Provider interaction with the electronic health record: The effects on patient-centered communication in medical encounters

2.1. Research setting and participants

Twenty-three care providers (PCPs) (21 physicians and 2 nurse practitioners; 74% female) at four VA primary care clinics in the San Diego area were recruited to participate in the project. The VA San Diego primary care clinics use the VA's paperless EHR system called VISTA Computerized Patient Record System (CPRS^®). The CPRS includes all provider chart notes, pharmacy data, laboratory data, radiology reports and images, and patient demographic data. It has computerized patient order entry (CPOE), clinical decision support, and basic key word search capability. All participating VA providers have undergone training in CPRS software and use it for clinical care via traditional desktop PC interface.

The study enrolled 125 established patients scheduled for routine follow-up visits, with up to six patients per PCP. Inclusion criteria for patients included the following: 18 years of age or older, must not have a significant communication disability, must be mentally competent to provide informed written consent, have an established relationship with their provider (a minimum of 2 prior visits within a period of 12 months), and have a minimum of 2 primary care clinic visits/year, based on historical clinic data. The study was approved by the UCSD Institutional Review Board.

Patients willing to participate in the study agreed to have their consultations video-recorded and, following the visit, complete a survey consisting of several post-consultation measures, including their satisfaction with their health care [13].

2.2. Nonverbal measures of EHR activity

Provider gaze at the computer was captured through analysis of video recordings of the interaction. Gaze behavior was coded directly from digital video clips by two trained coders using video coding software (Inqscribe^®). Time-stamped data for each visit was then exported for aggregate analysis (e.g., average time for each gaze behavior/consultation). Inter-rater reliability was performed via independent double coding of randomly selected 10% of visits. Agreement of gaze coding was measured by the proportion of time in agreement over total visit time and ranged from 0.58 to 0.96.

The MORAE usability software package (techsmith.com) was used to automatically capture discrete mouse click events. These measurements may be directly summarized into indicators of gross EHR activity and were used to create a measure of number of mouse clicks per minute during a visit.

Conversational dead space was measured as the proportion of time in the consultation when neither PCP nor patient was talking (state 0,0). Coders captured the data using a computer-based coding program as follows: a coder listens to the interaction, focuses on one of the participants (e.g., the PCP), and presses the mouse down when that participant talks, releases it when he/she stops speaking, presses it down when he/she speaks again, and so on for the duration of the consultation. The program's output includes a serial string of 1s (talking) and 0s (silence) every tenth of a second. The coder then goes back to the start of the consultation and does the same coding process with the other participant (e.g., the patient). Once this is complete, the two binary strings are merged onto the same timeline. The conversational dead space measure would be the proportion of 0,0 combinations for that consultation. Reliability for the measure in this study was established by recoding 11 consultations. Inter-coder reliability (assessed with intraclass correlation) was .68, indicating satisfactory consistency between the two coders.

2.3. Communication measures

2.4. Data analysis

We assessed three outcome variables: PCC ratings, patient involvement ratings, and ratio of PCP over patient (including companion) talk during the visit, and their associations with patient and PCP characteristics (except patient gender due to small number of females), including PCP's EHR use (total number of mouse clicks, percentage of gaze time at EHR) and percentage of silence time during the visit. To account for PCP's cluster effect, a linear mixed effects model was used. The univariate analysis was performed to study the association between each variable with outcome. Variables found to be significant at p < 0.15 were included in the multivariable analysis, and variables with p < 0.10 were kept in the final models. Because PCC and patient involvement often have reciprocal influence on each other, each was also added to the final model to assess its effect on the significance of other variables in the model. Finally, a simple random effects model was used to study the association between the patient satisfaction measure and rating of PCP patient-centered communication, rating of patient involvement, and PCP control of the conversation. Normal assumption of residuals in linear mixed effects model was examined using normal probability plot. All analyses were performed using statistical software R [19].

3.1. Overview

The final sample consisted of 125 patient visits. Table 1 presents the characteristics for the study sample. The study subjects were primarily men (96.8%, typical of VA clinics), mostly white (64.8%), just over 60 in average age (SD = 13.4, median = 62), and fairly well educated (81.6% had at least some college education or degree). The distribution of the communication measures, patient satisfaction scores, and PCPs' EHR usage are given in Table 2. Patient satisfaction was positively associated with observer ratings of PCPs' PCC (p = 0.0005) but was not related to ratings of patient involvement or the PCP conversational control measure.

3.2. Univariate analysis

Higher percentage of time looking at the EHR (b = −12.2, p = 0.009) and a higher percentage of silent time (b = −21.3, p = 0.0001) were significantly associated with lower ratings of PCPs' PCC. Longer visit time (b = 0.002, p = 0.052) also was significantly associated with more PCC. As expected, patient involvement (b = 0.38, p < 0.0001) was positively correlated with PCC.

Longer visit time (b = 0.002, p = 0.002), less percentage of silent time (b = −17.2, p < 0.0001), and higher patient education (college vs. high school or lower, b = 3.28, p = 0.048; more than college vs. high school or lower, b = 4.18, p = 0.012) were significantly related to more patient involvement.

We also found that a higher ratio of PCP over patient talk time was significantly associated with higher percentage of silent time in the consultation (b = 2.51, p = 0.009) along with a trend for less patient education (more than college vs. high school or lower, b = −0.66, p = 0.072) to be associated with more PCP control of talk time.

3.3. Multivariable analysis

Multivariable analysis results are given in Table 3. More PCP time spent gazing at the computer (p < 0.0001), less visit time (p = 0.0003), patients with less than $20,000 income (compared to > $60,000, p = 0.006), and higher percentage of silent time (p < 0.0001) remained predictive of lower ratings of PCPs' PCC. We also added the patient involvement rating scores to the fitted model and found a trend for patient involvement to correlate with PCC ratings (b = 0.19, p = 0.079), while results for other variables stayed similar.

Less visit time (p = 0.002) and a higher percentage of silent time (p < 0.0001) were associated with less patient involvement in the multivariable model. Regarding PCP control of conversational floor time, we found that only percentage of silence time was significantly predictive (p = 0.01) of higher ratio of PCP talk over patient talk.

4.1. Discussion

A major inefficiency of many current generation EHRs is the use of WIMP (Windows, Icons, Menus, Pointers) graphical user interfaces (GUI) and the use of menu- and form-based GUIs for data entry. Such systems require clinicians to navigate deeply nested menus and browse through long pull-down lists that are neither filtered nor contextualized. In addition, simple tasks are broken down into individual components requiring multiple points, clicks, and scrolls. Consequently, the usability of current EHR system can have real impact on clinicians' ability to communicate with patients while engaging with technology.

This study was one of the first of its kind to examine whether the ways in which PCPs' nonverbal engagement with the EHR (e.g., gaze, mouse click/keyboard activity, silence) affected the quality of their communication with patients. PCPs were rated by observers as displaying less patient-centered communication when they spent proportionally more time gazing at the computer and when there was more conversational dead space (silence) in the visit. More conversational dead space was also predictive of less patient involvement and more PCP control of the conversational talk time. The findings have important implications for future research and clinical training, particularly with respect to how the EHR can be managed in ways that do not interfere with patient-centered communication.

First, eye contact is an important indicator of a PCP's attentiveness to a patient and involvement in the interaction [20]. When a PCP spends more time looking at the computer, it may disrupt the flow of the conversation, and the PCP may be perceived as distracted, disengaged, and less patient-focused in the consultation [8]. While this may not be directly related to lower patient satisfaction per se, [21] it could affect how involved patients are in discussing their own concerns and needs [22].

However, the potential negative reaction to gaze directed at the computer and away from the patient is likely mitigated by a number of factors. For example, nonverbal engagement with the computer that takes away from a focus on the patient may be particularly important for patients who are new to the PCP. With an established patient-provider relationship or a history of previous visits (which was part of the inclusion criteria in this study), a patient might have previous experiences to draw upon in assessing the quality of a PCP's communication skills. In the absence of previous experiences (i.e., much like the observers used in this study), how that PCP acts and talks during that particular consultation may have a strong impact on how that PCP's communication is perceived. This may be the reason proportion of gaze at computer and conversational dead space were not predictive of patients' satisfaction (data not reported), but was predictive of observer ratings.

In addition, because eye contact during face-to-face conversation is a commonly accepted social norm (at least in the US and Europe), many PCPs may naturally compensate for the lack of gaze and avoid extended periods of silence through other conversational strategies, such as making small talk or providing the patient with accounts and explanations of what they are doing while they are entering or extracting data from the EHR.

The study had several limitations. First, both the patient (n = 125) and PCP (n = 23) samples were relatively small. Also, the patients were mostly male, white, and relatively well educated and the PCPs were disproportionally female (74%). Thus, the sample sizes and demographic distribution of patients and PCPs precluded additional exploration of subgroup differences in the ways clinicians use the EHR and patients react to its use. Second, PCPs in the study used the same EHR, and there may be differences in the ways PCPs interact with other types of EHRs. Third, we examined observers' ratings of communication quality and not the actual patients' perceptions of communication. However, consistent with other studies using analog patients, [23,24] observer ratings of PCC were correlated with a general patient satisfaction measure and, compared to patients' ratings, typically provide responses less skewed toward the positive extreme. Future research should additionally examine the verbal content of these encounters in order to analyze various communication strategies PCPs used while working in the EHR. Finally, when PCP and patient were looking at the computer screen together, we coded this as gaze away from the patient when in fact one could argue patient and PCP were mutually engaged with the same visual object. However, instances of shared screen were quite rare and did not affect the results of the analysis.

4.2. Conclusions

The way PCPs interact with EHRs in exam-room settings may interfere with their efforts to achieve patient-centered communication if the EHR use takes too much visual attention away from the patient and leads to conversational dead space. Because EHR use is common in outpatient consultations, more research is needed to determine how providers can more effectively engage patients verbally while simultaneously using the EHR.

4.3. Practice implications

These findings have important implications for clinician education on how to manage consultations with patients. For example, many patients believe PCP access to the EHR is beneficial and contributes to higher quality of care. However, the PCP must be able to maximize the advantages of having the EHR in the exam room (e.g., accessing information, efficient data entry), while at the same making sure it does not interfere with PCPs' ability to be attentive, informative, and maintain the flow of the conversation. Moreover, the computer could potentially be a facilitator of provider–patient conversation if, for example, information on the screen was shared visually with the patient for purposes of patient education. This suggests that an educational curriculum should include not only the traditional focus on communication skills, but also EHR management skills (e.g., shortcuts, keyboarding skills, sharing computer screen with patient) that will not impede and could even facilitate the interaction with the patient. Finally, next-generation EHR design must address better usability and clinical workflow integration in ways that can facilitate patient–clinician communication.