Aim and perspective

The main objective of the health economic evaluation was to assess the cost-effectiveness of treating displaced extra-articular fractures of the distal tibia using IM nail fixation versus locking plate fixation. The primary analysis was undertaken from the perspective of the NHS and PSS, as recommended by the National Institute for Health and Care Excellence (NICE).³¹ A societal perspective for costs was adopted for the sensitivity analysis and this included private costs incurred by trial participants and their families, as well as productivity losses and loss of earnings as a result of work absences.

The primary health economic outcome was the ICER attributable to IM nail fixation, which was calculated as the incremental cost per QALY gained 12 months after randomisation. No discounting of costs or health consequences was required as the trial-based economic evaluation was limited to a 12-month time horizon.

Measurement of resource use and costs

A comprehensive strategy was adopted to estimate the costs associated with distal tibia fixation using either locking plate or IM nail fixation. This included the (1) estimation of the initial fixation surgery costs and (2) estimation of broader health and personal social service resource inputs and broader societal resource inputs. All costs were expressed in pounds sterling and valued in 2014–15 prices. When appropriate, costs were inflated or deflated to 2014–15 prices using the Hospital and Community Health Services (HCHS) Pay and Price Inflation.³²

Costing of distal tibia fixation

The initial fixation surgery costs (intervention costs) were based on the initial hospital stay and associated operative costs, as reported in Table 22. The cost of distal tibia fracture fixation surgery was estimated using the NHS reference costs, specifically Healthcare Resource Group (HRG) code HT23D (major knee procedures for trauma).³³ According to this HRG code, operative costs for distal tibia fixation were £5315.47 for cases with a mean length of hospital stay of 5 days. Patient-specific costs for the initial operative period were identified using the average length of stay following primary surgery, as reported in the patient records. The mean length of hospital stay was 3.87 days for IM nail fixation versus 3.85 days for locking plate fixation and SEs were 0.34 and 0.33 days, respectively. The surgery cost of patients who stayed longer than 5 days was adjusted using the cost per excess bed-day figure of £327.00 for the same HRG code. For patients who stayed in hospitals < 5 days, we assumed that treatment costs were disproportionately weighted towards the first 3 days of each initial hospital admission. Thus, the cost to the NHS of a patient who stayed in hospital for 3 days was calculated as £5315.47 – (2 × £327), that is, the 5-day tariff minus the bed-day cost of £327 per each day not spent in hospital.

TABLE 22

Unit costs associated with initial operative procedures and initial hospital stay for IM nail and locking plate fixation

In addition, operative costs included the implants used during the surgery, namely nails, plates, locking screws/bolts, blocking screws and non-locking screws. Estimation of the number of each of these implants used involved prospectively recording the number of items used for each patient from the operation notes and radiographs. The total cost of implants for each patient was calculated by combining the resource inputs with their unit cost values; the unit cost values were derived from NHS trust finance departments.

Measuring broader resource use

Individual-level data on all significant health and personal social service and broader societal resource inputs were collected during the trial using follow-up questionnaires completed by trial participants. These data were collected at 3, 6 and 12 months post randomisation. The questionnaires captured the number and duration of admissions to inpatient hospital wards by ward type, number and type of diagnostic tests, use and type of outpatient hospital services, frequency of use and type of community-based health and social care services, medication use and aids and devices provided. In addition, the questionnaires captured the direct non-medical costs (including travel expenses) incurred by patients and their carers, as well as number of days off work and gross loss of earnings, attributable to the trial participant’s health state or contacts with care providers. Copies of the resource use questionnaires administered at each time point are provided in Appendix 3.

Valuation of resource use

The derivation of unit cost values was consistent with NICE’s Guide to the Methods of Technology Appraisal 2013³¹ and included values extracted from the Department of Health and Social Care’s Reference Costs 2014–15,³³ the Personal Social Services Research Unit (PSSRU)’s Unit Costs of Health and Social Care 2015,³⁴ NHS Prescription Cost Analysis and the British National Formulary.³⁵ Table 35, Appendix 4, summarises the unit cost values and data sources for the broader resource use categories.

Further inpatient admissions following the initial operation were costed as minor knee procedures for non-trauma (HRG code HN25A) if the inpatient care involved procedures of the leg. However, if the inpatient admission was related to a surgical complication of the primary surgery, individual HRG codes that related to the subsequent operation procedures undertaken (e.g. debridement, metalwork removal, revision of internal fixation) were derived using the NHS HRG4 2014/15 Reference Cost Grouper software version RC1415 (NHS Digital, Leeds, UK). The Department of Health and Social Care’s Reference Costs 2014–15³³ was used to assign the costs for each of the derived HRG codes. Subsequent inpatient care that was unrelated to procedures of the leg were also costed using Reference Costs 2014–15³³ (Table 35, Appendix 4). The same approach was taken to cost subsequent hospital outpatient care, with resource use data on frequency of outpatient care being combined with the relevant unit costs.

Costs for community-based health services and PSS were calculated by applying unit costs extracted from national tariffs to resource volumes. Costs of medications for individual participants were estimated based on their reported doses and frequencies, when these were available, or based on an assumed daily dose using British National Formulary³⁵ recommendations. When a dose range was reported as ‘as required’ or when the quantities were not recorded, we assumed a mean cost for that medication item based on the prescription cost analysis values (net ingredient cost per item). If the dose of the medication was missing, we assumed the patient received the same dosage as other trial participants who reported taking the same medication.

The costs of equipment that trial participants received to make their daily lives easier and manage their injury (aids and adaptations) were derived by combining the data on number and type of items received with their unit cost values. Unit costs were obtained from the NHS supply chain catalogue (https://my.supplychain.nhs.uk/catalogue; accessed 11 October 2017).

The costs of time taken off work were estimated by applying sex-specific median earnings data from the annual survey of hours and earnings for part-time and full-time work.³⁶ The employment status of trial participants was derived from self-reported work status information. Broader societal costs were calculated by combining the productivity losses and associated loss of earnings as a result of work absences and any privately incurred costs as a attributable to participants’ surgeries or impaired health states.

Calculation of utilities and quality-adjusted life-years

Participants’ health-related quality of life was assessed using the EQ-5D³⁷ obtained at baseline and at 3, 6 and 12 months post randomisation. The EQ-5D defines health-related quality of life in terms of five dimensions: (1) mobility, (2) self-care, (3) usual activities, (4) pain/discomfort and (5) anxiety/depression. Responses in each dimension are divided into three ordinal levels coded: (1) no problems, (2) moderate problems and (3) extreme problems.

The EQ-5D health states are converted into a single summary index by applying a utility algorithm, which attaches values (weights) to each permutation of responses to the EQ-5D descriptive system.³⁷ EQ-5D preference weights have been elicited from general population samples in the UK using the time trade-off method. The resulting utility scores range from –0.594 to 1.0, with 0 representing death and 1.0 representing full health; values below 0 indicate health states worse than death. QALYs were calculated as the area under the baseline-adjusted utility curve and were calculated using linear interpolation between baseline and follow-up utility scores.

Missing data

Incomplete data are a particular issue in within-trial health economic evaluations and can result from item-level missingness; for example, when data for visit 2 are missing but data for visit 1 and all visits after visit 2 are available.³⁸ Consequently, a base-case analysis was constructed when missing data were imputed using fully conditional multiple imputation made chained equations, under the missing at random assumption. Multiple imputation under the missing at random assumption provides unbiased estimates of costs and health consequences. The missing at random assumption was tested through logistic regressions of missingness of costs and QALYs against baseline covariates.

Regression models were used to impute unobserved costs and QALYs at each time point and by treatment allocation using the baseline covariates (age, sex) as predictor variables. Costs and EQ-5D utility scores at each time point contributed as both predictors and imputed variables. The imputation was run 50 times, following the rule of thumb that the number of imputations should be similar to the percentage of incomplete cases.³⁹ The multiple imputation generated 50 data sets using predictive mean matching. Predictive mean matching provides plausible values when costs and QALYs are non-normally distributed.⁴⁰ In line with best practice, the MI model was validated by comparing the distributions of the imputed data with the observed data.⁴⁰

The multiply imputed data sets were analysed independently with bivariate regressions using a seemingly unrelated regression model (Sureg) to estimate the costs and QALYs in each treatment group over the 12-month trial horizon. Non-parametric bootstrapping was used to generate joint distributions of costs and outcomes from the original data set, and changes in costs and QALYs were calculated for each sample. A total of 1000 bootstrap samples were drawn and means for both incremental costs and incremental QALYs (with associated 95% CIs) were calculated.

The final step involved combining estimates from each imputed data set using Rubin’s rule to generate an overall mean estimate of costs and QALYs and the SEs.⁴⁰ The SE calculated through Rubin’s rules reflects the variability within and across imputations.

Analyses of resource use, costs and outcome data

Resource use items were summarised by trial allocation group and follow-up period and differences between groups were analysed using t-tests for continuous variables and Pearson chi-squared (χ²) test for categorical variables. Means and SEs for values of each cost category were estimated by treatment allocation and follow-up period. Statistical differences in mean costs by treatment allocation were assessed using Student t-tests. Mean total costs by treatment allocation and follow-up period were also estimated. Statistically significant differences in the mean total costs were assessed using non-parametric bootstrapping, based on 10,000 replications.

We calculated the proportion of patients reporting suboptimal health for each of the five dimensions of the EQ-5D. Patients were considered to be in suboptimal health if they reported moderate or extreme problems. We explored whether or not any statistical differences in suboptimal health-related quality of life existed between the two treatment arms at the different time points, using a Pearson chi-squared (χ²) test.

Cost-effectiveness analyses

Cost-effectiveness results are expressed in terms of the ICER and calculated as the difference between treatments in mean total costs divided by mean total QALYs. The bootstrap replicates from the non-parametric bootstrapping, described in Analyses of resource use, costs and outcome data, were used to populate cost-effectiveness scatterplots. Cost-effectiveness acceptability curves, which showed that the probability that IM nail fixation is cost-effective relative to locking plate fixation across a range of cost-effectiveness thresholds, were also generated based on the proportion of bootstrap replicates with positive incremental net benefits. The net monetary benefit (NMB) of using IM nail fixation versus locking plate fixation was also calculated across three cost-effectiveness thresholds, namely £15,000 per QALY, £20,000 per QALY and £30,000 per QALY. A positive incremental NMB indicates that the intervention is cost-effective compared with the alternative at the given cost-effectiveness threshold.

Sensitivity and subgroup analyses

Several sensitivity analyses were undertaken to assess the impact that uncertain parameters had on components of the economic evaluation. These involved re-estimating the main cost-effectiveness outcomes under the following scenarios: (1) restricting the analyses to complete cases (i.e. those with complete cost and outcome data over the 12-month follow-up period); (2) adopting a wider societal perspective that included private costs incurred by trial participants and their families, as well as productivity losses and loss of earnings owing to work absences; (3) estimating the cost-effectiveness under a per-treatment analysis; and (4) additionally adjusting the baseline analysis for pre-injury health-related quality of life, which was assessed using the EQ-5D at baseline.

Subgroup analyses were also conducted for the main cost-effectiveness results to explore heterogeneity in the trial population. These were conducted by (1) age group (< 50 and ≥ 50 years) and (2) sex (male and female).

Longer-term economic modelling

The study protocol allowed for decision-analytic modelling to extend the cost-effectiveness of IM nail fixation, drawing on best-available secondary data sources, supplemented when necessary by expert opinion.

Health-care resource use

Table 8 summarises the key resource inputs associated with the initial treatment of displaced extra-articular fractures of the distal tibia. Table 37, Appendix 4, presents details of broader health and social care resource use over the 12-month follow-up period for complete cases, disaggregated by resource category and period of follow-up. Generally, resource use at the aggregate level was higher for participants allocated to the locking plate than those allocated to IM nail fixation, but this was not always statistically significant. The exceptions were differences in mean total inpatient stay at 6 months (0 vs. 0.11 months), which was statistically significant at the 5% level (p-value of 0.03), and mean total outpatient care contacts at 6 months (3.64 vs. 4.78 mean outpatient contacts), which was also significant at the 5% level (p-value of 0.04). The reported number of days taken off work was generally higher for the locking plate arm. However, the mean difference (46.12 vs. 54.46) was significant at a 10% significance level at 3 months (p-value of 0.07). Regarding components of the resource categories, patients in the locking plate arm were more likely to use walking frames at 3 months (0.20 vs. 0.34), utilise more NHS physiotherapy (1.84 vs. 2.53) at 6 months and report higher use of ‘other’ medicines at 12 months (0.08 vs. 0.29).

Costs

Table 23 summarises the total costs associated with resource use during the trial period among complete cases by cost category and follow-up period. The mean intervention costs from admission until discharge were £5585 for IM nail fixation compared with £5615 for locking plate fixation; the mean difference of £30 was not statistically significant. The mean total NHS and PSS cost throughout the first 6 months post randomisation was £5876 for IM nail fixation and £6814 for locking plate fixation; the mean cost difference of £939 was statistically significant at the 5% level. The mean total NHS and PSS cost for the entire 12-month follow-up period was £6107 for IM nail fixation and £7102 for locking plate fixation; the mean cost difference of £995 was statistically significant at the 10% significance level. Total societal costs among cases with complete data are summarised in Table 24. The mean total societal costs throughout the first 6 months was £9793 for IM nail fixation compared with £12,178 for locking plate fixation; the mean cost difference of £2385 was statistically significant at the 10% significance level. The mean total societal costs are £9490 for IM nail fixation and £12,886 for locking plate fixation; the mean cost difference of £3396 was statistically significant at the 5% significance level.

TABLE 23

NHS and personal social service costs for cases with complete data by trial allocation, study period and cost category (£, 2014–15 prices)

TABLE 24

Societal costs related to distal fracture fixation for cases with complete data by treatment arm (£, 2014–15 prices)

Health outcomes

Table 25 details the number and proportion of individuals reporting each of the EQ-5D dimension levels at each time point. The proportion of trial participants reporting suboptimal health (moderate to extreme health outcomes) is also indicated for each dimension and the difference between the two treatment arms shown by p-values. With the exception of mobility at 3 months [IM nail fixation (81%) vs. locking plate fixation (89%)], which was statistically significant at the 10% significance level, there were no significant differences in the proportions of individuals reporting suboptimal health within dimensions between the two arms at each time point.

TABLE 25

The EQ-5D descriptive measurements by trial allocation, study period and EQ-5D dimension

Cost-effectiveness results

The cost-effectiveness results are presented in Table 26 with locking plate fixation selected as the comparator. The analytic time horizon covers the entire 12-month follow-up period of the trial. Table 26 shows the joint distribution of incremental costs and outcomes for the base-case analysis, sensitivity analyses and subgroup analyses. The joint distribution of costs and outcomes for the base-case analysis, sensitivity analyses and subgroup analyses are graphically represented in Figure 9 using the NMB metric and in Figure 10 using scatterplots and cost-effectiveness acceptability curves. Plots for the remaining of the sensitivity analyses that are not shown in Figure 10 mirror the base-case analysis plot. As such, we elected not to present them.

TABLE 26

Cost-effectiveness, cost/QALY (£, 2015): IM nail fixation compared with locking plate fixation

FIGURE 9

Forest plots of base-case, sensitivity and subgroup analyses showing the impact on incremental NMB at cost-effectiveness thresholds of (a) £15,000 per QALY, (b) £20,000 per QALY and (c) £30,000 per QALY.

FIGURE 10

Cost-effectiveness scatterplots and acceptability curves at 12 months. Base-case analysis (NHS and personal social service perspective, imputed, intention-to-treat analysis): (a) scatterplot and (b) cost-effectiveness acceptability curve; complete-case (more...)

Subgroup analyses

The subgroup analyses for age and sex revealed that there was significant interaction between treatment effect and age group, but this was significant at the 10% significance level (p-value of 0.09); the sex subgroup analysis did not reveal any statistically significant interactions. IM nail fixation lowered costs for patients aged < 50 years (–£1468, 95% CI –£3547 to –£291) and moderately increased QALY benefits (see Table 26). For patients aged ≥ 50 years, IM fixation reduced costs and moderately reduced QALY outcomes, with an ICER of £60,000 per QALY that fell within the south-west quadrant of the cost-effectiveness plane, that is, the average costs were less and average benefits were also lower for IM nail fixation than for locking plate fixation. The 95% CIs for both the costs and QALY estimates suggest considerable uncertainty surrounding the effects of IM nail fixation for this category of patients; this is graphically depicted in the NMB forest plot (see Figure 9). Cost-effectiveness acceptability curves for the subgroup analyses show the differences in probability of cost-effectiveness of IM nail fixation versus locking plate fixation (Figure 11). According to Figure 11, the probability that IM nail fixation is more cost-effective is higher for individuals aged < 50 years. There is no observable difference in probability of cost-effectiveness of IM nail fixation between male and female participants.

FIGURE 11

Cost-effectiveness acceptability curves at 12 months (NHS and personal social service perspective, imputed). (a) Age; and (b) sex.