This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Personal health records (PHRs) have the potential to improve patient self-management for chronic conditions such as diabetes. However, evidence is mixed as to whether there is an association between PHR use and improved health outcomes.
The aim of this study was to evaluate the association between sustained use of specific patient portal features (Web-based prescription refill and secure messaging—SM) and physiological measures important for the management of type 2 diabetes.
Using a retrospective cohort design, including Veterans with diabetes registered for the My Health
By 2013 to 2014, 34.13% (38,113/111,686) of the cohort was using Web-based refills, and 15.75% (17,592/111,686) of the cohort was using SM. Users were slightly younger (
Although rates of use of the refill function were higher within the population, sustained SM use had a greater impact on HbA1c. Evaluations of patient portals should consider that individual components may have differential effects on health improvements.
Diabetes affects over 29 million Americans [
Patients with diabetes and other chronic diseases do not do well with episodic, transactional care limited to in-person visits. The Institute of Medicine [
Evidence for patient portal effectiveness for chronic disease management is limited, and association with outcomes is mixed [
Portals vary widely, adding to the difficulty in evaluating any effects they may have on patients’ health outcomes. Some are tethered to a health care system, others are not, some are disease specific, whereas most are not [
This study examines whether diabetes outcomes are improved for patients with type 2 diabetes who initiate use of key features of the MHV patient portal compared with similar patients with type 2 diabetes who are also registered for the portal but do not initiate use of any of these features. To answer this question, we focused on patients with a diagnosis of type 2 diabetes who had at least one uncontrolled physiological measure (hemoglobin A1c, LDL cholesterol, blood pressure) at baseline (2009-2010) to examine whether those who had used the portal’s Web-based prescription refill or SM features for the first time between 2010 and 2013 were more likely than nonusers to achieve control at follow-up (2013-2014). We also sought to explore both the separate and combined effects of Web-based refill and SM use on physiological measures and whether sustained use was associated with a greater probability of achieving control.
We conducted a 5-year retrospective cohort study of Veterans with type 2 diabetes registered for the MHV portal. Data for these analyses came from the Veteran’s Health Administration’s Corporate Data Warehouse, including administrative data, clinical records for inpatient and outpatient care, and MHV registration and use data. We used International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes (October 1, 2007-March 31, 2009) to determine type 2 diabetes diagnosis and determine patient characteristics at baseline. Data from April 1, 2009 through March 31, 2014 were used to assess MHV use over time. Intermediate physiological measures obtained during clinical care were obtained at baseline and follow-up. In addition, we linked income and educational attainment variables from the US Census Bureau’s 2007- 2011 American Community Survey (5-year estimates) to each Veteran via postal code.
We identified patients who had at least two outpatient records or one inpatient record with an ICD-9-CM diagnosis code for type 2 diabetes by March 2009 (N=1,207,703). Use of two or more diabetes-related ICD-9-CM codes from inpatient or outpatient visits has previously been determined to be the most accurate way to identify patients with diabetes in VA administrative data [
Cohort Selection.
We used the American Diabetes Association’s guidelines to define cutoffs for glycemic, cholesterol, and blood pressure control [
Among Veterans registered by April 1, 2013, we measured use of two key features of the portal, which had been available throughout the study period: the Web-based prescription refill feature and the SM feature and used a binary indicator of any use to describe the samples. We assessed how often each patient used each feature during the potential exposure period (April 2010-March 2013). At some facilities, patients were prompted to try these features (eg, send a test message to one’s primary care team) as part of a MHV training. We therefore defined “use” as two or more prescriptions filled online via the MHV portal per year or two or more SMs sent per year, to ensure we captured actual use and not just attendance at a training session. To measure dose of exposure, our primary measure of use for each portal feature was a categorical variable indicating whether a patient had used each feature two or more times per year over 1 year, 2 years, or for 3 or more years during the potential exposure period. A continuous variable measuring years of use (ie, years with 2+ refills or 2+ SMs sent) for each portal feature was used for tests for trend.
Other covariates we used included demographic characteristics such as patient age, gender, race or ethnicity, urban, suburban, or rural residence, educational attainment, and income. In multivariable models, we adjusted for age, gender, race, comorbidities, and available measures of socioeconomic status because these have been significantly associated with adoption of SM and patient portals in previous studies [
We characterized the overall cohort and examined means and distributions of patient demographic and clinical characteristics by use, both overall and for those with specific uncontrolled physiological measures at baseline. We calculated the proportion of patients with diabetes in our cohort using each feature over each year of the study and the average number of prescriptions refilled or secure messages sent during each year. Our primary goal was to assess the association of use of patient portal features with change in diabetes-relevant physiological measures (HbA1c, LDL, BP). To achieve this goal, we first calculated means and binomial confidence intervals (CIs) for the proportion of patients who were uncontrolled at baseline who achieved control at follow-up, stratified by the number of years of use of the SM or Web-based refill features. We then constructed a series of logistic regression models predicting control of each physiological measure at follow-up based on categorical measures of portal use (years of use of each feature), adjusting for the covariates described previously. All logistic regression models were adjusted for patient age; gender; race or ethnicity; eligibility for free VA health care; number of Elixhauser comorbidities at baseline; number of primary care visits at baseline (in 2009-2010) and during the study period (2010-2014); urban, suburban, rural, or highly rural residence; median income by postal code; and the percentage of college graduates in the patient’s residential postal code. In addition, models for control of blood pressure, cholesterol, and HbA1c at follow-up (2013-2014) were adjusted for the patient’s mean baseline blood pressure, LDL cholesterol, or HbA1c value in 2009 to 2010, respectively. Separate models were first run for each feature (Web-based prescription refill use and SM use) because there was a moderate correlation between uses of the two features. To further evaluate the independent effect of each feature, we also ran combined logistic models, which included both Web-based prescription refill use and SM use in the same models. To test for dose response, we then ran tests of trend treating the number of years of use of each feature as a continuous variable. We also conducted sensitivity analyses to see whether results changed depending on (1) our definition of use (ie, defining use as one or more uses of a feature in a given year) or (2) inclusion of patients who met other inclusion criteria but were controlled at baseline in the analysis sample.
Within our cohort of 111,686 patients (see
Characteristics of patients with type 2 diabetes registered for My HealtheVet, overall and by use or nonuse of the Web-based refill or secure messaging features as of March 2014.
Variables |
Overall | Used neither Web-based refill nor SM as of March 2014 (nonusers) | Used Web-based refill or SM or both as of March 2014 (users) | Difference between user and nonuser groups (Pearson’s chi-square or 2-sided |
|
111,686 | 61,204 | 50,482 | |||
62.05 (9.6) | 63.22 (9.6) | 60.63 (9.5) | |||
3.58% | 3.16% | 4.08% | χ21 = 67.2, |
||
White | 68.87% | 67.25% | 70.84% | Reference group for χ2 | |
African-American | 16.95% | 18.64% | 14.90% | χ21 = 280.3, |
|
Latino | 5.69% | 5.63% | 5.76% | χ21= 1.1, |
|
Native Hawaiian or |
1.10% | 1.08% | 1.12% | χ21 = 0.1, |
|
Asian | 0.81% | 0.73% | 0.90% | χ21 = 5.9, |
|
American Indian or |
0.71% | 0.73% | 0.70% | χ21 = 1.8, |
|
Unknown to patient, |
5.87% | 5.94% | 5.79% | χ21 = 9.1, |
|
20.43% | 21.71% | 18.87% | χ21 = 137.6, |
||
33,548.86 (8,926.98) | 33,532.67 |
33,568.51 (8,842.24) | |||
23.46% (12.7 ) | 23.39% (12.8) | 23.54% |
|||
Urban (%) | 73.23% | 73.27% | 73.19% | Reference group for χ2 | |
Suburban (%) | 13.40% | 13.46% | 13.32% | χ21 = 0.3, |
|
Rural (%) | 7.23% | 7.32% | 7.12% | χ21 = 1.3, |
|
Highly rural (%) | 6.14% | 5.95% | 6.38% | χ21 = 7.5, |
|
5.57 (2.5) | 5.56 (2.6) | 5.59 (2.5) | |||
4.40 (3.6) | 4.37 (3.6) | 4.43 (3.6) | |||
17.59 |
16.97 (12.6) | 18.34 (13.0) |
Compared with patients who did not use either of the features, patients who used Web-based refill or SM were slightly younger (60.6 years vs 63.2 years,
There were no significant differences in the number of Elixhauser comorbidities at baseline (
Further detail describing the characteristics based on each uncontrolled measure (ie, the sample for each logistic regression model) is summarized in
Demographics of patients with type 2 diabetes registered for My HealtheVet by uncontrolled physiological measure at baseline and by use of the portal.
Variables |
Uncontrolled Measure | ||||||
Hemoglobin A1c |
Low-density Lipoprotein |
Blood Pressure |
|||||
Registered, no use | Used SM or Web-based refill | Registered, no use | Used SM or Web-based refill | Registered, no use | Used SM or Web-based refill | ||
36,305 | 30,917 | 18,898 | 16,153 | 31,907 | 26,471 | ||
62.66 (9.2) | 60.28 (9.2) | 61.47 (9.6) | 58.68 (9.6) | 62.63 (9.7) | 60.13 (9.7) | ||
2.86% | 3.71% | 4.87% | 6.53% | 2.88% | 3.58% | ||
White | 66.88% | 70.98% | 64.21% | 67.62% | 64.55% | 68.44% | |
African-American | 19.12% | 14.84% | 22.14% | 17.56% | 21.38% | 17.25% | |
Latino | 6.08% | 6.03% | 5.79% | 6.27% | 5.53% | 5.63% | |
Native Hawaiian Pacific Islander | 1.08% | 1.11% | 1.04% | 1.08% | 1.08% | 1.21% | |
Asian | 0.75% | 0.89% | 0.71% | 1.00% | 0.72% | 0.87% | |
American Indian or Alaska Native | 0.73% | 0.70% | 0.69% | 0.84% | 0.71% | 0.70% | |
Unknown to patient, refused, or missing | 5.36% | 5.45% | 5.42% | 5.64% | 6.02% | 5.91% | |
22.38% | 19.40% | 21.61% | 18.76% | 21.93% | 18.96% | ||
33,453.54 |
33,548.58 |
33,111.41 |
33,197.85 |
33,364.98 |
33,424.07 |
||
23.12% (12.7) | 23.32% (12.6) | 22.95% (12.6) | 23.20% (12.4) | 23.27% (12.7) | 23.46% (12.6) | ||
Urban (%) | 73.21% | 73.12% | 73.32% | 73.43% | 73.40% | 72.91% | |
Suburban (%) | 13.34% | 13.15% | 13.64% | 13.15% | 13.54% | 13.47% | |
Rural (%) | 7.41% | 7.24% | 7.18% | 7.05% | 7.14% | 7.12% | |
Highly rural (%) | 6.04% | 6.50% | 5.86% | 6.37% | 5.93% | 6.50% | |
5.70 (2.6) | 5.72 (2.5) | 5.40 (2.5) | 5.45 (2.4) | 5.54 (2.5) | 5.51 (2.4) | ||
4.64 (3.8) | 4.66 (3.7) | 4.32 (3.5) | 4.38 (3.5) | 4.29 (3.5) | 4.30 (3.4) | ||
18.04 (12.9) | 19.17 (13.3) | 17.09 (12.4) | 18.29 (12.7) | 16.97 (12.5) | 18.17 (12.5) |
Use of Web-based refills and SM increased steadily from 2010 to 2014 (
Proportion of patients with type 2 diabetes registered for My HealtheVet and first using Web-based prescription refills or secure messaging after 2010, increase in feature adoption over time, and average number of uses per user per year.
The logistic regression results are presented in
Our single-feature logistic regression models (Models 1a-c and Models 2a-c) showed that patients with uncontrolled HbA1c at baseline (2009-2010) were significantly more likely to achieve glycemic control at follow-up (2013-2014) if they used SM for 2 or more years. The odds of having an HbA1c below 7.0% (53 mmol/mol) at follow-up were 22% higher (after 2 years of use, odds ratio: OR=1.22, CI: 1.13-1.32) and 28% higher (after 3 or more years, OR=1.28, CI: 1.13-1.44), for those using SM compared with those who never used it.
However, use of Web-based prescription refills was only associated with glycemic control at follow-up after 3 or more years of use (OR=1.07, CI: 1.01-1.14). Those with uncontrolled blood pressure at baseline were significantly more likely to achieve control at follow-up only with 2 (OR=1.06, CI: 1.01-1.12) or 3 or more (OR=1.05, CI: 1.00-1.11) years of Web-based refill use, compared with nonusers. Use of SM was not significantly associated with improvements in blood pressure control. Both Web-based refill use and SM use were significantly associated with improvements in LDL cholesterol levels at follow-up. Compared with nonusers, the odds of users having LDL cholesterol below 100 mg/dL (2.586 mmol/L) were 12% higher with 2 years of Web-based refill use (OR=1.12, CI: 1.05-1.20), 16% higher with 3+ years of Web-based refill Use (OR=1.16, CI: 1.08-1.24), 9% higher with 1 year of SM use (OR=1.09, CI: 1.01-1.18), 17% higher with 2 years of SM use (OR=1.17, CI: 1.07-1.27), and 22% higher with 3+ years of SM use (OR=1.22, CI: 1.06-1.40).
We also ran logistic regression models identical to those mentioned previously that included both years of SM and Web-based refill use in the same model (Models 3a-c), as well as logistic regression models that included years of SM or Web-based refill use as a continuous variable as a test for trend (Models 4a-c). The conclusions remained largely unchanged, although ORs for the association between SM use and LDL were more attenuated (and no longer significant with the exception of 2 years of SM use) in the combined model. The combined model (and test for trend) did not show a significant association between SM use and blood pressure control (
Adjusted odds of being in control at follow-up (OR (95% CI)) for a patient with uncontrolled physiological measures (HbA1c, LDL, or blood pressure) at baseline, based on years of portal feature use.
Modelsa | Health Outcomes in 2013-14 | |||
Web-based prescription refill use | ||||
None | Reference | Reference | Reference | |
1 year | 0.99 (0.93, 1.05) | 1.01 (0.95, 1.08) | 1.02 (0.97, 1.08) | |
2 years | 1.01 (0.95, 1.08) | 1.12 (1.05, 1.20)c | 1.06 (1.01, 1.12)b | |
3 or more years | 1.07 (1.01, 1.14)b | 1.16 (1.08, 1.24)d | 1.05 (1.00, 1.11)b | |
Secure messaging use | ||||
None | Reference | Reference | Reference | |
1 year | 1.03 (0.96, 1.10) | 1.09 (1.01, 1.18)b | 1.03 (0.97, 1.09) | |
2 years | 1.22 (1.13, 1.32)d | 1.17 (1.07, 1.27)c | 1.03 (0.96, 1.10) | |
3 or more years | 1.28 (1.13, 1.44)d | 1.22 (1.06, 1.40)c | 1.00 (0.90, 1.12) | |
Web-based prescription refill use | ||||
None | Reference | Reference | Reference | |
1 year | 0.96 (0.91, 1.03) | 1.01 (0.94, 1.08) | 1.02 (0.97, 1.07) | |
2 years | 0.96 (0.90, 1.03) | 1.13 (1.05, 1.21)c | 1.07 (1.01, 1.13)b | |
3 or more years | 1.00 (0.94, 1.07) | 1.13 (1.05, 1.22)c | 1.08 (1.02, 1.14)c | |
Secure messaging use | ||||
None | Reference | Reference | Reference | |
1 year | 1.04 (0.97, 1.12) | 1.05 (0.97, 1.14) | 1.00 (0.94, 1.07) | |
2 years | 1.24 (1.14, 1.34)d | 1.10 (1.00, 1.21)b | 0.98 (0.91, 1.05) | |
3 or more years | 1.28 (1.12, 1.45)d | 1.12 (0.96, 1.30) | 0.95 (0.85, 1.07) | |
Web-based prescription refill use | ||||
Secure messaging use |
aAll models adjust for patient characteristics in
bOdds ratios are significant at the
cOdds ratios are significant at the
dOdds ratios are significant at the
Proportion controlled at follow-up, out of all diabetics uncontrolled for that specific measure at baseline (proportion and binomial CIs).
We conducted sensitivity analyses to see whether our results would change with the inclusion of those whose physiological measures were controlled at baseline, but otherwise met criteria for inclusion. Although the ORs were attenuated, significant tests for trend revealed the same relationships between feature use and being in control at follow-up for all the measures. Similarly, when use was defined as use of a feature even once in a given year, ORs were again somewhat attenuated; however, the results, including the tests for trend, led to identical conclusions about the associations between feature use and controlled physiological outcomes at follow-up.
Within this cohort of patients with type 2 diabetes and uncontrolled physiological measures, we saw increasing activity on the MHV patient portal between 2010 and 2014. The rate of use and increase in use was greater for Web-based refills than for SM. We observed small, statistically significant, and potentially meaningful improvement in physiological measures among diabetic patients who initiated and sustained use of Web-based refills or SM or both via MHV. However, the association varied by specific MHV feature. Where a significant association was found, use of SM was associated with higher odds of improved outcomes than use of Web-based refills.
The association between use of SM and improved diabetes physiological measures is consistent with that of prior research [
One mechanism by which Web-based medication refills may affect health outcomes may be through improved adherence to prescribed medications. In prior work, MHV use has been associated with improvements in antiretroviral adherence [
SM has been shown to improve patient ratings of patient–provider communication [
This work also expands on previous research that has often focused generally on the patient portal or PHR use [
Patients who used one or both features during the study period were more likely to be younger, female, white, and were less likely to be socioeconomically disadvantaged than other patients with diabetes who met our inclusion criteria. Numerous studies have documented sociodemographic differences in patient portal access and adoption [
There are a number of limitations to this study. The VA patient portal has been deployed nationwide. As all patients are free to choose whether to use the patient portal, it is difficult to limit access or to randomize access to various features to conduct a randomized controlled trial. Because this is an observational study, it is impossible to ensure that the comparison group (ie, the nonusers) is similar in all ways to the portal users. As discussed, we limited the sample to those who had registered to use the portal to reduce heterogeneity in measured and unmeasured confounders. In our prior research [
Recognizing that our study is an observational study and that the associations cannot be considered causal, the availability of multiple years of observational data, detection of a dose response, and adjustment for patient characteristics known to influence technology use and diabetes outcomes strengthen the potential conclusions we can draw from this analysis about the differential effects use of patient portal features may have on physiological outcomes. The results in this study suggest that measuring the relative use and relative association of each feature of a patient portal is critical because each can have a different effect on changes in health care and health outcomes.
Future research should also focus on uncovering the mechanisms (causal pathways) through which portal use leads to physiological improvements. Does improved communication with providers via SM lead to greater patient engagement between visits, sustained behavior changes, better continuity of care, improved medication titration by the clinical team, or improved adherence to medications by the patients? What portion of the engagement might be explained by other portal features such as the ability to track and chart their blood glucose or blood pressure measurements? A study of adult diabetes patients at Kaiser Permanente found that both patient nonadherence to medications for glycemic, lipid, or blood pressure control and lack of provider treatment intensification occurred frequently among patients whose outcomes are above desired target levels [
blood pressure
hemoglobin A1c
International Classification of Diseases, Ninth Revision, Clinical Modification
low density lipoprotein
My HealtheVet
personal health record
secure messaging
Department of Veterans Affairs
Dr. Shimada was supported by a Career Development Award (CDA 10-210) from the Department of Veterans Affairs Health Services Research and Development Service and the VA eHealth QUERI (EHQ 10-190). The contents do not represent the views of the US Department of Veterans Affairs or the US Government.
None declared.