Mapping clinician navigation paths, click counts, and cognitive load across EHR workflow redesigns
Before any redesign can be justified, the existing workflow has to be measured honestly. A legacy EHR interface — built up over a decade of incremental feature bolt-ons — is instrumented click-by-click and screen-by-screen for a single, standard office-visit encounter, producing the baseline number every subsequent intervention is judged against.
A click-burden audit instruments a representative clinical task — most commonly a routine office visit — and records every discrete mouse click, keystroke shortcut, and screen transition required to complete it from chart opening to note sign-off. This is not an abstract metric: it is built from concrete navigation events such as opening the chart, reviewing the problem list, reconciling medications, documenting vitals, writing the History of Present Illness, placing orders, selecting billing codes, and closing the encounter.
Each of these steps in a legacy, un-redesigned EHR often requires its own screen, its own load time, and its own set of confirmation dialogs. A medication reconciliation step alone can require opening a separate module, scrolling a multi-page list, clicking to confirm or discontinue each entry individually, and clicking again to return to the main chart. Multiply that pattern across ten or more discrete workflow steps and the click count for what a patient experiences as a fifteen-minute visit climbs well past a hundred discrete interface interactions.
The audit methodology itself matters. Rigorous studies typically use screen-recording software combined with event-logging instrumentation embedded in the EHR itself, cross-validated against direct observation by a trained workflow analyst standing beside the clinician. This dual approach avoids both the underestimation that comes from self-report and the overestimation that comes from counting incidental clicks unrelated to the clinical task.
Independent audits have repeatedly found that primary care physicians generate more than 4,000 EHR clicks in a single ten-hour shift — a figure that has become a shorthand for the interface-friction problem in clinical informatics literature.
The National Institute of Standards and Technology's NISTIR 7804, "Technical Evaluation, Testing, and Validation of the Usability of Electronic Health Records," established one of the first formal frameworks for evaluating EHR usability against clinical workflow rather than against generic software usability standards. It defines a structured protocol: task analysis, heuristic evaluation against ten usability principles adapted for clinical software, and summative usability testing with representative clinician users performing representative tasks.
Critically, NISTIR 7804 treats click count and screen-transition count as proxy measures for a deeper construct — cognitive workload imposed by the interface — rather than as the end goal itself. A workflow with fewer clicks but poor information scent (where the clinician cannot predict what a given click will do) can still impose high cognitive burden. The audit stage therefore records not just raw counts but also navigation path shape: how often a clinician has to backtrack, how often the same information has to be re-entered across screens, and how often a screen transition interrupts an in-progress cognitive task like differential diagnosis formulation.
This baseline audit stage produces the "legacy path" — the longer, more circuitous navigation graph rendered in red in the simulation above — against which every later optimization stage in this simulator is measured.
A recurring failure mode in EHR usability improvement projects is redesigning based on stakeholder intuition about where the friction "must be" rather than on instrumented measurement of where it actually is. Clinicians reliably identify the most recently frustrating screen as the primary problem, which is a recency-biased sample of a much larger navigation graph.
Instrumented click audits instead surface the steps that consume disproportionate navigation cost relative to their clinical value:
• Redundant data entry — information already captured elsewhere in the chart (allergies, problem list, med list) that the interface forces the clinician to re-enter or re-confirm on a separate screen • Confirmation-dialog stacking — modal "are you sure?" prompts that exist for medico-legal or billing-compliance reasons but add a click with no clinical information content • Navigation dead-ends — screens that do not surface an obvious path back to the main chart, forcing the clinician to use browser-style back navigation or restart from the chart index • Context loss — screen transitions that discard scroll position or selected filters, forcing the clinician to re-locate information they had already found
Each of these categories has a distinct fix, and the audit stage is what tells the redesign team which categories dominate for a given specialty and a given legacy platform configuration.
A single-problem follow-up visit and a complex multi-problem visit are not the same workload multiplied by a constant factor. As active problems, medications, and required orders accumulate, the legacy interface forces the clinician back through chart-review, reconciliation, and coding loops for each additional item — producing click and time growth that outpaces the growth in clinical complexity itself.
In a well-designed workflow, adding a second active clinical problem to a visit should add roughly one additional documentation and ordering pass. In a legacy EHR with a rigid, module-based navigation structure, it frequently adds an entire repeated traversal of shared infrastructure: the clinician re-opens the problem list module, re-scrolls the medication reconciliation screen (now longer, because more medications are active), re-enters the order-entry module to place a second set of orders, and re-visits the coding screen to add a second diagnosis code and confirm the resulting level-of-service change.
This compounding effect means click count as a function of active problem count is closer to a step function with a steep slope than a flat linear increment. Empirical time-motion studies in ambulatory internal medicine and family practice settings have found that visits addressing three or more chronic conditions can consume more than 60% more total EHR interaction time than a comparable single-problem visit, even when total appointment duration is held constant — meaning the additional burden is absorbed as after-hours documentation, the phenomenon widely known in clinical circles as "pajama time."
Sinsky and colleagues (Annals of Internal Medicine, 2016) found that for every hour physicians spent in direct face-to-face contact with patients, they spent nearly two additional hours on EHR and desk work — a ratio that grows further, not linearly, as case complexity increases.
Two workflow steps are disproportionately sensitive to task complexity in legacy interfaces: medical coding and order entry.
• Coding: each additional diagnosis addressed in the visit typically requires its own code-selection screen interaction, and the resulting evaluation-and-management level frequently needs to be manually recalculated and re-confirmed rather than being derived automatically from the documented complexity — adding clicks that have nothing to do with clinical care and everything to do with legacy billing-logic separation from the clinical note.
• Order entry: legacy order-entry modules built around single-order workflows force the clinician to repeat the full order-placement sequence (select order type, select specific test or medication, select dosing or frequency, confirm, sign) for every discrete order in a complex visit, rather than allowing a single batched order-set selection that bundles the orders clinically associated with a given problem.
These two categories alone typically account for more than half of the incremental click growth observed between a simple and a complex encounter on an unoptimized platform, which is precisely why order-set design and coding-assist automation become the highest-leverage targets in the redesign stage that follows.
Measuring how click and time burden scale with complexity — rather than only measuring the average across all visit types — gives redesign teams a prioritization signal that a flat average obscures. A specialty with a high proportion of complex, multi-problem visits (primary care, geriatrics, endocrinology) will see disproportionate benefit from fixes targeted at the compounding steps described above, while a specialty dominated by simple, procedural, single-purpose visits may see more benefit from streamlining the baseline path itself.
This stage of the simulator lets the Task Complexity slider drive the length of the legacy navigation path directly: each complexity increment inserts additional problem-review, reconciliation, and order-entry hops into the rendered path, visually demonstrating the non-linear growth pattern described above rather than simply stating it as a number.
Reducing click burden is rarely about removing clinical steps — it is about collapsing redundant navigation, batching related actions, and pre-populating defaults so the clinician confirms rather than re-enters. Combined with newer ambient AI documentation assistance, these techniques can cut the legacy path by more than half without cutting a single clinically necessary decision point.
The single highest-leverage redesign intervention documented in EHR usability literature is the order set: a pre-built, clinically curated bundle of orders associated with a specific diagnosis, protocol, or visit type. Instead of placing five individual orders through five repetitions of the full order-entry sequence, the clinician selects one order set and confirms or deselects individual items within a single screen.
Structured documentation templates apply the same collapsing logic to note-writing. Rather than a blank free-text field requiring the clinician to type or dictate the entire History of Present Illness and Review of Systems from scratch, a structured template pre-populates the note skeleton with pertinent negatives and normal findings drawn from prior visit data, problem list, and standard-of-care checklists — leaving the clinician to confirm, modify, or add only what is clinically new.
The risk this technique introduces is well documented in the informatics literature as "note bloat": templates and copy-forward functionality can produce notes that are long, internally repetitive, and difficult for other clinicians to parse for what actually changed at this visit — even as click count for the authoring clinician goes down. A well-designed redesign therefore measures not just click count but note information density, ensuring click reduction is not achieved by simply auto-inserting boilerplate that shifts the burden onto whoever has to read the note later.
Smart order sets and structured templates are consistently the single highest-yield intervention in published EHR optimization case studies, often accounting for more than a third of total click reduction on their own.
A newer class of intervention, ambient clinical documentation assistants — voice-recognition and large-language-model systems that listen to the natural clinician-patient conversation and draft a structured note automatically — removes entire documentation screens from the navigation path rather than merely compressing them. Commercial systems in this category (frequently described in the literature by the shorthand "ambient scribe" or "DAX-style" assist, after early Dragon Ambient eXperience-class products) generate a draft History of Present Illness, assessment, and plan directly from the recorded encounter, which the clinician then reviews and signs rather than authors from a blank field.
Early published evaluations of ambient documentation assist report meaningful reductions in after-hours "pajama time" documentation and in self-reported cognitive burden, though the technology introduces its own new failure modes that a rigorous redesign process must audit for: transcription errors that propagate into the permanent medical record, over-generation of content not actually discussed in the visit, and a clinician tendency to under-review AI-drafted text before signing — a new source of documentation risk that trades click burden for review-vigilance burden.
The optimized navigation path rendered in green in this simulation reflects a composite of structured templates, order sets, and ambient assist collapsing the eleven-plus-screen legacy path down to roughly five to six screens: login, a unified smart chart view, a single structured note screen, a single order-set screen, and sign-off.
A quieter but equally important redesign principle is shifting the clinician's interaction model from data entry to data confirmation wherever clinically safe to do so. Smart defaults pre-select the most probable value (a standard dose, a standard follow-up interval, a normal-range vital sign) and require an explicit click only to change it, rather than requiring an explicit click to accept it.
This principle must be applied carefully — over-aggressive defaulting has caused documented patient-safety incidents where a default value was accepted without adequate clinician attention (a category of error sometimes called "alert fatigue" or "default fatigue" in the human-factors literature, closely related to the well-studied problem of alert-override rates in clinical decision support). Redesign teams following NISTIR 7804-style summative usability testing validate that defaults reduce click burden without measurably increasing override-without-review rates before shipping them broadly.
Click count matters clinically only insofar as it predicts something that actually harms clinicians and patients. This stage connects instrumented navigation-friction data to validated cognitive workload instruments and to the physician burnout literature, establishing that it is navigation friction — not documentation volume per se — that drives exhaustion.
Raw click count is an easy metric to instrument but an imperfect proxy for what actually matters to clinician wellbeing: cognitive workload. Human-factors researchers typically operationalize cognitive workload using validated instruments such as the NASA Task Load Index (NASA-TLX), adapted for clinical software contexts, which scores perceived mental demand, temporal demand, effort, and frustration across a task.
When click-burden audits are run alongside NASA-TLX-style workload surveys on the same clinicians performing the same tasks, the two measures correlate strongly — but not perfectly, and the gap between them is informative. A workflow with a moderate click count but poor information scent (screens that do not clearly signal what will happen next) or frequent context-switching between unrelated tasks (an inbox message interrupting an in-progress note, for example) can produce a cognitive load score disproportionately higher than click count alone would predict. This is why mature EHR optimization programs report both metrics side by side rather than optimizing for click count in isolation, which can produce a workflow that is technically shorter but subjectively more exhausting if the collapsed screens now cram unrelated decisions into a single dense view.
Surveys conducted by the American Medical Association and allied burnout researchers have repeatedly found that more than half of practicing physicians report at least one EHR-attributable symptom of burnout, with excessive documentation burden and poor system usability cited as leading contributing factors.
The American Medical Informatics Association (AMIA) has, over the past decade, elevated EHR usability and its relationship to clinician burnout into one of its core research priority areas, convening task forces and publishing position statements that explicitly link interface design quality to physician wellbeing outcomes, alongside more traditional patient-safety framing. This reframing matters organizationally: usability investment that might previously have been justified only on efficiency or error-reduction grounds can now be justified on clinician retention and burnout-mitigation grounds, which tend to carry more weight in health-system capital allocation decisions given the documented cost of physician turnover.
KLAS Research and related independent EHR usability rating organizations have similarly begun publishing vendor-comparative usability scores that hospital systems use in procurement decisions, creating a market incentive for EHR vendors to invest in the same click-reduction and cognitive-load-reduction techniques described in the previous stage, rather than treating usability purely as a customer-support cost center.
In this stage of the simulator, the cognitive load signal rendered on the canvas is derived from two inputs drawn directly from the navigation simulation: the frequency of screen-hop transitions per unit time (a proxy for temporal demand) and the total path length (a proxy for cumulative effort). As the UI Redesign Efficiency slider moves toward the optimized end, both inputs fall together, and the rendered cognitive-load gauge falls with them — visually reinforcing the finding that reducing navigation friction, not simply asking clinicians to "work more efficiently," is what moves the needle on the underlying workload construct that burnout research actually cares about.
This is the central argument for treating click-burden reduction as a clinical-quality and workforce-retention initiative rather than a cosmetic software-usability nice-to-have: the downstream outcome being optimized against is measurable clinician exhaustion, not merely a smaller number on an engineering dashboard.
A workflow optimization validated in one pilot clinic does not automatically hold across an entire health system. Sustained click-burden reduction at enterprise scale requires governance, staggered training, and continuous monitoring dashboards that catch configuration drift and workflow regressions before they quietly erode the gains measured in the pilot.
Rolling an optimized workflow out across an entire health system is a change-management problem as much as a technical one. A workflow redesign validated against a single pilot clinic's specialty mix, patient population, and existing local customizations cannot be assumed to transfer unchanged to a different department without local validation — a cardiology clinic's order-set needs differ substantially from a pediatric primary care clinic's, even on the same underlying EHR platform.
Mature rollout governance structures typically include a clinical informatics steering committee with representation from each major specialty group, a staged deployment schedule that sequences clinics by readiness and risk rather than deploying simultaneously system-wide, and a formal escalation path for clinicians who identify workflow gaps introduced by the new configuration that were not present in the pilot. Staged rollout also allows the informatics team to compare click-burden metrics across cohorts deployed at different times, isolating the effect of the redesign itself from secular trends like EHR vendor platform updates that would otherwise confound a simple before-after comparison at a single site.
The gains measured at go-live are not guaranteed to persist. Configuration drift — local super-users adding custom order-set variants, individual clinicians reverting to old habits available as alternate navigation paths, vendor platform updates silently altering default screen flows — can erode click-burden reductions gradually enough that no single event triggers investigation, while the aggregate metric quietly climbs back toward baseline over eighteen to twenty-four months.
Sustained programs address this with standing monitoring dashboards that track click count, time-per-encounter, and documentation-burden metrics on a rolling basis per clinic and per specialty, with automated alerting when a site's metrics drift outside an expected control-chart band relative to its post-rollout baseline. This is directly analogous to statistical process control in manufacturing quality assurance, applied to clinical workflow telemetry instead of production-line defect rates — the goal is catching regression early enough to intervene with targeted re-training or configuration correction before the site's clinicians experience it as "the system getting worse again."
Health systems that pair workflow optimization rollout with a standing monitoring dashboard report meaningfully better sustained click-burden reduction at the twelve-month mark than systems that treat optimization as a one-time project with no ongoing measurement.
The most durable enterprise rollouts treat EHR usability optimization as a standing program with a permanent team and budget line, not a single funded project with a defined end date. This program typically owns three ongoing functions: continuous click-burden and cognitive-load measurement across all clinics; a backlog of prioritized workflow friction points sourced from both instrumented telemetry and direct clinician feedback channels; and a change-control process that vets every proposed EHR configuration change against its projected click-burden impact before it ships, preventing well-intentioned local customizations from silently reintroducing the friction the original redesign eliminated.
This closes the loop back to Stage 1: the same instrumented audit methodology used to establish the original legacy baseline becomes the ongoing measurement infrastructure that keeps the optimized state from quietly regressing back toward it.