radiolucency subgroups were 19 points (10.6% density difference), 23 points (12.8% difference), and 37 points (20.6% difference) respectively. VOL. 96-B, No. 11, NOVEMBER 2014

1505

The mean greyscale value was recorded for each region of the radiograph, at each follow-up interval, for each patient. Progression of greyscale values over time was analysed using univariable linear regression. As each of the seven regions was analysed at every follow-up interval (two months, six months, one, three, five, seven and ten years), a total of 42 separate comparisons were made to compare the greyscale values in the two groups. There were too few knees remaining in the failure group at ten years to provide any meaningful results and therefore these are not reported. P-values for density changes were given by the least-squaremeans test in each of the 42 univariable linear regressions. Demographic variables were compared between matched knees by the paired student’s t-test. Equivalence testing was performed by the two one-sided tests (TOST) statement in the TTest Procedure in SAS 9.2.28 Statistical significance was defined as a p-value < 0.05. All analyses were performed using SAS version 9.2 (SAS Institute Inc., Cary, North Carolina) Results The control cohort (n = 54) demonstrated a significant decline in periprosthetic density in all regions over time from two months through to ten years (p < 0.001). Compared with the density observed at two months, the greatest decline in density was observed in the medial regions, followed by the lateral regions and the region distal to the keel (Table II). There was a trend towards increased periprosthetic BMD in the failing knees compared with controls. This reached statistical significance in M1 at all times beyond one year, and in M2 at one to five years (Table III). Compared with controls, knees failing by medial tibial collapse showed consistently higher densities in the M1 region at all times (Table III). There was a trend to higher bone density in the other medial regions but this was less consistent. Lateral and distal regions showed no significant difference in density compared with the control pool. While the periprosthetic BMD in knees failing secondary to instability was significantly lower in most areas at two months, this difference disappeared and from six months onwards (aside from borderline significance in D1 at one year) there were no significant differences between the knees failing by ligamentous instability and controls (Table III). In knees failing by loosening secondary to progressive radiolucency, there were significantly higher medial bone densities beyond one year in all medial regions. While there was a trend to higher bone densities in the lateral regions, this did not reach statistical significance at any point (Table III). For all failure mechanisms, in nearly every region, there was little fluctuation in numerical values for density over time, while non-failing knees declined in density as time increased (p < 0.001). In M1, knees eventually failing by instability did not statistically rise or decline in value and were statistically equivalent within 10% above or below the initial value (p = 0.034 at ten years).

1506

M. A. RITTER, K. E. DAVIS, S. R. SMALL, J. G. MERCHUN, A. FARRIS

Table II. Calibrated grayscale values are provided for all failure cohorts, measurement regions, and followup intervals. No data are available for the radiolucency five-year and seven-year follow-ups due to the small number of patients assessed Non-fail (n = 54)

L1

L2

L3

M1

M2

M3

D1

Two months Six months One year Three years Five years Seven years Ten years

199.2 194.5 192.8 183.4 184.5 175.7 179.6

203.3 197.7 197.5 189.1 190.3 183.6 185.1

211.1 207.5 207.7 200.7 202.5 196.5 199.7

185.4 178.1 173.5 163.4 160.3 153.2 153.9

191.7 186.1 181.2 171.9 171.2 163.6 164.2

195.6 192.4 189.2 179.5 179.9 172 172.5

208.6 206.6 205.9 198.8 201.6 193.4 194

All failures (n = 54) Two months Six months One year Three years Five years Seven years Ten years

L1 202.9 199.8 198.7 194.3 200.6 192.3 203.2

L2 203.6 201.6 201.2 196.6 204.1 193.5 202.8

L3 210.9 208.0 211.5 206.8 212.8 206.1 213.1

M1 193.1 188.5 196.5 188.0 192.8 181.6 185.9

M2 196.1 192.6 199.0 191.1 194.8 186.5 191.6

M3 198.0 194.5 199.8 193.8 196.3 189.1 194.3

D1 208.5 205.1 208.0 204.2 206.6 203.4 210.1

Instability (n = 16) Two months Six months One year Three years Five years Seven years Ten years

L1 194.7 199.9 192.4 193.2 200.3 190.5 199.8

L2 193.7 199.1 191.7 194.2 201.4 191.1 200.2

L3 198.7 205 201.2 202.3 211.7 203.3 210.9

M1 173.2 173.5 171.8 168.4 182.1 169.9 178.3

M2 178.7 181.7 177.1 173.5 184.6 176.9 184.1

M3 181.6 185.1 181.5 179.1 190.5 184.2 188.4

D1 195.9 198.6 195.3 197.5 206.1 199.7 205.5

Collapse (n = 26) Two months Six months One year Three years Five years Seven years Ten years

L1 204.3 202.2 193.2 189.6 197.1 192.5 216.1

L2 204.6 203.5 196.3 192.8 198.7 190.5 213.2

L3 215 209.4 208.9 206.1 207.6 202.6 223.3

M1 200.3 198.3 202.7 200.2 201.7 197.2 203.1

M2 202 199.9 201.8 200.1 197.5 192.4 206

M3 203.4 200.2 199.7 200 195.4 185 207.5

D1 212.1 207.6 204.6 205.6 203.2 200.9 220.4

Radioluncency and loosening (n = 5) Two months Six months One year Three years Five years Seven years Ten years

L1 208.4 201 213.2 214.8 X X 201.2

L2 211.7 205.5 217.4 215.5 X X 204

L3 217.2 213.9 225.9 225.4 X X 210.9

M1 207.3 195.4 209 209.9 X X 226.1

M2 208.5 193.7 218.9 221.2 X X 234.3

M3 209.5 198.9 219.9 220.5 X X 223.3

D1 214.7 210 227.7 224 X X 223.3

Discussion When measured using DEXA, periprosthetic bone density has been demonstrated to decline progressively over time following TKR.2-6 The amount which bone density declines varies between studies: it has been quantified as 5.1% at seven years4 and 36.8% at two years6 in different cohorts. In the present study, non-failing knees showed a significant decline in density across tibial regions ranging from 7.3% to 17.4% at seven-year follow-up on standardised radiographs, matching the general trend found in DEXA studies. This technique has been validated previously: a study comparing the use of standardised radiographs to DEXA in the measurement of bone density following TKR has demonstrated an intraclass correlation coefficient (ICC) of 0.989

(95% confidence interval (CI), 0.968 to 0.998).15 It is worth noting that one recent study involving the Oxford partial knee reported a relatively low amount of BMD change for partial knee arthroplasty at two years post-operatively,29 and appears consistent with the physiologically balanced behavior of BMD reported by Soininvaara.3 The use of standardised radiographs can never replace DEXA entirely. DEXA provides a physical density value in g/cm2, while standardised radiographs do not, and previous studies have suggested that the sensitivity of standardised radiographs to changes in density is poorer than that of DEXA.3,7,10 However, it is prohibitively difficult to perform a longitudinal study on failing and non-failing TKRs using DEXA. By analysing radiographs which are taken routinely THE BONE & JOINT JOURNAL

TRABECULAR BONE DENSITY OF THE PROXIMAL TIBIA AS IT RELATES TO FAILURE OF A TOTAL KNEE REPLACEMENT

1507

Table III. P-values from 42 separate univariable linear regression analyses for each tibial region and followup interval comparing all failures to controls, followed by subgroup analyses. Statistically significant differences (p < 0.05) are denoted in bold. All failures (n = 54) L1

L2

L3

M1

M2

M3

D1

0.5102 0.1447 0.5826 0.0665 0.0301 0.1231 x

0.8445 0.2853 0.9645 0.1363 0.0673 0.2715 x

0.7818 0.7354 0.7924 0.1433 0.1127 0.3024 x

0.1264 0.0639 0.0020 0.0010 0.0032 0.0227 x

0.4628 0.2630 0.0211 0.0091 0.0234 0.0867 x

0.8450 0.6963 0.2548 0.0419 0.0678 0.2704 x

0.7135 0.6981 0.7002 0.2904 0.3014 0.4263 x

L1 0.4224 0.3631 0.8543 0.2778 0.1983 0.3098 X

L2 0.0908 0.7611 0.3259 0.5107 0.3324 0.5125 X

L3 0.0204 0.759 0.2036 0.8018 0.3657 0.5211 X

M1 0.0454 0.5454 0.5232 0.6808 0.1823 0.2688 X

M2 0.0253 0.5703 0.3571 0.9415 0.3901 0.4258 X

M3 0.0152 0.3382 0.1444 0.8628 0.4395 0.4552 X

D1 0.0219 0.2377 0.044 0.7895 0.6349 0.6377 X

L1 0.328 0.1753 0.9461 0.5275 0.2962 0.3541 X

L2 0.7975 0.2831 0.7469 0.6241 0.4526 0.632 X

L3 0.4032 0.6706 0.8994 0.4153 0.5892 0.646 X

M1 0.0048 0.0046 0.0001 0.0001 0.012 0.0116 X

M2 0.0428 0.0524 0.0032 0.003 0.0717 0.104 X

M3 0.1316 0.2364 0.1302 0.0209 0.2337 0.5228 X

D1 0.49 0.8521 0.6745 0.358 0.8308 0.657 X

Radiolucency and loosening (n = 5) L1 Two months 0.332 Six months 0.5473 One year 0.0872 Three years 0.083 Five years X Seven years X Ten years X

L2 0.3646 0.4738 0.0869 0.1146 X X X

L3 0.4634 0.5397 0.0636 0.0906 X X X

M1 0.027 0.2168 0.0071 0.0268 X X X

M2 0.0778 0.5909 0.0035 0.0197 X X X

M3 0.1479 0.6212 0.0109 0.038 X X X

D1 0.5092 0.7732 0.0282 0.1187 X X X

Two months Six months One year Three years Five years Seven years Ten years Instability (n = 16) Two months Six months One year Three years Five years Seven years Ten years Two months Medial collapse (n = 26) Two months Six months One year Three years Five years Seven years Ten years

following TKR, the use of this technique has allowed such a study to be performed for the first time. By studying bone density longitudinally, both long- and short-term trends can be identified. This study has demonstrated a consistently elevated periprosthetic density in the medial tibial plateau of knees that progress to failure by medial collapse. In all three medial zones, density values remained high and relatively constant for TKRs failing by collapse compared with control TKRs, which declined as expected. While there was a trend to increased density in the lateral zones, none reached statistical significance. These findings suggest that the theory whereby stress shielding and low density lead to tibial collapse is wrong,1 instead, this study concurs with studies that suggest that increased stresses in the medial tibia are associated with tibial collapse.17,18,20 This supports the hypothesis that altered loading and bone remodeling are VOL. 96-B, No. 11, NOVEMBER 2014

associated with failure by medial collapse. What is unexpected is that the increases in density appear to begin as early as two months post-operatively, whereas previous studies found unusually prolonged migration and the associated density increases in problematic knees at or after the first post-operative year.18,25 There are many potential reasons for these findings. High BMI and varus medial component alignment, leading to overloading of the medial cancellous bone, have been shown to be directly associated with failures by medial tibial collapse.17 Patients with small tibiae and a high BMI have also been shown to be at higher risk of medial collapse in theoretical studies due to the presence of elevated forces at the joint.18 Biomechanical studies have shown a clear association between tibial strains and component malalignment,19,20 increased resection depth,20 size mismatches between femoral and tibial components,21 and the use of an

1508

M. A. RITTER, K. E. DAVIS, S. R. SMALL, J. G. MERCHUN, A. FARRIS

Calibrated areal density (grayscale value)

230 220 L1

210 200

Nonfail Instability Medial collapse Radiolucency

190 180 170 160 150

Calibrated areal density (grayscale value)

0

2

4 6 8 Follow-up (yrs)

10

230 220

M1

210 200 Nonfail Instability Medial collapse Radiolucency

190 180 170 160 150

Calibrated areal density (grayscale value)

0

2

4 6 8 Follow-up (yrs)

10

230 D1

220 210 200

Nonfail Instability Medial collapse Radiolucency

190 180 170 160 150 0

2

4 6 8 Follow-up (yrs)

10

Fig. 2 Graphs showing changes in density for each failure cohort at each follow-up interval. No data is available for five and seven year follow-up in knees which failed due to radiolucency.

all-polyethylene tibial component.22 Although a direct relationship between strain and failure has not been defined, the elevation in medial density observed in the present failure cohort is believed to be the result of increased stress, vascularity, and bone formation in the medial aspect of the proximal tibia, eventually proving to be unable to compensate and prevent the medial collapse of the tibial plateau. An alternative explanation is that, rather than abnormally increasing in density, the failing knees simply failed to display the normal decrease in density expected following TKR, perhaps as a result of disuse. If this were the case, the authors believe that far more than 26 of 7760 knees would have failed by medial collapse and conclude that this explanation is unlikely to be the case. In TKRs which failed secondary to the development of radiolucency, there was a similar increase in bone density prior to failure, but this seems to propagate further down the tibia than observed with the medial collapse group. At one year and three years, M1 had 20.4% and 28.4% higher

density than the non-failed group, compared with M2 at 20.8% and 28.7%, and M3 at 16.2% and 22.8% respectively. Knees that eventually failed by ligamentous instability initially exhibited significantly lower tibial densities at two months compared with control knees in the majority of the tibial regions. This difference appears because the prosthesis is not loose, and instead these TKRs exhibit compromised ligament function and stability that appears to be difficult to detect in early follow-up. As follow-up time increases, the density values in this group continue to remain relatively constant, with no observable decline. At follow-up periods after the first post-operative year, the mean densities of the control group appear to fall below those of the instability group in both the medial and lateral regions, but these differences were not found to be significant with the numbers available. It is worth noting that the average densities of the non-failed group were found to lie below the statistically equivalent region of the instability group in later follow-up (the reversal is visible at three, five, seven and ten years follow-up in Figure 2). The behaviour of the instability group does appear dissimilar to that of the non-failing group, due to the lower initial density at two months and the lack of any gradual reduction in density over time. This study had a number of weaknesses. This was a retrospective study and, while the patients are paired there is the risk of unmeasured confounding which would not be present in a randomised study. The subgroup analyses were not paired and there were significant differences between the baseline demographics of failing and control knees in some of the variables. This raises the possibility that the differences in density observed in the subgroup analyses were due to inadequate matching rather than being due to the failure mechanism itself. Unpaired analysis was conducted because the number of paired observations was low. This could have been avoided by matching patients at a ratio of two or three to one, but this was impossible due to time and resource constraints. The number of knees that failed was very small and therefore the power of some of the subgroup analyses (particularly the radiolucency-associated failure group) was very low, raising the probability of type II error. In the radiolucency-associated failure group a difference of < 37 points would have been undetectable with the cohort studied. There was a very high probability of type I error due to multiple testing. As each comparison was conducted separately with no post hoc correction attempted, at least two false positive p-values would be expected among the 42 tests conducted in each of the analyses. Also, since preoperative radiographs lack a metal prosthesis for the lightest point of calibration in the image, we were unable to match against the pre-operative density. This could be resolved with the inclusion of a metal measurement standard placed in the pre-operative radiological field at the time of exposure. Likewise, while post-operative radiographs THE BONE & JOINT JOURNAL

TRABECULAR BONE DENSITY OF THE PROXIMAL TIBIA AS IT RELATES TO FAILURE OF A TOTAL KNEE REPLACEMENT

were calibrated using the greyscale value of the femoral component, different sizes of femoral component may be more or less dense, and therefore compromise the calibration. However, the same femoral component design was used in all cases, and femoral component thickness changes minimally based on femoral component width in this design. Likewise, the same core group of radiographers was used over the 21-year period, minimising, though not eliminating, variability in exposure of the radiographs. In a previous study, the mean time to failure has been shown to vary by failure type, being 3.3 years for medial collapse, 6.3 years for ligamentous imbalance, and 3.7 years for progressive radiolucency.17 In a similar way, this study has demonstrated differences between knees failing by different mechanisms in terms of periprosthetic bone density. Knees failing by tibial collapse showed an elevated, constant density compared with non-failing knees throughout the follow-up period, starting at two months, particularly visible in the medial region just beneath the tibial plateau. Radiological failures also showed a similar constant elevation in densities in medial regions from two months following surgery until the time of failure. Knees failing by ligamentous instability showed a decrease in initial post-operative density in medial, lateral and distal regions at two months, then no decline in density across the post-operative period. While restricted by the limited number of patients evaluated, the present investigation appears to identify observable, distinct differences between mechanisms of failure in TKR shown in standardised radiographs. This raises the possibility that the risk of failure could be predicted by analysis of post-operative radiographs as early as two months following surgery. Further adequately-powered studies are required to support the findings of this study. In patients who exhibit these changes in density in the early post-operative period, a prospective study using DEXA could be useful in the development of this technique as a diagnostic tool. No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article. This article was primary edited by A. Liddle and first proof edited by G. Scott.

References 1. Lonner JH, Klotz MK, Levitz C, Lotke PA. Changes in bone density after cemented total knee arthroplasty. Influence of stem design. J Arthroplasty 2001;16:107–111. 2. Petersen MM, Nielsen PT, Lauritzen JB, Lund B. Changes in bone mineral density of the proximal tibia after uncemented total knee arthroplasty: a 3-year follow-up of 25 knees. Acta Orthop Scand 1995;66:513–516. 3. Soininvaara TA, Miettinen HJ, Jurvelin JS, et al. Periprosthetic tibial bone mineral density changes after total knee arthroplasty: one-year follow-up study of 69 patients. Acta Orthop Scand 2004;75:600–605. 4. Hernandez-Vaquero D, Garcia-Sandoval MA, Fernandez-Carreira JM, Gava R. Influence of the tibial stem design on bone density after cemented total knee arthroplasty: a prospective seven-year follow-up study. Int Orthop 2008;32:47–51. 5. Levitz CL, Lotke PA, Karp JS. Long-term changes in bone mineral density following total knee replacement. Clin Orthop Relat Res 1995;321:68–72.

VOL. 96-B, No. 11, NOVEMBER 2014

1509

6. Minoda Y, Kobayashi A, Iwaki H, et al. Comparison of bone mineral density between porous tantalum and cemented tibial total knee arthroplasty components. J Bone Joint Surg [Am] 2010;92-A:700–706. 7. Abu-Rajab RB, Watson WS, Walker B, et al. Peri-prosthetic bone mineral density after total knee arthroplasty: cemented versus cementless fixation. J Bone Joint Surg [Br] 2006;88-B:606–613. 8. Petersen MM, Nielsen PT, Lebech A, Toksvig-Larsen S, Lund B. Preoperative bone mineral density of the proximal tibia and migration of the tibial component after uncemented total knee arthroplasty. J Arthroplasty 1999;14:77–81. 9. Mintzer CM, Robertson DD, Rackermann S, et al. Bone loss in the distal anterior femur after total knee arthroplasty. Clin Orthop Relat Res 1990;260:135– 143. 10. Therbo M, Petersen MM, Varmarken JE, Olsen CA, Lund B. Influence of preoperative bone mineral content of the proximal tibia on revision rate after uncemented knee arthroplasty. J Bone Joint Surg [Br] 2003;85-B:975–979. 11. Hernandez-Vaquero D, Garcia-Sandoval MA, Fernandez-Carreira JM, et al. Measurement of bone mineral density is possible with standard radiographs: a study involving total knee replacement. Acta Orthop 2005;76:791–795. 12. Robertson DD, Mintzer CM, Weissman BN, et al. Distal loss of femoral bone following total knee arthroplasty. Measurement with visual and computer-processing of roentgenograms and dual-energy x-ray absorptiometry. J Bone Joint Surg [Am] 1994;76-A:66–75. 13. Nagamine R, Hanada Y, Kondo M, et al. Quantification of bone volume on radiographs using NIH Image. Mod Rheumatol 2000;10:220–224. 14. Seki T, Omori G, Koga Y, et al. Is bone density in the distal femur affected by use of cement and by femoral component design in total knee arthroplasty? J Orthop Sci 1999;4:180–186. 15. Small SR, Ritter MA, Merchun JG, Davis KE, Rogge RD. Changes in tibial bone density measured from standard radiographs in cemented and uncemented total knee replacements after ten years' follow-up. Bone Joint J 2013;95-B:911– 916. 16. Moreland JR. Mechanisms of failure in total knee arthroplasty. Clin Orthop Relat Res 1988;226:49–64. 17. Berend M, Ritter M, Meding J, et al. Tibial component failure mechanisms in total knee arthroplasty. Clin Orthop Relat Res 2004;428:26–34. 18. Berend ME, Ritter MA, Hyldahl HC, Meding JB, Redelman R. Implant migration and failure in total knee arthroplasty is related to body mass index and tibial component size. J Arthroplasty 2008;23(Suppl):104–109. 19. Green GV, Berend KR, Berend ME, Glisson RR, Vail TP. The Effects of Varus Tibial Alignment on Proximal Tibial Strain in Total Knee Arthroplasty. J Arthroplasty 2002;17:1033–1039. 20. Berend ME, Small SR, Ritter MA, Buckley CA. The effects of bone resection depth and malalignment on strain in the proximal tibia after total knee arthroplasty. J Arthroplasty 2010;25:314–318. 21. Berend ME, Small SR, Ritter MA, et al. Effects of femoral component size on proximal tibial strain with anatomic graduated components total knee arthroplasty. J Arthroplasty 2010;25:58–63. 22. Small SR, Berend ME, Ritter MA, Buckley CA. A comparison in proximal tibial strain and all-polyethylene anatomic graduated component total knee arthroplasty tibial components. J Arthroplasty 2010;25:820–825. 23. Taylor M, Tanner KE. Fatigue failure of cancellous bone: a possible cause of implant migration and loosening. J Bone Joint Surg [Br] 1997;79-B:181–182. 24. Taylor M, Tanner KE, Freeman MA. Finite element analysis of the implanted proximal tibia: a relationship between the initial cancellous bone stresses and implant migration. J Biomech 1998;31:303–310. 25. Ryd L, Albrektsson BE, Carlsson L, et al. Roentgen stereophotogrammetric analysis as a predictor of mechanical loosening of knee prosthesis. J Bone Joint Surg [Br] 2005;77-B:377–383. 26. Winsor RE, Scuderi GR, Moran MC, Insall JN. Mechanisms of Failure of the Femoral and Tibial Components in Total Knee Arthroplasty. Clin Orthop Relat Res 1989;248:15–20. 27. Ritter MA. The Anatomical Graduated Component total knee replacement: a long-term evaluation with 20-year survival analysis. J Bone Joint Surg [Br] 2009;91-B:754–759. 28. No authors listed. The TTest Procedure: TOST Equivalence Test. SAS/STAT(R) 9.2 User’s Guide, Second Edition. 2010, SAS Institute Inc., Cary, North Carolina:7414-7418. 29. Hooper GJ, Gilchrist N, Maxwell R, et al. The effect of the Oxford uncemented medial compartment arthroplasty on the bone mineral density and content of the proximal tibia. Bone Joint J 2013;95-B:1480–1483.