The size of each data marker is proportional to the weight carried by the corresponding study in the random-effects pairwise meta-analysis.
Circles represent individual results for each trial with the size of the circle being proportional to its weight in the random-effects meta-analysis. Each meta-regression line (solid lines) and its 95% confidence interval (shaded areas) were estimated using a random-effects linear meta-regression model with mean use of CPAP as the covariate. The dashed lines represent the overall difference. Treatment difference data are CPAP minus inactive control.
eMethods. Search strategy
eFigure 1. Network map showing the number of trials and patients in which CPAP, mandibular advancement devices and inactive controls were compared
eFigure 2. Comparison-adjusted funnel plots
eFigure 3. Association between the length of follow-up in each study and the reported effect of CPAP on BP vs inactive control
eFigure 4. Association between the reported average baseline BP in each study and the corresponding reported effect of CPAP on BP vs inactive control
eFigure 5. Association between the reported mean baseline apnea-hypopnea index (AHI) and the reported effect of CPAP on BP vs inactive control
eFigure 6. Association between the reported mean baseline oxygen desaturation index (ODI) and the reported effect of CPAP on BP vs inactive control
eFigure 7. Summary of proportion of trials at low, unclear and high risk of bias in each domain of the Cochrane Collaboration’s tool for assessing risk of bias (N=51)
eTable 1. Sensitivity analysis of the network meta-analysis on systolic blood pressure (SBP)
eTable 2. Sensitivity analysis of the network meta-analysis on diastolic blood pressure (DBP)
eTable 3. Results of the network meta-analysis common-heterogeneity model
eTable 4. Difference in the reported effects of CPAP vs inactive control on BP in trials using sham CPAP and any other placebo compared to trials using no placebo
eTable 5. Difference in the reported effects of CPAP vs inactive control on BP in trials from which morning, office or 24h BP data were extracted compared to studies from which daytime BP was extracted
eTable 6. Risk of bias of included trials evaluated using the Cochrane Risk of Bias tool
Customize your JAMA Network experience by selecting one or more topics from the list below.
Bratton DJ, Gaisl T, Wons AM, Kohler M. CPAP vs Mandibular Advancement Devices and Blood Pressure in Patients With Obstructive Sleep Apnea: A Systematic Review and Meta-analysis. JAMA. 2015;314(21):2280–2293. doi:10.1001/jama.2015.16303
Obstructive sleep apnea is associated with higher levels of blood pressure (BP), which can lead to increased cardiovascular risk.
To compare the association of continuous positive airway pressure (CPAP), mandibular advancement devices (MADs), and inactive control groups (placebo or no treatment) with changes in systolic BP (SBP) and diastolic BP (DBP) in patients with obstructive sleep apnea.
The databases of MEDLINE, EMBASE, and the Cochrane Library were searched up to the end of August 2015 and study bibliographies were reviewed.
Randomized clinical trials comparing the effect of CPAP or MADs (vs each other or an inactive control) on BP in patients with obstructive sleep apnea were selected by consensus. Of 872 studies initially identified, 51 were selected for analysis.
Data Extraction and Synthesis
Data were extracted by one reviewer and checked by another reviewer. A network meta-analysis using multivariate random-effects meta-regression was used to estimate pooled differences between each intervention. Meta-regression was used to assess the association between trial characteristics and the reported effects of CPAP vs inactive control.
Main Outcomes and Measures
Absolute change in SBP and DBP from baseline to follow-up.
Of the 51 studies included in the analysis (4888 patients), 44 compared CPAP with an inactive control, 3 compared MADs with an inactive control, 1 compared CPAP with an MAD, and 3 compared CPAP, MADs, and an inactive control. Compared with an inactive control, CPAP was associated with a reduction in SBP of 2.5 mm Hg (95% CI, 1.5 to 3.5 mm Hg; P < .001) and in DBP of 2.0 mm Hg (95% CI, 1.3 to 2.7 mm Hg; P < .001). A 1-hour-per-night increase in mean CPAP use was associated with an additional reduction in SBP of 1.5 mm Hg (95% CI, 0.8 to 2.3 mm Hg; P < .001) and an additional reduction in DBP of 0.9 mm Hg (95% CI, 0.3 to 1.4 mm Hg; P = .001). Compared with an inactive control, MADs were associated with a reduction in SBP of 2.1 mm Hg (95% CI, 0.8 to 3.4 mm Hg; P = .002) and in DBP of 1.9 mm Hg (95% CI, 0.5 to 3.2 mm Hg; P = .008). There was no significant difference between CPAP and MADs in their association with change in SBP (−0.5 mm Hg [95% CI, −2.0 to 1.0 mm Hg]; P = .55) or in DBP (−0.2 mm Hg [95% CI, −1.6 to 1.3 mm Hg]; P = .82).
Conclusions and Relevance
Among patients with obstructive sleep apnea, both CPAP and MADs were associated with reductions in BP. Network meta-analysis did not identify a statistically significant difference between the BP outcomes associated with these therapies.
Obstructive sleep apnea is characterized by recurring cessations or reductions in respiratory flow due to upper airway collapse during sleep. The estimated prevalence of symptomatic obstructive sleep apnea in Western countries is 2% to 4%; however, prevalence is increasing due to levels of obesity in these populations.1 The condition is associated with oxygen desaturations and arousals from sleep, which can lead to increases in blood pressure (BP) and risk of cardiovascular disease.2
Continuous positive airway pressure (CPAP) has been shown to be an effective treatment for improving symptoms of obstructive sleep apnea, such as daytime sleepiness,3 and meta-analyses have shown it to be associated with a reduction of about 2 mm Hg in BP.4-8 However, the estimated association of trial-level characteristics (mean nightly CPAP use in particular) with the effects of CPAP reported in individual randomized clinical trials (RCTs) has been less consistent.5,6
Alternative treatments often used by patients unable to tolerate CPAP are mandibular advancement devices (MADs), which work by protruding the mandible and tongue to keep airways open during sleep.9 The association of use of MADs with reductions in BP is less clear. A recent meta-analysis that included only 2 RCTs was inconclusive.10 To our knowledge, no meta-analysis has comprehensively compared CPAP vs MADs on change in BP, perhaps due to a lack of RCTs. The most recent meta-analysis5 briefly reviewed 2 trials,11,12 comparing the effects of CPAP with MADs on change in BP, and found conflicting results between the studies.
The primary aim of our study was to perform a network meta-analysis13 comparing the association of CPAP vs MADs and vs an inactive control (eg, placebo or no treatment) with changes in systolic BP (SBP) and diastolic BP (DBP) in patients with obstructive sleep apnea. A secondary aim was to explore the association of trial-level characteristics, such as mean nightly CPAP use, with the reported treatment effects of CPAP vs inactive control therapy on BP outcomes.
The studies must have randomized participants aged 18 years or older with a diagnosis of obstructive sleep apnea (defined by an apnea-hypopnea index of ≥5/h) to at least 2 of the following treatments: (1) CPAP, (2) MADs, or (3) inactive control, such as sham CPAP, placebo MADs, or conservative treatment (no active therapy). Trials must also have measured and reported data on SBP or DBP at a follow-up visit and preferably also at enrollment or randomization, or reported a treatment effect for either outcome. If 2 eligible studies contained a significant overlap in patients, the larger of the 2 studies was used in the analysis. The protocol for this meta-analysis appears in Supplement 1.
Literature searches were performed independently by 2 of the authors (D.J.B. and A.M.W.) using the databases of MEDLINE, EMBASE, and the Cochrane Library from inception to the end of August 2015. The RCTs were identified using the Cochrane Collaboration highly sensitive search strategy (sensitivity-maximizing and precision-maximizing version).14 The general electronic search strategy appears in Supplement 2. The bibliographies of all eligible trials and review articles were also screened for relevant trials that might have been missed in the database search. Inclusion was restricted to trials reported in English.
Two of the authors (D.J.B. and A.M.W.) assessed the eligibility of studies found in the literature searches. One author (D.J.B.) extracted the relevant data from eligible studies, which was then independently checked by another author (T.G.). Trial characteristics, such as sample size, length of follow-up, type of control group (eg, placebo or no treatment), and type of study (eg, crossover or parallel), were recorded. The main outcome of interest was the change in SBP and DBP between baseline and follow-up in the 3 treatment groups (CPAP, MADs, inactive control). Treatment effects were extracted directly from the studies along with standard errors, 95% confidence intervals, or P values. If treatment effects were not reported, other data, such as mean (standard deviation) for SBP and DBP for each treatment group at each visit or the change in SBP and DBP between visits in each group, were recorded and used to estimate the treatment effect of interest.
Measurements of BP during the daytime (while the patient was ambulatory), during the morning, or during an office visit were preferable (in that order). Otherwise, 24-hour ambulatory BP measurements were used. Summary statistics for the following baseline data were also recorded: age, body mass index (calculated as weight in kilograms divided by height in meters squared), apnea-hypopnea index, oxygen desaturation index, Epworth Sleepiness Scale score, and proportion of male participants. We also extracted the mean nightly use of CPAP from trials comparing CPAP with an inactive control.
The primary outcomes were the absolute changes in SBP and DBP from baseline to follow-up in each of the following treatment comparisons: (1) CPAP vs inactive control, (2) MADs vs inactive control, and (3) CPAP vs MADs. For the first comparison (CPAP vs inactive control), we also investigated the association of mean CPAP use with the treatment effects reported in each trial for both outcomes of SBP and DBP. We also explored the association between the reported treatment effects and mean baseline apnea-hypopnea index, oxygen-desaturation index, baseline BP, length of follow-up, type of control group (sham CPAP, no treatment, or other placebo), and type of BP measurement (daytime, morning, office, or 24 hour). Similar investigations for the second and third comparisons were not made due to the insufficient number of studies directly comparing these treatments.
Two authors (D.J.B. and T.G.) evaluated the risk of bias in each study included in the meta-analysis using the Cochrane Collaboration tool for assessing risk of bias.15 This tool assesses studies on different sources of bias (selection, detection, performance, attrition, or reporting bias) and categorizes studies by low, unclear, or high risk of bias in each domain. We then planned to compare the reported treatment effects on SBP and DBP in studies at low and high risk of bias in each domain using meta-regression.
If treatment effects or changes in SBP and DBP in each treatment group were not reported in studies, then the mean (standard deviation) values for each outcome, in each group, and at each visit were extracted and used to estimate treatment effects. To calculate standard error, an estimate of the correlation between SBP and DBP measurements at baseline and follow-up was required. Previous meta-analyses6,7 have assumed this correlation to be 0.5, which might not be appropriate. We instead estimated the correlation in all studies for which it was possible (ie, those reporting standard deviations or standard errors of treatment effects or changes during follow-up and the standard deviations at each visit) and used the mean correlation to impute the treatment effect standard error in studies for which estimation was not possible. To assess the sensitivity of our results to this correlation, we repeated the meta-analysis using the minimum and maximum correlations estimated from the studies. In crossover studies not reporting treatment effects from paired t tests (or tests accounting for the between-period correlation), the between-period correlation was assumed to be zero, slightly increasing the conservatism of the analysis.16 In studies not reporting data on baseline BP, conservative estimates of treatment effects were obtained by estimating the differences between treatment groups in follow-up BP measurements.
Separate meta-analyses of direct evidence only (pairwise meta-analyses) were conducted for each of the 3 treatment comparisons listed above using the metan command in Stata version 14.0 (StataCorp).17 Heterogeneity was assessed using the estimated between-study variance (τ2), Cochran χ2 test, and the I2 statistic.18 Only random-effects models were used to be consistent with the network meta-analysis. To assess the association of trial characteristics and risk of bias with the reported effects of CPAP compared with an inactive control, we used random-effects meta-regression (metareg command in Stata19), and only included studies directly comparing these 2 interventions. To be consistent with the pairwise random-effects meta-analyses, we performed each meta-regression without the use of the adjustment method by Knapp and Hartung,20 but also applied the adjustment in a sensitivity analysis.
Because there are relatively few trials directly comparing CPAP vs MADs on change in BP, a meta-analysis of only direct evidence is likely to lack power. To strengthen this and all other treatment comparisons, we performed a network meta-analysis. Unlike traditional meta-analyses, this method has the advantage of allowing trials comparing CPAP or MADs with some other common treatment (eg, placebo) to be incorporated into the analysis, thus increasing power and enabling a better comparison of CPAP and MADs to be made.13
We used multivariate, random-effects meta-regressions to perform each analysis using the network family of commands in Stata.21 We first fitted a consistency model, which assumes that treatment effects from direct and indirect comparisons are in agreement. An unstructured between-study covariance matrix was used to allow for the possibility of unequal levels of heterogeneity in each comparison. We also performed a sensitivity analysis for which heterogeneity was assumed to be the same in each comparison. To test for inconsistency, design × treatment interactions were added to the consistency model, in which design refers only to the set of treatments in a trial (4 sets in total). To further investigate the plausibility of the consistency assumption, we also checked whether potential effect modifiers were similar across different designs.22 If inconsistency was not rejected, we estimated the probability of each treatment having the strongest association with BP reduction by applying the parametric bootstrap procedure (with 5000 draws), which was described by White.23 Forest plots were used to summarize study level and pooled treatment comparisons and comparison-adjusted funnel plots were used to assess publication bias.24 All analyses were conducted at the 2-sided significance level of .05.
A total of 51 eligible studies (4888 patients) were identified and included in the network meta-analysis (Figure 1). Of these 51 studies, 44 compared CPAP with an inactive control (4289 patients),25-68 3 compared MADs with an inactive control (229 patients),69-71 1 compared CPAP with MADs (126 patients),72 and 3 compared CPAP, MADs, and an inactive control (244 patients)11,12,73 (eFigure 1 in Supplement 2). All included studies assessed SBP as an outcome. One study43 did not report DBP data. Summaries of the characteristics of the included trials appear in Table 1 and Table 2. There was some evidence of a difference in the mean body mass index between trials of different treatment comparisons (Table 3); however, the differences between the weighted means were small. In particular, mean baseline BP was similar across comparisons. Thus, the consistency assumption in the network meta-analysis was not majorly violated in this regard.
The mean between-visit correlation across the 16 trials for which it could be estimated was 0.69 (minimum, 0.47 and maximum, 0.89) for SBP and 0.74 (minimum, 0.53 and maximum, 0.85) for DBP with the mean values being used to impute treatment effect standard errors in 25 studies. No inconsistency was found in the network meta-analysis of SBP (χ23 = 0.54, P = .91) or DBP (χ23 = 2.25, P = .52) and so the main conclusions were drawn from the consistency model. Although these tests for inconsistency are likely to be underpowered due to the small number of studies for some treatment comparisons, the relatively small values of the χ2 statistics indicate that this is unlikely to be an issue in this case.
In the network meta-analysis and compared with an inactive control, CPAP was associated with a reduction in SBP of 2.5 mm Hg (95% CI, 1.5 to 3.5 mm Hg; P < .001) and MADs were associated with a reduction in SBP of 2.1 mm Hg (95% CI, 0.8 to 3.4 mm Hg; P = .002) (Table 4). In the network meta-analysis and compared with an inactive control, CPAP was associated with a reduction in DBP of 2.0 mm Hg (95% CI, 1.3 to 2.7 mm Hg; P < .001) and MADs were associated with a reduction in DBP of 1.9 mm Hg (95% CI, 0.5 to 3.2 mm Hg; P = .008). There was no significant difference between CPAP and MADs in their association with change in SBP (−0.5 mm Hg [95% CI, −2.0 to 1.0 mm Hg]; P = .55) or in DBP (−0.2 mm Hg [95% CI, −1.6 to 1.3 mm Hg]; P = .82).
The findings of the pairwise meta-analyses (Table 4) were similar to those of the network meta-analysis except for the comparison of MADs with inactive controls for reductions in DBP for which a smaller difference of −1.1 mm Hg (95% CI, −2.4 to 0.2 mm Hg; P = .11) was estimated. The results of each pairwise and network meta-analysis for SBP appear in Figure 2 and Figure 3 and for DBP in Figure 4 and Figure 5.
A sensitivity analysis using the minimum and maximum between-visit correlation estimates (instead of the mean estimate) to impute missing treatment effect standard errors did not greatly affect the results of the network meta-analysis (eTables 1 and 2 in Supplement 2). Another sensitivity analysis in which the level of heterogeneity was assumed to be the same across comparisons only led to a slightly smaller estimated difference between MADs and inactive controls (eTable 3 in Supplement 2). This was likely due to the increased weight of the smaller studies directly comparing these interventions, which tended to show treatment effects closer to the null than larger studies.
By applying a bootstrap procedure with 5000 draws to the main network analysis model, the probability of CPAP having the strongest association with SBP reduction was estimated to be 72%, whereas it was 28% for MADs. The probability of CPAP having the strongest association with DBP reduction was 58%, whereas it was 42% for MADs. Comparison-adjusted funnel plots for the network meta-analysis appear in eFigure 2 in Supplement 2. There is a small amount of asymmetry in the plot for DBP.
Mean CPAP use (hours/night) could be obtained from 44 of the 47 studies comparing CPAP with an inactive control. The associations between the mean CPAP use and the treatment effects on BP reported in these studies appear in Figure 6. A 1-hour-per-night increase in mean CPAP use was associated with an additional reduction in SBP of 1.5 mm Hg (95% CI, 0.8 to 2.3 mm Hg; P < .001) and an additional reduction in DBP of 0.9 mm Hg (95% CI, 0.3 to 1.4 mm Hg; P = .001). There was evidence of an association between length of follow-up and the reported effects of CPAP on SBP (slope, 0.2 mm Hg per 1-week increase [95% CI, 0.1 to 0.3 mm Hg]; P = .003) and on DBP (slope, 0.1 mm Hg [95% CI, 0 to 0.2 mm Hg]; P = .006) with the reported treatment effects trending toward zero as length of follow-up increased (eFigure 3 in Supplement 2). Studies with smaller sample sizes tended to report higher mean compliance (correlation with sample size, r = −0.30) and had shorter follow-up periods (r = 0.19). Thus, the reported treatment effect of CPAP was likely to be larger in smaller studies, which may be the reason for the asymmetry in the funnel plots in eFigure 2 rather than publication bias.
There were statistically significant associations between the reported effect of CPAP on SBP and mean baseline SBP (slope, −0.2 mm Hg [95% CI, −0.3 to 0 mm Hg]; P = .04) and between the reported treatment effect on DBP and mean baseline DBP (slope, −0.2 mm Hg [95% CI, −0.4 to 0 mm Hg]; P = .01) (eFigure 4 in Supplement 2). A total of 11 of the 47 studies (23%) comparing CPAP with an inactive control included only patients with some form of hypertension. In a post hoc analysis, we found no difference between the reported treatment effects in this subgroup of trials compared with those with no such inclusion requirement on SBP (−0.8 mm Hg [95% CI, −3.1 to 1.5 mm Hg]; P = .50) or on DBP (−0.6 mm Hg [95% CI, −2.2 to 1.0 mm Hg]; P = .47). There was no association of either the mean baseline apnea-hypopnea index or the oxygen desaturation index in each study with the reported treatment effects of CPAP on SBP and DBP (eFigures 5 and 6 in Supplement 2). Thus, despite there being some evidence of the mean apnea-hypopnea index differing between treatment comparisons (Table 3), this lack of association implies that the consistency assumption of the network meta-analysis was not likely to be violated.
Of the 47 trials comparing CPAP with an inactive control, 22 used sham CPAP as the comparator. The remaining 25 studies used either no treatment (n = 18), an oral placebo (n = 5), a placebo oral appliance (n = 1), or an expiratory nasal resistance valve placebo (n = 1). There were no statistically significant differences in the reported effects of CPAP on SBP or DBP between trials using sham CPAP, any other type of placebo or no placebo as the comparator (eTable 4 in Supplement 2).
Data on daytime ambulatory BP measurements could be extracted from 20 of the 47 studies comparing CPAP with an inactive control, and only 24-hour data could be obtained in 6 studies. A post hoc analysis showed that the association of CPAP with reduction in BP in these 6 studies did not differ compared with those in which daytime BP data was obtained; however, there was some suggestion that the effect of CPAP reported in the studies was larger in those in which morning BP data was extracted (eTable 5 in Supplement 2). Applying the adjustment method by Knapp and Hartung20 in a sensitivity analysis made little difference in the findings of the meta-regression analyses, increasing the standard error of the meta-regression coefficients by no more than 9% in the analyses for SBP and by no more than 12% for DBP.
No more than 5 of the included trials (<10%) were deemed to be at high risk of bias in only 3 domains (allocation concealment, incomplete outcome data, selective reporting) of the Cochrane Collaboration risk of bias tool (eTable 6 and eFigure 7 in Supplement 2). In most domains, the majority of trials were at low risk, except for the allocation concealment category in which most trials were at an unclear risk due to inadequate reporting of methods. Due to either the small number or the absence of any high-risk studies in each domain, reported treatment effects were not compared between trials at high and low risk of bias.
To our knowledge, this is the first network meta-analysis comparing CPAP, MADs, and inactive controls on BP in patients with obstructive sleep apnea. We found that both CPAP and MADs were associated with similar reductions in SBP and DBP compared with an inactive treatment. This is partly in contrast to a previous meta-analysis,10 which did not find a beneficial association with MADs, perhaps due to including only 2 RCTs and thus having inadequate power to detect a difference. Even though there was no statistically significant difference between the associations of CPAP and MADs with change in BP in our meta-analysis, CPAP had a considerably higher probability of having the strongest association with SBP reduction. The associations of both CPAP and MADs with DBP reduction were more similar; however, the association of CPAP with reductions of both SBP and DBP is likely to be greater in patients using CPAP for longer periods at night or in those with higher baseline BP levels.
Even though the results of the pairwise and network meta-analyses were mostly similar, the biggest difference was seen in the comparison of MADs with inactive controls on DBP with the network model estimating a larger association than the pairwise meta-analysis. This was most likely because the data from the direct comparisons of CPAP and MADs tended to favor MADs, and so incorporating these data in the network meta-analysis increased the difference between MADs and inactive controls on change in DBP. In addition, because the difference between CPAP and inactive controls in the pairwise analyses was substantially larger than that for MADs, the comparison between CPAP and MADs changed to favor CPAP in the network meta-analysis (albeit not to a statistically significant extent). The precision of the comparison between CPAP and MAD was slightly lower in the network meta-analyses than when considering direct evidence alone, which may be due to the large amount of between-study variation observed in the indirect comparisons being incorporated into the analysis.
Our meta-regression analyses showed that studies in patients with higher mean use of CPAP or higher baseline BP level tended to report more beneficial treatment effects of CPAP on SBP and DBP. However, it should be noted that meta-regression analyses of mean patient characteristics do not necessarily demonstrate the dose-response relationship at a patient level and so it is possible that the same associations found herein would not have been observed in many of the individual trials.74 For instance, in contrast to our findings, a previous meta-analysis75 investigating the effect of CPAP in asymptomatic patients using individual patient data did not detect a difference between outcomes in patients using CPAP more or less than 4 hours/night compared with controls. Repeating our network meta-analysis using individual patient data rather than aggregate data would improve the assessment of the association of CPAP use and other patient characteristics with each treatment comparison but would be challenging to conduct due to the large number of RCTs from which to acquire data. Alternatively, conducting an RCT in which patients are given various durations of CPAP therapy each night will provide an unbiased assessment of whether there is a dose-response relationship with BP but also might prove challenging to conduct.
Compared with the most recent meta-analysis5 on this topic with similar inclusion criteria, our study includes 18 more RCTs of CPAP and includes at least an additional 2700 patients. Therefore, we had considerably more power to assess the association of trial-level characteristics, such as mean CPAP use, with the reported treatment effects on SBP and DBP. We have also used data from 6 trials comparing MADs with an inactive control, which is considerably more than the 2 trials used in a previous meta-analysis.10 Although only 4 RCTs directly compared CPAP with MADs, we have attempted to strengthen this and all other comparisons by incorporating indirect evidence using a network meta-analysis. Furthermore, in contrast to separate pairwise analyses, we have been able to rank each treatment based on the strength of its association with reductions in SBP and DBP.
A limitation of our meta-analysis is that we only investigated 2 active treatments (ie, CPAP and MADs) and excluded other treatments, such as weight loss interventions,76 which are likely to have beneficial effects on BP because they have been shown to have favorable effects on decreasing the severity of obstructive sleep apnea. However, few trials of other interventions exist and so including them would have increased the sparseness of the network meta-analysis, which can lead to modeling problems, particularly with regard to estimating the between-trial variance of treatments that were not directly compared. Another limitation was that we were unable to extract daytime ambulatory BP data from all studies and thus had to use the available morning, office, or 24-hour measurements. Although this may have increased heterogeneity, it allowed all of the relatively few studies investigating MADs to be incorporated into the analyses. A subgroup analysis showed some evidence that studies from which we extracted morning BP reported slightly larger treatment effects than in other studies. However, there was no difference with studies in which we extracted 24-hour ambulatory measurements and so the effect of any nighttime BP variability was negligible. Future meta-analyses could analyze each BP measurement separately to better understand whether each treatment is associated with greater reductions in BP during the daytime or nighttime.
Our results were robust to the assumed between-visit correlation, which was estimated from parallel trials. However, for simplicity, we did not estimate the between-period correlation from crossover trials (reporting treatment effects from paired t tests) and then use that estimate when calculating treatment effects in other crossover studies. Although this could be deemed a limitation of our analyses, assuming a between-period correlation of zero is arguably reasonable when considering changes from baseline.77 In addition, because only a small proportion of crossover studies were treated this way, we do not believe that our results are sensitive to this assumption.
Among patients with obstructive sleep apnea, both CPAP and MADs were associated with reductions in BP. Network meta-analysis did not identify a statistically significant difference between the BP outcomes associated with these therapies.
Corresponding Author: Malcolm Kohler, MD, University Hospital Zürich, Rämistrasse 100, 8091 Zürich, Switzerland (email@example.com).
Author Contributions: Dr Bratton had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Bratton, Kohler.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Bratton.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Bratton.
Obtained funding: Kohler.
Administrative, technical, or material support: Gaisl, Kohler.
Study supervision: Kohler.
Conflict of Interest Disclosures: The authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: This research was supported by grant 32003B_143365/1 from the Swiss National Science Foundation and by funding from the University of Zurich Clinical Research Priority Program Sleep and Health.
Role of Funder/Sponsor: The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.