Abstract

IJPDS

International Journal of Population Data Science

IJPDS

2399-4908

Swansea University

10.23889/ijpds.v11i1.3350

11:1:50

Population Data Science

Medication Use Trajectories in Chronic Pain: Comparing Risk-Based and Count-Based Approaches Using Linked Prescription Claims Data

Nguefack

Hermine Lore Nguena

¹ Ménard

Nancy

² Beaudoin

Sylvie

² Pagé

M. Gabrielle

³⁴ Guénette

Line

⁵⁶ Hudon

Catherine

⁷ Lacasse

Anaïs

¹* 1Department of Health Sciences, Université du Québec en Abitibi-Témiscamingue (UQAT), 445 boulevard de l’Université, Rouyn-Noranda (QC), J9X 5E4, Canada 2Patient partner, Rouyn-Noranda (QC), Canada 3Research Center, Centre hospitalier de l’Université de Montréal (CHUM), 900 rue Saint-Denis, Montréal (QC), H2X 0A9, Canada 4Department of Anesthesiology and Pain Medicine, Faculty of Medicine, Université de Montréal, 2900, boulevard Édouard-Montpetit, Montréal (QC), H3T 1J4, Canada 5Population Health and Optimal Health Practices Research Unit, CHU de Québec Research Center – Université Laval, 1050 chemin Sainte-Foy, J1-08, Québec (QC), G1S 4L8, Canada 6Faculté de pharmacie, Université Laval, 1050 avenue de la Médecine, local 2345, Québec (QC), G1V 0A6, Canada 7Family Medicine and Emergency Medicine Department, Université de Sherbrooke, 3001 12e avenue Nord, Sherbrooke (QC), J1H 5N4, Canada

*Corresponding author: Anaïs Lacasse, lacassea@uqat.ca Statement on conflicts of interest

The authors have no conflicts of interest with any financial organisation regarding the material discussed in the manuscript.

04112026

2026

11 1

3350

This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

This article is available from the IJPDS website at: https://ijpds.org/article/view/3350

Abstract Introduction

In the context of chronic pain, no study has yet compared risk-based and count-based approaches for modelling medication use trajectories. Risk-based approach reflects the potential burden of medication-related adverse effects, whereas count-based approach reflects treatment burden and regimen complexity. Comparing these complementary measures may identify different patient profiles depending on the clinical or research objective.

Objectives

This study aimed to compare a composite medication risk score, the Medication Quantification Scale 4.0 (MQS-4.0), with medication count in examining medication use trajectories among individuals with chronic pain.

Methods

We conducted a cohort study using data from the TorSaDE Cohort, which links five cycles of Statistics Canada’s Canadian Community Health Survey (2007-2016 cross-sectional questionnaires) and Québec provincial health administrative databases (1996-2016 longitudinal data). We selected 8,760 adults with chronic pain who were covered by public prescription drug insurance. Pain-related medications used in the two-year period following survey completion were operationalised as monthly risk scores and monthly medication count. Growth Mixture Modelling was applied to each of these measures to identify groups with similar patterns over time (trajectories).

Results

Both the MQS-4.0 risk score and the medication count were effective in forming trajectory groups that were statistically robust (based on entropy and Bayesian information criterion) and clinically meaningful. However, the MQS-4.0 risk score approach was more discriminative, identifying differences between trajectory groups in terms of sociodemographic characteristics, pain profiles, health status, and healthcare utilisation. Individuals with the highest pain-related medication count did not necessarily have the highest MQS-4.0 risk scores, and some individuals using fewer medications were found to be at high risk.

Conclusions

Our findings suggest that a risk-based approach such as the MQS-4.0 may provide complementary information to a count-based approach for subgroup analyses, particularly when using health administrative databases to study pain-related medication use in individuals living with chronic pain.

chronic pain polypharmacy analgesics medications risk trajectories medication quantification scale growth mixture modelling health administrative data

Highlights

This study compares a risk-based (MQS-4.0) approach to a count-based approach for modelling medication use trajectories in chronic pain.

Both approaches identified statistically robust trajectory groups, but MQS-4.0 was more discriminative in differentiating sociodemographic, pain, and health characteristics.

The study highlights that high medication count does not always correlate with high risk, thus challenging traditional assumptions.

Results suggest that risk-based approach such as the MQS-4.0 may provide complementary information to a count-based approach for subgroup analyses.

Introduction

Chronic pain (CP) affects one in five adults [1, 2] and is clinically defined as pain that persists or recurs over more than three months [3]. In 2019, the annual financial burden of CP in Canada in terms of healthcare and lost productivity was estimated at $38.3–40.4 billion [4]. CP treatment involves a combination of physical and psychological therapy (physiotherapy, meditation, psychotherapy, etc.), in addition to medication [4]. Polypharmacy, defined as the concurrent use of multiple medications [5], is highly prevalent in this population [6, 7]. A recent Canadian study highlights that 71% of adults living with CP concomitantly use five or more medications (a common threshold for polypharmacy), while 26% concomitantly use 10 or more medications (i.e., excessive polypharmacy) [8]. Although polypharmacy can lead to positive outcomes by approaching pain through multiple mechanisms of action, many medications can pose substantial risks, including adverse effects [9, 10], drug interactions [9, 10], drug cascades [11], as well as depression [12], lower quality of life [13], and mortality [14, 15].

Seeing that individuals living with CP have varying medication regimens that involve different therapeutic classes and dosages that change over time [16, 17], it is essential to gain a better understanding of the risk associated with medication use among individuals living with CP over time (risk trajectories). Trajectory analyses rely on person-centred statistical approaches to identify subgroups of individuals with similar longitudinal patterns [18]. Identifying these distinct trajectories helps characterise individuals at greater or lower risk of adverse outcomes and provides insight into the modifiable factors associated with each trajectory. Such knowledge can guide the development of tailored interventions aimed at improving outcomes among high risk groups. In CP pharmacological treatment, a research tool designed to quantify the risk associated with pain-related medications, the Medication Quantification Scale (MQS), was developed in 1992 and various versions were validated [19–25]. Its fourth version (the MQS-4.0) has been validated in Canadian practice [22].

The relevancy of composite medication risk scores, such as the MQS-4.0, is expected as the associated risk of adverse effects is not always directly proportional to the number of medications taken. However, such scores sometimes require complex calculations. It is therefore worth questioning whether simply counting the number of medications offers comparable value for research purposes.

This study thus aimed to compare a composite medication risk score, the Medication Quantification Scale 4.0 (MQS-4.0), with a count-based approach in examining medication use trajectories among individuals with CP. Our intention was to examine how a risk-based measure and a count-based measure may differently classify medication use trajectories in a real-world chronic pain population. Although related, medication risk and medication count represent distinct constructs. Medication count primarily reflects treatment burden and regimen complexity, whereas the MQS-4.0 incorporates information on medication classes to better estimate the potential burden of medication-related adverse effects. Consequently, the choice of measure should depend on the research or clinical objective. A count-based approach may be appropriate when examining polypharmacy prevalence, healthcare resource use, or medication regimen complexity, while a risk-based approach may be more informative when studying medication safety, adverse effects, deprescribing priorities, or identifying individuals who may benefit from medication review. Specifically, we compared these two approaches based on: (1) their utility in creating trajectory groups (i.e., individuals with similar patterns over time) that are statistically robust and clinically interpretable; (2) their ability to generate trajectory groups that are distinct in terms of sociodemographic profiles, pain characteristics, health status, and healthcare utilisation; and (3) the overlap between the high risk or high-count individuals identified respectively through the MQS-4.0 risk score trajectory analysis and the count-based trajectory analysis. We hypothesised that heavy users of pain medication are not necessarily at higher risk of adverse effects per the MQS-4.0 score, and individuals using fewer medications may still face substantial risk.

Methods

This manuscript is reported in accordance with the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement.

Design

This retrospective cohort study used data from the TorSaDE Cohort (n = 102,148) [26], a research infrastructure linking five cycles of Statistics Canada’s Canadian Community Health Survey (2007-2008 to 2015-2016 cross-sectional questionnaires) and Québec provincial health administrative databases (1996-2016 longitudinal data).

Data Sources

The TorSaDE Cohort has been described in detail elsewhere [26].

Canadian Community Health Survey

This Statistics Canada survey collects health and socioeconomic data on Canadians aged 12 and older using probability sampling [27]. Standardised questionnaires are used and data quality is maximised through a variety of methods, including rigorous interviewer training and various control measures.

Health Administrative Databases

In Canada’s universal healthcare system, public health insurance covers medical consultations, emergency visits, hospitalisations, and procedures for all Québec residents, except those with temporary permits [28]. Prescription drug coverage is mandatory through public or private plans. The public prescription drug insurance plan covers individuals who: 1) lack private drug insurance; 2) are aged 65 or older; or 3) receive last-resort financial assistance. These groups represent approximately 46% of the Québec population [28], and 57 to 68% of individuals living with CP [29, 30]. The pharmaceutical services database contains detailed information on prescription drugs dispensed in the community and covered by the public prescription drug plan [31].

Linkage and Data Access

When taking part in the Canadian Community Health Survey, participants give informed consent to Statistics Canada for the linkage of their responses to provincial health administrative databases and use for research purposes. Any given survey year, 90% to 95% of survey participants consent to the linkage with data from other sources for research purposes. Linkage with administrative data was possible for more than 95% of participants (person-level linkage) [26]. The linkage was conducted by the Institut de la statistique du Québec, the official government agency responsible for collecting, producing, and disseminating statistical information about Québec. De-identified data from the TorSaDE Cohort was accessed via the secure server of the Institut de la statistique du Québec (ISQ), which acts as the mandated data access authority. The preparation and cleaning of the TorSaDE Cohort data then required over 1,000 hours of work by our team’s experienced biostatisticians specialised in the analysis of large-scale administrative databases, ensuring high data quality and reliability for research purposes. Ethical approval was obtained from the Commission d’accès à l’information du Québec (#1013990), the Institut de la statistique du Québec (ISQ), and relevant university Research Ethics Boards (Université du Québec en Abitibi-Témiscamingue: #2018-02-Lacasse, A.; Centre hospitalier universitaire de Sherbrooke (CHUS): #2017-1504).

Study Population and Selection Criteria

This study included 8,760 individuals selected according to the following criteria applied to the TorSaDE Cohort: (1) for individuals with more than one survey participation, only the first participation was retained; (2) individuals reporting CP : individuals who answered “No” to the survey question “Are you usually free of pain or discomfort?”—while this definition may differ from commonly used definitions of chronicity based on the duration of symptoms [3, 32], it has been used in many CP epidemiology studies [33–48] and provides prevalence estimates (17-22% [33, 35, 36, 46]) comparable to those obtained in studies using a more traditional definition (20-21% [1, 2]) during similar periods; (3) individuals aged 18 years or older; and (4) individuals covered by the public prescription drug insurance plan during the two years following their participation in the Canadian Community Health Survey.

Study Variables

Analgesic and co-prescribed medication use (hereafter referred to as pain-related medications) was operationalised using two approaches for comparison: the MQS-4.0 risk score and the number of medications used.

MQS-4.0 Risk Score

The MQS-4.0 is a validated research tool designed to quantify the risk associated with 36 medication subclasses used for CP management and associated comorbidities (e.g., sleep, mood) (Supplementary Appendix 1) [22]. The MQS-4.0 reflects the overall risk of short- or long-term adverse effects caused by medications [22]. Pain-related medications were assigned weights (points), which could then be summed to compute a composite risk score, where higher scores mean higher risk of adverse effects. Just like previous versions of the tool, the MQS-4.0 weighting system is grounded in the clinical judgement of prescribers. Specifically, weights were derived from the median scores on a 0–10 scale assigned to each medication subclass in a Canadian survey in which physicians, pharmacists, and nurse practitioners who prescribed or adjusted analgesic medications in their clinical practice evaluated the level of risk associated with different medications. The definition of the overall risk of medication was specified to participants before survey completion, that is, “overall risk of medication to cause a short- or long-term adverse reaction, such as organ-specific or systemic toxicity (gastrointestinal symptoms, central nervous system), medication interactions, physical or psychological dependence potential, abuse potential, insomnia, tolerance, increased pain perception over time (hyperalgesia), and memory or concentration problems”. Approximately 11% of respondents were chronic pain experts, while the remaining participants practised in primary care or other clinical specialties [22]. Consequently, the weights used to calculate MQS-4.0 scores reflect perspectives drawn from diverse real-world clinical practice settings. The Medication Quantification Scale (MQS), was developed in 1992 [21]. Updated in 1998 (MQS-II) [20], in 2003 (MQS-III) [19], and in 2022 by our team (MQS-4.0) [22], it has been shown to be reliable, valid, and sensitive to change [19–25]. The MQS was used as the quantification of the risk associated with pain medications use in a large number of pain studies between 2000 and 2026 [49–71], for instance as an outcome measure in various clinical trials on the impacts of interventions, or as a dependent, independent or control variable in observational studies. Our research team developed a valid SAS^® algorithm to apply MQS-4.0 risk weights to prescribed drug insurance data [22], utilising American Hospital Formulary Service (AHFS) codes, supplemented at times by non-proprietary name codes (Supplementary Appendix 1). In our study, we used prescribed drug insurance data to calculate monthly MQS-4.0 risk scores (30-day periods) for each participant over the two years following the survey completion date.

Number of Mdications (Medication Count)

Prescribed drug insurance data was used to calculate the monthly number of different pain-related medications used by each participant over the same two-year period. To enable an adequate comparison of the two approaches, only medications covered by the MQS-4.0 were included in this calculation, excluding those prescribed for other indications. For both approaches (MQS-4.0 risk score and medication count), participants were considered users of a specific medication in a given month if at least one claim for the medication was recorded during that month, regardless of prescription duration.

Furthermore, we assessed medication use status at the time of survey completion using health administrative data. For this assessment, we applied a 90-day window to account for potential refill gaps [72]. Then, using the number of different pain-related medications used at least once in the 90-day period following survey completion, we classified participants based on medication count: five or more medications (i.e., polypharmacy), and 10 or more medications (i.e., excessive polypharmacy) [5, 73]. We followed the same approach for medications used across all therapeutic indications. As polypharmacy can also be defined as the use of two or more medications for a single health condition [74], we also applied this cut-off to classify participants based on the number of pain-related medications used.

Covariables

Variable selection aimed to describe our sample in terms of sociodemographic status, health status, and healthcare utilisation. Such self-reported variables included age, sex at birth, education level, relationship status, annual household income, administrative region of residence, pain intensity (none/mild/moderate/severe), activities prevented by pain or discomfort (none/a few/some/most), and self-perceived general health or mental health (fair/poor/good/very good/excellent). We used health administrative data to calculate the following variables during the two-year period following survey completion: continuity of care index (percentage of visits to the same practitioner [75]), number of different prescribers, and number of visits to a general practitioner, to a specialist, or to the emergency department. Administrative data covariables also included the number of hospitalisations (episodes), benzodiazepine use, and opioid use during the two-year period. We also considered all-cause mortality (dying at some point, after the end of the two-year period during which the trajectories were modelled) as descriptive outcome to characterise participants.

Statistical Analysis

We first described participant characteristics using descriptive statistics. We described baseline MQS-4.0 risk scores, the number of pain-related medications, and various polypharmacy classifications for the whole sample and across sex groups according to best practices [76].

Trajectory Groups

Growth Mixture Modelling (GMM) [18, 77] was applied to the monthly MQS-4.0 risk scores, and then to the monthly number of pain-related medications over the two-year period following survey completion. This analytical approach allowed for the grouping of individuals with similar outcomes (MQS-4.0 risk scores or medication count) over the two-year period following survey completion time. We then used trajectory group membership as a categorical dependent variable. We tested models with 1 to 6 trajectories using linear, quadratic and cubic shapes. We first tested the 18 models for the monthly MQS-4.0 risk scores, and then for the monthly number of pain-related medications. The following criteria were used successively to select the best model: (1) at least 5% of individuals per group [18]; (2) entropy ≥ 0.8 (implying less uncertainty in the classification) [77–79]; (3) highest proportion of individuals per group with posterior probability group ≥ 0.8 [78]; (4) lowest Bayesian information criterion (BIC) in absolute value [18]. To enhance the robustness of the trajectory modelling, growth mixture models were estimated using 100 initial starting values. This approach reduces the risk of convergence to local maxima and increases the likelihood of identifying the global maximum likelihood solution, thereby improving the stability and reproducibility of the trajectory groups [80]. See Supplementary Appendix 2 for modelling fit indices.

We described the profiles of trajectory groups using descriptive statistics, and performed comparative analyses using bivariable statistical testing (Chi-square tests with Tukey style multiple comparisons of proportions [81] and Kruskal-Wallis test with Dwass-Steel-Critchlow-Fligner test [82] for post-hoc pairwise analyses), with statistical significance defined as a p-value < 0.05. As the aim was not to identify causal or modifiable risk factors, but to describe real-world differences between groups, we opted for bivariate comparisons as opposed to multivariable analysis. Lastly, a Venn diagram illustrated the overlap between high risk individuals identified by the MQS-4.0 trajectory and those identified by the pain medication count trajectory [83].

The proportion of missing data across outcomes and compared variables was low (0–3.7%; see footnotes of Tables 2–4); all analyses were conducted on the available data without imputation. Growth mixture models for MQS-4.0 risk scores were estimated using the hlme function from the R package lcmm (R Core Team, Vienna, Austria), which is restricted to Gaussian outcomes [84]. Given the distributional properties of medication count data (non-negative integers with a skewed distribution), an alternative modelling approach appropriate for count-type outcomes was used (FLXMRziglm function from the R package flexmix). All other analyses were carried out with SAS 9.4 (SAS Institute, Cary, NC, USA).

Results Sample

This study was conducted among 8,760 adults living with CP and covered by the public drug insurance plan during the two years following their participation in the Canadian Community Health Survey (Figure 1). Table 1 shows the profile of the study sample. Table 2 shows baseline MQS-4.0 risk scores, the number of pain-related medications, and polypharmacy classifications (2 or more, 5 or more and 10 or more).

Figure 1 Study Sample Selection Flowchart ^aNo chronic pain n = 64,885; missing response to the question about pain n = 21,118; chronic pain percentage calculated using non-missing data = 16,145^*100/(102,148-21,118)=19,9%.

Table 1Sample Characteristics

Characteristics^a (n = 8,760)	No. (%) of participants^b
Sociodemographic profile
Age (years) – mean ± SD	63.6	± 14.8
median (IQR)	66.0	(18.0)
Age (years) – n (%)
18-24	143	(1.6)
25-44	817	(9.3)
45-64	3,042	(34.7)
≥ 65	4,758	(54.3)
Sex – n (%)
Female	5,684	(64.9)
Male	3,076	(35.1)
Education level – n (%)
No secondary diploma	3,600	(41.7)
Secondary diploma	1,112	(12.9)
College diploma/Registered apprenticeship or other trades certificate or diploma	2,934	(33.9)
University diploma	995	(11.5)
Relationship status – n (%)
Not in a relationship	5,066	(57.7)
In a relationship	3,699	(42.3)
Annual household income (Can$) – n (%)
< 20,000	2,964	(33.8)
20,000 -39,999	3,329	(38.0)
40,000 – 59,999	1,391	(15.9)
60,000 -79,999	592	(6.8)
≥ 80,000	484	(5.5)
Living in a remote region – n (%)
No	6,384	(72.9)
Yes	2,376	(27.1)
Pain symptoms
Pain intensity – n (%)
Mild	1,759	(20.3)
Moderate	4,703	(54.2)
Severe	2,209	(25.5)
Activities prevented by pain or discomfort – n (%)
None	3,001	(34.4)
A few	2,881	(33.1)
Some	1,456	(16.7)
Most	1,373	(15.8)
Health profile
Self-perceived general health – n (%)
Excellent, very good or good	5,254	(60.1)
Fair or bad	3,491	(39.9)
Self-perceived mental health – n (%)
Excellent, very good or good	7,510	(89.1)
Fair or bad	923	(10.9)
All-cause mortality (died, at some point, after the 2-year period during which the trajectories were modelled) – n (%)
No	8,214	(93.8)
Yes	546	(6.2)
Healthcare and medication utilisation during the two years following survey completion
Continuity of care index (%) – mean ± SD	84.7	± 20.6
median (IQR)	100.0	(28.6)
Number of different prescribers – mean ± SD	4.2	± 3.5
median (IQR)	3.0	(4.0)
Number of general practitioner visits – mean ± SD	6.8	± 6.3
median (IQR)	5.0	(6.0)
Number of specialist visits – mean ± SD	8.2	± 12.3
median (IQR)	5.0	(9.0)
Number of emergency department visits – mean ± SD	1.5	± 2.5
median (IQR)	1.0	(2.0)
Number of episodes of hospitalisation – mean ± SD	0.6	± 1.1
median (IQR)	0.0	(1.0)
Filled at least one prescription for benzodiazepines during the two years following survey completion – n (%)	2,940	(33.6)
Filled at least one prescription for opioids during the two years following survey completion – n (%)	2,011	(23.0)

^aProportion of missing data across presented variables ranged between 0.0% and 5.4%. Listwise deletion was applied for the subsequent analyses. ^bUnless stated otherwise. IQR: Interquartile range. SD: Standard deviation.

Table 2Medications Used at the Time of Survey Completion

Medication variables	Whole sample	Females	Males
Risk associated with pain-related medications according to the Medication Quantification Scale (MQS-4.0)
Medication Quantification Scale (MQS-4.0) score –
mean ± SD	8.7 ± 9.3	9.4 ± 9.4	7.4 ± 8.8
median (IQR)	6.0 (13.0)	8.0 (14.0)	5.0 (12.0)
Pain-related medications used (those considered in the calculation of the MQS-4.0)
Number of different medications used –
mean ± SD	1.4 ± 1.7	1.5 ± 1.7	1.2 ± 1.53
median (IQR)	1.0 (2.0)	1.0 (2.0)	1.0 (2.0)
≥ 2 medications – n (%)	3,275 (37.4)	2,341 (41.2)	934 (30.4)
≥ 5 medications – n (%)	485 (5.5)	354 (6.2)	131 (4.3)
≥ 10 medications – n (%)	<1%^a	<1%^a	<1%^a
Medications used across all therapeutic indications
Number of different medications used –
mean ± SD	6.4 ± 5.1	6.9 ± 5.1	5.6 ± 5.0
Median (IQR)	6.0 (7.0)	6.0 (7.0)	5.0 (7.0)
≥ 5 medications – n (%)	5,180 (59.1)	3,577 (62.9)	1,603 (52.1)
≥ 10 medications – n (%)	2,152 (24.6)	1,552 (27.3)	600 (19.5)

IQR: Interquartile range. SD: Standard deviation. Proportion of missing data across presented variables: 0.0%.

^aWhen cell sizes are too small, the Institut de la statistique du Québec restricts data release for confidentiality reasons.

Table 3Patient Characteristics According to Trajectory Group Membership Determined using MQS-4.0 Risk Scores

Population characteristics	Low stable risk	Moderate stable risk	Increasing to high risk	High risk decreasing	p-value^a
	2,988 (34.1)	4,469 (51.0)	654 (7.5)	649 (7.4)
Sociodemographic profile
Age – n (%)^d
< 65 years old	1,721 (57.6)	1,668 (37.3)	320 (48.9)	293 (45.2)	< .0001
≥ 65 years old	1,267 (42.4)	2,801 (62.7)	334 (51.1)	356 (54.8)
Sex – n (%)
Female	1,774 (59.4)	3,021 (67.6)	443 (67.7)	446 (68.7)	< .0001
Male	1,214 (40.6)	1,448 (32.4)	211 (32.3)	203 (31.3)
Education level – n (%)
No secondary school diploma	926 (31.5)	2,064 (46.7)	327 (50.8)	283 (44.2)	< .0001
Secondary school diploma or more	2,012 (68.5)	2,355 (53.3)	317 (49.2)	357 (55.8)
Relationship status – n (%)
Not in a relationship	1,627 (54.5)	2,618 (58.6)	394 (60.2)	417 (64.3)	< .0001
In a relationship	1,359 (45.5)	1,848 (41.4)	260 (39.8)	232 (35.7)
Annual household income (Can$) – n (%)
< 20,000	769 (25.7)	1,649 (39.9)	268 (41.0)	278 (42.8)	< .0001
≥ 20,000	2,219 (74.3)	2,820 (60.1)	386 (59.0)	371 (57.2)
Living in a remote region
No	2,277 (76.2)	3,203 (71.7)	443 (67.7)	461 (71.0)	< .0001
Yes	711 (23.8)	1,266 (28.3)	211 (32.3)	188 (29.0)
Pain symptoms
Pain intensity – n (%)
Mild	826 (27.9)	769 (17.4)	87 (13.4)	77 (12.0)	< .0001
Moderate/Severe	2,137 (72.1)	3,646 (82.6)	562 (86.6)	567 (88.0)
Activities prevented by pain or discomfort – n (%)
A few	2,309 (77.7)	2,918 (65.6)	338 (52.2)	317 (49.1)	< .0001
Some/Most	663 (22.31)	1,528 (34.4)	309 (47.8)	329 (50.9)
General and mental health
Self-perceived general health – n (%)
Good/Very good/ Excellent	2,266 (76.0)	2,438 (54.6)	279 (42.8)	271 (41.8)	< .0001
Fair/Poor	717 (24.0)	2,024 (45.4)	373 (57.2)	377 (58.2)
Self-perceived mental health – n (%)
Good/Very good/ Excellent	2,689 (92.8)	3,788 (88.4)	527 (83.5)	506 (81.5)	< .0001
Fair/Poor	209 (7.2)	495 (11.6)	104 (16.5)	115 (18.5)
Healthcare and medication profile
Continuity of care^b – mean ± SD	83.4 ± 22.3	85.5 ± 19.8	84.5 ± 19.7	84.1 ± 19.7	0.0214
Number of different prescribers^b – mean ± SD	2.8 ± 2.5	4.6 ± 3.4	6.2 ± 4.3	6.0 ± 4.1	< .0001
Number of GP visits^b – mean ± SD	4.5 ± 4.8	7.6 ± 6.1	9.8 ± 8.1	9.1 ± 7.7	< .0001
Number of specialist visits^b – mean ± SD	5.3 ± 8.7	9.1 ± 13.2	12.4 ± 16.9	11.4 ± 11.4	< .0001
Number of ED visits^b – mean ± SD	0.9 ± 1.7	1.5 ± 2.5	2.8 ± 3.8	2.2 ± 3.4	< .0001
Number of episodes of hospitalisation^b – mean ± SD	0.3 ± 0.7	0.6 ± 1.1	1.1 ± 1.4	1.0 ± 1.5	< .0001
Polypharmacy (use of ≥2 medications for the same condition)^c – n (%)
No	2,876 (96.3)	2,165 (48.4)	288 (44.0)	156 (24.0)	< .0001
Yes	112 (3.7)	2,304 (51.6)	366 (56.0)	493 (76.0)
Excessive polypharmacy (use of ≥ 10 all-cause medications)^c – n (%)
No	2,869 (96.0)	3,048 (68.2)	406 (62.1)	285 (43.9)	< .0001
Yes	119 (4.0)	1,421 (31.8)	248 (37.9)	364 (56.1)
Use of benzodiazepines^b – n (%)
No	2,779 (93.0)	2,537 (56.8)	245 (37.5)	259 (39.9)	< .0001
Yes	209 (7.0)	1,932 (43.2)	409 (62.5)	390 (60.1)
Use of opioids^b – n (%)
No	2,684 (89.8)	3,374 (75.5)	325 (49.7)	366 (56.4)	< .0001
Yes	304 (10.2)	1,095 (24.5)	329 (50.3)	283 (43.6)
Mortality
All-cause mortality – n (%)
No	2,900 (97.1)	4,129 (92.4)	588 (89.9)	597 (92.0)	< .0001
Yes	88 (2.9)	340 (7.6)	66 (10.1)	52 (8.0)

^aChi-square tests or Kruskal-Wallis tests with Tukey style multiple comparisons of proportions or Dwass-Steel-Critchlow-Fligner tests for post-hoc pairwise analyses. ^bMeasured in the two years after survey completion. ^cMeasured in the year before survey completion. Proportion of missing data across presented variable ranges between 0.0% and 3.7%. P-values < 0.05 are highlighted in bold. ^dAge was dichotomised at 65 years, as most polypharmacy studies focus on older adults [85, 86].

Table 4Patient Characteristics According to Their Trajectory Group Membership Determined using the Number of Pain-Related Medications

Population characteristics	Zero pain-related medications	Occasional use of pain-related medications	Low stable number of pain-related medications	Moderate stable number of pain-related medications	High number of pain-related medications	p-value^a
	1,307 (14.9)	1,831 (20.9)	2,750 (31.4)	2,137 (24.4)	735 (8.4)
Sociodemographic profile
Age – n (%)
< 65 years old	575 (44.0)	847 (46.3)	1,240 (45.1)	992 (46.4)	348 (47.4)	0.7009
≥ 65 years old	732 (56.0)	984 (53.7)	1,510 (54.9)	1,145 (53.6)	387 (52.6)
Sex – n (%)
Female	820 (62.7)	1,227 (67.0)	1,780 (64.7)	1,391 (65.1)	466 (63.4)	0.1328
Male	487 (37.3)	604 (33.0)	970 (35.3)	746 (34.9)	269 (36.6)
Education level – n (%)
No secondary school diploma	487 (38.1)	773 (42.8)	1,154 (42.5)	890 (42.1)	296 (40.7)	0.0006
Secondary school diploma or more	792 (61.9)	1,034 (57.2)	1,561 (57.5)	1,223 (57.9)	431 (59.3)
Relationship status – n (%)
Not in a relationship	749 (57.3)	1,070 (58.5)	1,558 (56.7)	1,248 (58.4)	431 (58.9)	0.6360
In a relationship	558 (42.7)	760 (41.5)	1,191 (43.3)	889 (41.6)	301 (41.1)
Annual household income (Can$) – n (%)
< 20,000	377 (28.8)	657 (35.9)	945 (34.4)	735 (34.4)	250 (34.0)	< .0001
≥ 20,000	930 (71.2)	1,174 (64.1)	1,805 (65.6)	1,402 (65.6)	485 (66.0)
Living in a remote region
No	960 (73.5)	1,327 (72.5)	2,019 (73.4)	1,546 (72.3)	532 (72.4)	0.0116
Yes	347 (26.5)	504 (27.5)	731 (26.6)	591 (27.7)	203 (27.6)
Pain symptoms
Pain intensity – n (%)
Mild	254 (19.6)	372 (20.6)	542 (19.9)	438 (20.7)	153 (21.1)	0.8819
Moderate/Severe	1,042 (80.4)	1,438 (79.4)	2,180 (80.1)	1,678 (79.3)	574 (78.9)
Activities prevented by pain or discomfort – n (%)
A few	913 (70.2)	1,215 (66.8)	1,845 (67.5)	1,418 (66.8)	491 (66.9)	0.2566
Some/Most	388 (29.8)	604 (33.2)	888 (32.5)	706 (33.2)	243 (33.1)
General and mental health
Self-perceived general health – n (%)
Good/Very good/ Excellent	808 (61.9)	1,071 (58.6)	1,625 (59.2)	1,307 (61.2)	443 (60.4)	0.2539
Fair/Poor	498 (38.1)	756 (41.4)	1,118 (40.8)	828 (38.8)	291 (39.6)
Self-perceived mental health – n (%)
Good/Very good/ Excellent	1,103 (88.7)	1,581 (89.1)	2,343 (88.7)	1,833 (89.1)	650 (90.8)	0.5962
Fair/Poor	141 (11.3)	193 (10.9)	299 (11.3)	224 (10.9)	66 (9.2)
Healthcare and medication profile
Continuity of care^b – mean ± SD	84.0 ± 20.5	83.9 ± 21.0	85.2 ± 20.4	84.7 ± 20.7	85.6 ± 20.6	0.0470
Number of different prescribers^b – mean ± SD	4.43 ± 3.7	4.3 ± 3.6	4.2 ± 3.4	4.2 ± 3.4	4.1 ± 3.3	0.5699
Number of GP visits^b – mean ± SD	6.4 ± 5.8	7.2 ± 6.2	7.0 ± 6.7	6.6 ± 5.9	6.9 ± 6.4	0.0029
Number of specialist visits^b – mean ± SD	8.5 ± 10.6	8.3 ± 12.0	8.1 ± 11.2	8.3 ± 15.2	8.1 ± 9.6	0.8195
Number of ED visits^b – mean ± SD	1.3 ± 2.3	1.5 ± 2.6	1.6 ± 2.8	1.4 ± 2.4	1.4 ± 2.5	0.0396
Number of episodes of hospitalisation^b – mean ± SD	0.5 ± 1.0	0.6 ± 1.1	0.6 ± 1.1	0.6 ± 1.1	0.6 ± 1.1	0.2478
Polypharmacy (use of ≥2 medications for the same condition)^c – n (%)
No	1,307 (100.0)	1,023 (55.9)	1,557 (56.6)	1,188 (55.6)	410 (55.8)	< .0001
Yes	0 (0.0)	808 (44.1)	1,193 (43.4)	949 (44.4)	325 (44.2)
Excessive polypharmacy (use of ≥ 10 all-cause medications)^c – n (%)
No	980 (75.0)	1,362 (74.4)	2,075 (75.5)	1,644 (76.9)	547 (74.4)	0.3771
Yes	327 (25.0)	469 (25.6)	675 (24.5)	493 (23.1)	188 (25.6)
Use of benzodiazepines^b – n (%)
No	1,307 (100.0)	1,081 (59.0)	1,670 (60.7)	1,306 (61.1)	456 (62.0)	< .0001
Yes	0 (0.0)	750 (41.0)	1,080 (39.3)	831 (38.9)	279 (38.0)
Use of opioids^c – n (%)
No	1,307 (100.0)	1,353 (73.9)	1,998 (72.7)	1,551 (72.6)	540 (73.5)	< .0001
Yes	0 (0.0)	478 (26.1)	752 (27.3)	586 (27.4)	195 (26.5)
Mortality
All-cause mortality – n (%)
No	1,286 (98.4)	1,708 (93.3)	2,551 (92.8)	1,981 (92.7)	688 (93.6)	< .0001
Yes	21 (1.6)	123 (6.7)	199 (7.2)	156 (7.3)	47 (6.4)

^aChi-square tests or Kruskal-Wallis tests with Tukey style multiple comparisons of proportions or Dwass-Steel-Critchlow-Fligner tests for post-hoc pairwise analyses. ^bMeasured in the two years after survey completion. ^cMeasured in the year before survey completion. Proportion of missing data across presented variable ranges between 0% and 3.7%. P-values < 0.05 are highlighted in bold.

Trajectory groups obtained using MQS-4.0 risk Scores

When using the monthly MQS-4.0 scores to represent the risk associated with pain-related medications over the two-year follow-up period, GMM revealed that a linear four-group model best fitted the data (Figure 2): (1) ‘low stable risk’ group (n=2,988; 34.1% of the sample; the mean monthly risk score varied from 0.3±1.5 to 0.5±1.9 in this group); (2) ‘moderate stable risk’ group (n=4,469; 51.0% of the sample; the mean monthly risk score varied from 8.6±6.8 to 9.1±6.9 in this group); (3) ‘increasing to high risk’ group (n=654; 7.5% of the sample; the mean monthly risk increased from 9.7±9.9 to 18.7±10.1); (4) ‘high risk decreasing’ group (n=649; 7.4% of the sample; the mean monthly risk decreased from 18.3±10.8 to 10.8±11.0 over the two-year period).

Figure 2 Trajectory groups obtained using MQS-4.0 risk scores Black curve: ‘low stable risk’ group; Red curve: ‘moderate stable risk’ group; Green curve: ‘increasing to high risk’ group; Blue curve: ‘high risk decreasing’ group; Plain line: observed mean of risk associated with medication use. Dotted line: estimated mean of risk associated with medication use by the GMM. GMM: Growth Mixture Modelling. CCHS: Canadian Community Health Survey.

As shown in Table 3, the trajectory classification based on the MQS-4.0 risk scores allowed for the differentiation of sociodemographic characteristics, pain characteristics, health profiles, and healthcare utilisation. The participants of the ‘increasing to high risk’ group showed the highest proportion of the following characteristics: no secondary school diploma, residing in a remote region, number of different prescribers, general practitioner visits, emergency department visits, and episodes of hospitalisation. They also showed the highest prevalence of benzodiazepines and opioids use, and this group showed the highest prevalence of participants who died after the trajectory modelling period.

Trajectory groups obtained using medication count

When using the monthly number of pain-related medications over the two-year follow-up period, GMM revealed that a five-group model best fitted the data (Figure 3): (1) ‘zero pain-related medication’ group (n=1,307; 14.9%; the mean number of pain-related medications used was 0.0±0.0); (2) ‘occasional use of at least one pain-related medication’ group (n=1,831; 20.9%; the mean number varied from 0.1±0.4 to 0.2±0.5); (3) ‘low stable number of pain-related medications’ group (n=2,750; 31.4%; 0.9±0.7 to 1.0±0.7); (4) ‘moderate stable number of pain-related medications’ group (n=2,137; 24.4%; 2.2 ± 1.0 to 2.4±1.0); (5) ‘high number of pain-related medications’ group (n=735; 8.4%; 4.5±1.5 to±1.4).

Figure 3 Trajectories of pain-related medication use in the whole study sample Black curve: ‘zero pain-related medication’ group; Turquoise curve: ‘occasional use of at least one pain-related medication’ group; Red curve: ‘low stable number of pain-related medications’ group; Green curve: ‘moderate stable number of pain-related medications’ group; Blue curve: ‘high number of pain-related medications’ group; Plain line: observed mean number of pain-related medications used; Dotted line: estimated mean number of pain-related medications used by the GMM. GMM: Growth Mixture Modelling. CCHS: Canadian Community Health Survey.

As shown in Table 4, trajectory classification based on the number of pain-related medications appeared less effective in differentiating sociodemographic characteristics, pain characteristics, health profiles, and healthcare utilisation. In fact, there were no differences between the five groups for most participant characteristics. Only the ‘zero pain-related medication’ group showed differences compared to all other trajectory groups.

High risk/high-count groups identified by the two approaches

We analysed the overlaps between high risk/high-count individuals identified by the MQS-4.0 (’increasing to high risk’ and ’high risk decreasing’ groups) and those identified by medication count (’moderate stable’ and ’high number of pain-related medications’ groups) using a Venn diagram (Figure 4). While some overlap was observed, most trajectory groups did not match, suggesting that a higher medication count does not always indicate higher risk. Some individuals with fewer medications were at high risk. For instance, 872 (432 + 440) participants were exclusively in the high risk groups without being part of the higher medication count groups. These participants primarily used aspirin, pregabalin, oxycodone, clonazepam, and fentanyl. Conversely, 2,441 (1,813 +628) participants were exclusively in the moderate or high medication count groups without being in the high risk groups, with common medications including acetaminophen, amitriptyline, celecoxib, corticosteroid cream, and trazodone. Also, among individuals classified in high risk/low-count trajectories, the mean monthly number of medications ranged from 0.6 ± 0.7 to 0.7 ± 0.8, suggesting that elevated medication-related risk may occur even with relatively few medications. In contrast, among individuals classified in low-risk/high-count trajectories, the mean monthly number of medications ranged from 4.4 ± 1.4 to ± 1.4.

Figure 4 Overlap between the high risk/high-count individuals identified through the MQS-4.0 risk score trajectory analysis versus the trajectory analysis based on the number of pain-related medications Discussion

This study aimed to examine how a risk-based measure and a count-based measure may differently classify medication use trajectories in the real-world. We showed that both the MQS-4.0 risk score and medication count effectively created statistically robust (based on entropy and Bayesian information criterion) and clinically interpretable trajectory groups. However, MQS-4.0 was more discriminative in identifying differences in sociodemographic, pain, health, and healthcare characteristics. Our hypothesis was confirmed: higher medication count does not always equate to higher risk, and some individuals using fewer medications may be at higher risk. These findings suggest that composite risk scores may offer complementary value for subgroup analyses and personalised approaches in chronic pain research, depending on the intended research or clinical objective.

In our study, a higher number of pain-related medications did not automatically reflect increased risk levels. This supports the hypothesis underlying the justification for the use of the MQS-4.0 for CP populations [22]. Furthermore, this could be explained by the fact that certain medication classes are riskier than others [22]. For example, in a context where opioid prescriptions are highly debated for CP [87, 88], rational polypharmacy potentially offers an alternative strategy by integrating non-opioid medications (meaning sometimes a higher number of lower risk medications). This dissociation underscores the limitation of relying solely on medication count as a proxy for the risk associated with medication use; medication count alone may lead to the under-identification of individuals vulnerable to adverse effects. A parallel can be drawn here with studies outside the field of pain which have demonstrated the value of composite medication scores [89, 90]. However, although the MQS-4.0 has demonstrated validity across multiple versions and its various iterations have been widely used in CP research, its weighting system is based on healthcare professionals’ perceptions of medication-related risk rather than solely on pharmacological potency, which may result in similar detriment weights for medications with different clinical risk profiles. In addition, the MQS-4.0 does not explicitly account for dose quantity beyond medication subclasses. The limited overlap between medication count and medication-related risk classifications may suggest that these approaches capture distinct but complementary dimensions of medication use patterns, although we cannot exclude the possibility that some of the observed discordance may reflect methodological artefacts.

In terms of health inequalities, the trajectory groups obtained through MQS-4.0 risk score analysis are more discriminative than those obtained using pain-related medication count trajectory analysis. While differences in modelling approaches may contribute to some variation in the identified trajectories, the extent to which the observed differences are attributable to the underlying constructs, their distributions, or the modelling approaches themselves remains uncertain and distributions of the underlying measures. Nevertheless, MQS-4.0 risk score trajectory groups showed clearer differences in sociodemographic profiles, pain characteristics, general and mental health, healthcare utilisation, and even all-cause mortality. To our knowledge, no other study has compared, in this way, a medication risk score to the number of medications used.

Our study reinforces existing evidence supporting the value of risk-based approaches over count-based approaches in the study of medication use for CP. Our results support the use of such a tool to identify high risk subgroups, even among individuals with fewer medications, to achieve greater granularity in understanding medication-related risks. This is especially critical in CP research, where polypharmacy [8] and high risk medications like benzodiazepines and opioids are prevalent [91]. Allowing risk trajectory differentiation, the MQS-4.0 enables relevant exploration of medication use patterns that can present a hazard and their impact on patient outcomes. Furthermore, the observed differences in risk profiles highlight opportunities for researchers to investigate tailored interventions that could improve safety among individuals living with CP.

Strengths and limitations

This study used data from the TorSaDE Cohort, a unique probability-based longitudinal database with very few missing data. Its large, diverse community sample enhances external validity compared to pain research focused on tertiary care settings. The survey data also provides key covariates for integrating health administrative databases, the most valid way to study healthcare use and medication prescribing [92, 93]. However, limitations of our study include the definition of CP used in this study that differs from the traditional definitions. While some studies have shown this definition yields comparable results in terms of prevalence estimates [33, 35, 36, 46], the possibility of misclassification cannot be excluded. Moreover, the drug insurance claims data used in this study captures only community-acquired prescriptions, excluding over-the-counter and hospital medications. Additionally, Québec’s drug insurance data primarily covers lower socioeconomic groups; this affects the generalizability of the study [94]. Regarding groups profiles, the comparisons across trajectory groups were bivariate and aimed at evaluating the extent to which trajectory classifications derived from each approach captured sociodemographic and clinical differences between individuals. These analyses were not designed to identify independent factors associated with trajectory membership, and therefore no multivariable adjustment was performed. Future studies could build on these findings by using multivariable approaches to further examine factors independently associated with trajectory membership and to better understand modifiable factors underlying different medication risk profiles.

Conclusion

While both the MQS-4.0 risk scores and the medication count approaches generated statistically robust and clinically interpretable trajectory groups, the MQS-4.0 risk score captured more meaningful distinctions in terms of sociodemographic and clinical variables. Our study also highlighted that higher medication use does not necessarily indicate greater risk, as individuals with fewer medications can still present significant risk. Our findings highlight the potential value of combining composite risk scores and medication count measures when studying medication use among individuals living with chronic pain, as these approaches appear to capture different aspects of medication-related burden and may provide complementary value for subgroup analyses and personalised approaches in chronic pain research, depending on the intended research or clinical purpose.

Supplementary Files

Supplementary Appendices

Acknowledgements

We thank Emily-Jayn Rubec, who provided professional linguistic revision services when revising our paper. We would also like to thank Oumar Mallé Samb, Gabriel Gingras-Lacroix, and Joséanne Desrosiers, close collaborators on the qualitative phase of the larger project within which the present analysis is situated. We would also like to thank the members of the TorSaDE Cohort Working Group for sharing their methodologies and statistical expertise, which facilitated this project. The results reported herein correspond to the specific aims of grant number PJT 178216 awarded to principal investigator AL by the Canadian Institutes of Health Research (CIHR). HLNN, first author, is the recipient of Fonds de recherche du Québec – secteur Santé (FRQS) and Canadian Institutes of Health Research (CIHR) doctoral scholarships. The funders had no role in the design of the study, data collection, analysis, interpretation of data, or writing of the manuscript.

Ethics statement

Canadian Community Health Survey participants gave informed consent to Statistics Canada for the linkage of their responses to provincial health administrative databases and use for research purposes when they take part in the survey. De-identified data from the TorSaDE Cohort was accessed via the secure server of the Institut de la statistique du Québec (ISQ), which acts as the mandated data access authority. Ethical approval was obtained from the Commission d’accès à l’information du Québec (#1013990), Institut de la statistique du Québec (ISQ), and relevant university Research Ethics Boards (Université du Québec en Abitibi-Témiscamingue: #2018-02-Lacasse, A.; Centre hospitalier universitaire de Sherbrooke (CHUS): #2017-1504).

Data availability statement

Our dataset, the TorSaDE Cohort data that links Statistics Canada’s Canadian Community Health Survey (CCHS) data and Québec Health Ministry data, is not publicly available. Access must be granted by the Institut de la statistique du Québec (ISQ). Programming codes can be obtained directly from the corresponding author.

Authors contributions

AL, MGP, LG, and CH conceptualised the project and secured funding in collaboration with two patient partners (NM and SB). HLNN conducted the statistical analysis and drafted the manuscript in the context of her doctoral training supervised by AL. All authors critically revised the manuscript, gave final approval of the version to be published, and agreed to act as guarantors of the work.

AI in the writing process

During the preparation of this work the authors used ChatGPT version 4.o in order to assist with the linguistic revision as authors are French speakers. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

References 1

Moore, R.A., et al., The costs and consequences of adequately managed chronic non-cancer pain and chronic neuropathic pain. Pain Pract, 2014. 14(1): p. 79-94. 10.1111/papr.12050

Rometsch, C., et al., Chronic pain in European adult populations: a systematic review of prevalence and associated clinical features. PAIN, 2025. 166(4). 10.1097/j.pain.0000000000003406

Treede, R.-D.,

et al., A classification of chronic pain for ICD-11. Pain, 2015. 156(6): p. 1003-1007. 10.1097/j.pain.0000000000000160

Campbell, F.,

Hudspith, and

Choinière, La douleur chronique au Canada: jeter les bases d’un programme d’action. Rapport du groupe de travail canadien sur la douleur, 2019.

Sirois, C., et al., Polypharmacy definitions for multimorbid older adults need stronger foundations to guide research, clinical practice and public health. Pharmacy, 2019. 7(3): p. 126. 10.3390/pharmacy7030126

Ferrari, A., et al., Polypharmacy among headache patients: a cross-sectional study. CNS drugs, 2018. 32(6): p. 567-578. 10.1007/s40263-018-0522-8

Giummarra, M.J., et al., Polypharmacy and chronic pain: harm exposure is not all about the opioids. Pain Medicine, 2015. 16(3): p. 472-479. 10.1111/pme.12586

Zahlan, G., et al., Polypharmacy and Excessive Polypharmacy Among Persons Living with Chronic Pain: A Cross-Sectional Study on the Prevalence and Associated Factors. Journal of Pain Research, 2023: p. 3085-3100. 10.2147/JPR.S411451

Chang, T.I., et al., Polypharmacy, hospitalization, and mortality risk: A nationwide cohort study. Scientific reports, 2020. 10(1): p. 1-9. 10.1038/s41598-020-75888-8

Shi, S.,

Mörike, and

Klotz, The clinical implications of ageing for rational drug therapy. Eur J Clin Pharmacol, 2008. 64(2): p. 183-99. 10.1007/s00228-007-0422-1

Nunnari, P., et al., Prescribing cascades and medications most frequently involved in pain therapy: a review. Eur Rev Med Pharmacol Sci, 2021. 25(2): p. 1034-1041. 10.26355/eurrev_202101_24673

Palapinyo, S.,

Methaneethorn, and

Leelakanok, Association between polypharmacy and depression: a systematic review and meta-analysis. Journal of Pharmacy Practice and Research, 2021. 51(4): p. 280-299. 10.1002/jppr.1749

Schenker, Y., et al., Associations Between Polypharmacy, Symptom Burden, and Quality of Life in Patients with Advanced, Life-Limiting Illness. J Gen Intern Med, 2019. 34(4): p. 559-566. 10.1007/s11606-019-04837-7

Leelakanok, N., et al., Association between polypharmacy and death: A systematic review and meta-analysis. J Am Pharm Assoc (2003), 2017. 57(6): p. 729-738.e10. 10.1016/j.japh.2017.06.002

Gosselin, M., et al., Classifying Polypharmacy According to Pharmacotherapeutic and Clinical Risks in Older Adults: A Latent Class Analysis in Quebec, Canada. Drugs Aging, 2023. 40(6): p. 573-583. 10.1007/s40266-023-01028-2

Chou, R., et al., Opioid treatments for chronic pain. 2020.

McDonagh, M.S., et al., Nonopioid pharmacologic treatments for chronic pain. 2020.

Nguena Nguefack, H.L., et al., Trajectory modelling techniques useful to epidemiological research: a comparative narrative review of approaches. Clinical epidemiology, 2020: p. 1205-1222. 10.2147/CLEP.S265287

Harden, R.N., et al., Medication Quantification Scale Version III: update in medication classes and revised detriment weights by survey of American Pain Society Physicians. The Journal of Pain, 2005. 6(6): p. 364-371. 10.1016/j.jpain.2005.01.350

Kee, W., S. Steedman, and

Middaugh, Medication quantification scale (MQS): update of detriment weights and medication additions. Am J Pain Management, 1998. 8: p. 83-8.

Masters Steedman, S., et al., Chronic-pain medications: equivalence levels and method of quantifying usage. Clin J Pain, 1992. 8(3): p. 204-14.

De Clifford-Faugère, G., et al., The Medication Quantification Scale 4.0: An Updated Index Based on Prescribers’ Perceptions of the Risk Associated With Chronic Pain Medications. The Journal of Pain, 2023. 10.1016/j.jpain.2023.09.010

Karla

. Stormo, K.J., et al., Medication quantification scores and evaluation of patient pharmacology profiles. Current Review of Pain, 1998. 2(3): p. 171–174. 10.1007/s11916-998-0015-4

Gallizzi, M., et al., Medication Quantification Scale Version III: internal validation of detriment weights using a chronic pain population. Pain Pract, 2008. 8(1): p. 1-4. 10.1111/j.1533-2500.2007.00163.x

Gallizzi, M.A., et al., Use of a medication quantification scale for comparison of pain medication usage in patients with complex regional pain syndrome (CRPS). Pain Med, 2015. 16(3): p. 494-500. 10.1111/pme.12549

Vanasse, A., et al., Cohort profile: the care trajectories—enriched data (TorSaDE) cohort. International journal of epidemiology, 2021. 50(4): p. 1066-1066h. 10.1093/ije/dyaa167

Statistics Canada. Canadian Community Health Survey - Annual Component (CCHS) - Detailed information for 2012. 2012 [cited 2017 April, 15th]; Available from: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&Id=135927.

Régie de l’assurance maladie du Québec. La RAMQ en quelques chiffres. 2025 [cited 2025 January 30]; Available from: https://www.ramq.gouv.qc.ca/fr/donnees-statistiques/ramq-quelques-chiffres.

Lacasse, A., et al., Chronic pain patients’ willingness to share personal identifiers on the web for the linkage of medico-administrative claims and patient-reported data: the chronic pain treatment cohort. Pharmacoepidemiology and drug safety, 2021. 30(8): p. 1012-1026. 10.1002/pds.5255

Nguena Nguefack, H.L., et al., Distinct care trajectories among persons living with arthritic conditions: A two-year state sequence analysis. Frontiers in Pain Research, 2022. 3. 10.3389/fpain.2022.1014793

Tamblyn, R., et al., The use of prescription claims databases in pharmacoepidemiological research: the accuracy and comprehensiveness of the prescription claims database in Quebec. Journal of clinical epidemiology, 1995. 48(8): p. 999-1009. 10.1016/0895-4356(94)00234-h

Andrew, R., et al., The costs and consequences of adequately managed chronic non-cancer pain and chronic neuropathic pain. Pain Practice, 2014. 14(1): p. 79-94. 10.1111/papr.12050

Reitsma, M., et al., The epidemiology of chronic pain in Canadian men and women between 1994 and 2007: longitudinal results of the National Population Health Survey. Pain Res Manag, 2012. 17(3): p. 166-72. 10.1155/2012/875924

Ramage-Morin, P.L. and

Gilmour, Chronic pain at ages 12 to 44. Health Rep, 2010. 21(4): p. 53-61.

Reitsma, M.L., et al., The prevalence of chronic pain and pain-related interference in the Canadian population from 1994 to 2008. Chronic Dis Inj Can, 2011. 31(4): p. 157-64.

Gilmour, H., Chronic pain, activity restriction and flourishing mental health. Health Rep, 2015. 26(1): p. 15-22.

Ramage-Morin, P.L., Chronic pain in Canadian seniors. Health Rep, 2008. 19(1): p. 37-52.

Hogan, M.E., et al., Health utilities in people with chronic pain using a population-level survey and linked health care administrative data. Pain, 2017. 158(3): p. 408-416. 10.1097/j.pain.0000000000000776

Hogan, M.E., et al., Incremental health care costs for chronic pain in Ontario, Canada: a population-based matched cohort study of adolescents and adults using administrative data. Pain, 2016. 157(8): p. 1626-33. 10.1097/j.pain.0000000000000561

Csupak, B., et al., A population-based examination of the co-occurrence and functional correlates of chronic pain and generalized anxiety disorder. J Anxiety Disord, 2018. 56: p. 74-80. 10.1016/j.janxdis.2018.04.005

Grocott, B.,

J.L.

Sommer, and

El-Gabalawy, Usual presence and intensity of pain are differentially associated with suicidality across chronic pain conditions: A population-based study. J Psychosom Res, 2021. 148: p. 110557. 10.1016/j.jpsychores.2021.110557

Liu, L.J.W., et al., A Population Study of Prescribed Opioid-based Pain Reliever Use among Individuals with Mood and Anxiety Disorders in Canada. Drug Alcohol Depend, 2020. 216: p. 108229. 10.1016/j.drugalcdep.2020.108229

Men, F., et al., Food insecurity, chronic pain, and use of prescription opioids. SSM Popul Health, 2021. 14: p. 100768. 10.1016/j.ssmph.2021.100768

Qadeer, R.A., L. Shanahan, and M.A. Ferro, Chronic disruptive pain in emerging adults with and without chronic health conditions and the moderating role of psychiatric disorders: Evidence from a population-based cross-sectional survey in Canada. Scand J Pain, 2017. 17: p. 30-36. 10.1016/j.sjpain.2017.07.009

Roy, R., et al., Understanding correlates of suicidality among those with usual pain and discomfort: A Canadian nationally representative study. J Psychosom Res, 2021. 151: p. 110651. 10.1016/j.jpsychores.2021.110651

Shupler, M.S., et al., Pan-Canadian Estimates of Chronic Pain Prevalence From 2000 to 2014: A Repeated Cross-Sectional Survey Analysis. J Pain, 2019. 20(5): p. 557-565. 10.1016/j.jpain.2018.10.010

Sommer, J.L., C. Blaney, and

El-Gabalawy, A population-based examination of suicidality in comorbid generalized anxiety disorder and chronic pain. J Affect Disord, 2019. 257: p. 562-567. 10.1016/j.jad.2019.07.016

Su, Y.,

Meng, and

D’Arcy, Does Gender Moderate the Relationship Between Chronic Pain and Substance Use Disorder? Insights From a National Canadian Population Survey. Front Psychiatry, 2022. 13: p. 799655. 10.3389/fpsyt.2022.799655

Farhadi, K., et al., The effectiveness of wet-cupping for nonspecific low back pain in Iran: a randomized controlled trial. Complement Ther Med, 2009. 17(1): p. 9-15. 10.1016/j.ctim.2008.05.003

Guck, T.P., et al., Predictive validity of the pain and impairment relationship scale in a chronic nonmalignant pain population. Arch Phys Med Rehabil, 1999. 80(1): p. 91-5. 10.1016/s0003-9993(99)90313-1

Jacob, E., et al., Quantification of analgesic use in children with sickle cell disease. Clin J Pain, 2007. 23(1): p. 8-14. 10.1097/01.ajp.0000210938.58439.dd

Levesque, A., et al., Use of High-Concentration Capsaicin Patch for the Treatment of Pelvic Pain: Observational Study of 60 Inpatients. Pain Physician, 2017. 20(1): p. e161-e167.

McCormick, Z.L., et al., A Prospective Randomized Comparative Trial of Targeted Steroid Injection Via Epidural Catheter Versus Standard C7-T1 Interlaminar Approach for the Treatment of Unilateral Cervical Radicular Pain. Reg Anesth Pain Med, 2017. 42(1): p. 82-89. 10.1097/AAP.0000000000000521

Nguyen, J.P., et al., Treatment of refractory headache secondary to intracranial endovascular procedure by transcutaneous electrical nerve stimulation of the occipital nerve. Neurophysiol Clin, 2018. 10.1016/j.neucli.2018.05.041

Goudman, L., et al., Cross-Country Differences in Pain Medication Before and After Spinal Cord Stimulation: A Pooled Analysis of Individual Patient Data From Two Prospective Studies in the United Kingdom and Belgium. Neuromodulation, 2023. 26(1): p. 215-223. 10.1111/ner.13524

Kwon, H.J., et al., Comparison of Adjuvant Hypertonic Saline and Normal Saline for Epidural Block in Patients with Postherpetic Neuralgia: A Double-Blind, Randomized Trial. Pain Res Manag, 2022. 2022: p. 8081443. 10.1155/2022/8081443

Chae, J.S.,

W.J.

Kim, and

M.J.

Jue, Facet Joint Versus Transforaminal Epidural Steroid Injections in Patients With Cervical Radicular Pain due to Foraminal Stenosis: A Retrospective Comparative Study. J Korean Med Sci, 2022. 37(25): p. e208. 10.3346/jkms.2022.37.e208

Sim, J.H., et al., Effectiveness of Epidural Balloon Neuroplasty in Patients With Chronic Spinal Stenosis Accompanied by Redundant Nerve Roots: A Longitudinal Cohort Study. Pain Physician, 2022. 25(6): p. e841-e850.

Bian, J., et al., A Retrospective Comparison of Low-Temperature Plasma Ablation and Radiofrequency Thermocoagulation of the Thoracic Nerve Root for Refractory Postherpetic Neuralgia. Pain Physician, 2024. 27(4): p. 243-251.

Cheney, C.W., et al., Cooled radiofrequency ablation for the treatment of refractory postamputation neuroma-associated pain: A prospective pilot study. Pm r, 2026. 18(1): p. 15-23. 10.1002/pmrj.13412

De Clifford-Faugère, G., et al., Physicians’ and patients’ perceived risks of chronic pain medication and co-medications in Quebec, Canada: a cross-sectional study. BMC Prim Care, 2025. 26(1): p. 8. 10.1186/s12875-025-02704-5

Elango, Y., et al., Comparison of the analgesic efficacy of intra-articular steroid injections and its combination with suprascapular nerve block for adhesive capsulitis of the shoulder joint: a randomized clinical trial. Reg Anesth Pain Med, 2026. 51(4): p. 404-410. 10.1136/rapm-2024-105902

Kim, C.S., et al., Effects of Pulsed Radiofrequency Duration in Patients With Chronic Lumbosacral Radicular Pain: A Randomized Double-Blind Study. Neuromodulation, 2025. 28(2): p. 256-262. 10.1016/j.neurom.2024.03.006

Kumar, A., et al., Ultrasound-guided pulsed radiofrequency ablation of stellate ganglion in upper-extremity phantom limb pain: a case series. Anesth Pain Med (Seoul), 2024. 19(4): p. 349-352. 10.17085/apm.24035

Liao, C., et al., Ultra-Long-Term Real-World Outcomes of Lower Extremity Nerve Decompression for Painful Diabetic Peripheral Neuropathy: A Retrospective Cohort Study. Neurol Ther, 2026. 15(3): p. 1155-1175. 10.1007/s40120-026-00913-3

Lu, D., et al., High Burden and Phenotype-Specific Variability of Pain Polypharmacy in Early Autoimmune Rheumatic Diseases: A 15-Year Real-World Analysis. Res Sq, 2026. 10.21203/rs.3.rs-9152435/v1

Mohanty, A.J., et al., Prophylactic Regenerative Peripheral Nerve Interface Surgery in Pediatric Lower Limb Amputation Patients. Ann Surg, 2025. 282(2): p. 346-351. 10.1097/SLA.0000000000006327

Parks, T., et al., Exploring the impact of Namaste Care for individuals with advanced dementia: a systematic review of costs, effects and benefits. BMJ Open, 2026. 16(5): p. e102899. 10.1136/bmjopen-2025-102899

Ravaillault, S., et al., Occipital Nerve Stimulation for Refractory Occipital Neuralgia: A Multicenter, Randomized, Controlled Trial [StimO Study]. J Clin Med, 2026. 15(5). 10.3390/jcm15051922

Yildiz, G. and

O.T.

Akkaya, A Comparison Between the Efficacy of Trigeminal Ganglion Radiofrequency Thermocoagulation and Ultrasound-Guided Maxillary-Mandibular Nerve Pulsed Radiofrequency in the Treatment of Trigeminal Neuralgia: A Randomized Clinical Trial. Cureus, 2024. 16(6): p. e61565. 10.7759/cureus.61565

Yildiz, G., et al., Comparison of Selective Nerve Root Pulsed Radiofrequency Vs Paramedian Interlaminar Epidural Steroid Injection for the Treatment of Painful Cervical Radiculopathy. Pain Physician, 2024. 27(2): p. E221-e229.

Leckman-Westin, E., et al., Validation of a claims-based antipsychotic polypharmacy measure. Pharmacoepidemiology and drug safety, 2014. 23(6): p. 628-635. 10.1002/pds.3609

Hovstadius, B. and

Petersson, Factors leading to excessive polypharmacy. Clinics in geriatric medicine, 2012. 28(2): p. 159-172. 10.1016/j.cger.2012.01.001

Kingsbury, S.J.,

Yi, and

G.M.

Simpson, Psychopharmacology: rational and irrational polypharmacy. Psychiatric Services, 2001. 52(8): p. 1033-1036. 10.1176/appi.ps.52.8.1033

Jee, S.H. and

M.D.

Cabana, Indices for Continuity of Care: A Systematic Review of the Literature. Medical Care Research and Review, 2006. 63(2): p. 158-188. 10.1177/1077558705285294

Peters, S.A. and

Woodward, A roadmap for sex-and gender-disaggregated health research. BMC medicine, 2023. 21(1): p. 354. 10.1177/1077558705285294

Muthén, B., Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data., in Handbook of quantitative methodology for the social sciences,

Kaplan, Editor. 2004, CA: Sage: Newbury Park. p. 345-368.

van der Nest, G., et al., An overview of mixture modelling for latent evolutions in longitudinal data: Modelling approaches, fit statistics and software. Advances in Life Course Research, 2020. 43: p. 100323. 10.1016/j.alcr.2019.100323

Ram, N. and

K.J.

Grimm, Growth Mixture Modeling: A Method for Identifying Differences in Longitudinal Change Among Unobserved Groups. International Journal of Behavioral Development, 2009. 33(6): p. 565-576. 10.1177/0165025409343765

Hipp, J.R. and

D.J.

Bauer, Local solutions in the estimation of growth mixture models. Psychological methods, 2006. 11(1): p. 36. 10.1037/1082-989X.11.1.36

Elliott, A.C. and

J.S.

Reisch, Implementing a multiple comparison test for proportions in a 2xc crosstabulation in SAS^®. Proceedings of the SAS User’s Group International, 2006. 31: p. 204-231.

Elliott, A.C. and

L.S.

Hynan

, A SAS^® macro implementation of a multiple comparison post hoc test for a Kruskal–Wallis analysis. Computer methods and programs in biomedicine, 2011. 102(1): p. 75-80. 10.1016/j.cmpb.2010.11.002

Ho, S.Y., et al., What can Venn diagrams teach us about doing data science better? International Journal of Data Science and Analytics, 2021. 11(1): p. 1-10. 10.1007/s41060-020-00230-4

Proust-Lima, C., et al., Package ‘lcmm’. CRAN R, 2016.

Kurczewska-Michalak, M., et al., Polypharmacy Management in the Older Adults: A Scoping Review of Available Interventions. Front Pharmacol, 2021. 12: p. 734045. 10.3389/fphar.2021.734045

Keller, M.S., et al., Cumulative Update of a Systematic Overview Evaluating Interventions Addressing Polypharmacy. JAMA Network Open, 2024. 7(1): p. e2350963-e2350963. 10.1001/jamanetworkopen.2023.50963

A. Morrone, L., et al., Opioids resistance in chronic pain management. Current neuropharmacology, 2017. 15(3): p. 444-456. 10.2174/1570159X14666161101092822

Volkow, N.,

Benveniste, and

A.T.

McLellan, Use and misuse of opioids in chronic pain. Annual review of medicine, 2018. 69(1): p. 451-465. 10.1146/annurev-med-011817-044739

Häppölä, P., et al., A data-driven medication score predicts 10-year mortality among aging adults. Scientific Reports, 2020. 10(1): p. 15760. 10.1038/s41598-020-72045-z

Lu, C.Y., et al., Charlson and Rx-Risk comorbidity indices were predictive of mortality in the Australian health care setting. Journal of clinical epidemiology, 2011. 64(2): p. 223-228. 10.1016/j.jclinepi.2010.02.015

Nielsen, S., et al., Benzodiazepine use among chronic pain patients prescribed opioids: associations with pain, physical and mental health, and health service utilization. Pain medicine, 2015. 16(2): p. 356-366. 10.1111/pme.12594

Lix, L., et al., Features of physician services databases in Canada. Chronic Diseases and Injuries in Canada, 2012. 32(4).

Bellows, B.K., et al., Real-world evidence in pain research: a review of data sources. Journal of Pain & Palliative Care Pharmacotherapy, 2014. 28(3): p. 294-304. 10.3109/15360288.2014.941131

Bérard, A. and

Lacasse, Validity of perinatal pharmacoepidemiologic studies using data from the RAMQ administrative database. The Canadian journal of clinical pharmacology= Journal canadien de pharmacologie clinique, 2009. 16(2): p. e360-9.

Abbreviations AHFS: American Hospital Formulary Service BIC: Bayesian Information Criterion CP: Chronic Pain GMM: Growth Mixture Modelling IQR: Interquartile Range MQS: Medication Quantification Scale MQS-4.0: Medication Quantification Scale, version 4.0 SAS: Statistical Analysis System SD: Standard Deviation