Draft

188  Locke Fellow Research

188.1 Summary

  • fellow researchfrom a fellow fellow
  • 3 Components of a Project:
  • Research Question
  • Why do mentors mentor?
  • What Data Are You Going To Use
  • Databases
  • Methods:
  • How can I help?
  • Hypercapnic Respiratory Failure: Overall research premise
  • Completed Research
  • Proposed Research (at least 2025 – 2030):

188.2 Slide outline

188.2.1 Slide 1

  • fellow researchfrom a fellow fellow
  • disclaimer: what follows are my views and not the views of…
  • Brian Locke ### Slide 2
  • 3 Components of a Project:
  • Research Question
  • Data
  • Methods
  • Consider each before proceeding… ### Slide 3
  • Research Question
  • Significance
  • Contingency
  • Tractability
  • Neglected-ness
  • Leverage
  • x
  • Traditional Paradigm
  • ”Free Range” Approach
  • Find a mentor who is investigating what you are interested in.
  • Learn how to do what they are doing
  • Ultimately branch out later in your career
  • Formulate a question that interests you
  • Determine what data you can use to answer it
  • Find a group of collaborators
  • The only thing that works for
  • You’re limited to what people are doing
  • Might flop; big time investment ### Slide 4
  • TODO: No text extracted from this slide. ### Slide 5
  • Why do mentors mentor?
  • Intrinsic:
  • Training the Next Generation; Improve Medicine; Fun
  • Incentives:
  • Clinical Track (majority) – need impact outside of the institution (“excellence”) in 2 of: Clinical Care, Education, Administration, and Research
  • Tenure Track or similar (minority) – hired (and promoted) to get extra-mural grants: you are helpful to the extent you facility this (publications, preliminary work).
  • Favors executing previously-conceived projects; has ability/skill ### Slide 6
  • What Data Are You Going To Use
  • Database work
  • Chart Review
  • Meta-research
  • Survey
  • Prospective
  • Basic Science
  • Do we have access to it now?
  • How much data-cleaning is needed?
  • Can it answer the question?
  • How big?
  • How long will it take?
  • Can it be used for multiple things?
  • Systematic Reviews with or without meta-analysis
  • Who will be surveyed and how will it be done?
  • Do a feasibility study at most.
  • ??? ### Slide 7
  • What Data Are You Going To Use
  • Database work
  • Chart Review
  • Meta-research
  • Survey
  • Prospective
  • Basic Science
  • Do we have access to it now?
  • How much data-cleaning is needed?
  • Can it answer the question?
  • How big?
  • How long will it take?
  • Can it be used for multiple things?
  • Systematic Reviews with or without meta-analysis
  • Who will be surveyed and how will it be done?
  • Do a feasibility study at most.
  • ???
  • Data cleaning is time/expertise intensive
  • Chart Review is low yield
  • Time (and team) intensive
  • Unique methods expertise; limited impact
  • Inflexible time commitment, mentor dependent ### Slide 8
  • Databases
  • Health-Record Data
  • TriNetX -> Ram Gouripeddi
  • Epic Cosmos -> Vik Deshmukh (IM pays for expedited data requests)
  • ICU Databases:
  • Community Cohorts:
  • Trial Data: ### Slide 9
  • Methods:
  • Who is going to do the statistics?
  • Who will help you design the study?
  • “Traditional Mentorship Structure”: they provide
  • Otherwise… resources:
  • CTSI
  • MSCI ### Slide 10
  • How can I help?
  • What Can I Offer?
  • What I Can’t Offer
  • I have some ideas
  • I can do (and teach) the needed study design, epi/statistics, and database skills
  • I want (and am incentivized) to collaborate or advise on other projects
  • I think it is very helpful to have external advisors
  • I’ve (big) helped with Brittany and Jeff’s projects and (little) helped with Ryan, Jared, and Darren’s
  • I cannot provide resources (statisticians, study coordinators, grad students)
  • I cannot offer advocacy that carries weight
  • I would not be a wise primary-mentor for someone who wants a primary research career
  • But, I can help ### Slide 11
  • Hypercapnic Respiratory Failure: Overall research premise
  • inflicts a high burden of disease (prevalence morbidity)
  • is becoming a bigger problem (obesity, opiates, multimorbidity)
  • is (potentially) treatable with NIV, GLP-1Ra, etc.
  • but the current evidence base is inadequate
  • is comparatively neglected (biases against this type of patient)
  • Therefore, doing a better job recognizing hypercapnia will allow:
  • Better quantification (ie. research) about hypercapnia’s impacts
  • Better care for patients with known best management. ### Slide 12
  • Completed Research ### Slide 13
  • Proposed Research (at least 2025 – 2030):
  • OVERALL: Patients with Consequentially-Missed Hypercapnia can be Identified Using Health Record Elements
  • OVERALL: There is a significant burden of unrecognized hypercapnia among hospitalized patients.
  • 3 aims
  • Aim 1: Use NLP / AI to incorporate unstructured data
  • Aim 2: Prospective Validation of Positive Predictive Value
  • Aim 3: Assess Potential Impact of Unrecognized Hypercapnia
  • Hypothesis
  • RN/ED Triage note symptoms
  • simple NLP (CLAMP) vs
  • Large-Language Models
  • ‘Risk’-modeling can identify inpatients missed by their teams
  • Patients predicted likely to have had unrecognized hypercapnia suffer adverse outcomes, like readmission.
  • Rationale
  • Signs/Symptoms improve accuracy
  • e.g. STOP-BANG for OSA
  • Must be verified to identify unrecognized hypercapnia
  • Would expanded recognition lead to overdiagnosis?
  • Approach
  • Train & evaluate performance on retrospective, local health records
  • Prospectively apply TcCO2 to inpatients predicted likely to have hypercapnia – do they?
  • ↑readmission mortality In patients who likely had hypercapnia?

  • 189 likely hypercapnia admissions prior to actual recognition?

  • Payoff: Allow hypercapnic respiratory failure -not dependent on clinician identification- as an exposure or outcome
  • Career Development: Skillset bridging bioinformatics and clinical investigation required to validate promising technology ### Slide 14
  • In the interim, preliminary work needed…
  • strengthen the case to investigate hypercapnia
  • augment tools to study hypercapnia epi
  • Inpatient Hypercapnic Respiratory Failure
  • ABG CO2>45 vs VBG CO2>50
  • Who gets steroids / antibiotics / diuretics
  • National Consequence of ↑ICD
  • Who is diagnosed / managed as OHS?
  • Dysutility of FP vs FN during hypercap workup
  • Hypothesis
  • Vent Support, Dx is equivalent
  • Most get kitchen sink
  • High Readmission Rates
  • Conditions that exclude OHS vary
  • FN currently underweighted
  • Data
  • TriNetX (cleaning done)
  • National Inpatient Sample
  • Survey
  • Group Session Survey
  • Methods
  • Descriptive Stats, Regressions
  • Descriptive Stats, Regression
  • Randomized Vignette’s
  • Anchor-based
  • Team
  • Somya Mishra (Anesth/Sleep Fellow)
  • Chaiyakunapruk (Pharm/Econ) ### Slide 15
  • Immersive Virtual Reality – Joseph Finkelstein MD PhD MA, Dept Bioinformatics
  • VR: immersive (headset) vs non-immersive (computer game).
  • Immersion is the experience of being absorbed (termed, ‘presence’), forgetting their embodied presence and thus responding as if the environment is real.
  • ICU Use cases: Relaxation (several feasibility studies), Cognitive/physical mobilization (1 RCT), Distraction/pain control (1 RCT pre-op), Delirium, Sleep (1 +RCT)
  • Feasibility Study (done) → Next Steps? ### Slide 16
  • Hypercap. Diagnosis: ABG vs VBG
  • Problem: Emergency room providers often use VBGs to “diagnose” hypercapnia, but it’s unclear if they get managed like ABG-diagnosed
  • Study Question: Do patients admitted with VBGs suggestive of hypercapnia
  • Receive the same care (ie. ICD-codes of hypercapnic resp failure)
  • Have the same outcomes (ie. intubation rate, NIV use)
  • As those with ABGs confirming hypercapnia?
  • Data: Nationally aggregated EHR (TriNetX) of patients
  • Methods: Inverse Probability (of A/VBG)-Weighted Comparison ### Slide 17
  • Hypercapnia Diagnostics: Harm of FP vs FN
  • Problem: Thresholds for any test [e.g. HCO3- level, Modeled Pr(Hypercap)] should be defined to minimize the dis-utility of incorrect results. No evidence exists for hypercapnia diagnosis.
  • Study Question: How bad is it to miss hypercapnia in comparison to overtesting/overdiagnosing it?
  • Data: Survey (initially, providers – via Hosp/PCCM/Sleep groups)
  • Methods: survey ### Slide 18
  • Diagnostic Criteria of OHS:
  • Problem
  • Study Question
  • Data
  • Methods ### Slide 19
  • National Consequence of Hypercapnia Admissions
  • Problem
  • Study Question
  • Data
  • Methods ### Slide 20
  • Jeanette Brown MD PhD
  • VR: immersive (headset) vs non-immersive (computer game).
  • Immersion is the experience of being absorbed (termed, ‘presence’), forgetting their embodied presence and thus responding as if the environment is real.
  • ICU Use cases: Relaxation (several feasibility studies), Cognitive/physical mobilization (1 RCT), Distraction/pain control (1 RCT pre-op), Delirium, Sleep (1 +RCT)
  • Feasibility Study (done) → Next Steps?

189.1 Learning objectives

  • fellow researchfrom a fellow fellow
  • 3 Components of a Project:
  • Research Question
  • Why do mentors mentor?
  • What Data Are You Going To Use

189.2 Bottom line / summary

  • fellow researchfrom a fellow fellow
  • 3 Components of a Project:
  • Research Question
  • Why do mentors mentor?
  • What Data Are You Going To Use

189.3 Approach

  1. TODO: Outline the initial assessment or decision point.
  2. TODO: Outline the next diagnostic or management step.
  3. TODO: Outline follow-up or escalation criteria.

189.4 Red flags / when to escalate

  • TODO: List red flags that require urgent escalation.

189.5 Common pitfalls

  • TODO: Capture common errors or missed steps.

189.6 References

TODO: Add landmark references or guideline citations.

189.7 Slides and assets

189.8 Source materials