Slicing it thin: New methods for brief sampling analysis using RIAS-coded medical dialogue

Abstract

Objective

To explore the relationship between one-minute slices and full-session interaction and the predictive validity of the slices to ratings of affect and rapport.

Methods

Third-year medical students (n = 253) were videotaped during an OSCE. All interaction was coded using the Roter Interaction Analysis System (RIAS) and samples were drawn at minutes 1, 5, and 9 and extracted from the coded database. The slices were related in multivariate analysis to full-session interaction, corrected for slice content, and correlated with affect ratings of participants and independently rated judgments of rapport.

Results

One-minute slices explained 33% of full-session variance in student interaction and 30% of variance in standardized patient interaction. Slices were significantly correlated with affective ratings of participants and independent judgments of rapport in a similar pattern as full-session interaction analysis.

Conclusions

One-minute slices of interaction can provide a meaningful degree of insight into OSCE session communication with both concurrent and predictive validity to ratings of session affect and rapport.

Practice implications

Evidence of concurrent and predictive validity further supports use of this approach as a research tool that provides an efficient means of analyzing processes of care, examining variation in communication throughout a visit and predicting visit outcomes.

Introduction

Surprisingly small amounts of information derived from direct observation of behavior, often called “thin slices,” have been shown to accurately provide insights into the attitudes and emotions of individuals and the nature of the relationship between interactants [1], [2]. The assumption underlying thin slice analysis is that brief observational samples capture representative elements of the ongoing behavior stream and therefore provide a window into the totality of interaction [3].

A handful of thin slice studies have been conducted in the medical context using brief slices of medical interaction to predict a variety of outcomes. The majority of studies have focused on prediction of patient satisfaction and utilization [4], [5], [6], [7], [8], [9], but others reported correlates of interpersonal rapport [10], therapeutic efficacy [11], and physicians’ malpractice history [12].

Thin slices of communication behavior range in duration from less than one second to several minutes long. The thin slice samples are drawn in a variety ways, sometimes purposively throughout a visit, for instance, the first or last minute of a session or at set points like the middle of each third of the visit or randomly. Audio or video channels may be used in combination as the behavior would naturally appear, or the channels may be separated or manipulated in some way. The audio channel is sometimes altered by passage through an electronic bandpass filter that renders the literal content unintelligible but retains the vocal qualities that convey emotion. In some experiments the audio channel is completely silenced and only text, or silent video, is used. Typically, thin slice segments are rated by judges asked to assess the emotion or attribute conveyed in the communication [2].

In our previous study, trained observers rated rapport on three one-minute slices of medical student interaction during an OSCE with a standardized patient (SP) [10]. While the rapport ratings were based solely on the first minute, fifth minute, and ninth minute of the session, the ratings were related to interaction analysis based on the full OSCE session, coded with the Roter Interaction Analysis System (RIAS) [13]. We reported that medical students rated high on rapport were also rated as affectively positive by RIAS coders, gave relatively less medical information and used more positive statements. The SPs in high-rapport visits were similarly rated as affectively positive, were verbally active generally but asked relatively few questions. Notably, rapport ratings based on just the first minute of the session correlated with interaction almost as well as rapport ratings based on the combination of first, fifth, and ninth minutes. This finding suggests that the interaction patterns indicative of rapport manifest as early as the first minute of the session and may be predominantly conveyed through nonverbal channels associated with particular interaction categories.

This line of work was extended in several ways, including a novel approach to interaction analysis using thin slice segments of interaction. To this end, we explored the extent to which RIAS-based interaction reflected in the three one-minute slices described above relate to full-session interaction. We hypothesized that one-minute interaction slices would show a consistent pattern of correlation to full-session interaction and that these relationships would be substantially enhanced by additional slices.

Secondly, we examined the relationship between the interaction slices and global affect ratings hypothesizing that one-minute interaction slices would produce a similar pattern as full-session interaction in relation to ratings of medical student and standardized patient affect.

Finally, we explored the predictive validity of the thin slice segments by correlating them to an independent measure of rapport, hypothesizing that the one-minute interaction slices would show a similar pattern of relationship to rapport ratings based on full-session interaction.