Communication and patient participation influencing patient recall of treatment discussions

Abstract Context Patient recall of treatment information is a key variable towards chronic disease (CD) management. It is unclear what communication and patient participation characteristics predict recall. Objectives To assess what aspects of doctor‐patient communication predict patient recall of medication information. To describe lifestyle treatment recall, in CD primary care patients. Design Observational study within a RCT. Setting & participants Community‐based primary care (PC) practices. Family physicians (n=18): practicing >5 years, with a CD patient caseload. Patients (n=159): >40 years old, English speaking, computer literate, off‐target hypertension, type II diabetes and/or dyslipidaemia. Main variables Patient characteristics: age, education, number of CDs. Information characteristics: length of encounter, medication status, medication class. Communication variables: socio‐emotional utterances, physician dominance and communication control scores and PACE (ask, check and express) utterances, measured by RIAS. Number of medication themes, dialogue and initiative measured by MEDICODE. Main outcome measures Recall of CD, lifestyle treatment and medication information. Results Frequency of lifestyle discussions varied by topic. Patients recalled 43% (alcohol), 52% (diet) to 70% (exercise) of discussions. Two and a half of six possible medication themes were broached per medication discussion. Less than one was recalled. Discussing more themes, greater dialogue and patient initiative were significant predictors of improved medication information recall. Discussion Critical treatment information is infrequently exchanged. Active patient engagement and explicit conversations about medications are associated with improved treatment information recall in off‐target CD patients followed in PC. Conclusion Providers cannot take for granted that long‐term off‐target CD patients recall information. They need to encourage patient participation to improve recall of treatment information.

Many studies equate patient participation with patient question asking. 5,40 The effect of question asking on recall of information remains equivocal; some studies show no effects, 13,27,39,41 and others show positive effects 38 or even negative effects. 13,27,42 The main objectives of this study were to describe recall of lifestyle and medication treatment information and to assess what aspects of doctor-patient communication and patient participation predict patient recall of medication information, in off-target chronic disease primary care (PC) patients. This study is an observational study within a randomized control trial. Results from the randomized control trial concerning communication have been reported elsewhere. 43

| Study design
This study is an observational study within a randomized trial. The clinical trial, NCT00879736, was registered with ClinicalTrials.gov, and the protocol received ethics approval from the Institutional Review Board Services (IRB). Informed consent was obtained from physicians and patients. All participants were informed about the confidentiality of their data. There were no transcriptions of audiotapes. All direct identifying information was numerically coded.
The clinical trial was a three-arm parallel design that randomized patients into receiving one of the two communication interventions or usual care. Interventions were delivered either through the web only or combined with a workshop. The main objective was to assess the impact of the educational interventions compared to usual care (no additional material) on doctor-patient communication. Analysing group effects for medication information recall was not possible because of the quantity of missing data. Thus, this study disregarded original group assignment and reports on patients' correct recall of information about their chronic disease treatments.

| Setting and participants
Patients were recruited from nine urban and suburban communitybased PC practices in Ontario (Canada) between March and December 2009, by the study coordinator. The last outcome data were retrieved from patient charts in September 2010.
Participating family physicians (FPs) were chosen using a convenience sample and were considered eligible if they had been in practice at least 5 years, had a practice orientation towards an adult population, including chronic disease patients, agreed to the audio recording of one visit per participating patient. Once physicians consented, patients were approached by the study coordinator and enrolled if they consented and met the following criteria: (i) 40 years or more of age, (ii) ability to speak English, (iii) comfortable using a computer for routine activities such as regular access to the web and e-mail, (iv) have a routine follow-up visit scheduled within 3-4 months of study enrolment, (v) allow access to their medical records and (vi) have at least one of the following three CDs: hypertension, type II diabetes and/or dyslipidaemia. These CDs were diagnosed by their physicians as not meeting treatment targets set by relevant Canadian guidelines. [44][45][46] Pregnant patients or those actively treated for cancer were excluded.

| Procedures
Patients were recruited and enrolled at an initial visit. Participating FP completed a basic socio-demographic and practice profile questionnaire at study enrolment. Patients meeting inclusion criteria completed a baseline questionnaire including socio-demographic and clinical data such as length of FP-patient relationship. 47 Three to four months later, they returned for their follow-up visit with their FP following normal scheduling. This visit was audio-recorded. Immediately following this visit, patients completed a questionnaire that assessed their recall of treatment information they had just discussed with their FP.

| Variables & measures
Predictor and outcome variables that were used in analyses and their sources are described in Table 1.
Audio-taped encounters were coded with the Roter Interaction Analysis System (RIAS) 48 and MEDICODE. 49,50 These two validated coding systems and the following measures are described in detail in Lussier et al. 43 RIAS ascribes an interaction code to all utterances spoken by the physician and the patient during an entire encounter. Codes fall under two large categories: socio-emotional and instrumental utterances. Socio-emotional utterances are types of talk where the participant shows agreement, understanding, empathy, expresses concern, disapproval or reassurance, for example. Instrumental utterances are types of talk where the participant gives information or counsels on medical conditions, lifestyle issues or treatments. RIAS allowed us to estimate encounter length, measure physician dominance and patient communication control scores, the proportion of utterances that were socio-emotional and a PACE composite score assessing "ask," "check" and "express" utterances. MEDICODE is a content analysis system that codes communication about medications and medical conditions. MEDICODE codes the presence of discussions related to hypertension, diabetes and dyslipidaemia. Furthermore, discussions of lifestyle treatment, such as diet, exercise, stress management, tobacco and alcohol, were also coded. For medication discussions, six of the ten possible thematic metacategories were judged the most relevant for patients with chronic disease to understand and properly adhere to their medications: medication name, instructions (how to take medication), medication main effect (how it works), adverse effects, adherence and concerns. Themes such as "warnings" and "indications to reconsult" were deemed less pertinent in the context of follow-up care for long-term patients, where few prescriptions were new prescriptions. The average number of themes discussed per medication was calculated, giving an indication of the extent of the medication discussion.
Medication status was coded in MEDICODE as (i) active discussed: medications that patients are taking without need for a new script, (ii) renewed prescription: renewing a prescription for an active medication, (iii) new prescription and (iv) other: medications which are discussed but are not currently being taken. 51 For each theme discussed, MEDICODE also codes for interactions, such as who initiates medication discussions and whether there is a dialogue or a monologue. A dialogue score was calculated using the average level of dialogue on medication themes per medication discussed [0=monologue, 1=dialogue]. An initiative score was calculated in the same way [−1=patient initiative, 0=shared initiative, 1=physician initiative].
Three coders, who were blind to group allocation, received an intensive 1-month training in both methods and were supervised on a continuous basis during the coding by one of the authors (CR). Interrater reliability was calculated on 10% of encounters. Any discrepancies were resolved through group discussion. Average per cent agreement for RIAS categories was 90%. Mean Kappa value for MEDICODE was .83, showing good agreement between coders.

| Recall of treatment information
Immediately following the encounter, patients completed the postvisit questionnaire. They were asked whether they had discussed hypertension, diabetes or dyslipidaemia, and if so to elicit the name of all the medications they remembered discussing for each CD. Patients were also asked whether they had discussed the five other medication themes described above (yes/no) in relation to the named medication. Patients were asked whether they had discussed changing their diet, doing more exercise, reducing their stress, cutting down or quitting smoking and drinking less alcohol (yes/no). Patients' answers on the post-visit questionnaire were matched against the audiotape MEDICODE coding. A recall score was created for medication information, with one point awarded for each correctly recalled medication theme. Patients who did not recall the medication name received a score of zero because the questionnaire was only interpretable if it was clear which medication the patient was referring to. Unintelligible medication names, coded by MEDICODE, were excluded from analysis.

| Statistical analyses
Baseline characteristics and descriptive statistics were analysed using chi-square tests and ANOVAs. A linear mixed model was performed to see what variables predicted recall of medication information.
Medication was the unit of analysis clustered within patients. The dependent variable was the combined score of recall of medication information. Variables theoretically related to recall in Table 1 were tested in univariate analyses. Variables significantly related to recall in univariate regressions (P<.15) were inserted into the final multivariate regression as fixed effects. All tests used an alpha level of significance of .05.
The sample size used here was based on participants from a randomized control trial (n=221). 43

| Socio-demographic and clinical characteristics
A description of the randomization process for the original sample is reported in the primary paper. 43 Figure

| Recall of chronic medical conditions and lifestyle treatment discussions
Recall of whether participants discussed hypertension, diabetes and cholesterol is shown in Table 3. Hypertension, diabetes and cholesterol were discussed in 83.0% (n=183), 60.1% (n=133) and 50.2% (n=111) of the 221 interviews, respectively. Patients recalled approximately 88% of these discussions. Recall of lifestyle treatment is also shown in Table 3

| Recall of chronic disease medication (CDM) information
The 159 participants analysed for medication discussions discussed a total of 401 chronic disease medications (CDMs). Descriptive statistics regarding medication discussions are shown in Table 4. The majority of medications discussed during the encounter were of two main statuses, either actively being taken or represcribed, consistent with a CD population. Discussions were mostly initiated by physicians and had a low dialogue score. Medication discussions were not extensive in terms of the number of themes discussed. The most often discussed theme besides the name was instructions, mentioned in less than half of the medication discussions. Figure 2 shows recall of medication themes. Patients were able to correctly elicit the name of the medication for less than half of the 401 medications just discussed. When patients did not correctly identify the medication name, they were coded as incorrectly recalling the five other medication themes (in patterned red). Patients who did not remember discussing a specific medication theme, despite recalling the medication name, are also shown (block red). In green is correct recall.
Each theme was approximately correctly recalled a third of the time (eg 65 of 193 for instructions). Figure 3 shows the distribution of the number of themes correctly recalled. Patients recalled less than one theme on average (mean=0.97, SD=1.3, mode=0, median=0). Table 5

| DISCUSSION
Recall is an important mediating variable for improved treatment adherence and health outcomes. 4,6,11 This study focused on recall of both lifestyle and pharmacologic treatment information deemed crucial for the optimal management of CD. One of the strengths of this study is its focus on long-standing CD patients followed in PC and not reaching guideline-suggested outcomes. Recall was assessed immediately after the medical encounter.

| Type of recall measure
The heterogeneity of recall measurements has been recognized and has a major impact on the evaluation of recall. 11,12,52 The results of our study confirm this. Correct recall of treatment information in this sample ranged from <34% to 88% depending on what was being recalled and how recall was measured. This range is consistent with the bulk of published research. 6,11,12,19,38,[52][53][54][55] Recall was high for a general question such as remembering discussing hypertension. However, recall of specific discussions, such as exercise and diet in relation to the CD, dropped respectively to 70% and only 50% of patients remembering discussing these issues.
Recall scores are higher for yes-no or multiple-choice-type questions, than for open-ended questions asking patients to elicit information. 56 This has been replicated in our study, where eliciting the medication name had lower recall scores than yes-no questions regarding either specific medication themes or lifestyle issues. Considering that yes-no questions are more easily answered than open-ended questions, it is of potential concern that only half of patients remember whether or not they discussed diet. This makes one wonder how many would be able to remember specific elements of their diet discussion so important in the management of the three CDs this study focused on.

| Recall of medication information
For medication information, the recall score used here was relatively strict, because patients who did not recall medication names were automatically given a score of zero for the other themes. This way of scoring recall may in fact underestimate level of theme medication recall; however, questions were still yes-no as described above. Patients correctly remembered discussing one medication theme, when 2.5 themes were discussed on average. Patients were able to elicit less than half of the names of medications that they had just discussed with their providers. Medication names are frequently the best recalled items. 22,24,53,57 A common concept throughout history and across cultures is the idea that knowing the name of something gives one power over it. 58 Naming, thinking and apprehending reality are intimately related. 59 Medication names are an important way to establish shared language between patients and physicians and are instrumental in empowering patients to reflect upon, understand and manage their health. [60][61][62] Discussions of the six themes deemed most important for a patient with CD to properly take and adhere to their medication do not occur frequently. Less than half of medication discussions include instructions about how to take medications, and adherence is mentioned in only about a fifth of discussions. Although it is possible that FPs did not deem necessary to repeat instructions in these actively taken or represcribed medications, this is questionable considering these patients were not at target. Furthermore, patients only remember about a third of these discussions when they actually do take place.
Although concordant with the literature, the low recall scores of medication information are concerning considering the characteristics of our sample. These patients with CD were off target and in longterm relationships with their FPs. These are precisely the patients who need to be mobilized to properly manage their own care. Patients were generally discussing medications that they were actively taking. Many patients did not even discuss CD medications (47/221) during the audio-taped follow-up appointment. These patients were older than the rest of our sample, and it is troubling to see that medication discussions may be eschewed in older patients. Admittedly, one encounter does not capture continuous care. Yet, considering the low rate of recall, discussing relevant medications at each consultation would seem indicated for off-target CD populations.

| Patient participation and medication information recall
Beyond describing recall of treatment information, an aim of this study was to assess which communication and patient participation variables predict medication information recall. There is a need in the literature to be clear about how we define patient participation. 5,63 In this study, we used two different coding systems with different underlying assumptions about patient participation. RIAS is an interaction-based coding system. It codes each physician and patient speech act into two metacategories: instrumental and socio-emotional. Each speech act is then assigned to a large content domain such as medical condition, treatment, lifestyle or psychosocial. Participation is classically conceptualized in this system as an aggregate of different types of physician and patient utterances, such as physician dominance and communication control, describing a general pattern unrelated to specific content. Participation can also be defined, as we have done in this study, as PACE-like interactions reflecting specific information-seeking and clarification behaviours. MEDICODE, on the other hand, is a contentbased coding system. It first codes for specific content such as medication name, instructions, main effects. Each content element identified is then described in terms of interaction: who, of the two interlocutors, initiated the content, and how much dialogue occurred for that specific theme. Thus, measurement of participation in this second coding system is intimately linked to specific content discussions.
RIAS scores of socio-emotional utterances, physician dominance, communication control and the PACE score, measuring "ask", "check" and "express" utterances, were not found to be predictors of patient In this study, greater patient initiative and greater dialogue about medications, measured with MEDICODE, were predictors of patient recall. A known cognitive phenomenon, the "generation effect," stipulates that active involvement in producing information improves recall compared to passive reception. 65 Authors suggest that this effect may be due to an egocentric bias and a more accurate monitoring of one's contributions to the production of content. [66][67][68][69] When there is patient initiative and dialogue, patients produce information and are actively involved in the discussion. This finding lends importance to developing skills in improved information exchange. 70 Unfortunately, providers infrequently use recall-promoting techniques that aim to involve the patient, such as "teach-back". 10,54,71 Our sample of patients with CD who were familiar with their FPs benefited from greater extent of information, replicating findings from other studies. The more medication themes were discussed, the better patients recalled medication information. Other studies have found information provision to be an important predictor of recall. 27,42 However, too much information can overwhelm patients, 23,26,55 an effect that may be mitigated by an established relationship with FPs.
Discussing multiple themes surrounding a treatment deepens and adds density to the conversation. This extensive approach to the discussion can enable more conscious involvement.
Of note, new medication prescriptions were better recalled than active prescriptions, which may explain why Tarn et al. 55  Street has described three ways of assessing the quality of information exchanges. One is through message properties, such as the content and form of communication. One is through the process of co-construction of messages, and the third is through the outcomes of the information exchange. 73 We have included all three ways of measuring this information exchange, using RIAS and MEDICODE to assess the content and form, using dialogue and initiative scores to assess co-construction and by assessing recall as the outcome of the information exchange. This study shows the importance of having an information exchange that covers essential information about medications. Furthermore, information cannot simply be passively transmitted, as a physician initiated monologue, but needs to be co-constructed with the patient.

| Generalizability
This study had a heterogeneous multimorbid CD patient sample in

| CONCLUSION & PRACTICE IMPLICATIONS
The aims of this study were to describe treatment information recall and to assess which communication and patient participation variables are related to medication information recall for off-target CD patients. Providers cannot take for granted that long-term off-target CD patients know and remember treatment information. This is particularly relevant considering the ageing demographics of PC populations. Elderly patients were less likely to discuss their CD medications and had poorer recall when they did. Further research is needed to understand the dynamics between ageing patients, medication discussions and recall. There is a lack of shared language and empowerment in managing one's disease. Encounters with patients with CD need to engage in an explicit conversation about relevant lifestyle modifications and medications. 70 This conversation should ideally be thorough and engage patients in a true dialogue. Providers and patients need to engage more frequently about treatment information in ways that encourage patients to be active participants in the discussion. Improved discussions may impact treatment information recall. This is an important step towards improved self-management, adherence and eventually better health outcomes.

CONFLICTS OF INTEREST
This project was sponsored by Astra Zeneca Canada. The sponsor had no role in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the manuscript for publication.