What I learned

Disruptions are not random — they are tied to system design, role coordination, and communication structures. In trauma handoffs (Catchpole et al., 2013) the transition moment is when errors compound most. I learned to apply a sociotechnical systems lens to clinical environments.

Problem I solved

Synthesized two papers with different methodologies into a unified HF framework, identifying the shared root cause: inadequate system design, not individual failure. This reframing is central to how I approach any clinical workflow project.

What I learned

870 physicians (Melnick et al., 2020) linked poor EHR usability to professional burnout. In pediatric settings (Ratwani et al., 2018) those usability failures translated directly into patient safety events. Alert fatigue and information architecture failures compound silently over time.

Problem I solved

Connected burnout epidemiology with usability engineering to argue that EHR design is a patient safety issue, not just a UX inconvenience. This perspective directly informs how I evaluate digital health tools in consulting work.

What I learned

Transactive memory systems — where team members know who knows what — are a key determinant of ICU team performance. Time-limited trials (TLTs) are a practical HF intervention: they create structured decision checkpoints that reduce both cognitive overload and family distress.

Problem I solved

Applied systems ergonomics to an interprofessional care setting — bridging my clinical background as a physician with HF theory to articulate how shared mental models can be structurally enforced through communication protocols.

What I learned

Trust is not binary — it fluctuates with uncertainty, handover quality, and prefrontal cortex activation patterns visible in fNIRS. I learned to synchronize multimodal signals across instruments (BRITE23 fNIRS, Tobii Pro Glasses, Polar ECG) with different sampling rates, and interpret SHAP values to communicate which biosignal features were most predictive of trust. Supervised by Dr. Ranjana Mehta.

Problem I solved

Aligned three instruments with different sampling rates into a unified analysis pipeline using Python-based resampling and windowed epoch alignment, then validated with Bayesian inference to account for inter-subject variability.

What I learned

RULA and REBA are powerful but observer-dependent. I learned to score them using the angle measurement criteria for all body segments — then verified how MediaPipe’s 33-landmark skeleton can replicate those angles computationally, with trade-offs in occlusion handling and 2D vs 3D projection.

Problem I solved

Calibrated MediaPipe normalized-coordinate outputs to actual anatomical angles, and handled multi-task postures where the dominant arm and trunk position shift continuously — bridging the gap between a classic ergonomics method and a modern CV pipeline.

What I learned

Fatigue is detectable in EMG before the person feels it subjectively. As force drops over time, RMS increases (motor unit recruitment) while MPF shifts downward (slow-twitch fiber dominance). I learned to process raw EMG: remove noise, bandpass filter, compute rolling RMS, and run FFT for MPF — directly applicable to medical device ergonomics validation.

Problem I solved

Built an empirical dataset to validate the Rohmert curve with real Ultium EMG data. The key challenge was electrode placement consistency and normalizing MVC reference values between participants across sessions.

What I learned

Counterbalanced 3-condition designs are essential to avoid order effects. I learned to triangulate subjective (NASA-TLX), behavioral (gaze fixation, saccades), and physiological (HRV from ECG) data into a multi-dimensional cognitive load profile. The Jian Trust Scale and SART provided validated secondary task metrics.

Problem I solved

The study required 2 full data collection rounds to address confounds from the first. I redesigned the between-session protocol, recalibrated Tobii glasses for each participant, and documented the methodology for replication — a real lesson in experimental rigor.

What I learned

AI’s value in neurology is detecting disease years before clinical symptoms (voice biomarkers, gait analysis, retinal scans) and personalizing therapy in real time (adaptive DBS). I learned to evaluate AI claims critically: efficacy data, validation cohort diversity, and explainability requirements for physician adoption.

Problem I solved

Reconciled AI’s efficiency promise with the humanization imperative in chronic neurological disease. I developed a framework positioning AI as a decision-support tool — not a replacement — with the physician as the ethical bridge between algorithm and patient.

What I learned

Unsupervised clustering can reveal natural patient subgroups before labels are imposed, and supervised classifiers then predict intervention timing. Social determinants of health are often as predictive as clinical variables — a key insight for designing equitable health AI.

Problem I solved

Treated ‘high risk’ as a multi-criteria optimization: maximize preventive reach within healthcare capacity, using explainable model outputs to justify resource allocation to clinical stakeholders.

What I learned

Precision medicine restructures the physician-patient relationship, drug development economics, and what counts as evidence. I learned to analyze health innovation through three lenses simultaneously: clinical efficacy, implementation feasibility, and access equity.

Problem I solved

Argued that digital health infrastructure (wearables, FHIR-based EHRs, federated learning) is the enabler that makes precision medicine scalable — connecting the genomic layer to care delivery at population scale without sacrificing privacy.

What I learned

Adherence is not a behavioral failure — it’s a design failure. I learned to apply empathy mapping and patient journey analysis to identify friction points in glycemic monitoring workflows — from meal tracking burden to alarm fatigue in CGM devices.

Problem I solved

Proposed a redesigned digital companion integrating passive sensing (CGM + activity wearable) with context-aware reminders that adapt to the patient’s daily schedule — shifting from a compliance model to a support model.

What I learned

Most EHR integration failures happen at the semantic layer — systems exchange data but don’t interpret it the same way. True interoperability requires governance and terminology standards (SNOMED-CT, LOINC) as much as technical protocols like HL7 or FHIR.

Problem I solved

Traced a lab test request from physician order entry (HL7 ORM) through lab processing (HL7 ORU) to EHR integration (FHIR DiagnosticReport), identifying where legacy HL7 V2.x and modern FHIR-native applications need a translation middleware layer.

What I learned

Gamification fails when layered on top of behavior change — it must be built from motivation theory. I applied Self-Determination Theory (autonomy, competence, relatedness) and the Fogg Behavior Model to design game mechanics that reinforce intrinsic motivation, not just extrinsic rewards.

Problem I solved

Made health data collection invisible to the child by embedding physical activity challenges (step counts, balance tasks) directly into game progression — so data collection happens as a consequence of play, not a separate tracking task.

What I learned

Health entrepreneurship requires a different risk calculus than consumer tech: regulatory timelines, clinical validation, and reimbursement models must be designed in from day one. I learned to build an integrated Gantt + sprint hybrid plan and design KPI dashboards that track clinical and business metrics simultaneously.

Problem I solved

The IQVIA competitive analysis revealed that the telemedicine AI space is dominated by data intermediaries, not clinicians. My competitive advantage: a platform co-designed by a physician who understands both the care workflow and the data architecture.

What I learned

Internet penetration statistics tell an incomplete story: the real barrier to telemedicine is trust, digital literacy, and alignment with local healthcare culture. I learned comparative health systems analysis using a structured framework: infrastructure, regulation, reimbursement, workforce, and patient readiness.

Problem I solved

Designed country-specific adaptations: in Morocco, integrating community health workers as facilitators; in Mexico, building offline-capable tools for intermittent connectivity. Core insight: successful telemedicine requires local adaptation, not global replication.

What I learned

P&G’s New-Business-Creation teams taught me that sustained innovation in large organizations requires structural protection from the core business — exactly what medical device companies need when building digital health divisions. Audi vs. Mercedes showed that regulatory readiness matters as much as technical capability.

Problem I solved

Applied each case to digital health: what would P&G-style innovation look like inside a hospital? How should a telemedicine startup time market entry relative to regulatory milestones? The Apple ecosystem case became a lens for why EHR vendor lock-in persists despite interoperability mandates.

What I learned

Health data (Art. 9 GDPR) requires a specific legal basis from a closed list: explicit consent, vital interests, public health, or research. I learned to navigate the tension between data utility (AI training, federated research) and individual rights (right to erasure, portability). Geolocation combined with health data triggers additional ePrivacy protections.

Problem I solved

Designed a compliance architecture separating consent flows for clinical vs. location data, implementing pseudonymization at the data layer, and defining retention policies by data type — translating GDPR legal text into an engineering-readable specification.

What I learned

Dr. Lennon Rodgers’ prototyping seminar crystallized a principle I apply to all projects: domain expertise, interpersonal dynamics, and process planning are equally critical in medical device design. The FUS + microbubbles research taught me systematic scientific critique: hypothesis, controls, and the gap between animal models and human applicability.

Problem I solved

Across 7 seminars I built a personal framework for evaluating biomedical innovation: Is the hypothesis testable and falsifiable? Does the design control for confounds? What is the clinical translation pathway from animal study to first-in-human? This applies directly to how I evaluate digital health platforms and medical AI claims.

What I learned

Telemedicine saves time — but not equally. Efficiency gains accrue disproportionately to already-connected, health-literate populations. I learned to analyze technology adoption through a health equity lens: who benefits by default, and who requires deliberate design to be included.

Problem I solved

Developed a balanced evaluation framework for pre-deployment: clinical benefit vs. risk by patient segment, digital literacy requirements, connectivity assumptions, privacy architecture, and contingency protocols when technology fails. The output is a decision tool, not a verdict.