Identifying Adolescents at Risk for Substance Misuse Using Digital Tools: Why, Where, When and How

February 23, 2023

Information

YCSC Grand Rounds February 21, 2023

Kammarauche Aneni, MD, MHS, Assistant Professor of Child Psychiatry, Yale Child Study Center

ID9551

To CiteDCA Citation Guide

00:00Doctor Anani who used to be Doctor Isuzu
00:03but now she has her taken her married
00:07name and Annie is a good friend and and
00:10really one of our someone we're so proud
00:13to have as a trainee and now on faculty.
00:17Uche went to medical school in her
00:20native Nigeria, the University of Ibadan,
00:22and then in the states.
00:25She's been winding her way up
00:27the eastern seaboard,
00:28from Miami to Duke to Johns Hopkins.
00:35All of them are a bunch of losers
00:36because she ended up here with us,
00:37so we're delighted.
00:39And Uche has worked with a lot of the
00:42people in the Who's who of psychiatry,
00:45including I was just looking at
00:47her CV again with Doctor Charlie
00:49Nemeroff when he was in Miami.
00:51And it's just remarkable looking at your CV,
00:54which how quickly you've been publishing,
00:57getting grants, coming into your
00:59own right as an investigator,
01:01which is what you're going to show us today.
01:05Among her many interests,
01:07Uche has done some really important
01:10work on race, racism, race relations,
01:13including here in our inpatient unit.
01:17I was joking that I had the
01:19privilege of publishing, I think,
01:21Doctor Isuzus last paper,
01:23because Doctor Azusa is now Doctor Anani.
01:25But that paper is really important
01:27with many of our colleagues,
01:29including David Rees and Laurie Cardona,
01:31who's here in Amanda Calhoun,
01:33a qualitative study about racism.
01:35Race relations in an inpatient unit,
01:38a pretty brave type of work
01:41that is much needed.
01:43And Uche has moved into the digital
01:46world in a in in a major way.
01:50She is now part of a number of
01:53consortia and grants all together
01:55with the digital team here.
01:58She is now the associate director of the
02:00play to Prevent Group that we're going
02:02to be hearing about that brings technologies.
02:05To clinical needs and
02:08using all sorts of G Wiz.
02:11You know,
02:12toys towards very important purposes.
02:15So it's wonderful to be here.
02:17Oh,
02:18the most important thing is that I
02:19understand because I saw the list
02:21that her husband is watching you,
02:22so you better do a good job.
02:24And but the most important thing is
02:26that that which is also the proud mom
02:28of two beautiful kids who we have been
02:30seeing growing here, Daisy and Damien.
02:33So would you take it away?
02:44Thank you for that kind introduction.
02:49Umm.
02:54Ohh sorry. 2nd.
03:03All right. So I'm excited to talk to
03:06you all about my work investigating
03:09the utility of digital tools,
03:12not only for risk assessment,
03:13which is the focus of this talk,
03:15but also for intervention
03:18development and delivery,
03:20specifically focused on addressing
03:22adults and substance misuse.
03:25And all of this really is framed.
03:27Under the umbrella of increasing access to
03:30care, I think a lot about access to care.
03:33I grew up in a country where access
03:35to care was pretty challenging,
03:37and so I thought a lot about it growing
03:40up also as a resident and trainee,
03:44I was also really struck by how delaying
03:47access to care really leads to adolescence,
03:52presenting very late in disease,
03:54but also how difficult it is
03:56to manage when kids present.
03:58So within that framework is where
04:00where I where I land and how I
04:03will be framing the talk today.
04:05And I have an outline thinking
04:07together with you all about why,
04:09when,
04:09where and what are the barriers
04:11with identifying kids early.
04:13And I will focus mainly on 2 digital
04:15tools which have which I'm working on,
04:18which is game based and the electric the
04:20use of the electronic health records.
04:24And so when I say substance misuse,
04:25I mean unhealthy substance.
04:27Use of alcohol or drugs to relieve stress,
04:31alter reality or bring about pleasure.
04:33Using any way not prescribed by a doctor.
04:35Use without one's own prescription,
04:37using greater amounts, small often,
04:38or longer than told to take.
04:41And this affects 3.7 million adolescents,
04:43as is the national Survey of Drug
04:45Use and Health, which equates to
04:46about one to two in $10 cents.
04:48So it's common, although the average
04:50age of onset is about 15 to 17 years,
04:53it can occur earlier.
04:54And we know that younger age of onset
04:56is associated with a greater likelihood
04:59of developing a substance use disorder.
05:01Outcomes are worse.
05:03The consequences are dire.
05:04So in the last two years we've seen
05:07we're dealing with a crisis, right?
05:09And drug overdose rates has
05:11risen by 1 / 100%.
05:13If we look at the media
05:15monthly overdose deaths,
05:16it's associated with overdose,
05:17which can happen at the first time,
05:19the first time someone misuses a substance
05:22associated with comorbid mental disorders.
05:24Both as as increases the
05:27risk for substance abuse,
05:29but also as a consequence and
05:31of course associated with.
05:33Or functioning like school dropout
05:35legal problems for relationships.
05:37Also,
05:38the time between when disease staff
05:42seems sad and initial treatment seeking.
05:45Initial treatment seeking is on average.
05:48This study done by Tesla,
05:49which has reached really fund did they lead?
05:52The National Comorbidity Survey was 16 years.
05:57And I know you will agree that that's
06:01that's unacceptable and it underscores the
06:04importance and need to identify people.
06:07Adolescence, early substances used
06:09typically occur starts in adolescence,
06:11which is which is why the focus.
06:14Also less than 10% in need of
06:16treatment receive it in 2021.
06:18The number for adults since
06:1912 to 17 years was 3.5%.
06:22So.
06:23Adolescent substance misuse is a major
06:25public health problem problem and most in
06:28need of treatment are not receiving it.
06:31If we've got to think about how might we,
06:33you know, start to solve this problem,
06:35one of the things that we
06:36might think about like where,
06:37where,
06:38where are adolescents?
06:4196% are enrolled in schools,
06:43not necessarily attending schools,
06:44but I6I enrolled in school,
06:46so there might be a way to
06:48like engage contact them.
06:5095% have access to a smartphone,
06:5291% are seen in primary care annually.
06:5590% are lined daily and 90% play video games.
06:59So, umm, I think we need a systemic
07:01model for addressing substance misuse.
07:04And what I mean by systemic model is a model,
07:07a national model that
07:08screens all adolescents,
07:10triages them based on their risk of use.
07:12No substance misuse,
07:13and so therefore it needs primary prevention,
07:16at least substance misuse,
07:18secondary prevention,
07:18substance use disorder treatment. Umm.
07:22And I think what digital tools can do,
07:28we'll talk a bit about that,
07:29how they can help with these.
07:30But one of the questions that come up is,
07:32well we already have a lot of
07:34kids who are struggling and we
07:36don't have enough providers.
07:37So this is only going to increase the
07:40number of kids who who were identifying.
07:42But I I would say that I think that
07:44the fact that we're not identifying
07:46them early is contributing to the
07:48number of kids that we're seeing
07:49who are really ill and at that time.
07:52It's really difficult to address symptoms
07:55as opposed to like if you catch them early.
07:58One of the systematic reviews we did earlier,
08:00some years ago,
08:02showed that interventions digitally
08:04delivered interventions that deliver
08:06universal or targeted interventions were
08:08actually more likely to be effective,
08:11which I think speaks to other
08:13findings from other studies that
08:16data interventions can adequately
08:18target adolescents who have present
08:20with mild to moderate symptoms.
08:23As opposed to like severe disorders
08:25which need which you you need more than,
08:28you know, standalone digital intervention.
08:32And so, you know,
08:33the car as opposed to sustained model,
08:36what we do, what we have,
08:38where the model that we have is
08:40routine recommendation by Samsung,
08:41the American Academy of Pediatrics
08:43to screen routinely at annual
08:45primary care visits.
08:47And the way this happens is you,
08:49you know through the expert
08:50model where you screen,
08:51you have a proof intervention
08:53and then you refer to treatment.
08:54All of this is required.
08:56The primary care provider does this.
08:58Some systems do differently where there's
08:59someone in there who can deliver.
09:01Of intervention, but many,
09:02many times it falls on the shoulders
09:04of the primary care provider.
09:05There are no universal screening in schools.
09:07Schools do schools once,
09:08but they do whatever they can.
09:10It's not systemic at all.
09:13So what are the barriers?
09:15Many providers are not screening
09:17for substance misuse.
09:18There are time constraints,
09:20there's lack of reimbursement.
09:21It seems that every time we want
09:23to do something else that is
09:25prevention or preventive wise,
09:27we add some one more thing that
09:29primary care providers have to do.
09:31And so primary care providers have
09:32to triage and decide what they're
09:34going to prioritize in their visit,
09:36which might be appropriately informed
09:39by the presenting complaint and so.
09:42Provide us some providers have
09:44reported lacking knowledge
09:46about what screens to use the system.
09:48The workflow is in there to
09:50actually make this happen.
09:51What are the resources?
09:52I don't have resources of identifying
09:55how to figure out what to do.
09:56Occurrence not always aware of of of
09:59substance misuse among their adolescence.
10:01Adolescence are worried about disclosing
10:02this for the first time with their
10:05parents being present so they're not
10:07always forthcoming for many reasons.
10:08The stigma there's no privacy in the clinics.
10:12Umm, and so I'm saying I'm suggesting
10:14that you talk to us, mere Canadians,
10:16why some of these barriers?
10:18There's wide reach.
10:19You can automate this.
10:21You can alleviate provider burden.
10:23This can happen at any time.
10:26The study by night at all showed
10:28that adolescents are more honest
10:29about the substance misuse when
10:31completing digital screeners and
10:33paper screens compared to interviews,
10:35and there's potential for electronic
10:37health record integration.
10:39We could potentially build an
10:41automated system that identifies
10:42risk as it as it emerges,
10:44which would be ideal.
10:46Funnel appropriately,
10:47deliver an intervention if it's
10:48a standalone for multi moderate,
10:50figure out a way to deliver or funnel
10:53to to treatment for those who need.
10:56Specific treatment by a trained personnel.
10:59So what might an ideal digital
11:02Screener look like?
11:03I would suggest that it would be up,
11:05you know,
11:06objective as opposed to self report
11:07will be effective at identifying
11:09what you're trying to measure.
11:11So substance misuse in this case.
11:12It will be scalable and it will
11:15be accessible at all times by
11:17whoever needs needs to access it.
11:20So my my proposition is that games may
11:25offer an objective and effective method
11:27for identifying at risk adolescents.
11:29And this is based on the premise
11:32of of metrics which experts call
11:34digital biomarkers that are that are
11:37captured by the by the game software.
11:39And I like this definition of game
11:42digital biomarkers which is which
11:43defines it as consumer generated
11:45physiological and behavioral measures
11:47collected through connected digital
11:49tools that explain influence or
11:51predict health related outcomes.
11:53So one example that has been used now
11:56is being used is motor performance
11:58in games and the identification
12:00of Ellie Parkinson's. Umm.
12:02And so could we do something similar?
12:08Games collect a lot of metrics and
12:09we'll talk a little more about the
12:11work that I've been doing around this.
12:13But games collect a lot of metrics,
12:15and some of the metrics for example
12:17are like time to complete a task,
12:19accuracy of choices.
12:20Those task may be informed by
12:24cognitive processes such as maybe
12:26working memory or inhibitory control,
12:28which we know are implicated in development
12:31of substance misuse and are also in
12:35in impacted by misuse of substances.
12:38And so if we're going to use games to
12:40measure cognitive function or identify
12:42kids who are at risk for substance misuse,
12:45are they valid?
12:46So we did a systematic review and meta
12:48analysis to assess the validity of game
12:51based assessments of cognitive function.
12:53This has been accepted in
12:55progress in brain research.
12:57We looked at studies examining game
12:59based assessments among children
13:01and adults and zero to 17 years,
13:03but four major questions
13:05general game characteristics,
13:06cognitive functions that were measured,
13:09how, what was the validity
13:10and how did they compare.
13:11They used in the studies used pairwise
13:13correlations and we're looking at factors
13:15that may influence the validity of.
13:17Game just assessments would define
13:20validity by criterion validity.
13:22So how well does a new measure
13:25compared to an to a previously
13:27validated measure using you know?
13:30Here using pairwise correlation,
13:32so pairwise correlations between games
13:33that measured specific cognitive
13:35functions and traditional assessments.
13:37We did a meta analysis of specific
13:39to these games had many tasks and
13:41so we only did a meta analysis of
13:44games that measured specific tasks.
13:46So an impact task for working
13:48memory to a traditional and back
13:51task measuring working memory.
13:52We're also interested in things that would
13:55affect validity and we organized this
13:56around things at the person level like age,
13:59sex, race,
13:59how the game was delete,
14:02where the game was delivered,
14:03home school clinic,
14:04and how the how the game itself.
14:07Operated one of the things I'll talk
14:09about the scarring analytical method.
14:11So you could you could you could
14:13do a task and basically measure
14:15for the impact test for example
14:17how many errors or go no go task,
14:20how many errors were made when
14:21this person was doing the test.
14:23As opposed to like using collecting
14:25all the metrics in a game and
14:28using an analytic machine learning
14:30predictive model to predict cognitive
14:32function for example.
14:34So we extracted all of these metrics,
14:36type of game, duration of gameplay,
14:38narrative, storyline.
14:39So one ways in which people are trying
14:43to make screen assessments more
14:45palatable is to gamify, for example.
14:47And one way in which they gamify it
14:49is to include a narrative storyline,
14:51follow Mr.
14:51X as it goes on a plane,
14:53and then while that is happening
14:56you have different cognitive tests.
14:57And so I was interested in whether
15:01this influenced the validity of game.
15:03Especially since I'll talk about my
15:05study which actually uses a narrative
15:07based game and we extracted this
15:09current method like I talked about
15:11the study side mode of delivery and
15:14this year's traditional validated tests
15:16like the West, Kaufmanns, Baileys.
15:19There were eighteen games across
15:2020 studies, 17 serious games,
15:22which means that they were specifically
15:24designed to measure cognitive function,
15:26and one commercially available game,
15:28Minecraft, which was assessed for its
15:30utility in assessing cognitive function.
15:32And compared with traditional assessments.
15:34The duration of Gameplay varied,
15:37So we can do much, much more with that.
15:39There were five games to use narrative
15:41story Line 6 studies of 20 studies,
15:43which is the predictive model.
15:45You can see the wide range of sites,
15:47the delivery mods. Words.
15:49Most of them were through computers
15:52and we extracted shout out to Megan
15:54and ISA for helping with extracting
15:57all of this coral correlations.
15:59But we extracted.
16:02375 pairwise correlations across the
16:04street difference through 20 studies.
16:0775% of these were significant.
16:10Working memory was the most common
16:13cognitive function measured,
16:15followed by attention,
16:16inhibitory control and visual spatial skills.
16:19The meta analysis just quickly
16:22while on the low to medium range,
16:23which you might say compares
16:25to other validation studies.
16:27But for attention it was
16:280.3 inhibitory control 0.3,
16:32my working memory at the best at 0.4.
16:36We did very basic frequency high
16:40square frequency comparisons by by
16:42correlations by these different
16:44factors and we found no differences
16:48by sites and format of delivery.
16:50But we found that adolescents tended to
16:52play older adolescents tend to play better.
16:54As you would expect.
16:55As you grow older,
16:57you would you your cognitive
16:58process will get better and you
17:00will play better in the game.
17:01And also as you would expect,
17:03a producer of a prediction model was
17:05more likely to yield significant.
17:07Solutions as you would expect,
17:08since they were specifically
17:10identifying metrics, right?
17:13But also the inclusion of a narrative
17:15story lines seem to be more associated
17:17with non significant correlations.
17:18So I wonder about whether
17:21that was distracting.
17:22All this to say in general is that
17:25that that that there are factors
17:27that influence the the how valid
17:29game based games are for assessing
17:31cognitive function is is the takeover.
17:34We were interested in assisting for
17:37race and none of the studies did
17:41reviewed reported on race and ethnicity.
17:45Most studies as you can see here
17:48found effect differences for age.
17:49Gender was really around spatial
17:52reasoning and ability,
17:53which I think relates to that.
17:56Boys are more exposed to buy their
17:58toys special task as opposed to
18:00like a biological difference and
18:02then prior exposure to gaming.
18:04Technology also influence which
18:06we'll talk about also.
18:08So if.
18:09If you can learn from the game,
18:11is it is it still?
18:13Can it still be useful as a screening tool?
18:18And so I talked about this unexplored
18:20factors which we we need to be thinking
18:23about and we need to be assessing for
18:26for widespread we're talking about
18:28scalable and scalability of this tools.
18:30So in summary, many adults and such risk
18:33for substance misuse are not identified.
18:35The use of digital tools
18:37can alleviate some barriers.
18:38Cognitive dysfunction is associated
18:40with substance misuse.
18:42Video games can measure cognitive function.
18:45So it may be a potential tool for
18:47identifying adult center to risk
18:48for substance misuse via measure of
18:50indices that correlates with cognitive
18:52functions that are also implicated
18:54in development of substance misuse.
18:57So this is one of the basis for our
18:59proof of concept study and the central
19:01question is can data collected during
19:03gameplay be used to identify adult
19:05centered risk for substance misuse?
19:08Umm.
19:09Central hypothesis is that adolescence
19:11with higher risk of substance misuse
19:13will perform worse on the video game
19:15and demonstrate poorer cognitive
19:16function compared to adolescence at
19:18a lower risk for substance misuse.
19:20We've already established that
19:22cognitive function influences how you
19:24perform in a game and also influences
19:26your risk for substance misuse.
19:28So we're using play forward game
19:29developed in the play to Prevent Lab
19:31in narrative based game initially
19:33designed to target HIV and high risk
19:35behaviors like substance misuse.
19:36It has 12 levels and five mini games
19:39here shown at the bottom here.
19:41Each mini game has 10 levels.
19:43We'll talk a little bit about that
19:45and players and stars and points has
19:47been previously tested where we had
19:49160 where they were $166 cents each,
19:5311 to 14 who played play forward.
19:55There were 43% black adolescents.
19:5815% fifty 6% Hispanics, Hispanic adolescents.
20:03So in the I will try and walk through
20:05the point to show how perhaps some
20:08of these metrics may correlate
20:10with cognitive processes.
20:11And so in the no sense mini game
20:14an adolescent has a challenge and
20:16they have to decide the statement
20:18presented to them about drug use
20:20or drug misuse and they have to
20:22decide if this statement is true
20:24false and or an opinion and so you
20:27can imagine that they have some of
20:29it is influenced of course by their
20:31their prior existing knowledge but.
20:33How quickly can they retrieve
20:35the information they know?
20:36Are they going to do that or are
20:38they going to just choose an option?
20:40Impulsivity, perhaps,
20:41and how does that influence how well
20:44they perform as they're playing in the game?
20:47They have feedback,
20:48they get feedback.
20:49And so you can imagine that if that
20:51people will do better over time,
20:53unless they don't really don't
20:55care whether to do better or not.
20:56And can we seize those differences apart?
21:00Umm,
21:00in the People's Sense mini game this
21:03this player has to decide where to
21:06place their friend on friendship
21:08circles and how they decide that ideally
21:11should be influenced by people's skills,
21:13which may reflect whether they
21:15are low risk or high risk.
21:17So for example, how do you observe
21:19people's skills while Jaden is
21:20always hanging out with Dante,
21:21but Dante happens to be so
21:23wasted at the party last night.
21:25Is he someone you wants and
21:27you're very close circle?
21:29So when you decide that's the play,
21:31it gets more complex as you
21:33go through higher levels,
21:34but you can imagine that a player
21:36has to keep that information.
21:37You can say they can always go back
21:39and click and look at the risk,
21:41but how quickly you advance in this
21:43game depends also on how much of
21:46that information you can keep within.
21:48And when they finish placing
21:49friends in their circle,
21:51they then have invites that they
21:53get and the invites that they get.
21:56They have to decide whether
21:57this is a good invite.
21:59Very risky invites and depending
22:01on whether they decline or accept,
22:03they get stars.
22:04So if you accept like
22:05invites that are very risky,
22:07you can have three strikes in your house.
22:11And so I'm we myself,
22:15Megan Isabella played the game and
22:18page by page of the game reviewed.
22:21What are specific metrics in the game that
22:24may be influenced by cognitive processes?
22:27And so, for example,
22:28in the deciding your friends,
22:30for example,
22:30in people's sense,
22:31checking out like this time spent
22:34checking PS characteristics and
22:36there's you have to correctly set
22:38peers into right social circles,
22:39and we're hypothesizing that
22:42these domains are probably
22:44influencing those processes.
22:47And also when you are accepting and rejecting
22:50invites correctly accepting or declining.
22:53Um,
22:53that these processes,
22:55this cognitive processes and domains
22:57are influencing how well adolescence
23:00perform in this tasks in the game.
23:03You can also see that there are constructs,
23:05so there's a time construct,
23:07and there's also an accuracy conduct
23:09time constructing yellow and the
23:11accuracy constructing green.
23:14And so. The the first game from
23:18this study be butanol, that is,
23:20can we identify them metrics in the
23:22game that are predictive of substance
23:25misuse and can we derive a prediction
23:28model using those identified metrics?
23:30We used 166 participants.
23:32Like I talked about.
23:33We had two outcomes,
23:35substance misuse and self
23:36efficacy to refuse drugs.
23:38Sometimes it's used was already measured
23:40using the youth Risk Behavior Survey.
23:42There were twenty questions related
23:44to alcohol and drug use and we'll
23:46talk a little more about this,
23:48but some of those questions included
23:49if I tried using the cigarettes,
23:51which I think contributed to this
23:53being overall a low risk group in
23:55terms of like if you used once you
23:58were you were considered high risk.
24:00Um, based on just how many?
24:02The variance in the sample and then
24:06self efficacy to refuse drugs was
24:08measured using the door scale and
24:10we dichotomized these two outcomes.
24:14We use the the variance
24:16threshold metal method to drop
24:18all metrics with zero variance.
24:19So if they have no variance then they
24:21are likely to differentiate the two
24:23groups between low and high risk.
24:25And we also dropped some variables
24:27that had very high multicollinearity
24:30and used this machine learning
24:32technique called cross validation
24:33which splits the data and then splits
24:35the data even more and and checks
24:37to make sure that how the model
24:39performs in one set of data is the
24:41same across multiple sets of the data.
24:43It's a very small sample for people
24:45who do machine learning statistics as
24:47one single is like very small sample.
24:49Usually you want to be using
24:52samples 600 and above.
24:53And so we were limited in the amount of
24:56like can we examine race for example,
24:59which is something that I wanted
25:00to examine other other models,
25:01the same between black and
25:03white adolescents for example.
25:05And then we tested these six different
25:08models and computer AUC values.
25:10AUC value tells you how well
25:12your model is doing.
25:13And 0.5 usually means that
25:16it's not doing anything,
25:180.6 means that it's performing moderately OK,
25:20and 0.7 usually means this
25:22is a pretty good model.
25:25So we excluded 6 um log files that were
25:30corrupted or were from adolescence.
25:33We had there were mostly mostly corrupted
25:36or were incomplete and based on the final
25:39sample one in three adolescents had high
25:41risk or causing the high risk of substance
25:44misuse and wanting about wanting to
25:46have poor self efficacy to refuse drugs.
25:48We ultimately had 285 in
25:51game metrics after cleaning.
25:53So our first outcome,
25:55a model was not good.
25:57So the model didn't predict what didn't
26:01predict substance misuse among adolescents.
26:04Umm. And but the second outcome
26:08self efficacy to refuse drugs.
26:10The logistic regression model seem
26:12to perform relatively well across
26:15multiple and was stable across multiple
26:17cuts of the data with an AUC of 0.6.
26:22When we looked at the model,
26:24when we looked at what was contributing,
26:26what metrics were contributing
26:28to this prediction?
26:30What I take from here is that most
26:32of them happened at the beginning.
26:34So zero level 0, there were ten levels,
26:36zero level 0 to 9 and most of these
26:38were happening at the beginning
26:39of the of the game.
26:40Which again speaks to like can these be
26:43used as a can just really be used as a
26:46screening if it identifies if it can,
26:48if your performance can change over
26:50time and is if your performance
26:52changing over time.
26:53Is that reflective of an improvement
26:55in function that actually influences
26:57substance misuse or not?
26:58Many questions are raised right?
27:01So in summary I think there
27:02are outstanding questions.
27:03I think we found certain game metrics
27:06were predicted of self efficacy to refuse
27:08drugs among adolescents aged 11 to 14,
27:10but not drug misuse.
27:11I think this was an overall
27:13overall low risk sample.
27:15Game based features may be more
27:17useful as monitoring metrics
27:18during an intervention for example,
27:20like if you embed them as opposed
27:22to screening, but if you embed.
27:24Them into an intervention and you
27:26use machine learning algorithms
27:28to personalize interventions,
27:30and you've documented at baseline
27:32where people have deficits.
27:33Could you then use that as a monitoring
27:35over time of the improvement?
27:37But then you also have to show
27:39that the the improvement is related
27:42to actual improvement in risk.
27:46And then we probably need better game
27:48behavior that is more reflective
27:50of substance misuse.
27:52Some of the work that people are
27:54beginning to think about is like
27:56are there can you embed cues within
27:59a game and and can you use more
28:01biometric measures and would that
28:03be more reflective as opposed to,
28:05you know,
28:06how people are performing in a game,
28:08for example. So there's definitely more work.
28:11I think further investigation is
28:13needed at this algorithm is going to be.
28:15Finally,
28:15valid whether the hypothesis may
28:17be at play that account for these
28:20predictions algorithms performing
28:21similarly between among black
28:23adolescents as opposed to white adolescents.
28:26Umm.
28:28One of the things we're going to do so,
28:29so far we've we've looked at a precise
28:31prediction for substance misuse,
28:32right. But we're our hypothesis is that
28:35this is influenced by cognitive processes.
28:37But we haven't tested
28:38for cognitive processes.
28:39So we're going to embark on a
28:41pilot to see if those same mackers,
28:44we can compute the score using the
28:47logistic regression model and if those
28:49are actually associated with executive
28:51functioning measured by actual tasks
28:53like the impact test and the go,
28:55no GO task among.
28:5714 to 15 years old.
28:59So we'll see.
29:00We'll see what that data tells us.
29:04So in addition to games,
29:07we talked a lot about games.
29:09I think I think I got you know,
29:10I presented the slide where I say let's
29:13meet adolescence wherever they are,
29:14how they are engaging with the world.
29:17And so you know the games I one of
29:19the areas that I've been thinking
29:21about is how do we use the electronic
29:24medical record to identify risk.
29:26Umm, and our rush risk of substance
29:29misuse rationale is that the you know,
29:32the HR is already in use.
29:34It has vast amounts of data and
29:37these that are routinely collected.
29:39We don't need a different process
29:40for collecting this data.
29:41It's already happening.
29:44There are two types of data
29:46that occur that you can.
29:47I mean there's it.
29:48There's a large debate about
29:49the kinds of data in the HR,
29:51but largely there's two
29:52kinds of data in the EHR,
29:54structured data and unstructured data.
29:56Structured data you might say like things
29:59that someone selects from pre populated,
30:02it's already in the system
30:04and you select all.
30:05This person has an alcohol use
30:07disorder you selected as opposed to
30:09like a structure where a provider is
30:11imputing what they think and data.
30:14The type of data is important from
30:18a from a from the point of trust.
30:20And we talk a little bit about about trust,
30:23like what data can we trust?
30:25Is the problem,
30:26can we trust the problem list?
30:28Is it always complete 80%?
30:32There is not.
30:34And is it nice thing to load?
30:39Um, can we,
30:40you know or or do we trust the
30:42clinical notes at the clinical notes,
30:45what kind of information we
30:47get from the clinical?
30:48Are they more predictive?
30:49The consensus is that we should be
30:52using both all kinds of data that
30:54we can get from the EHR as long as
30:57we're intentional about why we're
30:59using them and also intentional and
31:01thoughtful about what we find from,
31:04you know, whatever models that we find from,
31:06from the use of this data.
31:08Prior studies have shown that
31:10we can use EHR data to predict
31:13mental health outcomes as you
31:15side health services research,
31:17suicide prediction, depression,
31:18anxiety and alcohol.
31:19And these are findings from two different
31:22studies on the left here as one study here,
31:25which was among young adults
31:27and this was among adolescents.
31:29And the models were pretty pretty good.
31:32Good enough, I would say,
31:34in identifying these disorders
31:36among adolescents.
31:38So it can be used.
31:40I think there are limited
31:41studies among the adults and
31:42population,
31:43particularly using unstructured data.
31:45Most of the studies use structured data.
31:48And I'm, I'm very interested, you know,
31:51I showed this slide about the median
31:54time between onset of symptoms and
31:58actual initial treatment contact.
32:00Is it possible to derive a model
32:05that identifies risk as the mergers
32:09and then funnel an adolescent
32:12appropriately to intervention?
32:14It would mean that you have a way
32:16to to to determine time to event or
32:19time between the onset of symptoms
32:22and when has recurrently defined
32:25substance use disorder, for example.
32:27So I'm I would like to explore
32:30that and investigate that and I
32:32think there's a lack of focus on
32:35disparities and why disparities.
32:37Um.
32:38Umm.
32:42Lack of focus on disparities can
32:45cause harm and machine learning.
32:46You know, a few years ago we said
32:48we thought machine learning was
32:49going to solve all our problems.
32:51There was truth in it and we would be
32:53able to identify every insoluble problem.
32:56And now we're finding that machine
32:58learning algorithms are inherently biased.
32:59And some are. Some are racist and.
33:03And so on the left,
33:04here we have Google apologizing for
33:06having an algorithm that then gets a bug,
33:08but then this bug uses for some reason.
33:12However it does this, it's able to.
33:14It's now identified that now
33:16identifies black people, wrongly,
33:18as guerrillas.
33:20And on the right is this famous people
33:23might have known about this study by Obama.
33:27Yeah, which?
33:29Was looking at the models used in a program,
33:33where the program was designed to
33:37automatically funnel adults into
33:39a program that helped them manage
33:42comorbid chronic conditions.
33:44And so the algorithm computer the score
33:47and the risk score was at 97 percentile,
33:51and if you reach that based on
33:53the number of chronic conditions,
33:54you were automatically funneled
33:56into this program.
33:58And so you can see that the the the
34:01couple line is for blacks and the
34:03orange line or white is for the yellows
34:06for whites that whites at a lower level,
34:09lower number of active chronic
34:11conditions were being funneled into
34:14the program earlier than blacks.
34:16And so, umm,
34:17when they when they risk
34:19wasn't accounted for.
34:20Race is not exactly a very good
34:23because we're learning it's not a very
34:25good metric for assessing racism.
34:27But I think it's a, it's a good start,
34:30especially if you're intentional
34:32about examining disparities.
34:34But they didn't account for that here.
34:36They simply just deployed an
34:38algorithm based on data that existed.
34:40But when they examined,
34:43they realized they realized that the.
34:46Metric, which is why I,
34:48you know,
34:49examined the metric like what
34:50metrics are actually contributing
34:52to our prediction model.
34:54So when they examine the metrics that
34:56we're contributing to this model,
34:57they found that the one of the largest
35:01contributor of the model was cost
35:03and that whites were more likely to
35:06spend more per chronic condition than blacks.
35:10Why?
35:10Some of the things that like there
35:12are many competing priorities,
35:14Blacks may not be able to take time off work.
35:17To go and see their doctor,
35:19whites were more likely to have procedures
35:23and inpatients appointment large.
35:25This is to say that if if
35:27we're not intentional like.
35:29Algorithms can do a lot of good,
35:32but they can also cost ham
35:34and we need to be thinking.
35:36Or at least I,
35:36as someone who is doing a lot
35:38of machine learning research,
35:40needs to be thinking about how
35:43how these models are used.
35:44What informs these models and can we,
35:46before we deploy them for, you know,
35:48deploy them for prime time?
35:50And so the question that we're asking is,
35:53can data collected routinely
35:55in the electronic health record
35:57be used to identify adolescents
35:59at risk for substance misuse?
36:01And now they are algorithmic
36:03differences by racial ethnic groups?
36:07We will identify electronic health
36:09record data features that predict
36:11substance use disorder will derive a
36:13model will determine if electronic
36:15health record features predictive
36:16of substance misuse disorder.
36:18So use this sort of default by racial
36:20ethnic groups and then we'll try to derive
36:22a lot of time so these these features
36:25are collected at different time points.
36:27When do at what point do you have
36:29enough features in the model that
36:31you can actually see this person
36:33should be further assessed and.
36:36What is the length of time between
36:38when those at documented and the first
36:41determination of substance use is so
36:43that they exist right now and also
36:45referral for behavioral health services
36:47and also determine if that length of
36:49time there any racial ethnic differences
36:51in those in that length of time?
36:53We're going to use data from the
36:56Fairhaven Community Healthcare.
36:56They have about 100 / 100,000
36:59records and we're going to use that
37:02training model for predicting SD.
37:04We will validate that model among
37:06adolescents 12 to 17 years and we'll use
37:09both structured and unstructured data.
37:11So right now we have IRB approval and
37:14we're working through data use agreements.
37:19So hope, hope, hopeful to have some
37:21of this data and stats get into the
37:24nitty gritty nephew in a few weeks.
37:27So, in conclusion, a dozen substance
37:31misuse is a major public health problem.
37:34There are myriad of barriers that
37:36preclude early identification.
37:37We need to identify adolescents
37:39wherever they are and as risk emerges.
37:44Talked specifically about video games.
37:46I'm interested in all things digital tools.
37:49Phones and a lot of there's a lot
37:52of work on ER and that's ecological
37:56momentary assessment MMA and so that
37:59there's there's a wide variety of
38:01how we can use digital tools mid
38:04adolescence where they are social
38:06media for example I was I was I was
38:09trying this I was looking yesterday
38:10as I was preparing and Googling on on
38:13Google like self harm and immediately
38:15I Google self harm like the 1st
38:18988 if you you know like there's.
38:20Regarding working,
38:21that is trying to like identify risk and
38:24trying to like deliver an intervention.
38:28And so I think,
38:29you know we need to be thinking about
38:31all of these different strategies have
38:34I think ideally have a systemic model.
38:37And umm,
38:38and refine the ones like there's some.
38:42I think there's a lot of work to do
38:44to refine the use of video games,
38:46refine the use of the electronic
38:49health record.
38:50So there's more research to be
38:51done in refining these tools.
38:53And ultimately we want,
38:54we want the adolescents to
38:56live healthy lives.
38:57We want them to live highly functional lives.
38:59And yeah,
39:00whatever we can do to make that a reality.
39:04So is all about. So thank you.
39:07I will.
39:12My family patients,
39:14study participants by mentors,
39:16collaborators, funders,
39:17members of the play to prevent lab.
39:20Shout out to Jenny and Fiza who
39:23keep the wheels running and yes,
39:26I'll take questions.
39:33Yes.
39:36Stolen.
39:46Richard, thank you so much.
39:48So as a CL psychologist, I'm absolutely
39:50thrilled by this study where
39:52you're looking at the electronic
39:54health record in primary care. So
39:57that is an overwhelmingly rich
40:01source of data, everything from,
40:03you know, social workers notes
40:05to the standardized instruments
40:07that are being used at Fairhaven.
40:10Do you have a sense of in that vast
40:12data set which initially are going to be
40:16part of your first pass
40:18because we know that at Fair Haven they are
40:20using standardized measures of depression,
40:22anxiety, suicide risk, social work notes.
40:25So I'm wondering how you're prioritizing
40:28that vast data in terms of your first pass.
40:32Are you using your kind of clinical intuition
40:34of what's because you're a clinician,
40:36you're amazing clinician of
40:37what's most likely to result?
40:39In that higher yield, right, right.
40:42I do, yes, we're using the,
40:44we're using the scales validated skills.
40:47So they use craft to identify, to identify,
40:50to identify the outcome as you know,
40:52as an outcome for prediction,
40:54like who is misusing substances.
40:57But we're also going to use like
40:59all primary care notes, I find.
41:01I think the notes are more something
41:03I find like the notes are more
41:06informative about presenting complaints.
41:07Sometimes you don't use the problem lists.
41:10Umm.
41:10But we're also going to use the problem
41:12lists and then we're going to use similar
41:15things that have been used in different
41:18studies like vital signs and encounters.
41:20I mean encounters they had in a year,
41:23did they go to the Ed,
41:24what did they go to the
41:26Ed for inpatient visits.
41:27So we're going to be using all of those
41:30and those are going to definitely be
41:32informed by our clinical knowledge.
41:34But also you know the I think
41:37the the thing about that is.
41:39Useful about machine learning is that
41:41it's at its nature is exploratory and
41:44hypothesis generating and so informed by
41:47that you also want to like get all that
41:49you can get because you can learn you.
41:55On the one hand you can your your clinical
41:58knowledge informs hypothesis which are
42:00already existing and you're testing them.
42:03But also there may be things that
42:04you haven't thought about that
42:06the machine learning helps you
42:07generate or think about them.
42:09Um, that we.
42:10So we're balancing,
42:11we're balancing those and we'll
42:13be collecting as much information
42:15as we have access to.
42:16Yeah.
42:23That was great Bouche and and you
42:27know it's such a huge problem that
42:29that we really wish you success
42:31because it's it's so important.
42:33So I was wondering what you thought
42:35about kind of what the factors are
42:38the skills are that help adolescents
42:40you know say no to drugs or or or
42:42or lessen their use and and we
42:44know that fear doesn't work right.
42:46That's been tested forever
42:48that that doesn't work.
42:49But is it I'm stuck with this is it
42:52like learning things or is it just
42:55peer group influence and and you know
42:57are there really things we could teach
43:00that get kids to you know that you
43:03could teach on a game or is it the
43:05way we get them to think differently
43:07after they're playing the game?
43:09I'm, I'm I'm just interested in what you
43:11what you think you're going to find.
43:15So I think that based on based on
43:17other studies that have been done,
43:19I think that you can one, you can model,
43:22you can model behavior in like the
43:24same ways that you expect, you know,
43:27adults, parents to model behavior.
43:29You can model those behaviors in game,
43:31in games. You can also teach within the
43:34game because they like in the mining games,
43:36they are practicing skills like
43:38they're practicing how do you refuse?
43:41Someone says, oh,
43:41let's go to like how do you say no,
43:43what things can you say?
43:45Um to to circumvent this.
43:48Um, sometimes you may not
43:50succeed and if you don't succeed,
43:51what are the consequences?
43:53And the idea is that if you're if
43:56you're engaging your cognitive
43:58processes in practicing this in
44:00within the veteran environment
44:02that you may be able to translate,
44:04depending you may be able to translate
44:06that you will translate this into real life.
44:08But I think also there's the
44:10part about habit,
44:11like the dose like how much,
44:12how much practice are they getting?
44:16Similarly to how I might learn math,
44:18for example.
44:18Like if I if I did more of the
44:20work of problem sets,
44:21then the likelihood that I would know
44:23what to do if I'm presented with a
44:25different problem set will be high.
44:27So how much? How much dose is enough?
44:29How much dose translates to
44:30a lowering your risk,
44:31how much of that then translates to the
44:33idea to the fact that these adults and
44:36might then practice this in in real life.
44:38But also there's also,
44:40I think he also alludes to a a big
44:43the issue of like this culture.
44:45Because there's a huge influence of culture,
44:48so you may learn all of this,
44:50but like the stigma that's peer like.
44:53I might confident enough that I can
44:56do that within all of the scenarios
44:59and we in view and raise the level of
45:02confidence within the games to then
45:05ensure that they can also translate
45:06this when there's high pressure.
45:08That's something that's I think
45:10that's something we need to test
45:11because indeed it is complex, I agree.
45:16Hi, nice talk.
45:19I had a couple clarifying questions
45:21about the study about the game, so.
45:25Can you explain again how you
45:27classified whether they were
45:29classified as using substances?
45:31Did you say it was one time
45:33using one cigarette? No, no, no.
45:35I there were many questions.
45:36There were twenty questions one
45:38and and it included cigarettes,
45:41alcohol and drugs and it included ever
45:44used and also past 30 days of use.
45:47OK. Yeah. And all of those.
45:49So the, it was such a low risk
45:51sample that all the all those
45:53who said no for all of them were
45:55essentially in the low risk group
45:57and anyone who endorsed any of
45:58that went into the high risk group,
46:00which yeah which is why we're
46:02going to do it more with the pilot.
46:04We're going to have a more high risk
46:06sample like everyone will be misusing
46:08substances at some degree of frequency.
46:10And are you thinking of because
46:12I also was thinking about the
46:14age range from 11 to 14 but.
46:16Yes there are some kids that
46:17are starting to use that young
46:19unfortunately more all the time but
46:20I think that they'll start you know
46:22there are a whole group of kids
46:24that don't start until they get
46:26to high school and that's really
46:28when it when there's more risk.
46:29So I just wondered if you're thinking
46:31of going up a little little higher
46:33in the age range we are so the
46:35pilots is going to be between 14
46:37to 15 year olds and we wanted to
46:39be careful not to you know we're
46:40I I showed how age can in the age
46:43can influence how they perform in
46:45the game and such that if it's.
46:46If you're using older kids,
46:47gonna be so easy that you're really
46:49not getting at the processes.
46:51And so we're we're looking at 14 to
46:5315 also because of the feasibility
46:55like you can at least get all
46:57of them in a high school,
46:58but they would be high school
47:00students who are 14 to 15 years.
47:02Last question, sorry.
47:03I just thought too about the the
47:06reasons that kids start to use right,
47:08so that we've talked a lot,
47:09it sounds like we're talking a
47:10lot about the pressures, right,
47:11of other in the peer situations,
47:13but there are kids that that's not
47:15necessarily how or why they start.
47:17And so I just wonder if that's
47:19something you've thought about
47:21in terms of how to integrate?
47:22Those sorts of risk questions
47:24related to coping strategies
47:26or stress levels or you know,
47:28reasons that kids start,
47:29you know,
47:30nipping out of their parents
47:31cabinet at home or stuff like that,
47:32that doesn't have anything to do with
47:34them being at a party or being with friends.
47:37Right.
47:37I think the interventions do all like address
47:40a lot of the different risk situations.
47:43We just,
47:43yeah,
47:43this was just a sample of,
47:45but intervention interventions
47:46usually address a wide range of of
47:49risk and influence by like
47:51focus groups that make.
47:52The game story lines more
47:54reflective of their own
47:56experiences, but yes, agreed.
48:01Sort of PBA on two of the previous questions.
48:05This is a very important initiative for
48:08us to be hearing about those of us who
48:11have been around and been struggling
48:13with this issue for 50 years or more.
48:15It's easy to get very habituated
48:18to the discouraging components of
48:21how we have failed repeatedly this
48:23population no matter what we've tried.
48:26So the the fresh creativity that you
48:28bring to this is extremely welcome and
48:31don't let anybody talk you out of it.
48:35And one of the questions that
48:37I'm raised in my mind is that.
48:39I think some of the best.
48:44Diagnostic work about substance abuse
48:46in this age group is done by peers.
48:50They're often extraordinarily accurate
48:52talking about their friends and talking
48:55about what they see in their friends,
48:58and I wondered if you have a multiplayer
49:00game in your future where you could expose
49:03this to a larger group of problem solvers.
49:08Love to I would love to. Yes.
49:11Um, multiplayer family based games.
49:13Yes. Yes I think I I think we should
49:16be meeting adolescence where they are.
49:19We should, it should be informed
49:20by what we know about risk.
49:22I think we should use multi modality.
49:24And yes, absolutely curious.
49:28We have a. Sure. Kim,
49:33can you hear us? Can you unmute?
49:38Hi, uchi. I have a question for you.
49:44I was really interested what
49:45you were saying about like uh,
49:47it was really fascinating that you
49:48were able to pick about apart a
49:50lot of this by kind of thinking
49:52through how kids need to adjust to
49:53game play before they, you know,
49:55thinking about how those biomarkers
49:57might be relevant to kind of address
50:00you know or identify at risk players.
50:03So I'm curious your thoughts when
50:05you have like if you have a brief
50:08intervention like a brief one or
50:10two hour game based intervention,
50:12how you might get over that barrier?
50:14Um, do you in terms of like collecting
50:16data to try to kind of gather that
50:18information around at risk youth,
50:20are you suggesting that?
50:21It may not be a good idea to look
50:23at that early game play and maybe
50:25looking like at the full game
50:27plays to change over time.
50:28Or are you or are you thinking
50:30it may be more valuable to look
50:31after somebody kind of salad and
50:33learn the ropes and gotten through
50:35several levels later in the game?
50:37I I think we should look at all
50:39of it because I I think that.
50:41So one of the questions I'm I'm
50:44wondering about is that is the
50:46difficulty of overcoming that initial
50:48difficulty is that influenced by
50:51difficult you know difficulties
50:53and cognitive processing or is it
50:55just that we're just trying to get.
50:57I'm just trying to learn how this game works.
51:00We haven't really tested if if there's
51:02a difference with that and if they settle,
51:06if they settle.
51:07Like is that?
51:09The rate of settling, is that also
51:12influenced by baseline cognitive function?
51:15I don't know.
51:16I think that those are all questions
51:18that need to be, need to be,
51:19and need to be answered for us
51:21to determine when.
51:23I do think, though,
51:25that if learning occurs.
51:27If learning from the game occurs,
51:29and it doesn't influence risk like that,
51:32that that risk doesn't get better as you're
51:35learning like it's the couple from it,
51:37then it's not a good metric for
51:39measuring risk for substance misuse.
51:46Thank you, Kim and thank you for
51:48helping being such a good mentor
51:50to which we have another question
51:52from Doctor Christine Emmons.
51:54Christine and if you
51:56do you have plans to develop
51:58games that tag your treatment.
52:00Oh, sorry about that.
52:00Christine do you want to
52:01ask your question and if you if you.
52:05Yes. So my question is do you have
52:07any plans to to create games at
52:10Target treatment or integrated
52:13diagnostics with treatment?
52:16Um. I mean, ideally we would have a model,
52:19we'll have. Personally,
52:22I think about prevention. I do.
52:25But but if we think about the
52:27problem of substance misuse,
52:28we should all be thinking about
52:30how we also target treatment,
52:32and so I think that those are
52:35possibilities, especially.
52:36If we can, if there's a model,
52:39you know my mind,
52:41there's a model that funnels
52:43appropriately and but we also need
52:45you know I also mentioned how self
52:48standalone digital interventions may
52:50not be very effective for treatments,
52:53you know, for a severe substance
52:55use disorder and treatments.
52:57So I think we need more.
52:59Most of the studies that have
53:00been done so far are showing that
53:02they are useful as adjunctive,
53:03especially if you're having like
53:05a maybe zoom televideo treatment.
53:07And then you have these as perhaps
53:10assignments at John Adjunctive that
53:12help a person practice some of these
53:14skills that we're talking about.
53:16So as an adjunctive treatment,
53:18I think those are,
53:19those are areas where due to interventions
53:21can be really high yield, I think, yeah.
53:25But of course I'm open.
53:27I'm open to a possibility.
53:30So maybe building on some of the
53:32comments earlier on and you talked
53:34about possibly using other digital
53:35technologies and integrating them
53:37into your future research program.
53:38I was very taken by reaching recent
53:41nature I think biotechnology paper that
53:43looked at Fitbit data in the context
53:46of COVID and predicted infection like
53:49many days before symptom onset and that
53:51was in a relatively small sample size,
53:53I think it was around 64 participants.
53:55And then the all of US initiative,
53:58you know they're recruiting.
54:00Million people,
54:01but their approach is just to say well
54:03if you have a Fitbit you know will
54:04you allow us to access your data.
54:06So they're not even providing Fitbits,
54:08but they receive the data.
54:09I just wondered if anyone is
54:11integrating wearable tech like Fitbit
54:13data in substance use in adults
54:15and whether or not that might be of
54:17value in in this population as well.
54:19I think I think they're doing.
54:20I think there's actually some research from.
54:23Integrating Fitbit or at least
54:26wearable technology, I would love to.
54:29I would love to integrate that especially.
54:34Because you know, one of the ways
54:35I think about it is that if, if,
54:37if there are areas where I think that
54:39where they've used them before in adult
54:41studies is that there are areas that.
54:44For example, if you're close to a
54:45shop where you can buy a vape right,
54:47like and then something will,
54:50you will get a notification that
54:51you're in a place where you might
54:53engage in a high risk behavior.
54:54What might you do?
54:56What resources do you have that
54:58you can employ in this moment?
55:00So I would I would love to be
55:02able to integrate the wearable
55:06technology in monitoring.
55:08Kids who are high risk and
55:10thinking about how we might,
55:11how we might measure those, how,
55:13how and what will be most useful
55:17as measures for identifying
55:19who might be struggling,
55:21who is struggling at the moment,
55:23who would be a risk for substance misuse.
55:25Yeah,
55:26and if the kids are bringing their
55:27own Fitbit, it would be quite
55:29cost effective. Question from you.
55:37Thank you very much. That
55:38was terrific presentation
55:39of of your amazing work.
55:42I had a weird idea that that I
55:44don't it's not terribly well formed,
55:46so I apologize for that.
55:47But on my mind is the recent CDC
55:51report about the prevalence of.
55:53Of anxiety and depression in
55:55youth in the United States,
55:57especially amongst.
55:59Girls and young women and but
56:03across the board I think really,
56:05and what occurred to me is that probably
56:09there's a connection between those
56:11dysphoric experiences and substance use,
56:14I mean, at least broadly.
56:16And I just wondered whether.
56:18You know,
56:19you may not have an answer right now,
56:20but whether in the data you collect,
56:23is there some way of collecting
56:26data about about those dysphoric
56:29experiences at the same time you're
56:31collecting things that might lead
56:33to directly to substance use?
56:35Right.
56:37So with the electronic
56:39health record data data,
56:41we're also going to be looking at
56:43predictions for depression and ID.
56:48So we, we would look at,
56:50we would look at any relationships and
56:52see if those features are also predictive.
56:54If the features that are predictive
56:56of substance misuse disorder are
56:58also the feature same features that
57:00are predictive of major depression,
57:01there's major depression occur
57:03before substance use disorder.
57:05Could we use that as a way to also screen,
57:07you know, for for substance misuse?
57:12I do think that there in terms
57:14of games or digital technology,
57:16there are ways in which you can.
57:20Measure like the ways in which behavior
57:23in a game can be reflective of affect.
57:27And so if, if, if, we can.
57:34If we can embed some of those as
57:37we develop games and we use that
57:39as monitoring for like if if we can
57:42monitor a change in your effects
57:44is that also improving your risk
57:46or lowering your risk.
57:48So I do think I mean and these
57:49are highly commoditized like the
57:51depression anxiety increase you're
57:52repeating as they increase your
57:54risk of substance misuse.
57:55So we do need to be addressing the
57:58both but also yes I agree thinking
58:00about how we how we use what we
58:03know about the physiological.
58:04Presentations,
58:05physiological manifestations of
58:07this disorders,
58:08how do we embed those in games
58:11that measure and monitor over time
58:13and how do we use that to monitor
58:16improvements as we as we address,
58:18as we address this underlying disorders?
58:22Which I think that there
58:23would be many more questions,
58:24but we're going to finish here.
58:26But thank you so much.