Medication management (1) appointments are psychiatric visits used to treat drug-responsive psychiatric disorders (2). They are the principal clinical service provided by U.S. psychiatrists during outpatient care. Medication management appointments last between 15 and 20 minutes, and patients in ongoing care are seen every one to three months. Currently, medication management—provided in the context of split treatment that incorporates psychotherapy by a nonpsychiatrist clinician (3–5)—dominates outpatient psychiatric practice in the United States (6,7).
Stopping recommended treatment before completion (8,9) or before improvement (9,10) is cause for concern. Treatment dropout can signal a decline in medication adherence (11) and an increase in the probability of hospitalization within the next 12 months (12,13).
Most effective treatment adherence interventions target patient behaviors (14,15). Another approach that has received little attention is enhancing verbal and nonverbal communication behaviors among psychiatrists. There is some evidence that appointment length, patient-centered communications, and positive affect by the provider during psychiatric and medical visits are related to greater patient appointment adherence.
Appointment length has special importance in psychiatry, which has been viewed as the most time-intensive medical specialty (16). In one study of a nationally representative sample of visits to office-based psychiatrists between 2001 and 2006, visits by African-American patients were significantly shorter than visits by whites (17). Considering that African-American patients have higher treatment dropout rates than white Americans (9,18,19), shorter visit duration could be one factor that negatively influences patient appointment adherence.
Patient-centered communications focus on eliciting patients’ psychosocial and general medical needs, encouraging them to disclose their concerns, conveying a sense of partnership in treatment, and actively facilitating patient involvement in treatment decision making (20,21). This style of communication has been shown to improve provider and patient satisfaction as well as the continuity of care (22,23) and patient outcomes (24).
A meta-analysis of visit communications during medical appointments found that when the doctor offered more information, asked fewer questions overall but more questions about adherence in particular, and was more positive and less negative (both verbally and nonverbally), subsequent appointment keeping was greater (25). In addition, several measures of conveyed affect in physicians’ voice tone have been shown to be related to patient appointment adherence (26,27).
Our recent studies of psychiatrist communication style during medication management appointments suggest that the interpersonal and communication dynamics of these visits are similar to those of primary care visits. For example, in our earlier work we found wide variation in psychiatrists’ communication styles and expression of affect during routine medication management appointments (28,29). There was a notable difference during these visits between the communication styles of psychiatric residents and attending physicians. Visits by residents were twice as long as and were significantly more patient-centered than visits by attending physicians. Global ratings of affect revealed that residents were perceived as sounding significantly more positive (friendly and sympathetic) and less dominant and rushed.
The objective of this study was to explore the relationship between appointment length and psychiatrists’ communication behaviors in the context of medication management visits and subsequent patient appointment keeping. More specifically, we hypothesized that longer visit length, more patient-centered communication, and more positive conveyed affect would be associated, together or separately, with higher rates of patient appointment keeping.
Unique psychiatrist-patient dyads were audio-recorded during visits falling between the patient’s first and sixth appointment at four ambulatory mental health clinics located in the Midwest. These clinics were part of a large, urban mental health care system affiliated with a university.
Psychiatrists were eligible if they practiced at a participating study clinic. Patients were eligible if they were 18 to 65 years of age, had a chart-recorded DSM-IV-TR diagnosis of major depressive disorder, depressive disorder not otherwise specified, or dysthymic disorder, were in treatment with a participating psychiatrist, and had been seen by their psychiatrist for their first appointment but had not attended more than five appointments. We limited diagnosis to depression to control diagnostic variation, and we limited patient participation to the second through fifth appointments because patients are more likely to drop out of treatment around the time of these early appointments (9,10,30).
Psychiatrists were recruited through presentations and individual meetings. Patients were either self-referred in response to a flyer or introduced to the study by their therapist or psychiatrist. Eligible patients gave consent and were enrolled in the study prior to a routine medication management visit at the clinic. Patients received a $10 grocery store gift card for study completion. Psychiatrists received no compensation. The study, including the informed consent form, was approved by the University of Pittsburgh Institutional Review Board. Recruitment occurred from March 2005 through May 2008.
At the start of the appointment, in front of the psychiatrist and the patient, research staff activated the recorder and left the office. For appointment keeping, we collected information from patients’ health records about patient attendance at the next scheduled appointment.
The appointment recordings were quantitatively analyzed by using the Roter Interaction Analysis System (RIAS) (31). The original deidentified recordings were sent to the RIAS communications laboratory at the Johns Hopkins Bloomberg School of Public Health. Coded analytic data were returned via encrypted electronic mail and merged with all other study data for final analysis.
Analysis of visit communication.
Visit communication was analyzed by using the RIAS, a system for characterizing medical dialogue that is widely used and that has well-established reliability and predictive validity (31). Each statement by the patient and psychiatrist defined as a complete thought was assigned to one of 41 categories of speech. Statements were coded directly from recordings without transcription, allowing consideration of the emotional context of the visit.
As in many studies using the RIAS, the large number of specific codes was reduced by combining codes into broad common groupings. In this study, seven groupings were created to construct the patient-centered talk variable: biomedically focused questions regarding treatment response, medication side effects, psychiatric history, and other medical problems; psychosocially focused questions regarding social and family relations at work and at home, performance of activities and functions related to daily living, and exchanges related to feelings; biomedical-focused information regarding psychiatric history, symptoms, therapeutic regimen, and other medical problems; psychosocial-focused information regarding social and family relations at work and at home, performance of activities and functions related to daily living, and exchanges related to feelings; rapport-building talk, such as reflecting laughter, compliments, and comments showing agreement or understanding and approval; partnership-building talk, such as asking for opinions and beliefs, asking for clarification of fact, and checking for understanding; and facilitators of speech, such as showing agreement and understanding, paraphrasing, checking for understanding, providing transitions, giving orientation, offering instructions, and making back channel statements, such as “uh-huh.” To adjust for variation in total talk associated with appointment length, we calculated the proportion of total speaker talk for each RIAS communication category.
After the groupings above were constructed, we then calculated the patient-centered talk variable. This variable was calculated, as in other communication research studies using the RIAS, as the ratio of the sum of psychosocial, rapport-building, and facilitative behaviors to the sum of biomedical questions, information giving, and closed-ended questions by clinicians and patients. A score >1 indicates that the visit tended to further the patient’s agenda, and a score <1 means the visit tended to further the clinician’s agenda (32–34).
Appointment length was calculated in minutes from the first to the last statement identified by the RIAS coders during the audio recording of the appointment.
In addition to analyzing verbal behaviors, RIAS coders used a 5-point scale (1–5) to rate the emotional tone conveyed by the psychiatrist or the patient on a variety of affective dimensions. A low value suggests minimal presentation of this affective dimension in the provider’s voice tone, while a high value suggests extensive presentation of this affective dimension. We performed a factor analysis to determine if the variability of the psychiatrist voice tone ratings was a consequence of a few unobserved variables. Factor analyses search for such joint variations in response to unobserved latent variables. Our factor analysis revealed a two factor solution. One factor was comprised of voice tone ratings that have been referred to in the extant literature as conveying positive emotions (dominance-assertiveness, which is reverse-coded, so high dominance expression is coded low; interest-attentiveness; friendliness-warmth; responsiveness-engagement; sympathetic-empathetic; interactivity; and respectfulness). We calculated the mean score for the positive voice tone rating factor and entered it as an independent variable in the main analyses.
Two experienced RIAS coders, unaware of the study hypothesis, analyzed the recordings. Interrater reliability for a random sample of ten recordings ranged from .85 to .97 (Pearson correlation coefficients) across the categories used in this analysis. Coder agreement on ratings of affect within 1 point on the 5-point scale ranged from 80% to 100%.
Patient Health Questionnaire−9.
Because there is some evidence from the psychotherapy literature that depression severity has an inverse relationship with treatment dropout (35), depression severity was measured with the Patient Health Questionnaire−9 (PHQ-9) depression module (36). The PHQ-9 is self-administered, and each DSM-IV criterion for depression is rated from 0, not at all, to 3, nearly every day. Possible PHQ-9 scores range from 0 to 27, with higher scores indicating more severe depression.
Patient attendance of the next scheduled appointment was extracted from patients’ health records. Instances in which the patient rescheduled the appointment and kept it were counted as a kept appointment.
Background and sociodemographic characteristics.
Patient background and sociodemographic characteristics were assessed immediately prior to recording the medication management visit.
The main independent variables for the analyses were appointment length, patient-centered talk ratio, and positive voice tone ratings. The dependent variable was kept or missed appointments.
We performed t test and chi square analyses to assess the relationship between appointment adherence and confounding variables that have been suggested to influence appointment adherence. They were psychiatrists’ gender (37–42); patients’ gender (25,43–45), PHQ-9 score (35), income (35,46–48), health insurance (46,47), and race-ethnicity (17); psychiatrist-patient race concordance (32); and the clinic where services were received (35). Confounding variables with p values <.1 were included in the adjusted multivariate analyses.
We performed logistic regression analyses, both adjusting and not adjusting for the influence of confounding variables, by using generalized estimating equations (GEEs). GEEs were preferred over traditional regression because they accounted for the clustering effects of any within-psychiatrist correlation and the different number of recorded psychiatrist-patient dyads for each psychiatrist (49,50). We assumed an exchangeable correlation structure and employed the Huber-White sandwich estimate of variance, which provides valid and robust standard error (SE) estimates, even if the correlation structure is misspecified (51). Because odds ratios overstate the probability of frequent events, we present estimated unadjusted and adjusted probabilities for logistic regression analyses instead. In our analysis, adjusted probabilities compared patients who kept versus those who missed their next appointment with our main independent variables, while holding all covariates in the model constant, either by using a mean (continuous variables) or the average probability of being in a particular category (dichotomous variables). This allowed us to compare two otherwise equivalent patients on the basis of the available data.
To test that the effect of longer appointments, more patient-centered talk, and more positive voice tone ratings on appointment adherence all equaled 0 (our null hypothesis), we performed a two-tailed multiple coefficient Wald chi square test after fitting our full model. Thereafter, single-coefficient Wald chi square tests were performed on any of our hypothesized variables that were shown by GEE to have a significant relationship with appointment adherence (p<.05). All statistical analyses were performed by using Stata software, version 12.1.
We recruited three community-based and one mood disorder research clinic. The psychiatrists’ role in the participating sites was limited to pharmacotherapy in split treatment appointments. At all clinics, psychotherapy was provided by master’s-level clinicians.
Two clinics provided a majority of the 83 recordings (N=65, 78%) used in the final analysis. One clinic recruited more white (N=15, 65%) than African-American (N=8, 35%) patients. This clinic was staffed with both attending and resident psychiatrists. The other clinic principally recruited African-American patients (N=37, 88%) and was staffed by resident psychiatrists under the supervision of an attending psychiatrist.
The additional two clinics predominantly recruited white Americans (N=8, 89%, and N=9, 100%, respectively) and were staffed by attending psychiatrists.
Psychiatrist and patient recruitment
Fifty-two psychiatrists were approached, 41 (79%) consented to participate, and 24 (46%) were recorded. Eleven psychiatrists who consented were not recorded because they did not have any patients within the study period who met inclusion criteria.
Of the 150 patients approached, 130 (86%) signed consent, and 91 (70%) were recorded. Thirty-nine of the patients who consented did not return. Six recordings were excluded because of inaudibility, one recording was excluded because the patient did not meet diagnostic inclusion criteria, and one recording was excluded because the findings from this recording had an unusually strong influence on the overall study findings (it was a high-leverage point) (51). The final analysis included recordings of appointments for 83 patients, representing 64% of the patients who consented.
On average, psychiatrists were middle-aged (mean age=38, range 27–60 years) (Table 1). A majority were male (63%), white (71%), married (71%), and residents-in-training (67%). The median number of recordings per psychiatrist was three (range 1–11).
Table 1Characteristics of 83 patients and 24 psychiatrists who participated in audio recordings of medication management appointments
| Add to My POL
| Range (years)||24–65|
| African American||46||55|
| Marital status|
| Employment status|
| Insurance status|
| Less than high school||17||21|
| High school diploma or GED||31||37|
| Range (years)||27–60|
| African American||2||8|
| Marital status|
| Psychiatric status|
The mean age of patients was 45 (range 24–65 years), and most were African American (55%), female (83%), unmarried (89%), and unemployed or disabled because of a mental disorder (74%). The mean±SD PHQ−9 score was 13±7.
Appointment length was 22±12 minutes (range 4–60). Patients’ appointment adherence was 63%±5%. Twenty-five (30%) appointments were for race-concordant white dyads, and five (6%) were for African-American dyads. Scores were 3.17±2.44 for patient-centered talk and 3.93±.47 for positive voice tone rating.
Analyses of independent variables for patients who kept versus missed follow-up appointments were significant for patient race (χ2=12.74, df=1, p<.001), clinic (χ2=15.67, df=3, p=.001), psychiatrist gender (χ2=8.00, df=1, p=.005), psychiatrist-patient race concordance (χ2=11.58, df=1, p=.001), and patient income (χ2=4.73, df=2, p=.09) (Table 2).
Table 2Characteristics of 83 patients and 24 psychiatrists who participated in audio recordings of medication management appointments, by adherence to a follow-up appointment
| Add to My POL
|Kept appointment||Missed appointment|
|PHQ-9 score (M±SD)b||52||12.0±6.6||31||14.6±7.9||.12|
| Race concordance||26||87||4||13|
| Race discordance||26||49||27||51|
| African American||21||46||25||54|
| Medicaid or Medicare||27||61||17||39|
Unadjusted and adjusted GEEs are presented in Table 3. The adjusted analyses showed that positive voice tones conveyed by the psychiatrist were significantly and positively associated with appointment adherence (β=1.83, SE=.87, 95% confidence interval [CI]=.13–3.53, p=.03). Our probability estimator (β) informed us that for a .47 increase (one SD above the mean) in positive voice tone ratings, there was a corresponding increase of 162% (1.83 SDs) in appointment adherence.
Table 3Relationship between characteristics of a medication management appointment and adherence to a follow-up appointment, by unadjusted and adjusted GEEa
| Add to My POL
|Characteristic||β||SE||95% CI||p||β||SE||95% CI||p|
|Appointment length||–.01||.02||–.06 to .03||.51||–.04||.03||–.10 to .02||.18|
|Patient-centered talk||.06||.06||–.07 to .18||.40||.13||.10||–.07 to .34||.20|
|Positive voice tones||1.00||.45||.12 to 1.89||.03||1.83||.87||.13 to 3.53||.03|
Wald chi square analysis of independent variables of interest together revealed that the null hypothesis could not be rejected. In tests of single coefficients, the null hypothesis was rejected for positive voice tone ratings (χ2=4.47, df=1, p=.03) but not for appointment length and patient-centered talk. This analysis confirmed the positive and significant relationship between positive voice tone ratings and appointment adherence.
Consistent with our hypothesis, positive psychiatrist affect was significantly associated with appointment adherence. Contrary to our hypothesis, logistic regression analyses showed that medication management appointment length and patient-centered talk were not significantly associated with appointment adherence, even after adjustment for patient race and income, psychiatrist gender, race concordance, and the clinic where services were rendered. We believe that the study findings are novel and may be important in understanding how therapeutic alliance influences treatment adherence. A positive therapeutic alliance in mental health care has been shown to be related to retention in treatment for drug (52,53) and alcohol abuse (54) as well as in family therapy (55). Although previous articles detailing core features of the therapeutic alliance have focused specifically on providers’ verbal communicative style, our findings highlight the potential importance of providers’ implicit communication behaviors.
We hypothesize two potential reasons for our negative findings. First, because our study observations consisted of split treatment medication management appointments, the results may have been affected by a misalignment between patient and provider expectations. Patients may have expected the appointment to focus solely on pharmacotherapeutic issues. However, psychiatrists may have expected their role in care to include a discussion of psychosocial issues (a more patient-centered approach), with appointments lasting longer than patients expected. This misalignment of expectations could have resulted in patients’ devaluing the appointment and could have negatively influenced their decision to attend future appointments.
Second, psychiatrists spend more time and utilize more patient-centered talk with problematic patients. Problematic patients are less likely to attend regularly scheduled appointments. Examples of problematic patients are persons with complex biopsychosocial issues, recent histories of suicide attempts (56), or frequent hospitalizations (12,13).
Our study had several strengths. First, this is the only study known to the authors that assessed the relationship between psychiatrist verbal and nonverbal communication behaviors during medication management appointments and patient appointment adherence. Second, our patient sample was made up of persons with diverse racial-ethnic and socioeconomic characteristics. Third, we recruited patients and providers from multiple clinics that are common within the mental health care system.
This study had a few limitations. First, one limitation common to all communications research is observation bias. The presence of recorders in appointments may alter the dynamics of care and affect how providers and patients communicate. In this study, concern about observation bias was tempered for two reasons. Several studies in the field of patient-provider communication have shown that videotaping of clinical encounters, which is more intrusive than making an audio recording, does not systematically alter physician communication or patient satisfaction (57,58). In addition, even if psychiatrist behavior is affected in some way, it would likely reflect best effort. In this circumstance, the study findings would have even greater implications for patient care.
Second, we were unable to assess a possible causal relationship between positive psychiatrist affect and patient appointment adherence because of the cross-sectional nature of our study and the lack of an assessment of the relationship between psychiatrist and patient affect. For example, patients who demonstrated positive nonverbal behaviors during the appointment may have had a greater likelihood of returning for future appointments. Also, their behavior may have caused their provider to demonstrate positive nonverbal behaviors. One approach to addressing the causal question would be to perform a randomized controlled prospective study testing the impact of a psychiatrist-directed positive nonverbal training enhancement intervention on patient appointment adherence. Third, patients were selected via provider referral or self-referral. It is possible that differences in the characteristics of a group selected in this way and of the general population of patients coming to these clinics may be relevant to this research. Last, the null finding related to patient-centered talk could be due to the inherent time constraint of medication management appointments.
Our study found that psychiatrist vocal qualities that convey friendliness, warmth, and empathy—an overlooked factor in the mental health services adherence research literature—may be important to improving appointment adherence among African-American and white patients with depression. This finding has potentially significant implications in terms of our understanding of therapeutic alliance and the development of psychiatrist-directed, evidence-based, positive nonverbal behavior training interventions that could improve outpatient treatment adherence rates.
This study was supported by grants P30 MH030915, P30 MH07944, R25 MH60473, and K23 MH071520 from the National Institute of Mental Health. The authors thank all of the psychiatrists and patients who participated in this study.
Dr. Roter is the author of the RIAS and holds the copyright for the system. Johns Hopkins University also has rights to enhancements of the system. Dr. Roter and Ms. Larson are coowners of RIASWorks LLC, a company that provides RIAS coding services for projects undertaken outside Johns Hopkins University. RIASWorks may benefit indirectly from dissemination of the current research. Neither author nor Johns Hopkins University collects royalties for use of the system in research. The other authors report no competing interests.