Secure Messaging for Diabetes Management: Content Analysis

Introduction

Approximately 29 million Americans have been diagnosed with diabetes [1]. The prevalence of diabetes among veterans is even higher than that in the general population, affecting nearly 25% of Veterans Health Administration patients [2]. Diabetes can lead to other serious problems, including cardiovascular disease, stroke, and loss of limbs [3]. Diabetes is estimated to cost the US health care system US $245 billion annually [2]. Effective management of diabetes is centered around glycemic control, requiring a multifaceted, continuous, and proactive approach that includes patient monitoring and education, lifestyle management, and pharmacologic therapy [4].

Online patient portals such as the Department of Veterans Affairs (VA) My HealtheVet, expand patients’ access to health care by facilitating communication with health care teams, such as through secure messaging. Secure messaging offers an electronic exchange of messages between patients and clinical team members. Through secure messaging, patients can inform clinical team members of their health status and progress and receive self-management or remote support. Secure messaging can increase patient engagement and self-management [5], and can further improve patient outcomes [6,7]. Previous work has found improved glycemic control in patients who used secure messaging for 2 or more years compared to nonusers of secure messaging [8]. Moreover, there appears to be a positive association between the frequency and intensity of secure messaging use and glycemic control [9]. However, the exact message content being exchanged between patients and clinical teams, and how that content supports improved diabetes self-management, is less clear.

Previous work has examined the content of patient-team secure messages. In a general patient population, one secure messaging content analysis across two VA primary care clinics documented that medication renewals and refills, scheduling requests, medication issues, and health issues were the most common patient-initiated requests via secure messaging [10]. To our knowledge, only one study to date has examined the content of secure messaging in patients with diabetes and the association with diabetes outcomes [11]. Specifically, Heisey-Grove et al [11] examined a sample of patient-initiated secure messages that were saved to the patients’ charts in one medical center. This analysis found greater improvements in glycemic levels (ie, decreased hemoglobin A_1c [HbA_1c]) when patients engaged in secure messaging where they sought information (ie, information-seeking messages) or received messages about procedures or treatments from clinical team members. Conversely, glycemic levels worsened (ie, HbA_1c increased) among patients who engaged in secure messages where they shared information with clinical team members [11]. The work of Heisey-Grove and colleagues thus offers foundational insights into how patients and their clinical teams leverage secure messaging for diabetes management.

The current study expands on this earlier work by examining a representative selection of patient- and team-initiated secure messages from a national sample of patients with type 2 diabetes. We sought to understand how patients and clinical teams use secure messaging to communicate with one another about diabetes management. Leveraging a mixed methods analytic approach, we adapted and applied a theoretically based taxonomy to analyze the content of secure messages between patients and clinical teams (described in further detail below). We examined differences in secure message content, the role of the secure message initiator (clinical team member role or patient), and associations with glycemic control.

Methods

Recruitment

We used data collected for a larger study performed between 2017 and 2020 examining patient portal use in diabetes management [5,12,13]. Participants included veterans with type 2 diabetes who had uncontrolled blood glucose (2012 mean HbA_1c>8.0% and less than 25% of the year with HbA_1c<8.0%), as well as repeated and current use of key portal features. Repeated use was defined as having at least two instances of each of the following in 2 out of 3 years between 2013 and 2015: requesting a prescription refill, viewing or downloading health information, and using secure messages. Current use included having sent at least four secure messages between January 2016 and June 2017. Surveys about My HealtheVet and diabetes management were mailed to a total of 1200 veterans and 448 were returned. Further details regarding the sampling methodology are available elsewhere [5].

Ethics Considerations

All survey respondents consented for their recent secure messages to be accessed and analyzed. The Institutional Review Board at VA Bedford Healthcare System approved this research (review number 0008).

Secure Message Thread Coding

Survey respondents’ five most recent secure messaging threads were pulled in 2018. Every message and/or thread was coded based on previously published message coding methods [10], which were inspired by elements of the Taxonomy of Requests by Patients [14]. Binary indicators were created to indicate which threads were initiated by the patient and which were initiated by the clinical team. At the message level, we used binary codes to indicate whether the message was related to a diabetes theme, including blood pressure, cholesterol, physical activity, diet/nutrition, mental health, and specific diabetes-related content. Messages were coded as including diabetes-related content if the message explicitly mentioned diabetes, blood glucose, “sugars,” insulin, endocrine, or other diabetes medications. Messages that were potentially related to diabetes but not explicitly tied to the condition by either the patient or provider were not coded as a diabetes-related message (eg, blood pressure, diet, exercise). Within each message that was specifically coded as diabetes-related, additional subcodes were applied to further describe whether the diabetes content related to one of the following topics: scheduling, referrals, or consults; medication renewals, refills, or other medication-related issues; test results; test issues; health issues; self-reporting; informational; My HealtheVet; life issues; complaints; establishing a personal connection; and care coordination.

Each message could be coded for more than one topic (eg, refills and scheduling). The research team met frequently to discuss coding questions, discrepancies, and codebook updates. A codebook was developed prior to reviewing messaging data and was further refined at the beginning of the coding stages. To ensure consistency, two of three research assistants independently coded each message. Interrater reliability (Cohen κ) was calculated iteratively (range 0.57-1.00 across the coding period), and joint coding continued throughout. Coding at the thread level was informed by the codes applied to the content at the message level (ie, if a thread contained one or more messages related to the specified topic, the thread was coded as discussing that topic). Coding at the individual level was informed by the codes applied to the thread level (ie, if a patient or clinical team member engaged in one or more threads related to the specified topic, the message initiator was coded as having discussed that topic over secure messaging).

Glycemic Control

To determine the percent time in glycemic control in 2018, we calculated the percentage of time in the baseline year a patient was estimated to have sustained control in HbA_1c (<8.0%). Time spent in control was calculated based on the Rosendaal method [15,16], using linear interpolation to assign values to each day between successive measurements. After interpolation, we calculated the percentage of time during the year that the interpolated values fell inside the region of control (eg, HbA_1c<8.0%). Participants whose HbA_1c was below 8.0% for more than 75% of 2018 were considered to be in “good” glycemic control. Conversely, participants who spent less than or equal to 75% of the year with an HbA_1c level under 8.0% were considered to be in “poor” glycemic control. The χ² test was used to examine differences in frequency of content codes between patients in good and poor glycemic control.

Results

Patient and Message Characteristics

The patients’ mean age was 67.4 (SD 7.5) years and the mean HbA_1c was 8.1% (SD 1.2). The sample was majority male (94%), white (84%), and married (50%). We examined secure messages across 2240 threads (5442 total messages) from 448 patients who responded to our survey. All 5442 messages were coded. Each thread contained a mean of 2.4 secure messages (SD 1.7), ranging from 1 to 24 messages per thread. Among the 2240 coded threads, 87.86% (n=1968) were initiated by the patient (n=1890) or the patient’s caregiver/proxy (n=78). The remaining 12.14% (n=272) of threads were initiated by the clinical team. Among the clinical team–initiated threads, registered nurses initiated the most threads (109/272, 40.1%), followed by physicians (32/272, 11.8%), licensed practical nurses (30/272, 11.0%), other VA staff (28/272, 10.3%), pharmacists (15/272, 5.5%), medical support assistants/health technicians (4/272, 1.5%), other providers (eg, psychologists; 4/272, 1.5%), and nurse practitioners (3/272, 1.1%). The clinical team member was not clearly identifiable in 17.3% (n=47) of the team-initiated threads.

Message Content

All patients (N=448) initiated at least one thread and 36.4% (n=163) received at least one team-initiated thread (Table 1). Most patients (n=323, 72.1%) engaged in at least one thread pertaining to diabetes-related content, 31.7% (n=142) of the sample engaged in at least one thread about blood pressure, 15.2% (n=68) engaged in at least one thread about cholesterol, 13.8% (n=62) engaged in at least one thread about mental health, 9.4% (n=42) about diet and nutrition, and 4.0% (n=18) about physical activity. Among patients who engaged in at least one thread related to diabetes content (n=323), 90.7% (n=293) engaged in at least one thread related to medication renewals or refills; 81.4% (n=263) discussed scheduling, referrals, or other administrative content; and 63.5% (n=205) discussed medication or equipment issues. Table 2 presents sample secure messages representing each content type as found within the first message in a patient-initiated thread.

Content contained in team-initiated threads was similar to that found in patient-initiated threads. Of the 163 patients that received a team-initiated thread, 74% (n=121) received at least one thread with content related to diabetes. This was followed by blood pressure (n=48, 29%), diet and nutrition (n=19, 12%), physical activity (n=12, 7%), mental health (n=24, 15%), and cholesterol (n=24, 15%). Among the 121 patients who received a team-initiated thread related to diabetes, 88% (n=107) received a thread about medication renewals of refills; 85% (n= 103) about scheduling, referrals, or consults; and 73% (n=88) about medication issues (Table 1). Table 3 presents sample secure messages of each content type as found within the first message in a team-initiated thread.

Association Between Content and Glycemic Control

From the initial 448 participants, 431 had at least one HbA_1c value in 2018. Of the 431, 44.6% (n=192) met our definition of good glycemic control and 55.5% (n=239) met our definition of poor glycemic control in 2018. A significantly larger proportion of patients with good glycemic control (n=72, 37.5%) engaged in at least one thread about blood pressure, compared to those with poor glycemic control (n=65, 27.2%; P=.02). Noting this significant difference, we further examined the content among blood pressure–related messages. Similar to the diabetes-related messages, we found that, among patients who engaged in at least one blood pressure–related message (n=142), most patients engaged in messaging related to medication renewals or refills (n=138, 97.2%); followed by scheduling, referrals, or other administrative content (n=108, 76.1%); and medication issues (n=86, 60.6%).

Among those who engaged in at least one thread related to diabetes content (n=312), 43.3% (n=135) were in good glycemic control and 56.7% (n=177) were in poor glycemic control. Significantly more patients in good glycemic control (32/135, 23.7%) engaged in at least one thread related to informing their clinical team about facts relevant to their health care compared to those in poor glycemic control (22/177, 12.4%; P=.009). There were no other significant differences related to the proportion of patients in good and poor glycemic control among other health topics of diabetes-related content codes (Table 4).

Among patients in good glycemic control (n=192), 39.9% (n=63) received at least one team-initiated thread. Among patients in poor glycemic control (n=239), 39.8% (n=95) received at least one team-initiated thread. However, this difference was not statistically significant (P=.14).

Discussion

Principal Findings

We used mixed methods to examine the content in a national sample of patient-team secure messages, and the relationship between content and glycemic control. Approximately one-third of the veterans in our sample received at least one secure message thread that was initiated by a member on their clinical team. The initiator on the clinical team was most frequently a registered nurse. We found that more than half of our sample used secure messaging to discuss diabetes-related content. Among those who used secure messaging to discuss diabetes-related content, more than half discussed medication renewals or refills, scheduling, and medication or equipment issues. The proportion of patients who discussed these topics was similar among patients who received at least one team-initiated secure message. This work is aligned with earlier content analyses [10]. Previous work, which was not specific to patients with diabetes, found that the most common messages are transactional (eg, requests for refills, appointment scheduling, administrative requests), followed by informational (eg, patients providing health measurements or updates on care) or interactional (eg, request for input on medical symptoms or medication issues) [10]. The content of the secure messages was similar in both patient- and team-initiated messages, although patients more frequently initiated a thread to seek a medication renewal, whereas team members most frequently initiated a thread to inquire about medication and medical equipment issues.

There were few differences in secure messaging content between participants in good and poor glycemic control. One exception was that compared with participants in poor glycemic control, significantly more participants in good glycemic control engaged in secure messaging related to blood pressure. There is a well-established risk inherent in the combination of poor glycemic control and uncontrolled hypertension in patients with type 2 diabetes [17]. As such, blood pressure control is also clinically recommended as an important target for diabetes management [18]. Post hoc analyses revealed that patients who engaged in secure messaging about blood pressure tended to discuss similar topics to those who engaged in secure messaging about diabetes-related content: medication renewals or refills, scheduling, and medication issues. It is possible that patients who were able to obtain good glycemic control are, in general, more active in their disease self-management, and thus leveraging secure messaging to manage other important aspects of their health (ie, blood pressure).

Compared with those in poor glycemic control, significantly more patients in good glycemic control used secure messaging to provide their clinical team with information simply for the sake of keeping them informed about their health status or health care received elsewhere. This contrasts with earlier work, in which Heisey-Grove et al [11], counter to their hypothesis, reported that patients sharing information with clinical team members experienced an increase in HbA_1c. A plausible explanation as to why our findings differ may be that our definition of information sharing was more constrained. In their analyses, they combined three subtypes of information sharing—clinical updates, self-reporting, and responses to clinician messages—and were unable to detect a statistically significant association for any of the three subtypes with glycemic control. Additionally, these represent both proactive and reactive messages, whereas we focused on proactive messages from patients to inform their teams, which they called “clinical updates.” Patients who engage in this type of proactive communication may be more activated patients, in the sense that they are engaged in managing their own condition and recognize the importance of care coordination. Heisey-Grove et al [11] did find that patients who initiated secure messaging threads experienced HbA_1c improvements compared to those who did not. Previous work has found that use of online patient portals and accompanying features such as secure messaging can increase measures related to patient activation [19] and subsequent self-management [5]. This is echoed in the eHealth Enhanced Chronic Care Model [20], which outlines the role that eHealth plays in supporting productive interactions between activated, informed patients and prepared, proactive clinical teams to support chronic disease outcomes. Patient and staff training is crucial to support portal use [21] and may in turn support patient activation and engagement in their care.

The current analysis expands on Heisey-Grove et al’s [11,22] earlier work by examining the association between information sharing and glycemic control with a more complete selection of secure messages (ie, all of a patient’s most recent secure messaging threads, not just those saved to the patient’s chart). We analyzed all secure messages that the patient and team engaged in from the patients’ most recent five secure messaging threads, whereas earlier work only examined secure messages that were selectively saved to the patients’ charts [11]. Secure messages that the clinical team chose to save to patients’ charts may be perceived to have greater clinical relevance, and therefore frequencies of message content and the associations between the content and clinical outcomes may also be biased. For example, at a site that saves only select messages, a team member may be more likely to save a secure message if the patient reported concerning HbA_1c levels and the team responded with recommendations for action. They may be less likely to do so if self-reported HbA_1c levels were not a cause for concern and no action or change is recommended. Thus, our study adds to the literature by presenting a less biased picture of how veterans living with diabetes and their clinical teams use secure messaging to support diabetes management.

The current analysis also expands on earlier work by examining both patient- and team-initiated secure messaging threads. Compared to a prior study examining the prevalence of team-initiated secure messaging threads in VA [10], we observed more team-initiated threads (10.8% of the threads) in this analysis of 2018 data than we did in 2013 (5.5%). This increase is promising given that clinical team–initiated secure messaging can significantly and positively influence diabetes self-management [5]. By including team-initiated messages, we enhance our understanding of patient-team communication for diabetes via secure messaging. Finally, another novel aspect of the current analysis is that we were able to leverage the United States’ largest integrated health system, the Veterans Health Administration, and examine secure messages from across the nation. Examining the content of secure messages within one medical center may not capture the heterogeneous needs of patients with diabetes. For example, a rural patient with difficulty accessing in-person care may be more reliant on using secure messaging than a patient who lives closer to the medical center [23].

Limitations

The strengths of the current analysis include the number of secure messages coded, rigorous coding methods, and link to clinical outcomes in a national sample of US veterans with diabetes. There are also several limitations. For one, the sample is predominantly white male US veterans and thus we are limited in the generalizability of these findings. Additionally, we focus on a set of specified topics related to diabetes, although we did not include a specific code about self-management that may have likely led to a change in clinical team action (eg, medication titration) or patient action (eg, decrease a medication dose). Future work should analyze other content areas beyond topic, such as shared decision-making, to perhaps further understand how other content may influence diabetes management. Another avenue for future work would be to explore the impact of training on patient and clinical team use of secure messaging. For example, teaching both patients and clinical teams how to use secure messaging to support disease management may support the adoption and effectiveness of secure messaging for disease management [24].

Another important consideration of this work is that we analyzed the content of secure messaging among patients who were sustained users of secure messaging. As such, our findings may not be generalizable to patients who use secure messaging less frequently or not as recently. Another limitation is that this research examines cross-sectional associations between secure messaging and glycemic control and cannot confirm a causal pathway. For example, patients who share information via secure messaging may be more proactive in their disease self-management, which may lead to improvements in glycemic control. Conversely, patients who share information may be struggling more with controlling their HbA_1c, which may be associated with worse glycemic control. Future work that examines how secure messaging adoption, and subsequent content of the messages, influences glycemic control will further our understanding of the directionality between content and clinical outcomes.

Conclusions

This is one of the first studies to perform a content analysis of secure messaging specific to diabetes, and to explore associations between message content and glycemic control. Focusing on a chronic condition where patients’ self-management behaviors drive outcomes allowed for a nuanced exploration of the relationship between secure messaging and health outcomes. In addition to reducing bias in how messages were sampled (ie, coding all patient-team messages as opposed to a sample saved to the patient’s chart), this work adds a complementary perspective to earlier content analyses [11] with supplementary qualitative examples of secure messages. We sought to understand whether secure message content was associated with good or poor glycemic control. It appears that, overall, the specific topic of the secure messaging may not be as clinically important for diabetes management. Rather, the act of engaging (compared to not engaging) in secure messaging may be most influential. Secure messaging makes it easier for patients to communicate with their clinical teams, and this may be the main driver of better clinical outcomes. Clinical teams should be encouraged to communicate, both responsively and proactively, with their patients through secure messaging to support their diabetes management.