Weighted predictive modelling estimation of patient acceptable symptom state for forgotten joint score, Oxford hip score, and EuroQol health index 3 and 12 months after total hip arthroplasty in a United Kingdom cohort

The study utilised data on unilateral primary total hip arthroplasty (PTHA) performed between April 2021 and March 2023 at the Golden Jubilee National Hospital, a high volume elective Scottish arthroplasty centre. Data on the PTHAs was identified retrospectively within the Clinical Outcome Report Structure (CORS) project database held at centre. There were 2793 unilateral PTHAs performed within the time period, of which 2786 and 2769 respectively were considered eligible for data extraction at 3 and 12 months follow-up. The ineligible cases were for patients who died prior to their 3 and 12 months follow-up (6 and 23 patients respectively) as well as one patient who was under 18 years of age at operation. In addition, 18 complex PTHAs were performed during the same period. Complex PTHAs are defined as procedures in which revision type implants were deemed necessary due to extensive bone loss or gross instability in a primary setting. Given that these procedures represent an extreme and are atypical of routine primary procedure, they were excluded to avoid skewing the data.

The study centre routinely provides a set of PROM questionnaires to patients (via post or telephone) before their procedure and subsequently at 3 months, 1 year, 3 years and 5 years after surgery. The exception to this is the FJS which is only completed postoperatively in keeping with the original rationale for its development as an assessment tool for “awareness” of the ‘artificial joint’. Responses to the questionnaires are recorded within the CORS project database held at the centre. Data on PROMs, patient characteristics and operative factors were extracted from the database. The research team handled all data in accordance with the United Kingdom Caldicott principles while adhering to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline [19].

All procedures at our high-volume teaching hospital were performed by 42 surgeons (17 consultants and 25 trainees). All our consultants are considered high-volume (> 30 cases per year). The majority of cases were performed by the consultant as first operator (69%), with a smaller proportion carried out by senior trainees under direct supervision. A standard posterior approach was used for 91% of the cases, the rest were done using an antero-lateral approach as per the routine practice of the surgeon. Fixation type varied in accordance with surgeon preference; hybrid (82%), cementless (13%), and cemented (5%). Early mobilisation was encouraged in line with an established departmental “Enhanced Recovery” protocol. If appropriate, some patients were placed on our “Day case” hip pathway, with the majority going home on post-operative day 1 or 2. Post-discharge, all patients were instructed on self-directed rehabilitation exercises and received a follow-up call from a trained arthroplasty practitioner at 10–14 days and 3 months to ensure there were no concerns. If necessary, patients were called back for face-to-face review or had the option of an “opt-in” review appointment at any stage if they had concerns. Indications for THA in majority of the cases was due to primary osteoarthritis (95%). Other indications were secondary osteoarthritis (1.1%), other inflammatory arthritis (1.3%), developmental dysplasia of the hip (0.9%) and avascular necrosis (0.8%). The overall recorded “complication” rate was 3% (71 out of 2035 cases). This included readmission (n = 18), dislocation/instability (n = 16), infection (n = 10), thromboembolic disease (n = 8), acute kidney injury (n = 6), revision (n = 5), wound complication (n = 4) periprosthetic fracture (n = 3) and vascular injury (n = 1).

The outcome measures assessed in this study are the FJS, OHS and EQ-5D-5L health index. The FJS comprises 12 questions about artificial joint awareness during daily activities, responses are scored on a 0 to 4 scale, then transformed to a 0 to 100 (worst-best) scale [20]. OHS is composed of 12 questions about pain and function, scored from 0 to 4 and summed to a total of 0 to 48 (worst-best) [21]. EQ-5D-5L measures five health dimensions (mobility, selfcare, usual activities, pain/discomfort, anxiety/depression) with five response levels. The EQ-5D-5L health index was calculated using the UK value set, generating an index from − 0.285 to 1 (death = 0, perfect health = 1, negative values indicate a state worse than death) [22].

An anchor question is an easily interpretable question which establishes a connection between a PROM score and the patients’ health condition [23]. It can be condition -specific or focus on general health but should correlate with the corresponding PROM. We used two anchors in this study, Patient satisfaction and EQ-VAS. The former is a single item questionnaire which measures patient satisfaction with their operated hip (“How satisfied are you with your operated hip?”) with responses recorded on a four-point scale namely, “very satisfied”, “satisfied”, “unsure” and “dissatisfied”. The EQ-VAS, also referred to as health score, is a visual health thermometer that assesses the current health state of patients and is measured on a 0 to 100 scale. A score of 0 represents the worst health the patient can imagine and 100 the best health. Patient satisfaction was chosen as the anchor for the two hip-specific PROMs (FJS and OHS) since it also assesses hip-specific outcome and is commonly used as anchor for estimating PASS for various joint-specific outcomes [23‐26]. Conversely, the EQ-VAS was used as the anchor for the EQ-5D-5L health index to match the domain of general health-related quality of life which the EQ-5D-5L health index also assesses. Previous studies have found that the most suitable anchors focus on the same domain (generic or condition-specific) as the corresponding PROMs noting that hip-specific outcomes cannot be used as anchor to assess general health PROMs [23, 27]. Responses to anchor questions were dichotomised into acceptable symptom state and unacceptable symptom state. Patients responding “very satisfied” or “satisfied” to patient satisfaction question were classified as reporting a hip-specific acceptable symptom state (PASS group) while those answering “unsure” or “dissatisfied” were classified as reporting a hip-specific unacceptable symptom state (Non-PASS group) [24, 25]. If a patient scored 80 or above in the EQ-VAS, they were considered to have an acceptable general health state and classified as PASS group. This cut-off is consistent with the UK population health state classification [28], EQ-VAS PASS validation studies [23, 27], EQ-VAS mean of 79 and median (IQR) of 80 (70–90) observed in the study cohort which also align with those from large national registries such as UK NHS PROMs and the Swedish hip arthroplasty registry [29, 30].

In order to estimate PASS values for the PROMs, the study utilised a new weighted predictive modelling (WPM) method recently developed as a simple and practical alternative to the predictive modelling (PM) method [12].

The WPM method [16] extends the PM method by applying a weighted likelihood function of the logistic regression using weights derived from the rank difference between PROM score and original anchor responses. The weighted approach assigns higher weights to data that exhibit strong positive correlation between the anchor and PROM score and lower weights to those exhibiting weak or negative correlation. The motivation for the weighted approach is to overcome the susceptibility of anchor-based methods to high misclassification rates (false positives and negatives) when correlation between the anchor and PROM score is weak (r < 0.4). Moreover, the WPM demonstrates high accuracy when the proportion of patients reporting satisfactory symptom state deviates from 0.5 under both normal and skewed distribution of PROM scores [16].

The PM and WPM methods were originally developed to assess threshold for meaningful within-individual change which describes the smallest meaningful change between pre- and post-operative outcome scores perceived by a patient after an intervention such as THA. The PM method is based on a logistic regression in which the dependent variable is a binary anchor variable and the independent variable is the difference between pre- and post-operative scores. PASS on the other hand defines a threshold in a post-operative outcome score above which a patient is considered to have experienced a satisfactory outcome following an intervention. The PM and WPM methods are easily adapted for PASS estimation by changing the independent variable in respectively logistic regression models from the change in outcome scores to the post-operative score at the follow-up time being considered. We used the original responses of the EQ-VAS with values ranging from 0 to 100 for specifying the weight in the WPM. However, for patient satisfaction, the dichotomised anchor responses were used to minimise any potential biases related to the lack of symmetry in the response scale.

The validity of the chosen anchors was assessed by calculating Spearman’s correlation between the dichotomised anchors and the PROM scores, and a correlation coefficient within the range of 0.3 to 0.7 was considered appropriate [31]. Non-parametric bootstrapping (n = 1000) was used to calculate 95% confidence intervals for the PASS estimates reported as 0.025 and 0.975 quantiles. Achievement rates of the estimated PASS thresholds calculated as percentage of patients whose scores were greater or equal to the thresholds were compared with the actual PASS percentage (percentage in the PASS group). Sensitivity and specificity were calculated using contingency tables. Empirical cumulative distribution functions (eCDFs) were plotted separately for the PASS and non-PASS groups while marking the PASS thresholds estimated in the study, as recommended by the U.S Food and Drug Administration [32]. To investigate whether PASS thresholds vary by patient characteristics, data was stratified by gender (male and female) and age (> 69 and ≤ 69 years) based on the median age of the study cohort, then PASS thresholds were estimated for each subgroup.

Patient characteristics were presented as median and interquartile range for continuous variables and frequencies and percentages for categorical variables. Normality was assessed using Shapiro-Wilk tests which found the data to show a non-normal distribution. Group differences were assessed using Wilcoxon Signed Rank test for continuous variables and chi-square test for categorical variables. All statistical analyses were performed using R version 4.4.3.

A total of 2793 unilateral PTHAs was performed between April 2021 and March 2023, of which 2786 and 2769 were eligible for extraction for estimating PASS at 3 and 12 months follow-up respectively. PROMS data completion rates ranged from 66% to 73% (Fig. 1). Patient demographics were recorded during preoperative assessment. Patients with complete data had a median age of 69 years, median BMI of 29 kg/m² and 57% were female. These patients were 3–4 years older and had lower BMI (1 point) than those with incomplete data (Table 1). The percentage of patients reporting acceptable symptom state at 3 and 12 months respectively were 94% and 91% for joint-specific PROMs and 64% and 62% for general health PROM (Table 2).

Moderate correlations were found between the PROMs and the dichotomised anchor with lowest correlation observed for FJS and data at 3 months follow-up (Table 3). PASS thresholds along with sensitivity and specificity are reported in Table 3. PASS achievement rates for the joint-specific PROMs were 76% for FJS and 77% and 75% for OHS at 3 months and 1 year respectively. For the EQ-5D-5L, 63% and 60% of patients as reached PASS at 3 and 12 months respectively.

The difference in percentage (percentage points [pp]) to the true PASS percentage was much larger for the joint-specific outcome scores compared to the generic EQ-5D-5L (Fig. 2). Precision as indicated by the 95% CIs tend to be lower in data where correlation between anchor and PROM score is weaker.

Estimated Cumulative distribution function (eCDF) plots stratified by PASS and non-PASS groups are presented in Figs. 3, 4 and 5. The proportion of patients in each group that achieved at least the estimated PASS thresholds can be read visually by tracing horizontally to the y-axis the point at which the eCDF plots intersect the vertical dashed line and subtracting the corresponding value from 1. For example, looking at Fig. 3, the proportions of patients in the PASS and non-PASS groups attaining the FJS PASS threshold are approximately 0.8 (1–0.2) and 0.2 (1–0.8) respectively.

PASS thresholds for age and gender subgroups are presented in Table 4. For gender, males exhibited higher thresholds in all three PROMs compared to females. Patients over 69 years had higher PASS thresholds in the FJS but lower thresholds in the OHS than those 69 years and under.