- Swedish
- PhD student in medicine (orthopedics)
- Statistician
- Use micro data from official sources (Statistics Sweden)
2018-09-09
Erik Bulow
Categprization
- Deterministic (defined mapping between codes and categories)
- Not statistical classification
TOC
- What’s the problem?
- How to solve it?
- The
coderpackage. - Example with comorbidity.
- Details about the package.
My problem
- Patients from one register
- Previous medical conditions (comorbidity) from another (much bigger) register.
## (polygon[GRID.polygon.1], polygon[GRID.polygon.2], polygon[GRID.polygon.3], polygon[GRID.polygon.4], text[GRID.text.5], text[GRID.text.6], lines[GRID.lines.7], text[GRID.text.8], lines[GRID.lines.9], text[GRID.text.10], text[GRID.text.11])
How can we do this?
- SAS macro exist (doh …!)
- Possible with some basic but messy R-script (18 hours to run)
- R-package
comorbidities.icd10(slow and complicated, not on CRAN) - R-package
icd(lacks some of the desired features, was initially slow and only contains specific coding schemes) - Let’s make another package!
coder
- Separate functionality from classification schemes
- Classification using regular expression
- Some use of matrix algebra when possible
- Rely on
data.table - Avoid other dependencies (for internal IT system)
- Keep it simple
- Pipe-friendly (or use just one function)
Centered around three objects
- Item data (the patients)
- Code data for the same items (the patient register)
- Classification scheme (based on reg-ex and with weighting schemes and additional conditions).
Codes
- Item data with many units (individuals)
- Code data with many instances (hospital visits) per item
- Some time frame of relevance (one year before surgery or 30 days after)
- Many codes per code instance
- Some of them relevant
- Some of them with additional characteristics (main vs additional diagnostic codes)
- Relevant codes grouped by category
- One code can belong to more than one category
- Weighting schemes for combined indices (total comorbidity burden based on individual comorbidities)
Example: Charlson Comorbidity
- Group of patients
- Link to National Patient Register for comorbidity data during the year before surgery
- Identify comorbidities based on the Charlson comorbidity classification scheme
- Calculate comorbity index due to one of the proposed standards
The coder package
Patients
Here by name, otherwise by personal identity numbers (PIN) for ((almost) perfect linkage.
head(ex_people)
## name surgery ## 1 Beaver, Tristin 2016-09-11 ## 2 Maestas, Lilibeth 2016-10-23 ## 3 Jung, Derek 2017-02-20 ## 4 Hayes, Kylihah 2016-12-31 ## 5 el-Riaz, Aadam 2016-04-19 ## 6 Sanchez, Dominique 2017-02-25
Medical records
head(ex_pardata)
## id variable code code_date hdia ## 1: el-Haider, Ruwaid hdia T840F 2012-01-04 TRUE ## 2: Roacho, Marla bdia2 Z510 1993-01-23 FALSE ## 3: el-Hariri, Huda hdia A469 2008-10-07 FALSE ## 4: el-Parsa, Ghazaala hdia M205 2010-10-28 FALSE ## 5: Martinez, Jonathan hdia 989,98 2008-02-12 TRUE ## 6: Martinez, Crystal hdia H401C 2006-09-05 TRUE
Charlson comorbidity
charlson_icd10[c(5, 19), 1:5]
## group regex charlson deyo_ramano dhoore ## 5 dementia ^(F0([0-3]|51)|G3(0|11)) 1 1 1 ## 19 AIDS/HIV ^(B2[0124]) 6 1 1
Visualize classification scheme
visualize(charlson_icd10, c("congestive heart failure"))
Summarise classification scheme
suppressPackageStartupMessages(library(tidyverse)) summary(charlson_icd10) %>% mutate(codes = substr(codes, 1, 30))
## group n codes ## 1 AIDS/HIV 22 B200, B201, B202, B203, B204, ## 2 cerebrovascular disease 82 G450, G451, G452, G453, G454, ## 3 chronic pulmonary disease 57 I278, I279, J409, J410, J411, ## 4 congestive heart failure 8 I099, I110, I130, I132, I500, ## 5 dementia 23 F000, F001, F002, F009, F010, ## 6 diabetes complication 71 E102, E102A, E102B, E102C, E10 ## 7 diabetes without complication 55 E100, E100A, E100B, E100C, E10 ## 8 hemiplegia or paraplegia 22 G041, G114, G801, G801A, G801B ## 9 malingnancy 525 C000, C001, C002, C003, C004, ## 10 metastatic solid tumor 29 C770, C771, C772, C773, C774, ## 11 mild liver disease 83 B180, B180A, B180B, B180C, B18 ## 12 moderate or severe liver disease 11 I850, I859, I864, I982, K704, ## 13 myocardial infarction 15 I210, I211, I212, I213, I214, ## 14 peptic ulcer disease 36 K250, K251, K252, K253, K254, ## 15 peripheral vascular disease 38 I700, I701, I702, I702A, I702C ## 16 renal disease 27 I120, I131, N032, N033, N034, ## 17 rheumatic disease 63 M050, M051, M052, M053, M058,
Add Charlson comorbidity
ex <-
ex_people %>% # Patient data from SHAR
categorize( # Categorize this data using ...
ex_icd10, # ... data from NPR
"charlson_icd10", # based on Charlson comorbidity from ICD-10
id = "name", # Identify id variable from SHAR
date = "surgery", # Identifiy date variable to relate to
days = c(-365, -1), # Only include comorbidity during this period
ind = "quan_original" # Calculate comorbidity index with specified weights
)
Result
ex[, c(1, 7, 16, 21:22)]
## name dementia malingnancy AIDS/HIV quan_original ## 1: Abzari, Joseph FALSE TRUE FALSE 2 ## 2: Alexis, Dayveon FALSE FALSE FALSE 0 ## 3: Anderson, Laquesh FALSE FALSE FALSE 2 ## 4: Armijo Jr, Kalynne FALSE FALSE FALSE 0 ## 5: Babcock, Landon FALSE TRUE FALSE 2 ## 6: Barela, Marco FALSE FALSE FALSE 0 ## 7: Barger, Willanna FALSE FALSE FALSE 0 ## 8: Barnes, Amos FALSE FALSE FALSE 0 ## 9: Bauder, Taylor FALSE FALSE FALSE 0 ## 10: Beasley, Michael FALSE FALSE FALSE 0 ## 11: Beaver, Tristin FALSE FALSE FALSE 0 ## 12: Bille, Cheyenne FALSE FALSE FALSE 0 ## 13: Boffill, Jordan FALSE FALSE FALSE 0 ## 14: Bradshaw, Noah FALSE FALSE FALSE 0 ## 15: Breshears, Kayla FALSE FALSE FALSE 0 ## 16: Bruce, Bryan FALSE FALSE FALSE 0 ## 17: Bruno, Jorge FALSE TRUE FALSE 3 ## 18: Burton, Danyalle FALSE FALSE FALSE 0 ## 19: Charlie, Mareno FALSE FALSE FALSE 0 ## 20: Deshazor, Kordell FALSE FALSE FALSE 0 ## 21: Dominguez Enriquez, David FALSE FALSE FALSE 0 ## 22: Edmondson, Christian FALSE TRUE FALSE 2 ## 23: Elder, Kevyn FALSE FALSE FALSE 1 ## 24: Gallegos, Diana NA NA NA NA ## 25: Ghimire, Chelsea FALSE FALSE FALSE 0 ## 26: Hall, Samorra FALSE FALSE FALSE 0 ## 27: Hayes, Kylihah FALSE FALSE FALSE 0 ## 28: Headden, Devon FALSE FALSE FALSE 0 ## 29: Henricks, Donovan FALSE FALSE FALSE 0 ## 30: Her, Lia FALSE FALSE FALSE 0 ## 31: Hicks, Camren FALSE FALSE FALSE 0 ## 32: Hoang, Elijah FALSE FALSE FALSE 0 ## 33: Jiles-Wright, Bralynd FALSE FALSE FALSE 1 ## 34: Jung, Derek FALSE FALSE FALSE 0 ## 35: Kane, Chapin FALSE FALSE FALSE 0 ## 36: Kim, Alisa FALSE FALSE FALSE 0 ## 37: King, Hali FALSE FALSE FALSE 0 ## 38: Laulu, Kajtshiab FALSE TRUE FALSE 8 ## 39: Lopez, Chiara FALSE TRUE FALSE 2 ## 40: Maestas, Lilibeth FALSE TRUE FALSE 2 ## 41: Mamet, A D FALSE FALSE FALSE 0 ## 42: Marquez, Jacqueline FALSE FALSE FALSE 0 ## 43: Martin, Elizario FALSE FALSE FALSE 0 ## 44: Martinez, David FALSE FALSE FALSE 0 ## 45: Martinez, Jessica FALSE FALSE FALSE 0 ## 46: Martinez, Juan FALSE FALSE FALSE 0 ## 47: Martinez, Nelson FALSE FALSE FALSE 0 ## 48: Mccarthy, Gabriel FALSE FALSE FALSE 0 ## 49: Mcginnis, Adunola FALSE FALSE FALSE 0 ## 50: Mcknight, Jaymes FALSE FALSE FALSE 0 ## 51: Miller, Darshay FALSE FALSE FALSE 0 ## 52: Moktader, Sameer FALSE FALSE FALSE 0 ## 53: Mosely, Jereek FALSE FALSE FALSE 0 ## 54: Nam, John FALSE FALSE FALSE 0 ## 55: Nelson, Phillip FALSE FALSE FALSE 0 ## 56: Okoye, Vanessa FALSE FALSE FALSE 0 ## 57: Parker, Pharo FALSE FALSE FALSE 0 ## 58: Perez Jr, Hannah NA NA NA NA ## 59: Perez, Kimberly FALSE FALSE FALSE 2 ## 60: Pham, Kyle FALSE FALSE FALSE 0 ## 61: Pronteau, Colton FALSE FALSE FALSE 0 ## 62: Reyes, Sylvia NA NA NA NA ## 63: Rodriguez, Alma FALSE FALSE FALSE 0 ## 64: Rowley Booneel, Hailee FALSE FALSE FALSE 0 ## 65: Rusnak, Kyle NA NA NA NA ## 66: Sanburg, Cassandra FALSE FALSE FALSE 0 ## 67: Sanchez, Dominique FALSE FALSE FALSE 0 ## 68: Santee, Holly FALSE FALSE FALSE 0 ## 69: Schell, Jack FALSE FALSE FALSE 2 ## 70: Schubert, Patrick FALSE FALSE FALSE 0 ## 71: Simpson, King FALSE FALSE FALSE 0 ## 72: Smith, John FALSE FALSE FALSE 0 ## 73: Smith, Ruasha FALSE FALSE FALSE 0 ## 74: Spoerl, John FALSE TRUE FALSE 2 ## 75: Stephan, Scott FALSE FALSE FALSE 2 ## 76: Suriwong, Sarah FALSE FALSE FALSE 1 ## 77: Thompson, Stephen FALSE FALSE FALSE 0 ## 78: Thwaites, Travis FALSE FALSE FALSE 0 ## 79: Todacheene, Erminio FALSE FALSE FALSE 1 ## 80: Tomlinson, Amanda FALSE FALSE FALSE 0 ## 81: Tuitele-Britton, Janet FALSE FALSE FALSE 0 ## 82: Valencia, Johnny FALSE TRUE FALSE 2 ## 83: Vallie, Mitchelle FALSE FALSE FALSE 1 ## 84: Vigo, Shi Hyung FALSE TRUE FALSE 2 ## 85: Ward, Harrison NA NA NA NA ## 86: Wilkerson, Teesa FALSE FALSE FALSE 0 ## 87: Wood, Elijah FALSE FALSE FALSE 0 ## 88: al-Abid, Asad FALSE FALSE FALSE 0 ## 89: al-Ahsan, Umar FALSE FALSE FALSE 0 ## 90: al-Eid, Abdus Samad FALSE TRUE FALSE 2 ## 91: al-Kassem, Humaidaan NA NA NA NA ## 92: al-Mansur, Kawkab FALSE FALSE FALSE 2 ## 93: el-Assad, Jubair FALSE FALSE FALSE 0 ## 94: el-Bina, Muntaha FALSE TRUE FALSE 2 ## 95: el-Kazi, Najeeba FALSE FALSE FALSE 0 ## 96: el-Masood, Nasreen FALSE FALSE FALSE 0 ## 97: el-Nagi, Waddaah NA NA NA NA ## 98: el-Riaz, Aadam FALSE FALSE FALSE 0 ## 99: el-Wakim, Nuha FALSE FALSE FALSE 0 ## 100: el-Zakaria, Saajid FALSE TRUE FALSE 2 ## name dementia malingnancy AIDS/HIV quan_original
More control
Function categorize takes care of everything. Alternative workflow:
ex_people %>%
codify(ex_icd10, "name", "surgery", days = c(-365, -1)) %>%
classify(charlson_icd10) %>%
index("quan_original")
## Alexis, Dayveon Bauder, Taylor ## 0 0 ## Beasley, Michael Deshazor, Kordell ## 0 0 ## Henricks, Donovan Marquez, Jacqueline ## 0 0 ## Martinez, David Mccarthy, Gabriel ## 0 0 ## Simpson, King Gallegos, Diana ## 0 NA ## Perez Jr, Hannah Reyes, Sylvia ## NA NA ## Rusnak, Kyle Ward, Harrison ## NA NA ## al-Kassem, Humaidaan el-Nagi, Waddaah ## NA NA ## Abzari, Joseph al-Abid, Asad ## 2 0 ## al-Ahsan, Umar al-Eid, Abdus Samad ## 0 2 ## al-Mansur, Kawkab Anderson, Laquesh ## 2 2 ## Armijo Jr, Kalynne Babcock, Landon ## 0 2 ## Barela, Marco Barger, Willanna ## 0 0 ## Barnes, Amos Beaver, Tristin ## 0 0 ## Bille, Cheyenne Boffill, Jordan ## 0 0 ## Bradshaw, Noah Breshears, Kayla ## 0 0 ## Bruce, Bryan Bruno, Jorge ## 0 3 ## Burton, Danyalle Charlie, Mareno ## 0 0 ## Dominguez Enriquez, David Edmondson, Christian ## 0 2 ## el-Assad, Jubair el-Bina, Muntaha ## 0 2 ## el-Kazi, Najeeba el-Masood, Nasreen ## 0 0 ## el-Riaz, Aadam el-Wakim, Nuha ## 0 0 ## el-Zakaria, Saajid Elder, Kevyn ## 2 1 ## Ghimire, Chelsea Hall, Samorra ## 0 0 ## Hayes, Kylihah Headden, Devon ## 0 0 ## Her, Lia Hicks, Camren ## 0 0 ## Hoang, Elijah Jiles-Wright, Bralynd ## 0 1 ## Jung, Derek Kane, Chapin ## 0 0 ## Kim, Alisa King, Hali ## 0 0 ## Laulu, Kajtshiab Lopez, Chiara ## 8 2 ## Maestas, Lilibeth Mamet, A D ## 2 0 ## Martin, Elizario Martinez, Jessica ## 0 0 ## Martinez, Juan Martinez, Nelson ## 0 0 ## Mcginnis, Adunola Mcknight, Jaymes ## 0 0 ## Miller, Darshay Moktader, Sameer ## 0 0 ## Mosely, Jereek Nam, John ## 0 0 ## Nelson, Phillip Okoye, Vanessa ## 0 0 ## Parker, Pharo Perez, Kimberly ## 0 2 ## Pham, Kyle Pronteau, Colton ## 0 0 ## Rodriguez, Alma Rowley Booneel, Hailee ## 0 0 ## Sanburg, Cassandra Sanchez, Dominique ## 0 0 ## Santee, Holly Schell, Jack ## 0 2 ## Schubert, Patrick Smith, John ## 0 0 ## Smith, Ruasha Spoerl, John ## 0 2 ## Stephan, Scott Suriwong, Sarah ## 2 1 ## Thompson, Stephen Thwaites, Travis ## 0 0 ## Todacheene, Erminio Tomlinson, Amanda ## 1 0 ## Tuitele-Britton, Janet Valencia, Johnny ## 0 2 ## Vallie, Mitchelle Vigo, Shi Hyung ## 1 2 ## Wilkerson, Teesa Wood, Elijah ## 0 0
Classification schemes
- Comorbidity
- Charlson (ICD-10)
- Elixhauaser (ICD-10)
- comorbidity-polypharmacy score (ICD-10)
- RxRiskV (based on ATC codes) (ICD-10)
- Adverse events
- After hip arthroplasty (ICD-10 and KVÅ)
- After knee arthroplasty (ICD-10)
- Example scheme for car brands
- S3 class mechanism to make tailored classification schemes
Where to find it?
- Github: www.github.com/eribul/coder
- Documented by
pkgdown: eribul.github.io/coder- Some vignettes exist and more are planned
- Plan to increase test coverage (now 65 %)
- Plan to release on CRAN
So … how fast is it?
- It’s quite fast!
- What took 18 hours before now takes 30 seconds!
- The newly updated
icdpackage is even faster though, but is not as generic and.
Thanks!
Questions?