Study design and data collection
The PaRIS-IOR is a prospective, single-site, cohort study that started on January 1, 2019, and consists of the administration of PROMs questionnaires investigating the quality of life (EQ-5D-3L, and EQ-VAS [18, 19]) and joint-specific functional outcomes (HOOS-PS [24,25,26]), to patients on the list for elective HA.
PROMs baseline questionnaires were administered to patients awaiting surgery by specifically trained researchers within 30 days before surgery. The follow-up questionnaires were mailed 6 and 12 months after surgery.
IOR hosts the Registry of Orthopedic Prosthetic Implants (RIPO). PROMs data were linked with those routinely collected by the RIPO [27] and other regional administrative data (i.e., hospital discharge records), in order to track patients’ medical history and to define patients’ health profiles.
Patients undergoing elective HA between January 1st and December 31st, 2019, constituted the baseline population. Data included patients’ demographics, pathology leading to joint replacement, type of surgical procedures, in-hospital complications, and the characteristics of the implant. Specifically, we collected and analyzed: (i) the patients’ characteristics and profile, including age and sex distribution, Body Mass Index (BMI), Elixhauser Comorbidity Index (ECI) [28], American Society of Anesthesiologists (ASA) score, Modified-Chronic Disease Score for clinical severity (M-CDS) [29]; (ii) the PROMs questionnaire total scores: EQ-5D-3L score (general range from less than 0, where 0 is a health state equivalent to death and negative values are valued as worse than death to 1, perfect health), EQ-VAS (range 0–100, where 0 is worst and 100 is best), HOOS-PS score (range 0–100, where 0 is worst and 100 is best) for HA patients. The ECI is a comorbidity index based on the International Classification of Diseases (ICD) diagnosis codes. It is obtained as an unweighted count of comorbid conditions [28]. The ASA score is a system for assessing the fitness of patients before surgery. In 1963 the American Society of Anesthesiologists adopted this five-category physical status classification system; a sixth category was later added. The ASA categories (1 to 6) are: Healthy person; Mild systemic disease; Severe systemic disease; Severe systemic disease that is a constant threat to life; A moribund person who is not expected to survive without the operation; A declared brain-dead person whose organs are being removed for donor purposes. The M-CDS [29] is a weighted chronic disease score based on 18 comorbid conditions derived from drug prescriptions that was developed as a prognostic score of 1-year mortality. This score is categorized into 6 classes (0–1, 2, 3–4, 5–6, 7–9, ≥ 10).
Inclusion criteria were age 18–95 years and elective primary hip arthroplasty; exclusion criteria were: severe cognitive impairment; arthroplasty for musculoskeletal cancer; patient not eligible for surgical procedures; hip arthroplasty in the 12 months prior to enrollment. The detailed study protocol, inclusion and exclusion criteria and other information are described in a previous publication [30]. This study follows the STROBE reporting guidelines for observational studies [31, 32].
Instruments
The choice of using the selected PROMs measures was based on consensus from the OECD’s PaRIS Initiative group, as reported in a previous publication [25, 30]. The validated EQ-5D-3L Italian version was used in this study [33, 34]. The EQ-5D-3L health status and quality-of-life measure is composed of five items (mobility, self-care, usual activity, pain/discomfort, and anxiety/depression) [34]. The 3L version describes health on three levels (no problems, some problems, and a lot of problems) resulting in 243(35) health states [33, 34]. The EQ-5D-3L index is calculated from the scores of the five dimensions, ranging from − 0.594 (worst) to 1.0 (best). Moreover, the EQ-5D-3L includes a VAS for rating of overall health status from 0 (worst imaginable health) to 100 (best imaginable health) [35].
The Italian validated version of the Physical function Short form of the Hip Disability and Osteoarthritis Outcome Survey (HOOS-PS) was used in this study [26]. It consists of 5 items (sitting, descending stairs, getting in/out of the bath/shower, running, twisting, or pivoting on the loaded leg) rated on a five-point Likert scale (none, mild, moderate, severe, extreme), where 0 indicates no problems and 4 extreme problems. The reliability of the 5 items was 0.80 (Cronbach’s alpha [36]. The total score can be transformed to a 0–100 scale, with 0 indicating the worst problems and 100 indicating no problems. The HOOS-PS has been used in subjects with hip disability with or without hip osteoarthritis and in patients with hip osteoarthritis pre- and postoperative total hip replacement. It is suitable for use in research as a disease-specific questionnaire [26, 37].
Statistical analysis
Baseline demographic and clinical characteristics were summarized using mean and standard deviation, median and interquartile range, or absolute and percentage frequencies, as appropriate. To determine whether patients completing the study questionnaires at 6 and 12 months were representative of the baseline sample, patients lost to follow-up and completers were compared at 12 months on age, gender, BMI, ASA score, region of residence, and primary diagnosis. Information about variable distributions and missing data can be found in the Additional file 1. Continuous variables were compared between groups using t-test and categorical variables using chi-square test or Fisher’s exact test, as appropriate. The significance level was set to 0.05.
Latent class growth analysis and growth mixture model
Latent class growth analysis (LCGA) was carried out as an initial modelling step to identify subgroups of patients with different trajectories of functioning and quality of life from pre-surgery and to 12 months following total hip replacement.
LCGA is a special type of Growth Mixture Modeling, whereby the variance and covariance estimates for the growth factors within each class are assumed to be fixed to zero [38]. By this assumption, all individual growth trajectories within a class are homogeneous. This technique allows the user to classify distinct subgroups that follow a similar pattern of change over time, hence it is appropriate for analyzing longitudinal data [39, 40]. Other longitudinal methodologies, such as conventional growth models, assume that individuals come from a single population and that a single trajectory can adequately summarize the entire population. Moreover, they assume that covariates that affect the growth factors influence individuals in the same way. However, we have theoretical reasons to assume that in a clinical population (and specifically among elderly patients) a single growth trajectory would be an oversimplification of the complex growth pattern that characterized changes among members of different groups.
LCGA can accommodate missing data at 6 and/or 12 months using the full information maximum likelihood algorithm (Additional file 1: Figure S1), thus allowing the user to define trajectories for the full set of patients [40, 41].
LCGA requires that assumptions are met concerning within-class conditional normality, a properly specified mean and covariance structure, the linearity of effects of exogenous predictors, a missing at random (MAR) mechanism underlying missing data, and that sample individuals are independent and self-weighted [38,39,40,41].
Standard inferential model fit indices were used to identify the best fitting models. Model fit indices included the Akaike Information Criteria (AIC) and Bayesian Information Criterion (BIC). These indices have no predefined cut-offs and can only be interpreted when comparing different models. Lower AIC and BIC indicate a better model fit. Other indices included entropy (values close to 1.0 denote excellent fit), no less than 1% total count in a class, and high posterior probabilities. In addition, Vuong-Lo-Mendell-Rubin likelihood ratio test was used to determine the number of classes. The final model was based on statistical considerations and clinical meaningfulness.
Growth mixture models were then used to estimate separate growth models for each latent class identified by the LCGA. Individuals were assigned to the most likely latent class based on posterior probabilities [32]. Lastly, we analysed the demographic and clinical predictors of the latent classes in GMM using a 3-step approach [42]. This approach takes into account the measurement error in the class assignment process and prevents defining class membership from being influenced by covariates. Specifically, we included the following demographic and clinical variables that are routinely recorded in the administrative databases or in the registry: age, sex (male as reference category), BMI (normal weight/underweight as the reference category), diagnosis (primary coxarthrosis vs. other diagnoses as reference categories), ASA score (1 as the reference category, 2, ≥ 3). Multicollinearity of variables was assessed using the variance inflation factor (VIF). No adjustment was made in the analyses for multiple outcomes.
Patients were cross-classified according to the trajectory group for two PROMs indicators: HOOS-PS and EQ-5D-3L. The former was chosen to account for patients’ reported functionality and mobility, while the latter for patients’ reported quality of life. This choice was made to investigate the distribution of patient reported QoL and functionality scores within the study population. We then analyzed differences in demographic and clinical characteristics between the sub-group of patients showing worse reported outcomes in both PROMs indicators.
All statistical analyses were performed using SPSS, version 25.0, R, version 4.1.0 and the package lcmm [41], and MPlus version 8.7.
Protocol registration
Protocol version (1.0) and trial registration data are available on the platform www.clinicaltrial.gov with the identifier NCT03790267, posted on December 31, 2018.