search

nalinimsingh / HST.953-AKI-Prediction

Investigating the relationship between mean arterial pressure and acute kidney injury in septic patients using the MIMIC-III Database. MIT HST.953 Fall 2016 Final Project.
13 stars 2 forks source link

Data Analysis #2

Closed nalinimsingh closed 7 years ago

nalinimsingh commented 7 years ago

Data manipulations before running analyses:

Chen:

Nalini:

Tables/figures to create for our paper:

Nalini:

Chen:

@cx1111 Are you interested in helping with creating the plots, etc.? If so, let me know how you'd like to split up the figures. No worries if not.

Let me know if I've missed anything or misinterpreted anything.

nalinimsingh commented 7 years ago

@matthieukomorowski @juliannc:

We mentioned at the end of today's meeting that the timespan for determining whether a patient has AKI is 72 hours starting at the time of the minimum observed mean.

For the features representing percent of time spent within each bin of MAP values, should these be computed over just those 72 hours, or over the entire ICU stay, or over the first 72 hours, or something else?

Thanks!

matthieukomorowski commented 7 years ago

@nalinimsingh "over the first 72 hours."

cx1111 commented 7 years ago

Some annoyances.

I actually just made 'admission creatinine' the closest creatinine value before intime, and if there is none before intime, the first one after. Out of these 7600 unique icustayids for which there is a creatinine, only 5700 values are taken before intime (not even applying the 0 to -4h window, and I have seen that many are several hours before intime. Conversely, many of those admission creatinines taken after intime are also many many hours after intime).

In addition, there are about 40 icustay_ids that were not excluded in our cohort, who have no creatinine measurements. These 40 were not filtered out in the final cohort.

I updated the tables in the google drive based on my above description.

juliannc commented 7 years ago

?Chen,

 I suspect that the 2000 individuals with no creatinine prior to admission came from an OSH and the creatine was done there. We have no way of obtaining OSH data retrospectively.

the 40 individuals with no record creatine are a surprise .

juliann

From: Chen Xie notifications@github.com Sent: Wednesday, November 30, 2016 8:38 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

Some annoyances.

I actually just made 'admission creatinine' the closest creatinine value before intime, and if there is none before intime, the first one after. Out of these 7600 unique icustayids for which there is a creatinine, only 5700 values are taken before intime (not even applying the 0 to -4h window).

In addition, there are about 40 icustay_ids that were not excluded in our cohort, who have no creatinine measurements. These 40 were not filtered out in the final cohort.

I updated the tables in the google drive based on my above description.

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D264053025&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=V8FW-m8fVE3BEB0od2IRr9IARFPBRwH9K8NlpLTVwpY&s=uJFXROIFYQ7X3SDwFGeZk659ANadd6azqdzPs8X_FQ4&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW-2DDGCsKX60e-2DtbZ-5FNlzJpjwEB43jks5rDiT7gaJpZM4LAXOo&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=V8FW-m8fVE3BEB0od2IRr9IARFPBRwH9K8NlpLTVwpY&s=p4ZWrc41s068xGeAbGW3oDjzosRaI27lximCR0JL3q4&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

cx1111 commented 7 years ago

Thanks Juliann. I guess it makes more sense to exclude all data from patients with no creatinine values because we will have no baseline creatinine with which to compute ESRD.

I remade all the tables by keeping only items with icustay admission creatinines<1.2. This means if they had no admission (or hence any) creatinine, they are not included. I also exported the table of admission creatinines including the minutes from icu admission.

If you guys want to you can take a look at these admission times and tell me whether I should reevaluate the 'admission creatinine' criteria. Reminder, I got the closest measurement before intime if any exists, or if not, the closest measurement after intime.

Total of 4524 icustays based on my above method.

@nalinimsingh I am happy to generate some of those tables. Please allocate me some. I'll be spending the weekend afternoons doing whatever data analysis I am capable of :) . Also I noticed that the biggest table has now shrunk to 43M. If you want me to upload them to github instead, tell me and remove the gitignore file.

juliannc commented 7 years ago

?Chen ,

That sounds fine. Thanks for all your work.

Juliann

From: Chen Xie notifications@github.com Sent: Thursday, December 1, 2016 5:43 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

So I guess it makes more sense to exclude all data from patients with no creatinine values.

I remade all the tables by keeping only items with icustay admission creatinines<1.2. This means if they had no admission (or hence any) creatinine, they are not included. I also exported the table of admission creatinines including the minutes from icu admission. If you guys like you can take a look at these admission times and tell me whether I should reevaluate the 'admission creatinine' criteria. Reminder, I got the closest measurement before intime if any exists, or if not, the closest measurement after intime.

Total of 4524 icustays.

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D264318732&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=Pol2NOycvRLISJKwqjo6wbvsC3O10RDKFZaI03GtXOI&s=ClYqAm4Xz0tPewlPXtE5faB5Ph7HupI5z9lyekfOgUA&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW3FNfLtUVvEy9DsXWEvwU7-5FwE8Z2ks5rD02AgaJpZM4LAXOo&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=Pol2NOycvRLISJKwqjo6wbvsC3O10RDKFZaI03GtXOI&s=sNtqrZcdR288WcsgtXKvEYS3To1kITIou1IAHk9-oK0&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

cx1111 commented 7 years ago

For the urine output criteria in the RIFLE system, does <.3ml/kg/h x 24h mean under 0.3ml per kg of patient weight per hour every single hour for a 24 hour period, or does it mean an average urine output of <0.3ml/kg/h for the 24 hours?

Also I noticed that 'table 1' from the RIFLE definition paper https://www.ncbi.nlm.nih.gov/pmc/articles/PMC522841/ gives rather large baseline creatinine estimations, many over 1.2. Could you tell me more about the cohort filtering by admission creatinine >1.2?

Thanks

nalinimsingh commented 7 years ago

@cx1111 -- sorry I somehow missed your comment from yesterday. I'm going to try to get the first two on the checklist done tonight or tomorrow morning and hopefully clean up the code so it's more readable for you than its current state.

If you'd prefer we both separately generate the tables/figures and compare the results, I'm fine with that, or if not, maybe you could start with the 3rd item?

cx1111 commented 7 years ago

I'll start with number 3 and 4. Cheers!

nalinimsingh commented 7 years ago

Ok, are you working right now? I'm in the process of re-calculating which patients have AKI, since I used different criteria before.

If you're working this afternoon, I can get you a table which has patients labeled as either having AKI or not.

If you're working now, maybe you could start implementing the urine output criteria?

cx1111 commented 7 years ago

Yes I'm working right now, I'm in the office. I was planning to do the urine output criteria actually. Need to pull weights and interpolate urine values. That's why I asked the clinicians whether the <.3ml/kg/h means for every hour or just the average.

Also are you using the admission creatinine table I exported?

I'll add another python noteboook so that our work doesn't clash. Will be a lot of code stolen from yours.

nalinimsingh commented 7 years ago

Ok that's perfect, sorry for the confusion on my end.

I'm just using creatinine.csv to get the first creatinine after admission to the ICU. Do I need to use the admission creatinine table? I thought the new creatinine.csv already had patients filtered based on the admission criteria -- let me know if that's not the case.

cx1111 commented 7 years ago

Yes they have been filtered. I just thought it was easier to use the admission creatinine table since those are already there. There's also a difference in that my admission creatinines are the first creatinine before admission if any exist, else the first after admission. I think that's what Juliann stated?

nalinimsingh commented 7 years ago

As I understand it, there are multiple potentially different creatinine values of interest:

  1. First creatinine before admission if it exists, otherwise the first after admission
  2. First creatinine after admission
  3. Max creatinine after the time at which the minimum MAP was achieved

(1) is used for including or excluding patients. (2) is used as the "baseline" creatinine to compare patients' later creatinine values against. (3) is the value compared to (2) in order to determine if the patient has AKI (based on the creatinine criteria) or not. Since (2) and (1) might not be the same, I need to calculate (2) separately.

Is this correct?

cx1111 commented 7 years ago

I thought (1) was meant to be defined as both the 'admission creatinine' to filter icustays and the 'baseline creatinine' with which to calculate AKI. Unless you guys refined that after I left.

Also I thought I would mention that there is an 'agebin' column in the 'cohort' table.

nalinimsingh commented 7 years ago

We didn't refine it after you left -- that was my interpretation of our discussion before. I'll text Juliann and ask.

In either case, should be easy enough to make the switch.

cx1111 commented 7 years ago

Thanks. Can you also ask her about the urine per hour thing?

nalinimsingh commented 7 years ago

Yes, will do!

nalinimsingh commented 7 years ago

Ok, you're correct about the creatinine values. I'll switch to using your admission creatinine table.

For the urine output, Juliann suggests we divide time up into 4 hour blocks, calculate the average urine output over the course of those 4 hours, and designate a patient as having AKI if they meet the criteria for three consecutive blocks.

Also, we didn't pull weight, did we? We will need that to check the urine output criteria.

cx1111 commented 7 years ago

Ok I can do that for the urine. I wonder why 3 blocks though and not for 24h according to the paper. Maybe there isn't enough quality data for that.

I'll get the weight for each icustay and add it to the cohort table.

nalinimsingh commented 7 years ago

Ah, maybe we discussed this after you left -- we're following both the Injury and Failure criteria, so for injury, you would check for 3 blocks (the Failure criteria are more strict than the injury criteria, so I think that should cover it).

cx1111 commented 7 years ago

For urine output, I don't think I can copy your map code exactly. I was thinking of linearly interpolating between all measurements and calculating the area under the trapezoids for total volume.

What do you think?

nalinimsingh commented 7 years ago

Yeah, I definitely don't think you can copy the MAP code exactly.

(EDIT: I fixed this description from my previous comment, which was incorrect.) I don't think you need to interpolate and then integrate -- it might be easier to construct one series of rectangles at each urine output event that extend until the next urine output event, another series of rectangles that go backward to the previous urine output event (basically left and right rectangular integral approximations), and just average the two. This should be the same as the trapezoidal method but easier to implement.

nalinimsingh commented 7 years ago

Also, data format wise, if you can create a dataframe indexed by icustay_id with at least one column with a boolean representing whether or not that patient had AKI, I can easily integrate this into the final dataframe used for the logistic regression.

Thanks!

matthieukomorowski commented 7 years ago

Sorry for the late reply. 0.3 ml/kg/h means 30 ml per hour if the patient is 100kg... does it make sense ?

On Sat, Dec 3, 2016, 20:04 Nalini Singh notifications@github.com wrote:

Also, data format wise, if you can create a dataframe indexed by icustay_id with at least one column with a boolean representing whether or not that patient had AKI, I can easily integrate this into the final dataframe used for the logistic regression.

Thanks!

— You are receiving this because you were mentioned.

Reply to this email directly, view it on GitHub https://github.com/nalinimsingh/HST.953/issues/2#issuecomment-264662175, or mute the thread https://github.com/notifications/unsubscribe-auth/AOp8Hey6T3CIz9tYqV_wHhadm2h5BMHDks5rEcs8gaJpZM4LAXOo .

cx1111 commented 7 years ago

I've generated the dataframe for the aki based on urine values but there are a few considerations left to address based on the underlying imported data:

I generated urine volumes in 4h windows for each patient, with the first window starting from the first urine measurement. Excluding the first urine measurement, I did left rectangular integral approximations and assigned the appropriate proportion of the volume to the current window, and any previous window if there is no other urine measurement that lies before it in the same window.

Actually as I'm typing this I realized that I didn't consider extending beyond the current and left adjacent window in cases where the previous urine measurement was >4h+ ago. I'll fix this and more tomorrow.

Let me know what you guys think about my bullet points.

juliannc commented 7 years ago

?? Nalini and Chen,

I urine output of 0 is not uncommon with a patient in AKI 0-10cc every 1-2 hours is often seen. I looked at your pull and I see that you ar excluding the first urine out put recorded on the patient which ofen reflects teh urien they putout prior to arriving the the ICU. Since you are doing that I do not junderstand why you are getting tens of thousands even with lasix an andividual would not putout that much urine in a day.

I think some how you are pulling an accumlative field of output. we have in metavision. Fluid volume input accumulative and output accumulative that includes all output not just urine.

Call me tommorrow and we will walk through to see were the problem is.

thank You

Juliann

781-385-9243

From: Chen Xie notifications@github.com Sent: Saturday, December 3, 2016 10:26 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

I've generated the dataframe for the aki based on urine values but there are a few considerations left to fix based on the underlying imported data:

I generated urine volumes in 4h windows for each patient, with the first window starting from the first urine measurement. Excluding the first urine measurement, I did left rectangular integral approximations and assigned the appropriate proportion of the volume to the current window, and any previous window if there is no other urine measurement that lies before it in the same window.

Actually as I'm typing this I realized that I didn't consider extending beyond the current and left adjacent window in cases where the previous urine measurement was >4h+ ago. I'll fix this and more tomorrow.

Sorry for the delay. Let me know what you guys think about my bullet points.

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D264681957&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=NxZTLNcyKcLbk8C6o-w_zGxHeRuKgToX4ZBkuugTbIY&s=bYeljhCsKXrMtwTEBCFLsPtnjeqMoR3s2zwaxOU433A&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW11F-5FyrKKYZ6l6hkfFy8JAFIPgrZks5rEjLzgaJpZM4LAXOo&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=NxZTLNcyKcLbk8C6o-w_zGxHeRuKgToX4ZBkuugTbIY&s=2F37SuD5dYE4G1PONqz8Ui2xVvfqqolLeSUQSZ-MPQI&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

matthieukomorowski commented 7 years ago

Regarding the large urine outputs, I have seen the same thing. What I did was plotting the distribution and excluding values exceeding a certain threshold, say 1L/h.

On Sat, Dec 3, 2016 at 10:26 PM, Chen Xie notifications@github.com wrote:

I've generated the dataframe for the aki based on urine values but there are a few considerations left to fix based on the underlying imported data:

  • Some urine values are ridiculously large. Order of tens of thousands. Should we impose a filter like we did for map?
  • Conversely, there are also a bunch of urine values = 0. Should we keep these?
  • We are missing weights for a few 10's of patients. In my code I just put some filler to make it run. Should we estimate weight?

I generated urine volumes in 4h windows for each patient, with the first window starting from the first urine measurement. Excluding the first urine measurement, I did left rectangular integral approximations and assigned the appropriate proportion of the volume to the current window, and any previous window if there is no other urine measurement that lies before it in the same window.

Actually as I'm typing this I realized that I didn't consider extending beyond the current and left adjacent window in cases where the previous urine measurement was >4h+ ago. I'll fix this and more tomorrow.

Sorry for the delay. Let me know what you guys think about my bullet points.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/nalinimsingh/HST.953/issues/2#issuecomment-264681957, or mute the thread https://github.com/notifications/unsubscribe-auth/AOp8HXO2ZMGOHsfQSmfd4uiv6ssqybB0ks5rEjLzgaJpZM4LAXOo .

--

Dr Matthieu Komorowski Clinical Research Fellow in Intensive Care PhD student, Dept of Surgery and Cancer & Dept of Bioengineering, Imperial College London Visiting Scholar, Laboratory of Computational Physiology, Harvard-MIT Health Sciences and Technology mkomo@mit.edu mkomo@mite.edu ; m.komorowski14@imperial.ac.uk Tel: +1 (857) 253 9547 ; +44 (0) 757 897 5175 ; SkypeID matthieukomorowski Twitter https://twitter.com/matkomorowski - LinkedIn https://uk.linkedin.com/in/matthieukomorowski - Research Gate https://www.researchgate.net/profile/Matthieu_Komorowski4 - Google Scholar https://scholar.google.co.uk/citations?user=xpAYtroAAAAJ

cx1111 commented 7 years ago

Hi guys

Sorry I must not have been looking at the final urine table. In that, there is just one ridiculous value of 160220 which I'll remove.

Aside from that, here's a histogram of the 1000/836000 urine measurements >1000ml. Seems alright. https://drive.google.com/open?id=0B-IsdtOumwVEOHZGOFc5dUVleFk

nalinimsingh commented 7 years ago

Hi all,

I have all the code written to get the covariate summary statistics and logistic regression results, and I ran a preliminary version of that using just the creatinine criteria to indicate whether or not a patient has AKI. So, once we have the urine criteria all ready to go, I can add it to the appropriate column, and rerun the results.

Chen -- can you call Juliann? I'm happy to listen in on the call, but since you did the urine output analysis, I think you're better informed to discuss that issue.

Thanks, Nalini

nalinimsingh commented 7 years ago

Oh, sorry -- I was writing as you posted! Let me know when you're ready for me to add in your values.

cx1111 commented 7 years ago

I've replaced the tables using admission creatinine < 1.5

juliannc commented 7 years ago

?Hi Chen and Nalini,

Just saw something intitaled final analysis is this what I should be looking at.

I have hurt my back and can not come in to MIT today can youplease call me so I can go over things with you

JUliann

From: Chen Xie notifications@github.com Sent: Wednesday, December 7, 2016 9:20 AM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

I've replaced the tables using admission creatinine < 1.5

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D265458670&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=XsADQQpKjSxf1zYnigyrYpta7XeNNBj7pXRf06GStBg&s=Tb_JDXzbr3rk2arqIp0lcMlEmZm1un_W_jDj7gqwQz8&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsWy8cEx6B4Hjm2o9fhVpRbTZUy2Rjks5rFsCdgaJpZM4LAXOo&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=XsADQQpKjSxf1zYnigyrYpta7XeNNBj7pXRf06GStBg&s=cmi1kziSsfmGAXFNUIFAU7QQ3UWcg-nYTv0EtskV9nQ&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

nalinimsingh commented 7 years ago

I just spoke with Juliann over the phone -- I'll re-run the results with admission creatinine < 1.5. We'll have a call to discuss the results around 10:30 if you want to join, Chen.

cx1111 commented 7 years ago

You guys can go ahead. I'll generate the 2 charts in the op.

nalinimsingh commented 7 years ago

Ok, I can take the one labeled "TBD." Thanks for your help incorporating the urine output into the LR.

cx1111 commented 7 years ago

I forsee a commit clash. So I'll make new files while you can keep using final_analysis.ipynb and process.py

nalinimsingh commented 7 years ago

Chen -- I realized you'll need the hour at which the minimum MAP was achieved for the second plot. Right now that's in the field 'hour' in the min_maps DataFrame. I never transferred it to the final dataframe, but you should be able to add it pretty easily.

cx1111 commented 7 years ago

For this plot: Bar chart of fraction of patients developing AKI within 72 hours for each minimum MAP bin , you don't actually want to restrict it to patients who develop AKI within 72h right?

Also I don't see how (Secondary) Scatter plot of time to AKI development based on minimum MAP value requires the hour at which the minimum map was achieved. *Edit: Do you mean T(aki_onset) - T(minimum_map)?

nalinimsingh commented 7 years ago

Looking over my notes, we said two conflicting things about your first point.

From the previous meeting, we did want to restrict it to patients who develop AKI within 72 hours to avoid counting patients who have long ICU stays and could develop AKI much later. However, I also have written down that we wanted to look at all patients who develop AKI. @juliannc @matthieukomorowski -- for that plot, should we restrict to patients who develop within 72 hours of achieving their minimum MAP?

For the second question: Yes, at the last meeting after you left, that was what we decided to plot.

cx1111 commented 7 years ago

Just pushed the plots. Let me know if they're ok.

juliannc commented 7 years ago

To clarify we want to look at 72 hour worth of data on all septic patient and identfy any that develope AKI with in that 72 hour window.

Here are my is interpretation of finding so far.

Findings

· Septic population relatively even dispersed continue to be true between those that do and do not develop sepsis AKI.

· The majority of the Population is Caucasian with 11% of the unknown again no change in dispersion noted between non AKI and AKI in septic population

· Age appears to be a factor in AKI development which is consistent with literature

· LOS is significantly less when AKI does not develop consistent with literature

· Lactate slightly higher in AKI than non AKI but not significant

· Use of vasopressor significant lower in Non AKI verses AKI

· Widen range of MAP variation from 30-102 there does not appear to be a relationship between MAP and the development of AKI which is a finding that is inconsistent with the literature.

· The finding do not help clarify what the optimal MAP is for septic patient to reduce risk of AKI.

· The Finding are inconsistent with the literature which suggest that MAP of > 65 at least are needed to reduce the risk of AKI.

Confounders?

· History of Hypertension

· True Baseline Creatinine is not known.

· More people ruled in for AKI based on urine output than Creatinine

· Many patient arrive from OSH where labs were drawn but are not available in MIMIC III data base.?

From: Nalini Singh notifications@github.com Sent: Wednesday, December 7, 2016 2:40 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

Looking over my notes, we said two conflicting things about your first point.

From the previous meeting, we did want to restrict it to patients who develop AKI within 72 hours to avoid counting patients who have long ICU stays and could develop AKI much later. However, I also have written down that we wanted to look at all patients who develop AKI. @juliannchttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_juliannc&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=7ti6KYHNxJinS-VB1Rb-DWF-WHGOal_J2cqsWCJqQYw&s=ntrmxLhE36VIMgNqbjC_JWPU-g_ayY3Y9oWks8LXxgQ&e= @matthieukomorowskihttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_matthieukomorowski&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=7ti6KYHNxJinS-VB1Rb-DWF-WHGOal_J2cqsWCJqQYw&s=c7IP0h91fDRfByaRqQLALPdJbFiuVfSkr4Ia1XxRrCM&e= -- for that plot, should we restrict to patients who develop within 72 hours of achieving their minimum MAP?

For the second question: Yes, at the last meeting after you left, that was what we decided to plot.

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D265550913&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=7ti6KYHNxJinS-VB1Rb-DWF-WHGOal_J2cqsWCJqQYw&s=G6hjRc2jDOJMEsYauBf7jJcXLz0P4D5NubkQAUXmhiQ&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW9T0m5BkkJEiPWekaEXtoVkslTwIks5rFwvBgaJpZM4LAXOo&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=7ti6KYHNxJinS-VB1Rb-DWF-WHGOal_J2cqsWCJqQYw&s=zhdrXtMf1eMkykYTT4OJvAmEwsz305TzzeXouDbHapA&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

nalinimsingh commented 7 years ago

Does the 72 hour window begin from the time of the reference creatinine measurement, admission to the hospital, or the measurement of the minimum MAP value?

juliannc commented 7 years ago

?I have to teach until 10pm will check in later . I am flying out in am but should be availble after 1300

Juliann

From: Chen Xie notifications@github.com Sent: Wednesday, December 7, 2016 2:58 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

Just pushed the plots. Let me know if they're ok.

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D265556056&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=UNSvKM-27kk9pu0AadjiZ1PWtwWEELZY4_ePt4JQL-c&s=q4_B8CPQwgyqLWj4cdTXBTQTK-nB9r5h3wD4Pyb2GB0&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsWytb6A4DaXxqgwdMK7T3VHKdxFrvks5rFw-5FlgaJpZM4LAXOo&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=UNSvKM-27kk9pu0AadjiZ1PWtwWEELZY4_ePt4JQL-c&s=lNdm78gH_LKxtJqQbr62TZCT8SYe78aYlzZnmxO1jOk&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

juliannc commented 7 years ago

?Admission to ICU

From: Nalini Singh notifications@github.com Sent: Wednesday, December 7, 2016 3:00 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

Does the 72 hour window begin from the time of the reference creatinine measurement, admission to the hospital, or the measurement of the minimum MAP value?

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D265556726&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=VmnILfDT98d1Oyd76fuUKRJhfnlpG4D0_Ug1tpVv2J4&s=ljGEdu0ie1GUJBLfhfPclOkvbL2y-zxpebjpz1qKHJI&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW7cocIj5ksOQclTCVihAAEny3h9Mks5rFxBogaJpZM4LAXOo&d=CwMCaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=VmnILfDT98d1Oyd76fuUKRJhfnlpG4D0_Ug1tpVv2J4&s=rR6bwVSq7pzIXr-itUCseAYZX-_4urbeDUSjqFYEwFo&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

nalinimsingh commented 7 years ago

@cx1111 wow, that bar graph for the proportion of patients developing AKI within each minimum MAP bin looks very clear. Could you add uncertainty bounds?

Since the 72 hour limit appears to make the MAP-AKI relationship pretty clear, I may redo the analysis, regression, etc. with that limit and if the correlation between MAP and AKI is stronger there.

cx1111 commented 7 years ago

What are the uncertainty bounds in this case? Also I noticed that I didn't remove urine measurements with t<0 from admission. I'm actually getting some AKI triggers from lack of urine output with t<0 due to those so I think I should remove them, unless we somehow want to use them to filter the entire cohort...

nalinimsingh commented 7 years ago

Hm, we could use bootstrapping to estimate a confidence interval? It's probably not that high priority, so don't worry about it if you don't want.

When you say you didn't remove the urine measurements -- I haven't looked at your code closely, does this mean it's possible that we're labeling people as having AKI based on urine events before admission? If so, yes, let's remove.

Thanks!

juliannc commented 7 years ago

Nalini and Chen,?

The weight may be cause some in acurracy.

When there are more than 1 weight on the patient it is the first weight recorded that should be used this is the dry weight. I think we are doing this justed wanted to make sure .

How many patient do not have weights? how are we extrapolating their weights .

Juliann

From: Nalini Singh notifications@github.com Sent: Wednesday, December 7, 2016 3:18 PM To: nalinimsingh/HST.953 Cc: Corey,Juliann (BIDMC - Nursing); Mention Subject: Re: [nalinimsingh/HST.953] Data Analysis (#2)

Hm, we could use bootstrapping to estimate a confidence interval? It's probably not that high priority, so don't worry about it if you don't want.

When you say you didn't remove the urine measurements -- I haven't looked at your code closely, does this mean it's possible that we're labeling people as having AKI based on urine events before admission? If so, yes, let's remove.

Thanks!

- You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_nalinimsingh_HST.953_issues_2-23issuecomment-2D265561490&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=sLy2HuSbVpHj30cRjv9dNWo0iTK2psJ59WLr4-NrNEY&s=pZ0KteJEAyCN8m-qcOc9foHEXVPiCX8Sas-TXV8O9iI&e=, or mute the threadhttps://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AVIsW6MoezyrwkK4kjvf-5FtBf0gS2k4j3ks5rFxSegaJpZM4LAXOo&d=CwMFaQ&c=WknmpdNpvrlj2B5K1aWVqL1SOiF30547pqSuOmtwXTQ&r=zdSU84yBy3j2OzZr9NEwh9M8teYlU_8vtGr4h9UKx-Q&m=sLy2HuSbVpHj30cRjv9dNWo0iTK2psJ59WLr4-NrNEY&s=QupoCfoBKPyGpTDrP-qlkyFE_f6JymZx6e8i88nbSuM&e=.

This message is intended for the use of the person(s) to whom it may be addressed. It may contain information that is privileged, confidential, or otherwise protected from disclosure under applicable law. If you are not the intended recipient, any dissemination, distribution, copying, or use of this information is prohibited. If you have received this message in error, please permanently delete it and immediately notify the sender. Thank you.

cx1111 commented 7 years ago

Yes I use the first weight. I interpolate under 100 out of 4000-5000 weights. The range of predicted weights was about 40 to 90 which seemed reasonable, given that the actual weights went from below 30 to above 120. I used age, age squared, sex, and height when available in a linear regression model.

Nalini, when you say bootstrap do you mean for each min map bin, draw N batches of M additional samples and use those N means to calculate the confidence intervals? Also do you want to meet tomorrow evening to discuss the presentation?