Annals of Internal Medicine
Literacy-Adapted Cognitive Behavioral Therapy Versus Education for
Chronic Pain at Low-Income Clinics
A Randomized Controlled Trial
Beverly E. Thorn, PhD; Joshua C. Eyer, PhD; Benjamin P. Van Dyke, MA; Calia A. Torres, MA; John W. Burns, PhD;
Minjung Kim, PhD; Andrea K. Newman, MA; Lisa C. Campbell, PhD; Brian Anderson, PsyD; Phoebe R. Block, MA;
Bentley J. Bobrow, MD; Regina Brooks; Toya T. Burton, DC, MPH; Jennifer S. Cheavens, PhD; Colette M. DeMonte, PsyD;
William D. DeMonte, PsyD; Crystal S. Edwards; Minjeong Jeong, PhD; Mazheruddin M. Mulla, MA, MPH; Terence Penn, BS;
Laura J. Smith, BA; and Deborah H. Tucker, MBA*
Background: Chronic pain is common and challenging to treat.
Although cognitive behavioral therapy (CBT) is efficacious, its
benefit in disadvantaged populations is largely unknown.
Objective: To evaluate the efficacy of literacy-adapted and sim-
plified group CBT versus group pain education (EDU) versus
usual care.
Design: Randomized controlled trial. (ClinicalTrials.gov:
NCT01967342)
Setting: Community health centers serving low-income patients
in Alabama.
Patients: Adults (aged 19 to 71 years) with mixed chronic pain.
Interventions: CBT and EDU delivered in 10 weekly 90-minute
group sessions.
Measurements: Self-reported, postintervention pain intensity
(primary outcome) and physical function and depression (sec-
ondary outcomes).
Results: 290 participants were enrolled (70.7% of whom were
women, 66.9% minority group members, 72.4% at or below the
poverty level, and 35.8% reading below the fifth grade level);
241 (83.1%) participated in posttreatment assessments. Linear
mixed models included all randomly assigned participants.
C hronic pain (that is, pain lasting beyond the usual
course of illness or injury, usually longer than 3 to 6
months) is widespread, challenging to treat, and a
growing problem (1–3). Approximately 20% of physi-
cian visits and 10% of drug sales are attributed to pain
(1, 2). More than 116 million Americans have chronic
pain, costing $600 billion annually (1, 3). Chronic pain
disproportionately affects economically disadvantaged
persons, ethnic minority groups, women, and older
adults (2, 4), who have higher rates of chronic pain,
greater risk for pain-related disability, and higher rates
of major chronic physical and psychological comorbid
conditions (1, 5– 8). Standard approaches to chronic
pain focus on expensive biomedical interventions and
analgesic medications that are fraught with undesirable
side effects, as exemplified by the spiraling opioid ep-
idemic. Recently published national clinical practice
guidelines stress nonpharmacologic, evidence-based
alternatives to pain medications (9 –11). However, many
providers are not familiar with such treatments as cog-
nitive behavioral therapy (CBT), and access to CBT is
limited, particularly in low-income communities.
Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018
ORIGINAL RESEARCH
Members of the CBT and EDU groups had larger decreases in
pain intensity scores between baseline and posttreatment than
participants receiving usual care (estimated differences in
change scores—CBT: ⫺0.80 [95% CI ⫺1.48 to ⫺0.11]; P = 0.022;
EDU: ⫺0.57 [CI, ⫺1.04 to ⫺0.10]; P = 0.018). At 6-month follow-
up, treatment gains were not maintained in the CBT group but
were still present in the EDU group. With regard to physical func-
tion, participants in the CBT and EDU interventions had greater
posttreatment improvement than those receiving usual care, and
this progress was maintained at 6-month follow-up. Changes in
depression (secondary outcome) did not differ between either
the CBT or EDU group and the usual care group.
Limitations: Participants represented a single health care sys-
tem. Self-selection bias may have been present.
Conclusion: Simplified group CBT and EDU interventions deliv-
ered at low-income clinics significantly improved pain and phys-
ical function compared with usual care.
Primary Funding Source: Patient-Centered Outcomes Re-
search Institute.
Ann Intern Med. 2018;168:471-480. doi:10.7326/M17-0972 Annals.org
For author affiliations, see end of text.
This article was published at Annals.org on 27 February 2018.
* Authors listed after Dr. Campbell are in alphabetical order.
Although CBT for chronic pain is effective (12–14),
rigorous trials have not been conducted in populations
with low socioeconomic status (1, 2). Pilot research
showed that group-administered CBT and pain educa-
tion (EDU), adapted to reduce cognitive demands,
were efficacious in persons with low socioeconomic
status who have chronic pain (15). That trial was under-
powered to detect differences between treatment
groups and lacked a usual care comparison group, pre-
cluding evaluation of whether the interventions could
augment existing practices (16).
The purpose of this study was to compare literacy-
adapted CBT, EDU, and usual care for treating chronic
See also:
Editorial comment . . . . . . . . . . . . . . . . . . . . . . . . . 517
Summary for Patients . . . . . . . . . . . . . . . . . . . . . . . I-18
Web-Only
Supplement
© 2018 American College of Physicians 471
 ORIGINAL RESEARCH
pain in a larger sample of patients with chronic pain.
We hypothesized that participants randomly assigned
to receive CBT or EDU would show greater posttreat-
ment improvements in pain intensity, physical function,
and depression than those receiving usual care and
that CBT would be superior to EDU and usual care in
reducing depression. Because of the highly disadvan-
taged nature of the population and anticipated diffi-
culty reaching participants for follow-up, we defined
“posttreatment” as our primary end point and did not
make hypotheses about maintenance of effects at 6
months.
METHODS
Design Overview
The LAMP (Learning About My Pain) study was a
3-group, parallel-design (1:1:1 allocation ratio), ran-
domized controlled trial comparing CBT, EDU, and
usual care in treating chronic pain (17). Participants
were recruited from 12 September 2013 to 1 Decem-
ber 2015, and follow-up visits were completed by 30
September 2016. Although the trial registration was
submitted to ClinicalTrials.gov in June 2013, required
revisions delayed its final publication. Sixteen of 290
patients (5.5% of the sample) were enrolled before trial
registration on 21 October 2013, but no outcome data
were collected. Procedures were approved by the Uni-
versity of Alabama Institutional Review Board (IRB); mi-
nor protocol changes during the study are described in
Appendix Table 1 (available at Annals.org). Informed
consent was obtained at enrollment.
Setting and Participants
Participants were recruited from 4 clinic sites in 36
separate cohorts at community health centers adminis-
tered by Whatley Health Services (WHS), a private non-
profit agency in western Alabama that predominantly
serves members of ethnic minorities with low socioeco-
nomic status. Two WHS employees served as paid par-
ticipant coordinators. Together, they recruited and
screened potential participants for eligibility, followed
all participants (including those in usual care), and facil-
itated retention. All WHS patients with a chronic pain–
related diagnosis were contacted to determine their in-
terest in participating in the study. Eligible persons
were at least 19 years of age; had a chronic pain–
related diagnosis, with pain most days of the month for
at least 3 months; were proficient in English; and pro-
vided a means of contact. Exclusion criteria were pain
related to cancer, clinically significant cognitive impair-
ment, current self-reported substance abuse, serious
uncontrolled mental illness, a reading level below first
grade, recent changes in pain or psychotropic medica-
tion (in the previous 4 weeks), and current psychosocial
treatment for pain (Figure 1).
Randomization and Interventions
The study statistician used statistical software to
generate a random-number table that stratified treat-
ment assignments by site (n = 4, with intentional
weighting toward the highest-volume site) and was bal-
472 Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018
Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018
Literacy-Adapted Psychosocial Treatments for Pain
anced by treatment group (n = 3). At each site, partici-
pants were assigned to cohorts of 7 to 9 persons per
group.
Cognitive behavioral therapy and EDU were ad-
ministered in group format by co-therapists. Both inter-
ventions were the result of adaptations begun in previ-
ous research to simplify treatment procedures and
revise patient workbooks to a fifth grade reading level
(15, 18). The patient workbooks and therapist supple-
ments are copyrighted but freely available at pmt.ua
.edu/publications.html. Each participant received a
workbook and weekly audio CD with a summary of that
week's material. For the CBT groups, CDs included in-
structions for skills practice and a relaxation exercise.
Participants were compensated $20 per session for
travel to the group sessions, with no additional com-
pensation for attendance.
CBT: LAMP Group
The CBT treatment manual used in this study is
standardized and supported by empirical studies (15,
19). Ten weekly 90-minute group sessions provided
simplified cognitive behavioral techniques, including
motivational reinforcement, pain education, and pain
management skills training (such as cognitive restruc-
turing, activity pacing, and relaxation).
EDU: LAMP Group
Ten weekly 90-minute group sessions provided rel-
evant pain-related information aimed toward pain self-
management and group discussion, but no specific
skill-building exercises were taught. The EDU program
was matched to the CBT intervention with regard to
format, discussion, and therapist engagement; it is
standardized and supported by empirical research (15,
18, 20).
Both Groups
All participants received usual medical care, which
may have involved any nonpsychological pain manage-
ment method, including medication, chiropractic, or
physical therapy, through WHS or external providers.
Those randomly assigned to usual care had parallel
contact with participant coordinators and assessors, as
well as intermittent phone contact to facilitate reten-
tion, but received no psychosocial treatment.
Therapists worked in pairs of 1 doctorate-level psy-
chologist and 1 predoctoral clinical psychology trainee
(total of 10 therapists: 5 doctoral, 5 predoctoral). Train-
ing for the CBT and EDU therapists, as well as supervi-
sion of ongoing therapy, was provided by using the
standardized CBT and EDU materials. All sessions were
video recorded, and a study investigator uninvolved in
treatment used a structured scale (21) to evaluate the
therapists' adherence to the protocol and the quality of
the intervention for 26.7% of the sessions (randomly
selected) (Appendix Table 2, available at Annals.org).
Annals.org
 Literacy-Adapted Psychosocial Treatments for Pain ORIGINAL RESEARCH

Figure 1. Flow of participants through the trial with usual care, CBT, and EDU.

Assessed for eligibility (n = 485)

Excluded (n = 195)
Randomly assigned (n = 290;
36 group cohorts)
Did not meet inclusion criteria: 17
Reading below 1st grade level: 8
Cognitive impairment: 2
Serious uncontrolled mental illness: 5
Substance abuse: 1
Acute pain only: 1
Declined to participate: 96
Unable to contact: 82

Assigned to CBT intervention (n = 95; 12 group
cohorts)
Received assigned intervention (>6/10
sessions): 71
Did not receive assigned intervention: 24
Did not complete any sessions: 7
Completed 1 session: 5
Completed 2 sessions: 2
Completed 3 sessions: 1
Completed 4 sessions: 1
Completed 5 sessions: 2
Completed 6 sessions: 6
Assigned to EDU intervention (n = 97; 12 group Assigned to usual care (n = 98; 12 cohorts)
cohorts)
Received assigned intervention (>6/10
sessions): 60
Did not receive assigned intervention: 37
Did not complete any sessions: 18
Completed 1 session: 7
Completed 2 sessions: 2
Completed 3 sessions: 3
Completed 4 sessions: 3
Completed 5 sessions: 2
Completed 6 sessions: 2

Completed posttreatment assessment (n = 83)
Lost to follow-up (n = 11)
Unable to contact: 11
Interested but unable to attend: 0
Discontinued intervention (n = 1)
Not interested: 1
Removed by researchers: 0
Completed posttreatment assessment (n = 80)
Lost to follow-up (n = 15)
Unable to contact: 11
Interested but unable to attend: 4
Discontinued intervention (n = 2)
Not interested: 2
Removed by researchers: 0
Completed posttreatment assessment (n = 78)
Lost to follow-up (n = 16)
Unable to contact: 12
Interested but unable to attend: 4
Discontinued study participation (n = 4)
Not interested: 3
Removed by researchers: 1

Completed 6-mo follow-up assessment (n = 71)
Lost to follow-up (n = 23)
Unable to contact: 23
Interested but unable to attend: 0
Discontinued intervention (n = 0)
Not interested: 0
Removed by researchers: 0
Completed 6-mo follow-up assessment (n = 68)
Lost to follow-up (n = 27)
Unable to contact: 27
Interested but unable to attend: 0
Discontinued intervention (n = 0)
Not interested: 0
Removed by researchers: 0
Completed 6-mo follow-up assessment (n = 71)
Lost to follow-up (n = 23)
Unable to contact: 23
Interested but unable to attend: 0
Discontinued study intervention (n = 0)
Not interested: 0
Removed by researchers: 0

Analyzed (n = 95) Analyzed (n = 97) Analyzed (n = 98)

Excluded from analysis (n = 0) Excluded from analysis (n = 0) Excluded from analysis (n = 0)

CBT = cognitive behavioral therapy; EDU = pain education.

Blinding Outcomes and Follow-up

Study participants and patient coordinators were
not blinded to treatment allocation but were unaware
of the study hypotheses. Therapists were not blinded to
participants' treatment assignment for the groups they
led. Study outcomes were collected by blinded, trained
assessors. Because the principal investigator super-
vised the therapists, she was not blinded to partici-
pants' assignment status; however, she did not conduct
outcome assessments, nor was she one of the medical
providers following the patients.
Annals.org
Assessors conducted face-to-face, 90-minute inter-
views, presenting assessment materials verbally. As-
sessment data were collected at baseline (before
randomization), 5 weeks after the start of treatment
(mid-treatment), at 10 weeks (posttreatment), and 6
months after treatment. Participants were compensated
$45 for each assessment.
Outcomes were prespecified as primary—pain
intensity, assessed by using the 4-item Brief Pain
Inventory–Short Form (BPI-SF) Pain Intensity subscale—
Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 473

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 ORIGINAL RESEARCH Literacy-Adapted Psychosocial Treatments for Pain

Table 1. Baseline Characteristics of Study Participants, by Treatment Group

Characteristic All (n ⴝ 290) UC (n ⴝ 98) CBT (n ⴝ 95) EDU (n ⴝ 97)

Sociodemographic characteristics
Mean age (SD), y 50.6 (8.9) 49.7 (8.7) 52.2 (8.5) 49.9 (9.2)
Women, n (%) 205 (70.7) 69 (70.4) 67 (70.5) 69 (71.1)
Education, n (%)
No degree 85 (29.3) 27 (27.6) 29 (30.5) 29 (29.9)
High school graduate/GED 112 (38.6) 36 (36.7) 36 (37.9) 40 (41.2)
Some college or vocational school 49 (16.9) 19 (19.4) 17 (17.9) 13 (13.4)
College graduate* 44 (15.2) 16 (16.3) 13 (13.7) 15 (15.5)
Race, n (%)
White/Caucasian† 96 (33.1) 38 (38.8) 20 (21.1) 38 (39.2)
Black/African American‡ 194 (66.9) 60 (61.2) 75 (78.9) 59 (60.8)
Marital status, n (%)
Single 71 (24.5) 27 (27.6) 24 (25.3) 20 (20.6)
Married or in a relationship >2 y 100 (34.5) 36 (36.7) 27 (28.4) 37 (38.1)
Divorced, separated, or widowed 119 (41.0) 35 (35.7) 44 (46.3) 40 (41.2)
Poverty status, n (%)§
Below 210 (72.4) 74 (75.5) 65 (68.4) 71 (73.2)
Above 70 (24.1) 20 (20.4) 25 (26.3) 25 (25.8)
Employed, n (%)兩兩 39 (13.4) 15 (15.3) 9 (9.5) 15 (15.5)
Insurance status, n (%)
No insurance 124 (42.8) 45 (45.9) 35 (36.8) 44 (45.4)
Private health insurance 23 (7.9) 4 (4.1) 11 (11.6) 8 (8.2)
Medicaid 68 (23.4) 28 (28.6) 21 (22.1) 19 (19.6)
Medicare 42 (14.5) 14 (14.3) 13 (13.7) 15 (15.5)
Combination¶ 33 (11.4) 7 (7.1) 15 (15.8) 11 (11.3)
Disability status, n (%)
Receiving disability benefits 137 (47.2) 44 (44.9) 49 (51.6) 44 (45.4)
Seeking disability benefits 103 (35.5) 38 (38.8) 30 (31.6) 35 (36.1)
Not receiving or seeking disability benefits 50 (17.2) 16 (16.3) 16 (16.8) 18 (18.6)
Mean WRAT GLE (SD) 7.4 (3.6) 8.2 (3.5) 6.7 (3.7) 7.3 (3.6)
GLE, n (%)
1–2 25 (8.6) 4 (4.1) 13 (13.7) 8 (8.2)
3–4 79 (27.2) 24 (24.5) 28 (29.5) 27 (27.8)
5–6 52 (17.9) 14 (14.3) 19 (20.0) 19 (19.6)
7–8 24 (8.3) 11 (11.2) 5 (5.3) 8 (8.2)
9–10 37 (12.8) 15 (15.3) 8 (8.4) 14 (14.4)
11–12 33 (11.4) 13 (13.3) 11 (11.6) 9 (9.3)
>12 40 (13.8) 17 (17.3) 11 (11.6) 12 (12.4)

Chronic pain history

Mean pain duration (SD), y 16.6 (12.2) 18.1 (13.1) 15.0 (10.6) 16.7 (12.8)
Primary pain site, n (%)
Lower back 144 (49.7) 55 (56.1) 44 (46.3) 45 (46.4)
Knee 37 (12.8) 11 (11.2) 19 (20.0) 7 (7.2)
Neck 23 (7.9) 5 (5.1) 5 (5.3) 13 (13.4)
Other** 86 (29.7) 27 (27.6) 27 (28.4) 32 (33.0)
Mean reported pain sites (SD), n 6.2 (3.1) 6.1 (3.1) 6.3 (3.2) 6.4 (3.1)
Type of pain, n (%)
Musculoskeletal†† 258 (89.0) 87 (88.8) 82 (86.3) 89 (91.8)
Arthritis‡‡ 220 (75.9) 67 (68.4) 80 (84.2) 73 (75.3)
Headache§§ 135 (46.6) 53 (54.1) 39 (41.1) 43 (44.3)
Pelvic 96 (33.1) 36 (36.7) 33 (34.7) 27 (27.8)
Nerve兩兩兩兩 92 (31.7) 37 (37.8) 20 (21.1) 35 (36.1)
IBS or abdominal 58 (20.0) 24 (24.5) 12 (12.6) 22 (22.7)
Chronic fatigue 41 (14.1) 13 (13.3) 15 (15.8) 13 (13.4)
Fibromyalgia 34 (11.7) 7 (7.1) 15 (15.8) 12 (12.4)
Mean reported pain type (SD), n 4.7 (2.7) 4.8 (2.9) 4.6 (2.6) 4.7 (2.7)

Baseline measures of pain treatment

Medications prescribed, n (%)¶¶
Opioids 191 (67.5) 67 (70.5) 59 (62.8) 65 (69.1)
Tramadol 149 (52.7) 51 (53.7) 43 (45.7) 55 (58.5)
Acetaminophen–hydrocodone 166 (57.2) 58 (61.1) 52 (55.3) 56 (59.6)
NSAIDs 162 (57.2) 50 (52.6) 54 (57.4) 58 (61.7)
Muscle relaxants 141 (49.8) 49 (51.6) 36 (38.3) 56 (59.6)
Nerve pain medications 73 (25.8) 27 (28.4) 24 (25.5) 22 (23.4)
Psychotropic medications 85 (30.0) 30 (31.6) 27 (28.7) 28 (29.8)
Continued on following page

474 Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 Annals.org

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Literacy-Adapted Psychosocial Treatments for Pain ORIGINAL RESEARCH

Table 1—Continued

Characteristic All (n ⴝ 290) UC (n ⴝ 98) CBT (n ⴝ 95) EDU (n ⴝ 97)

Self-reported treatments, n (%)***
Surgery 95 (34.2) 31 (32.3) 35 (40.7) 29 (30.2)
Nerve blocks 94 (33.8) 31 (32.3) 34 (39.5) 29 (30.2)
Chiropractor 129 (46.4) 45 (46.9) 39 (45.3) 45 (46.9)
Physical therapy 151 (54.3) 48 (50.0) 53 (61.6) 50 (52.1)
Psychological treatment 37 (13.3) 11 (11.5) 13 (15.1) 13 (13.5)

Baseline measures of primary outcomes

Mean BPI-Intensity score (SD) 6.5 (1.6) 6.5 (1.6) 6.5 (1.8) 6.5 (1.5)

Baseline measures of secondary outcomes

Mean BPI-Interference score (SD) 6.6 (2.0) 6.6 (2.1) 6.7 (2.1) 6.6 (1.9)
Mean PHQ-9 score (SD) 12.1 (6.4) 12.8 (6.4) 11.7 (6.1) 11.9 (6.8)
BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; GLE = grade-level expectation; IBS = irritable bowel
syndrome; NSAID = nonsteroidal anti-inflammatory drug; PHQ-9 = Patient Health Questionnaire–9; UC = usual care; WRAT = Wide Range Achieve-
ment Test.
* Indicates 2-y/technical graduate (n = 33), 4-y/college graduate (n = 10), and some graduate school (n = 1).
† Indicates white (n = 84) and white/Caucasian and Native American (n = 12).
‡ Indicates black/African American (n = 191) and black/African American and Native American (n = 3).
§ Missing data from 10 participants (UC, n = 4; CBT, n = 5; and EDU, n = 1).
兩兩 Indicates full-time employment, part-time employment, and homemaker.
¶ Indicates private health insurance and Medicaid (n = 1); private health insurance and Medicare (n = 5); Medicaid and Medicare (n = 25); and
private health insurance, Medicaid, and Medicare (n = 2).
** Indicates shoulders (n = 13), upper leg (n = 12), pelvis (n = 11), hands (n = 11), feet (n = 9), head (n = 9), lower leg/ankle (n = 6), upper back
(n = 6), unspecified (n = 6), arms (n = 2), and abdomen (n = 1).
†† Indicates low back (n = 246), neck (n = 158), soft tissue or muscle (n = 145), and spinal cord injury (n = 24).
‡‡ Indicates osteoarthritis (n = 141), rheumatoid arthritis (n = 56), and mixed arthritis (n = 48).
§§ Indicates tension headache (n = 81), migraine pain (n = 77), mixed headache (n = 26), and cluster headache (n = 12).
兩兩兩兩 Indicates neuropathic pain (n = 86) and chronic regional pain syndrome (n = 10).
¶¶ Indicates those prescribed within 1 y before study onset. This category is missing 7 participants (UC, n = 3; CBT, n = 1; and EDU, n = 3).
*** Missing 12 participants (UC, n = 2; CBT, n = 9; and EDU, n = 1).

and secondary—physical function, measured by the
BPI-SF Pain Interference subscale (22), and depression,
measured by the Patient Health Questionnaire–9,
(PHQ-9) (23). Patient Global Impression of Change (24)
was a secondary outcome evaluated in the study but
not reported here.
At baseline, sociodemographic and pain informa-
tion was obtained, including self-reported methods to
manage pain, and data regarding opioid prescriptions
were collected from medical records. Primary literacy
also was assessed with the Wide Range Achievement
Test 4 Word Reading Subtest (25). Treatment adher-
ence was defined as the number of CBT or EDU group
sessions attended, and adequate dosage was defined
as attendance at more than 6 sessions.
Self-reported adverse events (emergency room vis-
its, hospitalization, suicidal ideation) were queried
weekly by therapists for participants in the CBT and
EDU groups and by assessors at mid-treatment and
posttreatment for all participants (including those re-
ceiving usual care). Identified adverse events were doc-
umented in compliance with IRB policies, and IRB staff
ultimately determined whether each event was related
to the study. An external data safety officer monitored
all trial safety procedures and events, as did the IRB
and principal investigator.
Statistical Analysis
Sample size was estimated by using the smallest
expected between-group effect: the difference be-
tween groups across time in depressive symptoms
(␩p2 = 0.074) observed in the pilot trial (15). Partial eta
squared is the proportion of variance in the dependent
Annals.org
variable accounted for by different levels of an inde-
pendent variable. On the basis of an ␣ of 0.05 and a
power of 0.80, a total of 158 participants were needed.
Estimating an attrition rate of 15% and accounting for
aspects of the study design, we chose a target sample
of 294 (3 treatment groups × 12 cohorts × 7 to 9 par-
ticipants per cohort). Because negative health effects
were not anticipated, stopping guidelines were not
used. No interim analyses were conducted.
Using linear mixed models, we examined changes
from pre- to posttreatment and from posttreatment to
6-month follow-up, as well as differences between
groups in the primary outcome (BPI-Intensity) and sec-
ondary outcomes (BPI-Interference and PHQ-9). Three-
level analyses were conducted—repeated measures
(level 1) nested within participants (level 2) and partici-
pants nested within group cohorts (level 3)— by having
a 3 × 1 vector of outcome variables across time points
(pretreatment, posttreatment, and follow-up). To com-
pute effect sizes for the linear mixed models, Hedges ␦T
(26) (derived for nested designs) was calculated and
interpreted by using the Cohen d (27). For secondary
outcomes, a Benjamini-Hochberg– corrected ␣ was
used to protect against familywise ␣ inflation due to
multiple comparisons (28).
Additional analyses used logistic regression to ex-
amine the effect of the interventions on the binary
outcome of clinically meaningful improvement (CMI),
defined as 30% or greater improvement in each contin-
uous variable: 0 for no improvement, and 1 for improve-
ment (29). Odds ratios (ORs) and CIs and adjusted differ-
ences in predicted percentages of participants with CMI
Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 475

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 ORIGINAL RESEARCH Literacy-Adapted Psychosocial Treatments for Pain

Table 2. Estimated Differences in Change Scores for CBT and EDU vs. UC From Baseline to Posttreatment and Posttreatment
to 6-Month Follow-up*

Group Baseline to Posttreatment Posttreatment to 6-Month Follow-up

Within-Group P Value Change vs. P Value Within-Group P Value Change vs. P Value
Change (95% CI) UC (95% CI) Change (95% CI) UC (95% CI)
Pain intensity by
BPI-Intensity score
UC –0.26 (–0.61 to 0.08) 0.139 – –0.24 (–0.60 to 0.11) 0.173 –
CBT –1.06 (–1.65 to –0.47) <0.001 –0.80 (–1.48 to –0.11) 0.022 0.35 (–0.05 to 0.76) 0.085 0.60 (0.06 to 1.13) 0.028
EDU –0.83 (–1.14 to –0.41) <0.001 –0.57 (–1.04 to –0.10) 0.018 0.14 (–0.20 to 0.59) 0.42 0.38 (–0.11 to 0.87) 0.124

Physical function by
BPI-Interference score
UC –0.30 (–0.77 to 0.17) 0.21 – 0.18 (–0.25 to 0.62) 0.41 –
CBT –1.66 (–2.24 to –1.07) <0.001 –1.36 (–2.11 to –0.61) <0.001 0.54 (0.09 to 1.00) 0.020 0.36 (–0.27 to 0.99) 0.26
EDU –1.00 (–1.39 to –0.62) <0.001 –0.70 (–1.31 to –0.09) 0.024 0.25 (–0.19 to 0.69) 0.26 0.07 (–0.55 to 0.68) 0.83

Depression by
PHQ-9 score
UC –1.10 (–2.10 to –0.09) 0.032 – –0.24 (–0.83 to 0.35) 0.43 –
CBT –2.43 (–3.68 to –1.19) <0.001 –1.33 (–3.02 to 0.35) 0.120 0.61 (–0.49 to 1.70) 0.28 0.85 (–0.69 to 2.38) 0.29
EDU –2.24 (–3.51 to –0.98) 0.001 –1.15 (–2.71 to 0.41) 0.147 0.43 (–0.43 to 1.29) 0.33 0.67 (–0.18 to 1.53) 0.123
BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; PHQ-9 = Patient Health Questionnaire–9; UC = usual care.
* Negative values (i.e., lower or decreased scores) indicate improvement or advantage on the specified variable. To control for multiple compari-
sons for the secondary outcomes (physical function and depression), the Benjamini–Hochberg correction method was used, whereby the smaller
familywise P value was compared with ␣ = 0.025 and the larger P value was compared with ␣ = 0.050. Significance values on the primary outcome
(pain intensity) were compared with ␣ = 0.050. Estimated values that are statistically significant are in boldface. For the comparisons from post-
treatment to 6-mo follow-up, statistically significant values indicate that treatment gains were no longer maintained. All analyses were adjusted to
control for sex, minority status, years of education, and pain duration.

were used to report the likelihood of CMI for the treat-
ment groups compared with the group receiving usual
care.
Analyses included all randomly assigned partici-
pants and used all data available for each person. The
full information maximum likelihood estimation method
was used to account for possible nonresponse bias.
This method explains missing data by borrowing infor-
mation from the observed data and yields unbiased pa-
rameter estimates under a missing at random (MAR)
mechanism (30, 31). Linear mixed and logistic models
were adjusted for sex, minority status, years of educa-
tion, and pain duration. Inferential tests were 2-tailed.
Mplus, version 7.4 (32), was used to analyze linear
mixed and logistic regression models. Confidence in-
tervals for ORs and adjusted differences in predicted
percentages of participants with CMI for the logistic
models were calculated by using the CINTERVAL option
in Mplus.
Sensitivity analyses for the primary and secondary
outcomes were conducted by using a pattern-mixture
model to assess possible departures from the MAR as-
sumption for the linear mixed models (33–35). Pattern-
mixture modeling is based on the factorization of the
likelihood [y,d] = [y | d][d], where y is the response vari-
able and d is the dummy variable for dropout. We had
2 dropout dummy variables, d1 and d2, where dt is de-
fined as dt = 1 for a participant who drops out after time
t ⫺ 1 (t = 2, 3), t = 2 represents posttreatment, and t = 3
represents 6-month follow-up. We applied a simple
version of the pattern-mixture model by allowing the
random-effect means to vary as a function of the drop-
out dummy variables (35).
476 Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018
Role of the Funding Source
This work was funded by the Patient-Centered Out-
comes Research Institute (PCORI) and the University of
Alabama. The study design, conduct, and reporting
were solely the authors' responsibility.
RESULTS
Study Sample
Figure 1 shows enrollment details for all 290 study
participants, 241 (83.1%) of whom took part in the post-
treatment assessment. Of the 192 participants ran-
domly assigned to group interventions, 74.7% in the
CBT and 61.9% in the EDU group received an ade-
quate treatment dosage (>6 sessions). Table 1 provides
baseline information on participants, including demo-
graphic characteristics and pain-related variables, with
72.4% of participants at or below the poverty level,
35.8% reading below the fifth grade level, 82.7% re-
ceiving or seeking disability benefits, and 66.9% from
racial/ethnic minority groups. Participants' ages ranged
from 19 to 71 years. On average, the sample reported
more than 6 pain sites, more than 4 causes of pain, and
pain lasting longer than 15 years.
Outcomes
Within-group pain intensity scores decreased sig-
nificantly from pre- to posttreatment for both CBT and
EDU groups, with treatment gains maintained at follow-
up. The usual care group did not show changes in pain
intensity across time (Table 2). Compared with usual
care, CBT and EDU showed larger reductions in pain in-
tensity between baseline and posttreatment (estimated
Annals.org

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Literacy-Adapted Psychosocial Treatments for Pain ORIGINAL RESEARCH

Figure 2. Predicted mean pain intensity (BPI-Intensity), physical function (BPI-Interference), and depression (PHQ-9) scores, by
treatment group and time point from mixed linear models.

8 Usual care 8 13
CBT
EDU
7.5 7.5 12

7 7
11

BPI-Interference Score
BPI-Intensity Score

PHQ-9 Score
6.5 6.5
10

6 6

9
5.5 5.5

8
5 5

0 0 0
Baseline Posttreatment 6-Month Follow-up Baseline Posttreatment 6-Month Follow-up Baseline Posttreatment 6-Month Follow-up

BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; PHQ-9 = Patient Health Questionnaire–9.

differences—CBT: ⫺0.80 [95% CI, ⫺1.48 to ⫺0.11]; P = ⫺0.70 [CI, ⫺1.31 to ⫺0.09]; P < 0.001), and this prog-
0.022; EDU: ⫺0.57 [CI ⫺1.04 to ⫺0.10]; P = 0.018). At ress was maintained at follow-up (Table 2 and Figure
follow-up, treatment gains (compared with usual care) 2). The proportion of participants with 30% or greater
were not maintained for CBT but were still present for improvement in physical function score was 40.2%,
EDU (Table 2 and Figure 2). Appendix Table 3 (avail- 28.8%, and 18.4% at posttreatment and 34.8%, 20.6%,
able at Annals.org) provides effect sizes and associated and 15.7% at follow-up for the CBT, EDU, and usual
CIs. The Appendix Figure (available at Annals.org) illus- care groups, respectively. The pattern of differences
trates the cumulative proportion of percent changes in suggests that the CBT intervention consistently dif-
pain intensity by group (36). This Figure allows compar- fered from usual care at posttreatment and follow-up,
ison among groups across the entire range of response whereas EDU and usual care showed somewhat incon-
levels, demonstrating that the CBT and EDU groups sistent patterns for absolute differences and ORs (Table
had a consistently higher proportion of participants 3). Both CBT and EDU exceeded the minimally impor-
with improved pain intensity scores than the usual care tant change criterion of 14% (29) at posttreatment; only
group. CBT exceeded the criterion at 6 months (Appendix
The proportion of participants with 30% or greater Table 4).
improvement in pain intensity score was 30.5%, 20.0%, Within-group depression scores decreased signifi-
and 11.5% at posttreatment and 21.7%, 16.4%, and
cantly from pre- to posttreatment for the CBT and EDU
8.5% at 6-month follow-up for those receiving CBT,
groups, and these treatment gains were maintained at
EDU, and usual care, respectively. The pattern of differ-
follow-up. Participants receiving usual care did not
ences was consistent for both absolute differences and
show changes in depression across time (Table 2).
ORs (Table 3): Both the CBT and EDU groups had a
Change estimates for depression between baseline and
greater proportion of members with CMI at posttreat-
ment and follow-up than the usual care group. The CBT posttreatment did not statistically differ between usual
and EDU groups exceeded the minimally important care and either CBT or EDU (estimated differences—
change criterion of 10% (29) at posttreatment and 6 CBT: ⫺1.33 [CI, ⫺3.02 to 0.35]; P = 0.120; EDU: ⫺1.15
months, whereas the usual care group did not (Appen- [CI, ⫺2.71 to 0.41]; P = 0.147), nor did they differ at
dix Table 4, available at Annals.org). follow-up (Table 2 and Figure 2). The proportion of
Within-group scores for interference in physical participants with CMI was 39.5%, 42.5%, and 28.2%
function decreased significantly from pre- to posttreat- at posttreatment and 42.6%, 33.8%, and 31.0% at
ment for both CBT and EDU groups; treatment gains follow-up in the CBT, EDU, and usual care groups, re-
were not maintained at follow-up among CBT partici- spectively. The pattern of findings was consistent at
pants but were still present in the EDU group. Partici- posttreatment, with both CBT and EDU showing signif-
pants receiving usual care did not show a change in icant differences from usual care. At follow-up, the pat-
physical function across time (Table 2). Compared with tern of results showed inconsistencies between abso-
the usual care group, both CBT and EDU groups had lute differences and ORs but generally indicated that
greater improvement in physical function scores between CBT and EDU no longer differed from usual care (Table
baseline and posttreatment (estimated differences— 3). Appendix Table 5 (available at Annals.org) provides
CBT: ⫺1.36 [CI, ⫺2.11 to ⫺0.61]; P < 0.001; EDU: numbers and percentages of participants with depres-
Annals.org Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 477

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 ORIGINAL RESEARCH Literacy-Adapted Psychosocial Treatments for Pain

Table 3. CMI and Differences Among Treatments, by Treatment*

Variable Participants With CMI, n (%) Adjusted Difference in Percentage of Odds Ratio (95% CI)
Participants With CMI (95% CI)

UC CBT EDU CBT vs. UC EDU vs. UC CBT vs. UC EDU vs. UC
Pretreatment to (n = 78) (n = 82) (n = 80)
posttreatment
Pain intensity† 9 (11.5) 25 (30.5) 16 (20.0) 18.1 (8.5 to 28.8) 8.6 (4.4 to 12.8) 3.43 (2.72 to 4.32) 2.00 (1.52 to 2.64)
Physical function‡ 14 (18.4) 33 (40.2) 23 (28.8) 23.3 (15.0 to 32.2) 11.4 (7.1 to 16.2) 3.34 (2.26 to 4.93) 2.01 (0.84 to 4.78)
Depression§ 22 (28.2) 32 (39.5) 34 (42.5) 12.6 (5.3 to 20.1) 14.7 (9.5 to 20.0) 1.76 (1.37 to 2.28) 1.91 (1.46 to 2.50)

Pretreatment to (n = 71) (n = 69) (n = 68)

6-mo follow-up
Pain intensity† 6 (8.5) 15 (21.7) 11 (16.4) 10.0 (1.5 to 19.7) 7.5 (2.1 to 12.5) 2.70 (2.46 to 2.96) 2.16 (1.73 to 2.71)
Physical function‡ 11 (15.7) 24 (34.8) 14 (20.6) 19.6 (7.7 to 32.7) 4.6 (–1.0 to 10.4) 3.15 (2.38 to 4.17) 1.23 (1.04 to 1.44)
Depression§ 22 (31.0) 29 (42.6) 23 (33.8) 10.8 (–0.6 to 22.3) 2.9 (–5.4 to 11.2) 1.60 (1.41 to 1.82) 1.13 (0.96 to 1.33)
CBT = cognitive behavioral therapy; CMI = clinically meaningful improvement; EDU = pain education; UC = usual care.
* CMI was defined as ≥30% improvement since pretreatment. All analyses were adjusted to control for sex, minority status, years of education, and
pain duration.
† Measured by Brief Pain Inventory Intensity score.
‡ Measured by Brief Pain Inventory Interference score.
§ Measured by Patient Health Questionnaire–9 score.

sion scores above and below the “probable depres-
sion” cutoff, compared across treatment groups (23).
Sensitivity analyses showed that the results for pain
intensity, physical function, and depression were robust
to departures from the MAR assumption, because both
the participant and group cohort random-effect means
did not significantly vary at the 5% significance level as
a function of the 2 dropout dummy variables. Appendix
Tables 6 and 7 (available at Annals.org) provide the
results of the sensitivity analyses.
Adverse Events
During the trial, 9 of 95 participants (9.4%) in the
CBT, 17 of 97 (17.5%) in the EDU, and 18 of 98 (18.4%)
in the usual care group reported seeking emergency
department treatment. Causes were mostly temporary
pain exacerbations, with a few reported adverse events
due to infections or injuries and 2 based on suicidal
ideation. Six participants were hospitalized overnight
for non–pain-related reasons (such as asthma): 3 in the
usual care, 2 in the EDU, and 1 in the CBT group. No
events were determined to be caused by participation
in the trial.
DISCUSSION
Our study tested the efficacy of simplified CBT and
EDU in a socioeconomically disadvantaged population
and, as hypothesized, produced additional evidence
supporting the efficacy and acceptability of group CBT
and EDU for chronic pain. Within-group scores for pain
intensity, interference in physical function, and depres-
sion decreased substantially from pre- to posttreatment
for both CBT and EDU groups, with treatment gains
mostly maintained at 6-month follow-up. Participants
receiving usual care did not show meaningful
changes in outcomes across time. The CBT and EDU
groups both had a greater decrease in pain intensity
and more gains in physical function than the usual care
group. The hypothesis that CBT would be superior to
EDU in reducing depression was not supported in the
478 Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018
primary analyses, although both types of treatment
were more likely than usual care to yield posttreatment
CMI.
Focusing solely on a sample from a highly disad-
vantaged population, this trial demonstrates that
literacy-adapted CBT and EDU are suitable adjunctive
care options for adults with chronic pain attending low-
income clinics. Given the extent of the adaptations
made to the CBT intervention, our findings strongly
suggest that CBT can be simplified to improve its
accessibility while retaining its core principles and not
reducing its potency. Considering effect size compari-
sons (14, 37), CMIs (29), and comparisons with mini-
mally important change criteria (23, 29), CBT showed
some small clinical advantages over EDU with regard to
pain and physical function. However, given the greater
simplicity of EDU, meaningful treatment results for this
intervention are important.
Current biomedically focused treatments for the
complex and widespread public health problem of
chronic pain have not been largely effective and at
times have inadvertently caused harm (9, 38). These
observations have led to biopsychosocial recommen-
dations for treating chronic pain that involve nonphar-
maceutical, noninvasive therapies (1, 9 –11). Among
other approaches, CBT has shown efficacy in random-
ized trials (12–14). The potential positive effect on the
outcomes and economics of a safe and effective psy-
chosocial therapy is enormous. As just 1 example, us-
ing Medicare reimbursement rates and comparing the
cost of a typical lumbar fusion procedure versus a stan-
dardized 10-session group CBT approach, researchers
found that the surgery alone cost 168 times more than
CBT (39).
Other studies have used EDU as an attention con-
trol (40, 41). Our adaptations of this therapy included a
biopsychosocial focus on enhancing motivation for
pain self-management, emphasizing therapeutic alli-
ance and group cohesion and giving participants rele-
vant information to discuss and use but without explicit
Annals.org

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Literacy-Adapted Psychosocial Treatments for Pain
training in skills, such as those used in CBT. Our find-
ings suggest that providing a group venue with rele-
vant pain information and interactive discussion for pa-
tients with pain is useful. As an intervention involving
less expense, time, and effort on the part of patients
and staff, EDU (as compared with CBT) may be an ef-
fective alternative, or it might be used as part of a
stepped-care approach before CBT. Nonetheless, it is
important to emphasize that the EDU intervention fol-
lowed a structured protocol and would require sub-
stantial staff resources to achieve the potency observed
in this trial.
Study strengths include a large sample with ade-
quate statistical power to detect clinically meaningful
effects; close matching of the CBT and EDU interven-
tions with regard to format; and several protections
against bias, including a randomized controlled trial
design with blinded assessment of study outcomes
(17). Another unique feature is that our participants
were from a highly disadvantaged population; they had
severe and complex pain problems; and many repre-
sented the compounded risks associated with low in-
come, low literacy, and minority group (African Ameri-
can) membership. Low literacy, in particular, often
renders potential participants ineligible to take part in
health research. Although the LAMP study could not
change financial realities for study participants, the so-
cial and educational vulnerabilities of the sample
shaped the development and delivery of both biopsy-
chosocial interventions. The group format provided so-
cial support, and participants who did not read at the
fifth grade level or higher received oral assistance in
understanding the intervention materials and complet-
ing surveys. Given the economic, social, and educa-
tional challenges faced by many study participants, the
recruitment, attendance, retention, and treatment re-
sponse rates in the current trial demonstrate that health
research can be made more accessible to low-income
adults from all racial/ethnic backgrounds and educa-
tional levels.
Among the study's limitations was the enrollment
of participants from clinics within a single health care
system with limited resources, making generalizability
to other health care systems unclear. Furthermore, self-
selection bias was an unavoidable limitation. Given that
patients had mixed chronic pain conditions, we did not
determine whether treatment response differed ac-
cording to chronic pain type. We also could not docu-
ment whether changes in opioid use occurred as a re-
sult of the intervention because of concurrent initiatives
to curb analgesic use at the study sites.
Methodologically, some studies suggest that func-
tion and distress may be more important outcomes
than pain reduction (42), which might guide the selec-
tion of outcomes in future trials. Future studies also
should examine issues of durability (for example,
whether “booster” groups after treatment enhance long-
term outcome) and sustainability (such as identifying
and facilitating reliable mechanisms for reimbursement).
In conclusion, literacy-adapted group CBT and
comprehensive group EDU promoting pain self-
Annals.org
Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018
ORIGINAL RESEARCH
management showed significant improvements in pain
intensity and physical function scores compared with
usual care among low-income adults with chronic pain.
As our society struggles with health care costs and the
opioid crisis, these findings show promise for the safe,
effective, and efficient treatment of chronic pain.
From University of Alabama, Tuscaloosa, Alabama (B.E.T.,
J.C.E., B.P.V., C.A.T., A.K.N., P.R.B., M.M.M.); Rush University
Medical Center, Chicago, Illinois (J.W.B.); Ohio State Univer-
sity, Columbus, Ohio (M.K., J.S.C.); East Carolina University,
Greenville, North Carolina (L.C.C.); University of Alabama,
Tuscaloosa, Alabama, and Integrated Psychology Solutions,
Little Rock, Arkansas (B.A.); Arizona Emergency Medicine Re-
search Center, Tucson, Arizona (B.J.B.); Whatley Health Ser-
vices, Tuscaloosa, Alabama (R.B., T.T.B., C.S.E., D.H.T.);
University of Alabama, Tuscaloosa, Alabama, and Pacific Re-
habilitation Centers, Bellevue, Washington (C.M.D.); Univer-
sity of Alabama, Tuscaloosa, Alabama, and CHI Franciscan
Medical Group, Tacoma, Washington (W.D.D.); University of
California, Los Angeles, Los Angeles, California (M.J.); Univer-
sity of Alabama, Tuscaloosa, and University of Alabama at Bir-
mingham, Birmingham, Alabama (T.P.); and Virginia Com-
monwealth University, Richmond, Virginia (L.J.S.).
Disclaimer: The statements presented in this work are solely
the responsibility of the authors and do not necessarily repre-
sent the views of PCORI, its board of governors, or the meth-
odology committee.
Acknowledgment: The authors thank the many persons who
assisted with this study, particularly Madison Anzelc, Jacob
Baxter, Julie Cunningham, Laura M. Daniels, Julia Dodd,
Courtney Harhay, Tyler Jones, Wesley Korfe, Shweta Kapoor,
Megan Lyons, Hylton Molzof, Ian Sherwood, and Bryan Valladares.
Financial Support: Funded partially by a PCORI Research
Award (contract 941) and partially by the University of Alabama.
Disclosures: Dr. Thorn reports grants from PCORI and indirect
cost recovery for research expenses from the University of
Alabama during the conduct of the study and personal fees
from Guilford Publications outside the submitted work. Drs.
Eyer and Burns, Mr. Van Dyke, Ms. Newman, and Mr. Penn
report grants from PCORI during the conduct of the study. Dr.
Campbell reports grants from the University of Alabama and
PCORI during the conduct of the study. Dr. Cheavens reports
personal fees from the University of Alabama during the con-
duct of the study. Authors not named here have disclosed no
conflicts of interest. Disclosures can also be viewed at www
.acponline.org/authors/icmje/ConflictOfInterestForms.do
?msNum=M17-0972.
Reproducible Research Statement: Study protocol: See Sup-
plement (available at Annals.org). Statistical code: Not avail-
able. Data set: Available from Dr. Thorn (e-mail, bthorn@ua
.edu).
Requests for Single Reprints: Beverly E. Thorn, PhD, University
of Alabama, 348 Gordon Palmer Hall, Tuscaloosa, AL 35487;
e-mail, bthorn@ua.edu.
Current author addresses and author contributions are avail-
able at Annals.org.
Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 479
 ORIGINAL RESEARCH
References
1. Institute of Medicine. Relieving Pain in America: A Blueprint for
Transforming Prevention, Care, and Research. Washington, DC: Na-
tional Academies Pr; 2011.
2. Smedley BD, Stith AY, Nelson AR. Unequal Treatment: Confront-
ing Racial and Ethnic Disparities in Health Care. Washington, DC:
National Academies Pr; 2002.
3. Gaskin DJ, Richard P. The economic costs of pain in the United
States. J Pain. 2012;13:715-24. [PMID: 22607834] doi:10.1016/j.jpain
.2012.03.009
4. Green CR, Anderson KO, Baker TA, et al. The unequal burden of
pain: confronting racial and ethnic disparities in pain. Pain Med.
2003;4:277-94. [PMID: 12974827]
5. Tait RC, Chibnall JT. Racial and ethnic disparities in the evaluation
and treatment of pain: psychological perspectives. Prof Psychol Res
Pr. 2005;36:595-601.
6. Green CR, Ndao-Brumblay SK, West B, Washington T. Differences
in prescription opioid analgesic availability: comparing minority and
white pharmacies across Michigan. J Pain. 2005;6:689-99. [PMID:
16202962]
7. Hanson C. Care of clients in rural settings. In: Clarke M, ed. Nurs-
ing in the Community. Stamford, CT: Appleton & Lange; 1990.
8. McWilliams LA, Cox BJ, Enns MW. Mood and anxiety disorders
associated with chronic pain: an examination in a nationally repre-
sentative sample. Pain. 2003;106:127-33. [PMID: 14581119]
9. Dowell D, Haegerich TM, Chou R. CDC guideline for prescribing
opioids for chronic pain—United States, 2016. JAMA. 2016;315:
1624-45. [PMID: 26977696] doi:10.1001/jama.2016.1464
10. VA/DoD; Opioid Therapy for Chronic Pain Work Group. VA/DoD
clinical practice guideline for opioid therapy for chronic pain. 2017.
Accessed at www.healthquality.va.gov/guidelines/Pain/cot on 2 Au-
gust 2017.
11. Qaseem A, Wilt TJ, McLean RM, Forciea MA; Clinical Guidelines
Committee of the American College of Physicians. Noninvasive
treatments for acute, subacute, and chronic low back pain: a clinical
practice guideline from the American College of Physicians. Ann
Intern Med. 2017;166:514-30. [PMID: 28192789] doi:10.7326
/M16-2367
12. Aggarwal VR, Lovell K, Peters S, Javidi H, Joughin A, Goldthorpe
J. Psychosocial interventions for the management of chronic orofa-
cial pain. Cochrane Database Syst Rev. 2011:CD008456. [PMID:
22071849] doi:10.1002/14651858.CD008456.pub2
13. Ehde DM, Dillworth TM, Turner JA. Cognitive-behavioral therapy
for individuals with chronic pain: efficacy, innovations, and directions
for research. Am Psychol. 2014;69:153-66. [PMID: 24547801] doi:10
.1037/a0035747
14. Hoffman BM, Papas RK, Chatkoff DK, Kerns RD. Meta-analysis of
psychological interventions for chronic low back pain. Health Psy-
chol. 2007;26:1-9. [PMID: 17209691]
15. Thorn BE, Day MA, Burns J, et al. Randomized trial of group
cognitive behavioral therapy compared with a pain education con-
trol for low-literacy rural people with chronic pain. Pain. 2011;152:
2710-20. [PMID: 21920668] doi:10.1016/j.pain.2011.07.007
16. Freedland KE, Mohr DC, Davidson KW, Schwartz JE. Usual and
unusual care: existing practice control groups in randomized con-
trolled trials of behavioral interventions. Psychosom Med. 2011;73:
323-35. [PMID: 21536837] doi:10.1097/PSY.0b013e318218e1fb
17. Eyer JC, Thorn BE. The Learning About My Pain study protocol:
reducing disparities with literacy-adapted psychosocial treatments
for chronic pain, a comparative behavioral trial. J Health Psychol.
2016;21:2063-74. [PMID: 25712491] doi:10.1177/1359105315
570985
18. Kuhajda MC, Thorn BE, Gaskins SW, Day MA, Cabbil CM. Liter-
acy and cultural adaptations for cognitive behavioral therapy in a
rural pain population. Transl Behav Med. 2011;1:216-23. [PMID:
24073046] doi:10.1007/s13142-011-0026-2
19. Thorn BE, Pence LB, Ward LC, et al. A randomized clinical trial
of targeted cognitive behavioral treatment to reduce catastro-
phizing in chronic headache sufferers. J Pain. 2007;8:938-49.
[PMID: 17690017]
480 Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018
Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018
Literacy-Adapted Psychosocial Treatments for Pain
20. Ehde D, Jensen M, Engel J, Hanley M, Raichle K, Osborne T.
Education Intervention Therapist Manual: Project II: Role of Catastro-
phizing in Chronic Pain. Seattle: Univ Washington; 2005.
21. Barber JP, Liese BS, Abrams MJ. Development of the cognitive
therapy adherence and competence scale. Psychother Res. 2003;13:
205-21.
22. Cleeland CS, Ryan KM. Pain assessment: global use of the Brief
Pain Inventory. Ann Acad Med Singapore. 1994;23:129-38. [PMID:
8080219]
23. Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief
depression severity measure. J Gen Intern Med. 2001;16:606-13.
[PMID: 11556941]
24. Farrar JT, Young JP Jr, LaMoreaux L, Werth JL, Poole RM. Clini-
cal importance of changes in chronic pain intensity measured on an
11-point numerical pain rating scale. Pain. 2001;94:149-58. [PMID:
11690728]
25. Wilkinson GS, Robertson GJ. Wide Range Achievement Test 4
(WRAT4). Lutz, FL: Psychological Assessment and Resources; 2006.
26. Hedges LV. Effect sizes in cluster-randomized designs. J Educ
Behav Stat. 2007;32:341-70.
27. Cohen J. Statistical Power Analysis for the Behavioral Sciences.
2nd ed. Hillsdale, NJ: Lawrence Earlbaum Associates; 1988.
28. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a
practical and powerful approach to multiple testing. J R Stat Soc
Series B Stat Methodol. 1995:289-300.
29. Dworkin RH, Turk DC, Wyrwich KW, et al. Interpreting the
clinical importance of treatment outcomes in chronic pain clin-
ical trials: IMMPACT recommendations. J Pain. 2008;9:105-21.
[PMID: 18055266]
30. Enders CK. Applied Missing Data Analysis. New York: Guilford
Pr; 2010.
31. Schafer JL, Graham JW. Missing data: our view of the state of the
art. Psychol Methods. 2002;7:147-77. [PMID: 12090408]
32. Muthén LK, Muthen BO. Mplus User's Guide. Los Angeles:
Muthén & Muthén; 2010.
33. Demirtas H, Schafer JL. On the performance of random-
coefficient pattern-mixture models for non-ignorable drop-out. Stat
Med. 2003;22:2553-75. [PMID: 12898544]
34. Little RJ. Modeling the drop-out mechanism in repeated-
measures studies. J Am Stat Assoc. 1995;90:1112-21
35. Muthén B, Asparouhov T, Hunter AM, Leuchter AF. Growth mod-
eling with nonignorable dropout: alternative analyses of the
STAR*D antidepressant trial. Psychol Methods. 2011;16:17-33.
[PMID: 21381817] doi:10.1037/a0022634
36. Farrar JT, Dworkin RH, Max MB. Use of the cumulative propor-
tion of responders analysis graph to present pain data over a range
of cut-off points: making clinical trial data more understandable.
J Pain Symptom Manage. 2006;31:369-77. [PMID: 16632085]
37. Williams AC, Eccleston C, Morley S. Psychological therapies for the
management of chronic pain (excluding headache) in adults. Cochrane
Database Syst Rev. 2012;11:CD007407. [PMID: 23152245] doi:10
.1002/14651858.CD007407.pub3
38. Kolodny A, Courtwright DT, Hwang CS, et al. The prescription
opioid and heroin crisis: a public health approach to an epidemic
of addiction. Annu Rev Public Health. 2015;36:559-74. [PMID:
25581144] doi:10.1146/annurev-publhealth-031914-122957
39. Bruns D, Mueller K, Warren PA. Biopsychosocial law, health care
reform, and the control of medical inflation in Colorado. Rehabil Psy-
chol. 2012;57:81-97. [PMID: 22686548] doi:10.1037/a0028623
40. Ersek M, Turner JA, Cain KC, Kemp CA. Chronic pain self-
management for older adults: a randomized controlled trial
[ISRCTN11899548]. BMC Geriatr. 2004;4:7. [PMID: 15285783]
41. Turner JA, Mancl L, Aaron LA. Short- and long-term efficacy of
brief cognitive-behavioral therapy for patients with chronic temporo-
mandibular disorder pain: a randomized, controlled trial. Pain. 2006;
121:181-94. [PMID: 16495014]
42. Ballantyne JC, Sullivan MD. Intensity of chronic pain—the wrong
metric? N Engl J Med. 2015;373:2098-9. [PMID: 26605926] doi:10
.1056/NEJMp1507136
Annals.org
 Current Author Addresses: Dr. Thorn, Mr. Van Dyke, Ms. Tor-
res, Ms. Newman, Ms. Block, and Mr. Mulla: University of Ala-
bama, 348 Gordon Palmer Hall, Tuscaloosa, AL 35487.
Dr. Eyer: University of Alabama, 650 University Boulevard
East, Tuscaloosa, AL 35401.
Dr. Burns: Rush University Medical Center, 1645 West Jackson
Boulevard, Suite 400, Chicago, IL 60612.
Dr. Kim: Ohio State University, 127 Arps Hall, 1945 North High
Street, Columbus, OH 43210.
Dr. Campbell: East Carolina University, Rawl Building Room
104, Greenville, NC 27858.
Dr. Anderson: 1303 Stagecoach Village, Little Rock, AR
72210.
Dr. Bobrow: University of Arizona, Department of Emergency
Medicine, 1501 North Campbell Avenue, P.O. Box 245057,
Tucson, AZ 85724.
Ms. Brooks, Dr. Burton, Ms. Edwards, and Ms. Tucker: Whatley
Health Services, 2731 Martin Luther King Boulevard, Tusca-
loosa, AL 35401.
Dr. Cheavens: Ohio State University, 147 Psychology Building,
Columbus, OH 43210.
Dr. C.M. DeMonte: Pacific Rehabilitation Centers, 126 15th
Street Southeast, Puyallup, WA 98372.
Dr. W.D. DeMonte: Franciscan Medical Pavilion, 16045 1st Av-
enue South, Burien, WA 98148.
Dr. Jeong: University of California, Los Angeles, 3141 Moore
Hall, 457 Portola Avenue, Los Angeles, CA 90024.
Mr. Penn: University of Alabama at Birmingham, 201 Camp-
bell Hall, Birmingham, AL 35294.
Ms. Smith: 806 West Franklin Street, Box 842018, Richmond,
VA 23284.
Annals.org
Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018
Author Contributions: Conception and design: B.E. Thorn,
J.C. Eyer, B.P. Van Dyke, C.A. Torres, J.W. Burns, L.C. Camp-
bell, P.R. Block.
Analysis and interpretation of the data: B.E. Thorn, J.C. Eyer,
B.P. Van Dyke, C.A. Torres, J.W. Burns, M. Kim, A.K. Newman,
B.J. Bobrow, J.S. Cheavens, M. Jeong.
Drafting of the article: B.E. Thorn, J.C. Eyer, B.P. Van Dyke,
C.A. Torres, J.W. Burns, A.K. Newman, L.C. Campbell,
J.S. Cheavens.
Critical revision for important intellectual content: B.E. Thorn,
J.C. Eyer, B.P. Van Dyke, C. Torres, L.C. Campbell, P.R. Block,
T.T. Burton.
Final approval of the article: B.E. Thorn, J.C. Eyer, B.P. Van
Dyke, C.A. Torres, J.W. Burns, M. Kim, A.K. Newman, L.C.
Campbell, B. Anderson, P.R. Block, B.J. Bobrow, R. Brooks,
T.T. Burton, J.S. Cheavens, C.M. DeMonte, W.D. DeMonte,
C.S. Edwards, M. Jeong, M.M. Mulla, T. Penn, L.J. Smith, D.H.
Tucker.
Provision of study materials or patients: B.E. Thorn, J.C. Eyer,
B. Anderson, T.T. Burton.
Statistical expertise: J.C. Eyer, B.P. Van Dyke, M. Kim, M.
Jeong.
Obtaining of funding: B.E. Thorn, J.C. Eyer.
Administrative, technical, or logistic support: B.E. Thorn, J.C.
Eyer, C.A. Torres, L.C. Campbell, B. Anderson, P.R. Block, T.T.
Burton, C.M. DeMonte, W.D. DeMonte, M.M. Mulla, T. Penn,
D.H. Tucker.
Collection and assembly of data: B.P. Van Dyke, B. Anderson,
P.R. Block, R. Brooks, T.T. Burton, C.M. DeMonte, W.D. De-
Monte, C.S. Edwards, L.J. Smith.
Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018
 Appendix Table 1. IRB Modifications Throughout the LAMP Trial

Approval Description of Submission (Modification/Renewal)

Date
10/15/12 Personnel change to pilot protocol
2/13/13 Initial modification to pilot protocol
7/12/13 Renewal of pilot protocol but with major changes to use it for LAMP trial. From this point forward this is the official approval date of the LAMP
protocol. The LAMP IRB protocol differs from the ClinicalTrials.gov entry, the published protocol, and the current manuscript in terms of
specification of primary and secondary outcomes. The LAMP IRB protocol enumerated multiple primary and secondary outcomes. Prior to
collecting any outcome data, a decision was made to use a single primary outcome, which is a more rigorous design, and was specified in
the ClinicalTrials.gov entry, the published protocol, and the current manuscript. The IRB document was not subsequently revised to narrow
the primary outcome to one variable because the changes did not alter patient burden and because all of the previously approved
outcome variables were still collected for ancillary future analyses. Subsequent IRB revisions (see below) were made each time a measure
was added (due to increased participant burden), and each time a measure was revised or replaced with a different measure (see below)
8/14/13 Changes to exclusion criteria (reduce required reading level from fifth to first grade)
10/16/13 Personnel changes
12/17/13 Updates to Informed Consent, specification of referral process, addition of posttreatment qualitative interviews
12/19/13 Addition of Pain Conceptualization Questionnaire and Pain Coping Questionnaire (exploratory measures)
2/14/14 Personnel changes
4/3/14 Personnel changes
6/13/14 Renewal with changes to Informed Consent, addition of suicidality assessment measure, modifications of wording of qualitative interviews;
removed 12-mo follow-up; added measures of weekly affect questionnaires; addition of exploratory measures to be added mid, post,
follow-up assessments
8/1/14 Added patient consent to use specific video clips for training purposes
9/29/14 Personnel changes
10/2/14 Addition of questions to Working Alliance Questionnaire. Replaced MINI with more extensive interview
10/3/14 Update of Pain Conceptualization Questionnaire (exploratory measure) wording and questions
11/12/14 Personnel changes
12/22/14 Replaced Mood Disorders Questionnaire with WHO CIDI bipolar questionnaire
4/24/15 Personnel changes
5/18/15 Personnel changes
6/2/15 Personnel changes
6/12/15 Renewal without modifications
8/25/15 Personnel changes
4/20/16 Personnel changes
6/15/16 Renewal without modifications
4/25/17 Closure
CIDI = Composite International Diagnostic Interview; IRB = Institutional Review Board; LAMP = Learning About My Pain; MINI = Mini-International
Neuropsychiatric Interview; WHO = World Health Organization.

Appendix Table 2. Summary of Treatment Integrity Process for the LAMP Protocol*

Treatment Condition Treatment Fidelity

Mean Quality SD ⴞSE Mean Adherence SD ⴞSE Occurrences of Protocol

Rating Rating Deviation, %
CBT 2.75 0.36 0.07 100.00 0 0 0
EDU 2.53 0.33 0.6 97.00 5.57 1.02 3 (10)
CBT = cognitive behavioral therapy; EDU = pain education; LAMP = Learning About My Pain.
* A total of 63 sessions (30%) were randomly selected for review; however, 7 sessions were not rated because of technical difficulties (e.g., audio
unavailable due to technical issues at remote site). In total, 56 sessions (26.7%; 26 CBT and 30 EDU) were successfully reviewed and scored
for adherence, therapist competence, and quality of treatment. Any protocol deviations were also noted for each session reviewed. The Therapy
Adherence and Competence Scale is an adapted version of the Cognitive Adherence and Competence Scale (21) and was used to measure
adherence, therapist competence, and quality of treatment. Three sessions were randomly selected from each group cohort and rated on quality
on the basis of common therapeutic factors (e.g., co-interventionist alliance) and coverage of unique elements specific to each session and
treatment condition (e.g., Assertive Communication Skills Training— unique to session 5 in CBT). Each session received a quality rating based on a
4-point scale with end points ranging from “0 = poor to 3 = excellent.” Quality rating scores indicate that average ratings of both conditions were
in the good to excellent range. Adherence ratings were provided on the basis of demonstrated delivery of unique elements and treatment
components for each session. Adherence rating scores in EDU indicate that there were several instances in which interventionists did not demon-
strate coverage of a specific element for that session. Specifically, in 5 EDU sessions, coverage of a specific element was not demonstrated, the most
common missing element being “did not encourage participants to review materials during the coming week.” From the sessions reviewed, 3
incidents of protocol deviations occurred in EDU. These protocol deviations included 1 instance of a therapist discussing a unique element from
CBT in an EDU session (e.g. behavioral pacing), 1 instance of a therapist providing the wrong audio CD at the end of one EDU session, which
included a relaxation exercise, and 1 instance of 3 participants in the same EDU cohort receiving a faulty EDU manual that included a section of CBT
material. In the latter 2 cases, the incorrect audio CDs and the workbooks were replaced at the beginning of the next session. Of note, each protocol
deviation occurred at a different session and for different group cohorts.

Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 Annals.org

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Appendix Table 3. Effect Size Estimates (Hedges ␦T) of Baseline to Posttreatment and Posttreatment to 6-Month Differences in
Change Scores for CBT and EDU vs. UC and Associated 95% CIs*

Variable Within-Group Baseline to Within-Group Posttreatment to 6-Month

Baseline to Posttreatment Posttreatment to Change vs. UC (95% CI)
Posttreatment Change vs. UC 6-Month Change
Change (95% CI) (95% CI) (95% CI)
Pain intensity by BPI-Intensity score
UC –0.13 (–0.30 to 0.04) – –0.12 (–0.30 to 0.05) –
CBT –0.53 (–0.82 to –0.24) –0.40 (–0.74 to –0.06) 0.18 (–0.02 to 0.38) 0.30 (0.03 to 0.57)
EDU –0.41 (–0.57 to –0.21) –0.28 (–0.52 to –0.05) 0.07 (–0.10 to 0.29) 0.19 (–0.05 to 0.44)

Physical function by BPI-Interference score

UC –0.12 (–0.31 to 0.07) – 0.07 (–0.10 to 0.25) –
CBT –0.66 (–0.89 to –0.43) –0.54 (–0.84 to –0.24) 0.22 (0.04 to 0.40) 0.14 (–0.11 to 0.39)
EDU –0.40 (–0.55 to –0.25) –0.28 (–0.52 to –0.04) 0.10 (–0.07 to 0.27) 0.03 (–0.22 to 0.27)

Depression by PHQ-9 score

UC –0.17 (–0.33 to –0.02) – –0.04 (–0.13 to 0.06) –
CBT –0.38 (–0.58 to –0.19) –0.21 (–0.48 to 0.06) 0.10 (–0.08 to 0.27) 0.13 (–0.11 to 0.37)
EDU –0.35 (–0.55 to –0.15) –0.18 (–0.43 to 0.06) 0.07 (–0.07 to 0.20) 0.11 (–0.03 to 0.24)
BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; PHQ-9 = Patient Health Questionnaire–9; UC = usual care.
* To compute effect sizes for the linear growth models, Hedges ␦T (26) (derived for nested designs) was calculated for each effect estimate and
interpreted using Cohen's d (27) (small [0.2], medium [0.5], and large [0.8]). Negative values (i.e., lower or decreased scores) indicate improvement
or advantage on the specified variable. All analyses were adjusted to control for sex, minority status, years of education, and pain duration.

Appendix Table 4. Comparison of Change Estimates for Pain Intensity (BPI-Intensity Score) and Physical Function
(BPI-Interference Score) to Minimally Important Change Criteria*

Variable UC CBT EDU

BPI-Intensity score†
Change pre- to posttreatment –4.00 (–9.31 to 1.29) –16.28‡ (–25.32 to –7.23) –12.74‡ (–17.58 to –6.35)
Change pretreatment to 6-mo follow-up –7.75 (–18.46 to 2.94) –10.83‡ (–26.08 to 4.42) –10.58‡ (–20.69 to 2.71)

BPI-Interference score§
Change pre- to posttreatment –4.58 (–11.68 to 2.55) –24.75‡ (–33.48 to –16.01) –15.18‡ (–21.00 to –9.36)
Change pretreatment to 6-mo follow-up –1.80 (–15.47 to 11.86) –16.64‡ (–32.18 to –1.12) –11.39 (–23.82 to 1.05)
BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; UC = usual care.
* Values are percentages (95% CIs). These percentages were calculated by dividing the within-condition change estimates by the baseline scores.
For pre– 6-mo follow-up, the within-condition change estimates for pre- to posttreatment and posttreatment to follow-up were added and divided
by the baseline score.
† IMMPACT (Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials) (29) criteria for minimally important change for pain
intensity (BPI-Intensity score) >10%.
‡ Score exceeds guideline cutoffs.
§ IMMPACT (29) criteria for minimally important change for physical function (BPI-Interference score) >14%.

Annals.org Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Appendix Table 5. Numbers (Percentages) of the Sample
With PHQ-9 Scores Above the “Probable Depression”
Cutoff (≥10) at Baseline, Posttreatment, and 6-Month
Follow-up Compared Across Treatment Conditions

Depression by PHQ-9 Score* CBT EDU UC

Baseline†
Score <10 36 (37.9) 41 (42.3) 34 (34.7)
Score ≥10 59 (62.1) 56 (57.7) 64 (65.3)

Pre- to posttreatment‡
Score <10 24 (46.2) 24 (46.2) 11 (21.9)
Score ≥10 28 (53.8) 28 (53.3) 41 (78.8)

Pretreatment to 6-mo follow-up§

Score <10 22 (50) 13 (31.7) 13 (27.1)
Score ≥10 22 (50) 28 (68.3) 35 (72.9)
CBT = cognitive behavioral therapy; EDU = pain education; PHQ-9 =
Patient Health Questionnaire-9; UC = usual care.
* A score ≥10 indicates the presence of probable depression (23).
† Number of participants who completed a baseline assessment (CBT,
n = 95; EDU, n = 97; and UC, n = 98).
‡ Available number at posttreatment assessment for pre- to posttreat-
ment comparison (CBT, n = 52; EDU, n = 45; and UC, n = 52).
§ Available number at 6-mo follow-up assessment for pretreatment to
6-mo follow-up comparison (CBT, n = 44; EDU, n = 41; and UC, n = 48).

Appendix Table 6. Comparison of Estimated Effects From Linear Mixed Models to Results of Sensitivity Analyses Using
Pattern-Mixture Analysis

Group Original Analyses Sensitivity Analyses

Baseline to P Value Posttreatment P Value Baseline to P Value Posttreatment P Value

Posttreatment to 6-Month Posttreatment to 6-Month
Difference vs. Difference vs. Difference vs. Difference vs.
UC (95% CI) UC (95% CI) UC (95% CI) UC (95% CI)
Pain intensity by
BPI-Intensity score
CBT –0.80 (–1.48 to –0.11) 0.022 0.60 (0.06 to 1.13) 0.028 –0.80 (–1.49 to –0.12) 0.021 0.60 (0.06 to 1.13) 0.028
EDU –0.57 (–1.04 to –0.10) 0.018 0.38 (–0.11 to .87) 0.124 –0.57 (–1.04 to –0.10) 0.017 0.38 (–0.10 to 0.87) 0.121

Physical function by
BPI-Interference score
CBT –1.36 (–2.11 to –0.61) <0.001 0.36 (–0.27 to 0.99) 0.261 –1.38 (–2.12 to –0.63) <0.001 0.36 (–0.27 to 0.99) 0.262
EDU –0.70 (–1.31 to –0.09) 0.024 0.07 (–0.55 to 0.68) 0.831 –0.72 (–1.32 to –0.11) 0.021 0.07 (–0.55 to 0.68) 0.829

Depression by PHQ-9
score
CBT –1.33 (–3.02 to 0.35) 0.120 0.85 (–0.69 to 2.38) 0.280 –1.31 (–2.99 to 0.38) 0.128 0.85 (–0.67 to 2.38) 0.279
EDU –1.15 (–2.71 to 0.41) 0.147 0.67 (–0.18 to 1.53) 0.123 –1.14 (–2.85 to 0.57) 0.191 0.68 (–0.75 to 2.10) 0.352
BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education; PHQ-9 = Patient Health Questionnaire–9; UC = usual care.

Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018 Annals.org

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018

 Appendix Table 7. Effects of Dropout Dummy Variable Parameters in Sensitivity Analysis Using Pattern-Mixture Modeling*

Dropout Dummy Estimated Effect on Participant-Level P Value Estimated Effect on Cohort-Level P Value
Variables Random Effect (95% CI) Random Effect (95% CI)
BPI-Intensity score
d1 0.329 (–0.179 to 0.837) 0.20 1.033 (–17.254 to 19.320) 0.91
d2 –0.038 (–0.601 to 0.525) 0.89 –1.489 (–5.875 to 2.897) 0.51

BPI-Interference score
d1 0.835 (0.245 to 1.425) 0.006 –6.097 (–16.589 to 4.395) 0.26
d2 –0.094 (–0.807 to 0.619) 0.80 –5.853 (–13.387 to 1.681) 0.128

PHQ-9 score
d1 –0.671 (–2.807 to 1.465) 0.54 –32.557 (–124.828 to 59.714) 0.49
d2 –0.431 (–2.391 to 1.529) 0.67 –30.376 (–64.274 to 3.522) 0.079
BPI = Brief Pain Inventory; PHQ-9 = Patient Health Questionnaire–9.
* Sensitivity analyses for the primary and secondary outcomes assessed possible departures from the missing at random assumption. This was
accomplished by using a simple version of pattern-mixture modeling by allowing the random-effect means at participant and cohort levels to vary
as a function of the dropout dummy variables. Two dropout dummy variables (d1 and d2) were defined such that dt indicates a participant who
drops out after time t − 1 (t = 2, 3), where t = 2 represents posttreatment and t = 3 represents 6-mo follow-up.

Appendix Figure. Cumulative proportion of responders analysis of percent change in pain intensity (BPI-Intensity) scores per
treatment group.

100

90

80

70

CBT
60 EDU
Participants, %

Usual
50 care

40

30

20

10

0
0 10 20 30 40 50 60 70 80 90 100
Decrease in BPI-Intensity Score From Pre- to Posttreatment, %

BPI = Brief Pain Inventory; CBT = cognitive behavioral therapy; EDU = pain education.

Annals.org Annals of Internal Medicine • Vol. 168 No. 7 • 3 April 2018

Downloaded From: https://annals.org/ by a Kaohsiung Medica University User on 07/25/2018