An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Mary E. Reed, DrPH, Andrea Millman, Bruce Fireman, MA, Richard W. Grant, MD, MPH, Romain S. Neugebauer, PhD, John Hsu, MD, MBA, MSCE, Richard Brand, PhD, and Jie Huang, PhD.
Author Information and AffiliationsPersonal health records offer patients the option to access their medical information through a secure “patient portal” and the ability to manage their health care online through interactive tools.
In this study, our research questions asked the following: (1) Which patients use web portal tools, and which do not? Why or why not? (2) How does using web portal tools affect the patient health care experience?
Regular collaboration with a patient partner panel and a clinician and delivery system advisory board informed our study methods and interpretation of results. Via a patient survey we collected patient-reported preferences and experiences when using the patient portal, as well as patient-reported outcomes. We surveyed patients with a chronic condition, by mail, phone, or electronic questionnaire, and paid particular attention to oversampling patients with multiple complex conditions. Using electronic health records (EHRs), we also examined impacts of portal use on health care and health events, including doctor's office visits, emergency department (ED) visits, and hospital stays. Our analyses make rigorous statistical adjustment for sampling strategy and patient characteristics as well as time-changing clinical need.
A total of 1824 respondents completed the study survey (70% response rate), and more than 270 000 patients were included in automated data analyses. Patients who were male, older, identified as non-White, reported lower household income or education, and had lower internet access were less likely to use the patient portal. Patients with more complex conditions and very recent health care activity were more likely to use the portal. Many used the portal in their role as caregiver or family care partner, sometimes across a geographic distance. Among those not using the portal, many reported a preference for in-person or telephone-based health care access. Among portal users, most patients reported that it was a convenient channel for health care access, that the information provided was useful in managing their care, and that using the portal integrated well with other health care services. Those who experienced these portal use benefits also were more likely to report health improvements linked to using the portal. When patients checked their laboratory results through the portal, they were more likely to communicate with providers and to receive treatment intensification quickly. Portal use was associated with an increase in office visits, potentially a signal of increased engagement, and with decreases in downstream emergency visits and preventable hospitalizations, a signal of reduced health events, with greater reductions in patients with complex conditions than in patients with a single chronic condition.
Patient portals are a patient-centered tool with potential to improve health engagement and outcomes. Patient portal use varies by patient demographics, technology access, and clinical need. Using the portal, however, was associated with shifts toward greater communication and office visits with providers, patient-reported health improvements, increased timeliness of treatment, and decreases in health events, with greater improvements in patients with complex chronic conditions. Further research should examine the impact of targeted strategies to educate patients, providers, and health care delivery systems about the potential benefits of and barriers to portal use, including through dissemination of patient-reported experiences.
In survey data, we are limited to patient self-reported use and experiences and cannot establish causality. In the analyses based on automated data, because this an observational study, we cannot rule out unmeasured confounding.
Chronic diseases—including asthma, diabetes, congestive heart failure, coronary artery disease, hypertension, and cardiovascular event risk—account for significant levels of morbidity and mortality in the United States, with an increasing proportion of patients living with multiple complex chronic conditions.1-3 Patients with chronic conditions account for 84% of all US health care spending, more than half of which is among patients with multiple chronic diseases, presenting coordination challenges for both patients and health care providerrs.1,4 Treating complex chronic diseases often demands care strategies that coordinate and integrate multiple clinicians and sites of care.5-7 Without efficiently coordinated care, patients with complex care needs may receive more redundant care and require preventable acute services.1
Although patients with a complex burden of multiple chronic diseases are at higher clinical risk and experience more functional limitations, many of these patients may not receive optimal or safe treatment.8-22 Patients with complex chronic disease are more likely to have an adverse drug event, to be suboptimally treated, and to have preventive hospitalizations.7,23,24 Patient portal tools that improve patient access to their own health information and that help patients communicate asynchronously may offer an additional mechanism for delivering high-quality guideline-recommended care that can improve patient health.
Electronic patient portals, linked to the patient's clinical electronic health record (EHR) offer patients the option to access their own medical information through a secure website; some also offer patients the ability to manage their health care online at any time of day or night through interactive tools, including laboratory result review, visit summaries, secure email messaging to health care providers, medication refill ordering, and wellness programs. Patient access to a portal has great potential to improve patient engagement and change the way that health care is delivered. Secure patient portals have been proposed by the Institute of Medicine as a promising method of decreasing medical errors and increasing health care quality.25
Access to portal tools offers patients the option and ability to access some health services without visiting their providers and pharmacies in person or during business hours. For some patients this may offer convenient ways to avoid traveling to medical facilities and pharmacies, reduce time off from work (or caregiving), and increase communication with health care providers. However, for other patients, including those with limited access to the internet or those who prefer in-person communication with physicians and pharmacists, using the patient portal may be a cumbersome or an ineffective way to access health care. Our study aims to examine which types of patients use the patient portal and how it affects their health care experience.
We reviewed the literature (eg, PubMed, relevant journals, national reports) to identify evidence gaps related to the study's research questions. Existing evidence has shown that patient portals' use can improve patient satisfaction, patient–provider communication, and other patient-reported outcomes, but some studies have had limited scope and sample size.26-29 Patients with the option to email their provider were more satisfied with their care30 and with the convenience of communicating with their provider remotely.31 They reported that having access to their health information made them more engaged in their own care.32 Still, other small studies have found that some patients had concerns about confidentiality33 or worried that use of the portal would displace the person-to-person contact with their provider.34 Overall, the research evidence is mixed,35-37 with opportunities to contribute to the evidence through rigorous study design, examining more detailed characteristics of patients who do and do not use the portal, and in thorough measurement of health outcomes.
Our study examines the patient portal, a novel mechanism for delivering health care. Because its utility spans many conditions, the patient portal has substantial potential to affect large chronic disease burden. Specifically, our study examines 2 questions: (1) Which patients use patient portal tools, why (or why not), and when? (2) How does using patient portal tools affect the patients' health care access, experience, and health? We hypothesized that many factors might influence patients' portal use and the timing of this use, including sociodemographic characteristics, technology access and comfort, health engagement, and recent clinical need.
We hypothesized that the access to health information, tools to help self-manage their health care, and ability to communicate with providers directly through the portal would improve care quality, experience, and outcomes for some patients who use the portal. We hypothesized that patients with complex conditions would receive greater benefit from using the portal. The study findings will help adult patients with chronic conditions and other stakeholders understand how using a portal affects the patients' health care access, experience, and health outcomes. This may influence a relevant patient's decision whether to use a portal. The findings will also help inform health care organizations that are considering the implementation of a patient portal as a tool to facilitate management of patients with chronic conditions.
At the study outset, we built our Patient Partner Panel; this group collaborated with us throughout the research project. We recruited members via chronic conditions classes offered through the delivery system's health education department, a pre-existing delivery system online patient feedback panel, and word-of-mouth. We selected panelists primarily based on demonstrated interest in and commitment to contributing to the research as collaborators and advisors. We also selected panelists with the aim of constructing a diverse group with respect to age, gender, ethnicity, chronic condition diagnosis, and history of portal use.
The study Patient Partner Panel of 7 to 10 members met roughly biannually during the study period, and panelists were compensated $100 per meeting for their participation. We were pleased with the strong continuity in engagement over the course of the study, with several panelists joining all the meetings for the entire length of the study. However, we did recruit additional members when others were unable to join. The panelists helped select the study name: CONNECT (Caring for Chronic Conditions Through Interactive e-Healthcare Tools).
The patient engagement coordinator, who had experience leading participatory research activities, facilitated each meeting. Throughout the 3 years of the study, the meetings followed a consistent structure and upheld a philosophy that fostered a degree of familiarity as well as offered new content and tasks according to the study progress. At the start of each meeting, we allowed time for our participants to eat dinner and get reacquainted, which helped establish rapport and build relationships. Each meeting was then opened with introductions of the study team, the patient panelists, PCORI, and Kaiser Permanente Division of Research, and a review of the main research questions and study design. During the initial meeting the panelists devised a list of “Group Agreements”—shared expectations related to code of conduct during panel meetings. The list was then reviewed at each subsequent meeting. The remainder of the agendas typically had a 3-part structure that included (1) sharing study updates and, once available, the preliminary results; (2) research education; and (3) application of research concept – panelists engaged in activities that positioned them as experts.
Throughout the study, collaboration with the panel helped us edit and refine the data collection survey tool to be more credible and specific in capturing the breadth of the patient experience with the health portal, allowed us to better frame study results from the patient perspective, and generated some ideas for potential dissemination plans.
We recruited a group of 7 delivery system leaders and clinicians to the Clinician and Delivery System Stakeholder Advisory Group. We recruited participants for their experiences using and building the patient portal. The panel met yearly during the study and meeting structure generally included an overview of the study and an update on study progress to date. This group has acted as a sounding board, assisting us with learning about the operational and clinical context of our study and helping us position the research to be relevant to delivery system stakeholders.
We constructed the study design and detailed analysis plan a priori and to align with PCORI Methodology Standards.
We conducted our study within the patient population of Kaiser Permanente Northern California (KPNC), an integrated delivery system (IDS) providing comprehensive care for more than 4 million patient members, reflecting the general population in the geographic region. The IDS implemented a comprehensive web-based patient portal in 2005. All delivery system members can register (create an account) and have free access to personal medical information and interactive tools to manage their health care, including viewing laboratory results and visit summaries, securely emailing with health care providers, requesting medication refills, and scheduling appointments. Starting in 2011, members also had access to a mobile-enabled version of the portal and a downloadable mobile application.
Our study methods include 2 complementary primary approaches: (1) a patient survey about the patient experience (Appendix A); and (2) a retrospective analysis of automated, data-capturing, portal use patterns and health care data. Within these 2 types of studies, we describe details about study populations, eligibility (including dates), and analysis approaches below.
Throughout the study, to structure our analyses, we used a causal model about the patient factors associated with using the portal. We hypothesized that many patient characteristics (eg, age, sex, race) would act as baseline variables that would directly influence initial portal use; these would also, independently of portal use, affect a patient's health status, health engagement, internet access, etc, which are mediators along other pathways between the baseline factors and portal use. In the analysis of automated data, we also use precise time-varying longitudinal clinical measures to capture short-term changes in clinical need preceding technology use. In this observational study, we also chose rigorous statistical approaches for each specific analysis, to handle bias and confounding from differences in patient characteristics between portal users and nonusers. For example, to address potential bias from clinical needs that may be associated with both portal use and outcomes of interest, such as emergency visits or hospitalizations, we accounted for time-varying short-term clinical needs by using marginal structural models. The specific analyses describe represent key research questions examined (space limitations prevent descriptions of all subanalyses in this report).
In both survey and automated substudies we aimed to include diverse patient populations (by age, race/ethnicity, chronic conditions, etc) as they were available in the study setting. We defined statistical significance as P < .05 throughout our study. The study activities were approved by the Institutional Review Board of the Kaiser Foundation Research Institute.
In our initial meeting with the Patient Partner Panel and the Clinician and Delivery System Advisory Groups, we collected feedback about measures that represent outcomes of importance to these populations of interest. To develop the study survey instrument that captured these outcomes, the study team selected questions from previously published and validated items, where possible, and developed new items for many domains, when needed. We compiled a broad set of relevant survey items targeting the domains and important outcomes that we identified in initial discussions with both the Patient Partner Panel and the Clinician and Delivery System Advisory Group. We invited patient partners to offer feedback on initial survey drafts, keeping in mind their own experiences or possibly those of a friend or family member. We also asked the patient partners to discuss the survey draft and share any concerns and suggestions. Their feedback brought to light several questions that were unclear, highlighting the difficulty of designing survey questions that can apply generally to many different chronic conditions. The patient partners also discussed the ways that the survey did and did not adequately capture their own experiences using the patient portal. In response to this feedback, we modified survey questions to better measure the patient experience in more nuanced ways and included many opportunities for open-ended descriptions of patient experiences.
Engagement with both panels helped greatly improve the clarity, completeness, and user- friendliness of the study questionnaire; it also helped shape the data collection tool to be more credible and specific in capturing the breadth of the patient experience with the patient portal.
In the study survey, respondents were asked to describe their access to the internet (frequency and devices used) and whether they use the patient portal. Respondents who were not registered portal users were also asked to indicate which factors were important in their decision not to use the portal. Survey respondents reported if they had also accessed the portal for another person (acting as a “care partner”), their family relationship, and credentials used to log in (either formally, by using care partner credentials, or informally, by using the patient's credentials). Participants also reported their experiences acting as a care partner through the portal in regard to convenience, ease of organizing health information, and timeliness.
All survey respondents were then asked to report the first way they generally choose to contact the health system for a nonemergency question or concern. All survey respondents were also asked questions about the degree to which they are engaged in their health care; this included a question adapted from the Chat (Children's Asthma Treatment) survey about preferences for shared decision making in health care decisions, a question adapted from the Single Item Literacy Screener, which is designed to identify limited reading ability, and a question from the Patient Activation Measure, which is used to assess “an individual's knowledge, skill, and confidence” for managing one's health and health care.38,39
In addition, patients who use the patient portal were asked questions specifically designed for this study, including device used to access the patient portal, the usefulness of each tool on the patient portal, reasons for using the patient portal, barriers or concerns about using the patient portal, whether the patient portal has been helpful in managing a variety of health conditions (including blood pressure, cholesterol, and blood sugar), how using the patient portal has affected their use of other health care services, and how using the portal has affected their overall health. We grouped patient-reported reasons and benefits of using the portal under 3 general domains: convenience (faster, helps patient to miss less work/activities, costs less); data and information usefulness (access to information, organization of information, understanding conditions, improving decision making); and integration with other care channels (relationship with provider, overall quality of care, preparing for in-person visits, finding other services or classes, following through with treatments at home). Patients also reported barriers or concerns about portal use, including preferences for in-person care, concerns about privacy, uncertainty about what is available on the portal, concerns about adequate response electronically, and portal usability.
Furthermore, we collected the following patient characteristics: race/ethnicity, education, income, marital status, self-reported health status, and affiliation with a regular provider.
To extract the survey target population, we used the EHR and other automated databases to identify a stratified random sample from adult (aged 18 years and up) IDS members who were included in the health systems clinical chronic condition registries for asthma, diabetes, congestive heart failure, coronary artery disease, hypertension, and cardiovascular event risk (including diabetes, coronary artery disease, abdominal aortic aneurysm, and peripheral vascular disease) as of March 2015. Separate from this study, these registries are regularly and automatically generated for clinical care and outreach based on patient diagnoses and EHR; they are also used for quality reporting. We oversampled patients with multiple chronic conditions. We studied patients with chronic conditions to increase the likelihood that the patient would actively use the health care system during the study period. Our target sample for the survey, as originally funded, was 1500 total surveys.
Beginning in March 2015, we mailed a study introduction letter, survey, reply postcard, and return envelope to all potential participants. Recipients could decline participation via a postcard or telephone or complete the survey and return it by mail. To make the survey convenient and accessible to as many respondents as possible, regardless of their technology access or preferences, we gave participants the option to complete a mailed survey, a telephone interview with a research assistant, or a web-based survey. As a token of appreciation for participation, we mailed a $20 gift card to those who completed the survey.
Between March 2015 and April 2016, research assistants contacted nonresponders to obtained verbal consent and completed the interview by phone. To complete and/or clarify any missing items or conflicting responses, research assistants also called respondents who had mailed back the written survey. The research assistants attempted to reach potential participants up to 15 times at different times of the day, on weekdays, and on weekends. Near the end of the data collection period we sent another survey by mail to any remaining nonresponders.
Respondents were ineligible for the study if they could not complete the interview in English (materials and data collection were offered only in English due to project budget constraints), or if they could not be reached because of incorrect contact information or their mailed survey packet was sent back by the postal office. Of the 3000 potential respondents contacted, we were unable to complete the survey for 413 after 15 phone call attempts; 380 were ineligible for study participation (a language barrier or health problem prevented them from completing an English-language interview or mailed survey, or they could not be reached due to incorrect contact information); 1824 patients completed the study questionnaire (the response rate among eligible participants was 70%); and 383 declined to participate (Figure 1). Among all respondents, 20% completed the survey by telephone, 57% returned the survey by mail, and 23% completed the internet-based survey. Respondents and nonrespondents were similar in gender (52% female among respondents, 50% female among nonrespondents; P > .05), but different in terms of age composition (median age: 64 for respondents vs 69 for nonrespondents; P < .05).
Flow Diagram Illustrating Steps From Initial Patient Survey Sample to Responders Included in the Analysis.
We conducted bivariate analyses of patient characteristics and portal registration using chi-square tests. We then examined the association between patient characteristics and portal registration using multivariate logistic regression. We included all patient characteristics in a multivariate logistic regression model to assess the impact of portal use on patient health care experience. If applicable, we created a category for missing data in each variable. In sensitivity analyses, we excluded the observations with missing data and found that the results were comparable.
The project also included a retrospective observational study of automated data captured from portal use data, the EHR, and other delivery system automated databases. By using these data generated and used directly in clinical care, we assume that they completely capture all health care and outcomes within this setting. We report 3 primary analyses below, each with a detailed population and statistical approach tailored to the unique data and outcomes required. We implemented all analyses using STATA13 (StataCorp; 2013).Our aim throughout the automated data analyses was to be as thorough as possible in including potential variables associated with patient portal use, including patient characteristics, and clinical need. We developed an extensive set of time-varying automated data measures.
We used existing automated databases available within the health system to capture patient use of the portal tools, including the dates for which patients created an account to use the portal; the timing of each login; and the tools such as scheduling appointments, laboratory result review, prescription drug refill, and secure messaging through the mobile-optimized website or mobile app. In this study, our main outcome is first portal use in 2013. We explored different definitions for portal use in the sensitivity analyses.
We collected patient characteristics from the health system database and classified them into 4 domains: (1) patient demographics, (2) potential barriers to portal use, (3) potential drivers of portal use, and (4) experience with portal.
Using automated data, we designed an analysis to examine which patients use the portal and when. To focus our study on patients with at least some baseline need to interact with the health care system and portal, we examined patients with a chronic condition. To select a large population of patients with chronic conditions and multiple chronic conditions, our study population included all patients in the IDS clinical chronic disease registries for cardiovascular event risk (including diabetes, coronary artery disease, abdominal aortic aneurysm, peripheral vascular disease), including patients with multiple chronic conditions. These disease registries are designed to comply with quality reporting requirements (Centers for Medicare & Medicaid Services, Healthcare Effectiveness Data and Information Set, etc) and are used regularly in clinical practice for patient outreach and treatment. We required patients to be continuously enrolled in the health system in both 2012 and 2013 for complete capture of clinical information. We excluded patients who were <18 years of age from our study because these patients likely manage their health conditions and portal use with the help of a parent. We also excluded patients without a valid home address because they could not be geocoded to extract neighborhood SES or internet access (4%). We included 273 815 eligible patients in this analysis.
To allow for time-varying (monthly) explanatory variables that reflected patient status due to health and utilization changes, we used pooled logistic regression to examine the factors associated with first portal use in 2013 (January-December). This is equivalent to time-dependent covariate Cox regression analysis.41 For each patient, we created monthly observations for each month until the first use of the patient portal, or to the end of study period for those who did not use the portal in 2013. We calculated the number of events in the prior month for each patient-month observation to create a measure of recent clinical need. For a patient-month with no portal use, we calculated the recent clinical need measure using the number of events in the prior month; for a month with portal use that fell on a on a specific date, we used the specific 30-day period up to this first use date to define the recent clinical need measure. We included all variables of patient demographics, potential barriers, potential drivers, and experience with portal in the multivariable model. We adjusted standard errors by clustering at patient level.
Because portal use can be defined in different ways and characteristics associated with portal use might differ among patients who have already registered, in sensitivity analyses, we varied the outcome definition (first portal use in 2013) by examining patient portal registration (created an account before 2013) and any use of the patient portal in 2013. We used multivariate logistic regression, with patients as the unit of analysis, for these analyses. But this analysis cannot handle time-varying predictors, and therefore we did not include the time-varying recent clinical events variables. We also repeated all analyses among those who already created an account before 2013.
Over the course of this study, we found that patient partners were eager to share personal examples of instances when they had used the portal to view their own laboratory results, then had communicated with their providers about any potential elevated results, and then had participated in treatment decisions. To examine 1 example of this through the automated data available within our study patient population and data, we studied patients with diabetes who checked a common laboratory result (cholesterol) on the portal and how it changed the way they communicated with their doctor and then changed their treatment. We selected this treatment pathway and population because it was relevant to a large chronic condition population of relevance, with a relatively clear guideline-recommended treatment intensification pathway during the study period. We examined this example as a case study of many parallel experiences of patients accessing their own information and using it to shape their health care.
To study specific treatment pathways and how they might be influenced by portal use, we examined patient use of the portal to check an elevated cholesterol laboratory result and subsequent communication with providers and potential treatment intensifications.
We examined 139 331 adult patients (aged 18 or up) with diabetes who received an LDL cholesterol test result in 2013. For patients with more than 1 LDL cholesterol value received within the year, we examined the first value in the calendar year. According to clinical guidelines in 2013, to examine the subpopulation of patients whom we hypothesized a priori would be more likely to receive a guideline-recommended treatment intensification outcome, we stratified patients according to LDL cholesterol level into 3 groups (<100 mg/dL, 100-129 mg/dL, ≥130 mg/dL). Patients whose value was <100 mg/dL were considered to be under control and were not generally recommended for treatment intensification; however, patients with values of 130 or higher were recommended for treatment with a cholesterol-lowering medication or an intensification of this medication if they were already taking one (based on 2013 recommendations from the American Diabetes Association).
The primary outcomes in our study were evidence of a patient–provider communication and any cholesterol treatment intensification encounters after a cholesterol laboratory result. We measured communication by automated record and time stamp for any secure messaging via the portal, a phone visit, or an in-person office visit; however, we were unable to identify which communication events directly included a conversation about cholesterol test results or treatments. For all patients, we examined all follow-up encounters with their health care provider/primary care provider within 30 days after the LDL cholesterol value was either released (among those who did not view the result on the portal) or viewed through the portal (identified through portal automated records and time stamp). To define treatment intensifications, we compared the dose of lipid-lowering medications dispensed within 180 days before the laboratory test result was released (for patients who did not view the cholesterol test result through the portal) with the dose dispensed within 30 days after the test result was released (for patients who did view the results through the portal).
Using marginal structural models (MSMs), we assessed the treatment effect (viewing the laboratory results within 7 days of laboratory release through the patient portal) on follow-up care and treatment intensifications; we also implemented doubly robust estimation with a multivariate logistic regression (outcome model) weighted by inverse probability of treatment.42
The covariates in both outcome and treatment intensification model include patient age, gender, race/ethnicity, neighborhood SES, adherence to chronic disease medications in prior year (proxy for patient engagement), number of chronic diseases in prior year, neighborhood internet access, previous LDL cholesterol level, and medical center. As sensitivity analyses, we used standard logistic regression adjusted for covariates. We reported the adjusted percentages of follow-up care and treatment intensification as if everyone in the population had viewed the result on the portal and as if no one had viewed the result on the portal, and the average treatment effects (the difference between the adjusted percentages).
We designed an analysis to evaluate the impact of patient portal access on health events. In this analysis, we examined the time period in the first 2 years after portal implementation in 2006 and created monthly data for each patient from January 2006 to December 2007 or until the month of disenrollment or death, with information on patient baseline and time-varying covariates, patient portal registration status (0 in months before portal registration and 1 in and after the months of portal registration) and health events. We opted to constrain the patient population to reduce computer processing time in our data models, so we limited the analysis to patients with diabetes, including patients with multiple chronic conditions in addition to diabetes. We included 165 447 patients with up to 24 months of observation time (January 2006-December 2007) in this analysis.
In pilot patient surveys that informed the development of the original study proposal for this project, and in initial meetings with the Patient Partner Panel and the Clinician and Delivery System Advisory Groups, we understood that health care events—especially ED visits and hospital stays—were important outcomes to our population of interest. Using the clinical patient history data captured in the EHR, we examined outpatient office appointments, ED visits, and hospitalizations for ambulatory care sensitive conditions, including external claims submitted to KPNC, using administrative data. We calculated monthly counts of these events separately for each patient in the months before or after portal registration. In the months of registration, we counted the number of events after the registration in month, then divided the numbers by total days in the month after registration, and then multiplied by 30.
We calculated stabilized weights using the numerator and dominator predicted from the pooled registration. We employed pooled logistic regression to predict monthly probabilities of portal registration and censoring/attrition (disenrollment and death), using both baseline and time-varying covariates.42 Predictors in logistic regression for denominator include age, gender, race/ethnicity, neighborhood SES, prior year drug adherence, prior year health status (number of HCCs), neighborhood internet access level, use (office appointment, phone call, ED visit, hospitalization) in prior 30 days, use in the prior 2 to 6 months, and utilization in calendar month. Predictors in logistic regression for numerator include calendar month. We then used marginal structure models with inverse weighting (stabilized weights) estimation to evaluate the impact of patient portal access on health events. We repeated the analyses in 2 subgroups stratified by chronic condition complexity: patients with diabetes only and patients with diabetes and asthma, coronary artery disease, congestive heart failure, or hypertension. We examined the impact in these 2 subgroups because of an a priori hypothesis that patients with more complex conditions may potentially benefit more from any portal use. In sensitivity analyses, we calculated weights through Super Learner43 (a data-adaptive estimation approach), based on cross-validation and predictors defined by logistic regression and polychotomous regression.
A total of 1824 respondents completed the study survey (70% response rate among 2620 eligible patients) (Table 1).
Survey Respondent Characteristics.
In survey findings, 91% of people aged 18 to 44 years used the portal, compared with 84% of people 45 to 64 years old, 78% of people 65 to 74 years old, and 53% of people ≥75 years (P < .05). People who identified as White were more likely to use the portal compared with other groups (81% vs 66% of people identifying as Black; 72% of people identifying as Hispanic; and 78% of people identifying as Asian; P < .05). People who reported annual household income of $40 000 or more (88% vs lower: 61%, P < .05), or who reported at least some college education (85% vs high school or less: 62%, P < .05) were statistically significantly more likely to use the patient portal. Respondents who used the internet regularly (94% vs not regularly: 45%, P < .05), or with access to both a computer and a mobile device (96% vs computer only: 86%; mobile only: 77%, P < .05) were also statistically significantly more likely to use the patient portal. Even though the differences did not reach statistical significance, patients with higher engagement, literacy, or more chronic conditions showed higher rates of portal use.
Among users, 21% had also acted as a care partner in using the portal on behalf of a friend or family in the previous year, with 45% formally using care partner credentials to log in and 55% informally using the patient's credentials on the patient's behalf. Among those who had acted as a care partner through the portal (on behalf of another person), 62% used the portal for a spouse, 35% for a child/grandchild, and 11% for a parent/grandparent (including multiple family members). Among those using the portal as a care partner, 89% reported that it was more convenient than other ways of participating as a care partner in another person's health care, 81% reported that it helped in organizing the other person's health care information, and 85% reported that it was faster than other ways of participating in another person's health care.
As shown in Figure 2, among the respondents who did not use the portal, the most common reason reported was “I prefer to get care in person or over the phone instead” (56%); 44% of nonusers reported that they do not regularly use a computer with an internet connection, and 17% reported that they were not sure what was available on the portal.
What Was Important in Your Decision Not to Register for the Portal?
We grouped the survey items related to the patient-reported portal experience into the following domains: convenience, data and information usefulness, integration with other care, and barriers and concerns. As shown in Figure 3, among patients who had used the portal, 90% reported 1 or more aspects of convenience, including 81% reporting it helped find answers to questions more quickly and 66% reporting that it helped them miss less school, work, or other activities.
Why Do Patients Use the Portal?
Overall, 92% reported 1 or more aspects of data and information usefulness, including 89% reporting that they used the portal to gain better access to their own health information and 76% reporting that it helped them better understand their health conditions. Overall, 92% reported that using the portal integrated with their other health care experiences, including 82% reporting that sending emails improved the patient–provider relationship and 67% reporting that the portal helped them prepare for in-person visits. Still, even among portal users, 47% reported 1 or more concerns or barriers to portal use, with 35% reporting a general preference to get care in person or over the phone instead of through the portal and 17% reporting concerns about privacy (Figure 4). Patients who were male (odds ratio [OR], 1.39; 95% CI, 1.06-1.84 vs female) or who had less internet use (OR, 2.80; 95% CI, 2.00-3.91 vs more), fewer chronic conditions (OR, 1.39; 95% CI, 1.02-1.88 vs more), or lower patient activation (OR, 1.82; 95% CI, 1.34-2.48 vs higher) were statistically significantly more likely to report a portal-related concern or barrier.
What Concerns/Barriers Do Patients Have About Using the Portal?
Among portal users, the survey asked patients if using the portal had an overall impact on their health. While 67% reported no change in overall health, 2% reported worsening health, and 31% reported that using the portal had improved their overall health. When examining patient- reported experiences pertaining to portal-related overall health improvement, we found that patients who reported convenience (36% vs no: 9%, P < .05), data and information usefulness (34% vs no: 7%, P < .05), or integration with other care (35% vs no: 6%, P < .05) were statistically significantly more likely to report that the portal improved their health (Figure 5). These are likely pathways to health improvement. Patients reporting concerns about using the portal use (16% vs no: 36%, P < .05) were less likely to report that using it improved their health—these concerns may act as barriers.
Are Patient-Reported Experiences Associated With Reported Health Improvements?
Among 273 815 eligible patients, 68% had created a patient portal account before 2013, and 65% used the patient portal in 2013. More than half used secure messaging (57%) or checked laboratory results (56%); about a third refilled a prescription (35%) or scheduled an appointment (30%). Among those who had already created an account before 2013, 87% used the patient portal in 2013. Table 2 shows patient characteristics among portal users and nonportal users as well as use rate, by patient characteristics.
Patient Characteristics and Rate of Patient Portal Use in 2013.
The multivariate analyses showing the association between patient portal use in 2013 and patient characteristics are presented in Table 3. Complementing survey findings, patients of younger age (OR, 1.15; 95% CI, 1.12-1.17 vs older) and female (OR, 1.04; 95% CI, 1.01-1.06 vs male) were more likely to use the patient portal; patients of non-White race/ethnicity (OR, 0.78; 95% CI, 0.75-0.82 for Black; OR, 0.83; 95% CI, 0.81-0.86 for Hispanic; OR, 0.87; 95% CI, 0.84-0.89 for Asian vs White) were less likely to use the patient portal. Patients with potential barriers— those whose preferred written language was not English (OR, 0.78; 95% CI, 0.74-0.82 vs English), who lived in a neighborhood with low-SES (OR, 0.87; 95% CI, 0.85-0.90 vs higher SES), and who lived in a neighborhood with lower levels of internet access (OR, 0.89; 95% CI, 0.85-0.93 for low; OR, 0.96; 95% CI, 0.94-0.98 vs higher)—were also less likely to use the portal.
Association Between First Portal Use in 2013 and Patient Characteristics.
Patients with potential drivers—those with more baseline clinical condition complexity (OR, 1.75; 95% CI, 1.72-1.79 vs less), more recent 30-day clinical events (OR, 1.73; 95% CI, 1.71-1.76 for 1 additional office visit; OR, 0.61; 95% CI, 1.58-1.65 for 1 additional phone visit; OR, 1.69; 95% CI, 1.61-1.78 for 1 additional ED visit; OR, 1.14; 95% CI, 1.07-1.22 for 1 additional hospitalization)—were more likely to use the patient portal. Patients who had to pay a copayment for in-person visits (OR, 1.10; 95% CI, 1.05-1.14 vs no) or had higher engagement (OR, 1.03; 95% CI, 1.01-1.05 vs lower) were also more likely to use the patient portal.
During this study, many patients were eager to share personal examples of instances when they had used the portal to view their own laboratory results, had communicated with their providers about any potential elevated results, and had participated in treatment decisions. For instance, we studied patients who checked a common laboratory result (cholesterol) on the portal and how it changed the way they communicated with their doctor and then changed their treatment.
Overall, 139 331 patients with diabetes received a cholesterol test result and were included in our study. In the 7 days after their LDL cholesterol values were released to be viewed, 46% of all patients had viewed it on the portal, with 36% viewing the result in the first 24 hours and 40.7% viewing within 2 days of release. By cholesterol level, 47% of patients with a lower cholesterol level (<129 mg/dL) and 42% of patients with an elevated cholesterol level (≥130 mg/dL) had viewed the result at 7 days. Figure 6 shows that rates of secure messages from patients to providers were higher among patients who had viewed their laboratory results through the portal (between 33.0 to 38.8 higher across the LDL cholesterol value strata; P < .05).
What Percentage of Patients Emailed Their Provider After New Cholesterol Laboratory Results?
Patients who viewed laboratory results were slightly more likely to have an office visit in the 7 days after their laboratory results were released, a modest but statistically significant difference (between 3.1% and 3.5% higher in patients who had viewed the laboratory results across the LDL cholesterol value strata; P < .05). Phone calls were also modestly statistically significantly lower if patients had viewed their laboratory results (between 1.7% and 3.5% lower in patients who had viewed the laboratory results across the LDL cholesterol value strata; P < .05). There was no significant difference in treatment intensification rates among patients with lower cholesterol levels. Among patients whose LDL cholesterol was elevated (≥130 mg/dL), however, there was a statistically significant increase in treatment intensification (1.6%; P < .05) among patients who viewed their laboratory results as compared with those who did not (Figure 7).
How Does Viewing Cholesterol Laboratory Results on the Portal Affect Treatment Intensification?
In our examination of the association between portal use and subsequent office visits, ED visit rates, and hospitalization rates among 165 447 eligible patients, we find 2 different patterns. Use of the portal was associated with significantly more office visits (170 per 1000 patients per month; P < .05; Figure 8), even after adjustment for patient characteristics and both recent and historical time-varying clinical needs.
How Does Using the Portal Affect Visits?
In contrast, portal use was associated with significantly fewer ED visits (3.5 per 1000 patients per month; P < .05) and preventive hospital stays (0.8 per 1000 patients per month; P < .05), as measured by ambulatory care sensitive hospitalizations (Figure 9). We interpret this pattern as a potential indication of increased health engagement and the potential to meet previously unmet clinical needs in the outpatient setting. Together, direct use of the portal and the related increases in doctors' office visits appear to reduce downstream disease exacerbations and clinical events that result in ED visits and preventable hospital stays.
How Does Using the Portal Affect ED Visits and Preventable Hospital Stays?
We also examined the impact of portal use by chronic condition complexity. We found portal use associated with significantly more office visits in both patients with only 1 chronic condition (diabetes only) and multiple chronic conditions, and fewer ED visits and preventable hospital stays among patients with multiple chronic conditions, but not among patients with only 1 condition.
Patient portals can offer patients electronic access to their own health information (eg, laboratory results and visit summaries), access to some health care transactions (eg, pharmacy refills and appointment scheduling), and communication with health care providers (eg, secure message exchange). For patients with chronic conditions, we examined the association between a comprehensive set of patient characteristics (both patient reported and automated) and patient portal use. We also examined the impacts of portal use on patient-reported health, treatment, visits, and health events. The study results confirmed our hypotheses that patient portal use varied across patients with different demographics, reported barriers, existing and recent clinical needs, and health engagement. We found that patients who used the portal to access their health data were more likely to communicate with providers and to receive recommended treatment linked to elevated laboratory values. After using the portal, patients also had more office visits, likely as a sign of increased engagement with health care providers, but were less likely to have ED visits or preventive hospitalizations, suggesting a reduction in downstream clinical events associated with portal use.
In this population, many acted as a care partner44 in accessing the patient portal on a family members' behalf, both formally and informally. Most reported that the portal represents a convenient and faster way to organize health information as a care partner, including across geographic distance. While these family roles have been discussed anecdotally, our study offers one of the first data-supported examples of patient-reported use of portal tools for family members. As portal use expands, usefulness for family care partners is also likely to grow, to aid in health care coordination, information use, and communication with providers.
As others have reported across portals used in other settings, differences in patient portal use by race and ethnic group45 remained even after adjusting for neighborhood SES, preferred language, internet access,46 and other factors.29,47 While personal preference for in-person communication with physicians and pharmacists and/or cultural factors could be the possible explanation,48,49 further study is needed to investigate the cause. Use of the portal by patients with more recent office visits, phone visits, ED visits, and hospitalizations may indicate a change in medical condition and clinical needs. To the best of our knowledge, our study is among the few to examine the association of closely linked time-varying clinical need and portal use.50,51 Consistent with other reports of increases in office visits before portal use, we found that more recent office visits, more recent phone visits, and more recent ED visits were all statistically significantly associated with portal use. Research that does not adequately account for this factor may mistakenly attribute any increase in use of in-person health care services to the use of a patient portal, rather than to the new health need.
Our results provide useful information on potential barriers and drivers of portal use and may help inform portal implementation strategies for other delivery settings that are thinking of implementing such a tool. For example, providing a non-English version of the portal may reduce the language barrier for non-English speakers. More recently, since the completion of our study, a Spanish version of the portal has become available in the health system. In addition, our findings highlight a list of possible statistical confounders to take into consideration in designing observational studies that assess the impact of patient portal use on health outcomes.
We found that patients who checked an elevated laboratory result through the portal were more likely to have some type of follow-up encounter or communication with their provider, primarily through secure messages also sent through the portal. Those who had an elevated cholesterol level were also more likely to receive a timely guideline-recommended treatment intensification within 30 days. Patients were able to access their own health care information via the portal to engage with their health care providers and to receive more timely treatment than if they had not used the portal. We found quantitative evidence to support our finding that patient portal use enhances patient–provider communication.
Early in the project, the Patient Partner Panel and the Clinician and Delivery System Advisory Group collaborated with the study research team to formulate the research questions, survey items, study protocols, and patient-reported outcomes. We presented survey drafts to both panel groups and then collected feedback as to whether the survey questions and response options adequately captured their experience using the portal. The feedback from both panel groups helped greatly improve the clarity, completeness, and user-friendliness of the study questionnaire. Over the course of the project, our expectations and model for engagement evolved, and we challenged our panelists to engage with us at an increasingly higher level of complexity. Our goal was to foster an exchange of knowledge and awareness that crossed professional and socioeconomic lines and to assist the panelists in developing a deeper understanding of the collaborative work we were doing together. The patient panelists were confident in voicing opinions, making suggestions or challenging the research process. The engagement piece of this study kept us true to the concept of patient-centered research. On several occasions our panelists responded to our findings by sharing a personal experience related to a data figure, which helped us reframe our interpretation of the data and allowed us to see the data through a different lens. In future collaborations we hope to continue to build on this experience and engage with the patient research partners as much as possible.
Our finding of portal use impacts on health care events and office visit use offers a detailed examination of the complex impact of portal use in patients with complex chronic conditions. Our findings of increased office visits, paired with decreased emergency visits and preventable hospitalizations, signal that the portal may be increasing engagement in the outpatient setting, potentially addressing otherwise unmet clinical needs and thereby reducing downstream health events that lead to emergency and hospital care. We find these impacts to be stronger in patients with complex conditions, indicating that portal impacts may be greater or more beneficial in patient groups with more clinical and health care complexity.
There are several limitations to the generalizability of our study findings. Because we conducted the study in a single integrated health delivery system with a well-established patient portal, the results may not necessarily generalize directly to other settings that are not part of a similar integrated setting. Still, this study setting includes both commercial and public insurance enrollees, represents approximately 33% of the underlying population in areas served, and is highly representative of the surrounding and statewide insured population in race/ethnicity and SES, with some underrepresentation of those with extremely low incomes.52,53 Also, our results are consistent with findings in previous studies of differences in patient portal utilization by age and race/ethnicity.29,45,47,54 Also, since the portal tools offered in this setting were available free of charge to patients, findings may vary when costs for secure messages are applied. Because our study was focused on patients with chronic conditions, the study is limited in its generalizability to populations without chronic conditions.
We are limited to findings from a cross-sectional survey of self-reported use and experiences and cannot establish causality. Also, because the survey was offered only in English, we cannot report on portal experiences in non-English speakers. Future studies should examine patient experiences across multiple languages. Because we found that our survey nonrespondents were slightly older than respondents, we cannot be sure that the findings fully represent the experiences of all patients in this population.
Since this an observational study, we cannot rule out unmeasured confounding or establish causality in the analyses based on automated data. While our study accounts for many potential confounders and time-varying patient and clinical variables, other patient characteristics may be associated with using or not using the portal that we could not measure or account for in our analyses. Other clinical predictor or outcome measures of diabetes progression, chronic disease severity, or underlying health status that may not have been captured in our analyses. Also, some of the neighborhood-level measures of internet access and SES cannot be linked directly to the individual patients in the study.
Patient portals are innovative patient-centered tools with the potential to improve health engagement and outcomes. Patient portal use varies by patient demographic, technology access, and clinical characteristics. Using the portal, however, was associated with shifts toward greater communication and office visits with providers, patient-reported health improvements, increased timeliness of treatment, and decreases in health events, with greater health benefits in patients with complex chronic conditions. Further research should examine the impact of targeted strategies to educate patients, providers, and health care delivery systems about the potential benefits of and barriers to portal use, including through dissemination of patient-reported experiences.
Research reported in this report was [partially] funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#IH-12-11-4925) Further information available at: https://www.pcori.org/research-results/2013/use-patient-portals-people-long-term-health-problems
Patient Survey (PDF, 563K)
Reed ME, Millman A, Fireman B, et al. (2019). Use of Patient Portals by People with Long-term Health Problems. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/6.2019.IH.12114925
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
