Development of tools for accurate measurement of communication is in its infancy. Many such tools are based on traditional linguistic structures and are primarily descriptive of verbal content during a consultation or interview (Cecil, 1998; Shaikh, Knobloch, & Stiles, 2001). Observational schemas have developed some complexity over the years, but few depict process-based explanations and predictions of outcomes. Process models of communication are fundamental to understanding interactional behaviors and outcomes because they can account for static and dynamic factors (Bales, 1976; Berger & Calabrese, 1975; L. E. Rogers & Escudero, 2004). These types of communication models include individual, joint, and situational contributions to communication transactions as they unfold.

An important feature of communication process perspectives is the idea that communicators, through verbal and nonverbal communication, can convey multiple layers of messages. Early communication theorists conceptualized this idea as the content and relational levels of communication (Watzlawick, Beavin, & Jackson, 1967). In this sense, communication can be understood as providing two streams of meaning: one that regards the denotative content of a message and another that reflects the affective tone, and therefore the relational quality of the interactants. An observational scheme that captures both these streams of meaning may reveal important relationship dynamics than that revealed by a single analytic approach.

This article describes a computerized program to facilitate the analysis of health care conversations, the Siminoff Communication Content and Affect Program (SCCAP). Because health transactions are widely variable in form and function, this new program is designed to be adaptable to the goals and patterns of diverse health care contexts. The SCCAP builds on other well-known observational systems but emphasizes verbal and nonverbal communication behaviors drawn from the communication research literature. Specifically, the program captures the following: (a) task-driven information exchange among multiple interaction partners, (b) the affective and relational communication activities of all communicators, and (c) the social-influence tactics used in health care settings that contribute to decision making. Data are reported from initial tests using two different health communication settings: tissue donation telephone requests and oncologist-cancer patient treatment decision making. These different health care contexts tested the flexibility and adaptability of the SCCAP to capture multiple health communication scenarios.

The interaction process analysis provided a foundation for the Roter Interaction Analysis System, perhaps the most ubiquitous of communication coding schemes used in health communication research today (Roter & Larson, 2001). The system was designed to describe physician–patient interaction in terms of content and context of routine dialogue during medical care. General functions of the medical interview provide categories for coding physician–patient communication behavior. The first two categories, data gathering and educating/counseling, reflect general medical interview tasks. The second two categories, relationship building and partnership, are more affective in tone and reflect a broad conceptualization of nonverbal communication. This relational dimension of the Roter Interaction Analysis System accounts for a comparatively small segment of data and is often excluded from analyses or operationalized differently from study to study.

The Roter Interaction Analysis System has influenced the development of many communication coding schemes, including the SCCAP. Most of these offer twists on coding the content of the interaction but tend to minimize other communication functions. Gillotti, Thompson, and McNeilis (2002) used communication competence theory to create the Coordination and Competence System, which offers an approach to measuring participant responsiveness firmly grounded in conversational alignment. The Coordination and Competence System provides an excellent description of information exchange (i.e., “fit”), but stops short of capturing the affective component of relationship building. The Medical Interaction Process System accounts for some communicated affect via ratings of each utterance for “feeling” (Ford, Hall, Ratcliffe, & Fallowfield, 2000), but does not ground this activity in theoretical explanations of relational development. Dent, Brown, Dowsett, Tattersall, and Butow, (2005) revised an earlier coding scheme, the CN-LOGIT (Butow, Dunn, Tattersall, & Jones, 1995), to create the CanCode system, which includes ratings of discrete emotional content (e.g., anxious, sad) as well as various medical interview functions, similar to those in the Roter Interaction Analysis System. Although these coding systems are more descriptive of content than past coding schemes, these observational methods still lack systematic explanation and integration in the communication process.

One exception, an observational scheme developed by Street and colleagues, better captures the interrelationships of communication content and affect. This approach, similar to the Roter Interaction Analysis System, positions patient participation as a central mechanism for linking physician communication and measurable outcomes (Street, Gordon, Ward, Krupat, & Kravitz, 2005; Street & Millay, 2001). This coding scheme moves closer to fulfilling what Rimal described as “a typology of participation” (p. 96, Frankel, 2001).

Although theorists have offered several perspectives of relational communication, we focused on confirmation, vocal immediacy, and the more global construct of affiliation as important communication indicators of relationship (Andersen, Guerrero, Buller, & Jorgensen, 1998; Jones & Guerrero, 2001; Kiesler & Auerbach, 2003; L. E. Rogers & Farace, 1975; Watzlawick et al., 1967; Wilmot, 1980). These constructs reflect relational qualities derived from interpersonal theory and validated by research. Further, these constructs can be observed and measured with vocalic cues associated with speech. First, confirmation theory is used to conceptualize verbal relational messages in the SCCAP.

SCCAP is an application designed to accomplish coding of verbal conversations from audio. The audio portion is not integrated into the SCCAP; the researcher listens to and uses his or her audio data directly from his or her own audio equipment. This is advantageous because transcription is costly and time consuming. Nevertheless, researchers can choose to transcribe and use the written transcription to aid in the coding process if they so choose. The SCCAP program is designed for audio only, rather than audio and video. Although nonverbal communication presents a multitude of observable cues audio retains nonverbal vocal inflections that impart meaning to the spoken words. Some researchers have argued that inclusion of kinesic information does not necessarily add significant explanatory power to coding only vocalic information (Dent et al., 2005) and that vocal cues are a more reliable and valid channel for coding nonverbal cues (DePaulo, Rosenthal, Eisenstat, Rogers, & Finkelstein, 1978; Haskard, DiMatteo, & Heritage, 2009). Other recent work (Penner et al., 2007; Riddle et al., 2002) has indicated that nonverbal, observable behaviors have much to add to an understanding of a particular communication. However, recording of nonverbal data requires videotaping, which is often expensive or impracticable. Therefore, to maximize feasibility, reliability, and coding parsimony, our version of SCCAP is audio only. The flexibility of the program is such that future versions could include additional nonverbal coding sections (e.g., kinesics).

The program opens and runs from a main menu that offers several coder activities. These reflect three general sets of data. The first set, content themes, includes those activities that constitute the task or instrumental aspect of most medical transactions (e.g., providing treatment information). Content themes are delineated into general categories and then further refined into discrete communication behaviors or events (e.g., discuss treatment, side effects). Coders click each event or activity as it occurs in the interaction and the program automatically records speaker, topic, message form (statement or question) and sequence. The unit of analysis that the SCCAP developers have used to test the program is an utterance, that is, the smallest definable unit of meaning. We trained all coders to identify an utterance with accuracy and calculated reliability scores on that basis. However, the idea of the SCCAP was to provide researchers with as much flexibility as possible. Researchers need to choose their own unit of speech analysis as dictated by their theoretical framework and research questions. A highly focused research question examining specific content might only code every instance of a specific content item. Conversely, a study taking a holistic or linguistic approach might need to code every utterance. Researchers can decide how much communication data they need and use the program to obtain their desired level of detail as long as the coding unit of analysis is clearly defined and the coding is consistent.

The second set consists of communication types. These are the aspects of communication that indicate relational information or influence attempts. Within this group are nested additional menus for recording question types. Content themes and communication types are coded at the same time. As the coder assigns a specific content theme to an utterance, she toggles to another coding screen that offers a menu of various communication types. After coding, communication types can be analyzed as discrete entities or by how they are associated with the content codes.

The third set consists of observer speech and affect ratings, including emotions (e.g., anger, sadness) and more composite affect (e.g., composure). Coders are trained to observe nonverbal vocalic cues likely to indicate affective qualities of interaction (Frijda, 1989). Speech ratings include those cues associated with the immediacy construct, such as speech rate and timber (Bradac, Bowers, & Courtright, 1979; Kearney, 1994). Coders rate each participant after listening to and coding content aspects of the interaction. Each of these primary data sets—content themes, communication types, and speech/affect ratings—are subsequently described in more detail.

A second and more detailed examination of SCCAP was used to code communication in a sample of 1,200 requests for tissue from the families of deceased patients. The parent study was designed to understand how consent for tissue donation is obtained. Audiotapes of these conversations were obtained from 16 tissue banks in the United States and include next of kin who chose to donate as well as those who refused donation. Audiotapes were accessed after permission was obtained from deceased patients’ families. Five coders were originally trained to use the SCCAP application and reliability was assessed with a sample of 50 of the 1,200 audiotapes, randomly assigned to two coders who independently rated each audiotape.

The mean total number of utterances per consultation for Coder A was 177.30 (SD = 152.12), whereas the mean number for Coder B was 184.55 (SD = 152.70), Cronbach’s α =.998, p < .001. Interrater reliability for each content category, examining the tissue requester and next of kin utterances combined, ranged from 0.92 to 1.00 (see Table 3). For tissue requester communication, the interrater correlations ranged from 0.84 to 1.00. For next of kin communication, the interrater correlations ranged from 0.82 to 0.99. We also examined difference scores between Coder A and Coder B and these are presented in Table 4 and also indicate excellent interrater reliability.

Further examination of the SCCAP’s ability to enable prediction of an outcome of interest was conducted. Using all 1,200 coded tissue conversations, we tested predictors on family consent to tissue donation. As we originally hypothesized, we found that more requester affect was positively associated with consent to donation (31.1% vs. 29.1%; p < .001). Specifically, requesters who exhibited more involvement, sincerity, friendliness, and verbal expressiveness were more successful at obtaining consent to donation. Requesters who used disconfirmation were more likely to encounter family refusal of donation than consent (29.8% vs. 22.8%; p < .01). We also found that families were significantly more likely to consent when the requester did not cut off the family member when speaking (81.5% vs. 18.5%; p < .001) and when family members disclosed personal information about themselves (80.9% vs. 19.1%; p < .001).

Another test of the validity of the results obtained by SCCAP analysis was obtained by coding conversations between early stage breast cancer patients and their medical oncologists (Step, Rose, Albert, Cheruvu, & Siminoff, 2009). The sample consists of 39 oncologists, 180 patients, and 137 patient family members. Table 5 shows how much talk surrounded each function of the oncology interviews. Consistent with other health communication coding schemes (Frankel, 2001; Roter & Hall, 1993), medical management talk dominated the interaction. Also consistent with other studies (Frankel; Roter & Hall; Siminoff, Graham, & Gordon, 2006), oncologists talked more than patients about all topics with the exception of care logistics (i.e., how to get care). Psychosocial content (i.e., topics concerning, the patient’s lifestyle) and emotional topics were discussed but heavily dominated by oncologists.

An important feature of the SCCAP is that it also offers measures of relational communication tied exclusively to communication theory. This provides a means of describing sequenced cognitive and affective communication behaviors that can be analyzed at individual, dyadic, or group levels. We are currently planning analysis of the location and sequence of confirming and disconfirming communication to explore those effects on consult duration and patient participation in patient-physician encounters. This will allow us to find out whether physicians’ consistent use of relational messages (i.e., across content domains) has the same effect on select patient outcomes as does strategic use of relational messages (i.e., in response to patient concerns).

Building any coding scheme is a challenging endeavor and this one is no exception. Although reliability is generally high, training to obtain adequate reliability is lengthy, generally 2–4 weeks. Researchers should consider training challenges carefully when designing content templates. Longer coder training will be required for more nuanced content templates. The SCCAP allows each element to be independently analyzed and researchers are free to tailor what parts of the programs are useful for their research question.

Another important caveat is the need to use large samples in order to control subject differences within clinician samples. Observation research in medical contexts typically features multiple patients for each clinician in the sample. It is important to offset potential errors by systematic random sampling of clinicians in multilevel designs or controlling for random effects of clinician communication styles in analyses. Future work should offer examples of how sequenced data can be used, as well as how SCCAP can be used to model more complex communication processes.