AI-Assisted Quality Control for Clinical Programming
AI-Assisted Quality Control for Clinical Programming
GitHub Copilot Integration for Automated QC Procedures
This toolkit demonstrates comprehensive QC procedures using R with GitHub Copilot assistance for clinical programming validation.
Quality Control (QC) is a critical component of clinical programming that ensures data integrity, regulatory compliance, and statistical accuracy. This toolkit provides:
library(dplyr)
library(tibble)
library(stringr)
library(lubridate)
library(tidyr)
library(purrr)
cat("=== AI-Assisted Clinical Data Quality Control ===\n")=== AI-Assisted Clinical Data Quality Control ===cat("Comprehensive QC procedures with GitHub Copilot integration\n\n")Comprehensive QC procedures with GitHub Copilot integrationWe’ll create test data with various quality issues to demonstrate our QC procedures.
set.seed(2024)
# Create ADSL with various data quality issues
adsl_qc <- tibble(
STUDYID = "ABC-123",
USUBJID = c(
paste0("ABC-123-", sprintf("%03d", 1:18)),
"ABC-123-019", # Duplicate will be added
"ABC-123-019" # Duplicate subject
),
SUBJID = c(sprintf("%03d", 1:18), "019", "019"),
AGE = c(sample(18:75, 18, replace = TRUE), 65, NA), # Missing age
SEX = c(sample(c("M", "F"), 18, replace = TRUE), "X", "F"), # Invalid sex code
RACE = c(sample(c("WHITE", "BLACK OR AFRICAN AMERICAN", "ASIAN"), 18, replace = TRUE),
"white", "WHITE"), # Inconsistent case
TRT01P = c(sample(c("Placebo", "Study Drug 5mg", "Study Drug 10mg"), 18, replace = TRUE),
"PLACEBO", "Study Drug 5mg"), # Inconsistent treatment name
TRT01PN = c(sample(c(0, 5, 10), 18, replace = TRUE), 0, 5),
TRTSDT = c(
rep(as.Date("2024-01-15"), 18),
as.Date("2023-12-01"), # Treatment start before study start
as.Date("2024-01-15")
),
TRTEDT = c(
rep(as.Date("2024-03-15"), 18),
as.Date("2024-03-15"),
NA # Missing treatment end date
),
SAFFL = c(rep("Y", 19), ""), # Missing safety flag
ITTFL = c(rep("Y", 18), "N", "Y"),
COMPLFL = c(rep("Y", 17), "N", "Y", "Y"),
DTHFL = c(rep("N", 19), "Y"),
DTHDTC = c(rep("", 19), "2024-02-28"),
DTHDT = c(rep(as.Date(NA), 19), as.Date("2024-02-28"))
)
# Display the problematic data
print("ADSL Data with Quality Issues:")[1] "ADSL Data with Quality Issues:"print(adsl_qc)# A tibble: 20 × 16
STUDYID USUBJID SUBJID AGE SEX RACE TRT01P TRT01PN TRTSDT TRTEDT
<chr> <chr> <chr> <dbl> <chr> <chr> <chr> <dbl> <date> <date>
1 ABC-123 ABC-12… 001 19 F ASIAN Place… 5 2024-01-15 2024-03-15
2 ABC-123 ABC-12… 002 54 F ASIAN Study… 0 2024-01-15 2024-03-15
3 ABC-123 ABC-12… 003 62 M WHITE Place… 0 2024-01-15 2024-03-15
4 ABC-123 ABC-12… 004 34 M BLAC… Study… 0 2024-01-15 2024-03-15
5 ABC-123 ABC-12… 005 62 M ASIAN Place… 10 2024-01-15 2024-03-15
6 ABC-123 ABC-12… 006 49 M WHITE Study… 0 2024-01-15 2024-03-15
7 ABC-123 ABC-12… 007 58 M WHITE Place… 10 2024-01-15 2024-03-15
8 ABC-123 ABC-12… 008 75 F ASIAN Study… 0 2024-01-15 2024-03-15
9 ABC-123 ABC-12… 009 51 M WHITE Study… 5 2024-01-15 2024-03-15
10 ABC-123 ABC-12… 010 46 F BLAC… Place… 5 2024-01-15 2024-03-15
11 ABC-123 ABC-12… 011 28 F WHITE Study… 0 2024-01-15 2024-03-15
12 ABC-123 ABC-12… 012 33 F WHITE Place… 0 2024-01-15 2024-03-15
13 ABC-123 ABC-12… 013 46 F WHITE Study… 5 2024-01-15 2024-03-15
14 ABC-123 ABC-12… 014 31 F BLAC… Study… 10 2024-01-15 2024-03-15
15 ABC-123 ABC-12… 015 51 F BLAC… Place… 5 2024-01-15 2024-03-15
16 ABC-123 ABC-12… 016 43 M WHITE Place… 10 2024-01-15 2024-03-15
17 ABC-123 ABC-12… 017 61 F ASIAN Study… 5 2024-01-15 2024-03-15
18 ABC-123 ABC-12… 018 67 F WHITE Study… 10 2024-01-15 2024-03-15
19 ABC-123 ABC-12… 019 65 X white PLACE… 0 2023-12-01 2024-03-15
20 ABC-123 ABC-12… 019 NA F WHITE Study… 5 2024-01-15 NA
# ℹ 6 more variables: SAFFL <chr>, ITTFL <chr>, COMPLFL <chr>, DTHFL <chr>,
# DTHDTC <chr>, DTHDT <date>check_duplicates <- function(data, id_var = "USUBJID") {
cat("=== Duplicate Subject Check ===\n")
duplicates <- data %>%
group_by(!!sym(id_var)) %>%
filter(n() > 1) %>%
ungroup()
if (nrow(duplicates) > 0) {
cat("❌ ISSUE FOUND: Duplicate subjects detected\n")
cat("Number of duplicate records:", nrow(duplicates), "\n")
cat("Duplicate subjects:\n")
print(duplicates %>% select(!!sym(id_var), SUBJID, AGE, SEX))
return(list(
status = "FAIL",
count = nrow(duplicates),
details = duplicates
))
} else {
cat("✅ PASS: No duplicate subjects found\n")
return(list(status = "PASS", count = 0, details = NULL))
}
}
# Run duplicate check
dup_result <- check_duplicates(adsl_qc)=== Duplicate Subject Check ===
❌ ISSUE FOUND: Duplicate subjects detected
Number of duplicate records: 2
Duplicate subjects:
# A tibble: 2 × 4
USUBJID SUBJID AGE SEX
<chr> <chr> <dbl> <chr>
1 ABC-123-019 019 65 X
2 ABC-123-019 019 NA F check_missing_critical <- function(data, critical_vars = c("USUBJID", "AGE", "SEX", "TRTSDT")) {
cat("\n=== Missing Critical Variables Check ===\n")
issues <- list()
for (var in critical_vars) {
if (var %in% names(data)) {
missing_count <- sum(is.na(data[[var]]) | data[[var]] == "", na.rm = TRUE)
missing_pct <- round(missing_count / nrow(data) * 100, 2)
if (!is.na(missing_count) && missing_count > 0) {
cat("❌", var, ":", missing_count, "missing values (", missing_pct, "%)\n")
# Show which subjects have missing values
missing_subjects <- data %>%
filter(is.na(!!sym(var)) | !!sym(var) == "") %>%
select(USUBJID, !!sym(var))
issues[[var]] <- list(
count = missing_count,
percentage = missing_pct,
subjects = missing_subjects
)
} else {
cat("✅", var, ": No missing values\n")
}
} else {
cat("⚠️", var, ": Variable not found in dataset\n")
}
}
return(issues)
}
# Run missing values check
missing_result <- check_missing_critical(adsl_qc)
=== Missing Critical Variables Check ===
✅ USUBJID : No missing values
❌ AGE : 1 missing values ( 5 %)
✅ SEX : No missing values
✅ TRTSDT : No missing valuescheck_controlled_terminology <- function(data) {
cat("\n=== Controlled Terminology Validation ===\n")
# Define expected controlled terminology
valid_sex <- c("M", "F")
valid_race <- c("WHITE", "BLACK OR AFRICAN AMERICAN", "ASIAN",
"AMERICAN INDIAN OR ALASKA NATIVE", "NATIVE HAWAIIAN OR OTHER PACIFIC ISLANDER")
valid_flags <- c("Y", "N", "")
issues <- list()
# Check SEX
invalid_sex <- data %>%
filter(!SEX %in% valid_sex) %>%
select(USUBJID, SEX)
if (nrow(invalid_sex) > 0) {
cat("❌ SEX: Invalid values found\n")
print(invalid_sex)
issues$SEX <- invalid_sex
} else {
cat("✅ SEX: All values valid\n")
}
# Check RACE (case sensitivity)
invalid_race <- data %>%
filter(!RACE %in% valid_race) %>%
select(USUBJID, RACE)
if (nrow(invalid_race) > 0) {
cat("❌ RACE: Invalid/inconsistent values found\n")
print(invalid_race)
issues$RACE <- invalid_race
} else {
cat("✅ RACE: All values valid\n")
}
# Check flags
flag_vars <- c("SAFFL", "ITTFL", "COMPLFL", "DTHFL")
for (flag_var in flag_vars) {
if (flag_var %in% names(data)) {
invalid_flags <- data %>%
filter(!is.na(!!sym(flag_var)) & !(!!sym(flag_var) %in% valid_flags)) %>%
select(USUBJID, !!sym(flag_var))
if (nrow(invalid_flags) > 0) {
cat("❌", flag_var, ": Invalid flag values found\n")
print(invalid_flags)
issues[[flag_var]] <- invalid_flags
} else {
cat("✅", flag_var, ": All flag values valid\n")
}
}
}
return(issues)
}
# Run controlled terminology check
ct_result <- check_controlled_terminology(adsl_qc)
=== Controlled Terminology Validation ===
❌ SEX: Invalid values found
# A tibble: 1 × 2
USUBJID SEX
<chr> <chr>
1 ABC-123-019 X
❌ RACE: Invalid/inconsistent values found
# A tibble: 1 × 2
USUBJID RACE
<chr> <chr>
1 ABC-123-019 white
✅ SAFFL : All flag values valid
✅ ITTFL : All flag values valid
✅ COMPLFL : All flag values valid
✅ DTHFL : All flag values validcheck_date_logic <- function(data) {
cat("\n=== Date Logic Validation ===\n")
issues <- list()
study_start <- as.Date("2024-01-01") # Define study start date
# Check treatment start vs study start
early_treatment <- data %>%
filter(!is.na(TRTSDT) & TRTSDT < study_start) %>%
select(USUBJID, TRTSDT)
if (nrow(early_treatment) > 0) {
cat("❌ Treatment start before study start:\n")
print(early_treatment)
issues$early_treatment <- early_treatment
} else {
cat("✅ All treatment start dates valid\n")
}
# Check treatment end vs treatment start
invalid_treatment_duration <- data %>%
filter(!is.na(TRTSDT) & !is.na(TRTEDT) & TRTEDT < TRTSDT) %>%
select(USUBJID, TRTSDT, TRTEDT)
if (nrow(invalid_treatment_duration) > 0) {
cat("❌ Treatment end before treatment start:\n")
print(invalid_treatment_duration)
issues$invalid_duration <- invalid_treatment_duration
} else {
cat("✅ All treatment durations valid\n")
}
# Check death date logic
death_logic_issues <- data %>%
filter(DTHFL == "Y" & (is.na(DTHDT) | DTHDT == "")) %>%
select(USUBJID, DTHFL, DTHDT)
if (nrow(death_logic_issues) > 0) {
cat("❌ Death flag Y but missing death date:\n")
print(death_logic_issues)
issues$death_logic <- death_logic_issues
} else {
cat("✅ Death date logic consistent\n")
}
return(issues)
}
# Run date logic check
date_result <- check_date_logic(adsl_qc)
=== Date Logic Validation ===
❌ Treatment start before study start:
# A tibble: 1 × 2
USUBJID TRTSDT
<chr> <date>
1 ABC-123-019 2023-12-01
✅ All treatment durations valid
✅ Death date logic consistentcheck_age_validation <- function(data) {
cat("\n=== Age Validation ===\n")
issues <- list()
# Check age range
invalid_age <- data %>%
filter(!is.na(AGE) & (AGE < 18 | AGE > 100)) %>%
select(USUBJID, AGE)
if (nrow(invalid_age) > 0) {
cat("❌ Age outside expected range (18-100):\n")
print(invalid_age)
issues$age_range <- invalid_age
} else {
cat("✅ All ages within expected range\n")
}
# Age distribution summary
cat("\nAge Distribution Summary:\n")
age_summary <- data %>%
filter(!is.na(AGE)) %>%
summarise(
n = n(),
mean_age = round(mean(AGE), 1),
median_age = median(AGE),
min_age = min(AGE),
max_age = max(AGE),
sd_age = round(sd(AGE), 1)
)
print(age_summary)
return(issues)
}
# Run age validation
age_result <- check_age_validation(adsl_qc)
=== Age Validation ===
✅ All ages within expected range
Age Distribution Summary:
# A tibble: 1 × 6
n mean_age median_age min_age max_age sd_age
<int> <dbl> <dbl> <dbl> <dbl> <dbl>
1 19 49.2 51 19 75 15# Mock additional domain for cross-validation
adae_qc <- tibble(
USUBJID = c(
sample(adsl_qc$USUBJID[1:15], 10), # Some subjects from ADSL
"ABC-123-999" # Subject not in ADSL
),
AEDECOD = sample(c("Headache", "Nausea", "Dizziness", "Fatigue"), 11, replace = TRUE),
AESEV = sample(c("MILD", "MODERATE", "SEVERE"), 11, replace = TRUE),
AEREL = sample(c("RELATED", "NOT RELATED", "POSSIBLY RELATED"), 11, replace = TRUE)
)
check_cross_domain <- function(adsl_data, ae_data) {
cat("\n=== Cross-Domain Validation ===\n")
issues <- list()
# Check for AE subjects not in ADSL
ae_subjects <- unique(ae_data$USUBJID)
adsl_subjects <- unique(adsl_data$USUBJID)
orphan_ae <- setdiff(ae_subjects, adsl_subjects)
if (length(orphan_ae) > 0) {
cat("❌ AE subjects not found in ADSL:\n")
cat(paste(orphan_ae, collapse = ", "), "\n")
issues$orphan_ae <- orphan_ae
} else {
cat("✅ All AE subjects found in ADSL\n")
}
# Check for ADSL subjects with safety flag Y but no AEs
safety_subjects <- adsl_data %>%
filter(SAFFL == "Y") %>%
pull(USUBJID)
subjects_no_ae <- setdiff(safety_subjects, ae_subjects)
cat("Safety population subjects without AEs:", length(subjects_no_ae), "\n")
if (length(subjects_no_ae) > 5) { # Only show first 5
cat("Examples:", paste(head(subjects_no_ae, 5), collapse = ", "), "\n")
}
return(issues)
}
# Run cross-domain check
cross_result <- check_cross_domain(adsl_qc, adae_qc)
=== Cross-Domain Validation ===
❌ AE subjects not found in ADSL:
ABC-123-999
Safety population subjects without AEs: 9
Examples: ABC-123-005, ABC-123-010, ABC-123-011, ABC-123-013, ABC-123-015 generate_qc_summary <- function(dup_result, missing_result, ct_result, date_result, age_result, cross_result) {
cat("\n", paste(rep("=", 60), collapse=""), "\n")
cat(" QUALITY CONTROL SUMMARY REPORT\n")
cat(paste(rep("=", 60), collapse=""), "\n")
cat("Dataset: ADSL\n")
cat("QC Date:", format(Sys.Date(), "%Y-%m-%d"), "\n")
cat("Total Records:", nrow(adsl_qc), "\n")
cat(paste(rep("-", 60), collapse=""), "\n")
# Summary of issues
total_issues <- 0
cat("ISSUE SUMMARY:\n")
if (dup_result$status == "FAIL") {
cat("• Duplicate subjects:", dup_result$count, "\n")
total_issues <- total_issues + dup_result$count
}
if (length(missing_result) > 0) {
missing_count <- sum(sapply(missing_result, function(x) x$count))
cat("• Missing critical values:", missing_count, "\n")
total_issues <- total_issues + missing_count
}
if (length(ct_result) > 0) {
ct_count <- sum(sapply(ct_result, nrow))
cat("• Controlled terminology issues:", ct_count, "\n")
total_issues <- total_issues + ct_count
}
if (length(date_result) > 0) {
date_count <- sum(sapply(date_result, nrow))
cat("• Date logic issues:", date_count, "\n")
total_issues <- total_issues + date_count
}
if (length(cross_result) > 0) {
cross_count <- length(unlist(cross_result))
cat("• Cross-domain issues:", cross_count, "\n")
total_issues <- total_issues + cross_count
}
cat(paste(rep("-", 60), collapse=""), "\n")
cat("TOTAL ISSUES IDENTIFIED:", total_issues, "\n")
if (total_issues == 0) {
cat("🎉 QC STATUS: PASS - No issues found\n")
} else {
cat("⚠️ QC STATUS: REVIEW REQUIRED\n")
}
cat(paste(rep("=", 60), collapse=""), "\n")
}
# Generate summary report
generate_qc_summary(dup_result, missing_result, ct_result, date_result, age_result, cross_result)
============================================================
QUALITY CONTROL SUMMARY REPORT
============================================================
Dataset: ADSL
QC Date: 2025-11-07
Total Records: 20
------------------------------------------------------------
ISSUE SUMMARY:
• Duplicate subjects: 2
• Missing critical values: 1
• Controlled terminology issues: 2
• Date logic issues: 1
• Cross-domain issues: 1
------------------------------------------------------------
TOTAL ISSUES IDENTIFIED: 7
⚠️ QC STATUS: REVIEW REQUIRED
============================================================ Here are effective GitHub Copilot prompts for clinical data QC:
# Example Copilot prompts for QC programming:
# 1. Generate validation function
# Prompt: "Create a function to validate SDTM dates against FDA requirements"
# 2. Check data completeness
# Prompt: "Write R code to check missing values in clinical trial dataset by treatment group"
# 3. Controlled terminology validation
# Prompt: "Create validation rules for CDISC controlled terminology in R using dplyr"
# 4. Cross-domain referential integrity
# Prompt: "Write function to check that all subjects in AE domain exist in DM domain"
# 5. Statistical outlier detection
# Prompt: "Identify statistical outliers in vital signs data using R"# Statistical outlier detection for continuous variables
detect_outliers <- function(data, var_name, method = "iqr") {
cat(paste("Outlier detection for", var_name, "using", method, "method:\n"))
values <- data[[var_name]][!is.na(data[[var_name]])]
if (method == "iqr") {
Q1 <- quantile(values, 0.25)
Q3 <- quantile(values, 0.75)
IQR <- Q3 - Q1
lower_bound <- Q1 - 1.5 * IQR
upper_bound <- Q3 + 1.5 * IQR
outliers <- data %>%
filter(!is.na(!!sym(var_name)) &
(!!sym(var_name) < lower_bound | !!sym(var_name) > upper_bound)) %>%
select(USUBJID, !!sym(var_name))
if (nrow(outliers) > 0) {
cat("Outliers detected:\n")
print(outliers)
} else {
cat("No outliers detected\n")
}
}
return(outliers)
}
# Check for age outliers
age_outliers <- detect_outliers(adsl_qc, "AGE")Outlier detection for AGE using iqr method:
No outliers detectedThis QC toolkit provides a comprehensive framework for clinical data validation using modern R programming techniques with AI assistance.