Mayo Clinic is a large multispecialty academic medical center focused on integrated patient care, education, and research. Mayo Clinic has three main medical sites in Minnesota, Florida, and Arizona and Mayo Clinic Health System (MCHS). MCHS is as a network of community-based medical services and consisted of more than 40 hospitals and clinics in Minnesota, Iowa, and Wisconsin in 2021. Mayo Clinic’s patient portal (Patient Online Services) has been operational since 2010 [28]. We collected over 2 million portal messages generated by patients from the Epic Clarity database between February 1 and August 31, 2020. We filtered the patient-generated messages (PGMs) associated with COVID-19 using relevant keywords (eg, “COVID-19,” “Pandemic,” “Coronavirus,” “SARS-CoV-2,” and “2019-nCoV”) and their synonyms and morphological variations (see Table S1 in the Multimedia Appendix 1). We excluded the PGMs with empty message bodies and the PGMs requested by providers such as messages for preappointment COVID-19 screening and postdischarge COVID-19 symptom checks. We then identified 207,299 portal messages on COVID-19 generated by 102,470 patients from the Epic Clarity database. In addition, the patient portal provided an anonymous COVID-19 self-checker for patients’ self-assessment of COVID-19 symptoms on March 22, 2020. We collected 153,224 PGMs on COVID-19 symptom self-assessment during March 22 to August 31, 2020. Thus, a total of 360,523 PGMs were used for sequential analysis.

No patients were exposed to any intervention. We used the data from the Mayo Clinic Unified Data Platform for analysis. The study was approved by the Mayo Clinic institutional review board (19-002211).

Table 1 lists the demographic distribution of patient populations from Epic Clarity database by age, gender, marriage, ethnicity, race, language, and residence. The results of chi-square goodness-of-fit tests among the three patient role groups (COVID-19 message senders, general message senders, and general patients) were listed in Table S2 in Multimedia Appendix 1.

We found that both COVID-19 and general message senders had a significantly different distribution compared to all patients active on the portal (P<.001). The distribution of the COVID-19 message senders was also significantly different from that of general message senders in terms of age, gender, marriage, race, and residence (P<.001).

More than half (>55%) of patients were in the age groups of 50-64 years and ≥65 years. The proportion of patients in the age ranges of 30-39 years, 40-49 years, and 50-59 years was observed to have increased when looking at general portal users to general message senders to COVID-19–specific message senders. Meanwhile, the proportion of message senders in the age groups of <18 years and ≥65 years was the lowest in the COVID-19 message sender cohort.

More than half (>54%) of the patients were female and were married or had a life partner. The proportion of female patients (61% vs 54%) and married patients (64% vs 56%) also was highest in the COVID-19 message sender cohort when compared to the entire active portal user cohort. More than 90% of patients were of non–Hispanic or Latino ethnicity, White race, and spoke English. This proportion was also highest in the COVID-19 message sender cohort and lowest in the general portal cohort.

At least 61% of all patients assessed lived in the urban area. The percentage of urban patients increased to 70% among general message senders and 75% among COVID-19 message senders.

We illustrated the daily numbers and WSAs of PGMs on COVID-19 in Figure 1. The WSA of PGMs on COVID-19 started to increase at the end of February and quickly peaked at 3303.43 messages per week on March 30. Sequentially, the WSA of PGMs on COVID-19 decreased to a local minimum of 1750.29 messages per week on June 5. The WSA of PGMs on COVID-19 reached a local maximum of 2624.57 messages per week on July 12. This July peak was 79.45% of the March peak’s volume. The WSA of PGMs on COVID-19 displayed a consistently declining trend over time after this point. Table 2 lists the numbers and proportions of PGMs for COVID-19–related care and other health care issues.

Figure 2 depicts the daily numbers and WSAs of PGMs for COVID-19 symptom assessment via the self-checker (Figure 2A), COVID-19 symptom assessment via e-visits (Figure 2B), discussing the COVID-19 diagnostic tests and results (Figure 2C), and the care plan (Figure 2D). The top message concepts for COVID-19–related care were COVID-19 symptom assessment via the self-checker (42.50%) and COVID-19 tests and results (30.84%). The percentage of PGMs on COVID-19 symptom assessment associated with e-visits or the COVID-19 care plan was approximately 1%. Owing to the low use of COVID-19 symptom assessment via e-visits, the patient portal stopped the e-visit service for COVID-19 symptom assessment on August 1, 2020.

Similar to the total PGMs on COVID-19 in Figure 1, the PGMs on COVID-19 symptom assessment via self-checker and COVID-19 tests and results had analogous dynamic patterns: two peaks in late March or early April and late June or early July. These fluctuations were consistent with the surge in newly confirmed COVID-19 cases in the United States and in Minnesota (see Figure S1 in Multimedia Appendix 1) [11]. The PGMs for COVID-19 symptom assessment via e-visits had two similar peaks and surges before the termination of the service. The trend of PGMs related to COVID-19 care plans paralleled trends in newly hospitalized cases and deaths in the United States and in Minnesota (see Figure S1 in the Multimedia Appendix 1) [11].

Table 2 shows that the top message usage for other health care issues caused by COVID-19 was related to appointment postponement (7.47%), anxiety (5.94%), and appointment cancellation (5.27%). Among the studied mental health issues, the number of relevant PGMs decreased as the severity of health issues increased: anxiety (5.94%), depression (1.02%), and suicidal ideation (0.08%). Figure 3 depicts the numbers of PGMs related to COVID-19, which mentioned general issues (Figure 3A), appointment postponement (Figure 3B), appointment cancellation (Figure 3C), anxiety (Figure 3D), depression (Figure 3E), and suicidal ideation (Figure 3F). The curves in Figure 3 show a similar trend over time: the number of PGMs on COVID-19 started to increase in early March and quickly peaked within 2-3 weeks. However, after March 30, the number of PGMs on COVID-19 constantly decreased, although smaller upward fluctuations occurred from July to August in some curves; for example, such as PGMs for postponement, anxiety, and depression. These fluctuations paralleled those observed in PGM use for COVID-19 care plans, as shown in Figure 2D.

The COVID-19 pandemic and subsequent public health mitigation strategies, including stay-at-home orders and business restrictions, substantially impacted delivery of health care services. As the COVID-19 pandemic progressed in the United States, and specifically in Minnesota, newly confirmed cases had two peaks during February 1 and August 31, 2020, owing to the initial outbreak and late termination of stay-at-home orders. We observed similar dynamic patterns in PGMs on COVID-19, particularly, COVID-19 diagnosis and treatment, suggesting that patients actively used the portal messaging for addressing their concerns regarding the COVID-19 crisis [23]. Another previous study [26] was consistent with our findings, as they analyzed messages in an ambulatory practice network and determined that their inbox message usage patterns were also consistent with national trends. Patients sent portal messages mainly for COVID-19 symptom self-assessment and discussing COVID-19 tests and results. It appeared that patients preferred symptom self-assessment to e-visits for symptom assessment given the utilization rates. Thus, analyzing PGMs on COVID-19 symptom assessment via self-checker before diagnostic testing could serve as a timely surveillance of COVID-19. Prior work by Denis et al [29] utilized this association when developing a self-assessment web-based app to assess trends of the COVID-19 pandemic in France. We also determined that the PGMs related to COVID-19 care plans followed trends in newly hospitalized cases and deaths. The second relative maximum of PGMs on the COVID-19 care plan after May suggests a decline of COVID-19 risk during that time period.

Our findings also indicated that patients used the portal to report feelings of anxiety and depression about their existing medical conditions and potential contagious risks due to COVID-19 and seek support from their providers. Similar mental health concerns increased in the general population, according to a study of Twitter data, which showed an increased in tweets expressing mental health concerns due to infection risk and isolation strategies in the early stages of the pandemic [30]. After examining 100 samples of related portal messages, we found that patients often reported worry about worsening of their current illnesses without their typical in-person follow-up as well as concern for COVID-19 infection risk during their visits to the clinics or hospitals. Some patients were also nervous about falling ill because they were not able to afford health care services owing to loss of jobs and health insurances [31]. Under such stressors, some patients reported depressive symptoms, and a few indicated suicidal thoughts and requested medical advice from their providers [32,33]. Although the number of PGMs on these issues rose substantially at the beginning of March, the relevant PGM count was progressively declining over time after April. The trends of PGMs related to the mental health concerns were consistent with those of PGMs related to COVID-19 care plans. We speculate the findings on the decline in overall mental health issues due to COVID-19 among patient portal users may result from a reduction in COVID-19 risk, eventual management of medical conditions, adaption to mitigation activities, or support from their caregivers. For example, a study on web-based search behavior for mental health concerns in the United States demonstrated a significant flattening of the curve for searches for anxiety and suicidal ideation after implementation of stay-at-home orders in certain states [34].

After analyzing PGMs related to COVID-19 and unique patient senders from the Epic Clarity system (see Figure S2 in Multimedia Appendix 1), we found a low messaging rate per patient and a strong correlation between message count and unique patient count, which suggests that the volume of unique patients mainly contribute to the intensive utilization of portal messaging. Demographic analysis of patient populations shows a significant difference in the distribution of patient populations between general message senders and general portal users. Compared to general portal users, the frequent message senders were more so middle-aged adults (30-64 years old), female, married (or with life partners), non–Hispanic or Latino, White, English speakers, or urban residents. This proportion was ever more pronounced in COVID-19 message senders. Middle-aged (or female, married, White, and urban) patients were more inclined to use patient portals for addressing their issues regarding COVID-19. This phenomenon is interesting given the fact that the COVID-19 pandemic disproportionately affected racial minorities and rural populations, who are particularly vulnerable to severe outcomes of COVID-19 [35,36]. This may be a consequence of the patient population of the institution or rather that patient portal has some inherent bias toward more health-literate patient populations, which may not be the same populations as negatively affected by the pandemic.

Telehealth, including the use of patient portals, is transforming the delivery of health care [2]. An unintended effect of the growth of patient portal messages may be an increased workload for providers. A previous study reported that providers occasionally needed to reply to messages sent by patients after work hours in order to ensure timely response. Newer delivery models will need to properly distribute the communication load for better efficiency and avoid provider burnout [26,37]. Some health care systems will probably face this challenge of managing increasing volumes of patient messages in the near future [38]. These health care systems will require new billing models and practice metrics, or additional ancillary infrastructure, including support staff, to accommodate this growing trend of asynchronous communication. Evolving technologies in artificial intelligence and natural language processing tools may even be considered as technological support for care teams in secure messaging [39,40]. In addition, the observed disparities in use of remote patient care among these populations warrants attention from providers and researchers on designing inclusive as well as innovative solutions to achieve equity in health care service delivery [41].

There are several limitations to our study. First, the patient portal messages were collected at Mayo Clinic, a multispecialty academic medical center. The collected data might not be representative of different clinical settings or patient populations in other areas of the country. Second, keyword searching was carried out to identify patient portal messages associated with COVID-19, COVID-19–related care, and other health care issues due to COVID-19. Although the keyword sets cover a large number of relevant keywords, synonyms, and morphological variations, they may not be totally comprehensive; hence, bias could exist in our results. We are developing robust detection algorithms based on state-of-art deep learning techniques to accurately identify interesting health topics in portal messages during the pandemic. Finally, we investigated patient portal messages in the early stage of the COVID-19 pandemic. The analysis of patient portal messages in the following time or later stages of the pandemic is beyond the scope of this study but represents an important area for further exploration.

During the COVID-19 pandemic, patient portal utilization increased to address questions and concerns about the COVID-19 pandemic, revolving mainly around symptom self-assessment, tests, and results. The increased usage statistics for COVID-19 indicates the patient portal was a valuable web-based platform for patients to remotely discuss COVID-19 diagnosis and treatment as well as seek support for other health care issues impacted by the pandemic. The volume of PGMs on COVID-19–related care fluctuated as the pandemic developed. After initial increase in March, the PGMs regarding other health care issues such as appointment cancellations and anxiety about disease progression exhibited a declining trend. We observed differences in patient demographics between general portal users, general message senders, and COVID-19–specific message senders, mainly that the majority demographic took on a larger proportion of COVID-19 messages. There is still great potential to increase PGM engagement for minority populations and rural communities with regard to the COVID-19 pandemic. Time-series analysis of portal messages could offer us a timely surveillance of COVID-19 and its impacts on patients to improve patient-centered care related to the COVID-19 crisis.