Abstract
Objectives
Organizational strategies for implementing eHealth tools influence patient and provider use of portal technology. This study examines whether the intensity of bidirectional secure portal messaging is associated with improved clinical outcomes.
Material and Methods
This is a retrospective cohort analysis of 101 019 patients with diabetes or hypertension (11 138 active portal users) who received primary care within the Ochsner Health System between 2012 and 2014. Propensity score–adjusted multivariable fixed effects regression panel analysis was used to examine associations between intensity of “medical advice” portal messaging and glucose/blood pressure control.
Results
Most portal users rarely used medical advice messaging. A higher proportion of patients who were age 50 years and older, female, white non-Hispanic, and with co-morbid diabetes and hypertension had higher frequency and intensity of medical advice messaging. Study findings revealed a dose-response effect of the intensity of messaging on glucose control, whereby, compared to nonportal users, each level of messaging among portal users was associated with greater decreases in HbA1c (β estimate [95% CI]: none −0.28 (−0.34 to −0.22); low −0.28 (−0.32 to −0.24); medium −0.41 (−0.52 to −0.31); high −0.43 (−0.60 to −0.27), all P ≤ .001). There was no observed effect on blood pressure.
Conclusions
The digital divide exists not only between portal users and nonusers but also among portal users. Research exploring the relationship between intensity of bidirectional secure messaging and health outcomes for a broader scope of chronic conditions is needed. Future implementation research must also elucidate best practices that enhance not only the use of portals by patients and providers, but how they use portals.
Keywords: electronic medical record, patient portal, hypertension, diabetes
BACKGROUND AND SIGNIFICANCE
Over the last decade, adults in the United States have increasingly used the Internet for a variety of activities, including seeking online health information and interacting on social media.1–3 Incentivized by Meaningful Use initiatives under the Affordable Care Act, many health systems have integrated health information technology such as patient portals into care delivery.4,5 Recent studies suggest that patients are increasingly interested in asynchronous communication with health care providers to pose questions about health concerns, medications, test results, and appointments.6–8 Electronic communication such as secure portal messaging has also been shown to effectively support self-care management, improve select clinical outcomes, and increase patient satisfaction.6,9 Patient portal messaging has the promise of not only improving health outcomes, but also enhancing patient-provider communication.
Organizational strategies for implementing patient portals within a health care system undoubtedly impact providers’ and patients’ use of portal technology.10 The key to acceptance by providers is to make it easy for them to use without increasing their workload. Patient acceptance is heavily influenced by provider encouragement to use technology. Wolcott et al.,11 for example, recently demonstrated in a large military study that the level of usage by health care providers of portal messaging and their responsiveness to messages likely influence patient uptake and initiation of secure messaging. Presumably, responsiveness shapes patients’ perception of provider approachability and receptiveness to electronic communication.
The authors previously described a multicomponent assessment of Ochsner Health System’s portal technology implementation within the context of primary care reengineering efforts to improve accessibility to care.12,13 Workflow redesign to enhance primary care provider responsiveness to portal messaging was a key implementation strategy.13 The net result of this approach was that, among patients with hypertension or diabetes, portal users had increased primary care visits and telephone encounters as well as relative improvements in blood pressure and glucose control after initiating portal activity compared to portal nonusers. Since chronic disease management was not the initial focus of portal implementation, we postulated that patients who were already actively interacting with the health system were probably more likely to adopt portal use as another means of accessing services. However, in a subsequent survey of a sample of the study population, patients reported that the quality of interactions with providers or the care team via the portal (eg, responsiveness, consistency or lack thereof) influenced their use of the technology.12
OBJECTIVE
We conducted this follow-up study to assess whether the frequency and/or intensity of patient–care team portal messaging about medical concerns (“medical advice messaging”) was associated with the previously observed differences in clinical outcomes in our health system. Our hypothesis was that patients with high rates of bidirectional portal messaging would have better control of their hypertension or diabetes compared to (1) portal users who do not use secure messaging and (2) portal nonusers. We examined whether there was a dose-response effect of medical advice messaging among portal users compared to portal nonusers on blood pressure and glucose control.
MATERIALS AND METHODS
Study setting, population, and design
This retrospective observational cohort study was conducted at Ochsner Health System, southeast Louisiana’s largest nonprofit, academic, multispecialty health care delivery system. Ochsner owns, manages, or is affiliated with 29 hospitals and 60 health centers. In 2012, Ochsner implemented Epic Systems and later launched its electronic medical record (EMR)–tethered MyOchsner patient portal. This study was part of a multicomponent assessment of portal technology implementation within the context of adult primary care practice transformation, which focused on increasing patient accessibility to online appointment scheduling, medical records, and portal messaging.13
Primary care providers led the health system in promoting the use of MyOchsner by implementing lean management strategies to improve care teams’ response times to portal messages. All portal messages are automatically routed to nursing staff (mostly licensed practical nurses) assigned to specific providers to triage the messages. This workflow was purposely designed to control the volume of provider in-basket messages, recognizing that some patient messages may not require direct responses from providers. If direct provider response was deemed necessary or preferred, providers could e-mail patients directly after messages were forwarded to them.
The study population was limited to adult patients (age ≥18 years) with at least 2 primary care visits between July 2012 and December 2014 who had a diagnosis of hypertension and/or diabetes. The study was limited to this subpopulation because of primary care providers’ early adoption of patient portal workflows during this time period and the health system’s subsequent launching of a number of chronic disease management programs for patients with hypertension and/or diabetes. Among the 101 019 potentially eligible patients, only 11 138 were MyOchsner portal users. Portal use was defined as having portal activity documented during the study period. This study was approved by the Ochsner Health System Institutional Review Board.
Study variables
Dependent variables: disease control
The authors extracted data from the EMR for patients age ≥18 years who had a diagnosis of hypertension (International Classification of Diseases Ninth Revision code range 401.xx–405.xx) or diabetes (codes 250.xx, 648.0x, 775.1x). The research team collected HbA1c laboratory values and blood pressure readings documented longitudinally in the EMR 1 year prior to and after the index date. The index date for MyOchsner users was defined as the date when they first logged in to the patient portal. For nonusers, the index date was defined as the midpoint between their first and last contact with a primary care provider during the study period. For each patient, we extracted a longitudinal dataset of repeated measures of HbA1c and/or blood pressure (eg, discrete values with unique date and time of collection). HbA1c results were derived from lab-based assays conducted either within Ochsner’s laboratory system or at external laboratories, depending on the patient’s health insurance plan. Blood pressure readings are routinely measured and entered by clinic nursing staff (or medical assistants) using automatic machines. Per primary care protocol, abnormal blood pressure readings are remeasured manually by clinical staff.
Explanatory variable: medical advice messaging
When patients log in to MyOchsner, they have 3 options for sending portal messages: (1) “communicate with your provider,” (2) “request a medication renewal,” and (3) “customer service question” (billing, insurance, nonmedical concern). When a patient selects “communicate with your provider,” message subject line dropdown menu options include nonurgent medical, prescription, test results, visit follow-up, referral request, or health maintenance update. Portal messaging activity sent via “communicate with your provider” is coded as “medical advice” for the Epic Clarity data warehouse variable entitled “message type.” The research team therefore defined “medical advice messaging” as portal activity for message types coded as “medical advice” in the data warehouse.
We extracted portal data on the types of modules in which users had “read” and/or “write” activity and examined medical advice messaging initiated by patients. Frequency of medical advice messaging (or total count of messages) was defined as the number of unique patient-initiated portal messages. In contrast, intensity of medical advice messaging was defined as the total number of message threads. A message thread is a set of continuous 2-way exchanges between a patient and clinic staff related to an initial message. For example, if a patient sends 100 messages but never receives a reply, the frequency count is 100; however, the intensity of messaging is 0, because no 2-way e-mail exchange ever occurred between the patient and clinic staff. Intensity of messaging was therefore considered a measure of interactive communication. The authors examined histograms of the data to define categories of 6-month interval messaging activity. Frequency count was categorized as low (1–2 messages), medium (3–6 messages), or high (≥7 messages). Intensity was also categorized as low (0–1 threads), medium (2–3 threads), or high (≥4 threads). The 6-month interval for medical advice messaging was selected based on 2 assumptions: (1) patients with diabetes or hypertension should ideally see their primary care provider a minimum of twice a year, and (2) portal messaging activity level can change in the weeks leading up to and following a clinic visit.
Portal module use
We extracted portal data on types of modules in which users had “read” and/or “write” activity and categorized the modules into 9 broad groups: medical records (eg, lab results), communication (eg, advice messages, refill requests), appointments (eg, auto-scheduling), administrative (eg, account inquiry), encounter information (eg, after-visit summary), medical tools (eg, patient-entered flowsheets), editing information (eg, update allergies), information exchange (transmit visit summary), and other.
Covariates of interest: patient demographics and service utilization
Measures of health care utilization included number of (1) primary care provider (PCP) clinic encounters; (2) telephone encounters with primary care, endocrinology, cardiovascular service, nutritionist/dietician, or pharmacy; and (3) specialty clinic encounters with endocrinology or cardiovascular service (including chronic disease care management programs). Data on age, gender, race, insurance type, and diagnosis codes were also collected. Zip code–level household median income data were retrieved from the Uniform Data System Mapper.14
Data analysis
Descriptive statistics were performed for the entire sample. We employed the K means clustering algorithm to perform a cluster analysis of portal users to classify patients into different types of users in terms of what particular module or groups of modules in the system they tend to use more. The 8 clusters generated were (1) infrequent, low-intensity users of all portal modules, (2) electronic messenger–medical record reviewer–appointment maker, (3) electronic messenger–medical record viewer, (4) electronic messenger only, (5) medical record viewer only, (6) appointment maker only, (7) frequent high-intensity users of multiple modules, and (8) outlier patients with the highest frequency and intensity of portal use.
The authors then examined changes in blood pressure (BP) or HbA1c among patients who had at least one measurement documented in the pre-index period and one measurement in the post-index period. Direct regression modeling of medical advice messaging intensity and frequency with the outcomes of interest was conducted. Message intensity had the best model fit and strongest association. Bivariate analysis based on message intensity was then conducted with independent sample t-test comparing means and Wilcoxon rank sum tests comparing medians for continuous variables as appropriate, and Pearson’s chi-square test or Fisher’s exact test comparing proportions for categorical variables as appropriate. Propensity score–adjusted fixed effects multivariable linear regression panel analysis of longitudinal data was then conducted for the study endpoints, with adjustments for covariates of interest that were significant in the bivariate analysis or previously documented in the literature for being clinically and/or statistically meaningful. The propensity score was constructed based on the likelihood of primary exposure of variables being clinically and/or statistically important based on the bivariate results and the existing literature, and balance was verified based on the final number of blocks identified (9 total). The final variables included in the propensity score model were age, sex, race, insurance, income level, service utilization, and disease status (diabetes only vs hypertension only vs diabetes and hypertension). The authors used fixed effects regression panel analysis based on the assumption that there are time-invariant unobservable traits that would make one person respond to intensity of portal medical advice messaging differently than another person.
Stratified regression analysis was conducted based upon chronic condition, and biometric baseline levels of control for each condition (all patients with recorded HbA1c vs baseline HbA1c ≥8% [63.9 mmol/mol] for diabetes; all patients with documented systolic blood pressure [SBP] vs baseline SBP ≥140 mmHg for hypertension; all patients with documented diastolic blood pressure [DBP] vs DBP ≥90 mmHg for hypertension). The authors recognize that clinical guidelines recommend targeting A1c <7% (53.0 mmol/mol) for diabetes control; however, we targeted A1c ≥8% as a cutoff point for defining poor control given the complexity of care for our underserved population in Louisiana. The stratified analysis was conducted to address concerns that changes in BP and HbA1c could represent a tendency for outliers to regress toward the mean rather than the effects of using technology.
The final regression models were reviewed by 2 authors (EPH, DM) to ensure support by substantive clinical and statistical theory and evidence. Correlation matrix and variance inflation factors were utilized to ensure no multicollinearity in the final models. Hosmer-Lemeshow’s goodness-of-fit test was also conducted to determine if the final models fit the data well. Stratified subanalysis with the same regression technique was conducted based on diabetes and hypertension status for biometric endpoints. All regression estimates with 95% confidence intervals (CIs) are reported as fully adjusted results. Statistical significance was set at 2-tailed P < .05. All analyses were conducted using Stata 14.2 (StataCorp, College Station, TX, USA).
RESULTS
Patient characteristics
Among 101 019 primary care patients with hypertension or diabetes seen during the study period, only 11 138 patients were active portal users (Table 1). Compared to portal nonusers, portal users were younger, and a higher proportion were female, lived in zip code regions with higher average household incomes, and were commercially insured. Among portal users there were a lower proportion of black non-Hispanic patients, lower average Charlson comorbidity scores, and a smaller proportion of patients with baseline HbA1c ≥8% (63.9 mmol/mol) or SBP ≥140 mmHg. Portal users also had a lower average number of clinic visits with their PCP or specialists and a higher average rate of telephone calls, hospitalizations, and emergency department (ED) use.
Table 1.
Characteristics of primary care patients by portal use status
| Covariates | Sample | Portal use status |
P-value | |
|---|---|---|---|---|
| No | Yes | |||
| n = 101 018 | n1 = 89 880 | n2 = 11 138 | ||
| Demographics | ||||
| Age, mean (SD) | 63.15 (14.11) | 63.78 (14.12) | 57.79 (12.88) | <.001 |
| Female, n (%) | 56 447 (55.88) | 50 243 (55.50) | 6204 (59.11) | <.001 |
| Black non-Hispanic, n (%) | 33 106 (32.77) | 30 665 (33.88) | 2441 (23.26) | <.001 |
| Household income in $10 000 s, mean (SD) | 5.12 (1.34) | 5.10 (1.34) | 5.34 (1.37) | <.001 |
| Insurance, n (%) | <0.001 | |||
| Commercial | 49 232 (48.74) | 42 205 (46.62) | 7027 (66.95) | |
| Medicare | 50 091 (49.59) | 46 735 (51.63) | 3356 (31.97) | |
| Charlson Comorbidity Score, mean (SD) | 2.07 (2.39) | 2.13 (2.43) | 1.63 (2.00) | <.001 |
| Diabetes, n (%) | 35 343 (34.99) | 31 695 (35.01) | 3648 (34.76) | .601 |
| HbA1c %, mean (SD) | 7.57 (1.62) | 7.58 (1.63) | 7.46 (1.47) | <.001 |
| ≥8% (63.9 mmol/mol), n (%) | 9708 (29.87) | 8785 (30.22) | 923 (26.87) | <.001 |
| Hypertension, n (%) | 95 712 (94.75) | 85 791 (94.77) | 9921 (94.52) | .273 |
| SBP, mean (SD), mmHg | 136.26 (13.46) | 136.43 (13.58) | 134.80 (12.25) | <.001 |
| Baseline SBP ≥140 mmHg | 35 101 (37.07) | 31 973 (37.68) | 3128 (31.79) | <.001 |
| DBP, mean (SD), mmHg | 78.08 (8.48) | 77.92 (8.53) | 79.43 (7.91) | <.001 |
| Baseline DBP ≥90 mmHg | 8268 (8.73) | 7324 (8.63) | 944 (9.59) | .001 |
| Utilization rate, mean (SD) | ||||
| Primary care providers | 4.90 (3.42) | 5.04 (3.47) | 3.70 (2.60) | <.001 |
| Specialists (per 100 subjects) | 5.46 (24.04) | 5.86 (24.90) | 1.96 (14.28) | <.001 |
| Telephone calls | 4.45 (6.72) | 4.36 (6.84) | 5.20 (5.52) | <.001 |
| Hospitalizations (per 100 subjects) | 4.18 (25.92) | 4.08 (25.67) | 5.03 (27.98) | <.001 |
| ED visits (per 100 subjects) | 18.39 (70.84) | 18.07 (71.27) | 21.18 (66.98) | <.001 |
| Portal medical advice messaging over 6-month time interval, n (%) | ||||
| Portal nonusers | 89 880 (90.05) | 89 880 (100.00) | 0 (0.00) | |
| Portal users | ||||
| Portal users but no message use | 1811 (1.81) | 0 (0.00) | 1811 (18.24) | |
| Frequency of messaging | ||||
| Low (<3 messages) | 6443 (6.46) | 0 (0.00) | 6443 (64.90) | |
| Medium (3–6 messages) | 1261 (1.26) | 0 (0.00) | 1261 (12.70) | |
| High (≥7 messages) | 413 (0.41) | 0 (0.00) | 413 (4.16) | |
| Intensity of messaging | ||||
| Low (1 thread) | 7227 (7.24) | 0 (0.00) | 7227 (72.79) | |
| Medium (2–3 threads) | 667 (0.67) | 0 (0.00) | 667 (6.72) | |
| High (≥4 threads) | 223 (0.22) | 0 (0.00) | 223 (2.25) | |
Portal user medical advice messaging patterns
Most portal users were either “infrequent low-intensity users of all modules” or “electronic messenger–medical record reviewer–appointment makers” (50.7% and 43.7%, respectively). Most of the portal “write” activity (eg, editing) occurred within messaging and medication renewal modules. Seventy-three percent of portal users had activity documented within the medical advice messaging module. Most portal users were classified as low-frequency and low-intensity medical advice messengers when use patterns over 6-month time intervals (Table 1) were examined. However, there were demographic differences among patients with the highest frequency of messaging (age ≥50 vs <50: 4.6% vs 3.9%, P = .03; female vs male: 4.9% vs 3.8%, P < .01; white non-Hispanic vs black non-Hispanic: 4.9% vs 2.6%, P < .001; diabetes and hypertension vs hypertension only vs diabetes only: 6.1% vs 3.8% vs 2.8%, P < .001; data not shown in tables). Additionally, there were demographic differences among patients with the highest intensity of messaging (age ≥50 vs <50: 2.5% vs 2.1%, P = 0.6; female vs male: 2.8% vs 1.9%, P < .01; white non-Hispanic vs black non-Hispanic: 2.7% vs 1.3%, P < .001; diabetes and hypertension vs hypertension only vs diabetes only: 3.3% vs 2.1% vs 1.2%, P < .01; data not shown in tables).
Time trends in HbA1c
In the multivariate analysis of patients with diabetes, there was a dose effect of medical advice messaging, whereby increasing levels of message intensity was associated with greater decreases in HbA1c levels (Table 2). This pattern of association was not consistently observed in the stratified regression analysis of patients with baseline HbA1c ≥8% (63.9 mmol/mol). Trends in HbA1c were also associated with patient demographics and service utilization. Increasing age, female sex, and higher household income were associated with decreases in HbA1c, whereas black non-Hispanic race (compared to white non-Hispanic race), Medicaid insurance (compared to commercial), and higher Charlson Comorbidity Scores were associated with increases in HbA1c. The strength and direction of the relationship between service utilization and trends in HbA1c varied; however, increased rates of PCP, specialty, and ED visits were consistently associated with increases in HbA1c.
Table 2.
Propensity score–adjusted fixed effects multivariable regression panel analysis of portal medical advice messaging and HbA1c pre-post index change among patients with diabetes
| Covariates | Beta estimate (95% CI; P-value) |
|
|---|---|---|
| Patients with diabetes | Patients with baseline | |
| n = 35 343 | HbA1c ≥8% (63.9 mmol/mol) (n = 9708) | |
| Medical advice messaging, 6 months | ||
| No portal use | Reference | Reference |
| Portal use but no message use | −0.28 (−0.34 to −0.22)* | −0.56 (−0.75 to −0.38)* |
| Intensity low | −0.28 (−0.32 to −0.24)* | −0.01 (−0.14 to 0.11) |
| Intensity medium | −0.41 (−0.52 to −0.31)* | 0.06 (−0.43 to 0.55) |
| Intensity high | −0.43 (−0.60 to 0.27)* | −0.51 (−1.08 to 0.06) |
| Demographics | ||
| Age | −0.02 (−0.02 to −0.01)* | −0.01 (−0.02 to 0.00) |
| Female | −0.17 (−0.20 to −0.14)* | −0.12 (−0.21 to −0.04)** |
| Black non-Hispanic | 0.60 (0.48 to 0.73)* | 0.98 (0.64 to 1.33)* |
| Household income in $10 000 s | −0.05 (−0.07 to −0.04)* | −0.10 (−0.14 to −0.06)* |
| Insurance | ||
| Commercial | Reference | Reference |
| Medicaid | 0.15 (0.08 to 0.22)* | 0.31 (0.16 to 0.46)* |
| Medicare | −0.03 (−0.01 to 0.06) | 0.07 (−0.02 to 0.17) |
| Charlson Comorbidity Score | 0.02 (0.02 to 0.03)* | 0.03 (0.01 to 0.04)* |
| Utilization rate | ||
| Primary care provider | 0.02 (0.02 to 0.02)* | 0.01 (0.00 to 0.02)** |
| Specialist | 0.36 (0.33 to 0.39)* | 0.01 (−0.05 to 0.08) |
| Telephone | 0.01 (0.01 to 0.01)* | −0.00 (−0.01 to −0.00)** |
| Hospitalization | −0.04 (−0.06 to −0.01)** | 0.04 (−0.02 to 0.10) |
| ED | 0.03 (0.02 to 0.04)* | 0.05 (0.02 to 0.08)* |
*P ≤ .001; **P < .01; ***P < .05.
Sample interpretation of β estimates: An increase in (or comparatively higher) HbA1c is indicated by positive estimates. A decrease in (or comparatively lower) HbA1c is indicated by negative estimates.
Time trends in blood pressure
In the multivariate analysis of patients with hypertension, the strength and direction of the relationship between medical advice messaging and changes in blood pressure varied (Table 3). The results suggest that high-intensity messaging may be associated with substantial decreases in SBP; however, the proportion of patients characterized as high-intensity messengers was considerably low. The observed relationship between patient demographics and service utilization also varied. Older age was consistently associated with increases in SBP and decreases in DBP. Female sex (compared to male) was associated with decreases in SBP and DBP, although not always statistically significant. Black non-Hispanic race was consistently associated with increases in blood pressure. Living in regions with higher household incomes was associated with decreases in SBP, whereas Medicaid and Medicare insurance were associated with increases in SBP. Among service utilization, only increased rate of ED use was consistently associated with increases in both SBP and DBP.
Table 3.
Propensity score–adjusted fixed effects multivariable regression panel analysis of portal medical advice messaging and blood pressure pre-post index change among patients with hypertension
| Covariates | Beta estimate (95% CI; P-value) |
|||
|---|---|---|---|---|
| Changes in SBP among all patients with hypertension | Patients with baseline SBP ≥140 mmHg | Changes in DBP among all patients with hypertension | Patients with baseline DBP≥90 mmHg | |
| (n = 95 712) | (n = 35 101) | (n = 95 712) | (n = 8268) | |
| Messaging, 6 months | ||||
| No portal use | Reference | Reference | Reference | Reference |
| Portal use, but no messaging | 0.08 (−0.23, 0.39) | 1.20 (−0.69 to 3.08) | 0.21 (0.05 to 0.37)** | 1.01 (−0.62 to 2.65) |
| Intensity low | −0.32 (−0.49 to −0.15)* | 0.37 (−0.77 to 1.51) | 0.39 (0.30 to 0.49)* | −0.71 (−1.76 to 0.34) |
| Intensity medium | −0.09 (−0.49 to 0.30) | −4.40 (−9.05 to 0.25) | 0.66 (0.45 to 0.88)* | −2.65 (−0.846 to 3.17) |
| Intensity high | 0.14 (−0.46 to 0.74) | −17.00 (−32.09 to −1.92)*** | −0.53 (−0.86 to −0.20)** | 0.00 (0.00 to 0.00) |
| Demographics | ||||
| Age | 0.18 (0.17 to 0.20)* | 0.24 (0.14 to 0.34)* | −0.25 (−0.26 to 0.24)* | −0.10 (−0.20 to −0.00)*** |
| Female | −0.09 (−0.22 to 0.03) | −0.29 (−1.15 to 0.57) | −1.35 (−1.43 to −1.28)* | −0.60 (−1.47 to 0.27) |
| Black non-Hispanic | 5.79 (5.23 to 6.35)* | 6.81 (3.14 to 10.47)* | 1.39 (1.08 to 1.70)* | 0.68 (−2.80 to 4.17) |
| Household income in $10 000 s | −0.47 (−0.54 to −0.40)* | −0.77 (−1.21 to −0.33)* | 0.02 (−0.01 to 0.06) | 0.23 (−0.19 to 0.66) |
| Insurance | ||||
| Commercial | Reference | Reference | Reference | Reference |
| Medicaid | 0.30 (0.03 to 0.57)*** | 2.37 (0.09 to 4.65)*** | −0.85 (−1.00 to −0.70)* | 2.54 (0.71 to 4.38)** |
| Medicare | 0.65 (0.50 to 0.79)* | 1.08 (0.09 to 2.07)*** | −1.08 (−1.16 to −1.00)* | 0.29 (−0.85 to 1.43) |
| Charlson Score | −0.00 (−0.02 to 0.01) | 0.49 (0.36 to 0.62)* | −0.43 (−0.43 to −0.42)* | 0.05 (−0.16 to 0.26) |
| Utilization rate | ||||
| Primary care provider | −0.10 (−0.10 to −0.09)* | 0.61 (0.50 to 0.71)* | −0.04 (−0.05 to −0.04)* | 0.39 (0.24 to 0.55)* |
| Specialist | −0.47 (−0.60 to −0.33)* | 1.16 (−0.34 to 2.65) | −1.08 (−1.15 to −1.00)* | −0.52 (−2.39 to 1.35) |
| Telephone | −0.02 (−0.03 to −0.02)* | −0.07 (−0.13 to −0.01)*** | 0.01 (0.01 to 0.02)* | −0.04 (−0.12 to 0.05) |
| Hospitalization | −0.32 (−0.42 to −0.22)* | 1.01 (−0.60 to 2.62) | −0.39 (−0.45 to −0.34)* | 1.53 (−1.79 to 4.85) |
| ED | 0.54 (0.50 to 0.58)* | 1.45 (1.03 to 1.87)* | 0.18 (0.16 to 0.20)* | 1.26 (0.70 to 1.82)* |
*P ≤ .001; **P < .01; ***P < .05.
Sample interpretation of β estimates: An increase in (or comparatively higher) SBP or DBP is indicated by positive estimates. A decrease in (or comparatively lower) SBP or DBP is indicated by negative estimates.
DISCUSSION
In a previous study, we observed among primary care patients with hypertension and/or diabetes that portal users compared to nonusers had higher rates of primary care clinic visits and telephone encounters, along with greater decreases in BP and HbA1c levels.13 In this follow-up study, we examined the impact of medical advice messaging patterns to further explore the relationship between portal use and disease control. We observed a dose-response effect of the intensity of medical advice messaging (bidirectional portal communication) on decreases in HbA1c. We did not observe a consistent effect on changes in blood pressure. The frequency of portal medical advice messaging was not associated with trends in clinical outcomes.
Our findings confirm several recent studies examining associations between active portal messaging and diabetes control.15–18 Devkota et al.15 found that patients with more active portal e-mail communication are more likely to have HbA1c control compared to less active e-mail users (readers or nonusers). Petullo et al.16 demonstrated that portal messaging was associated with lower HbA1c and that the frequency of messaging was not associated with HbA1c, even after accounting for diabetes-specific messaging. Shimada et al.17 further suggested that sustained use of portal messaging for several years may have a major impact on outcomes.
This study did not demonstrate a dose-response effect of the intensity of medical advice messaging on blood pressure control. Similar to Manard et al.’s19 recent study, portal use is no longer associated with observed improvements in blood pressure after accounting for sociodemographics and service utilization. McClellan et al.20 suggest that while patient-initiated messaging may improve adherence to recommended tests, it is not sufficient to improve clinical outcomes for patients with diabetes or hypertension in the absence of additional interventions. Patient portal messaging in conjunction with case management appears to be a more effective strategy for chronic conditions (eg, diabetes, hypertension, depression) that require self-monitoring and medication dose adjustment.9,21–25
The association between patient portal use and health outcomes may vary based on the combination of different portal features used and how patients use each feature for self-management of their health conditions.17 Half of the patient portal users in our study population rarely used the technology. The authors previously reported on patients’ perceptions of facilitators of and barriers to adopting portal technology.12 Cumbersome processes for accessing portals and variations in provider availability for appointment scheduling and response times to messages disengage portal users. Patient stakeholders recommend that health systems screen for eHealth literacy, provide training and technical support, promote proxy users for older adults, and institute quality assurance that ensures patients’ experiences will not vary across the system.
Similar to the work of Mikles et al.,26 this study demonstrates that there are demographic differences in patterns of medical advice messaging among active portal users. Medical advice messaging patterns may reflect a combination of provider responsiveness and patients’ level of engagement in self-management. Wolcott et al.11 recently showed that provider messaging levels predict patients’ subsequent communication behaviors. To optimize technology-enabled disease management, it is important to identify which patients are most likely to adopt technology, understand how they actually use it, and optimize providers’ use of portal messaging to enhance communication.
Health system promotion of patient portals for enhanced access to care and communication can inadvertently widen the digital divide for patients least likely to use it. Patients’ use of portals is driven by attitudes about and preferences for using technology, perceived usability, provider endorsement, care received prior to accessing patient portals, demand for different types of health services, preferences on how to access services, and multiple sociodemographic factors.21,27,28 Health systems must develop effective solutions to surmount these barriers.
This retrospective observational study has several limitations. The study reflects the experience of one organization and may have limited external generalizability. The authors restricted the analysis to patients with diabetes or hypertension and cannot draw conclusions about the role of medical advice portal messaging and disease control for other chronic conditions. The overall proportion of portal users with medical advice messaging activity who had either HgbA1C >8%, SBP >140, or DBP >90 was lower than that of nonusers; therefore, the study may have been underpowered to consistently detect an association between messaging and clinical outcomes among these subpopulations. The research team was unable to access the contents of the e-mails; therefore, this study cannot decipher whether the association between the intensity of medical advice messaging and clinical outcomes was directly linked to disease management. Additionally, unmeasured patient- or provider-level factors that influence portal use patterns and/or clinical outcomes were either not collected as part of routine care in our health system (eg, health literacy, computer literacy, patient activation measures) or not retrievable as structured data (eg, provider use of medication adjustment algorithms, successes or challenges with incorporating portal messaging into care team workflows). This study’s focus on the intensity of bidirectional portal messaging as a proxy for interactive communication may have missed instances where providers or clinical staff reached out by telephone to patients to quickly follow up on a portal message. In fact, we demonstrated in a prior study of the same population that portal use was associated with an increase in telephone encounters.13 Finally, randomization of patients to portal use or nonuse during system implementation would have eliminated selection bias; however, doing so was simply not practical. Accordingly, this study used a propensity score–adjusted multivariable fixed effects regression panel analysis to improve the precision of estimating the association between medical advice messaging and disease control. The major strength of this study is that the statistical approach facilitated a comparison of portal nonusers to users who use the technology with varying degrees of frequency or intensity.
CONCLUSIONS
As our experience demonstrates, health systems will continue to face challenges with the digital divide not only between portal users and nonusers, but also among patients who have adopted the technology to varying degrees. There are multiple complex factors that impact the use of eHealth technology (including portal messaging). While a preponderance of studies has focused on deciphering characteristics of patients engaged or disengaged with using technology for self-care, few studies examine innovations in technology optimization targeting the community, health systems, providers, patients, and/or caregivers. Future research must focus more on implementation science that elucidates best practices for enhancing patient and provider use of eHealth technology while improving patient outcomes across a broader range of health conditions.
ACKNOWLEDGMENTS
Contributors: The authors wish to thank the Ochsner Center for Applied Health Services team members Timothy Hilbun (research department information analyst) and Jewel Harden-Barrios (clinical research coordinator) for their contributions to data management.
Funding/Support: None.
Prior Presentations: This work was presented as a poster abstract at the Southern Regional Meeting of the Southern Society of Clinical Investigations in February 2016.
COMPETING INTERESTS
None.
REFERENCES
- 1. Pew Research Center. Pew Internet and American Life Project. Washington, DC: Pew Research Center. [Google Scholar]
- 2. Perrin A, Duggan M. America’s Internet Access: 2000–2015. June 26, 2015. 2015. www.pewinternet.org/2015/06/26/americans-internet-access-2000-2015/. Accessed June 6, 2017. [Google Scholar]
- 3. Prestin A, Vieux SN, Chou WY. Is online health activity alive and well or flatlining? Findings from 10 years of the Health Information National Trends Survey. J Health Commun. 2015;207:790–98. [DOI] [PubMed] [Google Scholar]
- 4. Otte-Trojel T, de Bont A, van de Klundert J, Rundall TG. Characteristics of patient portals developed in the context of health information exchanges: early policy effects of incentives in the meaningful use program in the United States. J Med Internet Res. 2014;1611:e258. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Hsiao CJ, Hing E. Use and characteristics of electronic health record systems among office-based physician practices: United States, 2001–2013. NCHS Data Brief. 2014;143:1–8. [PubMed] [Google Scholar]
- 6. de Jong CC, Ros WJ, Schrijvers G. The effects on health behavior and health outcomes of Internet-based asynchronous communication between health providers and patients with a chronic condition: a systematic review. J Med Internet Res. 2014;161:e19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Lee JL, Choudhry NK, Wu AW, Matlin OS, Brennan TA, Shrank WH. Patient use of email, Facebook, and physician websites to communicate with physicians: a National Online Survey of Retail Pharmacy Users. J Gen Intern Med. 2016;311:45–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Haun JN, Patel NR, Lind JD, Antinori N. Large-scale survey findings inform patients’ experiences in using secure messaging to engage in patient-provider communication and self-care management: a quantitative assessment. J Med Internet Res. 2015;1712:e282. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Goldzweig CL, Towfigh AA, Paige NM et al. , Systematic Review: Secure Messaging Between Providers and Patients, and Patients’ Access to Their Own Medical Record: Evidence on Health Outcomes, Satisfaction, Efficiency and Attitudes. Washington, DC: US Department of Veterans Affairs; 2012. [PubMed] [Google Scholar]
- 10. Wells S, Rozenblum R, Park A, Dunn M, Bates DW. Organizational strategies for promoting patient and provider uptake of personal health records. J Am Med Inform Assoc. 2015;221:213–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Wolcott V, Agarwal R, Nelson DA. Is provider secure messaging associated with patient messaging behavior? evidence from the US Army. J Med Internet Res. 2017;194:e103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Price-Haywood EG, Harden-Barrios J, Ulep R, Luo Q. eHealth literacy: patient engagement in identifying strategies to encourage use of patient portals among older adults. Popul Health Manag. 2017;206:486–94. [DOI] [PubMed] [Google Scholar]
- 13. Price-Haywood EG, Luo Q. Primary care practice reengineering and associations with patient portal use, service utilization, and disease control among patients with hypertension and/or diabetes. Ochsner J. 2017;171:103–11. [PMC free article] [PubMed] [Google Scholar]
- 14. Health Resources and Services Administration. UDS Mapper [Web Application]. 2016. www.udsmapper.org. Accessed July 22, 2015. [Google Scholar]
- 15. Devkota B, Salas J, Sayavong S, Scherrer JF. Use of an online patient portal and glucose control in primary care patients with diabetes. Popul Health Manag. 2016;192:125–31. [DOI] [PubMed] [Google Scholar]
- 16. Petullo B, Noble B, Dungan KM. Effect of electronic messaging on glucose control and hospital admissions among patients with diabetes. Diabetes Technol Ther. 2016;189:555–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Shimada SL, Allison JJ, Rosen AK, Feng H, Houston TK. Sustained use of patient portal features and improvements in diabetes physiological measures. J Med Internet Res. 2016;187:e179. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Harris LT, Koepsell TD, Haneuse SJ, Martin DP, Ralston JD. Glycemic control associated with secure patient-provider messaging within a shared electronic medical record: a longitudinal analysis. Diabetes Care. 2013;369:2726–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Manard W, Scherrer JF, Salas J, Schneider FD. Patient portal use and blood pressure control in newly diagnosed hypertension. J Am Board Fam Med. 2016;294:452–59. [DOI] [PubMed] [Google Scholar]
- 20. McClellan SR, Panattoni L, Chan AS, Tai-Seale M. Patient-initiated electronic messages and quality of care for patients with diabetes and hypertension in a large fee-for-service medical group: results from a natural experiment. Med Care. 2016;543:287–95. [DOI] [PubMed] [Google Scholar]
- 21. Goldzweig CL, Orshansky G, Paige NM et al. , Electronic patient portals: evidence on health outcomes, satisfaction, efficiency, and attitudes: a systematic review. Ann Intern Med. 2013;15910:677–87. [DOI] [PubMed] [Google Scholar]
- 22. Ralston JD, Cook AJ, Anderson ML et al. , Home blood pressure monitoring, secure electronic messaging and medication intensification for improving hypertension control: a mediation analysis. Appl Clin Inform. 2014;51:232–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Ralston JD, Hirsch IB, Hoath J, Mullen M, Cheadle A, Goldberg HI. Web-based collaborative care for type 2 diabetes: a pilot randomized trial. Diabetes Care. 2009;322:234–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Magid DJ, Olson KL, Billups SJ, Wagner NM, Lyons EE, Kroner BA. A pharmacist-led, American Heart Association Heart360 Web-enabled home blood pressure monitoring program. Circ Cardiovasc Qual Outcomes. 2013;62:157–63. [DOI] [PubMed] [Google Scholar]
- 25. Simon GE, Ralston JD, Savarino J, Pabiniak C, Wentzel C, Operskalski BH. Randomized trial of depression follow-up care by online messaging. J Gen Intern Med. 2011;267:698–704. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Mikles SP, Mielenz TJ. Characteristics of electronic patient-provider messaging system utilisation in an urban health care organisation. J Innov Health Inform. 2014;221:214–21. [DOI] [PubMed] [Google Scholar]
- 27. Riippa I, Linna M, Ronkko I, Kroger V. Use of an electronic patient portal among the chronically ill: an observational study. J Med Internet Res. 2014;1612:e275. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Irizarry T, DeVito Dabbs A, Curran CR. Patient portals and patient engagement: a state of the science review. J Med Internet Res. 2015;176:e148. [DOI] [PMC free article] [PubMed] [Google Scholar]