Introduction
The adoption of electronic health record systems and web-based patient portals by hospitals has created new tools for patient-centered care with significant potential to improve clinical outcomes. Several studies have investigated the impact of patient portal use on patient-centered endpoints in various chronic diseases. In general, patient portal use was found to have a positive effect on metrics such as medication adherence and objective clinical outcomes.1–5 A recent systematic review by Han et al.1 evaluated patient portal interventions and their effects on clinical and psychobehavioral outcomes. They analyzed 24 studies, including randomized controlled trials, quasi-experimental studies, mixed methods studies using surveys and focus groups, and retrospective cohort studies. They found that patient portal interventions had a consistently positive effect on behavioral outcomes such as medication adherence and engagement in preventive health care screenings, such as colorectal cancer screening. However, the review also revealed mixed results regarding the impact of patient portal use on clinical outcomes and insufficient evidence of an effect on various clinical measures such as blood pressure or glucose control. These findings contrast with those from a systematic review performed by Alturkistani et al.,2 which reported overall improved glycemic control, reduced HgbA1c, and reduced hospitalizations in diabetic patients who actively use patient portals. Similarly, a review by Jeminiwa et al.3 demonstrated improved inhaled corticosteroid adherence in asthmatic patients who use patient portals versus those who do not. Comparatively, little research has examined the effect of patient portal use on outcomes in cancer patients.6
Furthermore, Han et al.1 noted that the studies in their review included primarily white, middle-aged, English-speaking populations and called for further investigation into the effects of patient portal use on outcomes among patients from more diverse backgrounds. The lack of diversity in these studies is consistent with previous findings of web-based portal underutilization by patients from minority groups and highlights a potential strategy for improving health outcomes in vulnerable populations.7,8 The goal of this study was to evaluate the impact of patient portal utilization on clinical outcomes in cancer care. Patient portal platforms provide patients with direct access to their providers through messaging, medication requests, and other tools. There is currently a knowledge gap in the literature regarding whether electronic patient portals enhance outcomes in cancer care. We hope to gain a better understanding of how the patient portal can be used to enhance care and improve patient outcomes. In this study, we evaluated the effect of patient portal use by patients with multiple myeloma on clinical outcomes and identified patient characteristics associated with active portal use. By understanding the web-based patient portal utilization characteristics of cancer patients and the effects of portal use on clinical outcomes, we can optimize web-based portal access and experience for patients to improve cancer outcomes, particularly for marginalized patients who often suffer from poor health outcomes.
Materials and methods
Sample and design
In this retrospective study, we analyzed data from 791 patients diagnosed with multiple myeloma, from the launch of the Scripps Health MyChart patient portal on April 1, 2017, through September 1, 2021. We compared the clinical outcomes of patients who utilized the MyChart portal to those who did not. Patient portal use was defined as patients who were actively enrolled in the patient portal platform. All patients with a diagnosis of multiple myeloma who were 18 years or older were included in the study. Data were collected from January 1, 2019, to January 1, 2021. Primary clinical outcomes were electronic health record (EHR)-documented unplanned hospital visits and all-cause mortality during the study period. Unplanned hospital visits were used as a surrogate for clinical outcomes, as other measures of disease control in myeloma (e.g., serum monoclonal protein, clonal plasma cells in bone marrow) cannot be easily extracted from the electronic medical record and would not be feasible for a study of this size. We used zip codes for regions defined by San Diego Health and Human Services to capture differences in socioeconomic status based on geographical location.
Statistical analyses
Comparisons between active portal users and inactive portal users were conducted using Chi-square tests or Fisher’s exact tests for categorical variables and t-tests or Mann-Whitney U tests for continuous variables. We analyzed differences in various demographic variables between active and inactive portal users. The differences between demographic variables and the primary outcomes of unplanned hospital admissions and death were analyzed using Chi-square tests and Mann-Whitney U tests. An exploratory analysis was performed using simple and multiple logistic regression to identify predictors of unplanned hospital visits. Univariable logistic regression assessed associations between individual patient and clinical variables and the outcomes, informing the variables to be used in multivariable logistic regression. Only variables with P-values less than 0.05 were considered in the multivariable models, along with any variables with a significant intercept. Additional exploratory analyses were conducted using Chi-square tests and logistic regressions to further investigate disparity variables. Two-tailed P-values less than 0.05 were considered statistically significant.
Many patients with multiple myeloma suffer from one or more comorbidities at the time of their cancer diagnosis; therefore, we included the age-adjusted Charlson comorbidity index (ACI) score as a variable in our analyses. The Charlson comorbidity index (CCI) is a validated tool that enables clinicians to predict the mortality of patients with multiple chronic conditions.9 The CCI consists of 17 comorbidities, with two subcategories for diabetes and liver disease.10 Comorbidities are weighted from one to six based on mortality risk and disease severity, and then summed to form the total CCI score.10 The ACI score incorporates age as an additional comorbidity by adding one point to the CCI score for each decade of age over 40 years.10 We evaluated the number of comorbidities and calculated the comorbidity burden using the ACI, categorizing it into scores of less than six and six or greater, as described by others.11,12 We assessed associations between various patient and clinical variables with low burden (ACI < 6) or high burden (ACI ≥ 6). If a variable was significantly associated with ACI burden (P < 0.05), it was analyzed in exploratory univariable and multivariable logistic regressions. Variables significant at the univariable level were added to the multivariable model, which included a significant intercept. We also estimated the predictive value of patient and clinical variables for a high ACI score as the outcome, as well as the predictive value of the ACI score for unplanned hospital visits as the outcome, in both univariable and multivariable analyses.
All analyses were conducted using R v. 4.0.3 in the R Studio environment. CCI scores were calculated using the comorbidity package in R.
Results
Sociodemographic characteristics of patients with active portal usage
Associations between portal use and sociodemographic variables are summarized in Table 1. Patients active on the portal were significantly younger (mean age of 71.8 ± 10.1 years) compared to inactive patients (mean age of 74.3 ± 11.7 years; P < 0.001). Males (53.02%) were also more likely to be inactive on the portal compared to females (P = 0.0434). Portal activity was higher among non-Hispanic patients (86.58%) than Hispanic patients (10.74%) (P < 0.001). The San Diego regions with the highest rates of portal utilization were the North Coastal region at 30.2%, followed by the North Inland region at 17.9%. These are more affluent regions of San Diego with the lowest levels of poverty, as shown in Figure 1. The regions with the lowest portal utilization were the more impoverished Central and Southern regions, with rates of 4.3% and 5.8%, respectively (Fig. 1). More active users lived in the North Coastal, North Inland, North Central, and Eastern regions, while more inactive users were present in the Southern and Central regions (P < 0.001). English-speaking patients utilized the portal at higher rates (88.81%) than Spanish-speaking patients (3.36%) (P < 0.001). Never smokers were more likely to be active portal users (P = 0.0104). Patients with private insurance utilized the patient portal at a higher rate (33.11%), whereas patients with Medicare had a higher rate of inactivity (66.89%) (P < 0.001).
Table 1Demographic characteristics by patient portal activity status
| Variable | Active (N = 447) | % or SD | Inactive (N = 344) | % or SD | P |
|---|
| Age (years), mean | 71.78 | 10.09 | 74.33 | 11.72 | 0.00060 |
| Sex | | | | | |
| Male | 237 | 53.02% | 208 | 60.47% | 0.03639 |
| Female | 210 | 46.98% | 136 | 39.53% | |
| Race | | | | | |
| White | 360 | 80.54% | 257 | 74.71% | 0.12670 |
| Asian | 24 | 5.37% | 22 | 6.40% | |
| Black | 21 | 4.40% | 29 | 8.43% | |
| Other/unknown | 42 | 9.39% | 36 | 10.47% | |
| Ethnicity | | | | | |
| Not Hispanic/Latino | 387 | 86.58% | 246 | 71.51% | <0.0001 |
| Hispanic/Latino | 48 | 10.74% | 81 | 23.55% | |
| Other/unknown | 12 | 2.68% | 17 | 4.94% | |
| Region | | | | | |
| Central | 26 | 5.82% | 40 | 11.63% | <0.0001 |
| East | 30 | 6.71% | 20 | 5.81% | |
| North Central | 75 | 16.78% | 55 | 15.99% | |
| North Coastal | 135 | 30.20% | 65 | 18.90% | |
| North Inland | 80 | 17.90% | 44 | 12.79% | |
| South | 19 | 4.25% | 52 | 15.12% | |
| Other | 82 | 18.34% | 68 | 19.77% | |
| Smoking status | | | | | |
| Never | 269 | 60.18% | 178 | 51.74% | 0.01041 |
| Former | 149 | 33.33% | 126 | 36.63% | |
| Current | 17 | 3.80% | 16 | 4.65% | |
| Unknown | 12 | 2.68% | 24 | 6.98% | |
| Language | | | | | |
| English | 397 | 88.81% | 279 | 81.10% | <0.0001 |
| Spanish | 15 | 3.36% | 48 | 13.95% | |
| Other | 7 | 1.57% | 6 | 1.74% | |
| Unknown | 28 | 6.26% | 11 | 3.20% | |
| Insurance type | | | | | |
| Private | 148 | 33.11% | 72 | 20.93% | 0.00020 |
| Medicare | 299 | 66.89% | 272 | 79.07% | |
| Currently insured | | | | | |
| Yes | 438 | 97.99% | 329 | 95.64% | 0.05641 |
| No | 9 | 2.01% | 15 | 4.36% | |
Clinical characteristics and outcomes for patients with active vs. inactive portal status
There were significant associations between portal use, ACI score, and the use of cancer-directed therapy, as well as the primary outcomes of unplanned hospital visits and death, as summarized in Table 2. Patients with a lower ACI score were more likely to be active portal users, whereas patients with a high ACI score were more likely to be inactive on the patient portal (P < 0.001,). The percentage of patients on cancer-directed therapy was higher among active portal users (53.67%) compared to inactive portal users (45.06%) (P < 0.001). A greater percentage of patients with unplanned hospital visits were inactive portal users, whereas a greater percentage of patients with no unplanned hospital visits were active portal users (P < 0.001).
Table 2Association of patient portal activity status with clinical characteristics and outcomes
| Outcome variable | Active (N = 447) | % | Inactive (N = 344) | % | P |
|---|
| Age-adjusted CCI score | | | | | |
| 0–5 | 160 | 35.79% | 72 | 20.93% | <0.0001 |
| 6+ | 287 | 64.21% | 272 | 79.07% | |
| Unplanned hospital visits | | | | | |
| Yes | 272 | 60.85% | 255 | 74.13% | 0.0001 |
| No | 175 | 39.15% | 89 | 25.87% | |
| Status | | | | | |
| Alive | 446 | 99.78% | 210 | 61.05% | <0.0001 |
| Dead | 1 | 0.22% | 134 | 38.95% | |
| Chemotherapy | | | | | |
| Yes | 256 | 53.67% | 155 | 45.06% | 0.0007 |
| No | 191 | 40.04% | 189 | 54.94% | |
Patient portal activity status and other predictors of unplanned hospital visits
Table 3 presents the findings from exploratory analyses of predictors for the primary outcome of unplanned hospital visits. Active portal use, older age, Hispanic ethnicity, current smoker status, living in the South region, Medicare, use of cancer-directed therapy, death, and high ACI burden were all individual predictors of having an unplanned hospital visit in univariable modeling. Notably, patients who were active on the patient portal had lower odds of having an unplanned hospital visit compared to those who were inactive (56.5% vs. 43.5%; odds ratio = 0.543; 95% confidence interval [0.398–0.736]; P < 0.001). In multivariable modeling, the use of cancer-directed therapy and an ACI score of six or greater remained predictors of unplanned hospital visits, while patient portal use was no longer significant (Table 3).
Table 3Exploratory logistic modeling of predictors of unplanned hospital visits
| Variable | n (%) | Univariable
| Multivariable
|
|---|
| OR | 95% CI | P | OR | 95% CI | P |
|---|
| Active status | | | | | | | |
| No | 344 (43.5) | REF | REF | REF | | | |
| Yes | 447 (56.5) | 0.5425 | 0.398–0.736 | <0.0001 | | | |
| Age (years), mean ± SD | 72.89 ± 10.89 | 1.035 | 1.021–1.050 | <0.0001 | | | |
| Sex | | | | | | | |
| Female | (346, 43.7%) | REF | REF | REF | | | |
| Male | (445, 56.3%) | 0.881 | 0.652–1.187 | 0.4050 | | | |
| Race | | | | | | | |
| Asian | (46, 5.8%) | REF | REF | REF | | | |
| Black | (50, 6.3%) | 0.667 | 0.294–1.542 | 0.3571 | | | |
| Other/unknown | (78, 9.9%) | 0.561 | 0.261–1.18 | 0.1323 | | | |
| White | (617, 78%) | 1.22 | 0.634–2.257 | 0.5415 | | | |
| Ethnicity | | | | | | | |
| Not Hispanic | (633, 80%) | REF | REF | REF | | | |
| Hispanic | (129, 16.3%) | 2.455 | 1.546–4.051 | 0.0002 | | | |
| Other/unknown | (29, 3.7%) | 0.203 | 0.083–0.449 | 0.0002 | | | |
| Smoking status | | | | | | | |
| Current | (33, 4.2%) | REF | REF | REF | | | |
| Former | (275, 34.8%) | 0.634 | 0.245–1.445 | 0.3061 | | | |
| Never | (447, 56.5%) | 0.523 | 0.205–1.170 | 0.1378 | | | |
| Unknown | (36, 4.6%) | 0.152 | 0.049–0.429 | 0.0006 | | | |
| Region | | | | | | | |
| Central | (66, 8.3%) | REF | REF | REF | | | |
| East | (50, 6.3%) | 0.667 | 0.300–1.474 | 0.3155 | | | |
| North Central | (130, 16.4%) | 0.759 | 0.388–1.443 | 0.4075 | | | |
| North Coastal | (200, 25.3%) | 0.797 | 0.422–1.459 | 0.4713 | | | |
| North Inland | (124, 15.7%) | 0.732 | 0.374–1.398 | 0.3524 | | | |
| Other | (150, 19%) | 0.417 | 0.218–0.772 | 0.0065 | | | |
| South | (71, 9%) | 2.953 | 1.218–7.730 | 0.0202 | | | |
| Insurance type | | | | | | | |
| Medicare | (571, 72.2%) | REF | REF | REF | | | |
| Private | (220, 27.8%) | 0.510 | 0.370–0.704 | <0.0001 | | | |
| Chemotherapy | | | | | | | |
| No | (380, 48%) | REF | REF | REF | REF | REF | REF |
| Yes | (411, 52%) | 2.867 | 2.113–3.911 | <0.0001 | 2.701 | 1.944–3.774 | <0.001 |
| Language | | | | | | | |
| English | (675, 85.3%) | REF | REF | REF | | | |
| Other | (13, 1.6%) | 1.788 | 0.541–8.031 | 0.3809 | | | |
| Spanish | (63, 8%) | 2.843 | 1.482–6.025 | 0.0032 | | | |
| Unknown | (39, 4.9%) | 0.858 | 0.446–1.702 | 0.6518 | | | |
| Death | | | | | | | |
| No | (656, 82.9%) | REF | REF | REF | REF | REF | REF |
| Yes | (135, 17.1%) | 7.090 | 3.829–14.677 | <0.0001 | 4.666 | 2.520–9.492 | <0.001 |
| ACI score | | | | | | | |
| 0–5 | (232, 29.3%) | REF | REF | REF | REF | REF | REF |
| 6+ | (559, 70.7%) | 4.639 | 3.353–6.449 | <0.001 | 3.905 | 2.776–5.524 | <0.001 |
We suspected that patients with more comorbidities would have more hospital visits, independent of their activity on the patient portal. An additional logistic regression analysis was conducted to control for the ACI score. Both active portal use and ACI score were found to be predictive of unplanned hospital visits. ACI was predictive in both univariable and multivariable models. Active portal users were less likely to have unplanned hospital visits, and this remained true when accounting for ACI.
Discussion
We showed that active patient portal utilization was associated with significantly lower rates of unplanned hospital visits and death in a diverse cohort of patients with multiple myeloma in San Diego County. Patient portal use remained an independent predictor of unplanned hospital visits in multivariable logistic regression analysis that included ACI. However, when additional significant variables were added to the model, cancer-directed therapy and ACI remained predictors of unplanned hospital visits, while patient portal use was no longer significant. Death was more likely in patients who had never enrolled in the patient portal. This finding is intriguing but warrants further investigation to better understand the association. We also observed that older patients were less likely to use the portal, which may correlate with a higher risk of death. Further prospective studies could elucidate outcomes in patients not utilizing the patient portal. While these findings do not prove a correlation between death and portal use, they support further research in this area. Studies measuring the impact of patient portal utilization in other diseases, such as diabetes, hypertension, and preventative care, have shown mixed results.1–5 The lower rates of unplanned hospital visits and deaths seen in our patient population may indicate a potential benefit of leveraging this technology to improve care for patients with multiple myeloma. Previous studies have assessed patient satisfaction, overall views toward the technology, improved communication, and other patient perceptions.6 However, no prior studies have evaluated the impact of patient portal utilization on outcomes in multiple myeloma or other cancers. There are several limitations to the retrospective design of this study, and while the results are promising, further investigation via prospective studies is warranted to explore the potential advantages of telehealth in reducing disparities in cancer care for vulnerable populations.
Patient portal utilization in our study cohort also highlights important healthcare disparities. Baseline differences in sociodemographic variables were observed; the groups with the lowest utilization of the online patient portal were older patients, Hispanic patients, Spanish speakers, Medicare recipients, smokers, and patients living in low-income areas, such as the U.S.-Mexico border community in Southern San Diego. Notably, elderly patients, Spanish-speaking patients, and those from lower-income areas had poorer clinical outcomes.
We also found that being inactive on the patient portal was an independent predictor of unplanned hospital visits. In multivariable analysis, portal activity status was predictive of outcomes independent of comorbidity burden.
The treatment of multiple myeloma and other malignancies can be complex, often involving multi-agent cancer-directed therapy with intravenous, subcutaneous, and oral medications. Communication between the patient and their hematology care team is crucial for ensuring compliance with treatment and follow-up. Our findings suggest that access to and interaction with patient portals may improve outcomes for patients with multiple myeloma, particularly in vulnerable populations.
Our study has several limitations. The grouping of patients into active and inactive users based on their MyChart patient chart status provided a limited understanding of portal use. We did not investigate specific within-group differences in portal use, such as the number of patient-to-provider e-messages, which may have offered insight into how portal use activity affects hospital visits. This approach also did not allow us to stratify the cohort by level of activity on the patient portal to investigate differences in hospital visits between patients who were regularly versus infrequently active on the portal compared to inactive users. The clinical outcome of unplanned hospital visits, while important, is limited in specificity. Understanding the reasons for each patient’s hospital visit would require a more thorough investigation beyond this study’s scope. Finally, although this retrospective study sheds light on the potential for patient portals to help address healthcare disparities in patients with multiple myeloma, further investigation through randomized controlled trials is needed to address potential confounding factors and provide a clearer understanding of how patient portal utilization affects clinical outcomes, given the inherent limitations of retrospective chart reviews.
Conclusions
Digital healthcare resources, such as online patient portals, are promising technologies that may reduce barriers to access and improve outcomes in vulnerable populations. While research into the impact of electronic health resources is growing, prior studies have not evaluated the effect of patient portal activity on health outcomes in cancer patients. This study provides evidence that culturally tailored programs to increase access to electronic resources in underserved populations are likely to help close the gap in patient outcomes in these communities.
Declarations
Ethical statement
This study was carried out in accordance with the recommendations of the Committee on Publication Ethics, the Declaration of Helsinki, and the recommendations for the conduct, reporting, editing, and publication of scholarly work from the International Committee of Medical Journal Editors. The study was approved by the Scripps Health IRB, and individual consent for this retrospective analysis was waived.
Data sharing statement
The data supporting the findings of this study will be made available by the corresponding author upon reasonable request after the publication of this manuscript.
Funding
This work was funded by a grant from the Research and Education Fund of Scripps Clinic Medical Group, La Jolla, CA [NIH/NCATS UL1TR002550].
Conflict of interest
The authors have no relevant conflicts to disclose.
Authors’ contributions
Conceptualization, methodology (EQ, AE), original draft preparation (EQ, AE, LP, SB, MX), draft review and editing (EQ, AG), formal analysis, data curation (LP, SB), and supervision (MX). All authors approved the final version of the manuscript for submission.