CAPER: patient preferences to inform nonsurgical treatment of chronic low back pain: a discrete-choice experiment

Abstract

Objective

We developed and used a discrete-choice measure to study patient preferences with regard to the risks and benefits of nonsurgical treatments when they are making treatment selections for chronic low back pain.

Methods

“CAPER TREATMENT” (Leslie Wilson) was developed with standard choice-based conjoint procedures (discrete-choice methodology that mimics an individual’s decision-making process). After expert input and pilot testing, our final measure had 7 attributes (chance of pain relief, duration of relief, physical activity changes, treatment method, treatment type, treatment time burden, and risks of treatment) with 3–4 levels each. Using Sawtooth software (Sawtooth Software, Inc., Provo, UT, USA), we created a random, full-profile, balanced-overlap experimental design. Respondents (n = 211) were recruited via an emailed online link and completed 14 choice-based conjoint choice pairs; 2 fixed questions; and demographic, clinical, and quality-of-life questions. Analysis was performed with random-parameters multinomial logit with 1000 Halton draws.

Results

Patients cared most about the chance of pain relief, followed closely by improving physical activity, even more than duration of pain relief. There was comparatively less concern about time commitment and risks. Gender and socioeconomic status influenced preferences, especially with relation to strength of expectations for outcomes. Patients experiencing a low level of pain (Pain, Enjoyment, and General Activity Scale [PEG], question 1, numeric rating scale score<4) had a stronger desire for maximally improved physical activity, whereas those in a high level of pain (PEG, question 1, numeric rating scale score>6) preferred both maximum and more limited activity. Highly disabled patients (Oswestry Disability Index score>40) demonstrated distinctly different preferences, placing more weight on achieving pain control and less on improving physical activity.

Conclusions

Individuals with chronic low back pain were willing to trade risks and inconveniences for better pain control and physical activity. Additionally, different preference phenotypes exist, which suggests a need for clinicians to target treatments to particular patients.

Introduction

Low back pain is a major health problem that affects up to 84% of adults at some point in their lives and incurs an annual economic burden of $177 billion in the United States.^1–2 Chronic low back pain (cLBP) has been defined as back pain persisting for at least 3 months and resulting in pain on at least half of the days in the prior 6 months.³ The prevalence of cLBP is approximately 23%, with 11% to 12% of the population experiencing cLBP-related disability.^2–4

The US Food and Drug Administration recognizes chronic pain as a patient preference–sensitive condition.⁵ Patient preferences are part of patient-centered care, which is widely accepted as a key element of high-quality health care.⁶ Shared clinical decision-making increasingly incorporates patient preferences, reflecting a growing recognition that patients have a unique experience with their medical conditions and have different views of risks and benefits. This is particularly true for chronic pain, the clinical management of which relies strongly on patient experience, with one trial showing significant improvements in pain and disability when patients were given their preferred treatment.⁷ As no single treatment has shown superior efficacy or effectiveness, clinical research is moving its focus to personalized treatments.^8–11

Previous studies on patient preferences for low back pain treatments are limited but identified multiple treatment attributes that affect these preferences. A study focusing on cognitive-behavioral treatments reported the importance of treatment delivery method and travel time,¹² while another study concluded that nonsurgical treatments are preferred to surgical treatments, with patients willing to wait for more ideal outcomes and preferred treatments.¹³ A preference study discovered that convenience was particularly important for exercise-based treatments,¹⁴ and another found that treatment modality was most important.¹⁵ Only one of these studies used previously validated discrete-choice protocols to study patient preferences or included treatment combinations.^13,15–21

The present study uses choice-based conjoint (CBC) analysis, a state-of-the-art discrete-choice method of measuring patient preferences, to assess patient choice preferences by using attributes relevant to most types and combinations of nonsurgical treatments for cLBP. Discrete-choice experiments measure tradeoffs, expressed in probability values, that patients are willing to make in a treatment decision, given its risks and benefits. The current “gold standard” for measuring patient preference uses behavioral economics theory and discrete-choice quantitative methods such as CBC from random-utility models.²²

The objective of the present study was to quantify how people with cLBP weigh risks and benefits of nonsurgical treatments when making their treatment decisions. In our CBC, the choice is between treatment scenarios described by varying levels along these treatment attributes (eg, chance and duration of pain relief, risks of injury with treatment). These treatment scenarios do not correspond to one specific treatment but are relevant to personalizing all kinds of available nonsurgical treatments. We first developed a tool, “CAPER TREATMENT” (Leslie Wilson), to measure patient preferences for the potential risks and benefits of nonsurgical treatments for cLBP. We then pilot-tested and updated the measure, and finally we implemented that measure to estimate utility preference scores in adults with cLBP.

Methods

Instrument development and pilot testing

Development of CBC attributes and levels

We developed a new CBC instrument, CAPER TREATMENT, to measure patient preferences for nonsurgical cLBP treatments. Among methods used to elicit health preferences, conjoint analysis has been shown to perform well for validity, internal consistency, reproducibility, theoretical basis, and acceptability to respondents, and CBC tools are increasingly used to understand health care decisions.^23–25

The CBC method is based on careful selection of a specific decision question (in this instance, a treatment choice decision), followed by the selection and definition of attributes, and levels within each attribute, that reflect the defining characteristics across those different treatments. We developed attributes and levels for CAPER TREATMENT on the basis of relevant literature and input from both clinical experts and individuals experiencing cLBP. An initial set of more than 50 attributes was developed after careful PubMed literature review on pain measurement, cLBP measures, outcomes of available treatment options, and preference studies evaluating pain management. We were careful to include risk and benefit attributes relevant to all types of physical therapy, behavioral approaches, and medication-focused treatment types. The attribute list was then modified by combining similar concepts and eliminating others on the basis of expert and patient input, resulting in 10 carefully defined attributes. The levels within each attribute were then selected and defined. This 10-attribute instrument was then used to conduct 8 one-on-one patient interviews to understand which attributes were most and least important to patients with cLBP and to refine the definitions and wording on the basis of their understanding, resulting in the final CBC tool with 7 attributes. Clinical experts and follow-up patient interviews provided additional input, resulting in final improvements. The final attributes include (1) type of treatment support, (2) time commitment, (3) method of treatment, (4) chance of pain relief, (5) duration of pain relief, (6) expected physical activity outcome, and (7) risk of adverse events (Table 1). These 7 final attributes of the CAPER TREATMENT instrument represent the most important concepts in managing cLBP both from the patient perspective and from documented evidence-based treatment outcomes.^21,26

Table 1.

CAPER treatment tool attributes and levels

Attribute	Levels
Type of treatment support (treatment) Concept: support	Self/family/friend support system with little direct contact with a trusted therapy provider Support of online education materials created by a trusted provider Small group treatment support guided by a trusted therapy provider Individual support of a trusted therapy provider at every treatment as needed
Time commitment (risk) Concept: patient burden	Scheduled therapy 2 times per week + daily self-directed therapy Scheduled therapy 4 times per week + daily self-directed therapy Scheduled therapy monthly + daily self-directed therapy
Method of treatment (treatment) Concept: treatment expectation	Mainly expect to relieve pain by living a valued life (holistic view), accepting pain experiences, and changing life habits Mainly expect to medicate symptoms of anxiety without taking strong pain medications Mainly expect to return to your daily physical strength
Chance of pain relief (benefit) Concept: pain intensity	Many (7 out of 10 people) report a moderate relief in pain intensity A few (4 out of 10 people) report a moderate relief in pain intensity Not very many (1 out of 10 people) report a moderate relief in pain intensity
Duration of pain relief (benefit) Concept: treatment response duration	1-month duration of pain relief 3-month duration of pain relief 6-month duration of pain relief
Physical activity outcome (benefit) Concept: pain interference	Limited to inside walking to do daily self-care activities given your pain intensity Limited to walking outside the house short distances given your pain intensity Limited to bike or walk of 1 mile given your pain intensity
Risk of adverse events (risk) Concept: adverse events	Rare risk of short-term serious injury, but treatable side effect with a bit more pain Some risk of short-term, mild, treatable side effect Some risk of mild, continuous bothersome long-term side effects Rare risk of permanent serious injury side effect (nerve damage)

Not all attribute levels were descriptive of all treatment types, but all treatments had some attributes and levels that were descriptive of that treatment. For example, the “method of treatment” attribute level for “medicate symptoms of anxiety without taking strong pain medications,” the “type of treatment support” attribute level for “self/family support system,” and the “risk of treatment-related adverse events” attribute level for “some risk of mild, continuous bothersome long-term side effects” are all related to the pharmacological approach to treatment. However, all levels of the “time commitment” and “physical activity outcome” attributes and the “method of treatment” attribute level for “mainly expect to return to your daily physical strength” are more specific to a physical therapy approach to treatment. Finally, behavioral treatment–specific attribute levels are to “mainly expect to relieve pain by living a valued life” and to have “individual support of a trusted provider.” One value of CAPER is that we can assess preferences for a mixture of attributes and levels that pertain across all treatment types and mixtures of treatments.

There were 3 types of attributes: (1) benefit attributes, (2) treatment experience attributes, and (3) risk attributes. The 3 benefit attributes correspond to the main outcomes expected across nonsurgical cLBP treatments, including (1) likelihood of relief of pain intensity, (2) duration of pain relief, and (3) an acceptable treatment outcome for physical activity limitation despite the pain. Treatment experience attributes were important characteristics that might differ across personalized treatment types, including (1) main expected method of the treatment expected to work best for you and (2) type of trusted support with treatment, defined as involvement and commitment of different groups of people that you trust. The 2 risk attributes were (1) patient burden from the time and effort required for treatment and (2) risk of treatment/therapy-related adverse events (Table 1, Figure 1, Supplementary Material).

Figure 1.

CAPER Treatment Survey—example choice task.

Design of CBC

After final attributes and levels had been selected, the CBC implementation was designed and validated with the Sawtooth Software and Discover Web Application (Sawtooth Software, Inc., Provo, UT, USA) and followed conjoint analysis good-practice guidelines.²⁷ We selected a random, full-profile design, which uses random sampling with replacement for selecting which attribute levels are shown for each choice-pair task, and allowed the same level of an attribute to appear within different tasks.²⁸ Full-profile designs always include all attributes in each treatment choice task. Sample size estimates were based on the number of attributes, levels, and choice tasks. With a target sample size of 120 patients, we chose the best relative design efficiency score (D-efficiency), resulting in 14 random choice pairs, given our 7 attributes with 3 or 4 levels per attribute.

Application of CAPER TREATMENT instrument

Study sample

Adult (≥18 years of age) English-speaking individuals with a self-reported or clinically documented diagnosis of cLBP were recruited between April 19, 2021, and June 13, 2021. cLBP was defined by the US National Institutes of Health (NIH) Task Force on Research Standards for cLBP,³ with the definition requiring back pain for more than 3 months on at least half the days in the prior 6 months. Before enrollment, individuals might have experienced any surgical or nonsurgical treatment type. All patients provided signed consent or implied consent by taking the survey and received $20 after completion. The study was approved by the University of California San Francisco institutional review board.

Patients were recruited at 4 academic medical centers and online through 5 strategies: (1) Patients with cLBP were identified from electronic medical records and contacted directly by the lead researcher (University of California San Francisco); (2) patients were contacted through the use of an existing health volunteer registry or research program (University of Michigan and University of Rochester); (3) subjects were recruited from an existing study (Back Pain Consortium [BACPAC]) that phenotypes a large cohort with cLBP (University of Pittsburgh); (4) flyers were posted at pain clinics (all 4 sites); and (5) advertisements were placed on carefully selected online platforms. Identified patients were emailed eligibility criteria, the study description, a website link, and the consent form. Those interested were emailed a specific study number and link to take the online CAPER TREATMENT survey. Online platforms included the US Pain Foundation social media postings and ResearchMatch, a national health volunteer registry supported by the US NIH. First, the ResearchMatch system identified and contacted volunteer patients with cLBP, and then if patients approved, we emailed study information and links.

Administration of CBC survey

Before completing the online treatment survey, participants completed a CBC training exercise that mimicked making an ice-cream choice. Participants were presented with 2 pairs of ice-cream choice scenarios with 4 ice-cream choice attributes (flavor, how it is served, number of scoops, and additions/toppings), each with 2 or 3 levels.

In the actual CAPER CBC treatment survey, participants expressed their treatment preferences for 14 different pairs of treatment scenarios by choosing between 2 hypothetical profiles consisting of randomly selected levels within each set of the 7 attributes described previously (Figure 1). All participants were emailed detailed attribute and level definitions, which also appeared within the CBC exercises when participants hovered the mouse cursor over the attribute and level text (Supplementary Material).

Measures

In addition to the CBC tool, study participants completed questionnaires on demographics, disease, and previous treatments, as well as several pain, functional ability, and quality-of-life instruments, including (1) the Pain, Enjoyment, and General Activity Scale (PEG-3); (2) the Oswestry Disability Index (ODI); (3) the Pain Catastrophizing Scale (PCS); (4) a modified version of the EuroQoL Visual Analogue Scale (EQ-VAS); (5) the Perceived Stress Scale (PSS-4); (6) the Patient-Reported Outcomes Measurement Information System (PROMIS-29), assessing 8 domains (physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, pain interference, and pain intensity); (7) the PROMIS Emotional Support 4a version 2 scale; and (8) selected questions from the Chronic Pain Acceptance Questionnaire-SF8.

Data analysis

Descriptive statistics were used to characterize demographic and clinical characteristics of respondents and the results of quality-of-life instruments and other patient-reported measurement scores. For CBC analysis, we estimated random-parameter logit models in R using the GMNL package with 1000 Halton draws. The dependent variable was the respondent’s choice of each hypothetical treatment scenario pair, and the independent variables were the attribute levels. We report preference weights (PWs) of each attribute level, which are the beta coefficients. Preference weights are a unitless measure representing the value of importance of one level relative to other levels for each attribute. A higher PW indicates a stronger preference attribute level, all else equal. A positive PW indicates a positive preference, whereas a negative PW indicates a negative preference.

We calculated PWs for the entire sample and for subgroups. Differences in PWs between subgroups were analyzed by demographic characteristics, including gender (male and female), income (less than $49,000, $50,000 to $99,000, and more than $100,000), and education levels (high school / some college and 4 or more years of college). We also divided respondents into low-pain (<4 on PEG-3) versus high-pain (>6 on PEG-3) groups and low/moderate (≤40 ODI) versus severe (>40 ODI) disability groups, as treatment preferences can vary across levels of pain intensity and disability.

Results

Patient characteristics

Emails were sent to 478 potential patient participants, of whom 217 completed the full CAPER TREATMENT CBC survey. Six were removed from analysis after they failed a reliability question that tested their understanding of the tool, which left 211 for analysis. The reliability question presented a choice between two sets of benefit/risk options, with the first option clearly having more beneficial attribute levels than the second option. If respondents chose the option with the worst levels, a lack of understanding of the tool or attention to the tool was assumed, and the respondents were removed.²⁹ Respondents were on average 53.3 years of age, were mostly female (66.8%), and were mostly white (85.3%), with a mixture of education and income levels. The majority were employed full-time or retired (72.5%), with most or all health care costs covered by insurance (72%) (Table 2).

Table 2.

Demographic characteristics of study participants

Characteristic	Number of participants (n=211)
Age, years, mean (range)	53.3 (20–83)
Height, feet and inches, mean (range)	5’5” (4’1”–6’7”)
Weight, lb, mean (range)	182.3 (85–350)
Gender identity, n (%)
Female	141 (66.8)
Male	67 (31.8)
Other	3 (1.4)
Race/ethnicity, n (%)
American Indian or Alaska Native	5 (2.4)
Asian	14 (6.6)
Black or African American	12 (5.7)
Hispanic	5 (2.4)
Native Hawaiian or Pacific Islander	1 (0.5)
White	180 (85.3)
Other	4 (1.9)
Educational level, n (%)
High school diploma or GED	8 (3.8)
Some college	60 (28.4)
Bachelor’s degree	64 (30.3)
Graduate degree	79 (37.4)
Employment, n (%)
Employed full-time	82 (38.9)
Employed part-time	24 (11.3)
Retired	71 (33.6)
Not employed for pay	28 (13.3)
Laid off because of COVID-19	6 (2.8)
Income level, n (%)
Less than $25 000	35 (16.6)
$25 000 to $49 999	50 (23.7)
$50 000 to $99 999	61 (28.9)
$100 000 to $199 999	45 (21.3)
$200 000 or more	20 (9.5)
Health insurance type, n (%)
Private insurance (HMO, PPO, etc.)	119 (56.4)
Medicare	62 (29.4)
Medicaid/Medical	21 (10.0)
VA	7 (3.3)
No insurance / self-pay	2 (1.0)
Marital status, n (%)
Single, never married	54 (25.6)
Married or domestic partnership	121 (57.3)
Widowed	11 (5.2)
Divorced/separated	25 (11.8)

Abbreviations: HMO = health maintenance organization; PPO= preferred provider organization; VA= Veterans Affairs.

Respondents primarily had had back pain for more than 5 years (65.4%) and had tried various therapy options before enrollment in this study, including active physical therapy (83.9%), adjustment/manipulation (68.7%), self-exercise routine (83.4%), and nonsteroidal anti-inflammatory drugs (75.8%). Almost a quarter of respondents had undergone low back surgery (24.2%). The most common comorbidities included depression (41.7%), osteoarthritis (46.9%), and hypertension (34.6%), and most respondents were regularly taking several pain medications, most commonly nonsteroidal anti-inflammatories and antidepressants (41.2%). (Table 3). Of the 211 respondents, 205 completed all quality-of-life questionnaires and other patient-reported measures (Table 4).

Table 3.

Clinical characteristics of study participants

Characteristic	Number of participants (n=211)
Duration of cLBP, n (%)
Less than 3 months	1 (0.5)
3 to 6 months	8 (3.8)
6 months to 1 year	15 (7.1)
1 year to 5 years	49 (23.2)
More than 5 years	138 (65.4)
Frequency of low back pain over prior 6 months, n (%)
Every day or nearly every day in the prior 6 months	125 (59.2)
At least half of the days in the prior 6 months	55 (26.1)
Fewer than half of the days in the prior 6 months	31 (14.7)
History of low back surgery, n (%)
One operation	31 (14.7)
More than 1 operation	20 (9.5)
No operation, but considering it	24 (11.3)
No operation, and not considering it	136 (64.5)
History of treatment use, n (%)	Use history	Use success
Back injection	99 (38.9)	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 / psychological counseling	55 (26.1)	6 (2.8)

Abbreviations: cLBP= chronic low back pain; NSAIDs= nonsteroidal anti-inflammatory drugs; OT= occupational therapy; PT = physical therapy; SNRIs= serotonin–norepinephrine reuptake inhibitors; SSRIs = selective serotonin reuptake inhibitors.

Table 4.

Patient-reported measure scores of study participants (n = 205)

Questionnaire	Median score (range)
Oswestry Disability Index (ODI)	30 (2–70)
Pain Catastrophizing Scale (PCS)	11 (0–46)
Pain, Enjoyment, General Activity Scale (PEG-3)	4.67 (0–10)
Perceived Stress Scale (PSS-4)	5 (0–16)
PROMIS
Physical function	15 (6–20)
Anxiety	7 (4–20)
Depression	6 (4–18)
Fatigue	11 (4–20)
Sleep disturbance	12 (4–20)
Ability to participate in social roles and activities	13 (4–20)
Pain interference	11 (4–20)
Pain intensity	5 (0–10)
Emotional support	16 (4–20)

Patient preferences for treatment attribute levels

Random-parameter logit analysis demonstrated that the strongest preference was for the 2 best chances of pain relief (β = 2.06, P < 0.001, and β = 1.18, P < 0.001). Lower, but still important, was a preference for longest duration of pain relief (β = 0.88, P < 0.001), which was similar to a preference for reaching the highest physical activity level of walking or biking for 1 mile (β = 0.81, P < 0.001). Risk avoidance was less important, showing that patients preferred to avoid “some risk of long-term side-effects” (β = –0.68, P < 0.001) and, interestingly, had a less strong preference to avoid a “rare risk of permanent serious injury” (β = –0.26, P = 0.032). The preferred “treatment method that worked best” was one that focused on improving physical strength (β = 0.48, P < 0.001), with a medication therapy option being nonsignificant. Preference for type of treatment support was low but strongest for the one-on-one individual “support of a trusted therapy provider at every session” (β = 0.22, P = 0.048), with all other treatment support options (group or online sessions) being nonsignificant (Figure 2, Supplementary Material). On a broader scale, assessing preferences by attributes alone using an importance score showed that patients placed highest importance on the “chance of pain relief” attribute, followed by duration of pain relief, and then reaching a physical activity outcome (Figure 3).

Figure 2.

Patient preference utility scores for persons with cLBP (n = 211). Bolded numbers in red represent preference scores that are statistically significant (P values <.05). For more detailed breakdown of utility scores, see Supplementary Material.

Figure 3.

Attribute importance score.

Sub-analysis

Gender-specific analysis revealed that both male (n = 67) and female (n = 141) patients placed high value on chance and duration of pain relief. Males, however, had a stronger aversion to a risk of “mild, continuously bothersome long-term side effects” (β = –1.25, P < 0.001) than did females (β = –0.61, P < 0.001). Females favored individual provider support (β = 0.34, P = 0.009) but were less willing to engage in the highest frequency of therapy attendance of 4 times per week (β = –0.42, P < 0.001). Males had a higher preference for improving physical strength (β = 0.89, P = 0.001) and being able to walk or bike for at least 1 mile (β = 1.31, P < 0.001) than did females (β = 0.44, P < 0.001, and β = 0.73, P < 0.001, respectively) (Table 5).

Table 5.

Patient preference utility scores of persons with cLBP analyzed by gender, annual reported income, and education level

	Female (n=141)	Male (n=67)	Income <$49,000 (n=85)	Income $50,000–$99,000 (n=65)	Income >$100,000 (n=61)	High school / some college (n=68)	Bachelor’s degree or higher (n=143)
Attribute and level	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.	Coef.
Type of treatment support
Self/family support system
Support of online educational materials created by trusted provider	0.02	–0.12	0.14	–0.28	0.01	0.28	–0.10
Small-group treatment support guided by trusted therapy provider	0.13	–0.41	–0.01	0.20	–0.15	0.12	–0.10
Individual support of a trusted therapy provider at every therapy session as needed	0.34**	–0.003	0.30	0.19	0.37	0.37	0.20
Time commitment (time and effort required until maximum effect or you switch to new treatment)
Scheduled therapy 2×/week + daily self-directed
Scheduled monthly + daily self-directed	–0.16	–0.02	–0.16	–0.33	0.01	–0.10	–0.07
Scheduled 4×/week + daily self-directed	–0.42***	–0.32	–0.36*	–0.42	–0.61*	–0.42*	–0.30**
Method of treatment you feel will work best for you
Relieve pain by living valued life accepting pain experiences and changing life habits
Medicate symptoms of anxiety without taking strong pain meds	–0.19	–0.21	0.005	–0.11	–0.50	0.23	–0.35*
Work toward improving your physical strength	0.44***	0.89**	0.33*	0.82**	0.79**	0.58**	0.45***
Chance of pain relief (before next cycle of treatment is needed in 3 months)
Not very many (1 out of 10 people)
A few (4 out of 10 people)	1.16***	1.58***	1.18***	1.47***	1.55***	1.28***	1.20***
Many (7 out of 10 people)	2.13***	2.61***	1.93***	2.53***	3.07***	2.13***	2.06***
Duration of pain relief
1 month
3 months	0.38***	1.03***	0.48***	0.59*	0.69**	0.35*	0.55***
6 months	0.89***	1.33***	1.01***	0.96**	1.22***	1.20***	0.84***
Physical activity outcome (an acceptable treatment outcome)
Limited to inside walking to do daily self-care activities given your pain intensity
Limited to walking outside the house for short distances given your pain intensity	0.41***	0.88***	0.46***	0.81***	0.52*	0.39*	0.59***
Limited to bike or walk of 1 mile given your pain intensity	0.73***	1.31***	0.33*	1.19***	1.75***	0.32	1.05***
Risk of treatment-related adverse events
Rare risk of short-term serious injury, but treatable side effect with a bit more pain
Some risk of short-term, mild, treatable side effect	0.07	–0.05	0.33*	0.25	–0.77**	–0.04	0.09
Some risk of mild, continuous bothersome long-term side effects	–0.61***	–1.25***	–0.72***	–0.46	–1.28***	–0.73***	–0.70***
Rare risk of permanent serious injury side effect	–0.18*	–0.77**	–0.003	–0.47	–1.04**	–0.16	–0.28*

^*P values <.05;

^**P values <.01;

^***P values <.001.

Income-specific analysis revealed that socioeconomic level influences patient preferences with regard to treatment of cLBP. Individuals in the low-income bracket (n = 85) had lower expectations across benefit and risk attribute levels than did individuals in the high-income group (n = 65). Those with low incomes were less willing to accept the higher therapy time commitments, all levels of chance and duration of pain relief, and all levels of adverse events, especially the risk of mild, continuously bothersome long-term side-effects. Both socioeconomic groups least preferred the risk of mild, continuously bothersome long-term side effects, but the low-income group had lower expectations of benefits (β = –0.72 (P < 0.001) than did the high-income group (β = –1.28, P < 0.001).

Individuals grouped by high and low pain scores showed heterogeneity, with the high-pain group (n = 58) caring more about all types of therapy support and caring less about the duration of pain relief than did the low-pain group (n = 54) (Figure 4). Individuals experiencing low pain had a stronger desire for maximum improved physical activity outcomes than did those in high pain (Figure 4). Individuals in both high and low pain strongly preferred a treatment type “which works toward improving physical strength,” but those with low pain showed a stronger preference score. The high-pain group had a much stronger preference for “medicating their symptoms of anxiety without taking strong pain medications,” which was relatively unimportant to those with low pain. Individuals with low/moderate-disability ODI scores cared much less about the duration of pain relief, while those with severe disability cared more about the chance of pain relief and much more about all types of treatment support (Figure 5).

Figure 4.

Patient preference utility scores of persons with cLBP analyzed by pain score. Bolded numbers in red represent preference scores that are statistically significant (P values <.05).

Figure 5.

Patient preference utility scores of persons with cLBP analyzed by disability level. Bolded numbers in red represent preference scores that are statistically significant (P values <.05).

Discussion

In this multicenter study of cLBP patient preferences, we developed CAPER TREATMENT, a broad discrete-choice CBC measure, and we performed an in-depth conjoint analysis that demonstrates patient preferences for attributes that can be used to personalize a variety of evidence-based, nonsurgical treatments for cLBP. Previous studies have either focused on measuring patient preferences for one specific treatment, such as cognitive behavior therapy or exercise programs, or instead focused on many treatment options for low back pain but were not exclusive to patients with cLBP.^{12–14,21,26} A recent Canadian discrete-choice experiment looked at primarily corticosteroid injections along with 5 other specific treatment modalities.¹⁵ Our study took the unique approach to include attributes and levels pertinent to a broad range of evidence-based, nonsurgical treatments for cLBP so that the results could be used to personalize treatments.

Treatment plans for cLBP can involve a combination of evidence-based approaches with different types of time commitment, risks of adverse events, and types and levels of outcomes. There is a broad range of cLBP treatment options with varying degrees of effectiveness, and the right treatment could depend on patient characteristics and patient preferences. The attributes and associated levels within each attribute of the CAPER TREATMENT instrument reflect the differences in each of these risks and benefits across a wide range of the most current evidence-based nonsurgical treatments.^30–33 Our results demonstrate a roadmap for personalizing treatments on the basis of individual preferences.

Given the heterogeneity of the population with cLBP and different risks and benefits of multiple treatment choices, our study revealed a more comprehensive patient perspective on cLBP treatment. The recommendations of the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) for determining the clinical importance of treatment outcomes for chronic pain suggest that the first next-step priority for research is to determine what patients themselves consider important. The present study demonstrates this patient perspective and allows a comparison of tradeoffs among risks and benefits of treatments.

Not surprisingly, people with cLBP showed a strong preference for the highest chance of pain relief. Historically, treatments for cLBP have been judged on their ability to improve pain scores without attention to pain duration.³⁴

We demonstrate that patients’ second-highest preference was for longest duration of pain relief, which had not been shown before and deserves more attention as a treatment goal. More surprising was patients’ strong preference for the highest physical functional outcome (walking 1 mile), rather than more modest physical goals. The similar preference scores demonstrate a willingness to trade duration of pain relief for reaching the highest physical outcome. Willingness to work toward these gains was demonstrated by the preference to trade 4-times-weekly therapy plus a daily self-directed therapy time commitment and potential treatment side effects for these gains. The therapy approach type strongly favored a preference for the maximum individualized provider support and therapies that improve physical strength. As more treatment becomes self-directed, the importance of also including an individualized support mechanism needs consideration. Respondents cared less about attributes more specific to behavioral treatment methods, although this response might potentially be due to less experience with this therapy.

Individual heterogeneity of responses was evident from the sub-analysis results, indicating that it could be possible to target a preferred treatment to a particular individual. Females, those with higher incomes, and persons with high school diplomas or some college education were more averse to the highest level of time commitment to therapy (4 times/week) than were their counterparts. Those with low incomes placed less weight on the chance and duration of pain relief and on physical activity outcomes than did the higher-income group. Similar preference patterns were observed for females in relation to males, but females valued the support of an individual therapy provider slightly more than males did. Individuals’ clinical characteristics also influenced preferences. Those with scores indicating high pain and severe disability placed higher importance on the chance and duration of pain relief than did patients experiencing less pain and low to moderate disability. The latter group placed more importance on physical activity outcomes instead. Further investigation is required to examine whether these treatment preferences might change longitudinally as pain improves or whether patient preferences are fixed.

Differences in preferences by demographic characteristics raise the issue of equity in cLBP treatment. Time and financial constraints imposed on certain patient populations play a role in treatment expectations and decisions. Expectations with regard to treatment risks and benefits were associated with income levels. Lower educational level was associated with lower expectations for physical activity outcomes. Differences in preferences by demographic characteristics might identify equity factors that are important to consider when clinicians work up a treatment plan. Understanding patient preferences and relating them to certain equity barriers could help target the right treatment to the right patient.

Strengths of our study include its focus on all types of nonsurgical treatments and recruitment from a diverse setting across the United States. This study has several limitations, as well. Despite our multicenter recruitment, our participants lacked diversity, being mostly white, female, and highly educated, and might not be representative of the population with cLBP. However, we had a relatively large sample size with enough diversity to make subgroup preference comparisons across several factors of diversity (education and income), which demonstrated important differences in preferences. Because we focused on all possible nonsurgical treatments by presenting treatment scenarios described by varying levels along treatment attributes, we did not specifically study preferences according to any specific available types of treatments. This prevents comment on preferences for a single treatment, but it allows personalizing treatments by using the strengths of preferences for attributes relevant to multiple treatment approaches and combinations of approaches. Despite this, as preference for a treatment approach that “medicates symptoms of anxiety without taking strong pain meds” is low compared with other alternative approaches, this shows a relative preference for other approaches in comparison. The only other discrete-choice experiment specifically examined treatments, with the lowest preference given to corticosteroid injections and psychotherapy, and heterogeneity described across 6 classes. Our results focus on attributes across all treatments more equally, and respondents showed stronger preferences for pain control and physical outcomes.¹⁵

Finally, individuals had a wide experience with treatment types before taking the survey, and this could affect their preference choices in an undefined way. We are also subject to the limitations inherent in the use of conjoint analysis methods, such as respondent fatigue and inability to test every attribute or value level that might have been relevant to mitigate this fatigue. Respondent fatigue rises as the number of attributes increases, with a maximum of 8–10 attributes recommended. We remained within this acceptable range with 8 attributes and only 3–4 levels for each. In limiting the number of attributes, we tried to mitigate loss of relevant choice factors by combining relevant features within higher-level constructs. For example, our “type of treatment support” attribute incorporated concepts of the amount of clinician support within individual and group treatment modalities. Finally, our comparisons between beta-coefficients were done without statistical significance testing. Next steps will include more statistical tests, such as latent class analysis and treatment scenario analysis, to further define treatment preference across groups.

Conclusions

In conclusion, we studied patient treatment preferences to develop a new CAPER instrument for personalizing the treatment of cLBP. We found that persons with cLBP were willing to trade risks, such as adverse events and time inconveniences, for a treatment that offers better control of their cLBP. They reported a strong preference for maximizing their chance for pain relief, maximizing its duration, and for working toward high physical outcomes. Patient preferences have been applied to clinical practice in general in 3 ways: (1) using preference instruments as a shared decision aid to support both the patient and clinician in a treatment discussion; (2) combining outcomes from preference instruments into treatment algorithms that predict treatment choices; and (3) creating software platforms, such as Annalisa, to combine population-based results with ranking exercises taken before a clinic visit, which use multi-criteria decision-making models to predict treatment choices displayed and discussed at the point of treatment.^35,36 To our knowledge, none of these efforts have been tried with cLBP patients.

In the clinic setting, the CAPER tool can provide a structure for treatment discussions by preparing and educating patients about their own preferences. The CAPER instrument will be used in a larger prospective, multicenter, longitudinal NIH-funded study comparing nonsurgical treatments for cLBP. The CAPER preference utility scores will be used as one variable in a multivariable treatment prediction algorithm, together with physical, psychological, biomechanical, and other patient characteristics. To further develop individualized treatments for cLBP, our immediate next step is to apply latent class analysis and Bayesian analysis with treatment prediction analyses to the current preference responses.

Supplementary material

Supplementary material is available at Pain Medicine online.

Funding

This work was supported by the National Institutes of Health HEAL Initiative under award number RM1-NOT-NS-21-066. The content is solely the responsibility of the authors and does not represent the official views of the National Institutes of Health or HEAL initiative.

Conflicts of interest: None declared.