Preferences for Risks and Benefits of Treatment Outcomes for Chronic Low Back Pain: Choice-Based Conjoint Measure Development and Discrete Choice Experiment

Abstract

Introduction:

Understanding patient preferences for chronic low back pain (cLBP) outcomes is essential for targeting available therapeutic options.

Objective:

We developed and tested a choice-based conjoint (CBC) measure to elicit what outcomes cLBP patients want to achieve and avoid.

Design:

We developed a survey based CBC measure to allow patients to make risk/benefit trade-off choices between possible treatment outcomes. After extensive literature, clinician, and patient input, our measure included seven attributes: fatigue, anxiety/depression, difficulty thinking/making decisions, pain intensity, physical abilities, change in pain, and ability to enjoy life despite pain. Each subject responded to 14 pairs of the same 7 attributes consisting of randomly selected levels within each attribute. Random-parameters logit models were used to estimate strength of preferences and latent class analysis was used to identify patient characteristics associated with distinct preference.

Setting:

Online study using the Sawtooth^™ web-based platform.

Participants:

211 individuals with cLBP recruited from online advertising and academic and private clinical sites.

Interventions:

Not applicable.

Results:

The most valued outcome was the highest level of physical activity (β=1.6–1.98;p<0.001), followed by avoiding cognitive difficulties (β=−1.48;p<0.001). Avoidance of severe pain was comparable to avoiding constant fatigue and near-constant depression/anxiety (β=−0.99,−1.02);p<0001). There was an association between preferences and current pain/disability status; patients with higher pain had a stronger preference to avoid severe pain, whereas those with higher disability stronger preferences for achieving physical activity. The latent class analysis identified two distinct groups: (1) more risk-seeking and willing to accept worse outcomes (56%); and (2) more risk-averse with a stronger preference for achieving maximum benefits (44%).

Conclusions:

Our study illuminated the heterogenous cLBP patient preferences for treatment outcomes. Patients stressed the importance of reaching high physical activity and avoiding cognitive declines, even over a desire to avoid pain. More work is needed to understand patient preferences.

1. Introduction

Chronic low back pain (cLBP) is a major problem in the United States, with 84% of adults being affected at some point.¹ Treatment options for cLBP are numerous and varied with uncertain targeted outcomes.² While a meaningful reduction in pain is the typical reason cLBP patients seek a clinical consultation, their goals may be tied to other outcomes within the domains of physical functioning, social and emotional well-being, and quality of life as well.³ The relative importance of these outcomes for cLBP patients can be invaluable information for the treating clinicians, allowing them to target a treatment, or a combination of treatments, to the patients’ preferences through shared decision-making.^{4,5 6, 7}

A few studies have identified some important outcome domains for assessing pain. The Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) consortium identified six core outcome domains to consider when designing chronic pain clinical trials: pain, physical functioning, emotional functioning, participant ratings of improvement, satisfaction with treatment, symptoms, and adverse events, and participant disposition.^7,8 Another study identified five most common goal domains in cLBP patients: physical activity, workplace, coping skills, relationships and sleep/energy.^9–10 Others proposed six main cLBP assessment domains: pain symptoms, function, wellbeing, disability, disability (social role), and satisfaction with care.¹¹ While clinicians can ask directly what health outcomes an individual patient cares about most, patients may feel that all outcomes are important to them; unable to attach relative importance to the outcomes. There is a need to further adapt these population patterns of preferences obtained here into a decision aid to guide individual patient choices.

A state-of-the-art method of measuring how individuals weigh risks and benefits when making choices they prefer is conjoint analysis using discrete choice (DCE). DCE methods use behavioral economics from random utility theory, assuming that individuals intuitively make decisions that maximize their utility or well-being.^{13, 14} Choice-based conjoint analysis (CBC; a type of DCE) measures trade-offs of positive and negative attributes and can yield more impactful findings for making treatment decisions where trade-offs must be made.¹⁵ Conjoint analysis has been increasingly used in health care,^16–18 but only a few studies address patient preferences for health outcomes as opposed to treatments.^{19, 20, 21} While several have identified important cLBP treatment elements, they did not address outcome preferences.^{17, 22, 23} Selection of best cLBP treatment approaches and combination of approaches to improve outcomes for those with cLBP should be informed by understanding their priorities for treatment outcomes.

Our focus on patient preferences for cLBP outcomes using conjoint analysis is unique and can build on current cLBP treatment preference studies. We aimed to: 1) to develop a CBC patient preference tool to measure preferences for outcomes for patients living with cLBP; 2) estimate utility preference scores characterizing trade-offs between outcomes in adults with cLBP; and 3) assess preference heterogeneity by identifying patient characteristics and latent classes that associate with distinct outcome preferences. CBC is a widely used population-based method for determining decision making preferences for health care, is easy for individuals to understand, and actively mimics how decisions are actually made through trade-offs.

2. Materials and Methods

2.1. Choice-based conjoint (CBC) tool development

Choice-based conjoint (CBC) experiments elicit stated preferences between two or more hypothetical alternatives that maximize their overall utility. Individuals were presented with a series of choice tasks and asked to choose between alternatives with a common set of attributes representing outcome risks and benefits. Attributes were defined by levels, i.e., values an attribute can take. Part-worth utilities, defined as the Beta coefficient for each level of each attribute, were measured by regression beta-coefficients of patient’s choices.

We developed attributes and levels for the CBC tool based on relevant literature review and input from both clinical experts in pain treatment and from individuals with cLBP, and following the Professional Society for Health Economics and Outcomes Research (ISPOR) guidelines.²⁴ An initial set of attributes and levels was developed after careful review of research on measurement of pain, chronic pain outcome domains, relevant treatment outcomes and pain patient preference studies. Feedback from clinical experts was then sought and the tool attributes and wording were modified, resulting in 10 attributes. Eight one-on-one patient interviews were then conducted using this interim version of the tool to understand which attributes were more and less important to cLBP patients, and to change or reword attributes and levels based on their input. The resulting seven attributes is within the recommended number to prevent testing fatigue.²⁴ Finally, we involved the clinical experts and patients again for any final improvements in attribute definitions and wording, resulting in the final CBC tool called a community-based ‘patient preference tool (CAPER)^® Outcome.

2.2. Description of CAPER^® Outcome

The final list of attributes and levels is provided in Table 1A. The seven attributes include: 1) fatigue, 2) anxiety or depression, 3) difficulty thinking and making decisions, 4) pain intensity, 5) ability to do physical activities, 6) empowerment to enjoy life, and 7) expected change in amount of pain. While clinicians within our group had reservations about lumping together anxiety and depression, two different clinical entities, the patients did not like to be labeled as depressed and encouraged the combination of anxiety with depression. These attributes encompass essential domains of cLBP and also relate to goals of the wide range of current treatments.^6–8 Each attribute has three or four levels that reflect the potential spectrum of that outcome.

Table 1.

Patient Characteristics: Demographics, Health Level, cLBP Experience, and Quality of Life Assessment

Characteristic	Number of respondents (%)*
Demographic and socio-economic characteristics (N=211)
Age (years), Mean, Range	53.3, 20–83
Sex (female)	141 (66.8%)
Race
American Indian or Alaska Native	5 (2.4%)
Asian	14 (6.6%)
Black or African American	12 (5.7%)
Native Hawaiian or Pacific Islander	1 (0.5%)
White	180 (85.3%)
Hispanic	5 (2.4%)
Other	4 (1.9%)
Education
High School Diploma or GED	8 (3.79%)
Some college	60 (28.4%)
Bachelor’s or Graduate Degree	143 (67.8%)
Employment
Full-time	82 (38.9%)
Part-time	24 (11.3%)
Retired	71 (33.6%)
Living Situation
Live alone	45 (21.3%)
Live with others	162 (76.8%)
Marital status
Single, never married	54 (25.6%)
Married or domestic partnership	121 (57.3%)
Widowed/divorced/separated	36 (17.1%)
Insurance type
Private insurance (HMO, PPO, etc)	119 (56.4%)
Medicare	62 (29.4%)
Medicaid/Medi-Cal	21 (10.0%)
VA	7 (3.3%)
No insurance/self-pay	2 (1.0%)
Insurance coverage
No coverage	3 ( 1.4%)
Some coverage	56 (26.5%)
Most coverage	98 (46.4%)
Full coverage	54 (25.6%)
Difficulty with affording care for back pain
Not at all difficult	111 (52.6%)
Somewhat difficult	83 (39.3%)
Very difficult	17 (8.1%)
Income
Less than $49,000	85 (40.3%)
$50,000 - $99,000	61 (28.9%)
Over $100,000	65 (30.8%)
Off work for 1 month or more due to low back pain	36 (17.1%)
Received or applied for disability pay due to back pain	36 (17.1%)
cLBP experience, treatment and co-morbidities (N=211)
Duration of cLBP
Less than 1 year	24 (11.4%)
1– 5 years	49 (23.2%)
Over 5 years	138 (65.4%)
Frequency LBP over past 6 months
Every day or nearly every day	125 (59.2%)
At least half of the days	55 (26.1%)
Less than half of the days	31 (14.7%)
Had low back surgery	51 (24.2%)
Treatments	Use History	Successful
Back injection	99 (46.9%)	48 (22.7%)
Opioid pain killers	100 (47.4%)	41 (19.4%)
SSRIs/SNRIs	40 (19.0%)	6 (2.8%)
Gabapentin or Pregabalin	101 (47.9%)	26 (12.3%)
Tricyclic antidepressants	23 (10.9%)	4 (1.9%)
NSAIDs	160 (75.8%)	34 (16.1%)
Adjustment/Manipulation	145 (68.7%)	50 (23.7%)
Active PT or OT	177 (83.9%)	57 (27.0%)
Passive PT	95 (45.0%)	18 (8.5%)
Self-exercise routine	176 (83.4%)	64 (30.3%)
Acupuncture	68 (32.2%)	11 (5.2%)
Mental health therapy/psych counsel	55 (26.1%)	6 (2.8%)
Currently taking regularly (at least 1x a week)
NSAIDs/acetaminophen	87 (41.2%)
Muscle relaxant	58 (27.5%)
Nerve pain/anti-convulsant medication	53 (25.1%)
Anti-depressant & related medication	87 (41.2%)
Opioid painkiller	31 (14.7%)
Sleep/Anxiety medication	75 (35.5%)
Topical pain medication (cream, gel, patch)	80 (37.9%)
Cannabis/marijuana	28 (13.2%)
Have a medical condition or treated for it:
Heart disease	27 (12.8%)
High blood pressure	73 (34.6%)
Lung disease	15 (7.1%)
Diabetes	13 (6.2%)
Ulcer or stomach disease	20 (9.5%)
Kidney disease	7 (3.3%)
Liver disease	8 (3.8%)
Anemia or other blood disease	14 (6.6%)
Cancer	16 (7.6%)
Depression	88 (41.7%)
Osteoarthritis, degenerative arthritis	99 (46.9%)
Back pain	195 (92.4%)
Rheumatoid arthritis	10 (4.7%)
Physical and behavioral health (N=211)
Height (feet, inches); mean (range)	5’5 (4’1 – 6’7)
Weight (lbs); mean (range)	182.3 (85 – 350)
Years with low back pain; mean (range)	13.7 (0 – 59)
Smoking
Never smoker	126 (59.7%)
Current smoker	15 (7.1%)
Used to smoke, but have now quit	70 (33.2%)
In the past year, drunk alcohol, smoked, or used drugs more than meant to
Never	136 (64.5%)
Sometimes	45 (21.3%)
Frequently	10 (4.7%)
Not applicable	20 (9.5%)
In the past year, wanted or needed to cut down on drinking, smoking, or drug use
Never	124 (58.8%)
Sometimes	38 (18.0%)
Frequently	5 (2.4%)
Not applicable	44 (20.9%)
Back pain characteristics (N=205)
Back pain spreads:
to buttock or thigh	137 (66.8%)
to legs	100 (48.8%)
Pain, disability and quality of life assessments (N=205)
Scale	Median score (range)
Oswestry Disability Index (ODI)	30 (2 – 70)
Pain, Enjoyment, General Activity Scale (PEG-3)	4.67 (0 – 10)

^*unless otherwise specified

2.3. Administration of CAPER^® Outcome

We used Sawtooth^™ software (v9.5.3) and followed conjoint analysis good-practice guidelines²⁵ to create a random, full-profile, balanced-overlap experimental design. The random CBC design employs random sampling with replacement for selecting which attribute levels are shown in each task (i.e., choice pair). Sampling with replacement allows the same level of an attribute to appear within tasks.²⁶ Full-profile designs allow the two outcome choices in each task to include all seven attributes. Sawtooth^™ software hosting platform also was used for the survey delivery.

In CAPER^® Outcome, participants expressed their preferences for outcomes by choosing between two hypothetical profiles consisting of randomly selected levels within each set of the seven attributes. Our question asked ‘what sets of outcomes cLBP patients are willing to live with on most days over the next year’. The respondents were presented with a series of 14 different pairs of outcome sets and asked to choose the preferred outcomes set in each pair (Figure 1A). Each choice pair had the same 7 attributes defined by a randomly selected level from among 3–4 options describing that attribute.

Figure 1A.

Preference Weights for Living with Outcomes of Chronic Low Back Pain. N=205 Note: The orange values are significant at P<0.05 level.

Subgroup analysis, by high vs low pain intensity High Pain Intensity n=54, Low Pain Intensity n=58; Note: The bolded numbers indicate a P value that is significant at at least 0.05 level

Subgroup analysis, by high vs low disability score Low Disability n=156, Severe Disability n=49; Note: The red numbers indicate a P value that is significant at at least 0.05 level

Low Income n=85; High Income n=65 Note: The bolded numbers indicate a P value that is significant at at least 0.05 level

All participants were sent detailed definitions of the attributes and levels via email; these definitions also appeared within CAPER^® Outcome when participants hovered their computer mouse over the attribute and level names. We also designed one scenario in which one choice had all the best levels and the other choice had all the worst levels in each outcome attribute to test for the validity of responses by eliminating respondents as non-attentive or not understanding if choosing all the worst levels option. Different combinations of choice tasks and sample sizes were pre-tested to generate the most statistically efficient design with a simulated logit standard error below 0.05 and with the highest relative D-efficiency score (487) given our 14 choice pairs and seven attributes with three or four levels each and with the target sample size of 120. Before completing the CAPER^® Outcome CBC survey online, the participants completed an online training CBC exercise that mimicked making a paired choice of ice-cream with four attributes and two to three levels to help participants understand how the tool works.

2.4. Additional Variables Collected

In addition to CAPER^® Outcome, the participants completed the CAPER ^®Treatment CBC survey ²⁷ (results published elsewhere) and questionnaires on demographics (age, gender, race and ethnicity, education, etc.), disease history and previous treatment characteristics, and several pain, functional ability and quality of life instruments. The additional instruments included PEG-3 screening tool of pain, enjoyment, and general activity ²⁸, the Oswestry Disability Index (ODI).^15,16

2.5. Study Sample

Adult patients with a cLBP diagnosis were recruited between April 19 and June 13, 2021 using four recruitment strategies: (1) identifying volunteer patients with a cLBP diagnosis in electronic records and contacting them directly, (2) using existing registries of patient volunteers who have or may have low back pain, (3) posting flyers at pain clinics, and (4) posting about our study on carefully selected online platforms. Recruitment was done at four academic medical center sites as well as at online sources. The United States (U.S.) Pain Foundation posted information about our study on their social media. ResearchMatch, a national health volunteer registry supported by the U.S. National Institutes of Health as part of the Clinical Translational Science Award (CTSA) program was searched for potential cLBP patients. For all recruitment methods, potential study volunteers were asked to contact the researchers. Once we confirmed eligibility and asked for consent, potential study volunteers were given the link to the survey and a unique study number to enter the online survey. Upon completion, patients received $20 for their participation. The study was approved by the University Committee on Human Subjects Research.

2.6. Data Analysis

Descriptive statistics were used to characterize demographic and clinical characteristics of respondents. For the CBC analysis, we estimated random-parameter logit models (RPL) with 100 Halton draws, where the dependent variable was the respondent’s choice of hypothetical outcome scenarios and the independent variables were the levels of the attributes. We report preference weights (PW) which are the beta coefficients showing the preference strengthof each attribute level, which are the estimated. For each attribute, preference weights are on a unit-less scale that measures the value of one attribute level relative to other attribute levels in the study. A higher PW (or beta coefficient) indicates a stronger preference and a positive PW indicates a positive preference, while a negative PW a negative preference.

We calculated PWs for the entire sample and separately for subgroups. Differences in PWs between subgroups were analyzed based on demographic characteristics including gender), income (less than $49k, $50k-99k, and above $100k) and education levels (high school/some college and four or more years of college). We also divided respondents into low-pain (<4 on PEG-3, question 1) versus high-pain (>6 on PEG-3, question 1) groups and low/moderate (≤40 ODI) versus severe (>40 ODI) disability groups, as treatment preferences can vary across levels of pain intensity and disability. The criteria for the cut points from surveys were the definitions from the surveys themselves. For patient characteristics cut points we used the categories asked in the survey questionnaires, sometimes combining categories to ensure an adequate sample size.

These cut-points were identified a priori based off of prior studies looking at cut-points for numerical pain score¹² and ODI.¹³ We also conducted a latent class analysis (LCA), both uncontrolled and controlled for 4 covariate variables that should be readily available at presentation to a clinical visit: gender, income, age and total PEG score. LCA allows identification of latent (unobserved) subgroups of respondents with similar patterns of preferences. All analyses were performed in R using the GMNL package. The R code is available on request from the authors.

3. Results

3.1. Patient characteristics

Emails were sent to 478 potential patients. 211 completed the full CAPER^® Outcome CBC survey, and 6 were removed after failing the fixed question, leaving 206 for analysis. Most patients were female (66.5%), white (85.4%), with a mixture of education and income levels. Average composite PEG score was 4.67. Most had back pain for over 5 years (65.5%) and had tried various therapy options including active physical therapy (83.5%), adjustment/manipulation (68.9%), self-exercise routine (83.5%), and nonsteroidal anti-inflammatory drugs (75.7%) (Table 1).

3.2. Patient preferences for outcome attribute levels

The strongest statistically significant (p<=0.05) outcome preferences were for reaching the three highest levels of physical activity: being able to walk at least 1 mile (β=1.98), walk outside the house for short distances (β=1.77), and walk inside and accomplish one’s basic household routines (β=1.60) (Figure 1A). Attaining physical function was the most important outcome to achieve for those with cLBP. Even just gaining the ability to walk inside and accomplish basic home routines was more important than avoiding moderate or severe pain intensity (β=1.60 as compared to 0.43 and 0.95 respectively). The next strongest preference was to avoid the highest level of difficulty with cognition (much difficulty with thinking, β=−1.48). This aversion to cognitive difficulties was much stronger than their aversion to to avoid the next three negative outcomes which were relatively similar: the highest levels of pain intensity (severe pain, β=−1.02), anxiety/depression (almost always feeling anxious or depressed, β=−1.01) and fatigue (always feeling fatigued, β=−1.00). Also, individuals with cLBP showed a similarly strong preference for being able to expect very much or much improvement in pain in the next 3 months (β=0.95). Preferences for other attributes such as fatigue, anxiety/depression, empowerment to enjoy life despite pain are as shown (Table 1).

3.3. Subgroup analyses

Subgroup analyses explored heterogeneity in preferences for some initial factors we identified that might affect preferences included demographic factors (age, gender), socio-economic factors (race and income), and clinical factors (pain, disability). Heterogeneity was also addressed by latent class analysis, which is a more accurate method of accounting for preference heterogeneity across multiple factors both known and unknown. .

Using the first pain question of the three-question PEG questionnaire, we divided patients into those scoring less than 4 (low pain level) and those scoring greater than 6 (high pain levels) (Figure 1B). A preference for achieving highest levels of physical activity was still the strongest preference for both groups. Those with high pain had only a slightly stronger preference (β =3.47) to avoid being with severe pain intensity than did those in the low pain group (β =3.41). However those in high pain were much more willing to accept the lower physical ability level ‘to walk outside the house’ (β=3.52 vs β=2.81), even more than their preference for the higher activity level ‘to walk or bike at least 1 mile’, than did those with low pain levels. Those in high pain also had almost a twice stronger positive preference to ‘expect to see a very much or much improvement in pain in the next 3 months’ (B β=2.06 vs β=1.4) and also a ‘minimal or no change in pain’ (vs worsening pain) than did those in low pain (β=1.06 vs β=0.53). Additionally, the high pain group showed more than a twice stronger preference to avoid the highest pain intensity compared to those in the low pain group (β=−2.88 vs β=−1.35) and also a much stronger preferences to avoid high fatigue, anxiety or depression and cognitive difficulties. Finally, those in high pain had a stronger preference to ‘be empowered to enjoy life despite the pain’ than did those in the low pain group (β=1.26 vs β=0.8)

A comparison of the preference scores across high (greater than 40) and low (less than 40) disability scores on the Oswestry Disability Index (Figure 1C) showed that those with high disability were more risk averse and sought better outcomes. Those with high disability scores had strongest preferences for acheiving the highest (β=2.49 vs β=2.04)., but also the more moderate (β=2.48 vs β=1.76) physical abilities than those with low disability. They also were more risk-averse, showing clearly stronger negative preference scores for three out of four negative outcomes (pain intensity, anxiety/depression, and fatigue) than those with lower disability scores. However, unlike those with low disability, those with high disability expressed the highest negative preference for avoiding severe pain intensity (almost twice as strong) (β=−1.73 vs β=−0.98). Among those with low disability, the strongest preference was for avoiding cognitive difficulty (β=−1.57), similar to main analyses. Those with high disability had a much stronger preference for ‘expecting a much improved or improved pain in the next 3 months’ (β=1.4 vs β=0.9) and for ‘the empowerment to enjoy life despite the pain’ (β=0.66 vs β=0.3) than those with low disability.

We also showed important difference in preferences by income with those with low annual incomes (<$50,000) expressing lower expectations for high outcomes and being more risk averse than those with high annual incomes (>$100,000) (Figure 1D). For example, although both groups still care most about increasing physical function, those with lower incomes had a weaker preference for all physical activity outcomes (β=1.37, β=1.57, β=1.47 vs β=2.0, β=2.1, β=2.4) and for expected pain improvements in 3 months than did the high high-income respondents (β=0.32, β=0.79 vs β=0.66, β=1.44). Those with low incomes also expressed weaker preferences to avoid anxiety/depression and severe pain intensity. Age and gender subanalysis demonstrated few differences in preferences.

3.4. Latent Class Analysis

The latent class analysis while controlling for 4 coefficients; gender, income, age and total PEG score showed two distinct groups or classes (Figure 2). Those age 30 (59%) vs age 75 and those with low incomes (65%) and with a high total PEG score (60%) were more likely to belong to class I (56% of the participants), which is more risk accepting and willing to accept lower outcome levels compared with class II individuals who were more risk averse and preferred to achieve maximum outcomes (44% of the participants). Those in class II had a four- to six-fold stronger preference for attaining all levels of physical activity (β = 3.23 vs 0.462 for ability to bike or walk at least 1 mile), a two-fold stronger preference for avoiding severe pain (β =−1.055 vs −0.666), and a two-fold stronger preference for avoiding cognitive difficulty (β =−1.579 vs −0.834 for avoiding much difficulty with thinking). Class I had stronger preference to avoid fatigue (β =−.301 vs −0.110 for avoiding fatigue sometimes and β=−0.877 vs −0.638 for avoiding fatigue all the time).

Figure 2.

Preference Weights, by Latent Class, with Controls as coefficients N=205 Note: The bolded numbers indicate a P value that is significant at the 0.05 level or greater

Note: Those age 30 (59%) vs age 75 and those with low incomes (65%) and with a high total PEG score (60%) were more likely to belong to class I (56% of the participants), which is more risk accepting and willing to accept lower outcome levels compared with class II individuals who were more risk averse and preferred to achieve maximum outcomes (44% of the participants). Those in class II had a four- to six-fold stronger preference for attaining all levels of physical activity (β = 3.23 vs 0.462 for ability to bike or walk at least 1 mile), a two-fold stronger preference for avoiding severe pain (β =−1.055 vs −0.666), and a two-fold stronger preference for avoiding cognitive difficulty (β =−1.579 vs −0.834 for avoiding much difficulty with thinking). Class I had stronger preference to avoid fatigue (β =−.301 vs −0.110 for avoiding fatigue sometimes and β=−0.877 vs −0.638 for avoiding fatigue all the time).

4. Discussion

This CBC demonstrates the overwhelming importance of reaching high levels of physical activity for most patients with cLBP, which dominates all other outcomes. In our participants, reaching the ability to walk or bike 1 mile was almost twice as important as avoiding severe pain, or maximum difficulties with fatigue and anxiety/depression, which showed the same importance. Avoiding cognitive difficulties as a result of treatment was consistently the strongest negative preference, a negative preference not focused on in existing literature. Our latent class analysis showed some heterogeneity; identifying two distinct groups of cLBP patients: one group that is risk accepting and willing to accept lower levels of outcomes, caring less about improving pain level at 3 months, avoiding severe pain and avoiding cognitive difficulty versus a second group that is risk averse and had stronger preferences for achieving maximum outcomes. Respondents who were younger, lower-income or have more pain (higher total PEG scores) were more likely to be in the first, risk accepting and lower outcome accepting group.

We uniquely examined cLBP patient preferences for outcomes they are willing to live with, rather than examining treatment preferences, as has been done previously ^{18, 27, 17, 18}. Treatments currently available for cLBP may differ in the outcomes that they target and their most likely adverse events. Whereas it is widely recognized that relevant cLBP outcomes encompass a broader range than pain intensity and physical functioning,^6–9,12, our study illuminates which outcomes cLBP patients prefer when they must make trade-offs among them. These outcome preferences are likely an important factor in treatment choices made between the patient and clinician. We believe a discrete choice based approach like ours is needed to promote a more informed patient-centered care and shared decision-making.

Our findings suggest that patients placed more focus on reaching higher levels of physical activity than even improving pain control. This contrasts prior findings both in cLBP and in other pain conditions. One of the few studies to employ DCE to elicit preferences for attributes of cLBP pharmaceutical treatments also found that most important attribute involved reducing risk of physical dependence (scaled conditional relative importance of 6.99)¹⁹. However, unlike our study, to those patients the most important attribute was improving symptoms control (scaled conditional relative importance of 10.00). Prior preference research in rheumatoid arthritis research also highlights that pain was the most preferred area of improvement across all subgroups of patients²⁰. This might be unique to our particular study population, which included not only patients from clinics but also from online patient registries. We are further validating this CBC tool in other study populations, including ones recruited from mainly clinical settings where patients may have worse pain.

One of the other most valued outcome attributes in our study – cognitive function – has previously not been given much attention. The IMMPACT consortium’s set of six core chronic pain outcome domains, described earlier, does not include cognitive function, which is listed as a supplemental outcome domain⁸. A recent paper that advocates for broadening the range of cLBP outcomes does not mention cognition.⁶ Cognitive outcomes in cLBP treatment warrant closer examination as some of pharmacologic treatments for cLBP may cause sedation. Being in chronic pain may also alter cognitive processes²¹ and some treatments can negatively impact cognitive performance, making it important to consider and discuss cognitive effects with a patient when directing to a best treatment.

Our subgroup analyses revealed that both high pain and high disability patients have stronger preferences for avoiding pain. Studies have shown that quality of life in those with chronic pain is more associated with pain beliefs than pain intensity²². While our analysis here did not focus on it, possibly, these patients with higher pain and disability are different from those with lower pain and disability given increased maladaptive pain beliefs or experience. It is also interesting that both the high pain and high disability patients also were more risk averse – wanting to avoid fatigue, anxiety/depression, cognitive changes. Fear avoidance is a known maladaptive pain belief that can prognosticate cLBP disability²³. Possibly this mindset is also associated with wanting to avoid other risks in life as well. Future studies should explore how pain beliefs and patient preferences correlate.

The latent class analysis identify two groups of patients that behaved similarly within groups and differently from each other. One group that is risk accepting and willing to accept lower levels of outcomes versus a second group that is risk averse and had stronger preferences for achieving maximum outcomes. Respondents who were younger, lower-income or have more pain and dysfunction (higher total PEG scores) were more likely to be in the first latent class anlaysis. It is somewhat unexpected that younger patients are willing to accept lower outomes such as more pain and less function and further examination may be required. It is potentially concerning that lower-income patients were more likely to be in the first group. Studies have shown that low socioeconomic status has been associated with maladaptive pain beliefs, poor coping strategies, and severe distress²⁴. Studies have shown that disadvantaged patients tend to have higher cLBP and perhaps these patients have internalized a tendency to be more accepting of more pain and disability. Finally, it is interesting that in this latent class analysis, those with higher pain and dysfunction are more likely to belong to the group that is risk accepting and willing to accept lower levels of outcomes when our subgroup analyses demonstrated that independently of other factors, patients with just high pain and separately patient with just high dysfunction both had stronger preferences to avoid pain. This suggests that patient preferences are highly complex, not only informed by individual levels of pain or dysfunction but by a combination of known and unknown preference factors captured in the latent class analysis. The analytic models used here allow a rich variety of information about behavior from our data and will encourage further exploration of what informs patient preferences for outcomes.

Our work contributes to the study of cLBP treatments in several ways. First, prior research has focused on cLBP treatment choice and cLBP treatment delivery setting choice,^{17,18,21,22, 27} and our study is unique to examine choices regarding cLBP outcomes that patients are willing to live with. This is particularly important given a lack of clearly superior efficacy of one nonsurgical treatment over another in reducing pain and achieving other outcomes. Second, results from this and further research could serve as the foundation for an algorithm designed to aid informed shared decision-making²⁵, or as one predictor in an algorithm predicting cLBP treatment success. Although there are few examples of the use of patient preference in low back pain to influence treatment recommendations, this has been successful in rheumatoid arthritis (RA), another preference-sensitive condition (one where patient preference impacts treatment choice) with similar chronic characteristics, and elements of uncertainty in treatment choice as cLBP. Hazlewood et al (2016) developed and tested a discrete choice instrument to estimate the average importance of the risks and benefits of treatments for RA²⁶. They then combined outcomes estimates with the patient preference scores to inform treatment recommendations for early RA and added these to the treatment recommendations^{27, 28}. A proof-of-concept study then used CBC choice tasks to relate a patient’s preferences to the actual treatment choices, demonstrating feasibility and predictive accuracy²⁹. Even before such findings can be implemented in a technological tool, our findings can help raise awareness among clinicians about the outcomes that are relatively more and less important to those with cLBP .

Our study has several limitations. First, the outcomes included in our study do not represent the whole range of possible cLBP outcomes. Some participants might prefer outcomes other than the selected set of attributes in our study. However, for a DCE to be feasible, it is recommended that studies do not include more than (7–8) attributes.²⁴ We selected the attributes via a rigorous iterative process, which involved reading clinical literature, consultations with clinicians caring for cLBP patients, and interviews with cLBP patients. Second, cLBP patients are a heterogeneous population and our sample may have excluded groups with other preferences. In order to better understand this heterogeneity we collected rich information on respondents’ experience of pain (Table 1) and conducted subgroup analyses across patient and pain characteristics. Additionally, the latent class analysis allowed us to identify two distinct groups each with their own pattern of outcome preferences. This can help clinicians identify with the patient, which of these outcomes are most important to them. Future research will examine a larger sample and test their preferences longitudinally as they experience different types of treatments for their cLBP.

5. Conclusions

This study uniquely examines patient preferences for cLBP outcomes in a discrete choice experiment. Our findings demonstrated that physical activity and cognitive functions were the two most valued outcomes, more important than avoiding pain. More attention should be given to the lower outcome expectations shown in those with lowest incomes. Discussions of these outcomes and how different cLBP treatments affect these outcomes should take place to enable shared decision-making that is sensitive to patient preferences. Given the two distinct preference groups we showed, clinicians can expect to encounter variation in risk avoidance, and strength of benefits sought across demographics, income, and pain and disability experience. Outcomes should be aligned with patient priorities and preferences in patient-centered care.