207 Locke RIP
207.1 Summary
- PGR Jan 30: What’s new with CO2Clinical Epidemiology of Hypercapnic Respiratory Failure
- Overarching hypothesis
- Background: Hypercapnic Respiratory Failure
- Hypercapnia Projects
- Utah Population Database
- All Payor Claims DataUtah licensed payers and Medicaid report information on their members’ eligibilities and paid medical and pharmacy claims.Pre-research request: summary-level data (600$+150/hr)
- How hypercapnic respiratory failure occurs
- Next Steps:
- Differences in the Diagnosis of Hypercapnic Respiratory Failure by Race and Ethnicity
- Analysis
- Main result
- Why?
207.2 Slide outline
207.2.1 Slide 1
- PGR Jan 30: What’s new with CO2Clinical Epidemiology of Hypercapnic Respiratory Failure
- Presenter: Brian W Locke MS MSCI
- Assistant Professor of Clinical Investigation
- Pulmonary and Critical Care
- Intermountain Medical Center
- COI: MTN, time-series machine learning on continuous sensor data.
- Funding:
- American Thoracic Society Academic Sleep Pulmonary Integrated Research/Clinical Fellowship
- NIH NHLBI T32 U of U PCCM
- Intermountain Fund PCCM Seed Grant: Follow-Up Needs after Discharge with Hypercapnia (FUND-Hypercapnia)
- To be reviewed K23: Bioinformatics Tools to Quantify the Disease Burden Associated with Hypercapnia ### Slide 2
- Overarching hypothesis
- Advice: ‘study what makes you mad’
- The inpatient to outpatient care transition for patients with hypercapnic respiratory failure is horrible.
- I suspect we’d achieve better outcomes by
- systematically identifying patients with hypercapnia
- Establishing an evidence base for undifferentiated ventilatory failure
- consistently implementing measures to mitigate their risks. ### Slide 3
- Background: Hypercapnic Respiratory Failure
- Population-standardized prevalence PaCO2 > 45 mmHg after excluding iatrogenic causes:
- 150 per 100,000 person/year
- Inpatient ICD code for Hypercapnic Respiratory Failure
- 23% 30d readmission rate
- Same as CHF, more than MI
- Usually (66%) with recurrent hypercapnia
- Common contributors to Hypercapnic Respiratory Failure:
- Advanced Age
- Opiate Use
- Obesity
- Advanced Lung Dz
- Multimorbidity ### Slide 4
- Hypercapnia Projects
- Hypercapnic Respiratory Failure Diagnostic Workups in Utah
- Racial inequity in diagnosis?
- Case definitions for hypercapnia are inconsistent
- Methods of identification: TcCO2 and machine learning. ### Slide 5
- Utah Population Database ### Slide 6
- All Payor Claims DataUtah licensed payers and Medicaid report information on their members’ eligibilities and paid medical and pharmacy claims.Pre-research request: summary-level data (600$+150/hr)
- Table 1: Extrapolated hypercapnic respiratory failure 1-yr period-prevalence by diagnosis codes in UT
- Year
- Hypercapnia Diagnosis
- (per 100k persons)
- 2016
- 133.5
- 2017
- 132.2
- 2018
- 143.3
- 2019
- 147.6
- ICD-10-CM Codes for Hypercapnic Respiratory Failure
- J96.02: Acute Respiratory Failure with Hypercapnia
- J96.12: Chronic Respiratory Failure with Hypercapnia
- J96.22: Acute and Chronic Respiratory Failure with Hypercapnia
- J96.92: Respiratory Failure, unspecified with Hypercapnia
- E66.2, Obesity with Hypoventilation
- Utah
- count
- Any Diagnosis
- 4064
- 4104
- 4520
- 4728 ### Slide 7
- How hypercapnic respiratory failure occurs
- Apply relevant guideline:
- CHEST 2023 NMD
- ATS 2021 Hypercap. COPD
- ATS 2019 OHS
- What workup and management is indicated?
- Hypothesis
- Reality
- Implied ### Slide 8
- TODO: No text extracted from this slide. ### Slide 9
- Next Steps:
- Individual patient data is in regulatory approval
- Recreate pre-research analysis, with additional covariates and outcomes
- Covariates: prior diagnoses, demographics
- Mediators: DME orders, Medication Fills, Specialist Referrals, Diagnostic Testing
- Outcomes: Presentation to ED/Urgent, Re-hospitalization, Mortality
- Linkage to Intermountain Health:
- cohort creation with lab data (ABG, VBG, Diagnostic Modeling) ### Slide 10
- Differences in the Diagnosis of Hypercapnic Respiratory Failure by Race and Ethnicity
- Diagnosis disparities in hypoxemic respiratory failure are well documented; less is known about hypercapnic respiratory failure
- Hypothesis: the method of diagnosis (ABG, VBG, or both) of hypercapnic respiratory failure differs between race/ethnicities, which has implications for treatment (device) qualifications
- TriNetX (aggregated health records from 76 hospitals in the United States)
- Adult encounters occurring between January 1 and December 31, 2022
- Ambulatory, emergency department, and hospitalized patients
- Non-missing race/ethnicity information ### Slide 11
- Analysis
- 1°: Among patients who receive a diagnosis code for hypercapnic respiratory failure based on day-1 evidence,
- Multinomial logistic regression, 3-outcomes ABG PaCO2 ≥ 45 mmHg only, VBG PCO2 ≥ 50 mmHg only, or ABG & VBG
- adjust for the type of encounter (ambulatory, emergency, inpatient), age, sex, and region of the U.S. (West, Midwest, South, or Northeast)
- 2°: Why?
- Difference in propensity for ABG?
- Different severity of disease?
- Different propensity to label as Hypercapnia? ### Slide 12
- Main result
- Day-1 Hypercapnia Evidence
- Estimate
- Non-Hispanic White
- Non-Hispanic Black
- Hispanic/
- Latino
- Non-Hispanic Asian
- Non-Hispanic AI/AN
- Non-Hispanic NH/PI
- ABG only
- Unadjusted
- 7,857 (54.4%)
- 1,579 (45.9%)
- 488 (45.2%)
- 160 (50.8%)
- 51 (38%)
- 23 (85%)
- ABG and VBG
- 4,189 (29.0%)
- 1080 (31.4%)
- 381 (35.3%)
- 104 (33.0%)
- 62 (47%)
- 1 (4%)
- VBG only
- 2,410 (16.7%)
- 784 (22.8%)
- 210 (19.5%)
- 51 (16.2%)
- 20 (15%)
- 3 (11%)
- Odds Ratio
- (Adjusted)
- 1.0 (Ref)
- 1.76
- [1.60-1.94]
- 1.04
- [0.88-1.23]
- 1.03
- [0.75-1.39]
- 0.8
- [0.5-1.3]
- 0.7
- [0.2-2.3]
- Average Marginal Effect (Adjusted)
- 0 (Ref)
- 8.8%
- [7.2% to 10.5%]
- 0.5%
- [-1.7% to 2.8%]
- 0.3%
- [-3.8% to 4.5%]
- -3%
- [-9% to 3%]
- -4.4%
- [-17% to 8%] ### Slide 13
- Why?
- Propensity for ABG & VBG
- Different disease severity
- Propensity for ICD label | CO2
- Logistic GEE (encounter type, region, age, sex, cat. BMI)
- Distribution of PCO2 | ABG, VBG
- Density Plot
- Logistic GEE (enctounter type, region, PCO2-spline, HCO3-spline) ### Slide 14
- Why?
- Propensity for ABG & VBG
- Different disease severity
- Propensity for ICD label | CO2
- Logistic GEE (encounter type, region, age, sex, cat. BMI)
- Distribution of PCO2 | ABG, VBG
- Density Plot
- Logistic GEE (encounter type, region, PCO2-spline, HCO3-spline) ### Slide 15
- Why?
- Propensity for ABG & VBG
- Different disease severity
- Propensity for ICD label | CO2
- Logistic GEE (encounter type, region, age, sex, cat. BMI)
- Distribution of PCO2 | ABG, VBG
- Density Plot
- Logistic GEE (encounter type, region, PCO2-spline, HCO3-spline) ### Slide 16
- Limitations
- Inability to account for hospital- and provider- level behavior
- Unable to control for all relevant confounders (eg disease severity)
- potential for incomplete ascertainment of outcomes.
- Difference in diagnosis vs Disparity in diagnosis?
- Does this translate to downstream consequences?
- Maybe VBG diagnosis is a good thing and device qualifications need to change? ### Slide 17
- Take home points
- Study Question: Are minoritized patients more likely to be diagnosed with hypercapnic respiratory failure on VBG-only evidence?
- Results: In nationwide aggregated health record data, Black patients with hypercapnic respiratory failure were 8.8% (7.2-10.5%) more likely to be diagnosed with VBG-only evidence. This may result from differing propensities to undergo VBG (higher) and ABG (lower) sampling, rather than differences in disease severity or diagnostic labeling.
- Interpretation: Where therapeutic interventions rely on results from arterial sampling (e.g. home NIV), Black patients may be at increased risk for lower-quality care. ### Slide 18
- The Consistency of Hypercapnic Respiratory Failure Case Definitions in Electronic Health Record Data
- Hypothesis: case definitions in use identify importantly different patients
- Literature Review
- Emulation Dataset
- Literature reviews of studies of any-cause hypercapnia
- Assess overlap of the case definitions in TriNetX database
- 15 studies using 13 case definitions to identify patients with hypercapnia.
- 3 included elements that couldn’t be emulated in EHR data (screen ABG, NIV on d/c, consultation)
- N445,686 U.S. adults from 76 institutions who had at least one characteristic that should make a reasonable clinician suspect hypercapnia may be present during 2022
- Cohen’s kappa (κ, agreement beyond chance) ### Slide 19
- Case Definitions for Hypercapnic Resp Failure
- Case Definition
- Adler
- PaCO2 ≥ 47.25 mmHg and either NIV or IMV procedure code
- Thille
- PaCO2 ≥ 45 mmHg, arterial pH < 7.35, and either NIV or IMV procedure code
- Ouanes-Besbes
- PaCO2 ≥ 45 mmHg and pH < 7.35
- Calvo
- PaCO2 > 45 mmHg, arterial pH < 7.35 and a NIV procedure code
- Bulbul
- CO2 ≥ 45 mmHg and pH < 7.45
- Meservey
- Any diagnosis code for Hypercapnic Respiratory Failure
- Vonderbank
- Either a PaCO2 ≥ 45 mmHg or venous CO2 ≥ 45 mmHg and venous pH > 7.35
- Wilson
- PaCO2 ≥ 50 mmHg and pH 7.35-7.45
- Cavalot
- Either PaCO2 ≥ 45 mmHg and arterial pH ≤ 7.35, or venous CO2 ≥ 50 mmHg and venous pH ≤ 7.34
- Chung
- PaCO2 ≥ 45 mmHg and pH ≤ 7.45
- Median κ 0.35 ### Slide 20
- Why do the case definitions identify different patients?
- Some difference in target population
- Sensitivity of diagnosis codes is low →
- Restricted cubic splines
- Method of testing differs (ABG, VBG) ### Slide 21
- Why do the case definitions identify different patients?
- Entire Enriched Cohort
- Only ABG obtained
- (first day)
- Only VBG obtained
- Both ABG and VBG obtained (first day)
- N515,286
- N132,193
- N94,614
- N55,049
- Setting
- Emergency encounter
- 33% (171,814)
- 16% (21,455)
- 42% (39,816)
- 14% (7,580)
- Inpatient encounter
- 67% (343,559)
- 84% (110,744)
- 58% (54,857)
- 86% (47,484)
- Critical care billed
- 23% (118,143)
- 36% (47,636)
- 21% (19,528)
- 48% (26,528)
- US region
- South
- 47% (242,266)
- 65% (85,849)
- 30% (27,918)
- 32% (17,574)
- Northeast
- 25% (129,446)
- 14% (18,937)
- 49% (46,609)
- 31% (17,300)
- Midwest
- 7% (38,471)
- 8% (10,013)
- 9% (8,046)
- 10% (5,534)
- West
- 20% (105,103)
- 13% (17,394)
- 13% (12,041)
- 27% (14,641)
- Proportion of patients from the enriched sample identified by each case definition
- Adler
- 4% (18,118)
- 8% (10,602)
- 0% (0)
- 14% (7,516)
- Thille
- 3% (16,263)
- 7% (9,790)
- 12% (6,473)
- Ouanes-Besbes & Bülbül
- 7% (37,336)
- 20% (25,812)
- 21% (11,524)
- Calvo
- 1% (6,566)
- 3% (4,094)
- 4% (2,472)
- Meservey
- 6% (29,009)
- 8% (11,080)
- 4% (4,032)
- 17% (9,196)
- Vonderbank
- 17% (86,137)
- 31% (40,411)
- 26% (24,267)
- 39% (21,459)
- Wilson
- 10% (13,113)
- 9% (5,005)
- Cavalot
- 11% (58,481)
- 21% (27,264)
- 13% (12,555)
- 34% (18,662)
- Chung
- 11% (57,813)
- 32% (17,402) ### Slide 22
- The case definitions identify different patients
- Case Definition
- n
- Female
- Black
- HCO3-
- PaCO2
- Death (2-mo)
- Adler
- PaCO2 ≥ 47.25 mmHg and either NIV or IMV procedure code
- 18,118
- 44%
- 16%
- 26.6 (±7.3)
- 66.5 (±23.7)
- 28%
- Thille
- PaCO2 ≥ 45 mmHg, arterial pH < 7.35, and either NIV or IMV procedure code
- 16,263
- 42%
- 24.8 (±6.8)
- 63.8 (±21.7)
- Ouanes-Besbes
- PaCO2 ≥ 45 mmHg and pH < 7.35
- 37,336
- 46%
- 14%
- 24.8 (±5.9)
- 60.2 (±19.6)
- 18%
- Calvo
- PaCO2 > 45 mmHg, arterial pH < 7.35 and a NIV procedure code
- 6,556
- 50%
- 17%
- 27.8 (±6.7)
- 64.7 (±16.5)
- 20%
- Bulbul
- CO2 ≥ 45 mmHg and pH < 7.45
- 38,504
- 47%
- 25.8 (±5.9)
- 62.7 (±16.5)
- Meservey
- Any diagnosis code for Hypercapnic Respiratory Failure
- 29,009
- 27.1 (±7.4)
- 55.1 (±19.5)
- 21%
- Vonderbank
- Either a PaCO2 ≥ 45 mmHg or venous CO2 ≥ 45 mmHg and venous pH > 7.35
- 86,137
- 26.5 (±5.5)
- 57.1 (±20.4)
- 13%
- Wilson
- PaCO2 ≥ 50 mmHg and pH 7.35-7.45
- 48%
- 28.6 (±5.1)
- 54.9 (±20.3)
- 12%
- Cavalot
- Either PaCO2 ≥ 45 mmHg and arterial pH ≤ 7.35, or venous CO2 ≥ 50 mmHg and venous pH ≤ 7.34
- 58,481
- 45%
- 56.8 (±20.1)
- Chung
- PaCO2 ≥ 45 mmHg and pH ≤ 7.45
- 58.813
- 26.3 (±6.1)
- 58.5 (±20.4) ### Slide 23
- Take-Home Points
- Study Question: Do the case definitions for hypercapnic respiratory failure in health record data identify the same, or similar, patients?
- Results: Applying ten case-definitions from the published literature to a large, aggregated nationwide dataset of health records identified different patients (median kappa of 0.35) who had widely variable characteristics and outcomes.
- Interpretation: Case definitions for identifying hypercapnic respiratory failure in health record data create substantively different cohorts, which hampers the synthesis of research findings. ### Slide 24
- How could we diagnose hypercapnic RF?
- Analysis
- Patients
- Model
- Discrim.
- Calibration
- Overall
- AUC
- E:O
- CITL
- CS
- Brier Score
- Primary
- ABG
- LASSO
- 0.763
- 0.986
- 0.028
- 1.110
- 0.180
- RF
- 0.758
- 0.969
- 0.062
- 1.127
- 0.184
- Silver Standard
- ABG or VBG
- 0.749
- 0.925
- 0.156
- 1.098
- 0.190
- 0.745
- 0.912
- 1.190
- 0.193
- Inverse Probability Weighted
- ABG (weighted to full sample)
- 0.792
- 0.982
- 0.041
- 1.227
- 0.178
- 0.782
- 0.957
- 0.094
- 1.251
- West
- Midwest
- South
- Northeast
- Training
- Validation
- n 16,277
- n 20,837
- n 70,921
- n 23,816 ### Slide 25
- How could we diagnose hypercapnic RF?
- SRMA of TcCO2-PaCO2 Agreement studies
- 7021 paired measurements from 2817 participants in 73 studies
- Mean bias: 0.09 mmHg lower
- Pop. Std. dev. (within- and between- study variance): 4.6 mmHg
- Simulate estimates for all 158k TriNetX patients w/ ABG in 2022:
- Sens: 84.2%
- Spec: 91.0%,
- NPV: 93.0% @ 30.3% prev.
- PPV: 80.4%. @ 30.3% prev
- Next: Context specific estimates
- Validation amongst no-ABG pts. ### Slide 26
- Next Steps
- Link Intermountain Health – Select Health
- Cost and utilization data
- Assess patient-, unit-, institution-variation in post-discharge management in the Intermountain Health System
- Outcomes: Select-Health and APCD
- UPDB-Intermountain linkage
- Effect of that variation on ED/readmissions, mortality ### Slide 27
- Overdiagnosis: a diagnosis that does not benefit the patient.
- Fair Umpires: existence and extent of overdiagnosis may be considered using the following questions:
- Are people diagnosed by diagnostic strategy 2 but not by diagnostic strategy 1 at an increased risk of disease specific clinical outcomes if left untreated compared to people not diagnosable by diagnostic strategy 1 or 2 (prognosis evidence)?
- Does diagnosis and treatment of people who are detected using diagnostic strategy 2 but not detected using diagnostic strategy 1 result in an increased beneficial reduction in disease specific clinical outcomes that outweighs the ensuing increased risk of harms, compared to no diagnosis or treatment (utility evidence)? ### Slide 28
- Credit to the crew
- Intermountain Research
- Sam Brown MD MSCI
- Ithan Peltan MD MSc
- Holly Frost MD PhD
- U of U DBMI
- Joseph Finkelstein MD PhD
- Ramkiran Gouripeddi MBBS MSc
- U of U PCCM
- Barb Jones MD MSCI
- Jeanette Brown MD PhD
- Rob Paine MD
- Dustin Anderson-Bell MD
- U of U Medical Machine Intelligence Lab
- Warren Pettine MD
- Matthias Christensen PhD
- U of U Epidemiology
- Benjamin Brintz PhD
- U of CA Davis PCCM
- Krishna Sundar MD
- Email: brian.locke@imail.org
207.3 Learning objectives
- PGR Jan 30: What’s new with CO2Clinical Epidemiology of Hypercapnic Respiratory Failure
- Overarching hypothesis
- Background: Hypercapnic Respiratory Failure
- Hypercapnia Projects
- Utah Population Database
207.4 Bottom line / summary
- PGR Jan 30: What’s new with CO2Clinical Epidemiology of Hypercapnic Respiratory Failure
- Overarching hypothesis
- Background: Hypercapnic Respiratory Failure
- Hypercapnia Projects
- Utah Population Database
207.5 Approach
- TODO: Outline the initial assessment or decision point.
- TODO: Outline the next diagnostic or management step.
- TODO: Outline follow-up or escalation criteria.
207.6 Red flags / when to escalate
- TODO: List red flags that require urgent escalation.
207.7 Common pitfalls
- TODO: Capture common errors or missed steps.
207.8 References
TODO: Add landmark references or guideline citations.