Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose...
Transcript of Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose...
![Page 1: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/1.jpg)
Explainable Prediction of Medical Codes from Clinical Text
James Mullenbach
Sarah Wiegreffe
Jon Duke
Jimeng Sun
Jacob Eisenstein
![Page 2: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/2.jpg)
The Clinical Coding Problem● Electronic Health Records have seen increasing
adoption in the last 5 years
2
![Page 3: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/3.jpg)
The Clinical Coding Problem● Electronic Health Records have seen increasing
adoption in the last 5 years● They abound with written physician notes
3
![Page 4: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/4.jpg)
The Clinical Coding Problem● Electronic Health Records have seen increasing
adoption in the last 5 years● They abound with written physician notes● ICD: taxonomy of diagnoses and procedures
4
![Page 5: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/5.jpg)
The Clinical Coding Problem● Electronic Health Records have seen increasing
adoption in the last 5 years● They abound with written physician notes● ICD: taxonomy of diagnoses and procedures● Human coding laborious, error-prone [Birman-Deych et
al., 2005]
5
![Page 6: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/6.jpg)
The Clinical Coding Problem, as an NLPer● Highly multi-label classification
○ Of nearly 9,000 labels, predict a subset
6
![Page 7: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/7.jpg)
The Clinical Coding Problem, as an NLPer● Highly multi-label classification
○ Of nearly 9,000 labels, predict a subset● Testbed for document representations
7
![Page 8: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/8.jpg)
The Clinical Coding Problem, as an NLPer● Highly multi-label classification
○ Of nearly 9,000 labels, predict a subset● Testbed for document representations● Documents are long and loosely structured
8
![Page 9: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/9.jpg)
The dataset
● Open-access, de-identified
9
![Page 10: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/10.jpg)
The dataset
● Open-access, de-identified● 47k admissions -> 47k documents for training
10
![Page 11: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/11.jpg)
The dataset
● Open-access, de-identified● 47k admissions -> 47k documents for training● Loosely structured:
11
Admission Date: [**2118-6-2**] Discharge Date: [**2118-6-14**]
Date of Birth: Sex: F
Service: MICU and then to [**Doctor Last Name **] Medicine
HISTORY OF PRESENT ILLNESS: This is an 81-year-old femalewith a history of emphysema (not on home O2), who presents...
![Page 12: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/12.jpg)
The dataset
● Open-access, de-identified● 47k admissions -> 47k documents for training● Loosely structured:
12
Admission Date: [**2118-6-2**] Discharge Date: [**2118-6-14**]
Date of Birth: Sex: F
Service: MICU and then to [**Doctor Last Name **] Medicine
HISTORY OF PRESENT ILLNESS: This is an 81-year-old femalewith a history of emphysema (not on home O2), who presents...
Long: Median post-processed document length: 1,341
![Page 13: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/13.jpg)
The dataset
● Open-access, de-identified● 47k admissions -> 47k documents for training● Loosely structured:
13
Admission Date: [**2118-6-2**] Discharge Date: [**2118-6-14**]
Date of Birth: Sex: F
Service: MICU and then to [**Doctor Last Name **] Medicine
HISTORY OF PRESENT ILLNESS: This is an 81-year-old femalewith a history of emphysema (not on home O2), who presents...
Long: Median post-processed document length: 1,341
Many labels: 519.1: ‘Other disease…’491.21: ‘Obstructive …’518.81: ‘Acute respir…’486: ‘Pneumonia, orga…’276.1: ‘Hyposmolality…’244.9: ‘Unspecified h…’31.99: ‘Other operati…’...
Median # labels: 14
![Page 14: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/14.jpg)
Prior approaches● Predict from a subset of labels
14
This work● Predict from all labels
![Page 15: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/15.jpg)
Prior approaches● Predict from a subset of labels● Sub-domain focus, e.g.
radiology
15
This work● Predict from all labels● General ICU setting
![Page 16: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/16.jpg)
Prior approaches● Predict from a subset of labels● Sub-domain focus, e.g.
radiology● Private dataset
16
This work● Predict from all labels● General ICU setting● Open-access data
![Page 17: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/17.jpg)
Modeling desiderata
...who sustained a fall at home she was found to have a large acute on chronic subdural hematoma with extensive midline shift...
17
● Focus on the parts that matter
![Page 18: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/18.jpg)
Modeling desiderata
...who sustained a fall at home she was found to have a large acute on chronic subdural hematoma with extensive midline shift...
18
E849.0: Home accidents
801.26: ...subdural, and extradural hemorrhage...
● Focus on the parts that matter● Treat labels individually
![Page 19: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/19.jpg)
Convolution
19
h1
h2
h3
patient
was
initially
placed
on
embed19
convolve
![Page 20: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/20.jpg)
Convolution + Attention
20
U
α11
α21
α31
vh1
h2
h3
patient
was
initially
placed
on
parameter
variable
embed convolve attend (separately for each label)20
![Page 21: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/21.jpg)
Convolution + Attention = the CAML model
21
U
α11
α21
α31
v
lossh1
h2
h3
patient
was
initially
placed
on
parameter
variable
classifyembed convolve attend21
β
![Page 22: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/22.jpg)
Dealing with the long tail● Huge label space (nearly 9,000 total)
2222
![Page 23: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/23.jpg)
Dealing with the long tail● Huge label space (nearly 9,000 total)● Many labels are similar250.00: “Diabetes mellitus without mention of complication, type II or unspecified type, not stated as uncontrolled”
250.02: “Diabetes mellitus without mention of complication, type II or unspecified type, uncontrolled”
2323
![Page 24: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/24.jpg)
Dealing with the long tail: DR-CAML
24
U
α11
α21
α31
lossh1
h2
h3
patient
was
initially
placed
on
parameter
variable
“Atrial fibrillation” Max-pool CNN Label embedding
L2 similarity
24
v
β
![Page 25: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/25.jpg)
MIMIC results● Few works use the whole
label space
2525
*[Scheurwegs et al., 2017]
![Page 26: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/26.jpg)
MIMIC results● Few works use the whole
label space● 20%, 8% relative
improvement over prior SotA
26
*[Scheurwegs et al., 2017]
26
![Page 27: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/27.jpg)
More MIMIC metrics
27
● Enable future comparison● Precision @ k: decision support use-case
27
![Page 28: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/28.jpg)
But, is the model focusing on reasonable areas?● We want interpretability, especially in the clinical domain
2828
![Page 29: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/29.jpg)
But, is the model focusing on reasonable areas?● We want interpretability, especially in the clinical domain● Physician evaluation: select all text snippet(s) that explain the code
2929
![Page 30: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/30.jpg)
But, is the model focusing on reasonable areas?● We want interpretability, especially in the clinical domain● Physician evaluation: select all text snippet(s) that explain the code● Baselines: LogReg, CNN, cosine similarity
3030
![Page 31: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/31.jpg)
But, is the model focusing on reasonable areas?● We want interpretability, especially in the clinical domain● Physician evaluation: select all text snippet(s) that explain the code● Baselines: LogReg, CNN, cosine similarity● 100 random label-document samples
3131
![Page 32: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/32.jpg)
But, is the model focusing on reasonable areas?● We want interpretability, especially in the clinical domain● Physician evaluation: select all text snippet(s) that explain the code● Baselines: LogReg, CNN, cosine similarity● 100 random label-document samples● “Informative” and “Highly Informative”
3232
![Page 33: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/33.jpg)
Physician evaluation example*Code: 575.4
Full descriptions: Perforation of gallbladder
“. . in the setting of gallbladder perforation secondary to acute acalculous cholecystitis after . . . . . . . . inhalation hospital1 times a day metronidazole mg tablet sig one tablet po tid times . . . . . . . to have an infection in your gallbladder requiring iv antibiotics and tube placement for . . .”
33
*not exact format used33
![Page 34: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/34.jpg)
Physician evaluation example*Code: 575.4
Full descriptions: Perforation of gallbladder
“. . in the setting of gallbladder perforation secondary to acute acalculous cholecystitis after . . . . . . . . inhalation hospital1 times a day metronidazole mg tablet sig one tablet po tid times . . . . . . . to have an infection in your gallbladder requiring iv antibiotics and tube placement for . . .”
34
*not exact format used
attention
34
![Page 35: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/35.jpg)
Physician evaluation example*Code: 575.4
Full descriptions: Perforation of gallbladder
“. . in the setting of gallbladder perforation secondary to acute acalculous cholecystitis after . . . . . . . . inhalation hospital1 times a day metronidazole mg tablet sig one tablet po tid times . . . . . . . to have an infection in your gallbladder requiring iv antibiotics and tube placement for . . .”
35
*not exact format used
attentionCNN
![Page 36: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/36.jpg)
Physician evaluation example*Code: 575.4
Full descriptions: Perforation of gallbladder
“. . in the setting of gallbladder perforation secondary to acute acalculous cholecystitis after . . . . . . . . inhalation hospital1 times a day metronidazole mg tablet sig one tablet po tid times . . . . . . . to have an infection in your gallbladder requiring iv antibiotics and tube placement for . . .”
36
*not exact format used
attention cosine sim.CNN
![Page 37: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/37.jpg)
Physician evaluation example*Code: 575.4
Full descriptions: Perforation of gallbladder
“. . in the setting of gallbladder perforation secondary to acute acalculous cholecystitis after . . . . . . . . inhalation hospital1 times a day metronidazole mg tablet sig one tablet po tid times . . . . . . . to have an infection in your gallbladder requiring iv antibiotics and tube placement for . . .”
37
*not exact format used
attention cosine sim.CNN
LogReg
![Page 38: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/38.jpg)
Physician evaluation results● Improves upon CNN,
LogReg
3838
![Page 39: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/39.jpg)
Physician evaluation results● Improves upon CNN,
LogReg● More experts needed!
3939
![Page 40: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/40.jpg)
Future work● Exploit the loose structure of discharge summaries
○ Some have already done this [Shi et al., 2017]
4040
![Page 41: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/41.jpg)
Future work● Exploit the loose structure of discharge summaries
○ Some have already done this [Shi et al., 2017]● Sparsify or structure attention to improve interpretability, accuracy
○ “fusedmax” [Niculae & Blondel, 2017]
4141
![Page 42: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/42.jpg)
Future work● Exploit the loose structure of discharge summaries
○ Some have already done this [Shi et al., 2017]● Sparsify or structure attention to improve interpretability, accuracy
○ “fusedmax” [Niculae & Blondel, 2017]● Develop methods for few-shot or zero-shot classification
4242
![Page 43: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/43.jpg)
Future work● Exploit the loose structure of discharge summaries
○ Some have already done this [Shi et al., 2017]● Sparsify or structure attention to improve interpretability, accuracy
○ “fusedmax” [Niculae & Blondel, 2017]● Develop methods for few-shot or zero-shot classification● Exploit hierarchy of codes
○ GRAM [Choi et al., 2017]
43
250.0 / \250.00 250.01
DM type 2 DM type 1
43
![Page 44: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/44.jpg)
Summary● ICD coding is valuable and challenging
4444
![Page 45: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/45.jpg)
Summary● ICD coding is valuable and challenging● Convolution + attention works well
4545
![Page 46: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/46.jpg)
Summary● ICD coding is valuable and challenging● Convolution + attention works well● Attention can explain the predictions
4646
![Page 47: Explainable Prediction of Medical Codes from Clinical Text · Future work Exploit the loose structure of discharge summaries Some have already done this [Shi et al., 2017] Sparsify](https://reader036.fdocuments.us/reader036/viewer/2022071015/5fce733d49952d5c681bc870/html5/thumbnails/47.jpg)
Summary● ICD coding is valuable and challenging● Convolution + attention works well● Attention can explain the predictions
4747
Thank you!Questions?
Code, data splits, pretrained models: github.com/jamesmullenbach/caml-mimic