Musculoskeletal disorders are associated with chronic and debilitating pain, account for 17% of all years lived with disability worldwide, and are particularly prevalent in higher income countries (Western Europe, 20.8%; United States, 25.8%; Japan, 26.2%) [1]. The recurrent pain experienced can have considerable consequences, causing anxiety and generally negatively impacting a patient’s quality of life [2]. With the global population of individuals over 60 years of age expected to increase significantly by 2050 [3], management of these disorders, which often progress with age, is particularly important. Two of the most common conditions causing musculoskeletal pain are lower back pain (LBP) and osteoarthritis (OA) [4].

LBP, the most common musculoskeletal disorder, is experienced by approximately 38.0% of the general population, and is the leading cause of disability worldwide [5,6]. A previous study has shown that most patients experience mild LBP (25.0%), with severe LBP experienced by the least patients (5.0%) [7]. With rates increasing in the United States by 62.3% between 1992 and 2006, and more patients seeking assistance from health care professionals (HCPs), the prevalence appears to be increasing [8]. LBP can result from acute trauma, such as strenuous physical activities like lifting, pushing, or pulling a heavy load, with pain intensity ranging from sharp shooting pain, arising during sudden strenuous activity, to chronic pain that develops slowly over time as a result of degenerative changes within the spinal cord [9,10]. However, the vast majority of cases are classified as acute nonspecific, as no definitive cause can be identified, and this will be the focus of this study [11]. Frequently, nonspecific LBP is acute or subacute in nature and may resolve within days to a few weeks; however, LBP is considered chronic if the pain persists for more than 12 weeks [12].

For patients with acute nonspecific LBP, effective treatment remains a challenge [12,13]. Management is often patient centered and involves patient education and reassurance. Nonpharmacological options, such as physical exercise and manual techniques, are usually considered prior to pharmacological therapies. For chronic LBP, multidisciplinary care, involving a number of specialists, may also be considered. Paracetamol or nonsteroidal anti-inflammatory drugs (NSAIDs) are usually the first-line pharmacological options. If ineffective, stronger medications, such as opioids, antidepressants, and gabapentinoids, can be used; however, the risk of an adverse reaction is higher, and regular patient re-evaluation is required. Invasive surgery is usually the last option and is considered on a case-by-case basis (Table 1) [12,14-19].

OA is characterized by the progressive destruction of articular cartilage associated with subchondral bone remodeling, formation of osteophytes, and secondary inflammation of synovial membranes [20,21]. Pain, mediated by a number of factors, including innervation and vascularization of the articular cartilage, is a principal symptom of OA [21,22]. Compressive forces, along with hypoxia, are believed to stimulate the development of nerves, causing pain even after inflammation has subsided [23]. OA-related pain includes both nociceptive and nonnociceptive components [24], and is associated with abnormally excitable pain pathways in the peripheral and central nervous systems [21].

Symptomatic knee OA is one of the more common forms of OA, experienced by approximately 22.9% of individuals over 40 years, and is a significant cause of disability [25,26]. In France, current OA management guidelines recognize the need for a combinational approach, using both pharmacological and nonpharmacological treatments tailored to each patient. First-line pharmacological options include paracetamol and NSAIDs, followed by opioids, symptomatic slow-acting drugs, and duloxetine (off-label use). Topical agents and intra-articular treatments have also been recommended (Table 2) [27]. Furthermore, these patients often have comorbidities, which may increase the risk of drug-drug interactions and limit the range of drugs that can be used. Age-associated differences in drug sensitivities should also be considered when treating older patients [28,29].

Given the similarities in the initial treatment recommendations for both LBP and OA, paracetamol is the most commonly employed first-line analgesic for both conditions, used by over 94.4% of patients with LBP and over 96.0% of patients with OA [32-35]. There are a variety of paracetamol brands available in France, and Doliprane is the most prescribed paracetamol (Table 3) [36,37]. This is backed up by an infodemiology study analyzing 44,283 social media posts, in which Doliprane was the most mentioned paracetamol with 31.7% of posts, followed by Dafalgan with 10.9% of posts and Ixprim with 9.8% of posts [38].

The use of long-term analgesia for patients over 65 years remains challenging, and HCPs must weigh the benefits and risks given the potential increased risk of adverse effects in these patients [39,40]. A high proportion of patients discontinue prescription analgesics within the first few months due to inadequate pain relief or intolerable side effects [41,42]. With few clinical trials enrolling patients over 65 years and fewer incorporating diverse races and ethnicities, there is limited data available for patients over 65 years. Infodemiology, a relatively new method allowing real-word data collection, may help fill the knowledge gap [28]. In this project, infodemiology encompasses electronic medical records (EMRs), interactive online medical forums, and social media (Twitter), allowing analysis of this publicly available, previously unexplored, valuable source of data [43].

As the prevalence of these diseases is expected to increase with the aging population, having a thorough understanding of the requirements of patients who seek treatment is important and may assist in the development of gold-standard care. This retrospective project was designed to provide an insight into the journey of French patients with OA or LBP treated with paracetamol.

Two retrospective noninterventional studies were performed in France, using EMRs, with a focus on patients with LBP and OA. These studies were distinct, and the data were analyzed separately, but the methods used to generate the results were the same.

The third study, an OA open-source study, used social listening. For this, publicly available data were analyzed, focusing on patients with OA, from both the United States and France.

For all 3 studies, the data collected were used to identify usage of analgesics in anonymized patients with either LBP or OA. This retrospective project analyzed data extracted from EMRs provided by doctors working in the community and discussions on social media. As such, no efficacy or safety data, or reasons for termination of medication were collected.

For the EMR studies, prescription data for approximately 3 million people were obtained from general practitioners (GPs) and rheumatologists in France. Comprehensive data on diagnoses, prescriptions, referrals, physician visits, and laboratory results allowed for analysis of pharmacological and nonpharmacological therapies. For patients with OA, data were collected between September 2013 and August 2018, and the following criteria were applied: (1) the patient must have visited a GP or rheumatologist between September 2013 and August 2018; (2) the patient must have at least one OA diagnosis, defined by International Classification of Diseases Version 10 (ICD-10) codes M15-M19; and (3) the patient must have at least one paracetamol prescription. For patients with subacute, acute, or chronic LBP, data were collected between October 2013 and September 2018, and the following criteria were applied: (1) the patient must have visited a GP or rheumatologist between October 2013 and September 2018; (2) the patient must have at least one LBP diagnosis, defined by ICD-10 code M54.5; and (3) the patient must have at least one paracetamol prescription. Distinguishing between different brands of paracetamol was not possible, so drugs were separated by brand name in order to understand usage patterns and distinguish preference.

Details on drugs prescribed in addition to paracetamol were extracted from the database. The drugs included in the analysis were the most frequently used and approved drugs for the treatment of either OA or LBP. For OA, this included ketoprofen, opioids, and NSAIDs in combination with opioids and some intra-articular steroidal injections. For LBP, this included ketoprofen, opioids, and NSAIDs in combination with opioids, intra-articular steroidal injections, antiepileptic drugs, and some antidepressants. In some cases, details of dose and the type of concomitant medication were not collected and therefore not reported in this analysis.

The OA open-source study, which focused on both French and United States populations, gathered data from only publicly available sources. Questions, comments, and posts related to back pain and OA from both English and French language public medical forums and Twitter were extracted into a text data set. For the United States only, data were extracted for the year 2016 from The Centers for Disease Control and Prevention National Health Interview Survey (CDC NHIS), which comprehensively captures major aspects of a respondent’s health and wellness condition; however, only relevant data, such as demographics, occupation, income bracket, and major medical conditions, were included.

The endpoint of this project was to define the demographics of patients with OA and LBP, and develop an understanding of the management of these conditions in France, including how prescribing methods differ between GPs and rheumatologists. In addition, this project aimed to evaluate the use of paracetamol in patients with OA and LBP within France.

For the EMR studies, anonymized patient data provided by the Decision Resources Group (DRG) [44] were loaded into the querying platform (Snowflake platform version 3.12.0), and analysis was performed using Structured Query Language. The following transformations and operations were applied to the data: demographics; comorbidities and clinical profiling; treatment persistence; and patient journey.

Patients were stratified by gender and age, as well as provider specialties. Comorbidities and clinical profiling were assessed by interrogating the medical history data that triggered an entry related to a prescription, procedure, or diagnosis in the physician’s office. Medical history data were captured as part of a patient’s anamnesis. Diagnostic data were captured on an ongoing basis, and medical history data and current diagnoses were consolidated to map a patient’s comorbidity profile.

To take into account gender and age, medical history and diagnostic data were assessed to determine the frequency and distribution of diagnoses based on ICD-10 codes (either for OA [M15-M19] or LBP [M54.5]), and these data were then further segmented into gender and age groups (10-year increments).

Treatment persistence was assessed for patients who had received a paracetamol prescription. For this, the first paracetamol prescription was defined as the initiation of therapy, and longitudinal prescription data were then assessed to account for further paracetamol prescriptions over time. For patients with more than 60 days without a prescription, defined by the difference between the first prescription plus days of supply and the subsequent prescriptions, 2 separate pain therapies were reported. Prescription for another main medication, either a different brand or active compound, was considered termination of the preceding pain medication and considered a switch event; however, no distinction between concomitant and new agents was made. Treatment duration was assessed and visualized across the entire population and analyzed in smaller cohorts based on age, gender, and comorbidity profile.

Assessment of the patient journey involved analyzing pain medication dynamics for patients with either OA or LBP across the entire patient population, as well as subcohorts over time. Switching dynamics assessed the share for each pain medication brand by month. This also segmented the patient share for “new” patients, who had no preceding pain medication prescription over the observation period; “switch” patients, who received other pain medication brands and whose treatment was switched during the respective month; and “repeat” patients, who were prescribed the same brand as in the preceding month. Where the volume of data was sufficient for product brands, the patient share and uptake rate analyses were segmented across age, gender, and comorbidity subcohorts.

For the OA open-source study, Python’s Natural Language Toolkit library was employed to extract nonrelevant discussions and responses from public medical forums and Twitter data, leaving only the key relevant “talking points,” which were then broken down into key words and phrases for analysis. For the CDC NHIS data, all unanswered fields in a record were floored to zero, categorized features were label encoded and dummied to separate columns, and prevalence of OA was selected as the target variable.

All analyses conducted in this work were descriptive, and no other confirmatory statistical tests were applied. For the EMR studies, patient data were primarily summarized with percentages, which were calculated using Excel (Microsoft Corp). In some instances, the median for the duration in days was calculated. The data are reflective of what was presented within the EMRs, and no missing values were imputed. Internal DRG therapeutic area experts who specialized in OA and LBP reviewed the results of the analysis for the respective OA or LBP data. For this analysis, in some cases, absolute numbers were recalculated from percentages.

For the OA open-source study, graph/network analysis was applied to study pairwise relationships between the following: symptom-condition, condition-treatment, symptom-medicine, symptom-treatment, symptom-diet and supplements, and symptom-activities. The relationship between 2 terms was determined by their co-occurrence in the same comment or post.

This noninterventional work was performed on fully deidentified and anonymized patient EMRs, and the methodology did not require any human interaction. Therefore, no notifications to the National Competent Authorities or Ethics Committee were required by national and local regulations and requirements.