An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Carol Greco, PhD, Lan Yu, PhD, and Paul Polonis, PhD.
Author Information and AffiliationsNonspecific, contextual factors are seldom measured in research trials or clinics, yet they can influence outcomes. This project evaluated the Healing Encounters and Attitudes Lists (HEAL) and PROMIS measures during pain treatment. HEAL includes Patient-Provider Connection, Healthcare Environment perceptions, Treatment Expectancy (TEX), Positive Outlook (POS), Spirituality (SPT), and Attitudes Toward Complementary/Alternative Medicine (CAM). We addressed heterogeneity of treatment effects (HTE), the methods gap, by comparing outcomes in patient subgroups: persons receiving CAM vs conventional treatments and, separately, patient subgroups who had higher and lower scores on HEAL TEX.
Specific research questions:
In this prospective observational study, persons with chronic pain starting a CAM (eg, acupuncture, chiropractic; n = 109) or a conventional medicine treatment (eg, physical therapy, medication management; n = 100) completed HEAL and PROMIS measures online at baseline and at 2 and 4 months later. They rated clinical global improvement (CGI) at follow-ups. Correlations and multiple regression were the main analytic strategies. We compared CAM and conventional patients as well as those with lower and higher HEAL TEX to evaluate HTE. A subset of patients (n = 44) and clinicians (n = 13) completed interviews on the clarity and clinical utility of HEAL, PROMIS, and ACPA Pain Log.
Baseline HEAL TEX, POS, and CAM were correlated with follow-up PROMIS Pain Intensity, and baseline HEAL TEX, POS, CAM, and SPT were correlated with PROMIS Pain Interference. All baseline HEAL scores except those for SPT were correlated with CGI at 2 and 4 months (Spearman rho, all P < .05). In final regression models, baseline HEAL TEX and baseline Pain Intensity accounted for 42% of the variance in 4-month Pain Intensity, with HEAL TEX contributing 2% additional variance over baseline Pain Intensity. Baseline HEAL CAM, SPT, and POS accounted for 2.1%, 1.7%, and 1.5%, respectively, of 4-month follow-up Pain Interference variance beyond that accounted for by baseline Pain Interference. Regarding subgroup comparisons based on receiving CAM vs conventional treatments, the conventional treatment group had higher PROMIS Average Pain, Pain Interference, and Pain Intensity; lower Physical Functioning; and poorer Overall Health compared with the CAM treatment group at all time points. The lower TEX group had higher pain and poorer health and functioning than the higher TEX group at all time points. In interviews, patients and clinicians reported HEAL and PROMIS to be relevant to their treatment, and patients made suggestions for the ACPA Pain Log.
Several HEAL measures were correlated with later PROMIS Pain treatment outcomes. Higher HEAL TEX and choosing CAM over conventional medicine were patient factors that positively influenced outcomes. Patients and clinicians found HEAL and PROMIS measures as well as the ACPA Pain Log to be potentially useful tools for enhancing communication in clinical settings.
This study included only patients with chronic pain; however, HEAL and many PROMIS measures apply to a broader range of patients and treatments.
In both health care and research, there is a growing recognition that important contributors to health and healing are the clinical context and patients' own beliefs and attitudes.1,2 The 2011 Institute of Medicine report Relieving Pain in America highlights the importance of the patient–provider bond, stating that “the effectiveness of pain treatments depends greatly on the strength of the clinician-patient relationship.” The 2011-2015 strategic plan of the NIH National Center for Complementary and Alternative Medicine (NCCAM) notes that “better understanding of the contributions of both specific and nonspecific effects influencing outcomes and the potential for insight into exploitation of either or both to improve symptom management or general health and well-being is needed.” Nonspecific aspects of treatment include factors such as patients' beliefs and expectations, patients' sense of connection or trust in the provider, and optimism and spirituality. These nonspecific, or contextual, factors may interact with the direct effects of various types of treatments, such as medication adjustment or physical therapy.
Historically, such complex and interactive factors have been unexamined or relegated to the “black box” of placebo response.1,3 More recently, however, patient-centered and contextual factors have been the subject of investigation, particularly in studies of complementary and alternative medicine (CAM). Factors that have been found to influence outcomes include patients' beliefs or expectations about treatment,4-6 patients' perceptions of the bond with a provider,7 negative affect,8,9 and spirituality.10-12 For example, 2 studies of openly labeled placebos support the role of patient beliefs and expectations. A study of more than 80 chronic low back pain patients assigned to treatment as usual (TAU) or to openly labeled placebo pills that included a statement about mind–body influences found statistically significant improvements in pain and pain-related disability in the placebo group compared with TAU.13 A similarly designed study with irritable bowel syndrome patients showed that open-label placebo resulted in reduced symptom severity (P = .03) and adequate symptom relief (P = .03) compared with an evaluation-only control group.14 Furthermore, in a randomized, double-blind placebo-controlled trial for gastroesophageal reflux disease, those randomly assigned to receive an expanded patient–provider interaction (including questions about lifestyle and symptom details) were more likely to have decreased gastroesophageal reflux disease (GERD) severity (P = .01) and were significantly more likely to report a 50% or greater reduction in GERD symptoms than those who received the nonexpanded visit (P = .01).15
Within-patient (spirituality, outlook) and contextual factors (treatment expectations, patient-provider connection) are often referred to as nonspecific factors, as they are not specific to a particular treatment. Although nonspecific factors are known to be important in health outcomes, the issue remains of how to measure these factors in a standardized and systematic way. Existing questionnaires are often long, difficult for low literacy patients to understand, and disease- or treatment-specific and thus of little use in comparative effectiveness research or general medical settings. To address this gap, our team developed a set of item banks to measure nonspecific factors in healing, the Healing Encounters and Attitudes Lists (HEAL) (NCCAM R01 AT006453, The Healing Context in CAM: Instrument Development and Initial Validation), using the rigorous methodology of the PROMIS instrument. PROMIS has developed and calibrated item banks assessing a wide variety of domains16 and represents the most ambitious attempt to date to apply models from item response theory to health-related assessment.17,18 PROMIS tools confer 2 major advantages: They are standardized to be comparable across diseases, treatments, and patients of various ages and literacy levels, and they can be presented as computerized adaptive tests (CATs), in which items are presented to test takers based on their responses to previously presented items. This reduces the number of items (typically 3-6) needed to determine a patient's score, thus minimizing patient burden. The efficiency of the PROMIS and HEAL measures makes it feasible to generate a rich health status profile and “snapshot” of the current treatment context at little cost.
We chose to evaluate the HEAL and PROMIS measures in the context of treatment for chronic pain because pain is a widespread problem associated with considerable personal and societal costs.2
Research question 1: Do nonspecific factors, as assessed by our HEAL measures, predict treatment outcomes?
We hypothesized that HEAL scores would predict significant variance in outcome on measures such as global impression of improvement, pain intensity, and pain interference with life activities.
Research question 2: Do the HEAL and PROMIS health status measures contribute to an improved understanding of the heterogeneity of patients and treatment effects?
Our person-centered approach was to compare patients based on their preferences for CAM vs conventional medicine. In addition, we wanted to evaluate the role of expectations of patients about their treatment, comparing patients with lower and higher positive Treatment Expectancy (TEX). We hypothesized that those with higher expectations of treatment would have greater symptom reduction than those with lower expectations.
Research question 3: Do patients and providers find the information provided by HEAL and PROMIS measures to be clear, relevant, and useful in the treatment setting? Do patients find the American Chronic Pain Association (ACPA) Pain Log clear and helpful as a communication tool in treatment?
To address these questions, we conducted cognitive interviews with patients and their providers. We predicted that patients and clinicians would endorse the HEAL instruments, the PROMIS health status instruments, and the ACPA Pain Log as useful tools to improve the quality of care.
This project, Measuring the Context of Healing: Using PROMIS in Chronic Pain Treatment, focuses on PROMIS health status measures of Pain Intensity (PINT) and Pain Interference (PI), Physical Functioning (PHYS), Sleep, Depression, and Anxiety (ANX), and a set of new instruments developed using PROMIS methods, HEAL. The HEAL measures include TEX, Patient-Provider Connection (PPC), Healthcare Environment perceptions (HCE), Positive Outlook (POS), Spirituality (SPT) and Attitude Toward CAM (CAM). Regarding the patient-centered questions that inspire patient-centered outcomes research (PCOR), 2 are most relevant to this study: “Given my personal characteristics, conditions, and preferences, what should I expect will happen to me?” and “What can I do to improve the outcomes that are most important to me?” The HEAL measures and other PROMIS health status instruments allow us to document patient-centered factors that may play a role in predicting outcomes of treatment for chronic pain.19-22 Our project also sought to inform the methodology gap identified by PCORI: heterogeneity of treatment effects (HTE). In this project, we used HEAL CATs and the selection of PROMIS health status CATs to better characterize individual differences among patients and their views of clinicians and health care settings that contribute to HTE. An equally important goal was to capture the patient voice in the development of measures. In the development of the HEAL measures (NIH R01 AT006453), 6 patient focus groups had helped identify important concepts in healing, and a diverse group of 48 community members participated in cognitive interviews to determine clarity and content validity of items. The qualitative review and cognitive interviewing led to enhanced ease of use, greater simplicity and clarity, and decreased literacy demands. Because PROMIS and HEAL CATs are relevant across medical conditions and treatments, including conventional medicine and integrative medicine, we expect the current project to have broad impact on both research and clinical treatment. When used clinically, HEAL instruments, in conjunction with other PROMIS measures, are expected to enhance treatment relationships, provide information about patients' trajectories of healing, and promote understanding of HTE.
Patient and clinician stakeholders were involved in this study from its inception. As we were creating the study proposal, we interviewed patients about study questions and outcomes that were important to them. This highlighted the significant outcomes of pain reduction and improved function, and the importance of clinicians' understanding of patients' preferences and everyday environment. We collaborated with patient advocacy organizations (ACPA and Chronic Pain Research Consortium) and integrative and conventional pain clinicians to shape the questions and methods.
Through these initial collaborations, we identified a set of stakeholders to serve on our patient/stakeholder advisory panel (PSAP). Our PSAP included 2 patients, a patient advocacy organization stakeholder, a pain research advocacy group organizer who is also a patient, and 4 clinician stakeholders. The PSAP met quarterly face to face as well as individually with researchers or in subcommittees, as needed, to work on activities such as the following:
In addition to engaging with our PSAP, we engaged with patients and clinicians during in-depth cognitive interviews. We conducted cognitive interviews with 44 patients and 13 providers to determine their views on whether HEAL and PROMIS can help patients and providers communicate and engage more fully in their treatment partnerships. We used feedback from the patient interviews to modify the HEAL and PROMIS reports. We also asked patients for feedback about the clarity and potential utility of the ACPA Pain Log items; their suggestions will be shared with the ACPA.
The research questions include: (1) Do nonspecific factors, as assessed by our HEAL measures, predict treatment outcomes? (2) Do the PROMIS health profile and HEAL measures contribute to the understanding of HTE? and (3) Do patients and their providers find the HEAL and PROMIS measures easy to understand and useful and relevant to their ongoing treatment? Likewise, do patients consider the ACPA Pain Log to be clear and useful as a communication tool in treatment?
This PROMIS-related project is a prospective cohort study.
Because our intention was to evaluate PROMIS and HEAL measures in clinical treatment for chronic pain, we invited patients who were starting a new treatment to complete the HEAL and PROMIS computerized assessments. Participants completed the online questionnaires at 3 time points: baseline (within 1 month of starting treatment), 2 months after baseline, and 4 months after baseline. For this study patients needed to have some experience with their provider and some knowledge about the treatment so that they could rate these areas on HEAL TEX and PPC. Therefore, the baseline was early in treatment rather than before treatment.
Participants received $30 for completing each of the 3 online assessments. Participants in Pittsburgh were also invited to complete individual cognitive interviews, in which they “think aloud” while reviewing HEAL, PROMIS, and ACPA Pain Log items, and rate items and summary reports. Participants and their clinicians, many of whom also completed individual interviews, received $40 for each interview. Enrollment began on July 10, 2015, and continued through August 30, 2016. The final online assessments were completed by December 9, 2016.
For recruitment, we used several methods for identifying potentially eligible and interested individuals. We routinely spoke with practicing pain clinicians about the study and supplied clinics with our flyers and information. In addition, clinicians at the Center for Integrative Medicine in Pittsburgh introduced interested persons to our research staff directly, in person. We regularly advertised electronically through university-wide email delivery systems (Read Green) and on electronic message boards in clinics. We utilized the services of the University of Pittsburgh's Clinical and Translational Science Institute's (CTSI) research registry. The CTSI research registry performed initial screening and provided contact information for potentially eligible participants. In addition, our stakeholder panel members helped by placing our recruitment flyers at their doctors' offices, gyms, and other public settings.
We enrolled patients who were seeking treatment in a variety of conventional medicine settings (pain clinics, primary care settings, and rehabilitation services) or integrative/CAM clinics. IRB approval was obtained from the University of Pittsburgh. Our 2 external collaborating integrative medicine clinics—Venice Family Clinic at UCLA and Allina Health System in Minneapolis—obtained IRB approval at their respective institutions.
We intentionally included a wide range of patients with many types of pain who received diverse treatments, because (1) a broad range of patients and treatments is needed to examine HTE, and (2) PROMIS and HEAL assessments are designed to apply broadly across patients and treatments. Recruitment materials were distributed at each research site. Interested patients who saw our advertisements or heard about the study at their health care offices telephoned or emailed our study staff to learn about the study and undergo screening, which required no more than 5 minutes of their time.
To be eligible to participate in the computerized assessments, patients must have had pain of at least 3 months duration and had started a pain treatment requiring at least monthly visits within the past month. They must have been 18 years of age or older and able to read and understand English as well as complete questionnaires on a computer. Individuals were ineligible if they were currently diagnosed (self-reported) with a psychotic disorder (eg, schizophrenia, schizoaffective disorder) or a bipolar disorder, or were unable to speak or read English.
Informed consent documentation as well as the main study questionnaires were completed online. Although participants were welcome to complete the assessments at our research offices or at their pain treatment sites, they could also complete them remotely. To minimize attrition, we tracked the time when second and third assessments were due and contacted participants via email and telephone, if necessary. In addition, the same research staff member contacted them across the 3 time points of the study. We found that careful monitoring and friendly connection with participants helped minimize attrition and loss to follow-up.
The schedule for assessment measures is provided in Table 1. All HEAL and PROMIS health status measures used in this project had been previously developed and validated, and were not developed as part of this study.19-24 The HEAL measures include the following: PPC, which assesses patient views of the relationship with the provider; TEX, which assesses patient expectations about whether the treatment will be helpful; HCE, which assesses patient perceptions of the health care provider's office and staff; POS, which assesses patient level of confidence and optimism in general; CAM, which assesses patient views about integrative medicine or CAM; and SPT, which assesses patient spiritual beliefs and experience of spiritual support. The HEAL responses are scored on a 1-to-5 scale representing frequency (“never” to “almost always”) or intensity (“not at all” to “very much”). Within each scale, raw scores are summed and converted to T scores with mean of 50 and SD of 10.
Assessment Measures.
PROMIS health status measures in the assessments were Depression (DEP), ANX, Fatigue (FAT), Sleep Disturbance (SLP), Alcohol Use, PHYS, PINT, and PI. The PROMIS measures listed previously are, like the HEAL scales, scored on 1-to-5 scales of frequency or intensity, and raw scores are summed and converted to T scores with mean of 50 and SD of 10. In addition, we collected PROMIS Average Pain (AvePoint) rating, which is a single item rated from none (0) to very severe (9-10), and PROMIS Overall Health (Health), a single item rating from excellent (1) to poor (5).
We administered the HEAL and PROMIS scales as CATs—except HEAL CAM, which is available only as a 6-item scale. CAT administration means that items are presented to the patient based on his or her responses to previous items; this reduces the number of items that a patient needs to answer.
The outcome measures for aim 1 were PROMIS PINT and PI, and the single-item rating of Clinical Global Improvement (CGI),25 which asks the patient to report his or her current symptoms compared with when he or she began treatment (eg, ranging from much worse to unchanged to much better). For aim 2, which evaluated heterogeneity of patient and treatment effects based on type of treatment (CAM or conventional) and, in separate subgroup analyses, HEAL treatment expectations, the outcome measures were PROMIS PINT, PI, PHYS, overall health, and AvePoint.
All measures used in this study were patient-reported outcomes chosen because people with chronic pain care about them. Thus, patients in treatment for chronic pain (along with their clinicians) are the best source of information for reporting about their pain, pain treatment, and the nonspecific factors assessed. We chose some of our measures, such as pain intensity, pain interference with life activities,23 and physical function based on interviews with patient stakeholders between February and April 2014.26 Our patient advocacy organization stakeholder, ACPA, has a pain log communication tool that we also included in the subsample of patients who participated in qualitative interviews, to document its clarity, ease of use, and patient perceptions of its utility. The PROMIS and HEAL measures were developed and tested using the rigorous instrument development procedures of the NIH PROMIS initiative, with validation studies published and in preparation.19,21,22
Additional measures included in the study are noted in Table 1 (PROMIS Alcohol Use, Press-Ganey Questionnaire, Opioid Prescription History). We included them based on specific input from stakeholders following the initial PCORI project submission. Patients at the conventional and integrative medicine sites completed the same online assessment protocol (through Assessment Center, the platform used by PROMIS). Each participant was assigned a study identification number. We used the identification number to link participants' data across the 3 online assessments (baseline, 2 months, and 4 months). We collected no medical record data for this study.
Eligible participants completed informed consent procedures and were administered the CAT versions of HEAL measures, a selection of PROMIS health status measures, and other demographic and clinical characteristics measures (see Tables 2 and 3).
Demographic Characteristics at Baseline, Overall, by Group, and by Low/High TEX.
Clinical Characteristics at Baseline, Overall, by Group, and by Low/High TEX.
Patients were eligible for follow-up assessments if they participated in treatment visits at least once after their baseline assessment. The reason for this requirement is that we wanted to be sure that they were actually in a treatment for their pain. We considered patients lost to follow-up if they did not complete any follow-up evaluations. Patients who did not complete the first follow-up (2 months) were still invited to complete the final (4 month) follow-up.
We conducted “think aloud,” or cognitive, interviews with a subset of patients to assess their understanding of the HEAL, PROMIS, and ACPA Pain Log questions and to determine whether they found the questions to be clear, relevant, and potentially useful for enhancing communication and understanding in clinical encounters. We set the maximum number of cognitive interviews at 25% of the overall sample (50 out of 200) based on guidelines that suggest that qualitative interviews should cease when no new information is uncovered (saturation)27 and the report that saturation may take place after approximately 20% of a sample has been interviewed.28 Participants in the assessment study were invited to participate in the cognitive interviews immediately after they completed their baseline assessment. We interviewed Pittsburgh participants and not participants from external integrative medicine sites (Los Angeles and Minneapolis) for feasibility reasons. As a part of the interview, both patients and clinicians rated the HEAL and PROMIS item banks on their understanding of the concepts, relevance of the concepts for their treatment, and usefulness of the information on a 0-to-4 ordinal scale (“not at all,” “a little bit,” “somewhat,” “quite a bit,” “very much”; see Appendix A for the cognitive interview guide and survey questions). Patients also completed the ACPA Pain Log and provided feedback on its content (at initial interview only). Patients were provided with graphic representations of their HEAL and selected PROMIS health status scores and were asked to provide feedback and suggestions to the research team.
We conducted follow-up interviews to assess patient perspectives on the personal and clinical utility of reviewing changes over time on the HEAL and PROMIS scales. The follow-up cognitive interview mirrored the baseline interview, including the same types of questions. Interviewees were encouraged to ask questions and engage in critical dialogue about the uses and implications of this type of assessment in clinical care settings. If the treatment was short term (eg, as with physical therapy, chiropractic treatment, or acupuncture), we conducted the second interview at the 2-month follow-up rather than at 4 months.
We approached all clinicians of the Pittsburgh patients who completed cognitive interviews to participate in clinician interviews. We enrolled several types of health care providers, including physicians, acupuncturists, massage therapists, and physical therapists. The clinicians were given their patients' summary scores from the online assessments and were encouraged to engage in a candid conversation about the directions that computer assessment might take in the treatment environment and what information is most beneficial for their purposes.
At follow-up interviews, the same clinicians were provided with longitudinal HEAL and PROMIS summary scores, which were briefly explained to them. The clinicians were asked to reflect on how the scores could potentially influence the clinical encounter over time. The topics mirrored those in the baseline interview, with additional attention paid to the potential value of having longitudinal health status measures available.
We predicted that correlations (Spearman's rank correlation) between HEAL scores and patients' reports of CGI at 2-month and 4-month follow-ups would be moderate or larger (≥0.50). These 2 predictions regarding HEAL and CGI at 2-month and 4-month follow-ups represent hypotheses 1a and 1b, respectively. Similarly, for hypotheses 1c and 1d (regarding HEAL and Pain outcomes), we predicted that correlations (Pearson) between HEAL scores and PROMIS PI and PINT at 2-month and 4-month follow-ups would be moderate to large (≥0.50). Regarding our hypotheses about regressions, with pain as outcomes and HEAL as predictors, we expected that HEAL scores would account for significant amounts of variance in PROMIS PINT and PI outcomes in preliminary, unadjusted models and also in final regression models that were adjusted for a baseline level of pain.
We specifically designed our study to address whether subgroups of patients experience different treatment effects. The main goals of aim 2 were to assess (1) the effect on treatment outcome of patients' choice to engage in CAM or conventional medicine, and, separately, (2) whether patients with higher and lower HEAL TEX at baseline had different pain treatment outcomes. For aim 2, we used stepwise multiple linear regression for each pain treatment outcome (ordinary least squares estimation), then used mixed models (ie, multivariate mixed effect linear regression models) to examine the cohort difference longitudinally. All predictors were fixed effects, and intercepts were random effects. For regression analyses, we examined whether HEAL scores were significant predictors of outcome in each model, with and without controlling for baseline status. For mixed models, we modeled HEAL scores at baseline, different cohorts (CAM vs conventional; high TEX vs low TEX), time, and baseline status simultaneously. In the series of mixed models, we examined (1) HEAL scores and time only; (2) HEAL scores, time, and time squared; and (3) HEAL scores, time, and baseline status. The outcome variables used in the models—and presented in Tables 8 and 9—are PROMIS AvePoint, PROMIS Overall Health, PROMIS PHYS, PROMIS PI, and PROMIS PINT. With the final models, we used backward selection by adding the significant (at P < .15) HEAL score predictors back to the model after controlling for the baseline status of the outcome variable, and time. We designed these analyses to increase understanding of the contributions of PROMIS and HEAL in the different cohorts: (1) CAM and conventional medicine, and (2) higher and lower TEX. We used the SPSS and STATA software programs for data analyses.
Multivariable Regression Final Models With Pain and Health Status Variables as Outcomes, and Treatment Type, Time, Baseline HEAL Scores, and Baseline Status as Predictors.
Multivariable Regression Final Models With Pain and Health status Variables as Outcomes, and High vs Low TEX, Time, Baseline HEAL Scores, and Baseline Status as Predictors.
For aim 3, we conducted individual cognitive interviews with patients and clinicians regarding clinical utility of the HEAL measures, PROMIS health status measures, and the ACPA Pain Log. At each of the 2 cognitive interviews (baseline and 4-month follow-up), we elicited several structured, evaluative ratings of clarity, relevance to the patients' treatment experience, and potential usefulness in the clinical setting, using 0-to-4 ordinal scales (eg, “never” to “very much”). We expected that patients and clinicians would rate the PROMIS and HEAL highly on clarity of language (ie, means ≥3.0 on a 0-to-4 scale). We expected that the patients would rate the relevance and usefulness of treatment-related HEAL items (eg, their views of the PPC, their perceptions of the Healthcare Environment, and their TEX) significantly higher than HEAL items that reflect personal characteristics (eg, POS, SPT) that may or may not be pertinent to clinical treatment. We compared mean scores on relevance of treatment-related vs within-patient HEAL measures using paired t tests.
Our data analysis plan for the aim 3 evaluation of the ACPA Pain Log included counts of comments regarding clarity and confusion related to the log's items, pictograms, response scales, and relevance to chronic pain. Thus, we obtained descriptive data during the cognitive interviews with patients with the intent to compile these data into charts with positive and negative comments and specific suggestions.
We implemented the study protocol outlined above as originally contracted, based on our original proposal and the data analytic requests made by PCORI scientific staff before the contract award. The protocol was approved by the University of Pittsburgh IRB and the IRBs of our contracted external integrative medicine sites: Venice Clinic in Los Angeles and Allina Health System in Minneapolis.
To briefly reiterate our aims, this project assessed (1) whether nonspecific factors assessed by HEAL measures predict pain treatment outcomes, (2) whether HEAL and PROMIS measures contribute to understanding of HTE, and (3) whether patients and their providers find HEAL and PROMIS measures, and the ACPA Pain Log, easy to understand and useful in their ongoing treatment.
The first 2 aims were addressed through our main study, which included 209 persons with chronic pain who were initiating a new treatment; the third question was addressed through interviews with a subset of patients and their health care providers.
Descriptions of study flow, participants, and assessments can be found in Tables 1 to 3 and Figure 1. The participants included 209 patients with chronic pain who had started a new complementary/alternative treatment (n = 109) or conventional medicine treatment (n = 100) during the previous month (Figure 1).
Participant Flow Diagram.
In terms of study flow, eligibility issues, study enrollment, and loss to follow-up, 42 eligible persons did not enroll or complete assessments. These 42 individuals were screened eligible and expressed interest in participating and were then emailed the link to enroll; however, they failed to complete enrollment after several contact attempts, apparently because they were no longer interested in participating. In addition, 3 participants completed the online consent and then withdrew without completing the questionnaires due to confusion about the technology or loss of interest. Five participants consented and then were deemed ineligible because they did not start treatment. One individual consented and completed the questionnaire but was later found to be ineligible due to an excluded psychiatric condition. See Figure 1 for details of eligibility, enrollment, and retention.
Regarding demographics, the participants' average (SD) age was 47.5 (15) years, 75% were female, 23% were non-White or multiracial, and 7% were of Hispanic ethnicity; further demographic and clinical characteristics can be found in Tables 2 and 3. CAM and conventional medicine participants did not differ on demographic characteristics (Table 2). Regarding clinical characteristics at baseline, CAM patients' PROMIS AvePoint was less than that of conventional medicine participants (P = .034).
Of our participants, 24% reported having 2 or more other medical comorbidities in addition to their chronic pain (Table 3). We also explored subgroups of patients who were using opioid medications and who were not (140 out of 209), as well as their satisfaction with these prescriptions. Only 23 out of 109 CAM patients used opioids, whereas 44 of 100 conventional medicine patients did (P = .001). Those in the high TEX group were less likely than low TEX participants to be taking opioid medications (20 out of 104 vs 47 of 105; P < .001).
Table 4 shows the variety of CAM and conventional treatments that participants received. The most common pain condition was back pain, reported by 46% (96 out of 209) of participants. The most frequently used CAM treatments were acupuncture (17.7%; 37 out of 209), chiropractic (13.9%; 29 out of 209), and massage/other body work (7.2%; 15 out of 209), and the most frequently used conventional medicine treatments were physical therapy (26.8%; 56 out of 209), medication (19.6%; 41 out of 209), and injection (7.2%; 15 out of 209). In addition to inquiring about current treatment, we asked participants to indicate number of prior treatments they had received for their chronic pain, and the vast majority—92% (192 out of 209)—had received 2 or more types of treatments previously for their pain.
Current Pain Treatment, Provider, and Pain Type, Self-reported at Baseline.
We found that the participants had improvement in PROMIS PINT and PROMIS PI from baseline to the 2-month assessment and from baseline to the 4-month assessment (all P < .001). On the CGI, 63% of participants reported improvement (“somewhat better” or “much better”) at both follow-up assessments (see Table 5). Thus, across multiple pain conditions and many types of conventional medicine and integrative medicine treatments, the participants as a group showed improvement.
HEAL, PROMIS, and CGI at Baseline, 2-Month, and 4-Month Assessments.
Our aim 1 research question addressed the issue of whether the HEAL nonspecific factors scores at baseline were associated with later improvement in (1) CGI, (2) PROMIS PINT, and (3) PROMIS PI with life activities.
All HEAL baseline scores were significantly and positively associated with CGI—except for HEAL SPT, which was not significantly associated with CGI. However, the HEAL baseline scores were not as strongly associated with CGI as we had hypothesized: The largest rho with CGI at 2 months was baseline HEAL PPC (Spearman rho = 0.257, 2-tailed P < .01), and the largest rho with 4-month CGI was baseline HEAL POS (Spearman rho = 0.252, 2-tailed P < .01; see Table 6).
Correlations Between HEAL Measures at Baseline, and CGI at 2 Months and 4 Months.
We found that several of the baseline HEAL scores were significantly associated with PROMIS PI at 2-month and 4-month follow-ups (see Table 7 for details): HEAL TEX, HEAL POS, and HEAL CAM. HEAL SPT was associated with 2-month ratings of PI but not with 4-month ratings. Baseline HEAL POS was associated with 2-month and 4-month PROMIS PINT. Baseline HEAL Attitude toward CAM and Baseline HEAL TEX were associated with 4-month PROMIS PINT but not 2-month PINT. Although the associations were statistically significant, ranging from −0.16 to −0.31, they were not of the magnitude that we had hypothesized (≥0.50). Thus, several of the HEAL measures were predictive of 2-month and 4-month pain outcomes—but modestly so.
Significant Correlations Between HEAL Measures at Baseline and PROMIS Pain Interference and Pain Intensity at 2 Months and 4 Months.
In the multiple regression analyses, 1 goal was to evaluate whether HEAL scores predicted pain outcomes in the entire group of 209 patients. As a preliminary step to the final models, we evaluated whether HEAL measures at baseline predicted pain outcomes at 2 months and 4 months, without adjusting for baseline levels of pain. In these unadjusted models, we found that HEAL POS and HEAL SPT accounted for 17.7% of the variance in PROMIS PI at 2 months, and 7.6% of the variance in PROMIS PINT at 2 months. Together, HEAL POS, HEAL SPT, and HEAL Attitudes Toward CAM accounted for 15.8% of the variance in PROMIS PI at 4 months, and HEAL POS and HEAL SPT accounted for 8.2% of the variance in PROMIS PINT at 4 months.
We adjusted the final regression models for baseline level of pain outcomes. This is important because the severity of pain at the first assessment is a confounder in the association of HEAL measures with pain outcomes. We adjusted the regression models by first entering the baseline value of the outcome (eg, PROMIS PINT). Thus, these analyses provide information about variance accounted for by HEAL over and above the influence of baseline level of pain symptoms on follow-up pain symptoms. Baseline PROMIS PI, HEAL SPT, and HEAL POS accounted for 46.7% of the variance in 2-month PROMIS PI, but most of this was due to baseline status: HEAL SPT contributed an additional 1.8% of variance beyond that accounted for by baseline PI. HEAL POS added an additional 3.7% of variance beyond the effects of baseline PI and HEAL SPT. When baseline PINT was added to the regression model predicting 2-month PROMIS PINT, we found that baseline PINT and HEAL TEX accounted for 47.5 % of the PROMIS PINT outcome at 2 months, with just 1.2% of this due to the addition of baseline HEAL TEX. The regression model predicting 4-month PROMIS PI adjusted for baseline PI resulted in a total of 46.2 % of the PROMIS PI variance predicted by baseline PI (40.9%), Attitudes Toward CAM (additional 2.1% of variance), HEAL SPT (additional 1.7% of variance), and HEAL POS (additional 1.5% of variance accounted for). The model predicting 4-month PROMIS PINT with baseline PROMIS PINT and HEAL scores as predictors resulted in a total of 41.2% of variance accounted for: Baseline PROMIS PINT accounted for 39.2 % of the variance, and an additional 2% of variance was accounted for by including HEAL TEX in the model.
For aim 2, we were interested in whether HEAL and PROMIS measures contribute to understanding of HTE. The goal was to determine HTE based on the type of treatment patients had chosen (subgroup analyses of patients receiving CAM/integrative medicine or conventional medicine). An additional goal was to determine heterogeneity of pain treatment outcomes based on the patients' baseline HEAL TEX scores (subgroup analyses of patients with higher vs lower TEX). In other words, the aim was to determine the effects on outcome of (1) choosing a CAM or a conventional treatment, and, in a separate subgroup analysis, determine effects of (2) initial treatment expectations (HEAL TEX). We conducted longitudinal analysis using mixed models to capture HTE based on type of treatment and HEAL TEX scores at baseline. For the CAM and the conventional medicine treatments, we plotted observed mean scores at baseline, 2-month follow-up, and 4-month follow-up for each of the 5 outcome variables: AvePoint, Overall Health (higher values = poorer health), PROMIS PHYS, PROMIS PI with life activities, and PROMIS PINT. The conventional group had a higher AvePoint score, poorer Overall Health, poorer PHYS, higher PI, and higher PINT than did the CAM group at each time point (see Figure 2). We used a series of mixed models to further examine these differences. We fitted the final most parsimonious model using backward selection on HEAL scores after controlling for baseline status. For AvePoint, the overall average across all time points for the CAM group was 0.2444 (ie, the intercept, P < .001). The AvePoint in the conventional group was 0.215 (P = .062), which was higher than that of the CAM group after controlling for time, baseline status, and baseline HEAL scores. A 1-unit change in time made a 0.462 (P < .001) decrease in AvePoint after controlling for other predictors in the model.
Observed Means of Outcome Measures by CAM and Conventional.
Also, a 1-unit change in baseline status on AvePoint made a 0.837 (P < .001) increase in AvePoint after controlling for other predictors in the model. Similarly, a 1-unit change in HEAL SPT scores at baseline made a 0.013 (P = .009) increase in AvePoint after controlling for other predictors, and a 1-unit change in HEAL TEX scores at baseline made a 0.035 decrease in AvePoint after controlling for other predictors. These baseline variables, HEAL SPT and HEAL TEX, were statistically significant predictors of AvePoint in the final model. See Table 8 for estimates for each of the 5 outcomes.
To examine the impact of TEX, we divided the sample into 2 groups using a median split (low: < 51.5 vs high: ≥51.5) of the HEAL TEX scores at baseline. Observed mean score plots showed that the low HEAL TEX group had higher AvePoint, Poorer Overall Health, poorer PHYS, higher PINT, and higher PI than did the high HEAL TEX (see Figure 3). The same series of mixed models was fitted to each of the 5 outcomes for TEX as that used for type of treatment (CAM vs conventional). The final most parsimonious model using backward selection on HEAL scores, after controlling for baseline status, showed that the overall average across all time points for the low TEX group was 2.590 (ie, the intercept, P < .001). The AvePoint in the high TEX group was 0.237 (P = .053) lower than the low TEX group after controlling for time, baseline status, and baseline HEAL scores. A 1-unit change in time made a 0.448 decrease in AvePoint after controlling for other predictors in the model; also, a 1-unit change in baseline status made a 0.851 increase in AvePoint after controlling for other predictors in the model. Similarly, a 1-unit change in HEAL SPT scores at baseline made a 0.016 increase in AvePoint after controlling for other predictors. A 1-unit change in HEAL PCAM scores at baseline, HEAL POS scores at baseline, and HEAL PPC scores at baseline made a 0.016 decrease, 0.013 decrease, and 0.018 decrease, respectively, after controlling for other predictors in the model. See Table 9 for estimates for each of the 5 outcomes.
Observed Means of Outcome Measures by TEX.
After we completed the mixed model analysis, we calculated post hoc power using Average pain. We observed a 0.7 AvePoint difference between CAM and conventional groups and a 0.5 difference across time points. Based on 100 simulations of a group sample of 100 participants, we achieved a 78% power to detect the difference of 0.7 between CAM and conventional groups and a 75% power to detect a difference of 0.5 across time points, with α = .05.
Aim 3 evaluated patients' and providers' views of the utility of HEAL, PROMIS health status measures, and the ACPA Pain Log in the treatment context. We conducted individual interviews with 44 of the aim 1 and aim 2 participants, and 13 of their CAM and conventional medicine practitioners. Of these 44 interview participants, 23% were male, 34% were non-White or multiracial, 5% were of Hispanic/Latino ethnicity, and 34% had less than a 4-year college education. During the interview, patients and clinicians were asked to rate the clarity, relevance, and usefulness of each of the HEAL and PROMIS measures on a scale of 0 (“not at all”) to 4 (“very much”).
Regarding clarity of the HEAL and PROMIS items, patients and providers found the items to be clear and easy to understand (means ranged from 3.5 to 4.0 on a 0-to-4 scale). PROMIS pain-related measures were rated as highly relevant in the treatment context (mean range, 3.8-4.0), and PROMIS PHYS, Depression, ANX, and SLP were also rated as clinically relevant (mean range, 3.3-3.8). As we expected, the clinical relevance of treatment-related HEAL items (eg, TEX, PPC) were rated significantly higher than within-person HEAL items (POS and SPT). Paired t tests revealed TEX (mean [SD], 3.5 [0.8]) relevancy ratings were significantly higher than POS ratings (mean [SD], 3.3 [1.0]) and SPT ratings (mean = 2.3, SD = 1.4); t(79) = 2.31, P < .03 and t(79) = 6.54, P < .001, respectively, when both patient and health care provider ratings were combined. PPC relevance ratings (mean [SD], 3.5 [0.7]) were higher than those for POS and SPT (t(79) = 2.43, P < .02 and t(79) = 7.29, P < .001, respectively) for the combined patient–provider sample. HCE was the only treatment-related HEAL item bank that did not significantly differ from the within-person HEAL ratings of POS and SPT. Overall, patients and HCPs were often in agreement of HEAL relevance to treatment; however, health care providers rated the relevance of the TEX items and SPT items lower than did patients (F[1, 79] = 9.90, P < .01, and F[1, 79] = 4.96, P < .03, respectively) (see Figure 4 for HEAL Short Form items).
HEAL Short Form Items.
Our aim 3 intention regarding the ACPA Pain log was to collect in-depth interview information concerning the log's graphic representations of symptoms and side effects, the log's response scales and verbal content, and the log's general relevance for enhancing patient–provider communication and understanding. Table 10 summarizes the types of comments patients made about the ACPA Pain Log. Our research team will report these results in detail to the ACPA along with any suggestions for changes to the log.
Preliminary Summary of Patient Stakeholder Concerns With the American Chronic Pain Association Pain Log.
Our study evaluated the roles of patients' perceptions, treatment-related beliefs, and personal characteristics in pain treatment outcomes. Of the patient-centered questions that inspire PCOR, 2 are relevant to the current study: “Given my personal characteristics, conditions, and preferences, what should I expect will happen to me?” and “What can I do to improve the outcomes that are important to me?” We found that nonspecific factors as assessed by HEAL-influenced pain treatment outcomes. The magnitude of baseline HEAL scores' influence on outcomes was small, although it was statistically significant for HEAL POS, SPT, PCAM, and TEX. Having higher expectations regarding the value and personal appropriateness of a particular treatment (eg, “This treatment is right for me”) was associated with better outcomes at our 2-month and 4-month follow-up. We also found that patients who initiated CAM treatments had more positive outcomes than did those who chose conventional medicine. Although the study design limits our ability to draw conclusions regarding the direct causes of improvement, the study results may inspire patients to become familiar with and educated about CAM treatments. And, if, once educated, they determine whether they could personally expect benefits (ie, have high TEX), they may engage in CAM fully and experience positive outcomes. Our results may also inform clinicians' decisions. Clinicians may be inspired in several ways: increased openness to CAM and increased recognition of the importance of the patients' beliefs about treatment and the connection between patient and provider. In some settings and with some patients, providers may recognize the value of talking about deeply personal factors such as spirituality. If health systems and payers recognize these results, increase their openness to CAM for pain, and increase their focus on nonspecific factors as important for outcomes, this could lead to decreasing the burden of pain in our society. It is important to recognize, however, that not all patients have access to CAM due to lack of availability and lack of insurance coverage for some CAM treatments. It is also important to note that, in our study, CAM and conventional medicine patients differed from each other on other factors, such as baseline average pain, and this may have also influenced the results.
The study addresses the methods gap of heterogeneity of patients and treatment effects, specifically evaluating how patient choices of CAM or conventional treatment modify outcomes and assessing the modifying effects of initial treatment expectations regarding pain treatment. To summarize what we have learned based on the HTE analyses, 3 main patterns are worth noting. First, as a group, the chronic pain patients in this study showed improvement over time in the following areas: AvePoint, PHYS, PI, and PINT. This was true for the group of participants overall, regardless of type of treatment they received and whether their treatment expectations were lower or higher than average. Second, participants' characteristics and treatment choices at baseline were strong predictors of their outcomes. Participants who engaged in CAM treatments had less severe pain and better physical function at baseline than did conventional medicine participants, and they also had greater improvements in these areas over time. Likewise, those participants who had higher HEAL TEX at baseline showed greater improvement in pain and function than those with lower HEAL TEX. Third, given the importance of baseline HEAL TEX and patient choices about treatment types, and considering the impact of time in predicting improvement, several other HEAL measures also emerged as contributing to treatment outcomes. Although the additional outcome variance predicted was relatively small, HEAL POS and HEAL SPT were statistically significant contributors to physical function and pain outcomes, and HEAL CAM and HEAL PPC were statistically significant contributors to several outcomes variables as well (see Tables 8 and 9).
There is growing recognition of the influence of nonspecific factors such as expectations, beliefs, ritual, and perceptions of the patient–provider connection in the experience of pain.29 Some studies have evaluated open-label placebo and found that patient beliefs and the ritual involved with taking the placebo pills led to improvements in pain relative to patients receiving usual care alone.13,14 A recent study used experimentally induced pain (a heat stimulus) and an opioid or a placebo infusion, and manipulated treatment expectations and attention through instructions to study participants. This study supported the role of expectations.30 Our study adds to this literature by measuring reports of nonspecific factors from the patient's perspective and evaluating their role in real-world pain treatment settings. In addition, we obtained patients' and clinicians' perspectives on the utility of using HEAL and PROMIS to enhance care in clinical settings.
Our aim 3 sought patients' and clinicians' perspectives on implementing PROMIS and HEAL assessments as part of clinical care. Patients viewed treatment-related assessments particularly favorably, rating them to be highly relevant. Regarding HEAL, patients made statements such as, “Integration between mental and physical is best” and “If I'm in a medical environment where people are spiritual, that adds another level to me; there is more trust.” However, some patients felt that HEAL SPT questions were irrelevant to their chronic pain condition, but “spirituality may be more relevant in a terminal situation,” and some providers said that they would not be comfortable discussing the HEAL PPC and HEAL HCE with their patients.
Conventional medicine providers also noted that the extra time required for viewing and discussing HEAL and PROMIS reports with patients would be a barrier to implementation in clinical settings. Thus, in implementing HEAL and PROMIS into clinical care in the future, it will be important for the assessments to be completed online before visits and for automatically generated summary reports to be easily available to clinicians as part of the medical record. Ideally, clinicians should receive training in how to discuss the report with their patients. And, preferably, patients and clinicians must have the freedom to choose which HEAL and PROMIS measures to use as communication tools in their treatment. Similarly, many patients found the ACPA Pain Log to be a potentially useful tool to enhance communication with their provider— given that some confusing elements of the log can be modified. The research team will suggest changes to the ACPA that should improve the log's clinical utility.
Disseminating the results of this study and encouraging other researchers to adopt HEAL into their studies were major activities of our research team and PSAP during recent months. Publications and presentations are listed in Section M of this report. Our dissemination of HEAL measures to other researchers is also provided in Section M.
We were able to recruit and retain a fairly diverse sample of persons with chronic pain, including 23% non-White or multiracial persons, 25% males, and approximately 50% with less than a 4-year college education; however, we had few persons of Hispanic/Latino ethnicity in the study (14 out of 209). We increased the breath of our sample by including patients from the Venice Clinic at UCLA, which is an integrative medicine clinic where patients receive treatment free of charge. Our results have reasonable generalizability to US populations seeking treatment for chronic pain. However, HEAL and many PROMIS measures are equally relevant for patients with a broad range of medical conditions and treatments, and the choice to study chronic pain limits the generalizability of our results to other patient populations.
Our study included subpopulations of persons seeking CAM treatment and conventional medicine treatment. Persons in CAM treatments tended to have higher expectations about their new treatment and better self-reported health and less pain than those in conventional treatments, and fewer of the CAM patients were also using opioid medications. Approximately a quarter of both CAM and conventional participants reported having 2 or more medical comorbidities in addition to their chronic pain; thus medically ill persons were represented. The questions regarding opioid use were suggested by a clinician member of our PSAP to explore the potential role of opioids in treatment satisfaction with pain treatment overall, which will require further study. We assessed the PROMIS health status measures of depressive and anxiety symptoms; these psychosocial factors are important to assess along with self-reports of pain and physical function. Although depression can contribute to poor pain treatment outcomes, our study found that, overall, in our diverse sample of persons who freely select their pain treatments, depressive symptoms were not significantly reduced over the 4-month period of study; however, pain was reduced over time to a statistically significant extent.
This study focused broadly on the measuring contextual factors in treatment using the HEAL and PROMIS health status instruments. We chose to study patients with chronic pain during their routine, self-chosen treatments. Thus, there are limits to the generalizability of this study beyond pain. And, due to the diversity of treatments, from acupuncture to physical therapy, we are unable to discern for which treatments contextual factors may be most important. HEAL and many PROMIS measures are relevant on a broad level to many types of medical conditions and treatments, however. In addition, in addressing the HTE methods gap, we chose to evaluate subgroups of patients based on type of treatment they had chosen (CAM or conventional medicine). In separate subgroup analyses, we also chose to evaluate outcomes in subgroups of patients with lower and higher levels of HEAL TEX. These treatment type and expectancy subgroupings were important factors in HTE; however, the roles of other potentially important factors remain unknown. A limitation of this study is that biases may be present in the analysis of HTE. Patients in the comparison groups of CAM vs conventional medicine and those showing low vs high HEAL TEX may diverge in many ways. Of note, CAM and conventional medicine patients did not differ on demographic grounds; however, in terms of clinical variables, there were differences. CAM participants reported less pain on average at baseline and were less likely to be taking prescription opioid medications. Thus, there may be bias in our HTE results. We addressed this through controlling for baseline level of AvePoint (and baseline levels of our other outcome variables) in our final multivariate models.
Some of the limits to the conclusions that can be drawn from this study are based on our chosen design. This study was a cohort study of patients who chose their pain treatments. Factors that may contribute to those choices—such as preference for CAM methods, types of pain, and severity of symptoms—were not controlled. The self-selection of treatments, while of real-world significance, is a limitation of this study. In contrast, in randomized controlled trials such potential sources of bias or confounding can be systematically addressed through stratification and other methods. Another limitation is that, given our simple subgroup analyses, we did not show that HEAL scores (such as TEX) identified a subpopulation for whom CAM therapies were more effective than conventional therapies. That said, despite limitations in the design, the study has value, as it was conducted in the real-world settings in which patients with chronic pain seek and obtain treatment.
Recommendations for future research include evaluating HEAL and PROMIS measures in populations with medical illnesses other than chronic pain to support generalizability. Feasibility, acceptability, and outcomes associated with implementing HEAL and PROMIS within the electronic medical record in real-world clinical settings should also be a focus of future research. In addition, HEAL measures could be a valuable addition to large placebo-controlled trials. In such a research design, investigators would be able to test, through patient-reported measures, the specific influences of treatment beliefs and personal characteristics as contributors to the placebo effect.
We found that the patients' perceptions of the treatment context and of themselves (eg, positive outlook and spirituality), as assessed by the HEAL measures, are associated with PROMIS pain treatment outcomes. This is consistent with recent literature on nonspecific and placebo research in pain.13,30 Factors such as expectations about treatment and participation in CAM vs conventional medical treatment were associated with significant differences in symptoms throughout the 4 months of our study follow-up. In interviews, patients and clinicians reported that the HEAL and PROMIS measures were easy to understand and useful for enhancing patient–provider communication during treatment. These results may be useful to the following health care decision makers: patients, clinicians, and health system leaders who are interested in enhancing treatment outcomes and fully engaging patients in treatment.
Research reported in this report was [partially] funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#ME-1402-10114) Further information available at: https://www.pcori.org/research-results/2014/using-surveys-assess-patient-centered-factors-may-affect-responses-chronic
Cognitive Interview Materials (PDF, 385K)
Glossary of Terms (PDF, 92K)
Greco C, Yu L, Polonis P. (2019). Using Surveys to Assess Patient-Centered Factors that May Affect Responses to Chronic Pain Treatment. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/4.2019.ME.140210114
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
