Background

Osteoarthritis (OA) is a highly prevalent disease, affecting approximately 10–12% of the adult population and more than 300 million people globally [1, 2]. In the absence of a cure, treatment addresses symptom management and maximising function [3]. Pain is the predominant reason for seeking medical care [4,5,6,7]. Core guidelines recommend non-pharmacological approaches such as education, exercise and weight management. For persistent and severe symptoms, pharmacological management is recommended, including analgesics (e.g., paracetamol), oral or topical non-steroidal anti-inflammatory drugs (NSAIDs), weak opioids (e.g., tramadol), stronger opioids or intra-articular corticosteroid injection, in conjunction with core treatments. However, age-related physiological changes and higher prevalence of multimorbidity increase vulnerability to adverse effects of commonly used medications in older adults [8].

Clinical practice guidelines from six professional associations recommend topical and oral NSAIDs as the mainstay of pharmacological management [9], with topicals preferred due to lower risk of adverse effects [10, 11]. Widespread and long-term use of oral NSAIDs is not recommended due to the risk of gastrointestinal, cardiovascular and renal adverse events [12, 13]. Paracetamol, which was previously consistently recommended as the first-line pharmacological option [10, 11, 14], is no longer widely recommended in guidelines [15], due to lack of effectiveness [5,6,7, 16, 17]. The small clinical benefit of oral or transdermal opioids (excluding tramadol) is outweighed by the risk of adverse events [18] and concerns about addiction [5], so these medications should only be considered when other pain management options have been exhausted [19]. Nutraceuticals employed as dietary supplements in OA therapy are not recommended due to lack of evidence [5,6,7].

Previous cross-sectional analysis of people with OA from wave 1 (2009–2011) of the Irish Longitudinal Study on Ageing (TILDA), a population cohort study, found that 63% (n = 660/1042) reported pain, compared with 25% (1521/5920) in those without OA [20]. NSAID use was associated with an almost 6-fold odds of having OA versus not having OA, although detailed analysis of pain medication usage in the OA cohort was not undertaken. Whilst numerous studies have reported pain medication usage in OA populations [21,22,23,24,25,26], few have explored patterns or combinations of use of different types of medications [27]. Various factors have been associated with OA, including age, female sex, obesity, previous injury, genetics [28], anxiety and depression [29], smoking and alcohol consumption [30], sleep disturbance [31], co-morbidities [32], and physical activity [33]. Investigating factors associated with pain severity along with medication use may help identify clinically important subgroups, thus leading to more targeted interventions.

Therefore, the aims of the current study were to (1) identify patterns of pain medication use and pain severity among people with OA aged ≥ 50 years in an Irish population cohort using latent class analysis (LCA) and (2) examine the relationship between medication pattern usage, pain and demographic and clinical factors.

Methods

Study design and population

TILDA is a nationally representative cohort study of ageing, which examines health, economic and social circumstances of people age 50 years or older living in the community in Ireland. It recruited approximately 8000 people, excluding those in institutional care. Participants were selected using a three-stage sampling process, using the Irish geodirectory sampling frame, a listing and mapping of all residential addresses in Ireland compiled by the Irish Postal Service and Ordnance Survey Ireland [34]. Data collection takes place every 2 years, with data collection for wave 1 carried out from October 2009 to February 2011. We conducted a secondary cross-sectional analysis of Wave 1 data (2009–2011) to determine baseline patterns prior to exploring longitudinal patterns of pain/pain medication use. The full TILDA study design is reported elsewhere [35]. Reporting was guided by the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [36].

TILDA data collection

TILDA data were collected using three methods. A computerised-assisted personal interview, which included questions related to socio-demographic characteristics, physical, cognitive, mental and behavioural health, lifestyle, social support and participation, health and social care utilisation including medication usage, was administered by trained interviewers. The self-completed questionnaire asked more sensitive questions, including loneliness, stress, anxiety, worry, quality of life (QoL), ageing perception and alcohol use. Participants attended a health centre for a comprehensive physical assessment.

OA was self-reported by asking ‘Has a doctor ever told you that you have any of the following conditions?’ If respondents answered yes to having arthritis, they were then asked ‘Which type or types of arthritis do you have? Response choices included ‘osteoarthritis’, ‘rheumatoid arthritis’ or ‘other types of arthritis’. Those reporting a diagnosis of OA were retained for this analysis.

Presence of pain was asked by the following question ‘Are you often troubled with pain?’ If the respondent replied ‘Yes’, they progressed to a question ‘How bad is the pain most of the time?’ Response options included ‘mild’, ‘moderate’ or ‘severe’. Respondents were asked in which part of the body pain was most severe. Pre-specified areas included back, hips, knees, feet, mouth/teeth, ‘other’ and ‘all over’. ‘Other’ referred to areas not pre-specified in the interview question.

Pain medication types

Participants showed interviewers the packaging of regular medications taken, including prescription and non-prescription medications, over-the-counter medications, vitamins, and herbal and alternative medications. Product names, but not medication strength or dosage, were recorded. Medication data were coded by an experienced pharmacist using the Anatomical Therapeutic Chemical (ATC) classification system (http://www.whocc.no/atc). The ATC codes were used to classify drugs into the following categories for this analysis: Analgesics (any paracetamol or aspirin product), Topical NSAIDs, Oral NSAIDs (including selective cox-2 inhibitors and non-selective NSAIDs), Opioids, Glucosamine/Chondroitin and combination drugs, which comprised any analgesic or NSAID combined with opioids. Due to the differentiation between tramadol and other opioids in clinical guidelines [7, 11, 37], we also subcategorised opioids into tramadol (as a single or combination product) or non-tramadol (any opioid other than tramadol).

Covariates

Socio-demographic explanatory variables included age, sex, education level, employment status and medical insurance cover (supplementary material 1).

General health, as self-reported by TILDA participants, was determined using the following categories ‘excellent’, ‘very good’, ‘good’, ‘fair’ and ‘poor’. Responses were coded as ‘excellent/very good’, ‘good’ and ‘fair/poor’. Co-morbidities were self-reported based on the following conditions: hypertension, diabetes, heart disease, cancer, lung disease, stroke, arthritis and osteoporosis and the number calculated and recoded to ‘0–1’, ‘2’ and ‘3 or more’. Sleep behaviour was ascertained through asking if there was difficulty sleeping due to arthritis, with response options of ‘yes’, ‘no’ and ‘sometimes’. Alcohol consumption was measured using the CAGE questionnaire, with a score of 2 or more suggestive of alcoholism [38]. Smoking behaviour responses were ‘never’, ‘past’ and ‘current’, with ‘current’ referring to smoking in the past 3 months. Depressive symptoms were measured using the 20-item Centre for Epidemiologic Studies Depression Scale (CESD), where scores of ≥ 16 indicate clinically relevant depression [39]. Anxiety was measured using the anxiety subscale of the Hospital Anxiety and Depression scale (HADS), with a cut-off of 11 used to indicate the presence of anxiety [40]. Self-reported physical activity levels were determined using the International Physical Activity Questionnaire (IPAQ) short version [41].

Statistical analysis

Socio-demographic and health-related characteristics were presented using frequency counts and percentages with 95% confidence intervals (CI) for categorical and ordinal data.

Latent class analysis

Medication variables and self-reported pain were used to extract latent classes representing distinct medication/pain profiles. We combined tramadol and non-tramadol opioids into a single opioid variable for the LCA, resulting in the following classes of medications: analgesics, topical NSAIDs, oral NSAIDs, opioids and nutraceuticals (glucosamine/chondroitin sulphate). Combination drugs were not included as they were already categorised in the original medication category. Two pain variables were used for LCA: presence of pain (yes/no) and pain severity (none, mild/moderate or severe).

Model selection

Determining the number of latent classes was assisted by the following model fit-criteria: Akaike’s Information Criterion (AIC), Bayesian Information Criterion (BIC), consistent AIC (CAIC) and adjusted BIC, while simultaneously assessing the classes’ distinctiveness via visual inspection, their sizes and interpretability. AIC and BIC are designed to strike a balance between model accuracy and overfitting, and lower AIC and BIC indicate better model fit [42]. Entropy was used to assess classification accuracy. While higher scores are indicative of better entropy, it should not be used to guide class-enumeration [42]. After class-enumeration, post hoc posterior probabilities of class membership are estimated for each individual, who is then assigned to the class for which they had the highest posterior probability. For simplicity, a classify-analyse approach was used for further statistical inferences: after class assignment classes were treated as known (i.e., handled as deterministic categories).

Class profiling

The relationship between additional covariates and the extracted latent classes (outcome variables) was assessed via multiple multinomial logistic regression models. Details of the variables used to characterise the latent classes for multinomial regression are available in Table 1. Inclusion of covariates in initial univariable models (Supplementary material 2) was based on previous research [22, 30, 43, 44], and known associations with OA [29, 30]. Variables that were statistically significant (p < 0.05) in the unadjusted models were added to for the multiple, multinomial logistic regression models. This forward approach was selected to avoid sparsity problems. The primary aim of the exploratory multivariable model was to investigate which covariates could be independently linked to class membership, irrespective of the direction and nature of the underlying association, for example, of the cause-effect or co-dependence. However, bearing in mind that mutual adjustments can inadvertently introduce collider bias, particularly when considering that some of these covariates are themselves health outcomes, we paid special attention to the nature and magnitude of changes in estimate (CIE). Therefore, covariates were roughly characterised into putative causes (group I) vs. co-dependents/consequences (group II) of the pain/medication profiles. Model building unfolded purposefully and hierarchically from univariable to multivariable models, first with mutual adjustment among variables in group I, among variables in group II, before fitting a final model (forest plot). For completeness, all hierarchical models are provided in supplementary material 1. Results were presented as odds ratios (ORs) and 95% confidence intervals (CIs). Analyses were conducted in Stata v15 (StataCorp, College Station, Texas, USA) and SAS v 9.4 (SAS Institute, Cary, North Carolina, USA) and Microsoft Excel (Microsoft Corp, Seattle, WA, USA) was used for generation of additional graphs.

Results

A total of 8174 people aged 50 years or over at Wave 1 were identified from the TILDA database. Following removal of those with no arthritis (n = 5919) or other types of arthritis (n = 1213), the total eligible sample for this analysis was 1042.

Table 1 shows the sample characteristics. Over two-thirds were female and 78.9% were overweight or obese. Most of the sample (57.8%) had three or more chronic conditions, with hypertension, lung disease and cardiovascular conditions most prevalent. A total of 21.2% (n = 221) had joint replacement surgery. General health was rated as very good or excellent by 43.6%, with 22.5% reporting fair or poor health, 34.7% reporting moderate or severe depressive symptoms and 9.1% reporting moderate/severe anxiety. A total of 63.5% (n = 660) reported often being troubled by pain, with over 50% reporting pain as moderate or severe, whilst 358 (34.4%) were taking some form of pain medication. The most commonly reported pain medications used were oral NSAIDs (n = 182, 17.5%; 95% CI 15.3–19.9%), analgesics (n = 119, 11.4%; 95% CI 9.6–13.5%) and opioids (n = 91, 8.7%; 95% CI 7.2–10.6%). Glucosamine/chondroitin use was reported in 8.6% (95% CI 7.08–10.50%) and topical NSAID use reported by 1.4% (95% CI 0.8–2.3%).

Table 1 Characteristics of the sample reported as having OA (n = 1042)
Full size table

Latent class analysis: class-enumeration

LCA was conducted with two up to five classes. Model fit criteria (supplementary material 2) suggested that a three-class model represented a good fit to the data (best BIC, adjusted BIC and CAIC), also yielding a solid classification accuracy (entropy = 0.87). Increasing the number of classes to four improved only one model fit criterion, AIC, but additional classes became prohibitively small (< 1%). Fit criteria deteriorated for larger number of classes (supplementary material 4).

Latent class analysis: Results

There was not muchheterogeneity in the unveiled cross-combinations of medication usage and self-reported pain, with the data showing, in particular, a low prevalence of medication usage. Yet, three drugs/pain profiles could be identified. The marking features of these classes are displayed in Fig. 1.

The item probability plot (IPP) shows the probability of individuals with OA selecting the answer categories for medication use and pain (x-axis) for each latent class, separately. While two classes (2 and 3), comprising approximately two-thirds of the sample, reported pain (ranging from mild to severe), medication use remained low for the vast majority of the sample. Only one group (class 3) showed a somewhat moderate use of opioids, analgesics and, less frequently, oral NSAIDs, consistent with the higher probability of reporting moderate and, in particular, severe pain. Given this picture, the classes were denominated as outlined in Table 2.

Fig. 1
figure 1

Item Probability Plot (IPP) illustrating the three extracted medication/pain profiles (with estimated sizes) identified from participants’ answers to medication use and pain questions. The y-axis shows the probability of OA individuals using the indicated medications and self-reporting the ‘pain’ categories for each latent class

Full size image
Table 2 Description of the final three latent classes
Full size table

Table 3 describes the characteristics of the final three classes. Whilst both class 2 and class 3 had a higher probability of pain than class 1, there are some noteworthy differences between class 2 and 3. Class 3 had a higher proportion of individuals aged ≥ 75 years, a higher proportion in receipt of a state-funded medical card, and lower proportion with private medical insurance. Whilst a similar proportion in both classes had moderate pain, a greater proportion in class 3 had severe pain than class 2. There were also differences in employment status, with fewer employed in class 3 than class 2. There was a higher number of co-morbidities in class 3 than class 2, along with a higher proportion of osteoporosis and peptic ulcer. More individuals in class 3 reported fair/poor health and a higher proportion had joint replacement surgery than in class 2. There were also some differences in pain locations between classes 2 and 3; a higher proportion had hip pain, back pain or pain all over in class 3, whilst feet pain was more prevalent in class 2.

Table 3 Characteristics of the latent classes
Full size table

Relationship between medication usage/pain profiles, and demographic and clinical factors

The unadjusted associations between socio-demographic/clinical variables and latent class membership are provided in supplementary material 2. Multicollinearity (Variance Inflation Factor (VIF) < 5) was tested and emerged non-problematic. Their adjusted counterparts, as estimated by the multiple multinomial logistic regression model, are displayed in Fig. 2. Results of the sequential multinomial regression models are also available in supplementary material 2.

Taking the low medication/no pain (class 1) as the reference class, females were more likely than men to be assigned to class 2 (OR 1.64, 95% CI 1.16–2.31). People aged 75 years or more were less likely to be assigned to class 2 (OR 0.51, 95% CI 0.32 to 0.83) than class 1, whilst those aged 65–74 years were less likely to be assigned to class 3 (OR 0.44, 95% CI 0.24–0.83).

Individuals with poor/fair self-reported health were more likely to be assigned to both class 2 (OR 2.90, 95% CI 1.80–4.68) and class 3 (OR 5.95, 95% CI 3.15–11.24) than class 1.

Sleeping difficulty was also associated with both pain classes (2 and 3). A clear difference in magnitude is noted; with a 2-fold increase in the odds of those reporting sleeping difficulty belonging to class 2 (OR 4.61, 95% CI 2.76–7.70) and class 3 (OR 10.08, 95% CI 5.34–19.02).

Furthermore, there was a larger representation in class 2 of those with moderate depressive symptoms (OR 1.49, 95% CI 1.02–2.18) and severe depressive symptoms (OR 1.96, 95% CI 1.15–3.33) compared to class 1.

Representation of the various socio-demographic and clinical factors across the three classes is shown in supplementary material 4.

Fig. 2
figure 2

Forest Plot of Multinomial Regression models comparing socio-demographic and clinical characteristics between Class 1 and Class 2, and Class 1 and Class 3

Full size image

Discussion

This study identified links between pain medications and pain severity in a population cohort with OA aged ≥ 50 years. The most common type of medication used included oral NSAIDs, followed by analgesics and opioids, aligning with previous reports [22, 45], indicating that many people with OA are not treated pharmacologically for pain.  Data from a primary care database in Ireland found topical NSAIDs comprised 20% of all NSAID prescriptions [17]. Use of topical NSAIDs was substantially lower than reported in more recent large cohort studies [45]. Despite the fact that clinical guidelines for OA preceding data collection in the current study recommended topical ahead of oral NSAID use [10, 11, 46]. Use of nutraceuticals in the current study is also considerably lower compared to other cohort studies [22], however, recommendations for glucosamine and chondroitin varied at the time of data collection [10, 11].

We identified three distinct pain medication and pain severity profiles. A total of 37% had low probability of pain and low pain medication usage. The largest class (50%) reported mild or moderate pain and low pain medication usage. The third class (13%) had moderate or severe pain, but higher probability of medication usage, especially analgesics or opioids. Both sleep difficulty and poor/fair health increased the likelihood of participants being assigned to ‘pain’ classes 2 and 3. Younger age, female gender and severe depressive symptoms were linked to low medication use and probability of mild/moderate pain (class 2), whilst fair/poor self-rated health, difficulty sleeping and being retired were differentially linked to high pain medication use and probability of moderate/severe pain (class 3), compared to the low medication/no pain class (class 1).

Previous analysis of this wave of TILDA data found the prevalence of polypharmacy (five or more medications) was significantly increased in those with OA (35%) compared to those without arthritis (15.5%) [20], likely compounded by the presence of common comorbidities of OA such as hypertension, heart disease and diabetes [47]. The current analysis provides further insight into the profile of pain medication usage, with class 3 demonstrating higher probability of taking multiple pain medications and the greatest proportion of those with three or more comorbidities (Table 3). This has implications for the pharmacological management of OA, in considering benefits, harms and interaction effects between medications.

An OsteoArthritis Initiative (OAI) cohort study [22] found that age, race, radiographic severity, BMI and co-morbidities were not associated with pain medication usage. Similar to our findings, frequent analgesic use was more common in women than men (OR 1.8, 95% CI 1.3-2.3) [22]. Higher pain medication use is well recognised in women, possibly due to genetic, hormonal and metabolic factors related to analgesic efficacy, and sex-related psychosocial factors [48]. In our analysis, age was differentially distributed across the three classes, with a lower probability of those aged ≥ 75 years being in class 2, and a lower probability of those aged 65–74 years being in class 3 compared to class 1. Older adults have increased risk of cardiovascular, gastrointestinal and renal adverse events with NSAID use [49]. Guidelines published in 2008 (preceding this data collection) [10] recommended topical NSAIDs as first-line pharmacological therapy for OA, due to emerging evidence of adverse events associated with oral NSAIDs, and withdrawal of some COX-2 selective inhibitors. This may explain this age differentiation. Difficulty sleeping and fair/poor self-rated health differentiated classes 2 and 3 from class 1. Self-rated health can predict health service use, mortality [50], social and physical health, sleep disturbance and depression among people with OA [51,52,53]. Up to 50% of people with OA have significant sleep difficulties [31, 54] and a reciprocal vicious-cycle relationship between pain and sleep can occur [55, 56].

Regular chronic use of opioids can interfere with sleep quality [57]. Whilst sleep disturbance and OA-related pain are associated with depression [58], these complex interactions are not fully understood [59]. Sleep disturbance mechanisms in OA-related pain may be associated with systemic inflammation, increased central pain processing and impaired pain inhibition driven by descending opioid and noradrenergic pathways within the central nervous system, although specific mechanistic pathways and interactions between these three factors are not fully known [56]. Age, as a confounder, increases risk of both OA and sleep difficulties [59].

Depressive symptom prevalence differed between class 2 and class 1, with a higher probability of severe depressive symptoms in class 2. This finding is contradictory to other research, showing strong associations between moderate-severe depressive symptoms and oral analgesic use, particularly opioids [44]. Depressive symptoms are common in OA, and are associated with higher pain and reduced function [29, 60]. Other factors including age, sex, ethnicity, education levels, employment status and socioeconomic status, as well as comorbidity, multiple pain sites, fatigue and sleep appear to be implicated in the relationship between OA and depression, and disentangling potential mediators in the pain-depression relationship requires further research [61]. Receipt of non-pharmacological treatment, such as exercise, which play a role in managing depressive symptoms in OA is unknown from this data [62]. However, other analysis from TILDA found that moderate and vigorous physical activity was associated with lower odds of depressive symptoms [63].

Further research is needed to determine reasons for participants’ medication-taking behaviours. Individuals with OA may be reluctant to take pain medications, commonly taking them at lower doses and less frequently than prescribed, or during pain flare-ups [64], with NSAIDs preferred over analgesics based on perceived effectiveness [65,66,67]. Reasons for reluctance to take pain medications may include perceived lack of benefit, fear of addiction, fear of running out of medication, comorbidities [67], fear of adverse events [68] and pain severity [64, 68], the latter two of which are major drivers of patient treatment preferences [68]. Research has shown that people with knee OA, when given different treatment options, preferred exercise to pharmacological management, with NSAIDs being the least preferred option, due to adverse effect concerns [69].

Qualitative research of general practitioners in Ireland highlighted challenges in prescribing medications, including NSAIDs, for older adults, due to concerns regarding medical complexity, chronicity and vulnerability to adverse drug reactions [70]. These factors may explain the low NSAID use in the current analysis.

Study strengths and limitations

Strengths include use of data from a nationally representative cohort of people aged ≥ 50 years, and the robustness of a large sample size [42]. Medication data, collected by in-home computer-assisted participant interview, demonstrates internal and external validity [71]. Using LCA provides a nuanced interpretation of pain medication usage to provide greater insights into medication used and how it relates to pain severity.

However, the cross-sectional design is limited in that it does not identify causal relationships. Longitudinal analysis across TILDA waves is ongoing, with a reduced sample size, due to loss of follow-up. Therefore, an initial exploration of baseline (Wave 1) cross-sectional data was deemed relevant. The diagnosis of OA was self-reported, based on doctor diagnosis.However, concordance between self-reported and clinically diagnosed OA has been demonstrated previously, with sensitivity ranging from 70–80% [72]. Time of diagnosis of OA is unknown as this was not asked in the TILDA survey. We excluded other arthritis types, but some of those excluded may have had co-existing OA. Medications could have included prescription or over-the-counter medications. Medication dosage and frequency was not available. Pain medication usage was based on medications taken at the time of interview, therefore there is the potential that the prevalence of medication use was underestimated by not capturing participants who were taking medications intermittently, but not at the time of interview. Similarly, pain was based on a once–off measure taken at the time of interview. Given that fluctuations in pain associated with OA are common, there is the potential for under/overestimating pain levels in this cohort.

We also cannot determine whether the pain was specifically related to OA, or if reported pain medications were specifically taken for symptoms associated with OA. Other medications used in OA management such as corticosteroid injection were also not available and we did not include other medication types, such as Serotonin-Noradrenaline Reuptake Inhibitors (SNRIs) (such as duloxetine), a common treatment of depression, which may be used for treatment of OA-related pain [73]. Use of non-pharmacological interventions for pain management was unknown. Medication usage patterns can vary between countries [21], therefore these findings may not be generalizable to other countries. As LCA assigns individuals to classes probabilistically, correct class assignment is not guaranteed [74]. The classify-analyse strategy, which discards the probabilistic assignment, may entail some classification error, which was considered negligible due to the high average posterior probabilities of assignment (> 0.90). Further, extracted classes are not real entities, but rather features of the data and therefore cannot be assumed to be concrete typologies.

Implications for research and practice

Whilst results suggest that pain medication usage in people with OA in Ireland is generally in line with international guidance concurrent with the data collection period, the low use of topical NSAIDs is noteworthy, in light of research and guidelines recommending them as an alternative to oral NSAIDs due to similar effects and better safety profile [10]. Research exploring low use of topical NSAIDs, particularly in low/moderate pain groups is warranted. Discordance between pain severity and pain medication usage in class 2 also suggests that current pharmacological management of OA is suboptimal, which requires further exploration from patient and clinician perspectives. Addressing risk factors such as obesity and physical inactivity as well as symptomatic relief is essential in OA management [75].

Associations between severe depressive symptoms and low medication use, despite moderate levels of pain, also warrants investigation. Non-pharmacological interventions such as exercise can play a role in managing depressive symptoms in OA [62]. Due to the sample proportion reporting moderate/severe depression and difficulty sleeping, screening for depressive symptoms and sleep disturbance in those with poorer self-rated health could identify treatment opportunities to address these impairments. Further understanding of the causal pathways between pain, sleep disturbance and depressive symptoms, all associated with increased levels of pro-inflammatory cytokines and altered central nervous system processing, could lead to treatments to interrupt the pain-sleep-mood dysfunction associated with OA.

Medication review is recommended in the 2022 NICE guidelines due to concerns regarding prescription of potentially harmful medications for OA [76]. Use of decision support systems and academic detailing for general practitioners, by providing evidence-based information to optimise prescribing, can enhance best-practice management [70].

Identification of specific subgroups/phenotypes is of interest in OA research [77], and the complexity of inflammatory, mechanical, cellular, metabolic, genetic and psychological factors poses considerable challenges in classifying people with OA into phenotypes for identification of therapeutic targets. However, greater understanding of pain profiles and pain medication use patterns may provide further insight in identifying clinical profiles. Numerous biomarkers associated with pain characteristics may contribute to the development of more targeted pharmacological and non-pharmacological strategies to manage OA-related pain [78].

Future research should explore longitudinal patterns of pain medication usage in OA due to the potential concerns regarding drug dependence, polypharmacy, co-morbidity and increasing life expectancy [79].

Conclusion

This study found that medication use among a population cohort of older adults with OA was generally low, and use of different medication types is broadly in line with international guidelines. By applying latent class analysis to pain-related variables and common medications used for OA in this cohort, we identified three clinical subgroups. The two subgroups exhibiting higher pain levels demonstrated poorer self-rated health and greater sleep disturbance. Identifying factors associated with pain severity and different pain medication usage may help to identify more targeted pain management interventions.