Effect of Electronic Messaging on Glucose Control and Hospital Admissions Among Patients with Diabetes

Background

Electronic patient portals have been increasingly utilized in response to the American Recovery and Reinvestment Act of 2009, which provides incentives for meaningful use of electronic health records (EHR). These systems allow patients to have access to their test results, refill prescriptions, set up appointments, access educational materials, and enable secure, usually asynchronous, electronic messaging (EM) with their healthcare provider.

EM has been investigated in the management of diabetes, where it has been assumed that EM could reduce the need for frequent office visits. However, studies show mixed results, with some reporting increased outpatient visits, ^{1 –3} and others showing no effect. ⁴ Some studies reported that EM is associated with lower HbA1c levels. ^{2,5 –8} Thus far, studies regarding EM have not differentiated between type 1 and type 2 diabetes, examined message content, or examined the impact of EM on hospital or emergency department (ED) visits.

This retrospective cohort study examines the relationship between EM and HbA1c levels, emergency room visits, and hospital admissions. Patients were stratified by type of diabetes and by EM use (active EM use [active users], activated account but not using [active nonusers], and not active at all [inactive users]).

Methods

Results

A total of 867 patients were considered active users, 1207 were active nonusers, and 1542 were inactive users. Inactive patients were older, more likely to be African American race, less likely to have private insurance, and had higher median area-level income (Table 1). Inactive users were also less likely to require insulin and more likely to have type 2 diabetes than type 1 diabetes (Table 1).

There were a total of 2748 message threads and 5264 distinct messages from all threads from the active users, with 19% of active users sending 1 message, 22% sending 2 messages, and 59% sending 3 messages or more. Of these total messages, 1600 were diabetes related and 1214 were medication related. Eighty percent of messages were two-way (sent by either the provider or the patient and accompanied by a response).

Discussion

This study demonstrates that EM is strongly associated with lower HbA1c, even after controlling for demographic, economic, and diabetes-related variables. However, the effect on HbA1c was modest and the intensity of use, determined with the number of EM messages among EM users, was not associated with HbA1c. The reasons for this differential outcome are unclear but may include (1) smaller sample size when data are restricted only to active users, particularly when only diabetes-specific EMs are included; (2) lack of specific data regarding message content; or (3) representation of residual bias or confounding. In addition, EM use was not associated with hospitalizations or ED visits after adjustment for relevant variables.

In this study, we found that EM use varied considerably by baseline characteristics, including age, race, and economic status. This likely represents differential access to electronic means of communication, which would expectedly be higher among younger (technology savvy) individuals or those with greater economic means. In fact, in a recent sample of 54 adults, a majority of participants had difficulty completing the tasks needed to utilize the EHR. ¹⁰ These baseline differences highlight concerns that rather than improving access, EM technology could widen disparities. These disparities have been previously documented. ^11,12 Nevertheless, it is a strength of the study that we were able to compare patients who have activated their account with those who have activated their account but not sent a message.

With time, access to electronic patient portals has increased, as mobile technology and associated health applications have become more pervasive. This is further fueled by meaningful use requirements and systematic efforts by medical centers to promote the use of patient portals. In fact, studies show that provider attitudes greatly influence personal health record use. ^13,14 Therefore, it is possible that provider enthusiasm may have affected EM use and utility in our study. To some extent, it is also possible that patient preference for a particular provider style may, in turn, influence adoption of EM. Other patient-reported reasons for lack of enrollment in an electronic patient portal include lack of information or motivation as well as negative attitudes toward technology, reasons that are reported more frequently than technological barriers. ¹⁵ Therefore, barriers to patient portal use could be effectively addressed with better patient education and engagement. Nevertheless, expanded access, not surprisingly, creates additional clinical work that is not reimbursable through traditional fee for service models. Therefore, it is critical that outcomes are established so that other clinical care models can be considered. This might include patient-centered medical/specialty homes, other chronic care models, and advocating for direct reimbursement for EM.

By comparison, a large Kaiser Permanente study of 35,423 patients with diabetes, hypertension, or both demonstrated in both log linear models and matched case–cohort analyses that EM use was associated with improved HbA1c and other Healthcare Effectiveness Data and Information Set (HEDIS) measures. ⁵ Furthermore, two or more EM threads in a 2-month period were associated with better HbA1c reduction compared with less EM use. In this study, 25% of participants had registered for EM but only 7.8% had actually sent a message. Insulin use was not reported. Harris et al. reported in a cross-sectional study of 15,427 health system patients with diabetes that 19% used EM, which was significantly associated with HbA1c <7% in MV analyses. ² The same group later reported that the frequency of EM use was associated with better HbA1c levels, even after adjustment for insulin use and other variables. ⁸ Lau et al. used propensity score matching in a sample of 157 patients with type 2 diabetes and determined that EM users were more likely to achieve an HbA1c of 7% or less at 6 months compared with nonusers. ⁶ However, message content and type of diabetes were not reported in any of these studies.

It is important to note that the availability of EM in isolation does not necessarily imply that it is utilized for the purpose of therapeutic interventions. This is particularly relevant for many of the aforementioned studies, which did not assess message content. In the current study, message content was assessed via categorization of the message subject, and we did not identify a difference in HbA1c by message subtype. For a more robust assessment, randomized controlled trials are necessary. Tang et al. reported a 12 month trial of 415 patients with type 2 diabetes who had baseline HbA1c ≥7.5% to a multifactorial intervention, of which EM was only one component. ⁷ The intervention resulted in better HbA1c reduction compared with the standard care at 6 months but was not sustained at 12 months. Of note, this study did not provide details of therapy, such as insulin use. Thus, it is unclear whether specific subgroups may have benefited. However, it suggests that initial enthusiasm for a technology may wane over time unless there is a strong effort for continued patient engagement. Importantly, patient engagement is likely to be even lower among the general population. To understand the impact of EM more completely, future randomized trials should be targeted toward a defined population of patients undergoing a specified target-driven intervention, such as initiation of basal insulin therapy among patients with type 2 diabetes. The study intervention would ideally also stipulate indications and timelines for EM use, seek to control for disparities in access to technology, and provide ongoing engagement throughout the study.

Several studies indicate that EM use did not reduce, and in some cases actually increased the frequency of office visits. ^{1 –4} The current analysis extends these observations by identifying no difference in hospitalizations or ED visits among EM users and nonusers.

This study has several limitations, mainly due to its retrospective nature. As a result, the observed effects, or lack thereof, after EM implementation cannot be assumed to be causal. Despite adjustment for multiple sociodemographic and illness-related factors, there could be other unmeasured factors involved in selection of patients who do or do not utilize EM. These unmeasured factors may cause residual confounding, which is not taken into account. Specifically, we were unable to account for hospitalizations or emergency visits at other hospitals. Information about duration of diabetes was not available. Moreover, changes in covariates, such as BMI or insurance status, were not tracked, and it is possible that this could affect results. In addition, even though EM use among patients with active portals was compared, there could still be residual selection bias. Message content was categorized by subject line, though actual content may have varied, particularly within a single thread, and it is not known whether a given message was initiated by the patient or the provider. It is for this reason that messages were primarily analyzed by message rather than by thread and why message count or content was not the primary exposure variable of interest. Although some message content may not be relevant for the outcomes studied here, it may be important for other outcomes, such as patient satisfaction. In addition, area-level data for income were utilized, which may be limited for predicting individual income, but may be more informative for determining the effect of living in an economically disadvantaged area. ¹⁶ Therefore, these findings warrant further investigation, ideally through randomized controlled trials.

In conclusion, EM use, but not EM frequency, is associated with lower HbA1c, but hospitalizations and ED visits are unaffected by EM use. However, significant disparities exist between patient portal users and nonusers, and this may profoundly impact the perception of outcomes attributed to EM. These findings warrant further attention in prospective studies in which the content of messaging is targeted and algorithm driven.