Primary Care Patient Portal Message Growth Statistics

Research question

What does published evidence actually establish about growth in patient portal messages, and which denominator should an independent primary care practice use?

The question matters because a headline count can collapse several kinds of work into one number. Independent practices have fewer layers between demand and the physician, so a vague denominator can lead to the wrong operational conclusion. This review asks what the cited evidence measures, what it leaves unresolved, and how a practice can build a truthful local comparison without presenting an external estimate as its own fact.

What the evidence shows

A JAMA Network Open study of 36.8 million messages reported a 157% increase in patient medical advice messages from the prepandemic to the pandemic period. The same study reported a smaller increase per patient, showing why total volume and patient-level intensity answer different questions.

A second multi-institution study found that message growth persisted beyond the initial shift to virtual care. It also showed variation across specialties and message types. Those findings support measurement by service line; they do not create one national message quota for every clinician.

Message counts can overstate or understate labor. One question may create several replies, while a complex request may be resolved in one long response after chart review. Thread counts, handling time, elapsed response time, and clinical escalation therefore belong in separate fields.

Taken together, the evidence supports a bounded conclusion: Portal workload rose sharply during the early pandemic period, but published studies measure different populations, message categories, and time windows. A practice should track threads, individual messages, unique patients, and clinician minutes separately. The finding does not prescribe staffing, promise time savings, or establish clinical policy. It identifies the units that should remain separate when an independent office studies its own workload.

Limits of the evidence

The sources use different designs. Surveys capture respondent experience and can cover large populations, but recall, response patterns, and question wording matter. EHR and transaction logs offer detailed timestamps but omit work outside the captured system and do not automatically distinguish active work from an open window. Observational studies see context but usually cover fewer sites and shorter periods. Policy documents define requirements or programs; they do not measure the effect in every practice.

Practice size is another limit. Results from academic centers and integrated systems may reflect centralized teams, mature analytics, and different patient mixes. National office-based data can include employed groups and multi-specialty organizations. None should be silently relabeled as a solo-practice average. The appropriate response is not to discard the studies, but to preserve their population when using them as context.

The publication date of a source and the date of its observations are also different. A recently published article may analyze earlier records, while an older study may remain useful for a narrowly defined mechanism. We prioritize authoritative and primary sources, state what they contribute, and avoid merging figures merely because they appear to concern the same broad topic.

How to interpret the evidence in an independent practice

Published evidence is a comparison set, not a performance standard for one office. Before comparing a local figure, confirm that the numerator, denominator, population, observation window, and exclusions match. A percentage calculated from encounters cannot be compared directly with a percentage calculated from patients, clinicians, workdays, or message threads. Means can also hide a long tail, so report a median and a high percentile when item-level data are available.

Use a representative baseline rather than a convenient week. Four weeks is often enough to expose weekday variation, but seasonal services may require a longer period. Record holidays, clinician leave, unusual closures, system outages, and campaign-driven volume. Those events do not invalidate the data; they explain it. Preserve raw counts beside rates so a changing denominator cannot create a misleading improvement.

Separate active handling time from elapsed time. A task may remain open while the practice waits for a patient, payer, pharmacy, laboratory, or outside clinician. Conversely, a short elapsed interval may conceal simultaneous work by several people. State whose time is measured and whether timestamps represent active use, clock time, or a manually observed interval.

Audit a sample of records to test the automated classification. System labels are created for workflow, not necessarily for research. Duplicate events, reopened items, corrected records, automated notifications, and actions performed outside the system can distort totals. Document exclusions before looking at the result and keep the same rules when the measure is repeated.

Administrative measurement must not erase clinical and privacy boundaries. A workflow category does not determine what may legally or safely be delegated. Diagnosis, interpretation, prescribing, triage, and medical advice remain with appropriately qualified professionals. Access should be role-based and limited to the minimum information required for the approved task.

Applying the finding without overstating it

Start with the research question, not a desired staffing or technology answer. Map the local process, identify each handoff, and choose one outcome plus a small set of balancing measures. Useful balancing measures include corrections, reopened work, clinician review, unresolved exceptions, patient follow-up, and safety events. Faster is not better if the change increases rework or loses clinical information.

When comparing periods, keep definitions stable. If a definition must change, show both calculations for an overlap period. Report absolute counts and rates, the observation window, and missing data. Label every locally calculated scenario as a calculation rather than a published benchmark.

The conclusion should remain proportionate to the design. A before-and-after operational audit can show association with a workflow change; without a comparison design it cannot establish that the change caused the outcome. National surveys provide context but do not predict a particular clinic. The most defensible use of external evidence is to improve local questions, definitions, and checks.

Methodology and conclusion

We reviewed the ten sources listed below, prioritizing peer-reviewed research and public material from federal agencies, AHRQ, ONC, CDC, CMS, HHS, and national evidence bodies. Sources were included when they clarified the workload, denominator, measurement method, or governance context. We did not average findings from unlike populations or clocks. Numeric examples in the key statistics retain the interpretation stated by their source; proposed local reporting fields are analytical guidance, not published national benchmarks.

The useful conclusion is narrow and actionable: Portal workload rose sharply during the early pandemic period, but published studies measure different populations, message categories, and time windows. A practice should track threads, individual messages, unique patients, and clinician minutes separately. An independent practice can use the cited research to define a better local audit, then report its own counts, rates, exceptions, and quality checks. That approach produces evidence fit for an operational decision without claiming that a national or health-system result automatically describes a solo physician's office.