Body mass index is not a clinically meaningful predictor of patient reported outcomes of primary hip replacement surgery: prospective cohort study

Abstract

Objectives

To describe whether body mass index (BMI) is a clinically meaningful predictor of patient reported outcomes following primary total hip replacement (THR) surgery

Design

Combined data from prospective cohort studies. We obtained information from four cohorts of patients receiving primary THR for osteoarthritis: Exeter Primary Outcomes Study (n=1431); EUROHIP (n=1327); Elective Orthopaedic Centre (n=2832); and St. Helier (n=787). The exposure of interest was pre-operative BMI. Confounding variables included: age, sex, SF-36 mental health, comorbidities, fixed flexion, analgesic use, college education, OA in other joints, expectation of less pain, radiographic K&L grade, ASA grade, years of hip pain. The primary outcome was the Oxford Hip Score (OHS). Regression models describe the association of BMI on outcome adjusting for all confounders.

Results

For a 5-unit increase in BMI, the attained 12-month OHS decreases by 0.78 points 95%CI (0.27 to 1.28), p-value 0.001. Compared to people of normal BMI (20 to 25), those in the obese class II (BMI 35 to 40) would have a 12-month OHS that is 2.34 points lower. Although statistically significant this effect is small and not clinically meaningful in contrast to the substantial change in OHS seen across all BMI groupings. In obese class II patients achieved a 22.2 point change in OHS following surgery.

Conclusions

Patients achieved substantial change in OHS after THR across all BMI categories, which greatly outweighs the small difference in attained post-operative score. The findings suggest BMI should not present a barrier to access THR in terms of PROMs.

INTRODUCTION

Total hip replacement surgery (THR) is a commonly performed and successful surgical intervention, providing substantial relief from pain and improvement in functional disability in patients with hip arthritis^1-3. The lifetime risk for undergoing a hip replacement in the UK is estimated to be 11.6% for women and 7.1% for men⁴. Recent studies have reported that around 10% of patients are not satisfied with their hip replacement within a year following surgery^5-8. It is generally acknowledged that the key indications for surgery include joint pain, functional limitation and radiographic evidence of arthritis⁹. There is no consensus as to the severity of symptoms that indicate surgery is required¹⁰, and no universally accepted criteria to determine the indications for surgery⁹.

Obesity is a known risk factor for the development of hip osteoarthritis¹¹, and it has been shown that obese patients have a greater clinical need for surgery¹². Data from the UK National Joint Registry¹³ show that the average body mass index of patients receiving hip replacement has been increasing steadily over time. Contrary to this, there is growing evidence in the UK that commissioners are restricting access to hip replacement for obese patients stating that obesity increases the risk of complications following surgery^14-25. Accordingly NICE clinical guidelines have stated that restriction of referral for surgery based on health issues such as body mass index (BMI) has no basis in evidence and that whilst the risks of complications may be slightly higher there is no evidence supporting this as a reason to deny treatment²⁶. Regarding patient reported outcomes, literature on the effect of BMI is conflicting. Some authors conclude that obesity is associated with worse pain and functional outcomes^27-29, whilst others have found no association^{18, 30-34}. Literature reviews conclude that observed differences in risk for obese patients are small, and they can still expect large symptomatic improvement following surgery³⁵. There are several limitations within the existing literature: the sample sizes of some studies are small with few patients in the morbidly obese groups; statistical methods used are weak, such as categorising BMI reducing statistical power and selection bias due to missing data; and most importantly limited adjustment for confounding.

To our knowledge, data from a single cohort study does not exist containing the required information to adjust for all important confounding variables, and multiple data sources are therefore necessary. Within the recent literature methodology has been developed in order to combine data from multiple sources in order to adjust for a wider range of confounding factors³⁶ or allow a wider range of variables to be included in a model³⁷.

Set against the conflicting literature regarding the influence of obesity on patient reported outcomes following hip replacement, and concerns that access to surgery is being restricted for obese patients, as part of the Clinical Outcomes in Arthoplasty Study (COASt) the aim of this paper was to provide a comprehensive assessment of the effect of obesity on patient reported outcomes of hip replacement, through combining data from large prospective cohort studies allowing us to take account of a wide and comprehensive range of important confounding factors.

METHODS

As part of the COASt study access was available to data from four large prospective cohorts of patients receiving primary hip replacement (THR) for osteoarthritis. The datasets have previously been reported elsewhere and are described in brief as follows: (1) The European collaborative database of cost and practice patterns of THR (EUROHIP) contains information on 1,327 patients receiving primary THR across 20 European orthopaedic centres in 12 countries in 2002³⁸; (2) Exeter Primary Outcomes Study (EPOS) is a prospective study of 1,431 patients with a primary diagnosis of OA who had THR between 1999 and 2002³⁹; (3) Elective Orthopaedic Centre database (EOC) – a purpose built Orthopaedic treatment centre opened in 2004 performing THR for four acute NHS Trusts in South West London, UK. The EOC database includes 2,832 patients receiving primary THR for OA between 2005 and 2008^{7, 40}; (4) St. Helier Hospital outcome programme – a district general hospital serving the London Boroughs of Sutton and Merton⁵. The dataset contains 787 patients with OA receiving primary THR whose operations were undertaken from 1995 to 2007.

The primary outcome of interest is the Oxford Hip Score (OHS)⁴¹, consisting of 12 questions asking patients to describe their hip pain and function during the past 4 weeks. Each question is on a Likert scale taking values from 0-4. The total score is created by summing the responses to each of the 12 questions, ranging from 0 to 48, where 0 is the worst possible score (most severe symptoms) and 48 the best score (least symptoms). Follow up OHS questionnaires were collected at 12-months in all four studies. However, in the EOC and EUROHIP cohorts the 12-month OHS was only collected for a minority of patients. The predominant follow up for EOC was the 6-month OHS, and for the EUROHIP study the 12-month WOMAC score. We therefore derived a 12-month OHS for both of these studies in the following way: (i) EOC – 250 patients in the EOC and St. Helier datasets completed both 6 and 12-month OHS scores. Using truncated regression modelling we derived an equation to predict the 12-month OHS from the 6-month OHS (R² 50.8%); (ii) EUROHIP – 110 patients completed both the OHS and WOMAC scores at baseline and 12-months follow up. Truncated regression models were used to predict the OHS from the WOMAC score at baseline (R² 75.5%) and 12-month follow up (R² 63.4%).

The main predictor of interest is pre-operative body mass index (BMI) treated as a continuous variable. Across the cohorts data was available on a wide range of patient and surgical variables. A-priori a list of these variables was circulated to co-authors and consensus obtained on the following extensive list of potential confounders: age, sex, SF-36 mental health score, comorbidities (deep venous thrombosis, pulmonary embolism, urinary tract infection, other musculoskeletal disease, neurological, respiratory, cardiovascular, renal, hepatic disease or treatment for other medical conditions), fixed flexion range of motion (degrees), analgesic use, college education, OA in other joints, expectation of less pain, radiographic Kellgren & Lawrence (K&L) grade, American Society of Anesthesiologists (ASA) status, years of hip pain, surgical approach (anterolateral or posterior) and femoral component offset size (millimetres offset). Each study collected data on age, sex, BMI and Quality of Life, however there were differences in the other confounders recorded (Table 1).

TABLE 1. List of confounding variables available within each of the four cohort studies and distribution of the extent of missing data in each study.

Variable	EPOS (n=1267)	EUROHIP (n=851)	EOC (n=1598)	ST HELIER (n=697)
OHS pre-op	100	100	100	100
OHS 12-month	100	100	100	100
Age at operation	99	99	100	100
Sex	99	95	100	100
Body Mass Index (BMI)	94	92	21	58
Education	0	88	0	0
IMD 2004 deprivation score	0	0	99	0
EQ5D anxiety/depression pre-op	0	99	96	0
SF36 mental health score pre-op	67	0	0	0
ASA grade	0	87	23	0
Years of Hip Pain	0	99	0	23
Fixed flexion	92	0	0	0
No. of Comorbidities	100	0	0	0
Pre-op expectation of less pain	0	100	0	23
Analgesic/NSAIDs medication use	100	81	0	0
K&L grade pre-operative	0	92	0	0
No. of joints with OA	0	100	0	0
Surgical approach	72	0	0	0
Stem size (mm offset)	100	0	0	0

Cells in the table represent the percentage of data available for analysis

Statistical Methods

In accordance with Katz et al⁴² we fitted two models to describe the association with BMI on: (a) the 12-month OHS as a measure of the level of postoperative pain and functional status achieved by 12-months (the destination). Linear regression modelling is used adjusting for the baseline OHS and confounding factors; (b) change in OHS between baseline and 12-months (the journey). A repeated measures linear regression model is fitted, where the outcome is the pre and postoperative OHS, and an interaction term fitted between BMI and time, to describe the change in OHS over time within BMI categories, adjusting for confounding factors.

Primary analysis

Each of the four cohort studies was analysed separately to describe the association of BMI on outcome. Models are adjusted only for confounders of age and sex in order to construct related hypotheses in each study. Fixed effects meta-analysis using inverse variance weights is used to combine results and estimate a common effect size of BMI on outcome. We tested for evidence of heterogeneity across studies.

Secondary analysis

As each study collected data on a different set of confounders, combining studies together results in a high proportion of missing data (Table 1). Within the literature methodology has been developed to combine data from multiple data sources to adjust for a wider range of variables^{36, 37}. We use the method of Multivariate Imputation by Chained Equations (MICE) to combine the data^{43, 44} (full details of the methods are provided in a supplementary file). We included all of the covariates (as listed earlier) together with the outcome variable in the imputation model as this carries information about the missing values of predictors. Regression models were then fitted to the combined dataset to describe the association of BMI on outcome adjusting for the full range of confounding factors.

RESULTS

Prior to surgery baseline data was available on 6,377 4,413 patients receiving primary THR for OA, of whom 4,413 (69.2%) completed both baseline and 12-month follow-up OHS and were included in the analysis. Table 2 describes the characteristics of patients in each of the four studies. There were small differences between patients that did, and did not, complete the follow up questionnaire. Those that completed the follow up had better pre-operative OHS in two of the studies (EUROHIP and EOC), and in the EPOS cohort those responding to the questionnaire were younger. Importantly, BMI was similar in both completers and non-completers across all four studies. For those included in the analysis, patients in the EOC cohort had better pre-operative pain and function (as measured by the OHS), whilst those in EUROHIP had more severe pre-operative symptoms. Patients in the EUROHIP and St. Helier cohorts were slightly younger than those in the other cohorts. There were a higher proportion of men in the EUROHIP study. The distribution of BMI was similar across all four studies (Table 3) with a median of 26.8 Inter-quartile range (24.3, 30.1) with a range from 14.6 to 54.3. Separating the distribution of BMI into WHO categories⁴⁵ 24 (0.9%) of patients were underweight (BMI <18.5), 864 (31.7%) normal (BMI 18.5 to 25), 1139 (41.8%) overweight (BMI 25 to 30), 502 (18.4%) obese class I (BMI 30 to 35), 150 (5.5%) obese class II (BMI 35 to 40), and 47 (1.7%) obese class III (BMI 40+).

TABLE 2. Descriptive statistics.

Study	Variables	Non-completers	Completers	P-value§
EPOS		(n=164)	(n=1267)
	OHS pre-op	15.81 (8.15)	16.49 (7.77)	0.46
	Age at operation:	72.64 (9.93)	68.80 (9.80)	<0.001
	Sex:
	Male	70 (43%)	462 (37%)	0.13
	Female	92 (57%)	800 (63%)
	Body Mass Index (BMI)	26.30 (23.62, 29.76)	26.72 (24.39, 30.04)	0.21

		Non-completers	Completers	P-value§
EUROHIP		(n=476)	(n=851)
	OHS pre-op	13.26 (8.43)	15.67 (8.61)	< 0.001
	Age at operation:	65.68 (11.22)	65.70 (10.67)	0.98
	Sex:
	Male	199 (43%)	360 (45%)	0.68
	Female	260 (57%)	448 (55%)
	Body Mass Index (BMI)	26.72 (24.02, 29.41)	26.94 (24.69, 30.12)	0.012

		Non-completers	Completers	P-value§
EOC		(n=1234)	(n=1598)
	OHS pre-op	18.38 (8.56)	19.51 (8.77)	<0.001
	Age at operation:	70.27 (11.30)	70.73 (10.35)	0.27
	Sex:
	Male	428 (35%)	577 (36%)	0.46
	Female	801 (65%)	1018 (64%)
	Body Mass Index (BMI)	27.34 (24.35, 30.98)	26.72 (23.89, 30.44)	0.032

		Non-completers	Completers	P-value§
ST HELIER		(n=90)	(n=697)
	OHS pre-op	17.47 (7.63)	17.52 (8.30)	0.95
	Age at operation:	66.30 (14.52)	66.54 (12.01)	0.88
	Sex:
	Male	37 (41%)	278 (40%)
	Female	53 (59%)	419 (60%)	0.82
	Body Mass Index (BMI)	27.00 (23.00, 30.00)	27.00 (24.00, 30.00)	0.34

For normally distributed continuous variables numbers represent mean (standard deviation), for non-normally distributed the median (inter-quartile range)

^§To compare characteristics of completers and non-completers, a chi-squared test is used for categorical variables, a two-sample t-test for normally distributed continuous variables, and Kruskal-Wallis test for non-normally distributed variables.

TABLE 3. Distribution of BMI across the four studies.

BMI categories	EPOS	EUROHIP	EOC	ST HELIER
Underweight (<18.5)	12 (1.0%)	0 (0.0%)	6 (1.8%)	6 (1.5%)
Normal (18.5 to 25)	370 (31.0%)	218 (27.8%)	116 (34.4%)	160 (38.8%)
Overweight (25 to 30)	509 (42.7%)	363 (46.3%)	123 (36.5%)	144 (35.0%)
Obese Class I (30 to 35)	215 (18.0%)	147 (18.8%)	69 (20.5%)	71 (17.2%)
Obese Class II (35 to 40)	70 (5.9%)	46 (5.9%)	12 (3.6%)	22 (5.3%)
Obese Class III (40+)	17 (1.4%)	10 (1.3%)	11 (3.3%)	9 (2.2%)

Cells in the table represent the number (percentage) of patients in each BMI category.

Primary analysis

The results of the analyses of each individual study are displayed as a forest plot in Figure 1. For the analysis of the effect of BMI on attained post-operative pain and function, each of the studies showed a small negative effect of BMI on outcome, but this did not reach statistical significance in all studies. The overall summary estimate from the meta-analysis was statistically significant suggesting that after adjusting for age and sex, for a 5-unit increase in BMI, the 12-month OHS decreases by 0.84 95%CI (0.59 - 1.08) points.

FIGURE 1. Results of Fixed-effects meta-analysis.

Secondary analysis

For the combined dataset, in the analysis of the effect of BMI on attained post-operative pain and function, adjusting for age and sex only, for a 5-unit increase in BMI, the 12-month OHS decreases by 0.96 units 95%CI (0.57 to 1.35), p-value <0.001. Adjusting for all potential confounding factors, the effect is attenuated to 0.78 units 95%CI (0.27 to 1.28), p-value 0.003. Although the effect size is small, it is important to note that there is a cumulative linear effect where the difference in post-op OHS becomes larger with increasing BMI. For example, compared to people with a normal BMI (20 to 25), those in obese class II (BMI 35 to 40) would have a post-operative OHS that is 2.34 units lower, and those in obese class III 3.12 units lower. There was no evidence of an interaction between BMI and the preoperative OHS.

Repeated measures analysis exploring the change in OHS between baseline and 12-months, suggests that patients achieved substantial improvement (change) in OHS, regardless of their pre-operative level of BMI (Table 4). After adjusting for all confounders, patients in the normal group (BMI 18.5 to 25) had a 23.0 point change in OHS between the pre- and 12-month postoperative assessment, those in the overweight group (BMI 25 to 30) a 22.4 point change, the obese class I group (BMI 30 to 35) 22.7 points, obese class II (BMI 35 to 40) 22.2 points, and obese class III (BMI 40+) 24.2 points. Hence, there is a substantial improvement in OHS after THR across all BMI categories, which greatly outweighs the small difference in attained postoperative OHS (Figure 2).

TABLE 4. Estimates of pre- and post-operative OHS from the repeated measures regression model, including an interaction of BMI with time.

	Adjusting for age and sex		Adjusting for all confounders
	Pre-op OHS	12-month OHS	Pre-op OHS	12-month OHS
BMI categories	Mean (95%CI)	Mean (95%CI)	Mean (95%CI)	Mean (95%CI)
Underweight (<18.5)	14.01 (9.54 to 18.48)	39.31 (34.93 to 43.68)	14.04 (9.56 to 18.52)	39.34 (34.97 to 43.71)
Normal (18.5 to 25)	17.02 (15.69 to 18.34)	40.04 (38.72 to 41.36)	16.83 (15.25 to 18.40)	39.85 (38.25 to 41.45)
Overweight (25 to 30)	16.65 (15.38 to 17.91)	39.01 (37.75 to 40.28)	16.79 (15.22 to 18.36)	39.15 (37.56 to 40.75)
Obese Class I (30 to 35)	14.23 (12.81 to 15.64)	36.95 (35.54 to 38.37)	14.93 (13.13 to 16.72)	37.66 (35.93 to 39.39)
Obese Class II (35 to 40)	13.69 (11.82 to 15.57)	35.90 (34.01 to 37.79)	14.71 (12.51 to 16.91)	36.92 (34.72 to 39.11)
Obese Class III (40+)	12.25 (9.02 to 15.49)	36.43 (33.10 to 39.76)	13.66 (10.24 to 17.07)	37.83 (34.25 to 41.41)

FIGURE 2. Change on Oxford Hip Score between baseline and 12-months follow up, stratified by body mass index categories.

DISCUSSION

Main findings

This study provides a comprehensive assessment of the association of BMI on PROMs of THR. Its strength lies in utilising data from four large, representative, prospective cohort studies that allowed us to accurately assess the size of effect of BMI on PROMs and to adjust for a full range of potential confounders. Our findings confirm a small statistically significant difference in the effect of BMI on post-operative pain and function (the destination), where compared to people of normal BMI (20 to 25), those in the obese class II group (BMI 35 to 40) would have a OHS that is 2.34 points lower. It has previously been suggested that the minimum clinically important difference (MCID) (the smallest amount of change in OHS that is likely to be clinically important) is around 5 points⁴⁶, based on observations that a half standard deviation in the change score has been shown to represent a meaningful difference⁴⁷, and that the MCID maybe be as low as 2 points such that even a small change may be clinically important Hence, although statistically significant, differences in attained 12-month OHS across different categories of BMI only approach potential clinical relevance in obese classes II and III. Regardless, this effect is greatly outweighed by the substantial improvement (change) in OHS seen across all categories of BMI (the journey) following surgery where in the obese class II group (BMI 35 to 40), after adjusting for all confounders, patients achieved a 22.2 point change (improvement) in OHS over the year following surgery. The findings are consistent across studies and robust to adjustment for a wide range of confounding factors. The findings suggest that BMI should not present a barrier to access THR in terms of PROMs.

Strengths and limitations

The strengths of this study include the relatively large sample size using four separate cohorts with data collected prospectively with a good rate of follow up, the use of a reliable, valid and responsive instrument for assessing outcomes of THR⁴¹, and the generalizability of the findings using data from the UK and Europe. A further strength is the comprehensive adjustment for confounding through combining data in a subset of the studies. A limitation of the study is that we have only evaluated short term outcomes at 12-months following surgery, and it is unclear what the effect will be in the long term. In addition, only 1.6% of patients within this study had a BMI of 40+, hence we did not have the power to evaluate the effect of this most severe category of obesity on outcomes, and further work is required to determine whether there is a threshold effect above which outcomes are worse. Across the four studies, patients were operated on during different periods of time and a limitation is that there are variations in the time spans of the various cohorts. The use of multiple imputation methods present potential limitations. They require us to make the assumption data are missing at random (MAR), which is plausible in the context of this study as the reason for missing data is due to variables not being collected in the study. Plausibility is further enhanced by inclusion of a wide range of covariates to ensure enough variables predictive of missing information are included. Further limitations are that in the EUROHIP study data on OHS was only collected in a minority of patients, the majority having a WOMAC score, hence we had to derive the OHS from the recorded WOMAC score. Inconsistency in the outcome measures collected across studies is not unexpected – this has lead to the development of a core set of outcomes being supported by the COMET initiative when collecting data for future studies⁴⁸. As highlighted in a recent systematic review⁴⁹ in situations where different outcome measures have been collected it is not uncommon to map one outcome measure to another⁵⁰, as we have done in this study. Reassuringly, the results from the EUROHIP study were entirely consistent with those of the other 3 studies.

What is already known

Patient Reported Outcome Measures (PROMs) are now the favoured measure of outcome used to see whether surgery has been successful from the patient’s perspective^51-53. The PROMS most often used to assess the effects of THR include well-validated self-assessment measures of OA severity such as the Oxford Hip Score (OHS)^{41, 54} or Western Ontario and McMaster Universities (WOMAC) OA index⁵⁵, which assess pain, stiffness and function and 12-months is a commonly used and appropriate time at which to consider whether surgery has been successful⁴⁰. There is growing evidence in the UK that commissioners are restricting access to hip replacement for obese patients^14-17, stating that increasing levels of obesity increases the risk of complications following surgery. Within the literature there is some evidence to support this view that obesity increases the risk of complications, although such risks are most consistently observed amongst the morbidly obese^{18-23, 56}. There is evidence of an increased risk of infection amongst obese patients^{18, 20, 22, 24, 25, 57, 58}, but the absolute risk is small, and this is in keeping with existing literature across other types of surgery^59-63. An increased risk of thrombo-embolic events has recently been observed⁵⁸, again consistent with other areas of surgery^63-65. Although the risks of complications may be slightly higher in these patients, in accordance with NICE guidelines²⁶, there is no evidence to support these as reasons to deny treatment. Of the studies that have explored the effect of obesity on prosthesis survival, most have focused on the mid term outcomes and found no difference in survival rates^{22, 66, 67}. A recent large study of longer term outcomes over 20-years following THR found a small, but significant, effect of BMI with an increased risk of revision, particularly in the morbidly obese (BMI 40+) group, although the absolute numbers of revisions were small⁶⁸.

Regarding patient reported outcomes of hip replacement surgery, evidence from the literature on the effect of BMI is conflicting. Some authors conclude that increasing levels of obesity are associated with worse pain and functional outcomes^27-29, whilst others have found no evidence of an association^{18, 30-34}. The general consensus being that any observed differences in risk for obese patients are small and they can still expect large symptomatic improvement following hip replacement, even though the may not attain the same level of post-operative pain and function³⁵. The main limitation of existing studies is the possibility of residual or unmeasured confounding. Given current attempts made by commissioners to ration access to hip replacement on the basis of BMI, and the conflicting results of existing studies, it is important to address this and other limitations in order to strengthen the evidence of whether or not a true association exists. Our findings confirm that whilst there are statistically significant differences in the attained level of post-operative pain and function, these differences are not clinically relevant, and are greatly outweighed by the substantial improvements (change) in pain and function seen across all pre-operative categories of BMI following surgery. These findings are robust to adjustment for an extensive range of confounding factors. The findings suggest that BMI should not present a barrier to access THR in terms of PROMs.

What this study adds

This study demonstrates that pre-operative levels of BMI should not present a contra-indication for hip replacement surgery on the basis of expected improvement in patients report pain and functional outcomes. Regardless of differences in pre-operative BMI these patients can expect to achieve substantial symptomatic improvement following surgery. Although there are small significant differences in attained post-operative scores, these differences are small and not clinically important.