February 27, 2024 • 29 mins
Dr. Hui (Henry) Luan joins EPITalk host and co-author, Dr. Patrick Sullivan, for a jam-packed discussion about their recent study on PrEP accessibility measurement. “Spatial accessibility of pre-exposure prophylaxis (PrEP): different measure choices and the implications for detecting shortage areas and examining its association with social determinants of health” can be found in the October 2023 issue (Vol. 86) of Annals of Epidemiology.
Read the full article here:
https://www.sciencedirect.com/science/article/pii/S1047279723001412
Link to the R script for implementing the Gaussian 2SFCA measure for PrEP accessibility:
https://figshare.com/articles/journal_contribution/R_script_on_Gaussian_2SFCA_R/21778493/1
Some resources on GIS/spatial statistics in health can be found below:
- https://www.cdc.gov/gis/index.htm
- https://epiresearch.org/annual-meeting/2022-meeting/2022-workshops/
- https://www.publichealth.columbia.edu/academics/non-degree-special-programs/professional-non-degree-programs/skills-health-research-professionals-sharp-training/bayesian-modeling
Episode Credits:
- Executive Producer: Sabrina Debas
- Technical Producer: Paula Burrows
- Annals of Epidemiology is published by Elsevier.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Patrick Sullivan (00:12):
Hello, you're listening to EPITalk
Paper, a monthly podcast fromthe Annals of Epidemiology.
I'm Patrick Sullivan,Editor-in-Chief of the journal,
and in this series we take youbehind the scenes of some of the
latest epidemiologic researchfeatured in our journal.
Today we're here with Dr.
(00:34):
Henry Luan to discuss hisarticle "Spatial Accessibility
of pre-exposure prophylaxis(PrEP): different measure
choices and the implications fordetecting shortage areas and
examining its association withsocial determinants of health.
" You can read the full articleonline in the October 2023 issue
of the journal at www.
annalsofepidemiology.
org.
(00:56):
I'll tell you just a bit aboutour guest today.
Dr.
Henry Luan is an assistantprofessor in spatial data
science and spatial epidemiologyin the Department of Geography
at the University of Oregon.
Dr.
Luan applies and developsBayesian spatial temporal
statistical models and advancedGIS approaches to investigate
how health phenomena, especiallyhealth events related to HIV,
(01:19):
vary over space, time and raceethnicity, and how socioeconomic
, demographic, physical andbuilt environmental factors
contribute to these space, timeand cross-race ethnicity
variations.
Dr.
Luan, thank you for joining ustoday.
Henry Luan (01:32):
Thank you for having me here.
I'm excited to be here.
Patrick Sullivan (01:35):
And I'm excited to talk some about your
study, which gets to geospatialepi, which I'm excited about,
and which gets to pre-exposureprophylaxis, which I'm excited
about.
So tell me a little bit aboutthe purpose of the study.
What were you really aiming toanswer with the analysis that
you did?
Henry Luan (01:51):
Yeah for sure.
So the major purpose of thisstudy is to compare different
spatial accessibility measuresof PrEP providers and then
examine how different choices ofthe measures affect the
detection of PrEP providershortage areas, as well as the
association between PrEPaccessibility and social
determinants of health variables, as suggested by the title of
(02:14):
the article.
So by something I would like toclarify here is the difference
between the accessibilitymeasures for individuals and for
areas or for a neighborhood,because they're different.
So for an individual, we canderive their activity spaces via
techniques such as GPS equippeddevices and then derive how a
(02:36):
PrEP provider is accessible toan individual.
But for an area we cannot dothat because we cannot track
every individual residing in thesame area and instead we can
only calculate PrEPaccessibility for the entire
area or calculate theaccessibility for a
representative point of thatarea, for example the geometric
(02:57):
centroid or the populationweighty centroid of an area.
So in this sense they aredifferent.
I would like to say that beforethe widespread use of GPS
equipped devices, so these arealevel accessibility measures are
used as proxies of individuallevel accessibility.
So to- in short, this articlecompares the accessibility
(03:20):
measures for areas rather thanfor individuals and the main
questions we aim to answer.
There are two questions we wantto answer.
First is how does the choice ofPrEP accessibility measures
affect PrEP shortage areadetection?
And second is how does thechoice of PrEP accessibility
measures affect the findingsregarding the association
(03:41):
between PrEP accessibility andsocial determinants of health
variables?
Patrick Sullivan (03:45):
That's really interesting.
I want to break down a couplepieces of this, just because I
think maybe we both think aboutprep and geospatial stuff.
But for people who may not beso familiar this idea of PrEP
accessibility in areas in whichPrEP is accessible, why is it
important to know that?
If I'm in Portland and you usethis method and you tell me
(04:05):
where prep is more or lessaccessible, how does that
translate to public healthaction?
How would that data be used byproviders, for example?
Henry Luan (04:13):
I think.
First because we care about whohave sufficient accessibility
to PrEP providers, especiallyfor those population who need
them most.
And if there is a mismatchbetween the supply and the
demand, so we want to make surethose demands are made.
And we know that these spatialaccessibility is not constant
(04:36):
over space.
There are always some areasthat are in short accessibility
and some areas where they havegood accessibility.
So that's, I think, from thisperspective, we need to
visualize and to know where theaccessibility is now and doesn't
meet the need.
Patrick Sullivan (04:51):
And it's very actionable, right?
Because if I'm responsible forHIV prevention in this city and
I can look and say, oh, I knowthe areas of the city where new
HIV diagnoses are occurring, andthen you are able to use these
kind of data to show where theremay be low service areas, and
if those two things line up likehigh risk and low service,
(05:12):
that's the most impactful place,exactly.
So you sort of mentioned themethods a little bit.
So you must have inputs ofwhere the providers are and then
what are the methods that sortof lead to identifying these
areas.
How does that work?
Henry Luan (05:28):
Yeah, actually like a great question, because we
compared four differentcategories of measures in our
study.
So the density-based, theproximity-based, the more
complicated two-step floatingcatchment area-based approach
and finally, the most complexone, the Gaussian version of the
two-step floating catchmentarea approach.
So I would like to brieflyintroduce the principles behind
(05:51):
these four categories ofmeasures.
First is the density-basedmeasures, so this is the
simplest one.
So we use the count of PrEPproviders within an area, so in
our context, the zip code, thezip code compilation areas.
So we use the count in thatarea over the total demand.
So this demand, in our study weuse two proxies One is the
(06:14):
total population, the other isthe total account of new HIV
diagnosis and we also comparedhow these different proxies of
population demand affect theidentification of PrEP shortage
areas.
So this is the first one, firstcategory.
The second category isproximity, so we simply
calculate the distance along theroad network, so it's not like
(06:38):
straight-line distance.
We do account for the network,like in structure in reality,
and then we calculate thedistance from the zip code
centroid to the nearest PrEPprovider, or the average
distance to the three or fiveclosed PrEP providers, because
we want to account for more thanone provider, because usually
(06:59):
they have been researched, thatreviewed, that the residents
probably do not necessarily goto the nearest PrEP provider.
So that's the second type.
And for the third type is morecomplicated, that's the two-step
floating catchment areaapproach.
So, as the name suggests, thereare two steps in the calculation
of this measure.
So first we create thecatchment area for each PrEP
(07:24):
provider, so that's from thesupply side.
So based on, for example, thetravel distance along the road
network, based on a specificcut-off value, for example 10
minutes driving, 15 minutesdriving or 30 minutes driving,
and then we calculate thesupply-demand ratio.
So for example, let's assume aprep provider can serve five zip
(07:46):
codes when we serve, that isthe centroid of the zip codes
for inside that catchment area.
And then, if we assume thateach zip code has one resident
only, and then the supply-demandratio will be 0.2.
That means not the fullcapacity will be used by each
zip code.
There is a computation along thesupply side but of course the
(08:09):
actual population can beaccounted for to calculate the
supply-demand ratio.
So that's the first step inthis method.
The second step is that foreach zip code, so that is at the
demand side, we identify whichprep providers are accessible to
that zip code.
And then we add up thesupply-demand ratios associated
(08:29):
with different accessible prepproviders so that the final
results are the accessibilitymeasures for each zip code.
So you can see that thismeasure accounts for the
computation from the demand sideas well as the spatial
variations in the demandpopulation.
So that's the third type ofaccessibility measures.
The last one, which is the mostcomplicated one, the
(08:51):
Gaussian-based two-step floatingcatchment area, goes one step
further compared with thestandard two-step floating
catchment area approach.
That is, we add Gaussianfunction to account for the
distance decay effect.
In other words, we account forthe fact that a prep provider
further away from the zip code'scentroid will have lower
(09:15):
accessibility.
So this is not the case for thestandard version of that method
.
So because as long as the prepproviders are within the cutoff
value in the standard method,they are assumed to be equally
accessible, but that's not thecase in the Gaussian version.
So that's the four categoriesof the measure that we used and
compared in our study.
Patrick Sullivan (09:37):
Thanks for that explanation.
From a methodologic point ofview, how did those measures
compare, because the bottom linehere is sort of trying to
identify these PrEP shortageareas.
So were they pretty muchaligned, or was there one that
you felt like ended up beingmore informative than the other?
Henry Luan (09:53):
I would say the most complicated one is more
informative than the other onesbecause, as based on the
introduction to the principlesof these methods, it accounts
for different factors into thecalculation.
But here I wouldn't conclude interms of which measures can
more accurately detect the PrEPshortage areas, because we don't
(10:14):
know the actual ground.
Choose which area has a higheraccessibility, which doesn't.
But in state I would comparethem based on different factors,
for example if they account forthe variations in the demands,
if they account for thecomputations among those demand
populations and if they areinsensitive to, for example, the
(10:36):
spatial scales, which is alsoone factor we explored in our
study.
So because based on thesecomparison criteria, like the
most complicated Gaussian-basedtwo-step floating catchment area
, is the most robust or mostconsistent.
So that's why I say that thisone is most robust measure to
(10:57):
quantify accessibility to PrEPproviders.
Patrick Sullivan (11:01):
Great.
Was there anything thatsurprised you by the findings of
this study, or were things sortof as you expected when you
went in?
Henry Luan (11:09):
To be honest, I am not surprised by the performance
of the most complicatedGaussian-based approach, but
what surprises me is the simpleproximity measures.
So if I go over our article, wesee that regression
coefficients identified fromthis measure is consistent, is
(11:30):
in the same direction, with theregression coefficients obtained
by using the most complicatedPrEP-accessibility measures.
So my interpretation is thatprobably is because this measure
also accounts for the distancedecay effect.
Again, further away that thezip code is from the PrEP
provider, the loweraccessibility.
(11:52):
So then, like this, distancedecay effect is accounted for in
that measure, and so that's whyI'm thinking that it is
challenging to implement thosemost complicated measures.
I would recommend to use theproximity measures as a starting
point and, in contrast, Iwouldn't recommend using the
(12:13):
density measures, especially forthose areas that are denated by
very arbitrarily definedboundaries for administrative
purposes.
Patrick Sullivan (12:23):
Yeah, because sometimes the responsible entity
, the public health entity, maybe responsible within those
boundaries, but the clinic thatmight serve people is just
across the board.
I mean, it's not all about justthe distance because of
administrative roles and wherethe public health response is
(12:43):
yeah.
So thank you so much for thatexplanation and, on a really
important issue, I want to pivota little bit and talk about
just your process and we callthis Behind the Paper is sort of
trying to understand, becausewe do work in these very
technical areas, but we're also,you know, people who have
colleagues and who haveinterests, and so I like to a
(13:06):
little sense of just how yourway of work and how this work
comes about.
So one of the things I noticedis that you and your co-authors
really have diverse academicbackgrounds and disciplines, and
so I wonder whether you thinklike having that kind of
multi-disciplinary teaminfluences the work, or is it
that people have been trained indifferent areas but also to
(13:26):
gravitate towards a very similarinterest in the end?
So does it make a differencethat there's this sort of
diversity of backgrounds on theteam?
Henry Luan (13:34):
Yeah, this is a great question.
Yeah, I have to admit that itdoes make like difference and
make impact on the finalpresentation or the findings of
our study.
So first here, like I wouldlike to explain this from
different perspectives.
From the statistical modelingperspective, for example, like
including the second author, Dr.
(13:54):
Li from the UK, who is astatistician.
I believe that makes ouranalysis statistical rigorous.
In fact, I had some questionswhere I was fitting the
marginalized two-part log-nodemodels.
It very sophisticated like aset of model.
So because that's the firsttime I fit the model, I had some
questions and that's why Ireached out to Dr.
(14:16):
Li, who I have known for a longtime, and he answered my
questions and make me moreconfident regarding the
statistical rigor of our study.
So that's from the statisticalmodeling perspective.
And, on the other hand, otherco-authors of this study,
including Dr.
Duncan, Dr.
Ransome and, of course, Patrick, are experts in like social
(14:38):
epidemiology, spatialepidemiology, and it's actually
research.
And you also significantlyhelped the presentation of the
final article, including thewriting style.
So how to make the article moreaccessible to the public, health
professionals and researcherswho might be interested in
implementing the methods fortheir own data sets and of
(14:59):
course also what kind ofinformation was the public?
Health professionals andresearchers might be interested
in knowing, so we should presentthem in the article and so on.
So one specific example I wouldlike to mention here is one
suggestion from Dr.
Ransome, from Yale so, whosuggested to include an abstract
video to show how the complexGaussian two-step for the
(15:22):
catchment area approach isimplemented in GIS and then
posted, for example, via YouTube.
So this could help, like folkswho are not trained in GIS but
interested in implementing themethod but don't know where to
start.
So I really like the idea.
But because the method wasimplemented in R rather than in
GIS, so I decided to post the Rscript online rather than making
(15:47):
a video.
But even for that we didn'tmake a video.
I really hope that sharing theR code that implemented the
method could encourage andinspire the practical usability
of our study.
So I'm really hoping that thesescript sharing could encourage
people to implement this methodfor their own data sets.
Patrick Sullivan (16:10):
Yeah, that's great, and we'll be sure to put
the link to that R code in theshow notes as well, so people
are listening, that can find thecode.
That's great.
I think just sharing your codeis great all the way around.
It's a little bit vulnerablebecause I feel like when I write
code it's not always the mostelegant, it's not always like
some people write and it's likethe absolute minimum number of
(16:33):
characters and steps and mine.
It more reflects how I'mthinking.
So it's kind of clunkysometimes, but it's a generous
thing to share your code.
Henry Luan (16:41):
Yeah, but I think that's totally fine, because we
are not trained as programmers.
As long as the function works,the job can be done in different
ways, right?
Patrick Sullivan (16:50):
and I think that's another good point is, I
mean, I love that you talk aboutthe collaboration with your
colleague who's like more on thestatistical side, because to do
this kind of work, where we'rereally trying to like take data
and make knowledge that improveshealth, you really have to
understand every piece of it andI'm not sure any of us by
(17:10):
ourselves could understand youknow, like, why people get PrEP
where they do and why somebodymight want to go to the closest
place and why somebody elsemight want to go to a place
that's not closest to theirhouse.
And and then, layering on allthese methods, it really has to
be done in teams, yeah for sure.
So I just want to sort of note,in terms of your own career,
(17:30):
that you've really I mean, we'retalking about HIV today, but
you've also been involved withother really important public
health challenges, like foodaccess, urban environmental
health, crime, like how do you?
So?
You have a set of methods here,but is there a common theme for
you about, like the areas thatyou've chosen to apply these
(17:50):
methods to to try to improvepublic health?
Because it does?
I mean, I see them unifying,like geography, GIS, expertise,
but what inspires you about someissues to tackle them?
Henry Luan (18:01):
Yeah, that's a great question.
Yeah, so I was also- sometimesI was reflecting what kind of
research I am, you know.
So what is like a narrowedtheme of my research.
So here, in short, I would sayI'm a data-driven and
system-modeling drivenresearcher in different health
phenomena, because for differenthealth data sets I address
(18:22):
different issues, for example,like in our article.
So when the density-basedmeasures are used to quantify
proper accessibility, we havethe skewed data sets with a lot
of zeros and we see that ifthese issues are not properly
addressed, they could lead tobiased results which could be
have like negative impacts onlike subsequent interventions.
(18:44):
Yeah, in terms of where and towhat population the HIV
interventions should be, likeallocated.
Yeah, so in this case, like we,that's why we propose these
complicated to mixture models,two-part mixture models, to
address this issue.
So another example is that so itis common that in HIV and in
(19:04):
other some other like similarsensitive data sets, if they are
released at a small area levels, we have this data spatial
issue because those values thatis smaller than a specific like
five below five, will besuppressed from releasing to the
general public.
So there are a lot ofuncertainties for to analyze
these incomplete data sets.
So that's why, like in myarticle, like I propose to use,
(19:27):
for example, bas approaches toaddress these data spatial issue
.
So to sum up, I would say againa data driven and a set score,
like quantitative method driven,like researcher, who are
interested in addressingdifferent data analysis problems
from different data sets.
That's why, like my application, like span from like food, hiv,
(19:49):
crime, urban environments, etc.
But I see that they are allclosely related to each other.
Patrick Sullivan (19:55):
So yeah, and then this element of space and
services right, I mean you cansee how that applies.
For example, the food access,for sure, the crime you know is
a place based.
There's a place basedconcentration of crime, for
example, of resources.
So see how it's set up at this.
But then there really are theselike sort of access and health
(20:18):
equity issues.
Yeah, that arise out of placeand I think that's an
interesting theme.
So do you have any advice?
For I think that the map basedand sort of place based research
is so of such interest topeople because it's accessible
and because the pathway betweenwhat you identify with your
analyses and how you make healthbetter, to me that's a short
(20:39):
jump.
Like once you see that app ofwhere the people who need the
thing is and where the servicelocations are, the fix is sort
of in the story, like in the map, right.
So if people are interested inthis but maybe a little
intimidated by the methods, whatadvice would you have for
students or early careerprofessionals who say like I'm
(20:59):
really interested in this kindof work but I'm not sure how to
get started?
How might somebody get anintroduction to this?
Henry Luan (21:08):
Yeah, so my, I do have some advices, like because
it's similar to where I startedto, for B is the system only
because I'm not a trendstatistician, so that I see that
there's like similarity betweenmy experience and this to this
question yeah, I do have somepiece of advice is, first, to
use those freely accessibleresources in this field, in GIS,
(21:31):
public health and theirapplications GIS and spatial
analysis applications in publichealth, because today there are
so many like free resources outthere, including, for example,
the textbooks, the journalarticles, youtube videos,
training programs, like atconferences or at different
institutions, so start fromthere.
So here I would like to providesome examples in terms of these
(21:54):
like resources, for example,like for those training programs
at conferences and institutions.
As far as I know, conferencessuch as SER, the Society for
Epidemiologic Research, andinternational conference on
environmental epidemiology theyare now having those pre
conference workshops on GIS,special analysis and the
(22:15):
applications in public health,and some institutions they also
have these like summer workshopslike on this field, from down
from Columbia University andDrexel University and so on, and
some textbooks.
They are also a freelyaccessible on the website
because of the widespread use oflike R and Python programming,
(22:36):
there are also a lot of samplecodes, like in those textbooks,
so which students, anyone canfollow and to reproduce the
results or replicate the study.
And for some journals they alsohave like freely accessible,
like articles on this topic, forexample, spatial and social
epidemiology, internationaljournal of health, geographics
(22:58):
and even like Annals ofEpidemiology, like they also
have like articles on thissubject.
So start from there, like Iread these articles and using
some simple data sets from theseresources.
So, and I believe that andthere is no doubt that with more
practice, like the students andanyone who would like to start
this topic will get to do moreabout this topic and get to be
(23:21):
more comfortable and confidentabout this topic.
And finally, I think anotherpiece of advice is that do not
be shy to reach out to theexperts in this field.
So my experience is that mostscholars would be happy if you
read their article, their workand reach out to them for
(23:41):
questions.
This is also my own experience.
For example, whenever I have aquestion on sesquimolony, I
reach out to those experts.
That's how I got connected withthe second author of this study
in this paper.
So that's my suggestions.
Patrick Sullivan (23:57):
Yeah, and what's amazing to me is when I
do that, people always respond.
I feel like in our world youcan reach out to someone who
you've never met before and sayI'm working on this analysis and
I saw you had worked on thesemethods and I have a few
questions and people areincredibly responsive.
Henry Luan (24:13):
Exactly, yeah, that's why, yeah, don't be shy,
just reach out, you'll get someresponses, great.
Patrick Sullivan (24:20):
Well, this has been a great conversation, both
about the substance of yourarticle and some ideas for these
sort of research collaborationsand how people might get
started.
Do you have any last thoughtsthat you want to share with our
podcast listeners of, eitherabout this analysis or advice
about doing these kinds ofanalyses?
Henry Luan (24:38):
Yeah, just do it like, go ahead.
We're simple because of the R&Rmovement in spatial analysis,
like that's reproducibility andreplicability.
So a lot of work, they have,shared the code they have and
some even share the data set.
If that's not, the data set isnot that sensitive.
So just go ahead and do somepractice and they will be like
(24:59):
at least like be more familiarwith this topic.
And yeah, that's my last pieceof suggestion.
Patrick Sullivan (25:07):
Yeah, and then once you have that level of
comfort, you can start keepingsome pieces of the code but
swapping in a different kind ofservice type, that's a great way
to start, I think the idea ofjust jumping in and seeing how
the code works, yeah, and thenand a lot of what makes research
interesting is bringing goodquestions.
Exactly, the methods are somemethods, but people who listen,
who may pick up.
(25:28):
You know, look at some of thiscode or play around with it,
then we'll bring that to anotherimportant question that they
understand, that I wouldn't knowabout.
You know, and that's that's howwe propagate this information.
Henry Luan (25:38):
Yeah, and something like I would like to say that
those interdisciplinarycollaboration and discussions
are also important, becausethat's also the way.
From these like brainstorming,talking with, like experts in
different fields, then we canthink of some innovative
solutions to a specific researchquestion, and sometimes even
the research questions arisefrom these interdisciplinary
(26:01):
discussions, yeah.
So yeah, go out to talk topeople.
Yeah, different topics, yeah.
Patrick Sullivan (26:08):
I do.
This does just make me thinkthat, like so much of my own,
like professional history wasaround going to certain
conferences mostly HIV andpublic health conferences for me
and meeting people like I hadread, you know, and whose names
I knew, and introducing myselfat posters, presentations, and,
and I think, in some ways, thatthe pandemic has made some of
(26:31):
the conferences and things moreaccessible because you can do
them online and you're nottraveling around.
But there is that piece of justlike making connections with
people, yeah, conferences andbuilding your network, that I
think that this, you know, thegenerations of researchers who
are really coming into this, yes, are going to have to develop
different tools in ways, becausefor me it was very much driven
(26:53):
by and that's why I went to theconferences, yes, to hear the
talks, but it was mostly to meetthe people giving the talks.
Henry Luan (26:58):
Yes, I can't agree more, to be honest.
Yeah, and I feel like onlythrough those in person
conversations we would be ableto know if I can get along with
this like person to do, tocollaborate with him or her, you
know.
So, yeah, to me this is likepersonal connections always
matters more than those on likevirtual conversations.
Patrick Sullivan (27:19):
Right.
Although in full disclosurewe're recording this today over
a zoom call, so it's hopefullythis sounds like we're sitting
in a room having a conversation.
Yeah, ups and downsides oftechnology.
Henry Luan (27:29):
Exactly yeah.
Patrick Sullivan (27:30):
Well, that brings us to the end of the
episode.
Thank you again, Dr.
Luan, for joining us today.
It was so nice to talk to youjust about your thoughts about
this field, and obviously I loveand I'm interested in the
application that you've used tobring about this knowledge and,
you know, I hope we haveopportunities to work together
(27:51):
again.
A lot of shared interests, so,but thanks for sharing this with
us.
Henry Luan (27:56):
Yeah, thanks for having me.
Yeah, I'm glad I, yeah, I havethis opportunity to share this
work, introduce the principlesbehind those methods and
hopefully this like episodewould inspire more researchers
to adopt this method in theirown work.
And don't hesitate to reach outto me if you need any
suggestions or like need, likeadvice is all implementing those
methods.
Patrick Sullivan (28:17):
So two things.
One, I'd encourage people tocheck the show notes.
We're going to put the link tothe video that you mentioned and
some of the other resources,and I'm going to follow up with
you in a month or two and justsee how many people have reached
out.
I think I have a goal of fivepeople who listened to this and
reach out and say help, get mestarted
Henry Luan (28:37):
So, yeah, great, let's see how that goes.
Patrick Sullivan (28:40):
All right, I'm your host, Patrick Sullivan.
Thanks again for tuning intothis episode and see you next
time on EPITalk (28:45):
Behind the Paper brought to you by Annals
of Epidemiology, the officialjournal of the American College
of Epidemiology.
Hello, you're listening to EPITalk
Paper, a monthly podcast fromthe Annals of Epidemiology.
I'm Patrick Sullivan,Editor-in-Chief of the journal,
and in this series we take youbehind the scenes of some of the
latest epidemiologic researchfeatured in our journal.
Today we're here with Dr.
(00:34):
Henry Luan to discuss hisarticle "Spatial Accessibility
of pre-exposure prophylaxis(PrEP): different measure
choices and the implications fordetecting shortage areas and
examining its association withsocial determinants of health.
" You can read the full articleonline in the October 2023 issue
of the journal at www.
annalsofepidemiology.
org.
(00:56):
I'll tell you just a bit aboutour guest today.
Dr.
Henry Luan is an assistantprofessor in spatial data
science and spatial epidemiologyin the Department of Geography
at the University of Oregon.
Dr.
Luan applies and developsBayesian spatial temporal
statistical models and advancedGIS approaches to investigate
how health phenomena, especiallyhealth events related to HIV,
(01:19):
vary over space, time and raceethnicity, and how socioeconomic
, demographic, physical andbuilt environmental factors
contribute to these space, timeand cross-race ethnicity
variations.
Dr.
Luan, thank you for joining ustoday.
Henry Luan (01:32):
Thank you for having me here.
I'm excited to be here.
Patrick Sullivan (01:35):
And I'm excited to talk some about your
study, which gets to geospatialepi, which I'm excited about,
and which gets to pre-exposureprophylaxis, which I'm excited
about.
So tell me a little bit aboutthe purpose of the study.
What were you really aiming toanswer with the analysis that
you did?
Henry Luan (01:51):
Yeah for sure.
So the major purpose of thisstudy is to compare different
spatial accessibility measuresof PrEP providers and then
examine how different choices ofthe measures affect the
detection of PrEP providershortage areas, as well as the
association between PrEPaccessibility and social
determinants of health variables, as suggested by the title of
(02:14):
the article.
So by something I would like toclarify here is the difference
between the accessibilitymeasures for individuals and for
areas or for a neighborhood,because they're different.
So for an individual, we canderive their activity spaces via
techniques such as GPS equippeddevices and then derive how a
(02:36):
PrEP provider is accessible toan individual.
But for an area we cannot dothat because we cannot track
every individual residing in thesame area and instead we can
only calculate PrEPaccessibility for the entire
area or calculate theaccessibility for a
representative point of thatarea, for example the geometric
(02:57):
centroid or the populationweighty centroid of an area.
So in this sense they aredifferent.
I would like to say that beforethe widespread use of GPS
equipped devices, so these arealevel accessibility measures are
used as proxies of individuallevel accessibility.
So to- in short, this articlecompares the accessibility
(03:20):
measures for areas rather thanfor individuals and the main
questions we aim to answer.
There are two questions we wantto answer.
First is how does the choice ofPrEP accessibility measures
affect PrEP shortage areadetection?
And second is how does thechoice of PrEP accessibility
measures affect the findingsregarding the association
(03:41):
between PrEP accessibility andsocial determinants of health
variables?
Patrick Sullivan (03:45):
That's really interesting.
I want to break down a couplepieces of this, just because I
think maybe we both think aboutprep and geospatial stuff.
But for people who may not beso familiar this idea of PrEP
accessibility in areas in whichPrEP is accessible, why is it
important to know that?
If I'm in Portland and you usethis method and you tell me
(04:05):
where prep is more or lessaccessible, how does that
translate to public healthaction?
How would that data be used byproviders, for example?
Henry Luan (04:13):
I think.
First because we care about whohave sufficient accessibility
to PrEP providers, especiallyfor those population who need
them most.
And if there is a mismatchbetween the supply and the
demand, so we want to make surethose demands are made.
And we know that these spatialaccessibility is not constant
(04:36):
over space.
There are always some areasthat are in short accessibility
and some areas where they havegood accessibility.
So that's, I think, from thisperspective, we need to
visualize and to know where theaccessibility is now and doesn't
meet the need.
Patrick Sullivan (04:51):
And it's very actionable, right?
Because if I'm responsible forHIV prevention in this city and
I can look and say, oh, I knowthe areas of the city where new
HIV diagnoses are occurring, andthen you are able to use these
kind of data to show where theremay be low service areas, and
if those two things line up likehigh risk and low service,
(05:12):
that's the most impactful place,exactly.
So you sort of mentioned themethods a little bit.
So you must have inputs ofwhere the providers are and then
what are the methods that sortof lead to identifying these
areas.
How does that work?
Henry Luan (05:28):
Yeah, actually like a great question, because we
compared four differentcategories of measures in our
study.
So the density-based, theproximity-based, the more
complicated two-step floatingcatchment area-based approach
and finally, the most complexone, the Gaussian version of the
two-step floating catchmentarea approach.
So I would like to brieflyintroduce the principles behind
(05:51):
these four categories ofmeasures.
First is the density-basedmeasures, so this is the
simplest one.
So we use the count of PrEPproviders within an area, so in
our context, the zip code, thezip code compilation areas.
So we use the count in thatarea over the total demand.
So this demand, in our study weuse two proxies One is the
(06:14):
total population, the other isthe total account of new HIV
diagnosis and we also comparedhow these different proxies of
population demand affect theidentification of PrEP shortage
areas.
So this is the first one, firstcategory.
The second category isproximity, so we simply
calculate the distance along theroad network, so it's not like
(06:38):
straight-line distance.
We do account for the network,like in structure in reality,
and then we calculate thedistance from the zip code
centroid to the nearest PrEPprovider, or the average
distance to the three or fiveclosed PrEP providers, because
we want to account for more thanone provider, because usually
(06:59):
they have been researched, thatreviewed, that the residents
probably do not necessarily goto the nearest PrEP provider.
So that's the second type.
And for the third type is morecomplicated, that's the two-step
floating catchment areaapproach.
So, as the name suggests, thereare two steps in the calculation
of this measure.
So first we create thecatchment area for each PrEP
(07:24):
provider, so that's from thesupply side.
So based on, for example, thetravel distance along the road
network, based on a specificcut-off value, for example 10
minutes driving, 15 minutesdriving or 30 minutes driving,
and then we calculate thesupply-demand ratio.
So for example, let's assume aprep provider can serve five zip
(07:46):
codes when we serve, that isthe centroid of the zip codes
for inside that catchment area.
And then, if we assume thateach zip code has one resident
only, and then the supply-demandratio will be 0.2.
That means not the fullcapacity will be used by each
zip code.
There is a computation along thesupply side but of course the
(08:09):
actual population can beaccounted for to calculate the
supply-demand ratio.
So that's the first step inthis method.
The second step is that foreach zip code, so that is at the
demand side, we identify whichprep providers are accessible to
that zip code.
And then we add up thesupply-demand ratios associated
(08:29):
with different accessible prepproviders so that the final
results are the accessibilitymeasures for each zip code.
So you can see that thismeasure accounts for the
computation from the demand sideas well as the spatial
variations in the demandpopulation.
So that's the third type ofaccessibility measures.
The last one, which is the mostcomplicated one, the
(08:51):
Gaussian-based two-step floatingcatchment area, goes one step
further compared with thestandard two-step floating
catchment area approach.
That is, we add Gaussianfunction to account for the
distance decay effect.
In other words, we account forthe fact that a prep provider
further away from the zip code'scentroid will have lower
(09:15):
accessibility.
So this is not the case for thestandard version of that method
.
So because as long as the prepproviders are within the cutoff
value in the standard method,they are assumed to be equally
accessible, but that's not thecase in the Gaussian version.
So that's the four categoriesof the measure that we used and
compared in our study.
Patrick Sullivan (09:37):
Thanks for that explanation.
From a methodologic point ofview, how did those measures
compare, because the bottom linehere is sort of trying to
identify these PrEP shortageareas.
So were they pretty muchaligned, or was there one that
you felt like ended up beingmore informative than the other?
Henry Luan (09:53):
I would say the most complicated one is more
informative than the other onesbecause, as based on the
introduction to the principlesof these methods, it accounts
for different factors into thecalculation.
But here I wouldn't conclude interms of which measures can
more accurately detect the PrEPshortage areas, because we don't
(10:14):
know the actual ground.
Choose which area has a higheraccessibility, which doesn't.
But in state I would comparethem based on different factors,
for example if they account forthe variations in the demands,
if they account for thecomputations among those demand
populations and if they areinsensitive to, for example, the
(10:36):
spatial scales, which is alsoone factor we explored in our
study.
So because based on thesecomparison criteria, like the
most complicated Gaussian-basedtwo-step floating catchment area
, is the most robust or mostconsistent.
So that's why I say that thisone is most robust measure to
(10:57):
quantify accessibility to PrEPproviders.
Patrick Sullivan (11:01):
Great.
Was there anything thatsurprised you by the findings of
this study, or were things sortof as you expected when you
went in?
Henry Luan (11:09):
To be honest, I am not surprised by the performance
of the most complicatedGaussian-based approach, but
what surprises me is the simpleproximity measures.
So if I go over our article, wesee that regression
coefficients identified fromthis measure is consistent, is
(11:30):
in the same direction, with theregression coefficients obtained
by using the most complicatedPrEP-accessibility measures.
So my interpretation is thatprobably is because this measure
also accounts for the distancedecay effect.
Again, further away that thezip code is from the PrEP
provider, the loweraccessibility.
(11:52):
So then, like this, distancedecay effect is accounted for in
that measure, and so that's whyI'm thinking that it is
challenging to implement thosemost complicated measures.
I would recommend to use theproximity measures as a starting
point and, in contrast, Iwouldn't recommend using the
(12:13):
density measures, especially forthose areas that are denated by
very arbitrarily definedboundaries for administrative
purposes.
Patrick Sullivan (12:23):
Yeah, because sometimes the responsible entity
, the public health entity, maybe responsible within those
boundaries, but the clinic thatmight serve people is just
across the board.
I mean, it's not all about justthe distance because of
administrative roles and wherethe public health response is
(12:43):
yeah.
So thank you so much for thatexplanation and, on a really
important issue, I want to pivota little bit and talk about
just your process and we callthis Behind the Paper is sort of
trying to understand, becausewe do work in these very
technical areas, but we're also,you know, people who have
colleagues and who haveinterests, and so I like to a
(13:06):
little sense of just how yourway of work and how this work
comes about.
So one of the things I noticedis that you and your co-authors
really have diverse academicbackgrounds and disciplines, and
so I wonder whether you thinklike having that kind of
multi-disciplinary teaminfluences the work, or is it
that people have been trained indifferent areas but also to
(13:26):
gravitate towards a very similarinterest in the end?
So does it make a differencethat there's this sort of
diversity of backgrounds on theteam?
Henry Luan (13:34):
Yeah, this is a great question.
Yeah, I have to admit that itdoes make like difference and
make impact on the finalpresentation or the findings of
our study.
So first here, like I wouldlike to explain this from
different perspectives.
From the statistical modelingperspective, for example, like
including the second author, Dr.
(13:54):
Li from the UK, who is astatistician.
I believe that makes ouranalysis statistical rigorous.
In fact, I had some questionswhere I was fitting the
marginalized two-part log-nodemodels.
It very sophisticated like aset of model.
So because that's the firsttime I fit the model, I had some
questions and that's why Ireached out to Dr.
(14:16):
Li, who I have known for a longtime, and he answered my
questions and make me moreconfident regarding the
statistical rigor of our study.
So that's from the statisticalmodeling perspective.
And, on the other hand, otherco-authors of this study,
including Dr.
Duncan, Dr.
Ransome and, of course, Patrick, are experts in like social
(14:38):
epidemiology, spatialepidemiology, and it's actually
research.
And you also significantlyhelped the presentation of the
final article, including thewriting style.
So how to make the article moreaccessible to the public, health
professionals and researcherswho might be interested in
implementing the methods fortheir own data sets and of
(14:59):
course also what kind ofinformation was the public?
Health professionals andresearchers might be interested
in knowing, so we should presentthem in the article and so on.
So one specific example I wouldlike to mention here is one
suggestion from Dr.
Ransome, from Yale so, whosuggested to include an abstract
video to show how the complexGaussian two-step for the
(15:22):
catchment area approach isimplemented in GIS and then
posted, for example, via YouTube.
So this could help, like folkswho are not trained in GIS but
interested in implementing themethod but don't know where to
start.
So I really like the idea.
But because the method wasimplemented in R rather than in
GIS, so I decided to post the Rscript online rather than making
(15:47):
a video.
But even for that we didn'tmake a video.
I really hope that sharing theR code that implemented the
method could encourage andinspire the practical usability
of our study.
So I'm really hoping that thesescript sharing could encourage
people to implement this methodfor their own data sets.
Patrick Sullivan (16:10):
Yeah, that's great, and we'll be sure to put
the link to that R code in theshow notes as well, so people
are listening, that can find thecode.
That's great.
I think just sharing your codeis great all the way around.
It's a little bit vulnerablebecause I feel like when I write
code it's not always the mostelegant, it's not always like
some people write and it's likethe absolute minimum number of
(16:33):
characters and steps and mine.
It more reflects how I'mthinking.
So it's kind of clunkysometimes, but it's a generous
thing to share your code.
Henry Luan (16:41):
Yeah, but I think that's totally fine, because we
are not trained as programmers.
As long as the function works,the job can be done in different
ways, right?
Patrick Sullivan (16:50):
and I think that's another good point is, I
mean, I love that you talk aboutthe collaboration with your
colleague who's like more on thestatistical side, because to do
this kind of work, where we'rereally trying to like take data
and make knowledge that improveshealth, you really have to
understand every piece of it andI'm not sure any of us by
(17:10):
ourselves could understand youknow, like, why people get PrEP
where they do and why somebodymight want to go to the closest
place and why somebody elsemight want to go to a place
that's not closest to theirhouse.
And and then, layering on allthese methods, it really has to
be done in teams, yeah for sure.
So I just want to sort of note,in terms of your own career,
(17:30):
that you've really I mean, we'retalking about HIV today, but
you've also been involved withother really important public
health challenges, like foodaccess, urban environmental
health, crime, like how do you?
So?
You have a set of methods here,but is there a common theme for
you about, like the areas thatyou've chosen to apply these
(17:50):
methods to to try to improvepublic health?
Because it does?
I mean, I see them unifying,like geography, GIS, expertise,
but what inspires you about someissues to tackle them?
Henry Luan (18:01):
Yeah, that's a great question.
Yeah, so I was also- sometimesI was reflecting what kind of
research I am, you know.
So what is like a narrowedtheme of my research.
So here, in short, I would sayI'm a data-driven and
system-modeling drivenresearcher in different health
phenomena, because for differenthealth data sets I address
(18:22):
different issues, for example,like in our article.
So when the density-basedmeasures are used to quantify
proper accessibility, we havethe skewed data sets with a lot
of zeros and we see that ifthese issues are not properly
addressed, they could lead tobiased results which could be
have like negative impacts onlike subsequent interventions.
(18:44):
Yeah, in terms of where and towhat population the HIV
interventions should be, likeallocated.
Yeah, so in this case, like we,that's why we propose these
complicated to mixture models,two-part mixture models, to
address this issue.
So another example is that so itis common that in HIV and in
(19:04):
other some other like similarsensitive data sets, if they are
released at a small area levels, we have this data spatial
issue because those values thatis smaller than a specific like
five below five, will besuppressed from releasing to the
general public.
So there are a lot ofuncertainties for to analyze
these incomplete data sets.
So that's why, like in myarticle, like I propose to use,
(19:27):
for example, bas approaches toaddress these data spatial issue
.
So to sum up, I would say againa data driven and a set score,
like quantitative method driven,like researcher, who are
interested in addressingdifferent data analysis problems
from different data sets.
That's why, like my application, like span from like food, hiv,
(19:49):
crime, urban environments, etc.
But I see that they are allclosely related to each other.
Patrick Sullivan (19:55):
So yeah, and then this element of space and
services right, I mean you cansee how that applies.
For example, the food access,for sure, the crime you know is
a place based.
There's a place basedconcentration of crime, for
example, of resources.
So see how it's set up at this.
But then there really are theselike sort of access and health
(20:18):
equity issues.
Yeah, that arise out of placeand I think that's an
interesting theme.
So do you have any advice?
For I think that the map basedand sort of place based research
is so of such interest topeople because it's accessible
and because the pathway betweenwhat you identify with your
analyses and how you make healthbetter, to me that's a short
(20:39):
jump.
Like once you see that app ofwhere the people who need the
thing is and where the servicelocations are, the fix is sort
of in the story, like in the map, right.
So if people are interested inthis but maybe a little
intimidated by the methods, whatadvice would you have for
students or early careerprofessionals who say like I'm
(20:59):
really interested in this kindof work but I'm not sure how to
get started?
How might somebody get anintroduction to this?
Henry Luan (21:08):
Yeah, so my, I do have some advices, like because
it's similar to where I startedto, for B is the system only
because I'm not a trendstatistician, so that I see that
there's like similarity betweenmy experience and this to this
question yeah, I do have somepiece of advice is, first, to
use those freely accessibleresources in this field, in GIS,
(21:31):
public health and theirapplications GIS and spatial
analysis applications in publichealth, because today there are
so many like free resources outthere, including, for example,
the textbooks, the journalarticles, youtube videos,
training programs, like atconferences or at different
institutions, so start fromthere.
So here I would like to providesome examples in terms of these
(21:54):
like resources, for example,like for those training programs
at conferences and institutions.
As far as I know, conferencessuch as SER, the Society for
Epidemiologic Research, andinternational conference on
environmental epidemiology theyare now having those pre
conference workshops on GIS,special analysis and the
(22:15):
applications in public health,and some institutions they also
have these like summer workshopslike on this field, from down
from Columbia University andDrexel University and so on, and
some textbooks.
They are also a freelyaccessible on the website
because of the widespread use oflike R and Python programming,
(22:36):
there are also a lot of samplecodes, like in those textbooks,
so which students, anyone canfollow and to reproduce the
results or replicate the study.
And for some journals they alsohave like freely accessible,
like articles on this topic, forexample, spatial and social
epidemiology, internationaljournal of health, geographics
(22:58):
and even like Annals ofEpidemiology, like they also
have like articles on thissubject.
So start from there, like Iread these articles and using
some simple data sets from theseresources.
So, and I believe that andthere is no doubt that with more
practice, like the students andanyone who would like to start
this topic will get to do moreabout this topic and get to be
(23:21):
more comfortable and confidentabout this topic.
And finally, I think anotherpiece of advice is that do not
be shy to reach out to theexperts in this field.
So my experience is that mostscholars would be happy if you
read their article, their workand reach out to them for
(23:41):
questions.
This is also my own experience.
For example, whenever I have aquestion on sesquimolony, I
reach out to those experts.
That's how I got connected withthe second author of this study
in this paper.
So that's my suggestions.
Patrick Sullivan (23:57):
Yeah, and what's amazing to me is when I
do that, people always respond.
I feel like in our world youcan reach out to someone who
you've never met before and sayI'm working on this analysis and
I saw you had worked on thesemethods and I have a few
questions and people areincredibly responsive.
Henry Luan (24:13):
Exactly, yeah, that's why, yeah, don't be shy,
just reach out, you'll get someresponses, great.
Patrick Sullivan (24:20):
Well, this has been a great conversation, both
about the substance of yourarticle and some ideas for these
sort of research collaborationsand how people might get
started.
Do you have any last thoughtsthat you want to share with our
podcast listeners of, eitherabout this analysis or advice
about doing these kinds ofanalyses?
Henry Luan (24:38):
Yeah, just do it like, go ahead.
We're simple because of the R&Rmovement in spatial analysis,
like that's reproducibility andreplicability.
So a lot of work, they have,shared the code they have and
some even share the data set.
If that's not, the data set isnot that sensitive.
So just go ahead and do somepractice and they will be like
(24:59):
at least like be more familiarwith this topic.
And yeah, that's my last pieceof suggestion.
Patrick Sullivan (25:07):
Yeah, and then once you have that level of
comfort, you can start keepingsome pieces of the code but
swapping in a different kind ofservice type, that's a great way
to start, I think the idea ofjust jumping in and seeing how
the code works, yeah, and thenand a lot of what makes research
interesting is bringing goodquestions.
Exactly, the methods are somemethods, but people who listen,
who may pick up.
(25:28):
You know, look at some of thiscode or play around with it,
then we'll bring that to anotherimportant question that they
understand, that I wouldn't knowabout.
You know, and that's that's howwe propagate this information.
Henry Luan (25:38):
Yeah, and something like I would like to say that
those interdisciplinarycollaboration and discussions
are also important, becausethat's also the way.
From these like brainstorming,talking with, like experts in
different fields, then we canthink of some innovative
solutions to a specific researchquestion, and sometimes even
the research questions arisefrom these interdisciplinary
(26:01):
discussions, yeah.
So yeah, go out to talk topeople.
Yeah, different topics, yeah.
Patrick Sullivan (26:08):
I do.
This does just make me thinkthat, like so much of my own,
like professional history wasaround going to certain
conferences mostly HIV andpublic health conferences for me
and meeting people like I hadread, you know, and whose names
I knew, and introducing myselfat posters, presentations, and,
and I think, in some ways, thatthe pandemic has made some of
(26:31):
the conferences and things moreaccessible because you can do
them online and you're nottraveling around.
But there is that piece of justlike making connections with
people, yeah, conferences andbuilding your network, that I
think that this, you know, thegenerations of researchers who
are really coming into this, yes, are going to have to develop
different tools in ways, becausefor me it was very much driven
(26:53):
by and that's why I went to theconferences, yes, to hear the
talks, but it was mostly to meetthe people giving the talks.
Henry Luan (26:58):
Yes, I can't agree more, to be honest.
Yeah, and I feel like onlythrough those in person
conversations we would be ableto know if I can get along with
this like person to do, tocollaborate with him or her, you
know.
So, yeah, to me this is likepersonal connections always
matters more than those on likevirtual conversations.
Patrick Sullivan (27:19):
Right.
Although in full disclosurewe're recording this today over
a zoom call, so it's hopefullythis sounds like we're sitting
in a room having a conversation.
Yeah, ups and downsides oftechnology.
Henry Luan (27:29):
Exactly yeah.
Patrick Sullivan (27:30):
Well, that brings us to the end of the
episode.
Thank you again, Dr.
Luan, for joining us today.
It was so nice to talk to youjust about your thoughts about
this field, and obviously I loveand I'm interested in the
application that you've used tobring about this knowledge and,
you know, I hope we haveopportunities to work together
(27:51):
again.
A lot of shared interests, so,but thanks for sharing this with
us.
Henry Luan (27:56):
Yeah, thanks for having me.
Yeah, I'm glad I, yeah, I havethis opportunity to share this
work, introduce the principlesbehind those methods and
hopefully this like episodewould inspire more researchers
to adopt this method in theirown work.
And don't hesitate to reach outto me if you need any
suggestions or like need, likeadvice is all implementing those
methods.
Patrick Sullivan (28:17):
So two things.
One, I'd encourage people tocheck the show notes.
We're going to put the link tothe video that you mentioned and
some of the other resources,and I'm going to follow up with
you in a month or two and justsee how many people have reached
out.
I think I have a goal of fivepeople who listened to this and
reach out and say help, get mestarted
Henry Luan (28:37):
So, yeah, great, let's see how that goes.
Patrick Sullivan (28:40):
All right, I'm your host, Patrick Sullivan.
Thanks again for tuning intothis episode and see you next
time on EPITalk (28:45):
Behind the Paper brought to you by Annals
of Epidemiology, the officialjournal of the American College
of Epidemiology.
Popular Podcasts
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy And Charlamagne Tha God!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Decisions, Decisions
Welcome to "Decisions, Decisions," the podcast where boundaries are pushed, and conversations get candid! Join your favorite hosts, Mandii B and WeezyWTF, as they dive deep into the world of non-traditional relationships and explore the often-taboo topics surrounding dating, sex, and love. Every Monday, Mandii and Weezy invite you to unlearn the outdated narratives dictated by traditional patriarchal norms. With a blend of humor, vulnerability, and authenticity, they share their personal journeys navigating their 30s, tackling the complexities of modern relationships, and engaging in thought-provoking discussions that challenge societal expectations. From groundbreaking interviews with diverse guests to relatable stories that resonate with your experiences, "Decisions, Decisions" is your go-to source for open dialogue about what it truly means to love and connect in today's world. Get ready to reshape your understanding of relationships and embrace the freedom of authentic connections—tune in and join the conversation!