262 Sleep Jc Sept 2022
262.1 Summary
- Sleep Journal Club
- Goals:
- Imagine it’s 1950. You discover that OSA is very common. How do you figure out if it matters?
- Structure of Observational Study ‘Argument’: Disjunctive Syllogism
- Association of obstructive sleep apnea with all-cause andcardiovascular mortality: A population-based study
- Statistical Teaching point:
- Design: Retrospective Cohort Cross-sectional
- Cross-sectional vs Retrospective Cohort
- Obstructive Sleep Apnea as a Risk Factor for Stroke and Death
- Community vs Clinical Cohort
- Prospective Study of Obstructive Sleep Apnea and Incident Coronary Heart Disease and Heart FailureThe Sleep Heart Health Study
- Confounders, Regressions
262.2 Slide outline
262.2.1 Slide 1
- Sleep Journal Club
- Sept 3 2022 ### Slide 2
- Goals:
- Understand that the disjunctive syllogism is the structure of a scientific argument
- Know the limitations of a cross-sectional design as compared to cohort as compared to randomized trial viz-a-viz causal inference
- Understand what a hazard ratio is and why they are so common in medical research
- Know the determinants of a studies power. ### Slide 3
- Imagine it’s 1950. You discover that OSA is very common. How do you figure out if it matters? ### Slide 4
- Structure of Observational Study ‘Argument’: Disjunctive Syllogism ### Slide 5
- TODO: No text extracted from this slide. ### Slide 6
- Association of obstructive sleep apnea with all-cause andcardiovascular mortality: A population-based study
- Design: Retrospective Cohort
- Population: NHANES (multistage-sampling survey) 2005-8 who had OSA status and CV status known (9000 pt)
- Exposed: Yes to ”Have you been told you have a sleep problem: OSA”
- Unexposed: no to the above
- Outcome: Prevalent HTN, CV disease; Incident CV-death and mortality
- Statistics: Logistic (yes/no) and Cox (Hazard Ratio) regressions accounting for demographics, health status.
- Findings: Prevalent HTN+CV indep assoc, CV-death and mort not ### Slide 7
- Statistical Teaching point:
- Ascertainment of exposure status
- Is “Have you ever been diagnosed with OSA?” accurate?
- How would you expect the group assignment to change if PSG was used?
- If you could design a perfect categorization of the groups, what would it be?
- How would this effect the study conclusions? ### Slide 8
- TODO: No text extracted from this slide. ### Slide 9
- Design: Retrospective Cohort Cross-sectional
- Population: Sleep Heart Health (1995-98), n6424 (-1200 missing data)
- Exposed/Unexposed: AHI quartile (0-1.3, 1.4-4.4, 4.5-11.0, 11+)
- Also, Sleep hypoxemia and arousal index
- Outcome: Prevalent odds of CVD (why not incident?)
- Statistics: Logistic regression with possible confounders
- Findings: Quartile III and IV of AHI and %Hypoxemia indepdently associated with prevalent CVD (particularly HF and Stroke).
- Sleep-disordered Breathing and Cardiovascular DiseaseCross-sectional Results of the Sleep Heart Health Study ### Slide 10
- Cross-sectional vs Retrospective Cohort
- Cross-sectional study: exposure and outcome assess simultaneously
- Cohort study: exposure assessed before outcome
- Can only investigate prevalence of disease
- Pros: easy, reliable if exposure doesn’t change (eg genetics), or prior exposure hard to remember (eg diet)
- Cons: can’t determine direction of causality, current exposuresurrogate for past exposure, overrepresent longer duration cases. ### Slide 11
- TODO: No text extracted from this slide. ### Slide 12
- Obstructive Sleep Apnea as a Risk Factor for Stroke and Death
- Design: Prospective cohort (clinical)
- Population: W/o pre-existing CVD referred for SDB eval @ Yale. 1997-2000, n1022
- Exposed: Referred to clinic, AHI > 5 events/hr
- Unexposed: Referred to clinic, AHI < 5 events/hr
- Outcome: incident stroke or death, median follow-up 3.4 yr
- Statistics: Cox-regression~Kaplan Meier/Log rank/proportional hazard
- Findings: OSA associated with HR 1.97 for stroke or death after adjusting for confounders ### Slide 13
- Community vs Clinical Cohort
- Community comparison:
- (C+D) vs A
- Clinical comparison:
- C vs B
- Any factor that changes the likelihood of B+C is a potential confounder
- If, rather than population burden, you are interested in how to care for B+C, this may be the correct method
- Population of interest
- Seeks care
- Has OSA
- A
- B
- C
- D ### Slide 14
- TODO: No text extracted from this slide. ### Slide 15
- Prospective Study of Obstructive Sleep Apnea and Incident Coronary Heart Disease and Heart FailureThe Sleep Heart Health Study
- Design: Prospective cohort (community)
- Population: Sleep Heart Health (1995-98), n6441 (-2000 excluded))
- Exposed: Mild (5-15), Moderate(15-30), Severe (30+) OSA by AHI. Mostly untreated (2.1% overall, 8.4% w mod-sev OSA treated)
- Unexposed: AHI < 5 events/hr
- Outcome: incident CHD before 4/1/06 (~8.7y)
- Statistics: Cox-regression (HR)
- Findings: More severe OSA independently associate with incident CHD (particularly CHF) in men, not women. ### Slide 16
- Confounders, Regressions
- Regression: estimate effect with “all other things held equal”.
- How do you decide what things to include?
- Confounders
- Mediators?
- Colliders?
- Is your understanding of what factors matter correct?
- Can you measure everything you want to measure? ### Slide 17
- TODO: No text extracted from this slide. ### Slide 18
- Sleep-Disordered Breathing and Mortality: A Prospective Cohort Study
- Design: Prospective cohort (community)
- Population: Sleep Heart Health (1995-98), n6441 (-147 excluded d/t OSA treatment)
- Exposed: Mild (5-15), Moderate(15-30), Severe (30+) OSA by AHI. Mostly untreated (2.1% overall, 8.4% w mod-sev OSA treated)
- Unexposed: AHI < 5 events/hr
- Outcome: (incident) death w/ ~8.2y avg follow-up
- Statistics: Cox-regression (HR)
- Findings: Mod-severe OSA (men) and severe OSA (women) associated with increased hazard of mortality. No relationship if age 70+ ### Slide 19
- What is a Hazard Ratio / KM / Cox-regression?We’re all dead in the long run
- 2 versions of Russian Roulette
- X: 1:6 w/ bullet, Y: 1:4 w/ bullet
- HR (1/6) / (1/4) 0.67
- 1000 people play 11 rounds of each
- On round
- P(Surviving 10 rounds?)
- X: (5/6)^10 16% -> 160 survivors
- Y: (3/4)^10 5.6% -> 56 survivors
- RR would be 84% / 94.6% 0.88
- Round 11: X 27 die, Y 14 die yet HR higher in group Y!
- What if Group Y says screw it after 7 rounds, but we want to maximize our data?
- HR allows for “censoring of data” ### Slide 20
- TODO: No text extracted from this slide. ### Slide 21
- Effect of continuous positive airway pressure therapy on recurrence of atrial fibrillation after pulmonary vein isolation in patients with obstructive sleepapnea: A randomized controlled trial
- Design: Open-label RCT (A3 study, Oslo)
- Population: new dx AHI 15+, ESS < 15 (etc), undergoing CA-PVI for Afib and able to tolerate CPAP during run-in period
- Exposed: randomized to auto-CPAP
- Unexposed: randomized to no CPAP
- Outcome: 2+min Afib months 3-12 after CA-PVI by loop recorder
- Statistics: logistic-regression yes/no recurrence
- Findings: Mod-severe OSA (men) and severe OSA (women) associated with increased hazard of mortality. No relationship if age 70+ ### Slide 22
- Negative RCT after positive observational study? power, adherence, & confounding
- Power ~ Sample size/$$$, Alpha, minimally important absolute effect size, info per data point (Continuous > HR > dichotomous), non-adherence/loss f/u
- ?
- Disease+
- Hypothesis+
- Disease-
- Hypothesis-
- Test+
- Study+
- TP
- FP
- PPV: TP / T+
- Test –
- FN
- TN
- NPV: TN / T-
- Se: TP / D+
- Power: Se
- Sp: TN / D-
- Alpha: 1-Sp ### Slide 23
- TODO: No text extracted from this slide. ### Slide 24
- Why do phase 2 trials: Why surrogates? Why process measures?
- Need to estimate each of the above parameters
- Surrogates: Chosen to show a larger effect size smaller trial
- Validity depends on them correlating with the outcome of interest
- Do NOT trust a positive clinical-outcome on a phase 2 trial powered to find a difference in a surrogate: increased error rates (Type 1 and 2)
- Power ~ Sample size/$$$, Alpha, minimally important absolute effect size, info per data point (Continuous > HR > dichotomous), non-adherence/loss f/u ### Slide 25
- Goals:
- Understand that the disjunctive syllogism is the structure of a scientific argument
- Know the limitations of a cross-sectional design as compared to cohort as compared to randomized trial viz-a-viz causal inference
- Understand what a hazard ratio is and why they are so common in medical research
- Know the determinants of a studies power.
262.3 Learning objectives
- Sleep Journal Club
- Goals:
- Imagine it’s 1950. You discover that OSA is very common. How do you figure out if it matters?
- Structure of Observational Study ‘Argument’: Disjunctive Syllogism
- Association of obstructive sleep apnea with all-cause andcardiovascular mortality: A population-based study
262.4 Bottom line / summary
- Sleep Journal Club
- Goals:
- Imagine it’s 1950. You discover that OSA is very common. How do you figure out if it matters?
- Structure of Observational Study ‘Argument’: Disjunctive Syllogism
- Association of obstructive sleep apnea with all-cause andcardiovascular mortality: A population-based study
262.5 Approach
- TODO: Outline the initial assessment or decision point.
- TODO: Outline the next diagnostic or management step.
- TODO: Outline follow-up or escalation criteria.
262.6 Red flags / when to escalate
- TODO: List red flags that require urgent escalation.
262.7 Common pitfalls
- TODO: Capture common errors or missed steps.
262.8 References
TODO: Add landmark references or guideline citations.