Heart-rate response
Mean pulse-rate rise following individual respiratory events.
Phenotype-guided estimation
Estimate the cardiovascular direction of CPAP effect
Explore how published treatment-effect modifiers may shift cardiovascular outcomes in obstructive sleep apnea.
Evidence applicability
Blood pressurei
/mmHg
Study modifier<140 / <90
High-risk phenotype = HRR >9.4 or HB >87.1
High-risk phenotype · Matched subgroup
High-risk OSA phenotype • CPAP-associated benefit signal
HR 0.65 (95% CI 0.48–0.89)
The most specific matched subgroup favored CPAP for MACCE prevention.
Favors CPAPNo effectFavors control
0.250.5124
Hazard ratio (log scale)
What the calculator can reproduce
Published subgroup estimates
The calculator applies the phenotype definition and the most specific matching hazard ratio reported by Azarbarzin and colleagues. It does not refit or extrapolate the trial data.
| Matched population | High-risk OSA | Low-risk OSA | Interpretation |
|---|---|---|---|
| All participants | 0.83 (0.66–1.05) | 1.22 (0.96–1.54) | Both CIs include 1.0 |
| Epworth <11 | 0.76 (0.60–0.98) | 1.30 (1.01–1.66) | Benefit/harm directions separate |
| BP <140/90 | 0.72 (0.54–0.95) | 1.33 (1.00–1.76) | Lower CI bound is 1.00 for low risk |
| Epworth <11 + BP <140/90 | 0.65 (0.48–0.89) | 1.35 (1.00–1.82) | Most specific secondary subgroup |
OR
Hypoxic burden
Event-associated desaturation area multiplied by apnea-hypopnea index.
High-risk OSA phenotype
Meeting either threshold classified a participant as high risk in the pooled analysis.
Second-paper context
Causal-forest feature review
Cohen and colleagues used 23 interacting features. The paper does not publish a deployable model object or sufficient weights to reproduce an individualized treatment-effect score, so this panel is descriptive—not a second calculator.
Enter available SAVE-model features
Values are compared with reported tertile summaries; they are never combined into a score.
No causal-forest score generatedOriginal fitted model required
How these features appeared in reported tertiles
- Absent
Prior MI was more prevalent in the harm tertile: 39.7% vs 30.6% in benefit.
- Absent
Prior revascularization was more prevalent in harm: 40.1% vs 30.2%.
- Present
Prior stroke was more prevalent in benefit: 46.7% vs 41.1%, with p=0.059.
- never
Smoking status differed across tertiles; never-smoking was most common in benefit.
Your valueBenefit meanHarm mean
AHI 2727.329.6
ODI 2626.028.9
T90 14%14.2%17.2%
HR SD 5.85.85.2
The authors explicitly note that isolated linear relationships cannot explain the causal forest. Similarity to a tertile mean must not be interpreted as an individual prediction.
Methods & safeguards
What this tool does—and does not do
Classifies the published phenotype
High risk requires heart-rate response >9.4 bpm or hypoxic burden >87.1 %·min/h. Strict inequalities match the paper.
Matches the narrowest subgroup
Epworth and blood-pressure inputs select the most specific applicable row from the published pooled-RCT table.
Reports—not predicts
The displayed hazard ratio is a subgroup estimate. It is not a personalized absolute risk, probability, or validated bedside recommendation.
Clinical guardrails
- Do not initiate, withhold, or stop CPAP based on this website.
- The evidence concerns cardiovascular outcomes in people with established cardiovascular disease; symptom relief and other CPAP indications are outside this estimator.
- Both analyses were post hoc; the causal-forest model lacked external validation, and the pooled-trial population was predominantly male and generally non-sleepy.
- Heart-rate response and hypoxic burden require event-linked sleep-study processing and should not be substituted with resting pulse or simple time below 90%.
Primary sources