OHDSI Home | Forums | Wiki | Github

Deriving pediatric growth charts from longitudinal health records

Researchers
#1

Hey all, I’m currently sitting in a fda workshop on longterm pediatric safety research (where I’ll be presenting ohdsi tomorrow). Here’s the info: http://www.fda.gov/NewsEvents/MeetingsConferencesWorkshops/ucm477639.htm

During the discussion, a participant claimed that the pediatric growth charts used by all pediatricians (from who or otherwise) were in fact originally derived from cross sectional survey data. He hypothesized that if we were to recalculate the growth charts using longitudinal data captured from sequential pediatric visits (height, weight, bmi, z score), that we might see different ‘norms’ which could subsequently change treatment paradigms. This is the first I’ve heard of this ‘problem’ but it screamed of something that the ohdsi community could do, if data were available. I’m not an expert in this space, so just dropping this note down while fresh in my mind to see if this resonates with anyone and if we should heed the call to action…

#2
  • the participant is correct (at least for the CDC 2 year old and
    higher) that the data are from sequential cross sectional surveys and are
    not based on longitudinal data. it is an interesting hypothesis that
    longitudinal data will tell us something different. certainly
    pediatricians create “models” in their heads for patients – johnny was at
    the 20th percentile at the last 4 encounters and now is at the 50th
    percentile for their age (humm!!!). It sounds like the suggestion is to
    create models for an individual patient based on longitudinal data. I’d be
    a bit surprised if there were big impacts on care decisions – of course
    the things we are surprised by are the most interesting.
#3

The two sets in common use are built from population-longitudinal, person-cross-sectional surveys. There’s been a fair amount of discussion among growth chart geeks about use of growth trajectories rather than point assessments, especially in infancy, which is how I think what you’re describing would play out in practice. The idea hasn’t really gotten widespread traction, though that’s hardly a good indication of its merit.

We’re gearing up to look at growth trajectories and relationship to antibiotic exposure using PCORnet, and we’d be happy to collaborate on the OHDSI side if there’re others interested in looking at this kind of question. There’re interesting conversations to be had about methodology that’re beyond my expertise, but happy to sit with the experts and learn.

The other recurring issue that arises is creation of growth norms for cohorts of children with a given chronic disease; this is another topic that the large populations in the OHDSI community could make tractable.

#4

Great point @bailey, why in the world do we assume that a growth chart is
one-size-fits-all? Say (hypothetically) a baby girl was born 6 weeks
premature and was found to have cardiovascular issues that resulted in
fluid restrictions and had a period of ‘failure to thrive’, then ‘normal’
in that case might be quite different from the post-term baby boy who comes
out healthy and ready to play linebacker for Ohio State.

a simple stratification by important diseases would seem a useful
contribution to the community…but I’ll defer to all the pediatric
researchers in the community to give a +1 on that…@overhage has a good
intuition about such things, so is probably right that is might be that we
show no difference…but I’d personally feel better empirically
demonstrating that than just assuming it’s the case…

i don’t have any data in US that could help with this, but I’ll take a look
to see if CPRD would have enough longitudinal height/weight measurements,
and if so, would be happy to work with others to put together a protocol to
tackle this.

#5

Hi,

We have published two papers about blood pressure norms based on cross sectional data:

  1. Blood pressure percentiles by age and height for children and adolescents in Tehran, Iran
  2. Blood pressure percentiles by age and body mass index for adults

In these two papers, we used GAMLSS (Generalized Additive Models for Location, Scale and Shape ) methods in R. In first one, we conclude that BP normal percentiles may differ between nations (by comparing USA , Germany and Iran norms) .In the second paper, we suggest that it is a good idea to calculate BP percentile based on age and BMI rather than age and height.

We also published another paper based on sequential cross sectional survey and estimate cohort effects based on APC (Age-Period-Cohort) method.

In this paper we conclude that the risk of hypertension (based on BP) may increase not only by increasing age, but also with birth cohort (only based on Temporal changes).

In changing treatment paradigms, these factors may also be considered.

Thanks.

#6

Well, let’s all look at our Achilles stats for that.

I see some data under these concept_ids

Do we all use 3013762? (Columbia?, Regenstrief?, Stanford?)

Yet another reason to publish an each data partner wiki page a subset of our non-sensitive Achilles data.

Well, looking at the age at observation - the study becomes a no-go.

Pasted image1818×905 76.9 KB

(btw, image upload in this forum platform (from clipboard) is VERY IMPRESSIVE and convenient)

#7

@Vojtech_Huser – try doing “blood pressure”. I believe we got ~90 LOINC codes when we tried it.

#8

Always fun. As one potential point of calibration, this document contains the concept IDs we’re using in PEDSnet for heights, weights, and SBP/DBPs: https://pedsnet.org/documents/11/PEDSnet_CDM_2.2pre_Spec.pdf . I’d show you cool Achilles pictures like @Vojtech_Huser , but our Achilles reports don’t cover Measurements yet. :blush:

#9

Hi all,

I just bumped into this topic and think I can contrubute some thoughts and graphs to the discussion. Hope you can use some of it.

I like the idea of generation a new growth chart from data. But it will be very important to pick the right database for it. Unfortunately in our own IPCI database (GP database) this can not be done. It will be heavily biased. You need to have a database where these things are recorded for all children in the population and not biased by any medical reason for a visit.In the Netherlands we should have such a registration because for the first 4 years after birth (almost) everyone goes to a specialized doctor because of national protocol. And here they monitor these things. Don’t know the English term but in dutch it is called the “consultatie buro”.

Bailey asks for profiling pictures for bloodpressure. Since Achilles does not support measurments (yet…) I have some graphs from our IPCI framework.
I’m currently working on a “cohort profiler” where we can compare characteristics of all of our codes for defined cohorts. I demonstrated a early version of this to Patrick some months ago. It has a lot of different overviews and graphs, I just picked two interesting ones for this topic. I just took two “cohorts”: male and females.

Here is a graph showing the patients age when registering the first systolic blood pressure (male in blue and females in red). Resolution and counts per month.

It clearly shows a huge peak around the age of 16. The reason for this is that it is part of the protocol when prescribing OAC for the first time. Most females start using this around this age. This reason for encounter is a pretty specif one because most of these patients are not there because of a medical reason in this case. The effect will be that the average in this age group will have a lot of “normal” values.

Other peaks at the ages of 60 and 70 because of protocols.

The next graph show the average bloodpressure by age of all measurements. Peaks at the extreme ages are caused by low number of patients.

And indeed: when we compare the SBP between males and females, we see a drop from this age and further. I’m not a cardiologist, but probably caused by the effect described above. So, for the females around the age of 16-20 the average value of the SBP would be much closer to the “average SBP of the total population”. But in the other agegroups the result will be more and more biased.

Similar things will happen to weight, length and BMI.

This little example shows the bias effect in our IPCI database. I hope this can contribute to your discussions.

I hope someone in the community has the right database to do this. And of course this is country depended since the Dutch population is the tallest around on this planet. :wink:

Greetings from the Netherlands,
Marcel de Wilde

#10

Hi again,

I just keep thinking about this topic and want to add two more graphs. Here are the graphs from the cohort profiler for male and female again, but now I selected the average weight and height by age.

Weight (in kg):

Weight.png700×428 20.4 KB

Height (in meters)

Height.png698×502 19.9 KB

I was amazed by the similarity under the age of 14 for both graphs and the sudden change after this age. Is this really like this? Also in other databases?

Cheers,
Marcel

#11

these graphs are spot on with clinical wisdom. the only thing that
surprised me a little is that we don’t see females surpassing males at age
12 or 13 for a while – there may be a hint of that but otherwise spot on.
puberty has powerful effects

#12

Hi Marc,

thanks for your reply. I assume your “spot on” comments are only referring to the weight and height graphs. Or can the effect on the systolic blood pressure also be caused by puberty?

Cheers,

Marcel

#13

yup to BP as well. probably different mechanisms (likely androgens like
testosterone for blood pressure). women catch up male bps in late life
thought to be due to shift is testosterone/estrogen balance after menopause

#14

Wow, no other physician I spoke about this graph has realized this effect before. Until now they all agreed with my amateur theory and this was biased by the OAC prescription. I have three weeks off from now, but after my holyday I will double check this by filtering out the “BP when combined with a OAC prescription” and see if I get the same lower results for females vs males. Cant wait to try. Thanks a lot for this explanation!!!

#15

you are not the first to look at this question. see for example
Tu W, Eckert GJ, Saha C, Pratt JH. Synchronization of Adolescent Blood
Pressure and Pubertal Somatic Growth. The Journal of Clinical
Endocrinology and Metabolism. 2009;94(12):5019-5022.
doi:10.1210/jc.2009-0997.
and
The Change in Blood Pressure during Pubertal Growth: The Journal of
Clinical Endocrinology & Metabolism: Vol 90, No 1 who report a prospective
cohort measuring height/weight/BP systematically across puberty

#16

Back from holidays and found some time to do some quick testing again. Not that I didn’t believe it, but I wanted to see it with my own data and eyes. I redid the graphs but now ignored all OAC related bloodpressures (range of 5 days around an OAC prescription). Result is that the peak between 12 and 20 years in the bloodpressure counting graph was completely gone, indicating that the frequency of the bloodpressure measurements was related to this OAC as expected. But… the graph of the average bloodpressure by age was exactly the same as before. The removal of all these OAC related bloodpressures didn’t change this graph (only very little changed when zooming in). The graphs of the uncorrected an “corrected” average bloodpressure where almost identical. I expected the corrected to come much closer to the male version, but this was clearly not the case (only a fraction closer when zooming but I think we can ignore this difference).

Thanks Marc, for this “hormone lesson”!!

t