The economic benefit of hip replacement: a 5-year follow-up of costs and outcomes in the Exeter Primary Outcomes Study.

OBJECTIVES: To assess changes in quality of life and costs of patients undergoing primary total hip replacement using the Exeter prosthesis compared with a hypothetical 'no surgery' group.
DESIGN: The incremental quality of life, quality-adjusted life years (QALYs) and cost of Exeter Primary Outcomes Study patients was compared with hypothetical 'no surgery' group over 5 years. Scores from annual SF-36 assessments were converted into utility scores using an established algorithm and the QALY gains calculated from pre-operative baseline scores. Costs included implant costs and length of stay.
SETTING: Secondary care hospitals. PARTICIPANTS: Patients receiving a primary Exeter implant enrolled in five of seven Exeter Primary Outcomes Study centres.
RESULTS: On average, patients gained around 0.8 QALYs over 5 years. Younger and male patients or those with lower body mass index and poorer Oxford Hip Scores were significantly associated with increased QALYs. Treatment costs for a primary episode of care were just over £5000 (95% CI £4588 to £5812) per patient. Compared with 'no surgery', the cost per QALY was £7182 (95% CI £6470 to £7678), and this remained stable when key cost parameters were varied. The most likely cost per QALY was between £7058 and £7220. Older patients (age 75+) cost more, mainly due to longer average hospital stays and had a higher cost per QALY, although this remained below £10 000.
CONCLUSIONS: 85% of cases had a cost of <£20 000 per QALY (with 70% having a cost per QALY under £10 000) compared with no surgery. Cases would be considered cost-effective under currently accepted thresholds (£25 000-£30 000) compared with 'no surgery'. However, depending on age and severity, younger patients and more severe patients had below average cost per QALYs. These results help to confirm the long-term benefits and cost-effectiveness of total hip replacement in a wide variety of patients using well-established implant models such as the Exeter. However, further and ongoing economic appraisal of this and other models is required for comparative purposes.

Introduction

Hip replacement is one of a few cardinal and successful operations in the NHS and yet it has mainly gone unchallenged from a cost-effective perspective since its inception. However, with the current financial environment in the NHS, it is important to reassess the costs and benefits and to dispel any uncertainties surrounding its cost-effectiveness for the majority of patients.

Long-term studies of hip replacements have not provided good conclusive economic proof, which is surprising, given the longevity of the procedure. However, implant models come and go and many get modified, so the field is not static for economic assessment. It is therefore important to assess those that are stable and have endured in practice over many years. The orthopaedic outcome literature is mainly based on studies of the implant's survival and few have incorporated a full economic analysis. Most economic studies nowadays rely on modelling, given the limitations of the randomised controlled trial design and well-designed studies compare alternative models.1–3 Good economic studies of total hip replacement (THR) also need an appropriate measure of patient outcome (preferably in terms of changes in health-related utility) as well as a robust costing.4 5 This study has addressed these issues but further work is now ongoing to assess the longer-term cost-effectiveness.

Patients and methods

The Exeter Primary Outcomes Study (EPOS) is one of the largest longitudinal studies of a single prosthesis undertaken in the UK and continues to follow-up recipients for 10 years post-operatively. In this seven-centre study, a case series of 1589 patients underwent hip replacement with the Exeter implant between March 1999 and February 2002. This retrospective economic study was undertaken at the 5-year follow-up stage in patients who had the required outcome data available. These patients were compared with a hypothetical ‘no surgery’ group in terms of their additional costs and outcomes (quality of life (QoL)) from baseline.

The cost of an episode of care (including the operation) was estimated using the average cost of the implant (in the year the operation was performed) and the actual number of bed days used per patient at the Department of Health reference cost per day (for the year in question). The SF-36 patient outcome score was collected annually in the main study. Provided patients did not die or were not lost to follow-up, the change in their QoL between pre-operative assessment and 5-year follow-up was assessed.

Due to a lack of a control group, the quality-adjusted life year (QALY) gain could only be compared hypothetically with the QoL estimates that might have prevailed without surgery. For this, we decided to use a patient's pre-operative QoL (as measured by the SF-36 score) as the counterfactual scenario. We recognised that in reality, without surgery, QoL might have improved or possibly even deteriorated.

The SF-36 is a multipurpose short-form health survey of 36 questions that yields an eight-scale profile of functional health and well-being as well as psychometrically based physical and mental health summary measures.6 However, in the EPOS, completion of this was optional and only 938 patients had sufficiently complete scores for the economic study. We calculated the overall SF-36 score following the user's manual guidance. However, the SF-36 does not directly provide a utility score, so we used the Brazier algorithm to convert SF-36 questions to respective utility scores.7 This utility score captures a patient's value for being in different health states on a scale from 0 (= death) to 1 (= perfect health). Where some patients did not have an SF-36 completed every year over the 5 years, we used their previous year's SF-36 score to derive the utility score for the missing year. This approach seemed reasonable, as after the first post-operative assessment, most utility scores for the group as a whole did not change greatly over remaining years.

We calculated QALY gains made each year compared with the pre-operative baseline using an ‘area under the curve’ approach (see figure 1). In reality, other treatments might have improved this baseline score and reduced the net potential utility gain found in this study. On the other hand, we might have assumed the condition worsened. The direction of change could not be known with any certainty and therefore we assumed that no change occurred in QoL from baseline over the 5-year period. The QALY gain per patient was calculated up to the 5-year follow-up period or until the last annual review before death or revision (whichever occurred first). As the EPOS excluded patients who had revisions, we could only assume a zero QoL gain after revision had taken place (although this would be likely to be higher).

Figure 1

Schematic calculation of quality-adjusted life years (QALYs).

As resource use data were not collected in the main EPOS, we retrospectively constructed a proxy cost per case. This was based on each episode of care and was a composite of the national variable cost per day (for the year in which the surgery was performed) adjusted for the patient's own length of stay plus the estimated total cost of the Exeter implant and other components added as the fixed cost element. An uplifted value of £7500 was used for the cost of a revision.3

Finally, the average cost per QALY was calculated for patients, assuming a zero cost for no surgery. Again this would have been unlikely but we felt it was a conservative assumption (as other treatment costs would be incurred in reality) reducing the cost difference between the two options.

Statistical methods

Analyses were carried out in Stata V.10.8 Bootstrapped, bias-corrected methods were used to calculate 95% CIs for costs per QALY.9 Multiple linear regression was used to model QALY gains, with standard t-test and F-test used to evaluate the significance of β coefficients and model fit.

SF-36 dimension scores were calculated using recommended procedures. Missing values were replaced by scale means if valid responses were available for at least half of the scale items. For the items used in the utility scores, we used chained equations (ICE) to estimate the missing values based on the values of all the other variables in the data set.10 This was carried out when fewer than half of the values for the items used in the calculation of a utility score (from the SF-36) were missing in any individual questionnaire. However, where more than half of the values needed were missing, questionnaires were excluded from the analysis.

Results

The EPOS data set contained complete information on 1589 patients who received the Exeter implant or partial component. However, only five of the seven centres collected QoL data on 1087 patients. Of these, 938 (86%) had sufficient data to be included in the economic study. During the 5-year follow-up, there were 4598 potentially useable questionnaires from the 1087 patients. One thousand and eight of these had completed a baseline score and at least one other follow-up questionnaire completed in sufficient detail for utility estimation. Use of the multiple imputation method (see above) was made to adjust 344 of these surveys. Seventy people with important missing prognostic indicators were excluded, leaving 938 subjects for the main economic analysis. There was good SF-36 completion at the 5-year follow-up date (77%) with 720 patients having this maximum follow-up score. There were 69 deaths and only 17 revisions in the study population within the 5-year follow-up period.

Table 1 shows the characteristics of patients included in the economic study.

Table 1

Characteristics of participants (n=938)

Characteristics		Minimum	Maximum
Age, mean (SD)	62.2 (11.5)	21.0	94.0
Body mass index, mean (SD)	27.2 (4.8)	15.6	53.3
Oxford Hip Score, mean (SD)	44.2 (8.0)	13.0	60.0
Male, n (%)	363 (38.7)

Nearly two-thirds of patients were women. The mean age of all cases was 62 years, but there was a large SD and upper and lower ranges. The average patient was not obese before surgery, although there was considerable variation in body mass index (BMI). The average patient fell into the third quintile of severity on the Oxford Hip Score (OHS).

Changes in utility scores from baseline to Year 5 are shown in table 2 and figure 2. The change in utility scores varied little after the initial large gain in the first post-operative year. The largest component of the increase in overall utility (around 0.18) was seen in the first year after operation. Much smaller changes were found in subsequent years. Both the overall SF-36 score and the individual dimensions that comprise it showed similar changes during this period. The largest changes on the SF-36 occurred in physical functioning (a 47 point increase), physical role functioning (+50 points) and pain (+48 points).

Table 2

Utility score and change from baseline utility by year

	N	Utility score			Difference from baseline utility
		Mean	SD	95% CI	Mean difference	SD difference	95% CI for difference
Baseline	938	0.537	0.113	0.530 to 0.544
Year 1	835	0.720	0.153	0.709 to 0.730	0.181	0.149	0.171 to 0.191
Year 2	728	0.709	0.159	0.698 to 0.721	0.166	0.151	0.155 to 0.177
Year 3	550	0.705	0.160	0.692 to 0.719	0.173	0.154	0.160 to 0.186
Year 4	389	0.712	0.159	0.696 to 0.728	0.170	0.150	0.156 to 0.185
Year 5	720	0.714	0.157	0.703 to 0.726	0.171	0.155	0.160 to 0.182

Figure 2

Utility score by baseline and subsequent year.

The QALYs gained was calculated over this 5-year period shown in figure 3. The majority of patients (90.7%) gained positive QALYs compared with no surgery. These gains were approximately normal distributed around a mean value of 0.8 QALYs (95% CI 0.76 to 0.84). However, a small group of patients (9.3%, n=87) lost QALYs (in a theoretical sense, they would have been better without surgery).

Figure 3

Quality-adjusted life years (QALYs) gained in Exeter Primary Outcomes Study patients up to 5 years.

In terms of estimating cost per episode, average length of stay was 10.8 days (SD 7.3) and the median estimated cost per patient was £5084 (IQR: £4588–£5812). The distribution of costs is shown in figure 4.

Figure 4

Cost of primary hip replacement in Exeter Primary Outcomes Study patients.

Table 3 shows the average QALYs gained and combined with the average cost to derive a cost per QALY. In order to take account of variation and uncertainty in these estimates, we calculated the associated CI using bootstrapping simulation methods. The average cost per QALY for all 938 subjects was £7182 (95% CI £6740 to £7678).

Table 3

Length of stay, estimated cost, QALY gain and cost/QALY for all subjects

	N	Length of stay/day		Estimated cost/£		QALY gain		Cost per QALY
		Mean	SD	Median	IQR	Mean	95% CI	Mean	95% CI*
All	938	10.8	7.3	5084	4588–5812	0.80	0.76 to 0.84	7182	6740 to 7678

Bootstrapped, bias-corrected CI, based on 10 000 bootstrap replications.

QALY, quality-adjusted life year.

We also analysed the QALYs and cost per QALY by age group as shown in table 4. As might be expected, QALY gains were significantly lower in older patients, with the largest gains found in younger patients. Conversely, cost per QALY increased in older age groups because of the increased length of stay combined with a lower QALY gain.

Table 4

Length of stay, estimated cost, QALY gain and cost/QALY by age group

	N	Length of stay/day		Estimated cost/£		QALY gain		Cost/QALY
		Mean	SD	Median	IQR	Mean	95% CI	Mean	95% CI*
23–49 years	82	9.5	3.0	4776	4356–5767	0.95	0.77 to 1.13	5545	4646 to 6861
50–59 years	149	9.4	5.4	4720	4356–5084	0.87	0.76 to 0.98	5902	5140 to 6824
60–64 years	126	10.3	5.1	4981	4391–5767	0.77	0.65 to 0.88	7410	6266 to 8986
65–69 years	190	9.6	2.8	4981	4588–5654	0.88	0.79 to 0.97	5937	5354 to 6631
70–74 years	162	10.5	4.8	5084	4588–5812	0.72	0.63 to 0.81	7944	6940 to 9234
75–90 years	229	13.7	12.1	5654	4812–6904	0.71	0.63 to 0.80	9570	8174 to 11300

Bootstrapped, bias-corrected CI, based on 10 000 bootstrap replications.

QALY, quality-adjusted life year.

With regard to the baseline OHS, we found that pre-operative severity was a good predictor of cost-effectiveness. The poorer the initial score on the OHS, the greater the QALY gain found and similarly the lower the cost per QALY. There were significant differences in the cost per QALY between quintile 1 (least severe) and quintiles 3, 4 and 5 (most severe) and also between quintiles 3 and 5 (see table 5).

Table 5

Length of stay, estimated cost, QALY gain and cost/QALY by Oxford Hip Score (OHS) quintile

OHS (score band)	N	Length stay/day		Estimated cost/£		QALY gain		Cost/QALY
		Mean	SD	Median	IQR	Mean	95% CI	Mean	95% CI*
Quintile 1 (13–37)	191	10.2	4.1	5084	4588–6075	0.61	0.52 to 0.70	9188	7893 to 10915
Quintile 2 (38–43)	206	11.3	10.2	4981	4588–5767	0.83	0.74 to 0.91	7102	6126 to 8186
Quintile 3 (44–47)	174	10.4	5.7	4981	4391–5812	0.70	0.61 to 0.80	7907	6814 to 9253
Quintile 4 (48–51)	191	11.0	7.0	5233	4720–6160	0.89	0.79 to 0.99	6628	5830 to 7577
Quintile 5 (52–60)	176	11.2	7.5	5084	4720–5812	0.98	0.87 to 1.08	5924	5189 to 6826

Bootstrapped, bias-corrected CI, based on 10 000 bootstrap replications.

QALY, quality-adjusted life year.

The QALY gain was approximately normally distributed and therefore linear regression could be carried out to determine patient and treatment characteristics associated with total QALYs. Age, BMI and OHS were significantly associated with QALYs gained (see table 6).

Table 6

Linear regression modelling of quality-adjusted life years (n=938)

	β Coefficient	SE (β)	p Value
Age (years)
Centred age	−0.009	0.002	<0.001
Centred age2/100	−0.027	0.011	0.012
Sex
Female	−0.086	0.045	0.053
Body mass index	−0.015	0.005	0.001
Oxford Hip Score	0.017	0.003	<0.001

Model F-statistic (5 df) = 12.74, p<0.0001, R2=0.064.

Given the limitations of the cost data on which the study was based, a sensitivity analysis was undertaken to determine robustness of the cost and cost per QALY results (see table 7).

Table 7

Sensitivity analysis for main results

	Estimated cost/£		Cost/QALY
	Median	IQR	Mean	95% CI*
Modelling cost entirely due to bed days (no fixed cost element)	4950	4240–5940	7220	6723 to 7748
Fixed cost element reduced by 50%	5012	4410–5870	7197	6725 to 7714
Fixed cost element reduced by 25%	5053	4499–5847	7193	6728 to 7687
Fixed cost element as used	5084	4588–5812	7182	6740 to 7678
Fixed cost element increased by 25%	5115	4669–5777	7167	6728 to 7653
Fixed cost element increased by 50%	5146	4758–5742	7160	6732 to 7641
Totally fixed cost (at 5516)	5516	5516–5516	7058	6679 to 7476

Bootstrapped, bias-corrected 95% CI, based on 10 000 bootstrap replications.

QALY, quality-adjusted life year.

Using various cost assumptions, mean estimated cost per case varied from £4950 to £5516. Given a small variation in cost, the cost per QALY remained relatively stable in the range of £7058–£7220. This confirmed that the cost-effectiveness results were robust and insensitive to some relatively large changes in cost assumptions. This is also reassuring in terms of potential variability in costs between treatment centres and/or surgical practice that occurs in practice.

A cost per QALY threshold analysis is shown in figure 5. Over 85% of cases had a cost per QALY of £20 000 or less with 70% of these having a cost per QALY under £10 000 thus making it very cost-effective when compared hypothetically with no surgery. However, 40 cases had a cost per QALY over £50 000. These patients were largely those where the QoL gain was very small rather than due to their cost being above average.

Figure 5

Probabilistic sensitivity analysis.

Discussion

We have shown that from the perspective of the absence of surgery, the majority of EPOS subjects were treated cost-effectively. The value to patients in terms of their health utility and QALY gains has been demonstrated.

Based on reasonably conservative assumptions, the mean QALYs gained was 0.8 QALYs (95% CI 0.76 to 0.84), while the mean cost per hospital stay was just over £5000 per patient. Although these costs would have been more accurate if the study had been undertaken prospectively, comprehensive data allowed us to build a reasonably accurate cost profile for each patient. Most of this cost could be attributed to length of stay, although the study could not directly account for variability in the price of implant costs. Uncertainty surrounding the fixed cost data was examined using sensitivity analysis and its cost was shown to possibly increase to £5500 per case. Bootstrapping techniques increased the robustness of these findings by reducing bias by multiple replications of the primary study results.

In terms of cost per QALY, we have shown that THR may be more sensitive to optimal treatment and care in the most appropriate patient groups than to local variations in cost. But such results should be treated with caution. An actual alternative comparator implant rather than no surgery would have been more realistic, but no such data existed in this study. However, we deliberately made highly conservative assumptions both about cost and any likely net QALY gain. Furthermore, a probabilistic sensitivity analysis demonstrated that THR is likely to be good value for money even when willingness-to-pay thresholds are set quite low.

This study confirms what is perhaps implicitly assumed in every day orthopaedic practice that hip replacement (using a reliable implant) is worth doing for the majority of patients. The EPOS patients had their pain, function and ultimately their QoL improved by the Exeter hip, even those with above average age, disability and BMI.

Recommendations

Further modelling studies are still needed to establish the longer-term cost-effectiveness of THR.9 The most cost-effective implants will be those with the best survival rates (and hence the fewest revisions), with the best patient outcomes and the least cost. More studies of a comparative nature incorporating economic evaluation would immensely improve the still imperfect knowledge of the cost-effectiveness of different THR implants in today's NHS.