"RCTs of Cardiovascular Outcomes in Obstructive Sleep Apnea: Is it Time for an Alternative Trial Design" Ulysses J. Magalang (11.18.2020)

November 20, 2020

ID5915

To CiteDCA Citation Guide

00:00Thank you and.
00:25Alright, I think we're ready
00:27to get started everybody.
00:28Hello, my name is Lauren Tobias
00:30and I'd like to welcome you.
00:32Doris Yale State Sleep seminar,
00:34Yale sleep seminar.
00:35This afternoon I have a few
00:36quick announcements before I
00:38introduce today's speaker.
00:39First, please take a moment
00:41to ensure that you're muted.
00:43Also, in order to receive
00:45CME credit for attendance,
00:46please see the chat room for instructions.
00:48You can text the unique ID for
00:51this conference anytime until
00:523:15 PM if you're not already
00:54registered with Chelsea and me,
00:56you will need to do that first.
00:59If you have any questions
01:00during the presentation,
01:01I encourage you to make use of the
01:03chat room throughout the hour and
01:06recorded versions of these lectures
01:07will be available on line within two
01:10weeks at the link provided in the chat.
01:12Finally,
01:13please feel free to share the
01:15announcements for our weekly lecture
01:16series to anyone else who may be interested,
01:19or contact Debbie Lovejoy to
01:20be added to our email list.
01:22So now I'm delighted to introduce Doctor
01:25Ulysses Magalang as our speaker today.
01:27Doctor Magalong is a professor of medicine.
01:29And neuroscience in the division
01:31of pulmonary critical care and
01:34Sleep Medicine at the Ohio State
01:36University and director of the
01:38OSU Sleep Medicine program.
01:40He is a member of the American
01:43Thoracic Society Scientific
01:44Advisory Committee and a founding
01:46member of the Sleep Apnea,
01:49Global Interdiscipline or
01:50Interdisciplinary Consortium,
01:51which promotes collaboration
01:52between international experts
01:54working in the field of genetics
01:57and genomics of sleep apnea.
01:59Doctor Magalong's is in an
02:02accomplished researcher whose work
02:04examines the effects of intermittent
02:06hypoxia on adipose tissue biology,
02:08particularly its effects on glucose
02:11control and diabetes and atherogenesis.
02:13His funding sources include the NIH an,
02:17the ASM Foundation,
02:18and he has projects including
02:21looking at the genetic,
02:23epigenetic,
02:23and metabolomic basis of different
02:26subtypes of OSA,
02:27and another project looking at.
02:30Transcranial direct current stimulation
02:32therapy for central hypersomnia.
02:33He regularly speaks nationally
02:35and internationally,
02:36and topics including phenotypes of
02:38sleep apnea and the neurobiology
02:41of breathing and I am delighted
02:43that he's here today to give a talk
02:46entitled RCT's of cardiovascular
02:48outcomes and obstructive sleep apnea.
02:50Is it time for an alternative trial design,
02:54and with that I will turn it over to you.
03:00Thanks,
03:00Lauren, good afternoon.
03:01Thanks for inviting me.
03:03So when I first received the
03:05invitation I thought about presenting
03:07some of the animal studies that
03:09we're doing here at Ohio State.
03:12However, you know,
03:13given the audience of this seminar series,
03:15I quickly changed my mind.
03:17So this afternoon we're going to
03:19talk about humans and that rodents.
03:24So this is my first slide.
03:27I have no conflict of interest to
03:31report in relation to this presentation.
03:34Let me start with this headline from 2017.
03:39From CNN Health stating that sleep apnea's
03:43CPAP machine doesn't cut heart risks.
03:47And of course the article is refering to.
03:51The now famous Safe Study,
03:54its largest trial so far up CPAP.
03:59In in cardiovascular disease that
04:02was published in the New England
04:05Journal of Medicine in late 2016,
04:07and the studies show that CPAP did
04:11not prevent cardiovascular events in
04:13patients with moderate to severe OSA
04:16and stab Lish cardiovascular disease.
04:19It did confirm results of prior studies
04:23that CPAP improve daytime sleepiness,
04:25health related quality of life.
04:29And mood.
04:31So in the next 40 minutes or So
04:34what I'm going to talk about?
04:37RCT's cardiovascular outcomes in
04:39OSA and biases in this RCT's.
04:42However, before before that,
04:45I'm gonna that's on a little
04:48bit about OSA disease,
04:50heterogeneity as well as some
04:53of the preclinical and large
04:56epidemiological studies that have
04:59been done and published that you're.
05:03All familiar with just as a
05:06review to put the.
05:08Results of the RCT's in
05:11the proper perspective.
05:13And then finally we will discuss
05:15alternative designs for future studies.
05:18In particular,
05:19we're going to touch on
05:22propensity score matching.
05:24So this is the famous Sir Bradford Hill
05:28who in the 60s published the criteria
05:32for the assessment of causation.
05:35What I have listed here are some of the
05:38criteria as well as the corresponding
05:41types of studies in the right side to
05:45fulfill the criteria for causation.
05:47So mechanistic and preclinical
05:50experiments both in animals.
05:52In humans are typically done to explore
05:56the biologic plausibility of causation.
05:59Epidemiological cross sectional
06:01studies showed the strength.
06:04Consistency and dose response of the
06:08Association while longitudinal studies.
06:11Show that the timing is right,
06:14you know the chronology is right,
06:16and then finally you have.
06:18Of course interventional studies
06:20which can be observation,
06:21ull or in the form of RCT.
06:24So randomized control trials that are
06:26used to evaluate the treatment effects.
06:31As far as treatment trials are concerned,
06:34you know proponents, of course of evidence
06:37based medicine state that there is a
06:40hierarchy of evidence with RCT's and
06:43systematic reviews and meta analysis
06:45occupying the top of the triangle.
06:47And the reason for that is indeed
06:51the quality of evidence shown with
06:55the error on the right hand side is
06:59higher from the bottom to the top.
07:02And mainly because of the effects
07:07of confounding.
07:08As shown on the left,
07:10the increasing arrows on the left hand side.
07:15However, there are situations where
07:18randomization is not possible or ethical.
07:22So for example,
07:24it would be unethical to randomize to
07:27no smoking versus versus versus smoking.
07:31And this is to illustrate this point.
07:34This is an article published in
07:36the Christmas edition of The BMJ.
07:38You know there they are known to publish
07:41this kind of articles around Christmas time.
07:45And this manuscript addresses the issue
07:48that parachutes reduces the risk of
07:51injury after gravitational challenge,
07:53but their effectiveness has
07:56not been proven by RCT's.
08:00So they perform a systematic review
08:02and found of course that know
08:04our cities have been performed,
08:06and they conclude that the basis for
08:09power should use is purely observation.
08:12ULL and it's apparent efficacy
08:14could potentially be explained
08:15by a healthy cohort effect.
08:18That is,
08:18those individuals who jumped from an
08:21airplane without without a parachute
08:24are likely to be mentally unhealthy.
08:27And that individuals.
08:29Who insist that all intervention
08:31interventions need to be validated by
08:33our CPS? Need to come down to earth.
08:36With a bomb.
08:39Of course they could have answered
08:42your question had they included all
08:45observational data and not only are cities,
08:48so it turns out that the US
08:51Parachute Association registers
08:52every single jump from an airplane.
08:57Of course, with the parachute.
09:00And in 2007 there were over 2 million jumps,
09:04resulting in 821 injuries and 18 deaths,
09:08so that's a relative risk
09:10reduction of about 99.9%.
09:13A huge can argue so.
09:15Huge effect size that cannot be ignored.
09:18So I have to be honest,
09:20I hesitated to use example
09:22because now it's probably the
09:24only the only slide that you
09:26will remember from this talk.
09:31But observation ULL study set value,
09:34but they still have to be.
09:36The methods should be rigorous.
09:40And I I would present argument that
09:43perhaps propensity score matching
09:46provides as meta methodology to robustly
09:50assess the cardiovascular benefit.
09:52Of CPAP in in real world patients.
09:57So let's talk about OSA disease heterogen.
10:00Nyati we've been at Ohio State.
10:05We've been actively participating in an
10:07international consortium called Sajik.
10:09As Lauren alluded to.
10:12You know there's there's.
10:14There's two sides in the US.
10:16There's two in Australia,
10:17a couple in Europe,
10:19and then the rest in in Asia.
10:22And one of the object objectives of Sajik
10:25is to establish a large multinational
10:28cohort with detailed phenotyping.
10:31To understand common and ethnicity specific
10:35always say presentations and risk profiles.
10:39So sleep apnea, of course,
10:41is a heterogeneous disease that's that.
10:44Is 2 patients with the same severity
10:47of the condition may present
10:49with totally different symptoms.
10:51And using cluster analysis we published
10:55an article about two years ago showing
11:00that indeed there are different
11:03symptom clusters of sleep apnea.
11:06And they are the consist of
11:10obstructive sleep apnea patients with
11:14predominantly insomnia symptoms.
11:17The typical OSA with excessive
11:20sleepiness as well as the third class
11:24are composed of relatively asymptomatic.
11:27Always say patience.
11:31So what is clustering?
11:32I probably don't need to.
11:35Tell this group about this since.
11:40Claire and doctors in truck.
11:43Actually at Publix up articles about
11:47clustering cluster analysis begins with
11:50a predefined set of input variables
11:54targeted to a specific question.
11:57Example other symptom based subtypes of OSA.
12:02You then apply a clustering algorithm
12:05and an many are available to group the
12:09patients such that within a cluster.
12:13Patients are as similar as possible
12:16and then between clastres they
12:18are as dissimilar as possible.
12:21The clustering method is unbiased,
12:24meaning it is unsupervised
12:27and typically uses the lowest.
12:30Value of the so-called BICR valuation
12:34information criteria to define the
12:37optimal number of number of clusters.
12:41So this table is is busy,
12:43but it's just meant to simply show the
12:47symptom questions and there's a variety.
12:50To define the clusters that
12:53was used in our study.
12:55Shows the characteristic's of the
12:57three clusters with the value side
13:00light that that helped define.
13:02You know this this clusters
13:04in different colors there.
13:09So the first clustering study was actually
13:12done in a clinical population in Iceland,
13:15and the results are shown here
13:18in the on the left hand side.
13:21What is known as the Ice Axe study?
13:26Showing indeed that there are three clusters
13:29and that that's shown on the left side.
13:33And what is not known after that article,
13:36as Publius is that if the classes
13:39are unique to Iceland with its
13:42relatively homogeneous population?
13:45We did find in our paper that the
13:47same 3 classers generalize in an
13:51international sample of clinic patients,
13:54although with some what you know,
13:57different prevalence of the insomnia
14:01and minimally symptomatic groups
14:03as shown in the figure here.
14:07The sleeping group remained
14:08constant at about 40%,
14:10and by the way,
14:11I just want to point out that
14:13the responses that defined the
14:15sleepy group was not solely on the
14:18basis of the Epworth score score,
14:20but that was all that was that
14:23was part of it.
14:27So this three symptoms sub
14:29subtypes are found in both.
14:32In both clinical and community
14:36based samples worldwide.
14:38So that's the original
14:40I sax study in Iceland.
14:42This is our study in Sajik
14:45which we compared to.
14:47Up the the nine Iceland,
14:50we also have a in the paper.
14:54There was a second group of Iceland
14:57patients that basically reproduced
14:59their their their original finding.
15:03And this is been shown also in a
15:07population based cohort in South Korea
15:10as well as in Europe and most recently,
15:15although this is not published yet,
15:18but it's been.
15:21Found and generalized,
15:23the three subtypes generalized
15:26to the Canadian biobank samples.
15:33And most importantly, so this is a.
15:38A study that was published
15:40in the Blue Journal by Diego
15:42Mazzotti out of the pen group.
15:45In 2019, recent analysis.
15:47So this is a re analysis of
15:50the Sleep Heart Health study.
15:53And this indicates that the
15:58increased cardiovascular risk.
16:01Would always say is driven by patients
16:04in the excessively sleepy subtype.
16:07So these are survival plots of knew,
16:11incident.
16:11Coronary heart disease,
16:13knew incident cardiovascular disease,
16:16and knew incident heart failure
16:19and after adjusting for covariates
16:23in the in the adjusted analysis,
16:26it's only this excessively sleepy
16:29subtype that predicted the occurrence
16:32of cardiovascular disease,
16:34and that's perhaps shown better here.
16:38In this figure.
16:40In the sleep part field study there was a.
16:45Another group called moderately sleepy,
16:48but it's the excessively sleepy
16:51subgroup where that had the increased
16:55cardiovascular risk, Interestingly.
16:58Sleepy patients or subjects without OSA?
17:06That wasn't there,
17:07not at risk for future garbage.
17:10Cardiovascular events.
17:12So, just to summarize,
17:14there are three symptom clusters
17:17that generalize the moderate severe
17:20OSA patients in both community
17:22and clinical samples.
17:24The OSA cardiovascular risk comes from
17:27only the excessively sleepy subtype.
17:30And sleepiness in those without OSA
17:33did not increase cardiovascular risk.
17:36Anne Anne it's conceivable that
17:38distinct molecular responses to OSA
17:41result in sleepiness and increased
17:44risk of cardiovascular disease in.
17:47And in certain patients,
17:49to this end,
17:51this was actually the basis of a.
17:56Dot Med grant application between Penn,
17:59Ohio State and University of British
18:02Columbia with innogy bias and Alan
18:05Pack that's looking at the molecular
18:07basis for differences between or say
18:10subtypes because that's not known right,
18:14we just this.
18:16This just got funded.
18:18And we got funded for 3000
18:21samples from patients with OSA.
18:23So basically the idea is a thousands
18:26of samples in its three subtypes,
18:29and the top Med program
18:32doesn't give you resources.
18:35For collection that they collecting
18:37the samples but it does give you
18:40resources for the following whole
18:43whole genome sequencing DNA methylation
18:45patterns as well as metabolomics,
18:48so they'll do that.
18:51Those three things in in
18:55all the 3000 samples.
18:58I believe this is going to be a good
19:01resource because you know that that
19:03Med program releases the data for
19:05two other two other investigators.
19:10So just quickly I'm going to touch on
19:13preclinical and epidemiological studies.
19:16You guys all know this.
19:20Numerous studies looking at biological
19:23plausibility of obstructive sleep
19:25apnea and cardiovascular disease,
19:28just to name a few increased
19:31oxidative stress through impaired
19:33vasoreactivity increase catecholamines.
19:36Increase platelet aggregation and
19:40increase inflammation and this
19:44been shown in many animals and.
19:48And human studies.
19:50They are small, but it does show
19:54a biological plausibility of the.
19:57Of obstructive sleep apnea
19:59causing cardiovascular.
20:01Events just to give you an example,
20:05you guys are very familiar with Seabass.
20:08Apollo skis.
20:09A paper that was published in the Blue
20:14Journal many years ago where he exposed.
20:18C57 Black 6 mice.
20:21Two chronic intermittent hypoxia and found.
20:26In panel D here that if you
20:29combine intermittent hypoxia,
20:31exposure with high cholesterol
20:33diet that the this is sections
20:36of the order that you will find
20:40atherosclerotic plaques or as
20:42all the other groups did not.
20:45This is our own metaanalysis
20:47also from out of the Sajik group,
20:50showing that the effects of CPAP
20:53on blood pressure in patients with
20:56resistant hypertension and the forest
20:58flat shown here shows the results
21:01of the randomized control trials.
21:04On 24 hour systolic blood pressure.
21:08And in this analysis we found that
21:11there is a large decreases in systolic
21:15blood pressure after CPAP use in the
21:19order about 7 millimeters Mercury.
21:25Just to summarize, in the interest of time.
21:30See you all know that large epidemiological
21:33studies consistently find that OSA is an
21:37independent risk factor for hypertension,
21:39coronary artery disease, heart failure,
21:41stroke and death, and death due to CBT.
21:47And that individuals effectively
21:49treated with CPAP have the same rate
21:52of cardiovascular events as age,
21:55sex and weight match controls
21:57with no apnea or snoring.
22:00I'm referring, of course,
22:02to the very famous study of Doctor Marin.
22:07Where he showed that severe always saying
22:11Christmas trees of cardiovascular events
22:13and that CPAP use reduces this risk
22:17because those patients and always say
22:20we'd always say on C pap have the same
22:24cardiovascular event rate as controls,
22:26an inflamed snores,
22:28and the more important thing is
22:31that I believe this is in a follow
22:34up paper where they showed that.
22:37Medication refill rates are similar
22:41in users and nonusers subsea of CPAP.
22:45Suggesting that healthy user bias,
22:47which is of course a big confounder
22:50in observation.
22:51ULL studies does not explain
22:54the observed benefit of CPAP.
22:57So if you then look at Sir Bradford
23:01Hill's criteria, you'll find that.
23:05All of this things had been have been shown.
23:11And except for our cities.
23:13So.
23:16Why is it that the three major are
23:19cities that have been published so
23:22far have been have been negative,
23:24and I'm talking about course the SAFE study,
23:28which is the largest?
23:29There's the re courage to study
23:32and then there's a dissect study
23:35that was published in Lancet
23:37respiratory medicine just this year.
23:40So I'm going to send a review
23:42real real quick.
23:43This three RCP's and we're going
23:45to discuss some of the biases.
23:47That we believe are present.
23:50So the same, of course,
23:53very briefly,
23:54is a study multicenter study of roughly
23:582700 adults with moderate to severe OSA.
24:02And it's they have coronary or
24:06cerebral cerebral vascular disease.
24:09They were randomized to see Pap less use,
24:13less useful care versus usual care alone.
24:16And of course the primacy of the
24:20primary composite endpoint scuse me
24:23was death from cardiovascular causes.
24:26Am I stroke?
24:29Or hospitalization for unstable angina.
24:34Heart failure ortie The mean follow-up
24:37was 3.7 years and the incidence of
24:41the primary endpoint did not differ
24:44significantly in patients who did
24:47versus those that did not receive C
24:50Pap with a hard hazard ratio of 1.1.
24:55And I mentioned earlier CPAP
24:57did improve daytime sleepiness,
24:59health related quality of life and and mood.
25:04The records study was published in the
25:07Blue Journal about four years ago.
25:09The single center RCT.
25:13There's it's a smaller study.
25:15Of course,
25:16there's 244 patients with newly
25:18revascularized coronary artery
25:20disease and moderate to severe
25:22OSA without daytime sleepiness.
25:24So this this patient also had stab Lish,
25:28coronary artery disease,
25:29and obviously they were randomized
25:32to sipat versus no see bat and the
25:35primary endpoint is listed there.
25:38It's again,
25:39it's a composite endpoint endpoint.
25:41Little bit longer follow up of.
25:444.75 years and again,
25:46the incidence of the primary endpoint
25:49did not differ significantly in patients
25:52who did versus those who did not receive
25:55a seat back with a hazard ratio of a .8.
26:02And in the third study,
26:05is that uh, is the Isak study
26:08that was published this year.
26:11It's a multi center RCT.
26:13This patients have were admitted
26:16for acute coronary syndrome.
26:18They were found to have moderate severe OSA,
26:22diagnosed during the first 24 to
26:2572 hours after admission and we
26:28without daytime sleepiness. Um?
26:31Of course you can question you know
26:34there's some data that says that.
26:38When you follow patients where
26:40admitted for acute coronary syndrome
26:43that perhaps there hi changes
26:45but nonetheless that was there.
26:50Entry criteria. Again,
26:53randomized to see that versus know
26:56Steve at about 600 in each arm.
26:59Again, it composite endpoint
27:01that's listed there with a
27:04median follow up of 3.3 years.
27:08And again, the primary endpoint
27:10did not differ significantly
27:12in patients who did versus
27:14those who did not receive.
27:17C pap therapy.
27:20So what are the biases in the in this
27:23published RCT's of cardiovascular outcomes?
27:26In no essay I I'm just going to
27:30touch on a couple. We believe that
27:34there is a sample selection bias.
27:37And and and there are a few things to
27:41consider here. But first thing is,
27:45are they recruited participants
27:47representative of real world and patients?
27:51And we believe the answer to this is no.
27:57Based on the data that I presented to you,
28:00they included.
28:02Non sleepy patients and excluded.
28:06The sleepy patients who are.
28:10The ones.
28:11That are primarily at risk of
28:16developing cardiovascular events.
28:18All these prior our cities were secondary
28:22prevention studies and and really that
28:24was done deliberately because she,
28:27you know they wanted a
28:29higher event rates force.
28:31But one of the downside of that
28:34would be that you know a lot of this
28:38patients were already being managed
28:40actively and they they are on statins
28:44and and perhaps the effect of.
28:47Uh, partly the reason why it's
28:50negative is that the effect of.
28:53Of C PAP may have invented.
28:57The largest issue, we believe,
29:00is that you know where and how
29:04this participants were recruited.
29:07So all these RCT's focus and diagnosing OSA.
29:13Among relatively asymptomatic individuals
29:16with stab Lish cardiovascular disease.
29:19As opposed to identifying adults
29:23with clinically diagnose.
29:25Always saying, then randomizing them.
29:28So there they are not from the sleep clinics.
29:34And we believe that symptomatic the bias
29:38occurs because symptomatic patients are
29:40less willing to be randomized to a study
29:43arm that receives no treatment for an
29:45extended period of time of follow-up,
29:48which is what you need for a
29:51study of cardiovascular events.
29:53Or their providers are less
29:56likely to recommend participation
29:58and such was the expiry.
30:00As in some NH sponsored trials.
30:02So for example, the Apple study had.
30:07You know, according to clip, Kushida had.
30:10Terrible time with recruitment in
30:13the sleep clinics and they had to
30:17resort to really large advertising.
30:20The other trial that comes to mind is
30:24nalaka Gooneratne's memories trial.
30:27Were he actually?
30:29You know providers were not
30:32willing to randomize.
30:34Their subjects with cognitive
30:37impairment if they have they
30:40were found to have sleep apnea.
30:43So we believe that these are
30:45not our patients.
30:46If you look at the exclusion criteria
30:50there that's listed and then the
30:52average on the right hand side.
30:54The. The average effort,
30:59sleepiness,
30:59scale score that all the recent RCT
31:04sub C Pap on cardiovascular events
31:07had had had had the same bias.
31:11And of course,
31:12you know they had to exclude this
31:14patients because it's unethical
31:16to randomize sleepy OSA patients
31:18to no treatment in cardiovascular
31:21trials of seed Bab,
31:22basically because of fear of automobile.
31:26Accidents as well as workplace accidents.
31:32But but the sample bias likely
31:35led to the very low adherence
31:38to CPAP that was reported.
31:41So that's another bias,
31:43because low adherence to therapy would
31:46tend to underestimate the effect size.
31:49And this is the summary.
31:53Of the adherence data.
31:57In in the three trials in the
32:00records actually separated their
32:02users and all patients here,
32:04so all patients here,
32:06these are the CPAP users in
32:09the regards the trial.
32:11The bottom line is after 20 four
32:15months roughly in the range of.
32:192.8 to 3 hours per nine.
32:25And you'd say, well, that's what
32:27you're going to get with C pap,
32:29but but there's some.
32:31Evidence that they may not
32:33be in our patients so.
32:36This is Peter's studies study that
32:39was published in 2019 using big data
32:44looking at CPAP usage in clinic patients.
32:48.6 million patients and and you can
32:52see that that indeed the device usage
32:55is roughly in the area of about 62.
32:59To 70% now, I mean this this study
33:02of course is limited because.
33:05You know these are they didn't.
33:07They didn't include those who did
33:10not drop up or return their seat
33:13that because that's going to be.
33:16They won't have the data and then.
33:20And and and it is.
33:22In addition, this was only the 90 days so.
33:26Off of therapy. But still some some evidence.
33:30Not great that perhaps our clinic patients.
33:33If you if you enroll them in
33:37a in a in a trial of C pap.
33:43Would perhaps use? Their seat belt
33:49more and then you know as I mentioned,
33:52likely some of the selection bias that accord
33:57resulted in the lowest seat belt usage.
34:01The two of these studies in fact did
34:04a propensity score matching in those
34:07who are adherent versus non adherent.
34:10Um? So the Save and Isaac did this
34:15and they got a point estimate of a .8.
34:19I would just point out and this is.
34:23This was I think it was Dan Gottlieb who,
34:26in an editorial in JAMA pointed
34:28this out at this point,
34:30estimates similar to the meta analysis of.
34:34That that's that's off of
34:36our cities Anstatt in trials.
34:41And but at the end of the day, you know this.
34:46This post hoc analysis using
34:48propensity score matching.
34:50Where or underpowered because of
34:53the event rate, the recuts a study
34:58did show that they if you separate
35:03out the users versus non users that.
35:08There was a difference in in
35:11cardiovascular events in a different
35:14versus non adherent subjects.
35:17So the question is.
35:20What are the alternative
35:23designs for future studies?
35:25If you think about,
35:26we believe there are three three
35:29ways of doing this. One is.
35:31We can include the excessive
35:34sleep patients in the trials,
35:37include them in the useful
35:39RCT you know the question is,
35:42is this ethical?
35:44And then there's also the question
35:46of whether symptomatic patients
35:48and their providers agreed to
35:51not being treated for years.
35:53The second one was actually published
35:56and was written by a doctor,
35:59Javaherian colleagues in,
36:01and they suggested that.
36:03Let's do the RCT with pharmacological
36:06management of sleepiness using using
36:10Modafinil. We don't think this to wait.
36:13Wait, are the way to go.
36:16It's probably we.
36:17We believe that using a study design
36:20using propensity score matching that
36:23allows the inclusion of excessively
36:26sleepy or Safeway patients.
36:28Most likely to show a cardiovascular
36:31benefit from CPAP and not only
36:33that because you're in a propensity
36:36score design and real world patient,
36:39you're going to compare users
36:41versus non users.
36:42You could examine the true benefit
36:45of C pap therapy on cardiovascular
36:48outcomes within real world clinical patients.
36:52And this is the the paper that I
36:55was alluding to that was published.
36:58They estimate that they would
37:00need a sample size,
37:02about 24,000 with 12,000 in each arm.
37:05Using pharmacological management of.
37:07Of sleepiness.
37:10This is the way we think that
37:14this the using propensity score.
37:17Should be done.
37:19You know you have include.
37:21Subjects who are seen in the clinic.
37:24So you have the inclusion criteria there.
37:27Of course there will be sleepy subjects
37:29based on the sleepy subtype and they will
37:32be treated with CPAP in all the patients.
37:35But the most important thing there
37:38in any propensity score design is to
37:41obtain the covariates and I'll explain
37:43that in a little bit and then you can.
37:46You can then compare those who are adherents.
37:50Versus those who declined
37:52therapy or non users.
37:53You could define this as less than
37:56two hours per night or you could say
37:59less than one hour per night without
38:01using the without CPAP you states
38:04in the last 30 days and then you
38:07can do a propensity score design.
38:11With an annual follow up of CPAP adherence,
38:15an major adverse cardiovascular events.
38:19For for a number of years.
38:22So this is crucial for any PS design study.
38:26You need to include a rich
38:28set of clinical relevant,
38:30clinically relevant covariates.
38:32Associated with.
38:33Basically we do things,
38:35the CPAP adherence and the outcome,
38:38and this reduces the bias associated
38:41with observed and unobserved covariates.
38:45And just in the interest of time,
38:47I'll you know these are the useful things.
38:52That we would think would be important
38:55as predictors of CPAP adherence.
38:57Including educational attainment.
39:01Social economic factors.
39:04Insoft,
39:05presence of insomnia and psychological
39:09problems but also include measures of.
39:14Self efficacy as well
39:16as medication adherence.
39:21The predictors of the events obviously
39:24are the useful things that we consider.
39:29Gender, obesity, prevalence, CVD, smoking.
39:34Lipids, family history and
39:37physical activity as well as Dyett.
39:41Assessment so what's propensity score?
39:47So the definition,
39:49the PS is the probability.
39:52Or being in the treated group conditional
39:56on all relevant baseline covariates.
39:59And at here's the.
40:00Is the formula there and basically
40:03it says that given two subjects with
40:06identical values of your propensity score,
40:09one from the treated group and
40:12one from the control group.
40:14If it's the same then analysis may proceed
40:17as if the subjects were randomized.
40:20And of course the key assumption is that no,
40:24there are no observed confounders.
40:27There's three types of PS design.
40:30She used stratification by PSR
40:32subclasses at one to one matching or
40:35there's a technique called inverse
40:38probability of treatment waiting,
40:40but the fondle fundamental considerations
40:43of this science is that the outcome days
40:48that data is not used in the PS design.
40:51So in regulatory studies.
40:55So FDA actually uses these two.
40:59To make a decision whether to
41:02approve surgeries or devices.
41:05It must be documented that the PS design.
41:09Start decision had no access to the
41:13outcome data and therefore the PS
41:16design faces a second design phase.
41:19Very briefly, this is just a schematic.
41:23You perform a observation,
41:25ULL study and you have the developed
41:29propensity scores using this
41:31techniques and then at the end of the
41:36day you got PS based matched pairs.
41:39So this is nothing new.
41:42Independent group has used this to assess
41:45CPAP treatment and fasting lipids.
41:48For example,
41:48and you'll see this is known as the lab plot,
41:53and here are the cold marriage
41:56and the PS design sample.
41:59As you can see,
42:01simulates that of us if you've
42:04done a randomized control trial.
42:07So we believe that the benefits
42:10of a PS assign.
42:13And obtain valid estimates of causal
42:15treatment effects in observation.
42:17ULL Data Bay creating covariate
42:19balance similar to or even better
42:22than under randomization.
42:24You can use real world patient data
42:27that is often not well represented in
42:31those that you choose to be randomized.
42:35You can include patients that cannot be
42:38otherwise ethically be randomized in RCT's,
42:40and you can evaluate benefits of
42:43treatment efficiently in larger samples.
42:45Because this is a pragmatic
42:47trial so you can just, you know,
42:50you can easily insert this within
42:53the context of clinical practice.
42:56And so,
42:57while an RCT provides the preferred
43:00level of evidence in ideal world,
43:03PS designs can achieve the same
43:06level of evidence.
43:08For treatment effects in the real world.
43:12And you know,
43:13I,
43:14I certainly am not an expert on the
43:16propensity score matching at the sign.
43:19Greg Maislin in our group is the one that.
43:24That that has worked with Donald Rubin,
43:27who is the inventor of the
43:30propensity score matching,
43:31and this this manuscript.
43:33He did a good job in explaining this.
43:37If you're interested, there's a
43:39recently accepted paper in sleep.
43:44That was accepted just.
43:47Last last week I believe,
43:50where he explained in detail more
43:54the propensity score matching.
43:57So the proposed clinical trial would
43:59be a multi center RCT of patients
44:02with moderate to severe severe OSA.
44:05We believe we can do this with
44:08either 10 or 1310 to 13 sites you
44:12offer seat up to all patients.
44:16The primary would be similar to the
44:18same a composet endpoint follow up of.
44:21Two to five years.
44:24And we believe that we with 11,000 subjects,
44:27and that includes additional 10% to
44:29maintain power after loss to follow
44:32up or trimming of patients in the
44:35PS design that you could do this.
44:37Now you say, well, that's a lot of subjects.
44:41We actually did a.
44:42So if you look at the number of subjects.
44:47We included this data in in a
44:50recent grant that we submitted.
44:52The total here is like this is the
44:55annual number of subjects in the centers
44:58and you have 7 to 6000 potentially.
45:04And rollable patients.
45:05So we believe that we could
45:08we could do this study.
45:10It's going to be a heavy lift. We we, we,
45:15we we think but but it's worth trying.
45:18So to summarize.
45:21Get few minutes for questions.
45:24Sleep apnea is heterogeneous disease symptom
45:28clusters of those with daytime sleepiness,
45:32insomnia, and asymptomatic groups.
45:34Are consistently shown in community
45:38and clinical samples worldwide.
45:41It's important because EDS we
45:43believe is a marker of cardiovascular
45:45risk in in those with OSA,
45:48but not in those without always say.
45:52And Publix are cities of cardio
45:55cardiovascular outcomes in OSA
45:57have been negative and inconsistent
46:01with the large epidemiological
46:03data because of major biases.
46:06That's primarily the sample selection bias
46:10and bias due to adherence to therapy.
46:14In future studies need to include
46:17and focus on sleepy subjects.
46:20Ethical lamp limitations,
46:22including this patients can
46:24be overcome with observation.
46:26ULL designs using propensity scores an
46:29to obtain a robust treatment effect.
46:33This designs need to directly
46:35ensure balance of covariates related
46:38to cardiovascular events,
46:40including measures of healthy
46:42used userin healthy adhere bias.
46:45In patients who are very compliant
46:48seat back compared to non users
46:50and I'm going to stop there.
46:53Thank you.
46:55Thank you so much Doctor Magalong,
46:58that was really a fantastic talk
47:00and I think really help to clarify
47:03some of the the residual questions
47:06that a lot of us had about how we
47:09should be characterizing the benefit.
47:12The cardiovascular benefit of CPAP
47:14for patients after these these recent
47:16trials I want to invite people to
47:19unmute themselves and ask questions.
47:21I expect there probably are some. Not
47:25really, I was going to say I'm not
47:27sure I have access to the chat room,
47:30but you could just tell us up, Garth. How
47:33are you? Thank you so much.
47:36That was a really thoughtful presentation.
47:38I'm so sorry we can't
47:40have you here in person,
47:42but we really appreciate you making
47:45the time and congratulations on the
47:47top Med project and I, you know,
47:49I agree with with so much of what
47:52you were saying and I think the
47:56propensity score matched approach
47:57is a great is a great idea and I
48:00I think I also want to emphasize.
48:03A point that you made which is you know,
48:07the trials that have been the three
48:10trials that you referenced that
48:12would really have been done to date,
48:15and I think we're really in the
48:18the infancy of doing randomized
48:20control trials in our field compared
48:22to the size of the trials that
48:26typically occur in cardiovascular
48:28disease are tiny and with so many
48:32pharmacological treatments available.
48:34That that actually reflects some of the
48:36biologic pathways by which sleep apnea
48:39can lead to cardiovascular disease.
48:41You really need so those large sample
48:43sizes to to demonstrate an additional
48:46benefit associated with CPAP therapy.
48:48But I think one point I would add
48:51is that I think the outcomes may
48:54be also different depending on the
48:56cardiovascular event that is chosen,
48:59and I think save may have pointed
49:02to this a little bit.
49:04Some of our studies and stroke have
49:07suggested this as well that there
49:09there may be a more robust affect in
49:12stroke for some reason compared to MI,
49:15and I think some of the observation.
49:18ULL data support that but.
49:21Another another approach I think to
49:24doing a randomized controlled trial.
49:27We've done is is more of a
49:30comparative effectiveness approach,
49:32and so you're not randomizing a
49:34patient that you have diagnosed
49:36with sleep apnea and not treated,
49:39but but rather randomizing to a diagnosis
49:42and treatment intervention strategy,
49:44trial versus the usual care approach,
49:47and I think that that might help
49:49to get through some of the ethical
49:52challenges and could be a potentially
49:55useful strategy in a very high
49:58pretest probability population.
50:01Thank you Clark.
50:02With it, you know I just didn't have
50:04the time to to go into those details,
50:07but that was those points.
50:09Your point about Cerebro
50:11vascular disease versus.
50:12You know, ameisen all those
50:15those kind of events?
50:17Certainly there is data to suggest
50:20that you'll have probably a greater
50:23effect on cerebral vascular effect
50:26events and and and the other issue
50:29of doing a comparative effectiveness.
50:32I didn't list it here,
50:35it was actually in the paper.
50:39Potentially you could say,
50:41well, let's do an enhance.
50:43Add CPAP adherence so that that way you can
50:48have a separation between with usage right.
50:52We believe that they may.
50:54That might actually affect the sample size,
50:57and you're going to because
50:58it's you're going to.
51:00You're probably going to need.
51:03A very large sample size,
51:04if that's the approach that you're going to.
51:08That you are going to take.
51:12But but those are very good points.
51:19Can I hire
51:20lease is high
51:21High made Nelson? How are
51:23you good? Thanks oh that was
51:25a great insightful talk.
51:27I'm just going to ask
51:29it kind of
51:30a different question.
51:31We're going to treat all patients with OSA
51:34that are sleepy because we have no
51:37other better treatment than CPAP.
51:40If that's a statement,
51:41then who cares about whether
51:43CPAP is going to reduce or not
51:46reduce cardiovascular events? OK,
51:48so the question is the
51:50non sleeping group that we
51:51don't really have the full
51:53confidence that whether they do or
51:56they do not have that
51:58increase risk. And that's the
52:01tough rope to trade with something
52:03like super, which lends itself to
52:06suboptimal adherence on
52:07a long term
52:08basis. How we gonna actually.
52:11Answered that question.
52:14Well, to the point of so the the first
52:16point or question is where are you
52:19going to treat this patient's anyway?
52:21Because they're sleepy is that is that,
52:23is that correct? Well, you know,
52:26we believe that there is a reason
52:29and one of them there are other.
52:31You know. There are several reasons,
52:33but the major one is that.
52:37You know right now I should know.
52:42Screening for or identifying.
52:46UH, patients, for example,
52:48a large scale in primary practice
52:51is is not recommended, right?
52:54So we believe that showing
52:57that sifat indeed impacts.
53:00On whether sudrow basket
53:03or or cardiovascular event
53:05would would sway you know.
53:08A people too.
53:13To identify more cases of sleep apnea and
53:17perhaps towards towards screening, although
53:20that's a different entirely different topic.
53:24The other thing is, as in other studies.
53:29That show that you know physician
53:32advocacy of treatment. For example,
53:35if if if they know that the treatment
53:40makes a difference, they would indeed.
53:44Outside of the excessive daytime sleepiness,
53:48they would indeed encourage identification of
53:51patients as well as US treatment of patients,
53:55in that I think that's well known in
53:59the in the cardiovascular literature.
54:03Your second point is about the non sleepy.
54:09Patients out how we're going to,
54:11how we're going to treat them. I.
54:16It's. I mean, that's as far as there
54:21are others who will argue with you.
54:23That if they are asymptomatic.
54:29Up at the present time,
54:31there is no rationale to treat them.
54:35I mean, I know,
54:37I know that's probably a very controversial
54:41statement given some of the guidelines.
54:46About at least the data that we have.
54:52In the Sleep,
54:53Heart tells Saudi and of course
54:55that needs to be replicated.
54:57It's it's.
54:58It's actually only the sleepy
55:00group that was that was at risk,
55:03or at least that was what
55:05was shown by the panel group.
55:08Yeah, the problem with the Epworth,
55:10which we use all of us use for assessing
55:14subjective sleepiness is very,
55:15very susceptible to
55:17false negative scores.
55:18Yeah, I pointed that out.
55:20I specifically said that actually
55:22that the subtype of sleep apnea is
55:26not only does that only include.
55:28The The Epworth Sleepiness Scale score.
55:31So determining those subtypes is actually
55:34there are other questions that were included
55:37that although it's it's the F word,
55:40was a component of defining the
55:42sleepy subtype. But it's it's not.
55:45It's not the F word. Alone.
55:50That defines the sleepy subtype.
55:52At least you know in, in, in,
55:55in the papers that we have
55:58established what we have popped. It
56:00have worked their real world situation.
56:03We use Epworth Aurora comperable type
56:05of a self administered questionnaire
56:07as opposed in
56:09a research based type of tools.
56:11So identifying those people with or without
56:14sleepiness is going to be
56:16prone to bias against or in favor of
56:20selecting people for
56:21treatments. Right after that,
56:24and then we actually so Brendan
56:28Keenan at Penn actually has created
56:32a so based on the on the studies
56:36that we publish it is there is a an
56:41app Web type app that you could.
56:46Plug in the answers to the
56:48questions and it will give you.
56:51The answer whether that patient
56:53belongs to a sleepy subtype,
56:55but you know whether that lends itself
56:57to the usual busy clinical practice.
57:00I I I agree with you. Yes,
57:05so why aren't we using objective
57:07measures of sleepiness?
57:09I mean, there's a big literature showing
57:12that subjective measures are terrible.
57:14An an an an so that's like, uh,
57:18that's that's a real problem and
57:20I think the other problem in a
57:22lot of these studies is that they
57:25are studying patients too late.
57:28So in the safe trial,
57:30the average patient was over,
57:32you know, 61 years old.
57:34By then the patient,
57:35already his cardiovasc he or
57:36her cardiovascular system,
57:38is already really abnormal.
57:40And for example,
57:41in in an art clinic in Canada,
57:44our average patient was 48 years old.
57:47And and and and and at the age
57:50of 48 they had already had
57:52symptoms for like 5 to 10 years.
57:55They already were very heavy
57:58users of health care resource
58:00is for five to 10 years,
58:01and that's the group that
58:03we ought to be studying,
58:05not the ones that already
58:07have a bunch of diseases.
58:10Yeah, that's certainly true.
58:12I mean, again, that May contributes
58:14to remember these are all
58:16secondary prevention trials, right?
58:19They had to have.
58:22CVD in order to be enrolled in in
58:25the in the in the Safe study and
58:28the other ones are they had acute
58:31coronary syndrome and then the
58:33the other study they you have to
58:36have a cast proven coronary artery
58:38disease and and I agree with that.
58:41Perhaps you know the the.
58:44Although I think the entry criteria
58:46of the age is about is is 18,
58:48but you're saying that the.
58:50The the the average age is there,
58:53they're older, they're older.
58:55Yeah, I mean it. It reminds me of
58:57the Women's Health Initiative study
58:59where the you know they were giving.
59:02They were treating women.
59:04You know, for menopause like 15 years
59:06after their menopause, Ann and Dan.
59:09And that's you know. In other words,
59:11we're treating patients way too late.
59:14We ought to be screening them earlier,
59:16and that's where I think the
59:19RTC should focus an in fact.
59:21There are several studies early,
59:23you know years ago that showed
59:25that that that the mortality
59:26of patients with sleep apnea,
59:29the older patients actually don't do so bad.
59:31You know it's the younger ones that have
59:34that seem to have the higher mortality.
59:37Yeah, that's because of this.
59:40Basically as survival effect,
59:42right? Yeah, yeah. Ulysses
59:45I have a question.
59:46This is Nancy Rediker High we met. I think
59:49that grant reviews.
59:50Hi, my question is about the
59:52mechanisms of the sleepy patients and
59:54CVD. So you've
59:56mentioned this study about the
59:57looking at genetics and and what
59:59could you let? You know there's,
01:00:01so there's obviously all different
01:00:03kind of genetic pathways,
01:00:04but is it
01:00:05possible that this is just,
01:00:07you know, the sleepy patient,
01:00:08it's just it's inflammatory or
01:00:10it's some other underlying process.
01:00:11It's causing the connections,
01:00:12so that's really just an
01:00:14epic phenomenon that there's
01:00:15inflammation going on anyway,
01:00:17or it matches what
01:00:18I'm guessing, but
01:00:19what? What kind of
01:00:20genetic pathways are you looking at?
01:00:23Well, that that that Grant is, we don't know.
01:00:26Basically, you see that it's it's there,
01:00:29but there are possible mechanisms and
01:00:31and the number one suspect will be.
01:00:34Of course what you mentioned is inflammation,
01:00:37right? There is some evidence of
01:00:40inflammation activity may cause
01:00:42you to be to be to be sleepy.
01:00:44Now to the point of.
01:00:48The PV, the objective evidence of sleepiness.
01:00:54We could potentially add results of,
01:00:56although it's not really sleeping
01:00:58as its vigilance would.
01:01:00That would be easy to incorporate,
01:01:02would be psycho motor vigilance
01:01:04testing for example.
01:01:05That might be.
01:01:07That that might be a that might
01:01:09provide really confidence on the
01:01:12defining the sleepy subtype.
01:01:14One Pvt is so easy to do now.
01:01:18I mean, we could do
01:01:20it on an iPad.
01:01:22We don't need a special device.
01:01:24Yeah, it feels like something
01:01:26that could readily be incorporated
01:01:29into clinical encounters.
01:01:32It is just going back to the to the
01:01:34top match of the the way that would
01:01:37that Grant was structured was that.
01:01:39We you know, it's basically
01:01:41we're going to do a whole genome.
01:01:44All all the mix an all the all the
01:01:48epigenetic things and see if there are
01:01:51any differences in the in the subtypes.
01:01:55Of course, when when the data is
01:01:58published in publicly available,
01:02:00there's a bunch of things that you
01:02:03could do with that with that data.
01:02:08Thank you so much for the accounts.
01:02:10I think as there are a few minutes
01:02:12past the hour and people hung around
01:02:14because this is such a compelling topic,
01:02:17but we should still cut it off here and thank
01:02:19you again. Needless yeah.
01:02:21Thanks for inviting me,
01:02:22I appreciate it. Thank you. Thanks.
01:02:26You like Michelle? Take care.