Fuzzy trace theory was used to develop a coding scheme that captures the gist that patients extract from information about medication risks and benefits and to explore the extent to which different patients extract different gist representations from the same information.
Data were collected from 2003–2007 in a study that included audiotape recording office visits that rheumatoid arthritis (RA) patients had with their rheumatologists. Each patient (n = 365) had up to 3 visits audiotape recorded. The audiotapes were transcribed to facilitate content analysis. Four patients with RA who did not participate in the original study guided development of the coding scheme and used it to code the transcripts.
The coding scheme contains 14 gist themes centering on medication effectiveness, need, and safety. There was considerable variation among the gist coders in the specific themes they extracted from individual transcripts. We observed the greatest intercoder agreement for the 4 gist theme variables related to whether the rheumatologist wanted to make changes to the medication regimen. Furthermore, the coders rarely used the “not clear” category to code these 4 variables. In contrast, intercoder agreement for the remaining gist themes, which were designed to capture issues central to the communication of information about medication risks and benefits, was low and the “not clear” category was used more frequently.
Our study findings suggest that different people exposed to the same information may form different gist representations. Patient–provider communication concerning medication risks and benefits might be enhanced by better understanding the factors that influence the gist extraction process.
Rheumatoid arthritis (RA) is a systemic, autoimmune disorder affecting ∼1% of the US population (). Despite recent advances in therapy, RA often leads to progressive joint destruction and significant functional impairment (). The primary therapeutic goals in managing RA are to control inflammation and pain, minimize joint damage, and prevent loss of function (). Over the past 15–20 years, the medical management of RA has changed dramatically because understanding of the pathophysiology and natural history of the disease has advanced. The current guidelines call for aggressive treatment of early RA with disease-modifying antirheumatic drugs (). However, the benefits associated with aggressive therapy are accompanied by risks (). Patients therefore need a clear understanding of both the risks and benefits associated with different treatment options in order to make informed decisions among therapeutic alternatives.
A large body of literature on perceived risk and risk communication exists (); however, little of this work has focused on patient–provider communication about medication risks, especially within the context of musculoskeletal disorders ([7, 8]). Therefore, we know little about the extent to which rheumatologists discuss medication risks with patients and even less about how patients extract meaning from the information provided during these discussions ().
In this study, we used fuzzy trace theory (FTT) to better understand the meaning that patients extract from information exchanged between patients and rheumatologists during routine office visits. Briefly, FTT is a dual-process model of memory, reasoning, and development that has been used to study how people make decisions that involve risk (). FTT was developed from findings in cognitive research conducted over the past 20 years suggesting that judgment and memory operate independently (). FTT posits that, when an individual is exposed to any meaningful stimulus (e.g., a statement made by a physician), 2 representations of the stimulus are encoded in memory: a verbatim representation and a gist representation. Verbatim representations capture the actual words and numbers used, whereas gist representations reflect the essential meaning of those words and numbers to the person (). Different people exposed to the same stimulus may form different gist representations depending on their preexisting knowledge, past experiences, emotional state, and developmental stage. A central tenet of FTT is that, when making judgments and decisions, people tend to rely on the gist representations that are stored in memory. Further, this preference for gist information processing increases with age and the acquisition of specialized expertise ().
In this study, we describe the development of a coding scheme to capture the gist representations that patients form in response to RA medication information exchanged between patients and rheumatologists during routine office visits. Because FTT suggests that different people may form different gist representations in response to the same information, the coding scheme did not use explicit coding rules and examples to define gist representations; instead, the coding scheme comprised a series of questions reflecting different gist themes. The coders were instructed to endorse themes that, in their view, best described the gist of what the rheumatologist said about a particular medication. We then used the coding scheme to examine the extent to which different patients extracted the same (or different) gist in response to the same information.
Box 1. Significance & Innovations
Research on fuzzy trace theory suggests that when making judgments and decisions, people tend to rely on gist representations that are stored in memory, and that this preference for gist processing increases with age and the acquisition of specialized expertise.
Within the context of patient–provider communication, clinicians should recognize that different patients may extract different gist from the same information.
Interventions targeting physicians might focus on the gist themes that we identified (e.g., physician concern about the safety of a particular medication) and training physicians to emphasize the gist of the information they provide in a clear and unambiguous manner. Interventions might also incorporate methods for the physician to assess the gist that patients extract from the information provided and address any errors in gist reasoning that are evident.
Interventions targeting patients might focus on helping them better understand the factors that physicians consider, and the reasoning principles that they use, when making therapeutic recommendations.
MATERIALS AND METHODS
The data analyzed were collected from 2003–2007 as part of a National Institute on Aging–funded study (Older Adults and Drug Decisions: Collaboration & Outcomes). The study used a randomized controlled trial design to evaluate an intervention that encouraged patients to talk with their doctor about their most important health concerns. The study was limited to patients who had RA (with the diagnosis confirmed by a rheumatologist), were ages ≥45 years, had no known terminal illnesses, could speak English, and were mentally competent. The clinic staff identified eligible patients prior to their next office visit. At the visit, a research assistant explained the study to the patient and obtained written informed consent. After providing informed consent at baseline, the participants met with their rheumatologist for a routine office visit, which was audiotape recorded. At the completion of the visit, the patients were interviewed and completed a brief questionnaire. The questionnaire included an item assessing general health status (i.e., 1 = excellent, 2 = very good, 3 = good, 4 = fair, and 5 = poor) and a visual analog scale assessing pain intensity during the past 24 hours (with the end points labeled 0 = no pain and 10 = worst pain ever). Each patient's rheumatologist also assessed the patient's functional status using the American College of Rheumatology (ACR) revised criteria for the classification of global functional status (). The patients were then randomly assigned to either the intervention group or control group. The intervention procedures were implemented at routine followup office visits that occurred ∼6 and 12 months following study entry. The data collection procedures for the 6- and 12-month followup visits mirrored those used at baseline.
A total of 18 rheumatologists participated in the study (8 in North Carolina and 10 in Wisconsin). Within each state, the rheumatology clinics were geographically dispersed and included both academic and community-based sites. In the current study, because RA patients who lived in or near Chapel Hill, North Carolina coded the data, 5 physicians local to Chapel Hill were excluded to ensure that the coders would not have had contact with any of the study participants. Therefore, data from 13 rheumatologists and 365 patients who participated in the Older Adults and Drug Decisions: Collaboration & Outcomes study were analyzed.
The gist coding scheme was developed through a series of group meetings involving 4 RA patients, who served as gist coders. The meetings, which took place from July to November 2011, were facilitated by the first author. One gist coder was a leader of a local support group; she identified the other 3 coders to participate in the project. All of the coders were white women.
Before the first meeting to develop the coding scheme, audiotapes of the office visits were transcribed and a random sample of the transcripts was used to develop the coding scheme. At the first 2 meetings, the gist coders were given 2–3 transcripts to read during the meeting. After reading each transcript, the coders were asked to identify and discuss 1) the main concerns expressed by the patient and rheumatologist during the visits, irrespective of whether the concerns were medication related; 2) the main points that the rheumatologist made related to medication safety/risks and medication effectiveness/need; and 3) the main concerns patients expressed regarding medication safety/risks and medication effectiveness/need. The discussion that occurred during these meetings was used to develop a draft coding scheme.
The coding scheme was revised through a series of iterations that involved the gist coders coding a limited number of transcripts (i.e., 5–25) between meetings and then convening to discuss the problems they had experienced using the coding scheme, including whether the coding scheme was capturing all of the important medication-related communication that occurred during the visit. The gist coders also suggested and discussed potential modifications that they believed would improve the coding scheme. Disagreements in codes assigned by different gist coders were discussed. However, because FTT posits that different people may extract different gist from the same information, the discussion focused on reaching a common understanding of the types of information that might be captured by each gist theme rather than reaching agreement concerning the presence or absence of a particular theme within a transcript.
After revisions were completed, an electronic version of the coding scheme was developed using Qualtrics software to facilitate data entry via the internet. The transcripts were converted to e-book format so they could be read on a tablet computer. This allowed the coders to highlight transcript passages for ease of reference. After reading each transcript, the coders connected to the internet and entered their codes into the Qualtrics database.
The coders used the gist coding scheme to identify the gist themes they extracted from the transcripts of the audiotape-recorded office visits. The remaining variables were obtained from the original study and were based on the responses of the patients and rheumatologists who participated in the Older Adults and Drug Decisions: Collaboration & Outcomes study.
The gist coders were asked to identify the arthritis or osteoporosis medication (e.g., methotrexate, or alendronate) discussed most during the visit. Osteoporosis medications were included because both RA and some medications used to treat RA increase the risk of bone loss (). In addition, medications used to control pain (e.g., narcotic analgesics) or treat sleep problems (e.g., cyclobenzaprine) could be identified if the coder considered the reason for the administration of the drug to be arthritis related. The coding form instructed coders not to identify medications that were used by someone other than the patient. Next, coders indicated whether the patient was 1) currently using the medication, 2) had used the medication in the past, or 3) had never used the medication. The coders stopped at this point if, in their judgment, very little about any medications had been discussed during the visit. Otherwise, the coders responded to a series of questions that asked them to select the answers that best described the gist of what the rheumatologist said about the medication they had identified. The questions included the following: 1) “the medicine has” (few side effects, many side effects, or not clear), 2) “the medicine has” (no serious side effects, some serious side effects, or not clear), 3) “compared to other medicines, this medicine is” (more safe, about the same, less safe, not clear, or no other medicines discussed), 4) “the patient can do things to decrease the risk of having side effects from this medicine” (no, yes, or not clear), 5) “the patient can do things that will increase the risk of having side effects from this medicine” (no, yes, or not clear), 6) “the patient can use this medicine as long as therapy is monitored carefully” (no, yes, or not clear), 7) “what risk is the doctor most concerned about?” (RA getting worse, both equally, medicine risks, or not clear), 8) “how much does the doctor think the medicine is helping (or is likely to help)?” (not at all, a little, a fair amount, a lot, or not clear), 9) “how much does the doctor think that the patient needs the medicine?” (not at all, a little, a fair amount, a lot, or not clear), 10) “how much is the doctor concerned about the safety of the medicine for this patient?” (not at all, a little, a fair amount, a lot, or not clear), 11) “does the doctor want to stop (start) using the medicine?” (no, yes, or not clear), 12) “does the doctor want to change the amount of the medicine that the patient is using?” (increase amount, stay the same, decrease amount, or not clear), and 13) “does the doctor want to change how often the patient uses the medicine?” (use more often, stay the same, use less often, or not clear).
After responding to these questions, the coders stopped if, in their judgment, very little was discussed about any other medicines during the visit. Otherwise, they identified a second medication discussed during the visit and completed the gist coding scheme for that medication.
Patient and physician sociodemographic characteristics
Patient sex, age, marital status, education, income, and race were assessed via the baseline questionnaire. The participating rheumatologists completed a questionnaire that assessed their sex, age, and race.
All analyses were performed using PC SAS (). Descriptive statistics (e.g., means and percentages) were used to describe the characteristics of the study participants and the frequency with which different gist themes were extracted by the gist coders from the information discussed during the audiotape-recorded visits. The results are shown aggregated across coders and stratified by coder. Because the coders could code different medications for a visit, primary analyses were based on the codes assigned for 264 medications that were coded by all 4 coders. Agreement among the codes assigned by the 4 coders was assessed by simple agreement and Cohen's kappa.
Characteristics of the study participants
As shown in Table 1, the mean ± SD age of the participants was 61.9 ± 9.7 years. Most participants were white (90.0%), women (77.0%), and in functional class II (54.0%) of the ACR revised criteria for the classification of global functional status (). The study participants were treated by a total of 13 rheumatologists. On average, the rheumatologists were a mean ± SD age of 47.6 ± 7.3 years. Most (84.6%) of the rheumatologists were white and nearly half (46.2%) were women.
Table 1. Characteristics of the study participants
Patients with a medication coded by all coders (n = 185)
aBecause of missing data, the numbers of patients included in each specific characteristic for the full sample and for those with a medication coded by all coders were: patient race (n = 359 and n = 181, respectively), patient education and marital status (n = 358 and n = 180, respectively), patient income (n = 333 and n = 164, respectively), general health status (n = 357 and n = 182, respectively), pain (n = 364 and n = 184, respectively), and patient functional class (n = 352 and n = 176, respectively).
Age, mean ± SD years
61.9 ± 9.7
61.4 ± 9.7
Eighth grade or less
Some high school
High school graduate
Some postgraduate work
Household annual income, %
General health status, mean ± SD
3.03 ± 0.98
3.09 ± 1.01
Pain, mean ± SD
3.65 ± 2.56
3.49 ± 2.56
Physician rating of functional class at baseline, %
Characteristics of medications discussed
Together, the gist coders coded a total of 4,178 medications. Each of the 365 study participants had at least 1 medication coded by at least 1 gist coder. A total of 264 medications were coded by all 4 gist coders. These medications accounted for 25.3% (i.e., [264 × 4]/4,178) of all the medications coded. The most frequently coded medications are shown in Table 2. Because multiple medications could be coded per patient, the 264 medications coded by all 4 gist coders were used by a total of 185 study participants. The characteristics of these 185 participants were similar to the full sample (Table 1). All 13 rheumatologists were represented in the subset of 185 participants.
Table 2. Medications most commonly coded by the gist coders*
Total number of times medication coded (n = 4,178)
Number of times medication coded by all gist coders (n = 264)a
Values are the percentage (number).
aThe remaining medications coded by all 4 gist coders were: minocycline (n = 4), piroxicam (n = 3), salsalate and rituximab (n = 2 for both), and valdecoxib, gold, ibuprofen, propoxyphene, nabumetone, acetaminophen, and rofecoxib (n = 1 for each).
Gist themes extracted
Figure 1 shows the proportion of times the coders extracted the various gist themes from the transcripts. These analyses were limited to the 264 medications coded by all 4 gist coders. The most frequently extracted gist themes were 1) the rheumatologist does not want to discontinue a medication used currently (87%); 2) when discussing a medication not used currently, the rheumatologist wants to initiate therapy with the medication (85%); 3) the rheumatologist wants to keep the dose regimen of a medication used currently the same (80%); 4) the rheumatologist is equally concerned about the risk of the RA getting worse and medication risks (60%); 5) the patient can use the medicine as long as therapy is monitored carefully (56%); and 6) the medicine has some serious side effects (52%). For 3 of the gist themes coded (i.e., whether the medication was more safe or less safe than other medications, whether the patient could do things to decrease the risk of medication side effects, and whether the patient might do things that would increase the risk of medication side effects), the modal code was “not clear.”
Variation in gist themes extracted across the gist coders
Table 3 shows the distribution of codes assigned by the gist coders. There was a high level of agreement among coders with respect to medication status (i.e., being used currently, used in the past, or never used). Simple agreement, averaged across coders, was 96% (κ = 0.88). There was a moderate level of agreement among coders with respect to the 4 gist themes that concerned the rheumatologist's desire to initiate regimen changes. The simple agreement for these themes ranged from 85.7–90.3% (κ = 0.49–0.78); however, for the remaining variables, there was substantial variation in the gist themes extracted by different gist coders. For example, the percentage of medications in which the gist coder indicated that the rheumatologist said the medication had few side effects ranged from 3.4% (coder 2) to 32.6% (coder 4). Similarly, the percentage of medications in which the gist coder indicated that the rheumatologist said the medication was helping (or would help) a lot ranged from 12.6% (coder 1) to 76.1% (coder 2). Furthermore, only one of the 10 general gist themes (i.e., the patient can use the medication as long as therapy is monitored carefully) had a kappa value >0.40.
Table 3. Proportion of times gist themes were extracted in discussion of the 264 medications coded by all 4 gist coders, by coder*
Simple agreement, %
Values are the percentage (number) unless otherwise indicated. RA = rheumatoid arthritis.
aThe intraclass correlation coefficient (ICC) was computed using the Shrout-Fleiss method for the reliability of a single score with the “not clear” category set to missing. The ICCs were 0.19, 0.27, and 0.19 for the amount the medication is helping, needed, and the rheumatologist's concern about medication safety, respectively.
Used in the past
Gist themes: medications used currently
Rheumatologist wants to stop the medication
Rheumatologist wants to:
Increase medication dose
Keep medication dose the same
Decrease medication dose
Rheumatologist wants patient to:
Take the medication more often
Keep the dose regimen the same
Take the medication less often
Gist themes: medications used in past or never used
Rheumatologist wants to start the medication
Gist themes: general
The medication has:
Few side effects
Many side effects
The medication has:
No serious side effects
Some serious side effects
Compared to other medications this medication is:
About the same
Concern about risks associated with RA versus medications
More concerned about risks of RA getting worse
Equally concerned about risks of RA getting worse and medication risks
More concerned about medication risks
The patient can do things to decrease the risk of having side effects from the medicine
The patient can use the medicine as long as therapy is monitored carefully
The patient may do things that would increase the risk of having side effects from the medicine
The findings from this study represent an important first step toward a better understanding of how patients extract gist representations from information concerning medication risks and benefits. The coding scheme developed included 14 themes centered on medication effectiveness, need, and safety. Involving patients in the development of the coding scheme helped to ensure that the themes would be patient centered. The themes were also consistent with previous studies suggesting that medication effectiveness, need, and safety are critical issues that patients consider when making decisions about whether to initiate and continue therapy ([15-18]).
As expected, there was considerable variation in the specific gist themes extracted from individual transcripts by the 4 gist coders. We observed the greatest intercoder agreement for the 4 gist theme variables related to whether the rheumatologist wanted to make changes to the patient's medication regimen, suggesting that rheumatologists may communicate this type of information more explicitly than other types of information, thereby leaving less room for different interpretations. Furthermore, the gist coders rarely used the “not clear” category to code these 4 variables. In contrast, for 6 of the remaining 10 gist theme variables, the “not clear” category was used >30% of the time, when averaged across the 4 coders. These variables were designed to capture issues central to the communication of information about medication risks, including the probability and severity of side effects, relative safety of therapeutic alternatives, and recommendations to minimize the risk of side effects. Furthermore, although the “not clear” category was used only 14% of the time for the gist themes concerning medication efficacy and need, this category was used 29% of the time for the theme that addressed rheumatologist concern about medication safety. Together, these findings suggest a quality gap in patient–rheumatologist communication concerning medication risks and risk management. If rheumatologists fail to provide this type of information clearly, it may leave patients vulnerable to information from less reliable sources and result in resistance to the escalation of therapy when indicated ([19, 20]); it may also lead to suboptimal adherence to recommended safety precautions (e.g., laboratory monitoring) throughout the course of therapy.
The variability among coders in the gist themes extracted from transcripts is consistent with FTT and suggests a need for further research to better understand how patients extract gist representations from verbatim information, whether communicated by their rheumatologist or accessed via other sources (e.g., the internet). Furthermore, although FTT suggests that patients may extract different gist representations from the same information (e.g., whether a 1% risk is interpreted as high or low), patients can still make errors when extracting the gist representation. For example, if 2 medications were identical on all safety parameters, except that the risk of liver toxicity is 5% for drug A versus 1% for drug B, forming a gist representation that drug A is safer than drug B would be incorrect. Two studies that examined the factors associated with errors in gist reasoning found that white race, higher education, and greater numeracy skills were associated with fewer errors when extracting gist from information presented in a standardized stimulus ([21, 22]). One of these studies () also found that men made fewer errors than women. However, it seems likely that many other factors influence the gist extraction process. At the patient level, factors such as prior knowledge and beliefs, past experiences, affective state, current disease activity, personality characteristics, and social resources may be important. Beyond patient-level characteristics, factors likely to influence the gist extraction process include the information source, characteristics of the information itself (e.g., complexity and clarity), and context in which the information is provided (e.g., the presence of competing demands for attention). Further research is needed to better understand the effect of these types of factors on the gist extraction process, particularly as they relate to errors patients may make in gist reasoning.
This line of research also has important implications for developing strategies to improve patient–provider communication concerning medication risks and risk management and to enhance a patient's ability to understand and evaluate this type of information when obtained from other sources. Because FTT suggests that people tend to rely on gist representations when making judgments and decisions, interventions should focus on a patient's ability to extract gist that is consistent with the verbatim information provided, rather than their memory of precise details (e.g., the probability of a particular side effect). Interventions targeting physicians might focus on the gist themes that we have identified (e.g., physician concern about the safety of a particular medication) and training physicians to emphasize the gist of the information they provide in a clear and unambiguous manner. Interventions might also incorporate methods for the physician to assess the gist that patients extract from the information provided during an office visit and address any errors in gist reasoning that are evident. Other novel interventions might be developed to enhance the gist reasoning skills of patients. A study on FTT found that gist reasoning ability improved with maturation and expertise (); therefore, patient–provider communication might be improved by interventions that help patients better understand the factors that physicians consider, and the reasoning principles they use, when making therapeutic recommendations. Finally, decision tools that incorporate gist principles may also help to enhance patient understanding and decision making ().
Our results should be interpreted in light of several study limitations. First, the patient coders were not the same patients whose office visits were audiotape recorded. In the future, we plan to administer the gist coding scheme immediately following an audiotape-recorded visit. This will allow us to assess the extent to which patients extract gist that is consistent with the information exchanged during the visit and the extent to which patient and contextual factors affect gist processing. Second, all of the coders were well-educated white women; similarity among the coders likely enhanced intercoder agreement. Greater heterogeneity among the coders may have resulted in the identification of additional gist themes. Third, the participants in the Older Adults and Drug Decisions: Collaboration & Outcomes study were predominantly white women, and most of the rheumatologists were also white; this may also have limited the diversity of the gist themes included in the coding scheme. Fourth, the coders worked from written transcripts of audiotape-recorded visits. Although we did this to protect the anonymity of the study participants, it prevented the coders from hearing the tone of voice that patients and rheumatologists used to express concerns. In addition, because the office visits were not videotape recorded, the coders did not have access to information communicated nonverbally (e.g., via facial expressions). Had this type of information been available to coders, it may have improved intercoder agreement. Fifth, data collection for the Older Adults and Drug Decisions: Collaboration & Outcomes study ended in 2007. Since that time, new medications have entered the market and information concerning medication risks has continued to evolve. However, because the gist themes included in the coding scheme were general in nature, we would expect the basic structure of the coding scheme to remain sound as therapeutic advances are made.
In conclusion, given the number of therapeutic options available to patients with many chronic conditions, including RA, the need to effectively communicate information concerning medication risks and benefits has become increasingly important. We hope that the coding scheme developed in this study will lead to a better understanding of the factors that influence how patients extract gist representations from information concerning medication risks and benefits that their physician provides, as well as information that patients encounter from other sources. We also anticipate that this line of research will lead to novel intervention and communication strategies that enhance a patient's ability to make informed decisions concerning therapeutic options and engage in safe medication use practices throughout the course of therapy.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Blalock had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Blalock, Brenda M. DeVellis, Robert F. DeVellis, Chewning, Sleath.
Acquisition of data. Blalock, Chewning, Jonas, Sleath.
Analysis and interpretation of data. Blalock, Slota, Brenda M. DeVellis, Robert F. DeVellis, Sleath.
The authors would like to thank Dr. Valerie Reyna for reviewing a draft of the coding scheme and providing advice for revisions prior to the initiation of coding procedures. We also wish to thank the 4 patients with RA who served as gist coders.