Dr Feng Pan

Osteoarthritis (OA) is the most prevalent form of arthritis in Australia and worldwide. OA is a chronic and painful disease of the synovial joint, and a leading cause of disability(1). Knees, hips and hands are the most commonly affected joints. Pain is the most prominent symptom which drives patients to seek healthcare. Thus, OA represents an enormous health and economic burden on patients and societies. There were 100,000 knee and hip replacement procedures (mainly due to OA) performed in Australian hospitals during 2015, with a cost of $2 billion a year(2). Current pain control is unsatisfactory. There are no proven strategies to prevent, slow, halt or reverse the OA progression. Current OA management is mostly palliative and focuses on pain relief. ‘First-line’ agents, such as paracetamol and non-steroid anti-inflammatory drugs (NSAIDs), have only small to moderate efficacy, with >75% of patients reporting the need for additional symptomatic treatment(3). Furthermore, there are increasing safety concerns about their use. For example, NSAIDs have long been associated with gastrointestinal adverse events, both minor and life-threatening. A more in-depth understanding of the contribution of the central nervous system to chronic musculoskeletal pain offers additional appreciation for the exploration of centrally-acting medications in OA. Nonetheless, pain control remains a substantial unmet need in OA treatment. In addition to the need to develop drugs that can slow disease progression, there is a critical need to develop more effective pain relief agents for OA patients.

What is needed to develop effective treatments for OA?-Current OA management and ongoing clinical trials tend to use a ‘one size fits all’ approach. This partly explains the overall lack of treatment efficacy and frequent negative findings from OA clinical trials. OA is a heterogeneous condition characterised by a complex and multifactorial nature (4). This leads to the large variation in clinical presentations (pain) and responses to OA treatment. Therefore, in order to optimise treatment effects in OA, the ‘one size fits all’ treatment approach needs to change to more tailored managements and strategies targeting specific subgroups/phenotypes. It is likely that different subgroups/phenotypes will differ in the underlying causes, mechanisms and health outcomes, and will consequently require different therapeutic approaches.

There are homogenous pain subgroups/phenotypes that exist within knee OA populations. Pain experience is a complex phenomenon which is affected by factors across multiple domains–peripheral, psychological, and neurological(5). This heterogeneity could explain poor responses to pain treatments (6). Although research studies have examined these factors on an individual basis, no study has examined all of these factors in the same population. Discoveries from my own research (7-9) have demonstrated the existence of distinct OA subgroups. Moreover, my most recent work, recently received a revision in the Rheumatology journal (Top 3 journal in Rheumatology field), which is the first study to comprehensively classify pain phenotypes using a wide spectrum of factors such as neurological, psychological factors and structural damage on MRI, identified three distinct pain phenotypes. These three pain phenotypes that potentially represent subgroups may benefit from different treatment approaches in general practice, and have different health outcomes.

My research tackles the urgent unmet need for OA pain control by identifying clinical subgroups characterised by different phenotypes. The findings will facilitate the development of targeted therapies to this highly heterogeneous condition. My research, therefore, has great potential to improve quality of life and save substantial healthcare costs as a result of a reduction in joint replacements due to pain relief.

The aims of the grant are to identify OA phenotypes, understand the mechanisms underlying these phenotypes, develop targeted treatments for subsets of patients with different phenotypes and investigate the impact of pain on patients’ health outcomes.

Knee OA is a heterogeneous condition, which involves variations in pathophysiologic aetiologies, disease progression, and factors specific to patient’s pain complaint (10). There has been great interest in the OA field to identify homogenous subgroups or phenotypes that exist within knee OA populations (10, 11). Some studies have considered disease or structural progression to identify knee OA phenotypes, while others have considered pain perception (4). The latter is more clinically relevant as the ultimate goal of identifying clinical phenotypes is to improve the effectiveness of clinical interventions for knee OA patients (12). The pain experience is a complex phenomenon which is affected by factors across multiple domains – peripheral, psychological, and neurological (5). This complexity has hindered the identification of pain phenotypes in prior work. Peripheral structural damage in the knee has historically been thought to be the key origin of pain in knee OA; however, there is a weak association between radiographic disease and knee pain (13). Magnetic resonance imaging (MRI) studies show there is moderate level of evidence for an association between MRI structural abnormalities and knee pain (14). A recent study showed that 20-35% of the variation in pain is explained by structural damage (13). Psychological and neurological factors are also important contributors to pain and worse clinical outcomes in knee OA (6, 15). This heterogeneity could explain poor responses to pain treatments (6). Latent class analysis (LCA) is an effective approach for identifying underlying unobserved subgroups of individuals based on multiple observed individual characteristics within a heterogeneous population (16). It is based on a finite mixture model and hypothesises that a population is consisted of two or more underlying latent groups which may differ in the response to prevention or treatment. LCA is superior to traditional cluster analyses as it can provide a model-based clustering approach using probability and additional assessment for model fit such as maximum likelihood (17). Given the heterogeneity of pain in knee OA, successful identification of pain phenotypes using LCA while considering peripheral, psychological, and neurological domains has the potential to further our understanding of the pain experience and improve pain management in knee OA patients. There is, however, only one cross-sectional study identifying pain phenotypes related to knee OA across multiple pain-related domains (18). Furthermore, to our knowledge, no studies have validated whether identified pain phenotypes were accurate through examining the clinically relevant association with distal pain features. Therefore, the aims of this study were to identify and validate ‘knee pain phenotypes’ in an older population with an average follow-up time of 10.7 years (range 9.2-12.5 years) across different pain-related domains.

Musculoskeletal pain is a leading cause of disability in the elderly and has an adverse impact on other health outcomes such as poor physical function and reduced quality of life (19). Its prevalence is high, with an estimate of 74% in community-dwelling older adults (20). The causes of musculoskeletal pain encompass a spectrum of conditions where osteoarthritis (OA) is the most common cause of pain (21). Studies have shown that 20% of musculoskeletal pain is ascribed to OA (22) and the proportion increases markedly with age, with 78% of knee pain attributed to radiographic knee OA (ROA) (23).

OA in the knee is a disabling condition ranked as the 11^th highest contributor of the 291 conditions to global disability (24). Pain in the knee results in a significant impact on individuals’ physical and psychological outcomes (25). However, pain relief remains an unmet clinical need, with modest efficacy of currently used pain medication (26). Lack of in-depth understanding of pain mechanisms might partly explain the reduced effectiveness of current treatments.

Pain is generally considered to gradually get worse in parallel with structural changes over time (27). While worsening of pain is often noted at the group level, recent studies have found a high variation in the individual course of structural progression and pain with some patients remaining stable, experiencing improvement or worsening (27, 28). This suggests that not all patients with pain have progressive symptoms. Within this highly heterogeneous population there may be homogenous subgroups of patients who follow distinct trajectories. In this context, there have been some attempts to identify knee OA pain trajectories and their risk factors over periods of 5-8 years (27-32). These studies identified at least three trajectories with the most common factors related to pain trajectories being higher BMI (27-30), lower education level (27-30), presence of psychological problems (27-30) and comorbidities (27, 29, 30). Although peripheral structural factors have traditionally thought to be an important factor in the genesis of pain, results about the association between radiographic severity and knee pain trajectories have been mixed. No study has examined whether structural damage on magnetic resonance imaging (MRI) can predict pain trajectory, and whether it is independent of other pain dimensions.

Therefore, this study aimed to identify distinct knee pain trajectories over 10.7 years in an older population and those with ROA, and to examine predictors of identified pain trajectories including MRI-detected structural pathology, demographic, psychological, lifestyle and comorbidities.

Musculoskeletal pain is common in western countries with a prevalence estimated as high as 70% in the general population (20). It leads to restrictions in physical function and activity, reduced quality of life and disability (33). Musculoskeletal pain has become a major public health burden worldwide. A recent study of the global burden of the 328 diseases and injuries reported that low back pain, neck pain, other musculoskeletal disorders and osteoarthritis were ranked 1^st, 6^th, 7^th and 12^th, respectively, for years lived with disability (YLDs) (34).

In pain research, the concept of ‘multi-site’ or ‘multiple site’ pain (MSP) has been proposed; defined as musculoskeletal pain occurring at more than one site, although, currently, an exact definition is still unclear. The prevalence of MSP is approximately 41-75% depending on study population and number of painful sites measured (35). It has been found to be associated with poorer physical and psychological health, worse health-related quality of life, and more severe depressive symptoms as compared to single-site musculoskeletal pain in both cross-sectional and longitudinal studies (36, 37). In addition, several studies also reported the adverse effects of MSP on other health outcomes, including risk of falls (38), cognitive impairments (39) and sleep quality (40). There is also evidence of a strong dose-response relationship between the extent of pain and these outcomes. Many of these outcomes may result from and lead to reduced physical activity (PA).

Low PA is the fourth leading cause of mortality worldwide. Lack of PA is associated with an increased risk for cardio-metabolic disorders (41) such as diabetes and heart diseases; and common mental disorders (42), such as depression and anxiety. A recent meta-analysis of seven studies found that older people with musculoskeletal pain are less likely to engage in PA than those without musculoskeletal pain (43). All included studies have relied on a self-reported PA from which it is hard to quantify total PA across different domains. Self-reported activity levels are however poorly correlated with objective measures of PA participation, i.e. accelerometer, with self-reported PA estimates more likely to be higher than those measured by objectively measured PA (44). This highlights the need for accurate and reliable measurements of PA in assessing the relationship between PA and health outcomes. To our knowledge, there are no previous studies reporting on the relationship between pain at multiple sites and objectively measured physical work capacity (PWC) and PA. Therefore, the aim of this study was to describe the association between MSP and objectively measured levels of PWC and PA in a population-based sample from the UK biobank.

Metabolic syndrome (MetS) has been suggested as having a role in osteoarthritis (OA) pathogenesis. No study has assessed whether MetS and its components are associated with pain severity and number of painful sites (NPS) and their courses over time. We aimed to examine the association of MetS and its components with trajectories of pain severity and NPS in people with radiographic knee OA (ROA) over 10.7 years. 1,099 participants (mean age 63 years) from a population-based older adult cohort study were recruited at baseline. 875, 768 and 563 participants attended years 2.6, 5.1 and 10.7 follow-up, respectively. Data were collected on demographic, psychological, lifestyle and comorbidities, blood pressure, glucose, triglycerides, and high-density lipoprotein (HDL) cholesterol. MetS was defined based on National Cholesterol Education Program-Adult Treatment Panel III criteria. ROA was assessed by X-ray. Knee pain was measured by Western Ontario and McMaster Universities Osteoarthritis Index pain questionnaire at each time-point. Presence/absence of pain at the neck, back, hands, shoulders, hips, knees and feet was collected by questionnaire at each time-point. Group-based trajectory modelling was applied to identify pain trajectories. Multi-nominal logistic regression was used for the analyses. 60% of participants had ROA and 32% had MetS at baseline. Three pain severity trajectories were identified: ‘Marginal pain’ (50%), ‘Mild pain’ (35%) and ‘Moderate pain’ (15%). Three NPS trajectories were identified: ‘Low NPS’ (10%), ‘Medium NPS’ (38%), and ‘High NPS’ (52%). In univariate analyses, MetS was associated with increased risk of both ‘Mild pain’ (relative risk: 1.47, 95%CI: 1.10−1.96) and ‘Moderate pain’ (2.22, 1.54−3.20) relative to ‘Marginal pain’. It was also associated with increased risk of both ‘Medium NPS’ (2.25, 1.11−4.54) and ‘High NPS’ (3.36, 1.70−6.63) compared to ‘Low NPS’. In multivariable analyses, abdominal obesity was associated with increased risk of both ‘Mild pain’ (1.70, 1.17−2.49) and ‘Moderate pain’ (2.75, 1.63−4.64), and MetS and low HDL were associated with ‘Moderate pain’. Abdominal obesity was the only component associated with increased risk of both ‘Medium NPS’ (2.82, 1.39−5.70) and ‘High NPS’ (3.60, 1.79−7.24), and MetS was only associated with increased risk of ‘High NPS’. However, these associations became non-significant after further adjustment for body mass index, but hypertension became protective with ‘Mild pain’. MetS is predominantly associated with trajectories of pain severity and number of painful sites through abdominal obesity, suggesting that weight management is the most important way of controlling OA pain.

Pain is common in older adults typically involving multiple sites. Pain at multiple sites has been shown to be associated with a number of adverse health outcomes; however, whether pain at multiple sites is associated with fractures, and whether this association is dependent of falls risk and bone mineral density (BMD) are unknown. This study aimed to examine the associations of number of painful sites with prevalence and incidence of fractures, and to explore whether pain at multiple sites is an independent marker for fractures. Data from a longitudinal population-based study of older adults (mean age 63 years, 51% female) were utilized with measurements at baseline (n=1086), and after 2.6 (n=875) and 5.1 years (n=768). Presence/absence of pain at the neck, back, hands, shoulders, hips, knees and feet was assessed by questionnaire at baseline. Fractures were collected by questionnaire at each time-point. BMD was measured by Dual-energy X-ray absorptiometry (DXA). Baseline demographic, lifestyle and fall risks data were also obtained. A total of 385 fractures were reported at three time-points and 86 incident fractures were observed over 5.1 years. Greater number of painful sites was associated with higher prevalence of fractures at any sites, fractures occurring at vertebral, non-vertebral and major sites (including the femur, radius, ulnar, vertebrae, rib and humerus) within 5.1 years in univariable and multivariable analyses with adjustment for age, sex, body mass index, smoking history, physical activity, pain medication, BMD and falls risk. There was a dose-response relationship between number of painful sites and prevalence of fractures at these sites (all P for trend <0.05). In addition, risks of incident fractures occurring at any sites and major sites over 5.1 years increased with greater number of painful sites in a dose-response manner before and after adjustment for confounders (all P for trend ≤0.15). No significant association between number of painful sites and prevalent and incident hip fractures was observed. Pain at multiple sites is associated with a higher risk of prevalent and incident fractures, which is independent of falls risk and BMD, suggesting that widespread pain may be a useful marker for screening older population at high risk of fractures. Counting number of painful sites and managing MSP are particular of importance in clinical practice to prevent fractures and reduce health burden.

I have secured two CIA grants and NHMRC ECF fellowship in 2018 (still under embargo, please do not make any announcements on social media until NHMRC advise that the embargo has been lifted)

Improving outcomes in patients with knee osteoarthritis: identifying pain phenotypes with relationships to long-term health outcomes.

I am not a real statistician, so took me long time to learn the sophisticated statistics (latent class analysis and group based trajectory model which were used for identifying distinct pain phenotypes and trajectories). In particular, traditional methods for dealing with missing data cannot be used for trajectory models, it’s very complicated to address the influence of missing data on the results while using the trajectory models.