Total hip arthroplasty restores population health-related quality of life norms

Abstract

Aims

This study compares health-related quality of life (HRQoL) between patients undergoing primary total hip arthroplasty (THA) for osteoarthritis (OA) and a propensity-matched general population cohort. We also aimed to clarify the relationship between BMI and postoperative improvements, mediated via preoperative HRQoL.

Methods

In this retrospective study using the Edinburgh Arthroplasty database (1 January 2013 to 31 December 2022; n = 3,495) and Health Survey for England data (2010 to 2012; n = 25,320), propensity score matching (1:1) was performed based on age, sex, and BMI. The primary outcome was EuroQol five-dimension three-level questionnaire (EQ-5D-3L) index score. Secondary outcomes included EuroQol-visual analogue scale (EQ-VAS) and mediation analysis examining how preoperative EQ-5D-3L mediated the relationship between BMI and postoperative improvement.

Results

Preoperatively, THA patients had significantly lower EQ-5D-3L scores compared with matched general population (median difference: 0.280, bootstrapped 95% CIs; 0.258 to 0.306; p < 0.001). At one-year follow-up, THA patients exceeded population norms (THA median: 0.814 vs general population: 0.796, p = 0.014). Patients aged > 85 years showed the greatest magnitude of improvements, restoring EQ-5D-3L scores equivalent with their age-matched general population peers (preoperative: 0.189 vs postoperative: 0.796, general population: 0.696). Mediation analysis revealed that BMI’s negative direct effect on improvements in EQ-5D-3L was counterbalanced by stronger indirect effects transmitted through preoperative scores (indirect effects: obesity I (30 to 34.9 kg/m²): β = 0.038, p < 0.001; obesity II (35 to 39.9 kg/m²): β = 0.086, p < 0.001; obesity III (≥ 40 kg/m²): β = 0.123, p < 0.001).

Conclusion

THA was shown to restore HRQoL to that expected of a matched normal population, but in younger patients this was less than expected. Patients aged > 85 years had the greatest magnitude of restoration. Postoperative HRQoL improvement was predominantly influenced by preoperative functional status, rather than BMI alone. These findings challenge current BMI-based eligibility thresholds and support surgical prioritization based on functional impairment severity.

Cite this article: Bone Jt Open 2026;7(1):90–101.

Introduction

End-stage hip osteoarthritis (OA) profoundly diminishes health-related quality of life (HRQoL), with one-fifth of patients reporting a subjective health state ‘worse than death’.^1,2 While joint registry studies have captured the effects total hip arthroplasty (THA) has on HRQoL,^3-5 the magnitude of these improvements relative to population norms remains poorly defined. This represents a critical gap, as such comparisons provide essential context for quantifying both the burden of disease preoperatively and the degree to which THA restores HRQoL to levels of the general population.

The relationship between BMI and arthroplasty outcomes remains contentious, as current practice often employs arbitrary BMI thresholds for surgical eligibility.^6,7 Patients with elevated BMI typically begin with lower baseline HRQoL scores, yet experience greater relative improvements compared to patients with a normal BMI.^8-10 Arbitrary BMI thresholds may therefore result in the exclusion of patients who substantially benefit from arthroplasty.¹¹ Farrow et al¹² suggested that preoperative functional status may better predict postoperative improvements and guide resource allocation. However, the causal mechanisms by which BMI influences THA outcomes remain incompletely understood.

This study addresses these gaps through three objectives: 1) to quantify the HRQoL deficit in patients awaiting THA compared to a propensity score matched general population cohort; 2) to determine whether THA restores HRQoL to age, sex, and BMI-matched population norms; and 3) to clarify the causal pathways through which BMI influences postoperative improvement, transmitted through preoperative scores. Through investigating these relationships, we aim to provide an evidence-based framework to aid in guiding patient prioritization and targeted preoperative optimization, while providing information for expectation-setting across diverse demographic groups.

Methods

Study design

This retrospective study compares EuroQol five-dimension three-level questionnaire (EQ-5D-3L)¹³ scores in patients undergoing primary THA for OA and the general population. Data for analyses were sourced from two distinct databases: pre- and postoperative patient data were extracted from the Edinburgh Orthopaedic Research Database (2013 to 2022), whereas population-level EQ-5D-3L scores were obtained from the Health Survey for England (HSE, 2010 to 2012). This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.¹⁴

Primary THA cohort

The Edinburgh Orthopaedic Research Database was prospectively compiled using data collected between 1 January 2013 and 31 December 2022. The THA cohort was composed of patients undergoing primary THA for OA who completed EQ-5D-3L questionnaires during preoperative assessment and at one-year follow-up (Research Ethics Committee approval: 20/SS/0125). A flow diagram of patients included is displayed in Figure 1.

Fig. 1.

Flow diagram of included patients and analysis pathway. HRQoL, health-related quality of life; HSE, Health Survey for England; OA, osteoarthritis; THA, total hip arthroplasty.

General population cohort

Annually, the National Centre for Social Research conducts a national survey aimed at monitoring trends in health, lifestyle, and wellbeing across adults and children living in England to produce the HSE. The responses from this survey are representative of the general population and can be accessed through the UK Data Service.^15-17 HSE data (2010 to 2012) were used for direct comparability due to routine EQ-5D-3L collection. Inclusion was restricted to respondents aged ≥ 16 years (Figure 1).

Patient characteristics

Baseline characteristics of matched groups are displayed in Table I.

Table I.

Propensity score matched cohorts 1:1 – general population (n = 25,320) vs preoperative group (n = 3,495).

Covariates	General population		THA cohort	SMD
	Unmatched	Matched		Unmatched	Matched
Total, n	25,320	3,495	3,495
Median age, yrs (IQR)	49.0 (35.0 to 64.0)	69.0 (61.0 to 76.0)	69.1 (61.2 to 76.3)	0.783	0.003
Male	1,1204	1,393	1,393	0.059	0.000*
Female	14,116	2,102	2,102	0.059	0.000*
Median BMI, kg/m2 (IQR)	26.7 (23.7 to 30.3)	27.9 (24.9 to 31.4)	27.6 (24.5 to 31.2)	0.113	0.027

Standardized mean difference (SMD) values represent effect sizes indicating balance between groups; SMD < 0.1 indicates good balance.

^*Exact matching was used for sex (SMD = 0.000).

THA, total hip arthroplasty.

Primary outcome

Our primary objective was to compare the difference in EQ-5D-3L scores between the THA and general population, pre- and postoperatively. EQ-5D-3L scores were calculated from the five dimensions: mobility, pain/discomfort, anxiety/depression, ability to perform daily activities, and self-care.¹⁸ The EQ-5D-3L scores were calculated by applying UK-specific value sets to the health state data, reflecting the relative importance the UK population assigns to different health conditions.¹⁹ Scores were generated using the Time Trade-Off value set for the UK,²⁰ with values ranging from -0.594 to 1, where 1 represents ‘perfect health’ and negative values indicate health states considered worse than death.

Secondary outcomes

Secondary outcomes included comparisons of EuroQol visual analogue scale (EQ-VAS) scores between the THA cohort and the general population. As a component of the EQ-5D-3L questionnaire, the EQ-VAS captures respondents’ overall health assessment on a scale from 0 (‘worst health imaginable’) to 100 (‘best health imaginable’).²¹ We performed a mediation analysis to uncover the underlying relationship between BMI and postoperative improvements in EQ-5D-3L.

Missing data handling

For our main analysis containing both the THA and HSE cohorts, analysis identified structured patterns of missing data that would have resulted in exclusion of 26.6% of cases through listwise deletion (Figure 1, Supplementary Material). Little’s missing completely at random (MCAR) test (chi-squared test = 7,153.8, degrees of freedom (df) = 1,207, p < 0.001), combined with significant associations between missingness and observed variables (p < 0.001), indicated missing at random (MAR) rather than missing completely at random. For the mediation analysis using only the THA cohort with additional comorbidity covariates (diabetes (DM), myocardial infarction (MI), congestive heart failure (CHF), peripheral vascular disease (PVD), chronic obstructive pulmonary disease (COPD), cerebrovascular disease (CVD), other joint pain), missing data analysis revealed that listwise deletion would have resulted in 36.6% due to comorbidity data. Little’s MCAR test (chi-squared test = 1,592.2, df = 421, p < 0.001). In each dataset associations between observed and missing variables (p < 0.001), supported the assumption that data were missing at random.²² We used multiple imputation by chained equations using the mice package in R (R Foundation for Statistical Computing, Austria),²³ with a fully conditional specification model adapted for HRQoL data. Predictive mean matching was used for continuous variables (height, weight, EQ-VAS),²⁴ while ordinal EQ-5D-3L dimensions were imputed separately using proportional odds models, and logistic regression for comorbidity variables. We generated 100 imputations with 50 iterations for each dataset to ensure stable estimates.²⁵ Convergence was verified using the Gelman-Rubin statistic (PSRF = 1.0). Derived variables were calculated post-imputation.²⁶ Sensitivity analysis confirmed consistency between complete-case and imputed datasets, with minimal differences (< 1% change).

Propensity score matching

Using MatchIt,²⁷ we performed 1:1 nearest-neighbour matching with a 0.2 SD calliper without replacement based on age, BMI, and exact matching for sex, using logistic regression modelling.²⁸ Two matched cohorts were created following multiple imputations (preoperative THA vs general population and postoperative THA vs general population), producing 3,495 in each group, in each imputed dataset. Balancing of matching was assessed by estimating the standardized mean difference (SMD) for each covariate. Sensitivity analyses were completed comparing 1:1 vs 1:2 vs 1:3 matching, with 1:1 matching selected for optimal balance (maximum SMD: 0.027) while maintaining adequate power.

Mediation analysis

We constructed separate mediation models for each BMI category; underweight (≤ 18.4 kg/m²), overweight (25 to 29.9 kg/m²), obesity class I (30 to 34.9 kg/m²), obesity class II (35 to 39.9 kg/m²), and obesity class III (≥ 40 kg/m²), with normal BMI (18.5 to 24.9 kg/m²) as the reference category. We hypothesized that elevated BMI (exposure) leads to lower preoperative EQ-5D-3L scores (mediator), which in turn leads to altered capacity for improvement post-THA (outcome). This specification is supported by biomechanical evidence,²⁹ observational studies,^30,31 and longitudinal data showing that BMI remains relatively stable before and after THA.^32,33 We investigated: BMI category’s effect on preoperative EQ-5D-3L (Path A), preoperative EQ-5D-3L’s influence on postoperative change (Path B), and BMI category’s direct effect on postoperative change controlling for preoperative EQ-5D-3L (Path C’). This disentangled BMI’s direct compared with indirect effects through preoperative status. All models were adjusted for age, sex, DM, MI, CHF, COPD, CVD, PVD, and any other joint pain.

Indirect effects (A × B) were calculated using the product-of-coefficients method with bias-corrected and accelerated 95% CIs from 5,000 bootstrap resamples.^34-36 Analyses were conducted on each of the 100 imputed datasets, with results pooled using Rubin’s rules. The Benjamini-Hochberg procedure controlled for multiple comparisons.³⁷ Model assumptions were tested with Shapiro-Wilk and Breusch-Pagan tests, applying heteroscedasticity-consistent (HC3) robust standard errors where needed.^38,39 Multiple statistical approaches showed consistent estimates (< 1.5% variation), and alternative model specifications produced minimal changes (< 6%) despite violations of distributional assumptions. Multicollinearity was assessed using variance inflation factors (VIF) with a threshold of < 5. All VIF values were < 1.1, including the relationship between BMI and baseline EQ-5D-3L, indicating no evidence of problematic collinearity. Sensitivity to unmeasured confounding was quantified using E-values for mediation analysis.⁴⁰ Our approach follows A Guideline for Reporting Mediation Analyses guidelines.⁴¹

Statistical analysis

All statistical analyses were conducted using R version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria). Differences between cohorts were assessed based on the median difference and its 95% CI. For our primary analyses, in each of the imputed datasets (m = 100), 5,000 BCa bootstrap resamples were completed with replicates pooled across imputations. The median of the pooled distribution served as our point estimate; BCa 95% CIs and two-tailed p-values were calculated. If the BCa algorithm failed to converge in any dataset, we employed a hierarchical fallback strategy. For stratified analyses involving multiple comparisons, bootstrap p-values were adjusted using the Benjamini-Hochberg method, p < 0.05 was considered significant.³⁷ In our primary analysis comparing EQ-5D-3L of the pre- and postoperative THA cohorts with the general population, propensity score matching was used so that meaningful comparisons could be made. Comparisons of EQ-5D-3L dimensions between the THA cohort and general population controls were completed using the Mann-Whitney U test. The mediation analysis was conducted using the THA cohort without propensity score matching with a control group, as this analysis examined within-group relationships rather than between-group comparisons.

Results

Matched cohorts: preoperative THA and general population

Patients awaiting THA had significantly lower EQ-5D-3L compared with the matched general population (median difference (MD) 0.280, bootstrapped 95% CI 0.258 to 0.306; p < 0.001) (Table II, Figure 2). This deficit was most pronounced in younger patients (aged < 45 years: MD 0.707, 95% CI 0.484 to 0.912; p < 0.001) and females (MD 0.532, 95% CI 0.280 to 0.601; p < 0.001). Similarly, EQ-VAS scores were significantly lower across all preoperative groups (Table III, Figure 3). Preoperative patients had higher rates of extreme problems across all EQ-5D-3L domains, indicating functional impairment beyond pain and mobility (Table IV, Figure 4).

Table II.

Summary of EuroQol five-dimension three-level index scores: matched cohorts (n = 3,495 in each group).

Variable	Median EQ-5D-3L (IQR)			Difference (95% CI)*	p-value†	Median postop EQ-5D-3L (IQR)	Difference (95% CI)‡	p-value†
	General population (unmatched)	General population (matched)	Preop
Sex
Male	1.000 (0.796 to 1.000)	0.848 (0.725 to 1.000)	0.587 (0.088 to 0.691)	0.261 (0.255 to 0.307)	< 0.001	0.850 (0.691 to 1.000)	-0.002 (-0.087 to 0.033)	0.578
Female	0.850 (0.727 to 1.000)	0.796 (0.691 to 1.000)	0.264 (0.055 to 0.691)	0.532 (0.280 to 0.601)	< 0.001	0.796 (0.691 to 1.000)	0.000 (-0.003 to 0.015)	1.000
Age, yrs
Under 45	1.000 (0.848 to 1.000)	1.000 (0.809 to 1.000)	0.293 (-0.016 to 0.691)	0.707 (0.484 to 0.912)	< 0.001	0.796 (0.620 to 1.000)	0.204 (0.054 to 0.273)	0.003
45 to 54	1.000 (0.796 to 1.000)	1.000 (0.727 to 1.000)	0.516 (0.088 to 0.691)	0.484 (0.261 to 0.636)	< 0.001	0.848 (0.691 to 1.000)	0.152 (-0.035 to 0.204)	0.348
55 to 64	0.848 (0.725 to 1.000)	0.848 (0.725 to 1.000)	0.516 (0.055 to 0.691)	0.332 (0.282 to 0.405)	< 0.001	0.848 (0.691 to 1.000)	0.000 (-0.087 to 0.034)	0.887
65 to 74	0.796 (0.710 to 1.000)	0.796 (0.691 to 1.000)	0.587 (0.088 to 0.691)	0.209 (0.177 to 0.234)	< 0.001	0.848 (0.691 to 1.000)	-0.052 (-0.087 to -0.018)	0.021
75 to 84	0.760 (0.656 to 1.000)	0.760 (0.656 to 1.000)	0.260 (0.055 to 0.691)	0.500 (0.244 to 0.601)	< 0.001	0.814 (0.691 to 1.000)	-0.054 (-0.087 to 0.000)	0.097
85 and older	0.725 (0.587 to 0.850)	0.696 (0.612 to 0.848)	0.189 (-0.003 to 0.620)	0.507 (0.175 to 0.568)	< 0.001	0.796 (0.620 to 1.000)	-0.100 (-0.123 to -0.019)	0.142
Total	1.000 (0.760 to 1.000)	0.796 (0.691 to 1.000)	0.516 (0.055 to 0.691)	0.280 (0.258 to 0.306)	< 0.001	0.814 (0.691 to 1.000)	-0.018 (-0.052 to -0.016)	0.014

All analyses were performed on multiply imputed datasets (m = 100). Median differences were calculated based on pooled medians for each group across imputations. CIs and p-values were calculated from pooled bootstrap replicates (from 5,000 resamples per imputation). CIs were primarily derived using the bias-corrected and accelerated (BCa) method, with fallback to alternatives when necessary. p-values smaller than 0.001 are reported as < 0.001. CI, bootstrapped 95% CIs (5,000 resamples).

^*General population (matched) vs preoperative THA; positive values indicate higher EQ-5D-3L scores in general population.

^†Bootstrap-based p-values for stratified comparisons (by sex or age) were adjusted using the Benjamini-Hochberg method.

^‡General population (matched) vs postoperative THA; negative values indicate higher EQ-5D-3L scores in postoperative THA.

EQ-5D-3L, EuroQol five-dimension three-level questionnaire; THA, total hip arthroplasty.

Fig. 2.

Box and whisker with density plot of EuroQol five-dimension three-level questionnaire (EQ-5D-3L) Index scores by group (matched cohorts).

Table III.

Summary of EuroQol visual analogue scale scores: matched cohorts (n = 3,495 in each group).

Variable	Median EQ-VAS (IQR)			Difference (95% CI)*	p-value†	Median postop EQ-VAS (IQR)	Difference (95% CI)‡	p-value†
	General population (unmatched)	General population (matched)	Preop
Sex
Male	80.0 (70.0 to 90.0)	80.0 (65.0 to 90.0)	70.4 (54.0 to 81.0)	9.6 (8.4 to 10.0)	< 0.001	81.2 (70.0 to 90.5)	-1.2 (-4.0 to -0.4)	< 0.001
Female	80.0 (70.0 to 90.0)	80.0 (60.0 to 90.0)	70.0 (50.0 to 80.5)	10.0 (9.2 to 12.2)	< 0.001	80.6 (70.0 to 90.9)	-0.6 (-3.7 to -0.2)	0.009
Age, yrs
Under 45	80.0 (70.0 to 90.0)	80.0 (70.0 to 90.0)	65.0 (50.0 to 80.0)	15.0 (10.0 to 24.0)	< 0.001	80.0 (66.0 to 90.0)	0.0 (-0.5 to 7.0)	1.000
45 to 54	80.0 (70.0 to 90.0)	80.0 (70.0 to 90.0)	70.0 (53.5 to 84.0)	10.0 (10.0 to 11.0)	< 0.001	86.0 (74.0 to 92.0)	-6.0 (-10.0 to -2.0)	0.003
55 to 64	80.0 (70.0 to 90.0)	80.0 (70.0 to 90.0)	70.2 (50.0, 81.8)	9.8 (8.0 to 10.0)	< 0.001	86.5 (70.1 to 93.0)	-6.5 (-10.0 to -2.0)	< 0.001
65 to 74	80.0 (69.0 to 90.0)	80.0 (65.0 to 90.0)	70.4 (59.0 to 81.0)	9.7 (8.0 to 10.0)	< 0.001	81.1 (70.0 to 91.0)	-1.1 (-4.2 to -0.6)	0.001
75 to 84	75.0 (60.0 to 85.0)	75.0 (60.0 to 85.0)	70.0 (50.2 to 80.0)	5.0 (0.0 to 8.0)	0.093	80.0 (70.0 to 90.0)	-5.0 (-10.0 to -2.0)	0.003
Over 85	70.0 (50.0 to 80.0)	70.0 (50.0 to 80.0)	69.9 (50.0 to 80.0)	0.2 (-10.0 to 5.0)	1.000	80.0 (69.8 to 90.0)	-10.0 (-20.0 to -10.0)	< 0.001
Total	80.0 (70.0 to 90.0)	80.0 (60.0 to 90.0)	70.0 (50.3 to 80.7)	10.0 (9.6 to 10.5)	< 0.001	81.0 (70.0 to 90.8)	-1.0 (-2.0 to -0.4)	< 0.001

All analyses were performed on multiply imputed datasets (m = 100). Median differences were calculated based on pooled medians for each group across imputations. CIs and p-values were calculated from pooled bootstrap replicates (from 5,000 resamples per imputation). CIs were primarily derived using the bias-corrected and accelerated (BCa) method, with fallback to alternatives when necessary. p-values smaller than 0.001 are reported as < 0.001.

CI, bootstrapped 95% CIs (5,000 resamples).

^*General population (matched) vs preoperative THA; positive values indicate higher EQ-VAS scores in general population.

^†Bootstrap-based p-values for stratified comparisons (by sex or age) were adjusted using the Benjamini-Hochberg method.

^‡General population (matched) vs postoperative THA; negative values indicate higher EQ-VAS scores in postoperative THA.

EQ-VAS, EuroQol visual analogue scale; THA, total hip arthroplasty.

Fig. 3.

Box and whisker plot EuroQol visual analogue scale (EQ-VAS) scores by group and sex (matched cohorts).

Table IV.

Descriptive summary of EuroQol five-dimension three-level questionnaire components: matched cohorts.

Component	General population, n (%)	Preop, n (%)	p-value*	Postop, n (%)	p-value†
Mobility			< 0.001		0.565
1	2,249 (64.3)	399 (11.4)		2,222 (63.6)
2	1,236 (35.4)	2,979 (85.2)		1,262 (36.1)
3	10 (0.3)	117 (3.3)		11 (0.3)
Problems‡	1,246 (35.7)	3,096 (88.6)		1,273 (36.4)
Self-care			< 0.001		< 0.001
1	3,130 (89.6)	2,237 (64.0)		2,898 (82.9)
2	342 (9.8)	1,202 (34.4)		570 (16.3)
3	23 (0.7)	56 (1.6)		27 (0.8)
Problems‡	365 (10.4)	1,258 (36.0)		597 (17.1)
Usual activities			< 0.001		< 0.001
1	2,491 (71.3)	493 (14.1)		2,143 (61.3)
2	892 (25.5)	2,435 (69.7)		1,268 (36.3)
3	112 (3.2)	567 (16.2)		84 (2.4)
Problems‡	1,004 (28.7)	3,002 (85.9)		1,352 (38.7)
Pain and discomfort			< 0.001		< 0.001
1	1,654 (47.3)	78 (2.2)		1,944 (55.6)
2	1,577 (45.1)	1,922 (55.0)		1,429 (40.9)
3	264 (7.6)	1,495 (42.8)		122 (3.5)
Problems‡	1,841 (52.7)	3,417 (97.8)		1,551 (44.4)
Anxiety and depression			< 0.001		< 0.001
1	2,613 (74.8)	2,195 (62.8)		2,862 (81.9)
2	811 (23.2)	1,158 (33.1)		564 (16.1)
3	71 (2.0)	142 (4.1)		69 (2.0)
Problems‡	882 (25.2)	1,300 (37.2)		633 (18.1)

EuroQol five-dimension three-level questionnaire domain level definitions: 1 = no problems, 2 = moderate problems, 3 = extreme problems (n = 3,495 in each group).

^*p-value from pooled Mann-Whitney U test comparing distributions between General population and Preoperative groups across 100 imputations.

^†p-value from pooled Mann-Whitney U test comparing distributions between General population and Postoperative groups across 100 imputations.

^‡Combined moderate and extreme problems (levels 2 and 3).

Fig. 4.

a) Radar plots of a) postoperative and b) preoperative EuroQol five-dimension three-level (EQ-5D-3L) questionnaire domain scores: matched cohorts. Each pentagon represents possible domain scores (inner = no problems, middle = some problems, outer = extreme problems).

Matched cohorts: postoperative THA and general population

At one-year follow-up, THA not only restored but exceeded population-normative HRQoL, with patients achieving significantly higher EQ-5D-3L scores than matched controls (median: 0.814 vs 0.796, MD -0.018, 95% CI -0.052 to -0.016; p = 0.014) (Table II). Age-stratified analysis showed older patients (aged ≥ 85 years) demonstrated the greatest magnitude of improvement from preoperative scores (preoperative: 0.189 vs general population: 0.696, postoperative: 0.796), while those aged < 45 years remained below population norms despite substantial improvements (MD 0.204, 95% CI 0.054 to 0.273; p = 0.003). This pattern of exceeding population norms was also observed in EQ-VAS scores (MD -1.0, 95% CI -2.0 to -0.4; p < 0.001) (Table III), with the oldest patients (aged > 85 years) reporting the most substantial improvements compared with their age-matched peers (MD -10.0, 95% CI -20.0 to -10.0; p < 0.001). Dimensional analysis revealed domain-specific recovery patterns: postoperative mobility problems matched population rates, pain/discomfort and anxiety/depression improved beyond population levels (p < 0.001), while self-care and usual activities showed significant improvement but remained below population benchmarks (p < 0.001).

Mediation analysis

Baseline characteristics by BMI category are displayed in the Supplementary Material. Mediation analysis revealed that preoperative EQ-5D-3L scores mediated BMI’s influence on outcomes (Table V, Figure 5 and Figure 6). Baseline scores showed a clear gradient: normal (median 0.587; IQR 0.088 to 0.691), obesity I (0.433; IQR 0.055 to 0.691), obesity II (0.159; IQR -0.016 to 0.689), and obesity III (0.091; IQR -0.004 to 0.620). Path A (BMI to preoperative score) demonstrated significantly lower baseline scores with increasing obesity (class I: β = -0.046, p < 0.01; II: β = -0.105, p < 0.001; III: β = -0.150, p < 0.001).

Table V.

Summary of EuroQol five-dimension three-level questionnaire (EQ-5D-3L) by BMI category and mediation analysis of BMI category (independent) effects on postoperative change in scores (dependent): preoperative EQ-5D-3L as mediator (reference category = normal BMI) (n = 3,495).

BMI category	N (%)	Median preop EQ-5D (IQR)	Median postop EQ-5D (IQR)	Median difference (95% CI)*	Path A (95% CI)†	Path B (95% CI)†	Path C’ (95% CI)†	A × B indirect effect (95% CI)‡	Total effect (95% CI)	Proportion mediated, %	Mediation type
Normal	974 (27.9)	0.587 (0.088 to 0.691)	0.850 (0.710 to 1.000)	0.310 (-0.168 to 1.016)	-	-	-	-	-	-	Reference
Underweight	49 (1.4)	0.516 (0.088 to 0.691)	0.805 (0.638 to 1.000)	0.309 (-0.092 to 0.941)	-0.002 (-0.105 to 0.102)	-0.822 (-0.846 to -0.797)§	-0.064 (-0.141 to 0.013)	0.001 (-0.071 to 0.075)	-0.063 (-0.176 to 0.051)	2.0	Partial
Overweight	1,295 (37.1)	0.587 (0.088 to 0.691)	0.849 (0.691 to 1.000)	0.309 (-0.175 to 1.074)	-0.006 (-0.036 to 0.023)	-0.822 (-0.846 to -0.797)§	-0.001 (-0.022 to 0.020)	0.005 (-0.017 to 0.027)	0.004 (-0.028 to 0.036)	122.4	Inconsistent
Obesity I	737 (21.1)	0.433 (0.055 to 0.691)	0.796 (0.691 to 1.000)	0.380 (-0.250 to 1.068)	-0.046 (-0.080 to -0.012)¶	-0.822 (-0.846 to -0.797)§	-0.026 (-0.050 to -0.001)**	0.038 (0.012 to 0.064)§	0.012 (-0.026 to 0.049)	316.1	Inconsistent
Obesity II	324 (9.3)	0.159 (-0.016 to 0.689)	0.796 (0.620 to 1.000)	0.428 (-0.279 to 1.177)	-0.105 (-0.151 to -0.060)§	-0.822 (-0.846 to -0.797)§	-0.064 (-0.097 to -0.032)§	0.086 (0.050 to 0.123)§	0.022 (-0.027 to 0.071)	390.3	Inconsistent
Obesity III	116 (3.3)	0.091 (-0.004 to 0.620)	0.727 (0.587 to 1.000)	0.526 (-0.428 to 1.083)	-0.150 (-0.222 to -0.079)§	-0.822 (-0.846 to -0.797)§	-0.083 (-0.136 to -0.031)¶	0.123 (0.072 to 0.174)§	0.040 (-0.036 to 0.116)	307.4	Inconsistent

Partial mediation: the direct (Path C’) and indirect (A × B) effects have the same direction, with the mediator explaining only part of the total effect.

Inconsistent mediation: the direct (Path C’) and indirect (A × B) effects have opposite directions, causing the indirect effect to counteract (suppress) the direct effect. This results in a total effect smaller than either the direct or indirect effect, leading to a proportion mediated > 100%.

^*Preoperative vs postoperative EQ-5D-3L scores.

^†Mediation paths A, B, C, and indirect effects are presented with 95% CIs. All estimates were derived from multiple linear regression models adjusted for age, sex, diabetes, myocardial infarction, congestive heart failure, chronic pulmonary obstructive disease, peripheral vascular disease, cerebral vascular disease, and other joint pain; pooled across 100 imputations using Rubin's rules.

^‡CIs for indirect effects were calculated using bias-corrected bootstrapping with 5,000 resamples.

^§p < 0.001.

^¶p < 0.01.

^**p < 0.05.

Fig. 5.

a) Mediation analysis pathway diagram, displaying the relationship between BMI category (obesity classes I to III), preoperative EuroQol five-dimension three-level questionnaire (EQ-5D-3L) score, and postoperative change in EQ-5D-3L scores, adjusting for age and sex. b) Mediation analysis pathway diagram, displaying the relationship between BMI category (obesity classes I to III), preoperative EQ-5D-3L score, and postoperative change in EQ-5D-3L scores, adjusting for age and sex. c) Mediation analysis pathway diagram, displaying the relationship between BMI category (obesity classes I to III), preoperative EQ-5D-3L score, and postoperative change in EQ-5D-3L scores, adjusting for age and sex.

Fig. 6.

Direct compared with indirect mediation effects of obesity classes on postoperative EuroQol five-dimension three-level questionnaire scores following total hip arthroplasty, showing inconsistent mediation.

Path B (preoperative score leads to postoperative score change) showed a consistent negative relationship (β = -0.822, p < 0.001), indicating greater improvement potential with lower baseline scores. Path C’ (BMI leads to postoperative score change) was negative and increased with BMI severity (obesity I: β = -0.026, p = 0.040; II: β = -0.064, p < 0.001; III: β = -0.083, p = 0.002), suggesting obesity directly impairs improvement potential.

The indirect effects (Path A × B) operated in opposite directions to direct effects (Path C’), becoming increasingly positive with higher BMI (obesity I: β = 0.038, 95% CI 0.012 to 0.064; p < 0.001; II: β = 0.086, 95% CI 0.050 to 0.123; p < 0.001; III: β = 0.123, 95% CI 0.072 to 0.174; p < 0.001), demonstrating that lower preoperative scores associated with obesity created greater potential for relative improvement. This pattern represents inconsistent mediation, where direct and indirect effects have opposite signs, resulting in suppression of total effects (Figure 6). E-values for the indirect effect showed increasing resistance to unmeasured confounding with higher BMI (obesity class I: 1.71; class II: 2.45; class III: 3.11).

Discussion

This study quantifies the substantial HRQoL deficits among patients awaiting primary THA for OA, compared to propensity score matched general population controls. At one-year follow-up THA was shown to restore HRQoL to that expected of a matched normal population, and in some this was exceeded, highlighting the broader impact of THA beyond pain relief and improved mobility alone. Yapp et al⁴² demonstrated similar restorative effects of total knee arthroplasty (TKA) for OA, collectively reinforcing the fact that both THA and TKA are transformative interventions capable of restoring profound HRQoL deficits to population levels.

Older patients (aged > 65 years), particularly those aged > 85 years, demonstrated the greatest magnitude of improvement from baseline scores, restoring HRQoL to their age-matched general population peers (EQ-5D-3L preoperative: 0.189 vs postoperative: 0.796, general population: 0.696). Aalund et al⁴³ have noted that older adults experience greater improvements in HRQoL than younger populations. In a prospective study as part of the Special Orthopaedic Geriatrics trial, Reinhard et al⁴⁴ reported similar substantial improvements in HRQoL among 101 geriatric patients (mean age 78.1 years (SD 4.9)) while also acknowledging the increased risk of complications in this multimorbid and frail cohort. Consequently, the authors strongly advocate for preoperative Comprehensive Geriatric Assessment to identify modifiable risk factors and optimize comorbidities, in conjunction with postoperative Geriatric Co-Management, which has been shown to improve outcomes.^44-46

We observed that preoperatively, female patients experienced considerably lower HRQoL compared to males, consistent with Steinbeck et al,⁴⁷ but also compared with the matched female general population. Hawker et al⁴⁸ found females with more severe symptoms were less likely to undergo arthroplasty. The underlying cause of this sex disparity is unclear, but may stem from female patients with chronic pain receiving inferior care due to sex-related bias,⁴⁹ contributing to inequalities in referral,⁵⁰ or patient concerns that treatment may lead to disruptions in their role as family caregiver.^51,52 Encouragingly, at one-year follow-up, female patients reported dramatic relative improvements, and THA restored HRQoL to general population levels.

Conversely, younger patients (aged < 45 years) showed the greatest preoperative HRQoL deficit and, despite remarkable improvements, uniquely remained below normative values postoperatively. Rolfson et al⁵ in their Swedish Registry study reported that patients aged < 44 years had lower relative improvements when compared with older patients, possibly reflecting higher functional demands and expectations of younger adults that exceed even modern prostheses’ capabilities. Nevertheless, despite achieving lower EQ-5D-3L scores, these younger patients reported postoperative EQ-VAS scores equivalent to their age-matched peers, suggesting meaningful perceived improvements even if functional recovery was incomplete.

Our mediation analysis revealed that preoperative functional status significantly mediates the relationship between BMI and improvement after THA. For patients with BMI ≥ 30 kg/m², inconsistent mediation was observed, where obesity directly impairs improvement (negative direct effect) while simultaneously creating greater improvement potential through lower baseline scores (positive indirect effect), with the latter effect predominating in determining relative improvements. Our findings provide an explanation of the clinical paradox where obese patients show larger relative HRQoL gains despite achieving lower absolute outcomes.^8,12 Elcock et al,¹¹ in their study of patients with higher BMIs undergoing TKA, reported that despite the increased rate of complications in obesity class III patients, TKA remained a cost-effective treatment. Targeted preoperative optimization of high BMI patients may enhance absolute outcomes while reducing complications,⁵³ as demonstrated by recent evidence showing that glucagon-like peptide 1 (GLP-1) receptor agonists before arthroplasty significantly reduced postoperative complications.⁵⁴ These insights challenge arbitrary BMI-based eligibility thresholds, suggesting surgical prioritization should focus on functional impairment severity.

Our study’s strengths include the robust methodological rigour using propensity matching, statistical validation, and a large representative patient cohort. Additionally, the novel mediation analysis provides valuable insights into BMI’s complex relationship with surgical outcomes, directly informing clinical decision-making. Limitations include only one-year follow-up which, although clinically meaningful, means that we did not capture longer-term functional outcome trends. While E-values generated for our mediation analysis suggest acceptable robustness for indirect effects, we acknowledge that unmeasured confounding cannot be definitively excluded in observational research. The EQ-5D-3L instrument, while widely used, is known for its ceiling effects, particularly in patients achieving higher functional outcomes. However, in our study we observed a varied distribution of postoperative scores below the ceiling across age, sex, and BMI categories. We recognize the temporal mismatch between the THA (2013 to 2022) and HSE (2010 to 2012) cohorts. However, the documented relative stability of EQ-5D-3L in the general population of the UK across most age groups, over two decades, suggests this mismatch is unlikely to significantly impact our findings.⁵⁵ We acknowledge that the temporal relationship between preoperative HRQoL and BMI category cannot be proven given this study design. Patients awaiting THA may gain modest weight due to immobility, however the combination of BMI stability over time;^32,33 the clear dose-response gradient of decreasing EQ-5D-3L with increasing BMI; and consistent observational evidence of lower EQ-5D-3L in higher BMI patients supports our specification of BMI as the exposure rather than the mediator.^30,31 The reverse causation hypothesis that functional impairment causes obesity would require implausibly precise mechanisms whereby specific EQ-5D-3L scores produced large, predictable, and temporally consistent weight gains sufficient to shift patients across BMI categories. Finally, BMI-based selection bias may exclude patients who could benefit from surgery, potentially underestimating true benefits.

Future studies should include longer follow-up periods and EQ-5D-5L to enhance clinical applicability. Our mediation analysis has revealed inconsistent mediation, with the direct effect of BMI acting in opposition with the indirect effect, mediated through preoperative EQ-5D-3L score, on change in postoperative improvement. Future research should aim to identify whether the suppressive direct effects in higher BMI categories on postoperative improvement may be overcome by structured weight management programmes, which include the use of GLP-1 agonists and their variants. These studies should take note of the induced catabolic state association with the use of these medications, and aim to identify optimal windows and safe rate of weight loss, preoperative medication cessation timing, effects of sarcopenia, and longer-term arthroplasty-related risk profiles.

In conclusion, this study highlights the profound HRQoL deficits experienced by patients awaiting THA for OA, compared to matched general population norms. At one-year follow-up, THA restored HRQoL to normative general population levels, with particularly notable improvements observed among older adults and female patients. Our mediation analysis demonstrates that postoperative improvements are predominantly driven by preoperative functional impairment rather than BMI alone. These findings challenge current BMI-based surgical eligibility thresholds, suggesting instead that clinical prioritization should focus on preoperative functional status.

Take home message

- Total hip arthroplasty restores health-related quality of life to population norms across propensity score matched sex and age stratifications at one year postoperatively.

- Our mediation analysis of BMI reveals that postoperative improvement is driven primarily by preoperative functional impairment, rather than BMI alone.

- These findings challenge BMI-based eligibility thresholds and support improving equitable access to total hip arthroplasty.

Data Availability

The datasets generated and analyzed in the current study are not publicly available due to data protection regulations. Access to data is limited to the researchers who have obtained permission for data processing. Further inquiries can be made to the corresponding author.