Outcome assessment after hip fracture

Data

Data were available from two distinct, but related, studies.

The Warwick Hip Trauma (WHiT) study¹¹ was a single centre, parallel group randomised controlled trial. Between September 2009 and April 2011, 225 participants aged 65 years and over with an intracapsular fracture of the hip were recruited. All participants were followed-up at six weeks, 12 weeks and one year post-operatively and completed the EQ-5D at each time-point. In addition, basic demographic data were recorded on admission (baseline), and trial participants were asked retrospectively to report score data to assess function and quality of life before the fracture. Full details of the study and treatment effects are described in detail elsewhere.¹²

The Warwick Hip Trauma Evaluation (WHiTE) study¹³ is a prospective cohort study performed at the same trauma centre, in an analogous manner and as an immediate extension of the WHiT study. Recruitment began in January 2012, and is on-going, with data currently available from 249 participants, all aged 60 years and above with all types of hip fractures. Following consent, all participants report their pre-operative hip function (OHS), quality of life (EQ-5D) and capability (ICECAP-O) retrospectively assessed through questionnaires. This is followed by telephone questionnaires at four weeks, four months and 12 months post-operatively. For those participants with cognitive impairment (defined as an abbreviated mental test (AMT) score less than 8),¹⁴ this assessment is limited to the EQ-5D, which is reported by an appropriate proxy.

Outcomes

EQ-5D is a validated, generalised and standardised instrument comprising a visual analogue scale (VAS) measuring self-rated health and a health status instrument, consisting of a three-level response (no problems, some problems and extreme problems) for five domains related to daily activities¹⁰; (i) mobility, (ii) self-care, (iii) usual activities, (iv) pain and discomfort and (v) anxiety and depression. Responses to the health status classification system are converted into an overall score using a published utility algorithm for the UK population.¹⁵ A respondent’s EQ-VAS gives self-rated health on a scale where the endpoints are labelled ‘best imaginable health state’ (100) and ‘worst imaginable health state’ (0).

OHS was introduced in 1996 and uses 12 multiple choice questions to quantify disability secondary to hip osteoarthrosis.⁸ It has good evidence supporting its validity^8,16,17 and is entirely patient-reported, thus can be administered by post or telephone as well as in the clinical setting, without the need for an assessment by a clinician. Each of the 12 items comprising the OHS is assessed on a scale from zero to four, where zero represents, for instance, no pain or difficulty in undertaking the task in question. The scores for the 12 items are summed for each patient to give an overall score that quantifies hip function on a scale from 0 to 48, where 0 indicates excellent hip function and 48 indicates very poor hip function.

ICECAP-O is a measure of capability in older people.⁹ It comprises five attributes that are important to older people in the UK; (i) attachment (love and friendship), (ii) security (thinking about the future without concern), (iii) role (doing things that make you feel valued), (iv) enjoyment (enjoyment and pleasure) and (v) control (independence). Each attribute is assessed using four ordered levels from the lowest capability (level 1) to the highest capability (level 4), and scaled using UK index values¹⁸ to ensure overall ICECAP-O values were constrained between zero (no capacity) and one (full capacity).

Statistical analysis

Descriptive data summaries (mean, standard deviation [sd], median and range) were calculated at each occasion to assess overall population changes and variability. Standardised effect sizes, calculated as the difference in means between the index occasion and the baseline assessment divided by the baseline standard deviation, were estimated in order to assess responsiveness for all of the available outcome measures. Paired t-tests were also calculated to assess the significance of changes in mean outcome measures from baseline. For WHiT, receiver operator characteristic (ROC) curves were constructed for six weeks of EQ-5D using the defined gold-standard outcomes of subsequent revision and death within 12 months. Areas under the curves, and 95% bootstrapped confidence intervals, were calculated to assess the predictive power of post-operative EQ-5D. Scatterplots were used to assess ceiling and floor effects visually. Construct validity was also assessed by scatterplot and by pairwise correlations between outcome measures, which were quantified using Pearson’s correlation coefficient. This was further explored for OHS and EQ-5D data, available from WHiTE, by calculation of item-to-item Spearman correlation coefficients; that is correlations between all pairs of individual questionnaire items. Responsiveness was quantified for proxy and patient scored EQ-5D for WHiTE by calculating appropriate statistics for demented and non-demented groups. EQ-5D and ICECAP-O data for this population were compared with previously reported normative values for the UK population. Statistical significance was set at the 5% level and all analyses were undertaken using the statistical software package R.¹⁹

Characteristics of the sample

Baseline characteristics for WHiT participants are shown in Table I and summaries of outcome measures, EQ-5D and EQ VAS, in Table II. Table II shows estimates of the means, medians and standard deviations at baseline and each of the follow-up time points at six weeks, 12 weeks and 52 weeks. Baseline characteristics for WHiTE participants are shown in Table III and summaries of outcome measures, OHS, EQ-5D, EQ VAS and ICECAP-O, in Table IV. Table IV shows estimates of the means, medians and standard deviations at baseline and each of the follow-up time points at four weeks and four months. Also shown are effect sizes, which are used to assess responsiveness to change from baseline, and paired t-tests at four weeks and four months.

Longitudinal change in outcome scores

There was clear evidence from both WHiT and WHiTE of a smooth increase in EQ-5D scores in the post-operative period, as patients returned to comparatively normal function (Table II and Table IV); that is scores gradually returned to their pre-operative (baseline) level.

The estimated (standardised) effect size in the WHiT Study decreased from a medium to a small level during follow-up for EQ-5D (0.68 at six weeks, 0.32 at 12 weeks and 0.27 at 52 weeks), consistent with the change in mean scores. EQ-VAS proved to be unresponsive to change from baseline. ROC curves for six week EQ-5D based on estimated sensitivity and specificity for subsequent revision and death within 12 months are shown in Figure 1. The estimated area under the ROC curve (AUC) for death was 0.72 (95% CI; 0.57, 0.85 for 2000 bootstrap replicates) and the area under the ROC curve for revision was 0.49 (95% CI; 0.38, 0.62). The AUC measures discrimination, that is, the ability of EQ-5D to classify correctly those who will die or be revised within 12 months; a value of 0.5 indicates no predictive value. Therefore, EQ-5D is not predictive of revision, but is moderately good predictor of death within 12 months.

Fig. 1

Receiver operating characteristic (ROC) curves for WHiT EQ-5D at six weeks and subsequent death and revision within 12 months; data (solid) and smoothed curves (dashed)

The estimated effect size in the WHiTE Study decreased from a medium to a small level during follow-up for EQ-5D, consistent with the change in mean scores. There was an analogous decrease in effect sizes, from large to small, from four weeks to four months for OHS. EQ-VAS proved to be unresponsive to change from baseline and ICECAP-O was moderately responsive at four weeks only.

Proxy reported EQ-5D

For those participants with cognitive impairment (nominally referred to here as ‘demented’) EQ-5D was reported by an appropriate proxy. Temporal trends in scores for demented and non-demented groups, shown in Figure 2, were similar with scores being lower overall for the latter than the former group. Estimated effect sizes at the four week assessment for the two groups were 0.85 and 0.58 and of approximately the same magnitude as those observed for the full population (Table IV).

Fig. 2

Boxplots for WHiTE for EQ-5D at baseline, four weeks and four months for demented and non-demented patients; symbols (●) and bars show means and 95% confidence intervals

Comparison between EQ-5D and other instruments

Pair-wise correlations for EQ-5D, OHS and ICECAP-O at baseline, four weeks and four months (Table V) showed that there were significant strong positive correlations, > 0.7, between EQ-5D and OHS at all occasions. Correlations between EQ-5D and ICECAP-O, and between ICECAP-O and OHS were also all significant, although smaller than between EQ-5D and OHS, at baseline and four weeks. The correlations between EQ-5D and ICECAP-O, and between ICECAP-O and OHS failed to reach significance due to the paucity of available data at this occasion. This is apparent in the scatter-plots between pairs of outcome measures at each occasion (Figure 3).

Fig. 3

Scatter-plots between pair-wise combinations of EQ-5D, Oxford Hip Score (OHS) and ICECAP-O from WHiTE at baseline (top row), four weeks (middle row) and four months (bottom row); solid lines show linear regression fit, with 95% prediction intervals shown as dashed lines

Item-to-item correlations between EQ-5D and OHS at four weeks showed a strong correlation (0.72; 95% bootstrapped CI 0.51 to 0.85) between the EQ-5D Self Care domain and OHS item 2 (Have you had any trouble washing and drying yourself because of your hip?). There were similar strong associations between the EQ-5D ‘Mobility’, ‘Usual Activities’ and ‘Pain’domains and those aspects of the OHS that were relevant to these construct (e.g. walking or climbing stairs for ‘Mobility’). There were generally many fewer significant associations between the EQ-5D ‘Anxiety’ domain and OHS questionnaire items, consistent with what one might expect for this construct that does not feature explicitly in the OHS.

Recall of pre-fracture status

The mean baseline EQ-5D scores were 0.66 (95% CI; 0.61 to 0.71) and 0.55 (95% CI; 0.50 to 0.59) for the WHiT and WHiTE studies respectively. A more informative categorisation for the WHiTE study gave means of 0.39 (95% CI; 0.33 to 0.45) and 0.66 (95% CI; 0.60 to 0.71) for the demented and non-demented groups. The UK normative data^20,21 derived from face-to-face interviews with randomly selected individuals from England, Scotland and Wales, suggest a mean EQ-5D score of approximately 0.680 for the 80+ age group, based on tabulated data that consists of marginal proportional splits by level and EQ-5D category. This is close to our reported values for both WHiT and WHiTE, taking the non-demented group as a more realistic point of comparison for the latter group, and provides strong evidence for the veracity (reliability) of retrospective assessments of EQ-5D in this population.