WellNet program - Overview, intervention, and study design
The ‘WellNet’ program developed by Sonic Clinical Services (SCS) is a general practitioner (GP) led, multidisciplinary team-based (MDT) care delivery model within primary care settings. The 12-month program is built upon best practice clinical care models, including the Patient-Centred Medical Home (PCMH), which aims to deliver care that is tailored to individual risk and comorbidity burden [27].
The enhanced primary care program is designed to provide individualised ‘whole-person’ care with focus on self-management support, health coaching and education, care coordination, shared decision making, and long-term continuity of care. Ongoing support and monitoring were provided through a total of 14 possible consultations with the care team in the form of in-practice visits and telephone contacts throughout the 12-month period. In addition, patients were also supported with a user-friendly online platform called ‘GoShare’ providing patient-tailored educational materials and a mobile application ‘MediTracker’ enabling access and reminders to the next scheduled GP appointments and prescriptions. Further details on how CCs monitored usage of GoShare and MediTracker are reported elsewhere [27].
Patients were recruited between December 2016 and October 2017 using a targeted convenience sampling technique if they met the eligibility criteria. Targeted convenience sampling is a commonly used non-probability sampling in clinical research where members of the target population that meet certain practical eligibility criteria are included for the purpose of the study [28]. A case-series study design was used to determine changes in the HRQoL before and after WellNet care among patients enrolled in six primary care practices across Northern Sydney, Australia. Informed consent was obtained from all participants upon enrolment into the WellNet program.
Participants
A computerised algorithm was executed to identify potentially eligible patients from the electronic medical records of SCS GP practices. The overarching criteria for eligibility include patients aged 40 years and over; having one or more chronic condition/s with or without one or more elevated clinical risk factors; and had visited a general practitioner (GP) at least thrice in the previous two years. Patients living in nursing homes and those with severe cognitive impairment or terminal illness (n=10) were excluded. More details on the risk algorithm, enrolment, and data collection are reported elsewhere [27]. Of the 636 consenting participants, 616 who completed the baseline EQ-5D questionnaire were analysed in this study. Flowchart of the enrolment outcomes is shown in Figure 1.
EQ-5D-5L instrument
The HRQoL was measured using the standardised UK version of the five dimensions and five levels of the EuroQol (EQ-5D-5L) instrument [29]. The questionnaire covers five dimensions of health: mobility, self-care, usual activities, pain or discomfort, and anxiety or depression. The levels of severity range from no problems to extreme problems for each of the five dimensions are recorded. The raw scores are then converted to a single EQ-5D index value (Time-trade off value) using a scoring algorithm ranging from 0 (worst perceived health state) to 1 (best perceived health state) [30]. In the WellNet program, EQ-5D-5L questionnaire was recorded at baseline and 12 months (at program completion).
KOOS and HOOS assessments (short versions)
The Hip disability and Osteoarthritis Outcome Scores (HOOS) and Knee injury and Osteoarthritis Outcome Scores (KOOS) are shortened but validated versions of the HOOS and KOOS surveys indicated for patients with a positive diagnosis of osteoarthritis and reporting different forms of hip and knee disability [31, 32]. These surveys are intended for use over short and long-term intervals to assess patient-reported changes in the quality of life in terms of changes in the levels of function, symptoms, and pain induced by a particular treatment [31, 32]. Both the questionnaires contain subscales of items recording patient’s QoL in terms of pain, function, daily living, and stiffness. The scores range from 0 to 100: 0 indicates total disability, whilst 100 indicates perfect functionality. As MCIDs were not established at the start of this study, previous body of literature estimations were adopted where minimal clinically important difference (MCID) values ranging between 9.6 and 16.2 for HOOS; and between 8 and 10 for KOOS [33, 34]. In this study, the HOOS and KOOS scales were only used as a supplement to EQ-5D instrument.
Study outcomes and exploratory variables
The primary outcome of interest was changes in the EQ-5D index value and increase in the proportion of patients at a lower severity level in all five dimensions at follow-up. The secondary outcomes included: 1) predictors of change in EQ-5D index over time; 2) adjusted mean difference in KOOS and HOOS scales recorded among subsample of patients diagnosed with osteoarthritis.
The explanatory or predictor variables analysed and adjusted for in this study as follows: age, gender, diagnosis of chronic conditions, number of co-existing conditions, private insurance status (PHI), and number of scheduled consultations.
Data analysis
Descriptive statistics for continuous variables are expressed as mean and standard deviation (SD) whereas frequency counts of categorical variables are shown in percentages. Normality of distribution was assessed using the Shapiro-Wilk test for normality and by analysis of normal quantile-quantile plots. Independent samples t-tests and Pearson’s chi-square tests were conducted to determine significant differences between completers and those who withdrew (non-completers) before program completion. Additionally, Pearson’s product-moment correlation coefficient was conducted to determine the level of association between EQ5D scores and different chronic condition groups at baseline.
Two models were employed for primary analyses: 1) per-protocol model (Model 1) – those who reported both baseline and follow-up EQ5D scores and 2) imputed model (Model 2) where missing follow-up scores are replaced by valid values using other available information from the dataset as a result of repeated draws from the fully conditional specification (FCS) of Markov Chain Monte Carlo (MCMC) algorithm [35]. A total of 25 iterations were computed and pooled estimates of the imputed models were reported. The multiple imputation model allowed for inclusion of dependent variable of follow-up EQ5D scores as a covariate to enable unbiased estimates of model coefficients [36]. The rationale for imputing missing data was supplemented with primary analysis to test the hypothesis whether the overall WellNet sample (N=616), including patients who withdrew prematurely before program completion without reporting their follow-up EQ5D scores would have reported similar trends in EQ5D scores, if they had completed the program as the cohort of patients who did complete the program.
Unadjusted mean differences between baseline and follow-up were computed using paired samples t-test and adjusted mean differences for both models and instruments were determined by using the repeated measures ANCOVA whilst adjusting for potential baseline covariates. Additionally, a subgroup analysis was also conducted to evaluate adjusted differences in EQ5D scores up among proportion of patients with two or more conditions and patients who had more than median contacts (≥12 contacts) with WellNet care team.
To determine predictors of change in EQ-5D over time, multivariable linear regression models were employed using post-EQ5D index scores as outcome variable. Post-EQ5D index scores were preferred over change scores (follow-up minus baseline) as outcome variable because change scores fail to allow for optimal control of the baseline imbalance owing to potential regression to the mean [37, 38]. Univariate linear regression was conducted for each variable separately and variables with p-value<0.20 were included in the multivariable model. The backward stepwise regression approach was used to reduce and create the final model while simultaneously assessing the fitness of model in order to avoid dropping of non-significant variables that may affect the model fitness. The final model constitutes variables, which when excluded, cause a prominent deviance change (p < 0.05) as compared to the corresponding X2 test statistic on the relevant degrees of freedom.
Finally, the internal consistency of EQ-5D, KOOS, and HOOS scales in this study were evaluated using Cronbach's alpha. All analyses were conducted using SPSS (version 25) and R statistical software.