Author ORCID Identifier


Defense Date


Document Type


Degree Name

Doctor of Philosophy


Health Related Sciences

First Advisor

Jonathan DeShazo

Second Advisor

Laura McClelland

Third Advisor

Leonard Friedman

Fourth Advisor

Kevin Jackson

Fifth Advisor

Alexander Tartaglia



In the upcoming chapters, we present our study findings as three papers ready for submission to peer-reviewed journals. The first paper describes the associations between taxa and the characteristics of the patients and clinic staff who exchange those messages. The second paper explores the associations between those taxa and patients’ healthcare utilization. The third paper presents associations between taxa and patient health outcomes for diabetes and hypertension. We conclude with how the three papers are related and highlight the importance of this research.

Across the three papers, we reference a theory-based taxonomy we developed specifically for secure messaging. A number of researchers have created taxonomies to classify secure message content. Although these contained common themes, many were used only once or twice in published research and few classified content generated by clinic staff. We built our taxonomy upon commonly used themes from these existing classification systems. In contrast with other researchers, however, we leveraged theoretical constructs to group taxa and identify the concepts within messages that must be present for logical linkages between message content and patient outcomes. To identify why patients might outreach to clinicians during times of uncertainty, we referenced Mishel’s Uncertainty in Illness Theory (Mishel, 1988, 1999). We leveraged the framework developed by Street, Makoul, Arora, and Epstein (2009) to highlight patient task-oriented requests that might manifest in secure messages (e.g., to support self-care, satisfaction), and clinician-generated content that might support improved patient health outcomes. Our three papers present the first reports using this taxonomy and are the first to explore associations between taxa, patient outcomes, and the senders’ and receivers’ characteristics.

We sampled patients with diabetes and/or hypertension to demonstrate that our taxonomy could be applied to different health conditions, and to highlight any differences in taxa use based on health condition. We included threads initiated and completed between January 1 and December 31, 2017. Our study included 2111 patients, of whom 49 percent initiated 7346 threads that included 10163 patient-generated messages and 8146 messages generated by 674 unique clinic staff (hereafter referred to as clinician-generated messages).

Patient and Clinic Staff Characteristics Associated with Message Content

In the first paper, we described the coding process and interrater and intrarater reliability derived from that process, and then presented our findings on the characteristics of the senders and receivers associated with selected taxa. We estimated both unadjusted and adjusted differences in characteristics associated with the use of each taxon. We assessed taxon use as a dichotomous variable that was positive if the patient or clinician sent or received at least one message coded with the selected taxon. For patient-generated taxa, we explored associations with the characteristics of the sender (which types of patients sent these taxa) and receiver (which types of clinic staff received these types of content). Similarly, we explored the associations between clinician-generated taxa and the characteristics of the sender (what types of clinic staff sent these taxa) and receiver (what types of patients were the recipients of this content). We created separate regression models for patient characteristics (demographic, geographic, health condition and status, and thread volume) and staff characteristics (staff type, specialty, and message volume). Our patient-level analyses included only the 1031 patients who initiated message threads using the patient portal.

Our analyses found differences in taxa use by patients’ age, sex, race, health condition and status, and distance from clinic. Younger patients and females were less likely to share certain types of information with clinic staff (clinic updates among younger patients OR=0.77; 95% CI: 0.65-0.91; self-reporting biometrics by women OR=0.78; 95% CI: 0.62-0.98). Use of certain types of task-oriented requests varied by age (younger patients’ prescription refills OR=0.77; 95% CI: 0.65-0.90 and scheduling requests OR=1.41; 95% CI: 1.19-1.68) and race (black vs white requests for preventive care appointments OR=2.68; 95% CI: 1.30-5.51, requests for a new or changed prescription OR=0.72; 95% CI: 0.53-0.98, and laboratory or other diagnostic procedures OR=0.66; 95% CI: 0.46-0.95). Younger and uninsured patients were less likely to receive medical guidance from clinic staff (OR=0.84; 95% CI: 0.71-0.99 and OR=0.21; 95% CI: 0.06-0.72, respectively), but patients with public payers were two times more likely to receive medical guidance compared to patients with private payers (95% CI: 1.27-3.24). Females were less likely to receive confirmation that requests were fulfilled (OR=0.81; 95% CI: 0.68-0.97).

These findings highlight differences in how patients used secure messaging to communicate with their clinic staff, which could result in differential access to care. Further, the differences in taxa use by clinic staff by patients’ characteristics might further exacerbate existing disparities in care and highlight opportunities for training and education to reduce these discrepancies.

Healthcare Services Utilization Associated with Message Content

The Street, Makoul, et al. (2009) framework highlights access to care as an intermediate outcome in the pathway between health outcomes and communication functions such as information exchange, enabling self-care, and making decisions. Patients reported that effective communication delivered through secure messaging prevented unnecessary appointments (Alpert, Markham, Bjarnadottir, & Bylund, 2019); however, prior studies that explored links between secure messaging and healthcare utilization only considered message volume, not what was said in those messages. Our second paper is the first to explore whether content is associated with healthcare utilization. We measured utilization in four ways: number of outpatient visits, number of emergency department visits, number of inpatient visits, and medication adherence. We created separate medication adherence dichotomous variables for diabetes and hypertension, based on having an average condition-specific medication possession ratio greater than 0.8 (Clifford, Perez-Nieves, Skalicky, Reaney, & Coyne, 2014; Khunti, Seidu, Kunutsor, & Davies, 2017; Krass, Schieback, & Dhippayom, 2015; Schulz et al., 2016). We measured our independent variables as the taxon prevalence among patient- or clinician-generated taxa, as appropriate. Our covariates included the patient characteristics described in the first paper. To estimate incidence rate ratios for the three visit dependent variables, we conducted Poisson regressions with robust variance estimation (Hilbe, 2014). We estimated the odds of medication adherence associated with each taxon using logistic regression.

In unadjusted analyses, we found that patients who initiated message threads had higher numbers of outpatient visits (p<0.0001) and better hypertension medication adherence (p<0.01), compared to patients who did not initiate threads. Among patients who initiated message threads, we identified a positive association between emergency department visits and prevalence of request denials from clinic staff (IRR=1.18; 95% CI: 1.03, 1.35) and patients’ requests for follow-up appointments (IRR=1.15; 95% CI: 1.07-1.23), as well as between clinic non-response and the number of outpatient visits (IRR=1.02; 95% CI: 1.00, 1.03). We identified an inverse association between hypertension medication adherence and patients’ appointment reschedule requests (OR=0.87; 95% CI: 0.79-0.96). These findings highlight opportunities for future research about the use of secure messaging to influence care delivery and access to care.

Patient Health Outcomes Associated with Message Content

Patients whose uncertainty in their illness is addressed experience less stress, leading to better health outcomes (Mishel, 1988). Through appropriate communication functions with clinicians, patients develop better understanding of their condition and how to manage it and may have improved access to care and self-care skills, which leads to better outcomes (Street, Makoul, et al., 2009). Our third paper describes the first study to assess the types of message content associated with improved health outcomes. We examined changes in patients’ glycemic index (A1C) for patients with diabetes and changes in diastolic (DBP) and systolic blood pressure (SBP) among patients with hypertension, comparing patients who sent or received messages with selected taxa to (1) those who sent other types of messages and (2) those who did not initiate threads in 2017. We measured outcome changes as the difference between baseline (the last measured value in 2016) and endpoint (the first measured value reported in 2018) measures. Similar to the analyses conducted for Paper 2, our independent variables were the prevalence of each taxon by patient, where the denominator was the number of patient- or clinician-generated taxa, as appropriate for the selected taxon. Analyses included only patients with the selected condition: 811 patients with diabetes only, 787 patients with hypertension only, and 513 patients with both conditions. We used linear regression to identify associations between the outcomes and each taxon.

In unadjusted analyses, we found that patients who initiated threads had lower endpoint A1Cs (p=0.01) compared to patients who did not initiate threads. We observed improvements in A1C among patients who sent information seeking messages (b=-0.07; 95% CI: -0.13, -0.00). We also observed improved SBP associated with clinic non-response to patients’ threads (b=-0.30; 95% CI: -0.56, -0.04), staff acknowledgement and fulfillment of patients’ requests (b=-0.30; 95% CI: -0.58, -0.02), and patients’ complaints (b=-4.03; 95% CI: -7.94, -0.12). Poorer outcomes were associated with information sharing messages among patients with diabetes (b=0.08; 95% CI: 0.01, 0.15), and deferred information sharing by clinic staff among patients with hypertension (SBP b=1.29; 95% CI: 0.4-2.19). In addition, among patients with either condition, we observed positive associations between outcome and patient- and clinician-generated appreciation and praise messages with effect sizes ranging from 0.4 (A1C) to 5.69 (SBP). These findings demonstrate associations between outcomes and message content and further emphasize the need for training and education of clinic staff on appropriate use of secure messaging to prevent exacerbation of health disparities due to differential communication delivered through this modality.


We identified patient characteristics associated with patients’ use of taxa; not surprisingly, patients’ use of taxa varied by age, sex, and race. Taxa use varied by clinic staff characteristics consistent with the triage systems employed by most healthcare organizations (Heyworth et al., 2013; Ozkaynak et al., 2014; Wooldridge, Carayon, Hoonakker, Musa, & Bain, 2016). We also identified differences in staff’s taxa use based on the characteristics of the patient to whom they were sending the message. We further identified associations between taxa and healthcare utilization and health outcomes. If certain types of patients use taxa less frequently, and these taxa are associated with better outcomes or more appropriate utilization, then that presents opportunities to target those populations for education to shift their use of secure messaging. Further, if clinician-generated message content is associated with improved outcomes and clinic staff are not equitably sharing that content with all patients, there is an opportunity for education and training. Our research is a set of first-of-its-kind analyses that highlight differences in taxa use by both patients and clinicians and demonstrates the associations between those taxa and patient outcomes. Healthcare administrators and clinic staff should be aware of these associations and consider mitigation strategies to improve equitable secure messaging use by their staff and across their patient populations.

The studies shared several limitations discussed in more detail in the papers themselves. These limitations included a need for more specificity in the taxa definitions and more rigorous coding processes, the lack of temporal indicators in the analysis, and limited patient and clinical characteristics. The analyses that incorporated A1C measurements suffered from significant missing data. Sample size for some taxa was limited so that the algorithms did not converge. The analyses were based on single taxa, which represented only one component of the overall thread discussion. Finally, our message sample included only those messages saved to patients’ charts, which likely led to an underrepresentation of taxa and clinic non-response.

We highlighted a number of opportunities for future research across the three studies. Consideration should be given to refining taxa definitions and applying more rigorous coding practices, incorporating temporal elements into the analyses to provide context and support assessments of causality, adding relevant covariates such as message reading level or patients’ health literacy levels, and exploring other proximal and intermediate outcomes identified in the Street, Makoul, et al. (2009) framework. We also strongly recommend examining the impact of taxa pairings: analyses that consider the call-and-response nature of the full conversation occurring within the thread.


© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission



View graphic versionView graphic version