Data-driven Identification of Urgent Surgical Procedures For Use in Epidemiological Studies in Trauma

Matthew Miller, Louise Jorm and Blanca Gallego

citation: Miller M, Jorm L, Gallego B. Data-driven identification of urgent surgical procedures for use in trauma outcomes measurement. Trauma Surg Acute Care Open. 2025;10:e001783. doi:10.1136/tsaco-2025-001783

The accompanying series of R scripts are used to create a list of urgent surgery from surgical booking priority. They can be used for all surgical patients, or limited to those admitted with trauma. Options are included to change the agreement cut off, consensus cut off, age group, and whether to use trauma-only or all surgical patients. Instructions are also given on how to alter the classification of urgency to suit bespoke definitions. R code to generate lists for specific age groups and genders are also provided.

Note: unfortunately the ACHI procedure code and ICD diagnosis code tables need to be obtained separately as we don't have a licence to distribute them here.

These scripts are written for an Australian study and modifications will be needed for other countries/hospitals. In particular:

1. Trauma is defined here as any patient admitted with a diagnosis code from the ICD-10-AM (11th edition) Chapters 19 (Injury, Poisoning, and certain other consequences of external causes) and 20 (External causes of morbidity and mortality). Codes related to hospital complications are excluded as part of the script. This will need to be altered for international codes such as the ICD-10-CM in script 2.
2. The procedure codes used in these scripts are for the Australian Classification of Health Interventions. To use ICD-10-AM procedure codes or ICD-11 Classification of Health Intervention codes you will need to modify script 2.
3. The surgical priorities will be institution specific and will need updating and collapsing. In our institution these were: "<=15minutes-Immediate life threatening", "A-1hr Life threatening incl obstetric", "<=1hr-Life threatening conditions", "B-2hrs Highly critical organ/limb threat", "<=4hrs-Organ or limb threatening", "C-4hrs Critical", "<=8hrs-Non-critical but emergent" , "D-8hrs Urgent", "<=24hrs-Non critical, non-emergent", "E-24hrs Semi-Urgent", "<=72hrsCannot discharge before procedure", "F-72hrs Non-Urgent". You will need to modify script 1, starting around line 33, and change the case_when criteria. The right hand arguments must remain as "<1hr","<4hrs", "<8hrs", "<24hrs","<72hrs".

Structure of raw data files

The tables below outline the columns in the databases, including names and levels, that are needed for the scripts to work. Examples using synthetic (fake) data are also given in the file repository.

Study Data

Column name	Description
study_id	Anonymised ID
sex	M, F
asa	"1 - Healthy" "2 - Mild Systemic Disease" "3 - Moderate to Severe Systemic Disease" "4 - Severe Systemic Disease" "5 - Moribund" "See anaesthetist record" "No anaesthetist" "6 - Organ Donor/Transplant"
age	In years
admit_dt_tm	POSIXct YMD HMS
case_date	Date
emerg_book_dt_tm	POSIXct YMD HMS
admit_to_book_pd	From script 1. "same day", "same week", "over 1 week"
specialty	Bespoke to your data
procedural_consultant	Alphanumeric from script 1
planned_procedure	Free text
primary_procedure	Free text
emerg_priority	Bespoke from your data
emerg_priority_collapse	From script 1. Must be one of "<1hr","<4hrs", "<8hrs", "<24hrs","<72hrs"
episode_number	numeric
coded_primary_procedure	ACHI / Procedure code
episode_start_date	Date
additional_procedure_1	ACHI / Procedure code
additional_procedure_2	ACHI / Procedure code
additional_procedure_3	ACHI / Procedure code
additional_procedure_4	ACHI / Procedure code
additional_procedure_.....	ACHI / Procedure code. No limit on additional columns
principle_diagnosis	ICD code
additional_diagnosis_1	ICD code
additional_diagnosis_2	ICD code
additional_diagnosis_3	ICD code
additional_diagnosis_4	ICD code
additional_diagnosis_......	ICD code. No limit on additional columns

Procedure data.

Column name	Description
achi_proc_code	ACHI procedure code
procedure_text	text
block_number	Number (from ACHI block)
block_text	text
procedure_chapter	Text
procedure_sub_chapter	Text
icd10_ver	ICD-10 version

ICD-10 data

Column name	Description
four_dgt_no	ICD-10 four digit code
four_dgt_text	Text
three_dgt_no	ICD-10 three digit code
three_dgt_text	Text
diagnosis_chapter	Text
diagnosis_subchapter	Text

The Scripts.

Script 1. Preprocessing the data.

The script focusses on preprocessing and summarising data related to study admissions and procedural priorities, collapsing and categorising emergency priorities, and ensuring data security by hashing certain variables. The aim is to make the urgency factor levels consistent with the remaining scripts, to anonymise the procedural consultant and convert the admission time to "same day" or not categories for filtering later.

1. Vector Definition:
   • Defines a character vector id_chrs containing alphanumeric characters (0-9 and a-z), to be used in creating study IDs.Urgency Levels:
   • Defines a vector urgency_levels_collapsed representing collapsed urgency levels.
2. Reads Study Database:
   • Reads a study database from an Excel file named "study_db.xlsx" into a data frame called study_db.
3. Calculates Admission to Booking Period:
   • Computes the time difference in hours between the columns emerg_book_dt_tm and admit_dt_tm and creates a new variable admit_to_book_pd representing the admission to booking period.
4. Categorizes Booking Periods:
   • Creates categories for the admission to booking period based on time duration ("same day," "same week," or "over 1 week").
5. Frequency Table:
   • Prints a frequency table for the variable admit_to_book_pd to summarize the distribution of booking periods.
6. Filters and Collapses Data:
   • Filters the study database to include only cases with the admission to booking period as "same day."
   • Groups the data by study_id, age, and admit_dt_tm.
   • Selects the first case within each group based on the case_date.
   • Collapses emergency priorities into broader categories (e.g., "<1hr", "<4hrs") using the emerg_priority_collapse variable.
   • Converts emerg_priority_collapse into a factor variable with specified levels.
7. Anonymise Procedural Consultant:
   • changes the procedural_consultant variable to an alphanumeric character.

Script 2. Processing procedure data and creating the function.

This is divided into three parts.

Part 1

processes procedure data, cleans and matches procedure descriptions.

1. Loading Study Data:
   • The study data is loaded into a variable named study_db_all.
2. Downloading ACHI and ICD Codes:
   • ACHI codes are stored in the achi_db variable.
   • ICD-1-AM codes are stored in the icd_db variable.
3. Removing Unwanted Procedure Codes:
   • Certain procedure codes related to anesthesia, airway management, ventilatory support, vascular access device, etc., are identified and removed from the achi_db, combined into a single list called codes_to_remove.
4. Extracting Primary Procedure Data:
   • The primary procedure data is extracted from the study_db_all variable and stored in study_db_coded_primary.
5. Matching Procedure Codes:
   • Procedure codes are matched based on both planned and primary procedures as we had two columns that describe the booked procedure. You may need to use only the first one.
   • Stopwords are removed, and text cleaning is performed on free-text descriptions.
   • Common words between planned and actual procedures are identified for both planned and primary procedures.
   • The resulting data frames are named planned_achi and primary_achi.
6. Creating ACHI Coded Database:
   • A new database (achi_study_db) is created by combining the primary procedure data with planned procedures for each study.
   • Duplicate entries are removed, and the column is renamed as booked_procedure. The booked_procedure column is added back to the original study database (study_db_all) and unwanted codes specified in codes_to_remove are filtered out.
7. Counting Procedure Additions:
   • The number of procedures for each study is counted, and the maximum count is retained.
   • The script prints the count of procedures for each study.
8. Checking for missing Values:
   • The script checks for missing values in the booked_procedure column. The final step involves filtering out rows where the booked_procedure column is missing.

Part 2

is an R function, named get_list_of_procedures, summarising surgical procedure data. It has the following options:

cohort	can be "all patients" or "trauma" . If all patients, then an extra step in the function classifies the patients as trauma or not for the final table.
agreement_threshold	0-100
consensus_threshold	0-1
urgency	can be 1,4,8,12,24 or 72 depending on the time period you are considering as urgent. For example, if it is all procedures booked within a 4-hour time period, then urgency = 4.
show_figures_tables	TRUE or FALSE if you want some summary information

The steps are:

1. Trauma Identification:
   • The function starts by identifying trauma cases in the dataset. It uses specific diagnosis codes to classify episodes of care as trauma or non-trauma.
   • The identified trauma study IDs are stored in the variable trauma_study_ids.
2. Filtering Dataset:
   • The function then filters the original study database (study_db) to include only cases with trauma study IDs (study_db_trauma).
3. Main Functionality:
   • The main functionality of the function is to create a table of procedures and urgency levels based on specified criteria such as agreement and consensus thresholds.
   • It provides an option to filter data based on age range and urgency level.
   • It calculates various statistics, including the number of unique procedures, unique booking proceduralists, and individual patients.
   • It generates tables and figures summarizing the urgency levels, agreement, and consensus among proceduralists.
   • It assesses the distribution of urgency levels across all booked procedures.
4. Trauma Analysis (Optional):
   • If the analysis is for the entire cohort (cohort == "all patients"), the function further analyzes trauma cases within the dataset.
   • Trauma cases are identified based on diagnosis codes, and the results are added to the main dataframe.
5. Primary Procedure Identification:
   • The function identifies primary procedures by comparing booked procedures with a list of primary procedures.
   • It calculates confidence intervals for the proportion of urgent cases and adds this information to the main dataframe.

Part 3.

runs the get_list_of_procedures function. Code is also provided to create urgency tables and save them to an excel workbook, and to create age group and gender specific lists.

1. Run Procedure Analysis Function for Trauma:
   • The get_list_of_procedures function is applied.
2. Join ACHI Code Descriptions:
   • The resulting list of procedures (full_list) is joined with the ACHI code descriptions (achi_text), providing additional information about each procedure.
3. Filter Procedures by Urgency (Agreement and Consensus):
   • Two subsets of procedures are created based on urgency classification:
     • full_list_urgent: Procedures classified as urgent by agreement.
     • full_list_urgent_consensus: Procedures classified as urgent by consensus.
4. Frequency Tables for Trauma and Primary Procedures:
   • Frequency tables are generated for trauma cases and primary procedures for both the agreement and consensus subsets.
5. Generate Lists for Different Age Groups:
   • Lists of procedures are generated for different age groups, including pediatrics (paeds_list), adults (adult_list), and the elderly (elderly_list).
   • Urgent procedures for each age group are also identified and categorized.
6. Identify Differences in Procedures Between Age Groups:
   • Differences in procedures between age groups are identified, and separate data frames are created to show procedures specific to each age group.
7. Frequency Tables for Differences in Age Groups:
   • Frequency tables are generated to show the distribution of procedures that are specific to age groups, considering trauma and primary procedure classifications.
8. Generate Lists for Different Gender Groups:
   • Lists of procedures are generated for male and female trauma patients.
9. Write Results to Excel Workbook:
   • The results, including urgent procedures, additional procedures by consensus, and various age and gender-specific lists, are written to an Excel workbook named "urgent_surgery_lists.xlsx."

Script 3. Demographic analysis and regression modelling.

This part produces demographic data and some regression analysis. The final part aims to analyse and visualise the distribution of urgency levels based on different consensus cutoffs in a trauma cohort, to help decide an appropriate consensus threshold.

Demographic Analysis

1. Study Group Selection:
   • A trauma-only study group (study_db_trauma) is created by filtering the original study database (study_db) to include only trauma cases.
2. Demographics Data Preparation:
   • Unique entries of trauma study IDs are selected from the trauma-only database (study_db_trauma).
   • Non-trauma patients are identified by subtracting trauma patients from the original study group.
3. ICD Code Processing:
   • Injury and mechanism codes are combined, and specific codes are removed based on predefined sets (ts, ys, us, xs).
   • Trauma identification is performed using injury and mechanism codes, and a new dataset (trauma1) is created.
4. Demographic Variables Selection:
   • Relevant demographic variables are selected, including age, sex, ASA (American Society of Anesthesiologists) score, specialty, and emergency priority.
5. Demographic Summary Statistics:
   • Summary statistics and histograms are generated for age.
   • Frequency tables are created for ASA score, trauma status, sex, and emergency priority.
6. Specialty Data Processing:
   • Specialty names are standardized, and a frequency table is generated.
7. Urgent Specialty Analysis:
   • A list of urgent specialty frequencies is created based on urgent procedures identified earlier.
8. Results Export:
   • Lists of specialty frequencies for urgent and entire study groups are saved to an Excel workbook (specialty_lists.xlsx).

Regression Analysis for Trauma Cohort

1. Procedure Selection for Regression:
   • Procedures with a minimum number of raters and those with both male and female patients are identified.
2. Multilevel Logistic Regression Models:
   • Several logistic regression models are fitted, considering different fixed and random effects.
   • Null model is assessed for Intraclass Correlation Coefficient (ICC).
   • Models are compared using ANOVA.
3. Odds Ratios Calculation:
   • Odds ratios with confidence intervals are calculated and presented for significant predictors.

Regression Analysis for Entire Cohort

1. ICD Code Processing for Entire Cohort:
   • Similar ICD code processing is performed for the entire cohort (study_db_all).
2. Procedure Selection for Regression (Entire Cohort):
   • Procedures with a minimum number of raters are identified.
3. Multilevel Logistic Regression Models (Entire Cohort):
   • Similar steps are followed as in the trauma cohort analysis.
4. Odds Ratios Calculation (Entire Cohort):
   • Similar odds ratios and confidence interval calculations are performed.

Threshold Analysis

1. A function named table_of_n is defined to calculate and create a table of procedure counts based on specified agreement and consensus thresholds.
2. Two vectors (a_th for agreement thresholds and c_th for consensus thresholds) are defined.
3. The map2 function is used to apply the table_of_n function to all combinations of agreement and consensus thresholds, and the results are combined into a table (table_of_counts).

Consensus Analysis and Visualization

1. Urgency levels (<1hr, <4hrs, <8hrs, <24hrs, <72hrs) are defined.
2. Different consensus thresholds are evaluated (0.6, 0.65, 0.7, 0.75), and procedures meeting the urgent surgery criteria are filtered accordingly.
3. The results are combined into a data frame (consensus_plots) and visualized using density plots and jitter plots for each urgency level.
4. Two plots are generated: one displaying the distribution of proportion of responses for each urgency level by consensus cutoff, and another showing the distribution for each urgency level, with each point representing a surgical procedure.