Sherley Project — Data Management

1 Variable Name Dictionary

Variables with the same name as the PDF / REDCap CSV are not listed. Only renamed and newly created variables are shown.

1.1 Renamed Variables

PDF / REDCap	In Code
bl_date	baseline_sample_date
fu_date	last_follow_up_date
baseline_creatinine	baseline_cre
cys_c_baseline	baseline_cys_c
er_status_90d	ed_status_90d
er_visit_date	ed_visit_date
er_visit_diff	ed_visit_diff
other_ed_visit	other_ed
hosp_visit_diff	hosp_visit_0_diff
hosp_visit_reason___1–11	hosp_visit_0_reason_1–11

1.2 Labeled REDCap Variables

These variables are stored as numeric codes in the REDCap CSV and labeled in section 5.7.

In Code	REDCap Coding
sex	1=Male, 2=Female
race	1=White, 2=Black, 3=Asian, 4=Other, 5=Unknown
ethnicity	1=Non-Hispanic, 2=Hispanic, 3=Unknown
tumor_stage	1=Local/neoadjuvant/early (stage I–II), 2=Advanced/metastatic (stage III–IV)
platin_group	1=Carboplatin, 2=Cisplatin, 3=Unknown
ecog_score	REDCap stores 1=ECOG 0, 2=ECOG 1, …, 7=N/A. Remapped to actual ECOG (0–5); 7 → NA
ecog_estimate	Same remapping as ecog_score; shown only when ecog_score = N/A

1.3 Derived — eGFR & Kidney Function

In Code	Description
ckd_epi_gfr_cre_cys	CKD-EPI 2012 combined creatinine + cystatin C eGFR (mL/min/1.73m²)
ckd_epi_gfr_cre_cys_unindex	Combined CKD-EPI eGFR adjusted for body surface area (mL/min)
baseline_cre_egfr_unindexed	CKD-EPI creatinine eGFR adjusted for body surface area (mL/min)
baseline_cys_c_egfr_unindexed	CKD-EPI cystatin C eGFR adjusted for body surface area (mL/min)

1.4 Derived — RPDR Pre-baseline Labs (within 45 days before baseline)

In Code	RPDR Group_Id	Units
pre_CRE_45days	CRE	mg/dL
pre_HGB_45days	HGB	g/dL
pre_PLT_45days	PLT	×10³/µL
pre_ALB_45days	ALB	g/dL
pre_GHBA1C_45days	GHBA1C	%

1.5 Derived — Exposure

In Code	Description
cr_cys	Creatinine-to-cystatin C ratio: baseline_cre / baseline_cys_c
cr_cys_7	Primary exposure (binary): 1 if cr_cys < 0.7, 0 otherwise
cr_cys_cat	4-level categorical: <0.5 / 0.5–0.7 / 0.7–0.9 / ≥0.9
exposure_epi_30	Binary: 1 if cystatin C eGFR < 0.7 × creatinine eGFR (CKD-EPI based)

1.6 Derived — Adverse Events (within 90 days of baseline)

In Code	Description
Anemia_g2_90days	Binary: 1 if HGB < 8 g/dL within 90 days (CTCAE grade ≥ 2)
Anemia_g3_90days	Binary: 1 if HGB < 6.5 g/dL within 90 days (CTCAE grade ≥ 3)
Throm_g2_90days	Binary: 1 if PLT < 75 ×10³/µL within 90 days (CTCAE grade ≥ 2)
Throm_g3_90days	Binary: 1 if PLT < 50 ×10³/µL within 90 days (CTCAE grade ≥ 3)
AKI_g2_90days	Binary: 1 if creatinine > 1.5× pre-baseline within 90 days (CTCAE grade ≥ 2)
AKI_g3_90days	Binary: 1 if creatinine > 3× pre-baseline within 90 days (CTCAE grade ≥ 3)
AKI_kdigo_90days	Binary: 1 if CRE ≥ 1.5× pre-baseline OR ≥ 0.3 mg/dL increase within any 2-day window, within 90 days (KDIGO)
grade_2_aes	Composite binary: 1 if any of anemia/throm/AKI grade ≥ 2
grade_3_aes	Composite binary: 1 if any of anemia/throm/AKI grade ≥ 3
time_to_anemia_2 … time_to_aki_3	Days from baseline to each AE (90 if no event)
time_to_grade_2_aes / time_to_grade_3_aes	Days to earliest composite AE (min across anemia/throm/AKI)

1.7 Derived — Hospitalizations & ED

Hospitalization reasons 1–11 per REDCap codebook: 1=Nausea/vomiting/GI, 2=Failure to thrive/fatigue/dehydration, 3=AKI, 4=Electrolyte abnormalities, 5=Cytopenia, 6=Febrile neutropenia, 7=Infection/sepsis, 8=Cancer progression, 9=Immune-related AE, 10=Planned surgery/port, 11=Other

In Code	Description
event_hosp_1 … event_hosp_11	Binary: 1 if hospitalized for that reason within 90 days
time_to_event_hosp_1 … time_to_event_hosp_11	Days from baseline to hospitalization by reason (90 if none)
event_hosp_immune_other	Binary: 1 if hospitalised for Immune-related AE (hosp_9) OR Other (hosp_11)
time_to_event_hosp_immune_other	Earliest of time_to_event_hosp_9 and time_to_event_hosp_11 (90 if neither occurred)
side_effect	Binary: 1 if any hospitalization for reasons 1–7 (platinum-related AEs)
time_to_side_effect	Days to first platinum-related hospitalization (90 if none)
extra_ed_n	Count of additional ED visits (beyond the first)
extra_hosp_n	Count of additional hospitalizations (beyond the first)

1.8 Derived — Mortality

In Code	Description
days_last_follow_up	Days from baseline to last follow-up date
death_90	Binary: 1 if died within 90 days (REDCap pt_status = 3 “Dead”)
time_to_death_90	Days to death, capped at 90 (90 if alive/censored)
time_to_death_90_status	0=censored (alive at last follow-up), 2=dead within 90 days

1.9 Derived — Comorbidities & Categories

In Code	Description
score	Charlson comorbidity index weighted score (from RPDR ICD codes via comorbidity package)
cci	Charlson category: “mild” (score 1–2), “moderate” (3–4), “severe” (≥5)
thyroid	Binary: 1 if thyroid disease in RPDR diagnoses or medications
smoking	Binary: 1 if smoking/tobacco use in RPDR diagnoses
steroids	Binary: 1 if corticosteroid dispensed within 30 days before baseline
hiv_rpdr	Binary: 1 if HIV in RPDR diagnoses or medications
bmi	Body mass index: weight(kg) / height(m)²
bmi_cat	1=Underweight (<18.5), 2=Normal (18.5–25), 3=Overweight (25–30), 4=Obese (≥30)
age_cat	1=≤65 years, 2=66–80 years, 3=>80 years
ici	Binary: 1 if receiving immune checkpoint inhibitor (Atezolizumab, Pembrolizumab, or Durvalumab)
cancer_type_cat_7	Factor from tumor_type: Thoracic(1), Breast(2), GI(3,4), GU(5), GYN(7), HEENT(8), Other(6,9)

1.10 Derived — Platinum Dosing

In Code	Description
carbo	Binary: 1 if platin_group = Carboplatin
cis	Binary: 1 if platin_group = Cisplatin
carbo_AUC	Carboplatin AUC (from platin_dose for carboplatin patients; NA for cisplatin)
cis_dose	Cisplatin dose in mg/m² (from platin_dose for cisplatin patients; NA for carboplatin)
high_dose	Binary factor: 1=High-dose (carbo AUC ≥ 5 OR cisplatin ≥ 75 mg/m²), 0=Low-dose
carbo_AUC_cat	Carboplatin AUC category: “<3” or “>=3”
cis_dose_cat	Cisplatin dose category: 1=<75 mg/m², 2=≥75 mg/m², 0=not cisplatin
cis_dose_high	Cisplatin dose factor: “<75” or “>=75” (NA for non-cisplatin)
cis_group	Cisplatin dose tier: 1=<40 mg/m², 2=40–60 mg/m², 3=≥60 mg/m²

1.11 Derived — Carboplatin Dose Discrepancy

Carboplatin is dosed via the Calvert formula: Dose(mg) = target AUC × (GFR + 25). The actual administered dose is in REDCap. CKD-EPI Cr-Cys eGFR predicts what the dose should be — the discrepancy between actual and predicted may relate to adverse outcomes.

In Code	Description
carbo_dose_redcap	Actual administered carboplatin dose (mg) from REDCap
carbo_dose_cre_cys	Predicted dose (mg) from CKD-EPI Cr-Cys: AUC × (eGFR Cr-Cys unindexed + 25)
dose_discrepancy	Actual − predicted (mg): carbo_dose_redcap − carbo_dose_cre_cys
dose_diff_pct	(actual − predicted) / predicted × 100 (%)
dd_gt25	Binary factor: 1 if dose_discrepancy > 25 mg, 0 otherwise
dd_gt50	Binary factor: 1 if dose_discrepancy > 50 mg, 0 otherwise
ecog_score_cat2	ECOG binary: “0” vs “>=1”
bmi_30	BMI binary: “<30” vs “>=30”
age_70	Age binary: “<70” vs “>=70”

1.12 Derived — Drug Indicators & Competing Risk

Drug indicators (binary 0/1) parsed from regimen_dose text by create_drug_indicators(): Bevacizumab, Paclitaxel, Pemetrexed, Durvalumab, Doxorubicin, Pembrolizumab, Etoposide, Gemcitabine, Atezolizumab, Pertuzumab, Trastuzumab, Docetaxel, Fluorouracil, Methotrexate, Vinblastine, Administered_alone (1 if no co-administered drugs detected), other_chemo (1 if any of Fluorouracil, Methotrexate, Vinblastine)

Competing risk variables: cmp_time_<outcome> / cmp_status_<outcome> for each outcome. Status coding: 0=censored (alive, no event by 90 days), 1=event occurred, 2=death (competing event)

---

2 Setup & Libraries

rm(list = ls())
library(readxl)
library(openxlsx)
library(dplyr)
library(tidyverse)
library(tableone)
library(writexl)
library(gtsummary)
library(survival)
library(ggplot2)
library(ggpubr)
library(splines)
library(lattice)
library(mice)
library(DescTools)
library(survminer)
library(finalfit)
library(tidycmprsk)
library(ggsurvfit)
library(comorbidity)
library(readr)
library(cowplot)
library(scales)

---

3 Functions (all from Copy 709)

Each function is defined in its own labeled cell.

3.1 standardize_colnames

standardize_colnames <- function(df) {
  # Replace non-alphanumeric characters with underscore
  names(df) <- gsub("[^[:alnum:]_]", "_", names(df))

  # Reduce multiple underscores to a single underscore
  names(df) <- gsub("_+", "_", names(df))

  # Remove underscore if it's at the end of the name
  names(df) <- gsub("_$", "", names(df))

  # Return the cleaned dataframe
  return(df)
}

3.2 create_events_and_times_1_11

Creates event_hosp_1–event_hosp_11 and time_to_event_hosp_1–time_to_event_hosp_11 from hosp_visit_0_* (baseline) and extra hosp_visit_1_*–hosp_visit_5_* rows.

create_events_and_times_1_11 <- function(data) {
    # Initialize hospital event and time variables
    for (i in 1:11) {
        data[[paste0("event_hosp_", i)]] <- 0
        data[[paste0("time_to_event_hosp_", i)]] <- NA
    }

    # Process each row
    for (i in seq_len(nrow(data))) {
        # If baseline hospital visit is missing, set all to 0/NA
        if (is.na(data[i, "hosp_visit_0_diff"])) {
            for (j in 1:11) {
                data[i, paste0("event_hosp_", j)] <- 0
                data[i, paste0("time_to_event_hosp_", j)] <- NA
            }
        } else {
            # Baseline hospital visit
            for (j in 1:11) {
                if (!is.na(data[i, paste0("hosp_visit_0_reason_", j)]) &&
                    data[i, paste0("hosp_visit_0_reason_", j)] == 1) {
                    data[i, paste0("event_hosp_", j)] <- 1
                    data[i, paste0("time_to_event_hosp_", j)] <- data[i, "hosp_visit_0_diff"]
                }
            }
        }

        # Additional hospital visits
        if (!is.na(data[i, "other_hosp"]) && data[i, "other_hosp"] == 1) {
            extra_hosp_n <- as.numeric(data[i, "extra_hosp_n"])
            if (!is.na(extra_hosp_n) && extra_hosp_n > 0) {
                for (visit in 1:extra_hosp_n) {
                    for (j in 1:11) {
                        if (!is.na(data[i, paste0("event_hosp_", j)]) &&
                            data[i, paste0("event_hosp_", j)] == 0 &&
                            !is.na(data[i, paste0("hosp_visit_", visit, "_reason_", j)]) &&
                            data[i, paste0("hosp_visit_", visit, "_reason_", j)] == 1) {
                            data[i, paste0("event_hosp_", j)] <- 1
                            data[i, paste0("time_to_event_hosp_", j)] <- data[i, paste0("hosp_visit_", visit, "_diff")]
                        }
                    }
                }
            }
        }
    }

    return(data)
}

3.3 create_drug_indicators

Parses regimen_dose free text to create 0/1 indicators per co-drug. Administered_alone = 1 when none of the 15 drugs are present.

create_drug_indicators <- function(data) {

  # Create indicator for Bevacizumab
  data$Bevacizumab <- ifelse(grepl("Bevacizumab|Avastin", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Paclitaxel
  data$Paclitaxel <- ifelse(grepl("Paclitaxel|Taxol|Abraxane", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Pemetrexed
  data$Pemetrexed <- ifelse(grepl("Pemetrexed|pem\\b", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Durvalumab
  data$Durvalumab <- ifelse(grepl("Durvalumab|Durva", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Doxorubicin
  data$Doxorubicin <- ifelse(grepl("Doxorubicin|Adriamycin", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Pembrolizumab
  data$Pembrolizumab <- ifelse(grepl("Pembrolizumab|Pembro", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Etoposide
  data$Etoposide <- ifelse(grepl("Etoposide|etop\\b", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Gemcitabine
  data$Gemcitabine <- ifelse(grepl("Gemcitabine|\\bGem\\b", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Atezolizumab
  data$Atezolizumab <- ifelse(grepl("Atezolizumab", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Pertuzumab
  data$Pertuzumab <- ifelse(grepl("Pertuzumab", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Trastuzumab
  data$Trastuzumab <- ifelse(grepl("Trastuzumab|Traztuzumab|Transtuzumab|Herceptin",
                                   data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Docetaxel
  data$Docetaxel <- ifelse(grepl("Docetaxel|Taxotere", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Fluorouracil
  data$Fluorouracil <- ifelse(grepl("Fluorouracil|5FU", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Methotrexate
  data$Methotrexate <- ifelse(grepl("Methotrexate", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Vinblastine
  data$Vinblastine <- ifelse(grepl("Vinblastine", data$regimen_dose, ignore.case = TRUE), 1, 0)

  # Create indicator for Administered alone (absence of all the above drugs)
  all_drugs <- c("Bevacizumab", "Paclitaxel", "Pemetrexed", "Durvalumab", "Doxorubicin",
                 "Pembrolizumab", "Etoposide", "Gemcitabine", "Atezolizumab", "Pertuzumab",
                 "Trastuzumab", "Docetaxel", "Fluorouracil", "Methotrexate", "Vinblastine")

  data$Administered_alone <- ifelse(rowSums(data[, all_drugs]) == 0, 1, 0)

  # Other chemotherapy: any of Fluorouracil / Methotrexate / Vinblastine, i.e.
  # the partner drugs not already broken out individually in Table 1.
  data$other_chemo <- ifelse(
    data$Fluorouracil == 1 | data$Methotrexate == 1 | data$Vinblastine == 1,
    1, 0
  )

  return(data)
}

3.4 cmp_variable

Builds competing-risk time and status for a single outcome. Status coding: 0 = censored, 1 = event occurred, 2 = death (competing event). If death occurs first, the death time and status (0/2) override the event.

cmp_variable <- function(data, event_time, event_status, name) {
  # Enquote the time and status variables to allow their evaluation within the data context
  event_time <- enquo(event_time)
  event_status <- enquo(event_status)


  # Create dynamic names for the new columns
  cmp_time_name <- paste0("cmp_time_", name)
  cmp_status_name <- paste0("cmp_status_", name)

  # Add new columns to the dataset using mutate()
  data %>%
    mutate(
      !!cmp_time_name := ifelse(time_to_death_90 <= !!event_time ,time_to_death_90,!!event_time),
      !!cmp_status_name := ifelse(time_to_death_90 <= !!event_time,time_to_death_90_status,!!event_status)
    )
}

3.5 pre_lab

Finds the closest lab value on or before index_date within the time window. Returns one row per patient (most recent qualifying lab result).

See Section 6.1 for the bug found in Sherley’s version of this function and how Copy 709 correctly implements it.

###### pre lab  index data included closest date
pre_lab <- function(data, month, time, index_date, Group_Id) {
  index_date <- enquo(index_date)
  if (month) {
    time_value <- time * 30
  } else {
    time_value <- time
  }


  lab_name <- paste0("pre_", Group_Id, "_", time_value, "days")
  lab_date_name <- paste0("pre_", Group_Id, "_", time_value, "days_date")

  result <- data %>%
    filter(Group_Id == !!Group_Id, grepl("^\\d+\\.\\d+$", Result)) %>%
    filter(!is.na(Result)) %>%
    mutate(Result = as.numeric(Result)) %>%
    filter(Seq_Date_Time >= (!!index_date) - time_value, Seq_Date_Time <= !!index_date) %>%
    arrange(EMPI, desc(Seq_Date_Time)) %>%
    distinct(EMPI, .keep_all = TRUE) %>%
    rename(!!lab_name := Result,
           !!lab_date_name := Seq_Date_Time) %>%
    select(EMPI,
           !!lab_name, !!lab_date_name)

  return(result)
}

3.6 post_lab_less_than_cutpoint

Finds the first post-baseline lab value below a clinical cutpoint within days. Used for post-baseline lab cutpoint outcomes.

## post lab cutoff function
post_lab_less_than_cutpoint <- function(data,days,index_date,Group_ID,Group_ID_name,outcome_name,cutpoint) {
  index_date <- enquo(index_date)
  lab_name <- paste0("post_", Group_ID_name, "_cutpoint_", days, "days")
  lab_date_name <- paste0("post_", Group_ID_name, "_cutpoint_", days, "days_date")

  result <- data %>%
    filter(Group_Id %in% Group_ID) %>%
    filter(grepl("^[><]?\\d+\\.?\\d* ?$",Result)) %>%
    mutate(Result=as.numeric(parse_number(Result))) %>%
    filter(!is.na(Result)) %>%
    filter(Seq_Date_Time > (!!index_date), Seq_Date_Time <= !!index_date + days) %>%
    filter(Result<cutpoint) %>%
    arrange(EMPI,Seq_Date_Time) %>%
    distinct(EMPI, .keep_all = TRUE) %>%
    rename(!!lab_name := Result,
           !!lab_date_name := Seq_Date_Time) %>%
    select(EMPI,
           !!lab_name, !!lab_date_name)
  return(result)
}

3.7 dia_list_name

Identifies patients with a given diagnosis (and optionally medication) on or before index_date. Supports excluding specific diagnosis name strings.

dia_list_name <- function(data, listed_names, index_date, excluded_names = NULL, med_data = NULL, med_names = NULL) {
  #
  if (!is.null(excluded_names)) {
    excluded_pattern <- paste(
      gsub("([][()])", "\\\\\\1", excluded_names),  #
      collapse = "|"
    )
  }

  if (is.null(excluded_names)) {
    dia_result <- data %>%
      filter(Date <= {{index_date}}) %>%
      filter(grepl(paste(listed_names, collapse = "|"), Diagnosis_Name, ignore.case = TRUE)) %>%
      distinct(EMPI)
  } else {
    dia_result <- data %>%
      filter(Date <= {{index_date}}) %>%
      filter(
        grepl(paste(listed_names, collapse = "|"), Diagnosis_Name, ignore.case = TRUE),
        !grepl(excluded_pattern, Diagnosis_Name, ignore.case = TRUE)  #
      ) %>%
      distinct(EMPI)
  }

  if (!is.null(med_data)) {
    med_result <- med_data %>%
      filter(Medication_Date <= {{index_date}}) %>%
      filter(grepl(paste(med_names, collapse = "|"), Medication, ignore.case = TRUE)) %>%
      distinct(EMPI)
  } else {
    med_result <- NULL
  }

  final_result <- rbind(dia_result, med_result) %>%
    distinct(EMPI)

  return(final_result)
}

---

4 Data Loading

4.1 REDCap export

master <- read_csv('Data/CystatinCInPatientsO_DATA_2026-05-26_1104.csv')

whole_cohort <- read_excel('Data/Patient IDS_Tianqi_20May2026.xlsx', sheet = 'Sheet3')

# Rename empi to EMPI (uppercase for RPDR compatibility)
master <- master %>% rename(EMPI = empi)

# Keep only patients whose EMPI appears in the whole cohort list
master <- master %>% filter(EMPI %in% whole_cohort$EMPI)

4.2 Standardise column names, rename ED / hosp variables

master <- standardize_colnames(master)

# Rename ED variables: PDF uses er_* prefix → code uses ed_*
# Rename hosp_visit_* → hosp_visit_0_* (baseline) and other_hosp_* → hosp_visit_*
master <- master %>%
   rename(ed_visit_diff  = er_visit_diff,
          ed_visit_date  = er_visit_date,
          ed_status_90d  = er_status_90d,
          other_ed       = other_ed_visit) %>%
  rename_with(~ gsub("^hosp_visit_", "hosp_visit_0_", .x), starts_with("hosp_visit_")) %>%
  rename_with(~ gsub("^other_hosp_", "hosp_visit_", .x), starts_with("other_hosp_")) %>%
  rename_with(~ gsub("_visit", "", .x) %>%
                gsub("_(\\d+)_", "_\\1_reason_", .), starts_with("other_ed_visit_")) %>%
  rename_with(~ gsub("^ed_visit_", "ed_visit_0_", .x), starts_with("ed_visit_")) %>%
  rename_with(~ gsub("^other_ed_", "ed_visit_", .x), starts_with("other_ed_")) %>%
  mutate(other_hosp = ifelse(other_hosp != 1 | is.na(other_hosp), 0, 1),
         other_ed   = ifelse(other_ed   != 1 | is.na(other_ed),   0, 1),
         extra_hosp_n = ifelse(other_hosp == 1,
                               rowSums(across(hosp_extra_1:hosp_extra_5), na.rm = TRUE), 0),
         extra_ed_n   = ifelse(other_ed == 1,
                               rowSums(across(ed_extra_1:ed_extra_3)), 0))

4.3 Create hospitalization event / time variables (reasons 1–11)

master <- create_events_and_times_1_11(master)

4.4 Create platinum-related side_effect variable (hosp reasons 1–7)

# side_effect = 1 if any of hosp reasons 1–7 (platinum-associated AEs) occurred
side_effect <- master %>%
  select(record_id, matches("^event_hosp_[1-7]$|^time_to_event_hosp_[1-7]$")) %>%
  rowwise() %>%
  mutate(
    side_effect = if_else(any(c_across(starts_with("event_hosp_")) == 1, na.rm = TRUE), 1, 0),
    time_to_side_effect = {
      if (side_effect == 1) {
        times <- c_across(starts_with("time_to_event_hosp_"))[c_across(starts_with("event_hosp_")) == 1]
        if (any(!is.na(times))) min(times, na.rm = TRUE) else NA_real_
      } else {
        NA_real_
      }
    }
  ) %>%
  ungroup() %>% select(record_id, side_effect, time_to_side_effect)

# Censor at 90 days
side_effect <- side_effect %>% mutate(
  side_effect = case_when(
    side_effect == 1 & time_to_side_effect > 90 ~ 0,
    TRUE ~ side_effect
  ),
  time_to_side_effect = case_when(
    side_effect == 1 & time_to_side_effect > 90 ~ 90,
    side_effect == 0 ~ 90,
    TRUE ~ time_to_side_effect
  )
)

4.4b Create overall hospitalization variable (any reason 1–11)

# any_hosp = 1 if hospitalized for ANY reason (1–11) within 90 days
overall_hosp <- master %>%
  select(record_id, matches("^event_hosp_\\d+$"), matches("^time_to_event_hosp_\\d+$")) %>%
  rowwise() %>%
  mutate(
    any_hosp = if_else(any(c_across(starts_with("event_hosp_")) == 1, na.rm = TRUE), 1, 0),
    time_to_any_hosp = {
      if (any_hosp == 1) {
        events <- c_across(starts_with("event_hosp_"))
        times  <- c_across(starts_with("time_to_event_hosp_"))
        valid  <- which(events == 1 & !is.na(times))
        if (length(valid) > 0) min(times[valid], na.rm = TRUE) else NA_real_
      } else {
        NA_real_
      }
    }
  ) %>%
  ungroup() %>% select(record_id, any_hosp, time_to_any_hosp)

# Censor at 90 days
overall_hosp <- overall_hosp %>% mutate(
  any_hosp = case_when(
    any_hosp == 1 & time_to_any_hosp > 90 ~ 0,
    TRUE ~ any_hosp
  ),
  time_to_any_hosp = case_when(
    any_hosp == 1 & time_to_any_hosp > 90 ~ 90,
    any_hosp == 0 ~ 90,
    TRUE ~ time_to_any_hosp
  )
)

4.5 Create final_master (merge side_effect and drug indicators)

# Cohort filtering (excl. missing cystatin C / platinum) happens in section 5.4.
# eGFR and AE outcomes are now computed from RPDR in sections 4.8–4.9.
final_master <- master %>%
  merge(side_effect, by = "record_id", all.x = TRUE) %>%
  merge(overall_hosp, by = "record_id", all.x = TRUE)

final_master <- create_drug_indicators(final_master)

4.6 Rename date variables; coerce types

# bl_date → baseline_sample_date
# fu_date → last_follow_up_date
final_master <- final_master %>%
  rename(
    baseline_sample_date = bl_date,
    last_follow_up_date  = fu_date
  )

final_master$last_follow_up_date  <- as.Date(final_master$last_follow_up_date)
final_master$baseline_sample_date <- as.Date(final_master$baseline_sample_date)
final_master$platin_dose          <- as.numeric(final_master$platin_dose)

4.7 Load RPDR files

# Diagnoses
dia <- read.table('/Users/to909/Partners HealthCare Dropbox/Tianqi Ouyang/Thoracic Oncology Spreadsheets/Carboplatin Paper/RPDR 2026-03-19/mes98_03182614492143216_Dia.txt',
  sep = "|", fill = TRUE, quote = "", header = TRUE)
dia$Date <- as.Date(dia$Date, "%m/%d/%Y")

# Labs (HGB, PLT, ALB, CRE, K, MG, NA — single pull)
lab <- read.table('/Users/to909/Partners HealthCare Dropbox/Tianqi Ouyang/Thoracic Oncology Spreadsheets/Carboplatin Paper/RPDR 2026-03-19/mes98_03182614492143216_Lab.txt',
  sep = "|", fill = TRUE, quote = "", header = TRUE) %>%
  select(EMPI, Seq_Date_Time, Result, Group_Id, Test_Description)
lab$Seq_Date_Time <- as.Date(lab$Seq_Date_Time, "%m/%d/%Y")

# Medications
med <- read.table('/Users/to909/Partners HealthCare Dropbox/Tianqi Ouyang/Thoracic Oncology Spreadsheets/Carboplatin Paper/RPDR 2026-03-19/mes98_03182614492143216_Med.txt',
  sep = "|", fill = TRUE, quote = "", header = TRUE) %>%
  select(EMPI, Medication_Date, Medication)
med$Medication_Date <- as.Date(med$Medication_Date, "%m/%d/%Y")

# Patient index: EMPI + baseline date (needed by pre/post lab functions)
master_list <- final_master %>% select(EMPI, baseline_sample_date)

master_lab     <- merge(master_list, lab,     by = "EMPI", all.x = TRUE)
master_dia     <- merge(master_list, dia,     by = "EMPI", all.y = TRUE)
master_med     <- merge(master_list, med,     by = "EMPI", all.y = TRUE)

# Exclude patients with no RPDR labs
empis_with_labs <- unique(lab$EMPI[lab$EMPI %in% final_master$EMPI])
empis_no_labs   <- setdiff(final_master$EMPI, empis_with_labs)
n_no_labs       <- length(empis_no_labs)

cat("Patients before RPDR lab exclusion:", nrow(final_master), "\n")

Patients before RPDR lab exclusion: 169 

cat("Patients with no RPDR labs (excluded):", n_no_labs, "\n")

Patients with no RPDR labs (excluded): 0 

if (n_no_labs > 0) {
  cat("  record_ids:",
      paste(final_master$record_id[final_master$EMPI %in% empis_no_labs], collapse = ", "), "\n")
}

final_master <- final_master %>% filter(EMPI %in% empis_with_labs)
master_lab   <- master_lab %>% filter(EMPI %in% empis_with_labs)

cat("Patients after RPDR lab exclusion:", nrow(final_master), "\n")

Patients after RPDR lab exclusion: 169 

4.8 Pipeline functions: eGFR formulas

Variables produced by this pipeline:

In Code	Formula	Description
male	sex == "Male"	0/1, required by CKD-EPI
BSA	DuBois	Body surface area (m²)
baseline_cre_egfr	CKD-EPI 2012 Cr	Indexed creatinine eGFR
baseline_cys_c_egfr	CKD-EPI 2012 CysC	Indexed cystatin eGFR
ckd_epi_gfr_cre_cys_unindex	CKD-EPI 2012 combined, unindexed	Primary combined eGFR for exposures
baseline_cre_egfr_unindexed	Cr eGFR × BSA/1.73	Body-adjusted
baseline_cys_c_egfr_unindexed	CysC eGFR × BSA/1.73	Body-adjusted
pre_CRE_45days	pre_lab()	Baseline CRE for AKI grading
post_Anemia_2/3_cutpoint_90days	post_lab_less_than_cutpoint()	Grade 2/3 anemia (HGB)
post_Throm_2/3_cutpoint_90days	post_lab_less_than_cutpoint()	Grade 2/3 thrombocytopenia (PLT)
post_AKI_2/3_cutpoint_90days	CRE > 1.5×/3× pre-baseline CRE	Grade 2/3 AKI (CTCAE)

# CKD-EPI 2012 creatinine-based eGFR (race-free CKD-EPI equation)
ckd_epi_gfr <- function(cre, male, age) {
  k <- ifelse(male == 1, 0.9, 0.7)
  a <- ifelse(male == 1, -0.302, -0.241)
  m <- ifelse(male == 1, 1, 1.012)
  gfr <- 142 * min(c(cre/k, 1))^a * max(c(cre/k, 1))^-1.2 * 0.9938^age * m
  return(gfr)
}

# CKD-EPI 2012 cystatin C-based eGFR (Inker et al. 2012)
# Age coefficient = 0.996 per official NKF/Inker 2012 (not 0.9961, which is the 2021 Cr+CysC value)
ckd_epi_cys_gfr <- function(cys, male, age) {
  gfr <- 133 * min(c(cys/0.8, 1))^(-0.499) * max(c(cys/0.8, 1))^(-1.328) *
         0.996^age * ifelse(male == 0, 0.932, 1)
  return(gfr)
}

# Combined creatinine + cystatin C eGFR — exact coefficients from R code summary.Rmd
# Note: coefficients differ from CKD-EPI 2021 race-free (a=-0.144/-0.219, max_cr=-0.544,
# min_cys=-0.323, max_cys=-0.778, age=0.9961, female=0.963)
ckd_epi_cre_cys_gfr <- function(creatinine, cystatin_c, male, age) {
  k      <- ifelse(male == 1, 0.9, 0.7)
  a      <- ifelse(male == 1, -0.144, -0.219)
  female <- ifelse(male == 0, 0.963, 1)
  gfr_cr_cys <- 135 *
    min(c(creatinine/k, 1.0))^a  *
    max(c(creatinine/k, 1.0))^(-0.544) *
    min(c(cystatin_c/0.8, 1.0))^(-0.323) *
    max(c(cystatin_c/0.8, 1.0))^(-0.778) *
    0.9961^age * female
  return(gfr_cr_cys)
}

# Cockcroft-Gault creatinine clearance (mL/min)
# Female factor = 0.85; male = 1.0
Cockcroft <- function(age, weight, male, cre) {
  if (male == 0) {
    value <- (140 - age) * weight * 0.85 / (72 * cre)
  } else {
    value <- (140 - age) * weight / (72 * cre)
  }
  return(value)
}

# Adjust indexed eGFR for body surface area
unindex <- function(value, bsa) {
  return(value * bsa / 1.73)
}

4.9 Pipeline execution: pre-labs, eGFR, AE outcomes

# male (0/1) and BSA are required by all eGFR formulas
final_master <- final_master %>%
  mutate(
    male = case_when(
      sex %in% c("Male", "male", 1)    ~ 1,
      sex %in% c("Female", "female", 2) ~ 0,
      TRUE ~ NA_real_
    ),
    BSA = 0.007184 * (weight ^ 0.425) * (height ^ 0.725)
  )

# pre_CRE_45days: baseline creatinine (AKI grade 2/3 reference denominator)
pre_CRE_45days <- pre_lab(master_lab, month = FALSE, time = 45,
                           index_date = baseline_sample_date, Group_Id = "CRE")

# Haematology / biochemistry pre-baseline labs
pre_HGB_45days <- pre_lab(master_lab, month = FALSE, time = 45,
                           index_date = baseline_sample_date, Group_Id = "HGB")
pre_PLT_45days <- pre_lab(master_lab, month = FALSE, time = 45,
                           index_date = baseline_sample_date, Group_Id = "PLT")
pre_ALB_45days <- pre_lab(master_lab, month = FALSE, time = 45,
                           index_date = baseline_sample_date, Group_Id = "ALB")
pre_GHBA1C_45days <- pre_lab(master_lab, month = FALSE, time = 45,
                              index_date = baseline_sample_date, Group_Id = "GHBA1C")

# Merge all pre-baseline labs into final_master
final_master <- final_master %>%
  merge(pre_CRE_45days,    by = "EMPI", all.x = TRUE) %>%
  merge(pre_HGB_45days,    by = "EMPI", all.x = TRUE) %>%
  merge(pre_PLT_45days,    by = "EMPI", all.x = TRUE) %>%
  merge(pre_ALB_45days,    by = "EMPI", all.x = TRUE) %>%
  merge(pre_GHBA1C_45days, by = "EMPI", all.x = TRUE)

# All eGFR variables computed via mapply() (scalar functions over rows).
# Uses PDF names baseline_creatinine and cys_c_baseline (renamed in section 5.8).
final_master <- final_master %>%
  mutate(
    cys_c_baseline      = as.numeric(cys_c_baseline),
    baseline_creatinine = as.numeric(baseline_creatinine),
  ) %>%
  mutate(
    baseline_cre_egfr            = mapply(ckd_epi_gfr, baseline_creatinine, male, age),
    baseline_cys_c_egfr          = mapply(ckd_epi_cys_gfr, cys_c_baseline, male, age),
    ckd_epi_gfr_cre_cys          = mapply(ckd_epi_cre_cys_gfr,
                                          baseline_creatinine, cys_c_baseline, male, age),
    baseline_cre_egfr_unindexed  = mapply(unindex, baseline_cre_egfr, BSA),
    baseline_cys_c_egfr_unindexed = mapply(unindex, baseline_cys_c_egfr, BSA),
    ckd_epi_gfr_cre_cys_unindex  = mapply(unindex, ckd_epi_gfr_cre_cys, BSA),
    ckd_epi_gfr_cre_cys_unindex_cap125 = pmin(ckd_epi_gfr_cre_cys_unindex, 125),
    cockcroft                          = mapply(Cockcroft, age, weight, male, baseline_creatinine)
  )

# Post-baseline AE lab outcomes (from RPDR lab — original pull).
# Dates are already Date type from section 4.7; no further conversion needed.

Anemia_g2_value <- post_lab_less_than_cutpoint(
  data = master_lab, days = 90, index_date = baseline_sample_date,
  Group_ID = "HGB", Group_ID_name = "Anemia_2", cutpoint = 10)   # HGB < 10 g/dL

Anemia_g3_value <- post_lab_less_than_cutpoint(
  data = master_lab, days = 90, index_date = baseline_sample_date,
  Group_ID = "HGB", Group_ID_name = "Anemia_3", cutpoint = 8)    # HGB < 8 g/dL

Throm_g2_value <- post_lab_less_than_cutpoint(
  data = master_lab, days = 90, index_date = baseline_sample_date,
  Group_ID = "PLT", Group_ID_name = "Throm_2", cutpoint = 75)    # PLT < 75×10^9/L

Throm_g3_value <- post_lab_less_than_cutpoint(
  data = master_lab, days = 90, index_date = baseline_sample_date,
  Group_ID = "PLT", Group_ID_name = "Throm_3", cutpoint = 50)    # PLT < 50×10^9/L

# AKI grade 2: post-baseline CRE > 1.5 × pre-baseline CRE (CTCAE)
AKI_g2_value <- master_lab %>%
  merge(pre_CRE_45days, by = "EMPI", all.x = TRUE) %>%
  filter(Group_Id == "CRE") %>%
  filter(grepl("^[><]?\\d+\\.?\\d* ?$", Result)) %>%
  mutate(Result = as.numeric(parse_number(Result))) %>%
  filter(!is.na(Result), !is.na(pre_CRE_45days)) %>%
  filter(Seq_Date_Time > baseline_sample_date,
         Seq_Date_Time <= baseline_sample_date + 90) %>%
  filter(Result > 1.5 * pre_CRE_45days) %>%
  arrange(EMPI, Seq_Date_Time) %>%
  distinct(EMPI, .keep_all = TRUE) %>%
  select(EMPI, Result, Seq_Date_Time) %>%
  rename(post_AKI_2_cutpoint_90days      = Result,
         post_AKI_2_cutpoint_90days_date = Seq_Date_Time)

# AKI (KDIGO): Rule 1 — CRE >= 1.5 × pre-baseline CRE
#              Rule 2 — any 2-day window with CRE increase >= 0.3 mg/dL
# Either rule triggers AKI; first qualifying date is used.
aki_kdigo_cre <- master_lab %>%
  merge(pre_CRE_45days, by = "EMPI", all.x = TRUE) %>%
  filter(Group_Id == "CRE") %>%
  filter(grepl("^[><]?\\d+\\.?\\d* ?$", Result)) %>%
  mutate(Result = as.numeric(parse_number(Result))) %>%
  filter(!is.na(Result), !is.na(pre_CRE_45days)) %>%
  filter(Seq_Date_Time > baseline_sample_date,
         Seq_Date_Time <= baseline_sample_date + 90) %>%
  arrange(EMPI, Seq_Date_Time)

# Rule 1: CRE >= 1.5 × pre-baseline
aki_rule1 <- aki_kdigo_cre %>%
  filter(Result >= 1.5 * pre_CRE_45days) %>%
  distinct(EMPI, .keep_all = TRUE) %>%
  select(EMPI, aki_date_r1 = Seq_Date_Time, aki_value_r1 = Result)

# Rule 2: within any 2-day window, CRE increased by >= 0.3 from lowest CRE in prior 2 days
aki_rule2 <- aki_kdigo_cre %>%
  group_by(EMPI) %>%
  mutate(
    lowest_cre_2d = purrr::map_dbl(seq_along(Seq_Date_Time), function(i) {
      window <- Result[Seq_Date_Time >= (Seq_Date_Time[i] - 2) &
                       Seq_Date_Time < Seq_Date_Time[i]]
      if (length(window) == 0) NA_real_ else min(window, na.rm = TRUE)
    })
  ) %>%
  ungroup() %>%
  filter(!is.na(lowest_cre_2d), Result >= lowest_cre_2d + 0.3) %>%
  arrange(EMPI, Seq_Date_Time) %>%
  distinct(EMPI, .keep_all = TRUE) %>%
  select(EMPI, aki_date_r2 = Seq_Date_Time, aki_value_r2 = Result)

# Combine: earliest date from either rule
AKI_kdigo_value <- merge(aki_rule1, aki_rule2, by = "EMPI", all = TRUE) %>%
  mutate(
    post_AKI_kdigo_cutpoint_90days_date = pmin(aki_date_r1, aki_date_r2, na.rm = TRUE),
    post_AKI_kdigo_cutpoint_90days = case_when(
      is.na(aki_date_r2) ~ aki_value_r1,
      is.na(aki_date_r1) ~ aki_value_r2,
      aki_date_r1 <= aki_date_r2 ~ aki_value_r1,
      TRUE ~ aki_value_r2
    )
  ) %>%
  select(EMPI, post_AKI_kdigo_cutpoint_90days, post_AKI_kdigo_cutpoint_90days_date)

# AKI grade 3: post-baseline CRE > 3 × pre-baseline CRE (CTCAE)
AKI_g3_value <- master_lab %>%
  merge(pre_CRE_45days, by = "EMPI", all.x = TRUE) %>%
  filter(Group_Id == "CRE") %>%
  filter(grepl("^[><]?\\d+\\.?\\d* ?$", Result)) %>%
  mutate(Result = as.numeric(parse_number(Result))) %>%
  filter(!is.na(Result), !is.na(pre_CRE_45days)) %>%
  filter(Seq_Date_Time > baseline_sample_date,
         Seq_Date_Time <= baseline_sample_date + 90) %>%
  filter(Result > 3 * pre_CRE_45days) %>%
  arrange(EMPI, Seq_Date_Time) %>%
  distinct(EMPI, .keep_all = TRUE) %>%
  select(EMPI, Result, Seq_Date_Time) %>%
  rename(post_AKI_3_cutpoint_90days      = Result,
         post_AKI_3_cutpoint_90days_date = Seq_Date_Time)

# Merge AE outcomes into final_master
final_master <- final_master %>%
  merge(Anemia_g2_value, by = "EMPI", all.x = TRUE) %>%
  merge(Anemia_g3_value, by = "EMPI", all.x = TRUE) %>%
  merge(Throm_g2_value,  by = "EMPI", all.x = TRUE) %>%
  merge(Throm_g3_value,  by = "EMPI", all.x = TRUE) %>%
  merge(AKI_g2_value,    by = "EMPI", all.x = TRUE) %>%
  merge(AKI_g3_value,    by = "EMPI", all.x = TRUE) %>%
  merge(AKI_kdigo_value, by = "EMPI", all.x = TRUE)

---

5 Data Cleaning & Variable Creation

5.1 Mortality variables

# days_last_follow_up: numeric days from baseline to last known follow-up date
final_master <- final_master %>%
  mutate(days_last_follow_up = as.numeric(last_follow_up_date - baseline_sample_date))

# decode pt_status (PDF name kept) to character label
final_master <- final_master %>%
  mutate(pt_status = case_when(
    pt_status == 1 ~ "Alive",
    pt_status == 2 ~ "On Hospice",
    pt_status == 3 ~ "Dead"
  ))

# time_to_death_90: follow-up time capped at 90 days
# time_to_death_90_status: 0=alive/censored, 2=dead (competing-risk coding)
# death_90: binary 0/1
final_master <- final_master %>%
  mutate(
    time_to_death_90 = ifelse(days_last_follow_up > 90, 90, days_last_follow_up),
    time_to_death_90_status = case_when(
      days_last_follow_up > 90                          ~ 0,
      days_last_follow_up <= 90 & pt_status == "Dead"  ~ 2,
      days_last_follow_up <= 90 & pt_status != "Dead"  ~ 0
    ),
    death_90    = ifelse(time_to_death_90_status == 2, 1, 0)
  )

5.2 Censor hospitalization time variables at 90 days

# For each reason 1–11: no-event patients get time = 90 (censored)
for (i in 1:11) {
  event_col <- paste0("event_hosp_", i)
  time_col  <- paste0("time_to_event_hosp_", i)
  final_master[[time_col]][final_master[[event_col]] == 0] <- 90
}

5.3 Adverse event outcome variables (from RPDR pipeline)

# AE dates come from RPDR (Seq_Date_Time), already Date type — no conversion needed.

# Binary AE flags: 1 if cutpoint value is non-missing (event occurred within 90 days)
final_master <- final_master %>%
  mutate(
    Anemia_g2_90days = ifelse(!is.na(post_Anemia_2_cutpoint_90days), 1, 0),
    Anemia_g3_90days = ifelse(!is.na(post_Anemia_3_cutpoint_90days), 1, 0),
    Throm_g2_90days  = ifelse(!is.na(post_Throm_2_cutpoint_90days),  1, 0),
    Throm_g3_90days  = ifelse(!is.na(post_Throm_3_cutpoint_90days),  1, 0),
    AKI_g2_90days    = ifelse(!is.na(post_AKI_2_cutpoint_90days),    1, 0),
    AKI_g3_90days    = ifelse(!is.na(post_AKI_3_cutpoint_90days),    1, 0),
    AKI_kdigo_90days = ifelse(!is.na(post_AKI_kdigo_cutpoint_90days), 1, 0)
  )

# Time-to-AE in days; NA (no event) replaced with 90 (censored)
final_master <- final_master %>%
  mutate(
    time_to_anemia_2 = post_Anemia_2_cutpoint_90days_date - baseline_sample_date,
    time_to_anemia_3 = post_Anemia_3_cutpoint_90days_date - baseline_sample_date,
    time_to_throm_2  = post_Throm_2_cutpoint_90days_date  - baseline_sample_date,
    time_to_throm_3  = post_Throm_3_cutpoint_90days_date  - baseline_sample_date,
    time_to_aki_2    = post_AKI_2_cutpoint_90days_date    - baseline_sample_date,
    time_to_aki_3    = post_AKI_3_cutpoint_90days_date    - baseline_sample_date,
    time_to_aki_kdigo = post_AKI_kdigo_cutpoint_90days_date - baseline_sample_date
  ) %>%
  mutate(across(
    c(time_to_anemia_2, time_to_anemia_3,
      time_to_throm_2,  time_to_throm_3,
      time_to_aki_2,    time_to_aki_3,
      time_to_aki_kdigo),
    ~ ifelse(is.na(.), 90, as.numeric(.))   # no-event patients censored at 90 days
  ))

# grade_2_aes: composite — any of anemia / throm / AKI grade 2
# grade_3_aes: composite — any of anemia / throm / AKI grade 3
final_master <- final_master %>%
  mutate(
    grade_2_aes = case_when(
      Anemia_g2_90days == 1 | Throm_g2_90days == 1 | AKI_g2_90days == 1 ~ 1,
      TRUE ~ 0
    ),
    grade_3_aes = case_when(
      Anemia_g3_90days == 1 | Throm_g3_90days == 1 | AKI_g3_90days == 1 ~ 1,
      TRUE ~ 0
    )
  )

5.4 Exclusion filter: missing cystatin C or platinum type

# Exclude patients missing baseline cystatin C or platinum group
final_master <- final_master %>%
  mutate(cys_c_baseline = as.numeric(cys_c_baseline))

cat("Missing cystatin C — record_id:",
    paste(final_master$record_id[is.na(final_master$cys_c_baseline)], collapse = ", "), "\n")

Missing cystatin C — record_id: 883 

cat("Missing platinum group — record_id:",
    paste(final_master$record_id[is.na(final_master$platin_group)], collapse = ", "), "\n")

Missing platinum group — record_id:  

final_master <- final_master %>%
  filter(!is.na(cys_c_baseline)) %>%
  filter(!is.na(platin_group))

5.5 Exposure variables

# cr_cys: raw ratio (creatinine / cystatin C)
final_master$cr_cys <- final_master$baseline_creatinine / final_master$cys_c_baseline

# cr_cys_cat: four ordered categories
final_master <- final_master %>%
  mutate(cr_cys_cat = case_when(
    cr_cys < 0.5                  ~ 1,
    cr_cys >= 0.5 & cr_cys < 0.7 ~ 2,
    cr_cys >= 0.7 & cr_cys < 0.9 ~ 3,
    cr_cys >= 0.9                 ~ 4
  ))
final_master$cr_cys_cat <- factor(
  final_master$cr_cys_cat,
  levels = c("1", "2", "3", "4"),
  labels = c("< 0.5", "0.5-0.7", "0.7-0.9", "≥ 0.9")
)

# cr_cys_7: primary binary exposure — 1 if cr_cys < 0.7
final_master$cr_cys_7 <- ifelse(final_master$cr_cys < 0.7, 1, 0)

# Alternative eGFR-based exposure
final_master <- final_master %>%
  mutate(
    # 30% discrepancy: cystatin eGFR vs creatinine eGFR (CKD-EPI)
    exposure_epi_30 = ifelse(baseline_cys_c_egfr < 0.7 * baseline_cre_egfr, 1, 0)
  )

table(final_master$cr_cys_7)

  0   1 
108  60 

5.6 BMI, age categories, ICI, CKD stage

# bmi: calculated from height (cm) and weight (kg)
# bmi_cat: 1=underweight <18.5, 2=normal 18.5–25, 3=overweight 25–30, 4=obese ≥30
# age_cat: 1=≤65, 2=66–80, 3=>80
# ici: immunotherapy co-administration (Atezolizumab OR Pembrolizumab OR Durvalumab)
final_master <- final_master %>%
  mutate(
    bmi = weight / (height / 100)^2,
    bmi_cat = factor(case_when(
      bmi < 18.5              ~ "Underweight (<18.5)",
      bmi >= 18.5 & bmi < 25 ~ "Normal (18.5-25)",
      bmi >= 25   & bmi < 30 ~ "Overweight (25-30)",
      bmi >= 30               ~ "Obese (>=30)",
      TRUE ~ NA_character_
    ), levels = c("Underweight (<18.5)", "Normal (18.5-25)", "Overweight (25-30)", "Obese (>=30)")),
    age_cat = factor(case_when(
      age <= 65             ~ "<=65",
      age > 65 & age <= 80 ~ "66-80",
      age > 80              ~ ">80"
    ), levels = c("<=65", "66-80", ">80")),
    ici = ifelse(Atezolizumab == 1 | Pembrolizumab == 1 | Durvalumab == 1, 1, 0)
  )

5.7 Label REDCap coded variables

# Apply human-readable labels to REDCap numeric-coded variables (per data dictionary)
final_master <- final_master %>%
  mutate(
    sex        = factor(sex, levels = c(2, 1), labels = c("Female", "Male")),
    race       = factor(race, levels = 1:5, labels = c("White", "Black", "Asian", "Other", "Unknown")),
    ethnicity  = factor(ethnicity, levels = 1:3, labels = c("Non-Hispanic", "Hispanic", "Unknown")),
    tumor_stage = factor(tumor_stage, levels = c(1, 2), labels = c("Local/Early", "Advanced/Metastatic")),
    platin_group = factor(platin_group, levels = 1:3, labels = c("Carboplatin", "Cisplatin", "Unknown")),
    # ecog_score: REDCap codes 1=ECOG 0, 2=ECOG 1, ..., 7=N/A
    ecog_score = case_when(
      ecog_score >= 1 & ecog_score <= 6 ~ ecog_score - 1,  # remap to actual ECOG 0–5
      ecog_score == 7 ~ NA_real_,                           # N/A → missing
      TRUE ~ NA_real_
    ),
    ecog_estimate = case_when(
      ecog_estimate >= 1 & ecog_estimate <= 6 ~ ecog_estimate - 1,
      TRUE ~ NA_real_
    )
  ) %>%
  droplevels()

5.8 Cancer type categories (7 groups)

# cancer_type_cat_7: recode REDCap tumor_type (numeric 1–9) into 7 clinical groups
# REDCap coding (from data dictionary):
#   1 = Lung → Thoracic
#   2 = Breast
#   3 = Upper GI/esophagus/gastric  → GI
#   4 = Lower GI/colon/rectum      → GI
#   5 = Genitourinary/RCC/bladder   → GU
#   6 = Mesothelioma                → Other
#   7 = Gyn/ovarian/uterus/cervix  → GYN
#   8 = HEENT/Laryngeal/head & neck → HEENT
#   9 = Other                       → Other
final_master <- final_master %>%
  mutate(
    cancer_type_cat_7 = factor(case_when(
      tumor_type == 1             ~ 1,   # Thoracic (Lung)
      tumor_type == 2             ~ 2,   # Breast
      tumor_type %in% c(3, 4)    ~ 3,   # GI (Upper + Lower)
      tumor_type == 5             ~ 4,   # GU
      tumor_type == 7             ~ 5,   # GYN
      tumor_type == 8             ~ 6,   # HEENT
      tumor_type %in% c(6, 9)    ~ 7,   # Other (Mesothelioma + Other)
      TRUE ~ NA_real_
    ), levels = 1:7, labels = c("Thoracic", "Breast", "GI", "GU", "GYN", "HEENT", "Other"))
  )

table(final_master$cancer_type_cat_7)

Thoracic   Breast       GI       GU      GYN    HEENT    Other 
     168        0        0        0        0        0        0 

5.8 Platinum dosing variables

# carbo / cis: binary platinum type (platin_group is now a labeled factor)
# carbo_AUC: carboplatin AUC (from platin_dose); cis_dose: cisplatin mg/m²
# cis_group: cisplatin dose tier (1=<40, 2=40–60, 3=≥60 mg/m²)
# dose_1: high-dose flag (carbo AUC≥5 OR cisplatin≥75 mg/m²)
# carbo_AUC_cat: "<3" vs ">=3" (factor, NA if not carboplatin)
# cis_dose_cat: 1=<75 mg/m², 2=≥75, 0=not cisplatin
final_master <- final_master %>%
  mutate(
    carbo        = ifelse(platin_group == "Carboplatin", 1, 0),
    cis          = ifelse(platin_group == "Cisplatin",   1, 0),
    carbo_AUC    = case_when(carbo == 1 ~ platin_dose, TRUE ~ NA_real_),
    cis_dose     = case_when(cis == 1 ~ platin_dose, TRUE ~ NA_real_),
    cis_group    = case_when(
      cis_dose < 40                    ~ 1,
      cis_dose >= 40 & cis_dose < 60  ~ 2,
      cis_dose >= 60                   ~ 3,
      TRUE ~ NA_real_
    ),
    high_dose = factor(case_when(
      carbo_AUC >= 5 | cis_dose >= 75 ~ 1,
      carbo_AUC <  5 | cis_dose <  75 ~ 0,
      TRUE ~ NA_real_
    )),
    carbo_AUC_cat = factor(case_when(
      carbo_AUC < 3  ~ "<3",
      carbo_AUC >= 3 ~ ">=3",
      TRUE ~ NA_character_
    ), levels = c("<3", ">=3")),
    cis_dose_cat = case_when(
      cis_dose <  75 ~ 1,
      cis_dose >= 75 ~ 2,
      TRUE ~ 0
    ),
    # Labeled cisplatin dose factor for Table 1 (NA for non-cisplatin patients,
    # symmetric to carbo_AUC_cat which is NA for non-carboplatin patients)
    cis_dose_high = factor(case_when(
      cis_dose <  75 ~ "<75",
      cis_dose >= 75 ~ ">=75",
      TRUE ~ NA_character_
    ), levels = c("<75", ">=75"))
  )

table(final_master$high_dose)

  0   1 
 30 138 

5.9 Carboplatin dose discrepancy variables (from Leyre dose.Rmd)

# Carboplatin is dosed via Calvert formula: Dose(mg) = target AUC × (GFR + 25)
# The actual administered dose is recorded in REDCap (dosage_mg).
# CKD-EPI Cr-Cys eGFR predicts what the dose should be — the discrepancy
# between actual and predicted may relate to adverse outcomes.
#
# carbo_dose_redcap:  actual administered dose from REDCap (dosage_mg)
# carbo_dose_cre_cys: predicted dose using CKD-EPI Cr-Cys unindexed eGFR
# dose_discrepancy:   actual - predicted (mg); positive = patient got more than CKD-EPI predicts
# dose_diff_pct:      (actual - predicted) / predicted × 100

final_master <- final_master %>%
  mutate(
    carbo_dose_redcap  = as.numeric(dosage_mg),
    carbo_dose_cre_cys = carbo_AUC * (ckd_epi_gfr_cre_cys_unindex_cap125 + 25),
    dose_discrepancy   = carbo_dose_redcap - carbo_dose_cre_cys,
    dose_diff_pct      = (carbo_dose_redcap - carbo_dose_cre_cys) / carbo_dose_cre_cys * 100,
    dd_gt25            = factor(ifelse(!is.na(dose_discrepancy) & dose_discrepancy > 25, 1, 0)),
    dd_gt50            = factor(ifelse(!is.na(dose_discrepancy) & dose_discrepancy > 50, 1, 0)),
    ecog_score_cat2    = factor(ifelse(ecog_score == 0, "0", ">=1"), levels = c("0", ">=1")),
    bmi_30 = factor(ifelse(bmi < 30, "<30", ">=30")),
    age_70 = factor(ifelse(age < 70, "<70", ">=70"))
  )

5.10 Rename baseline lab variables (Copy 709 convention)

# baseline_cre: from baseline_creatinine (PDF name)
# baseline_cre: from baseline_creatinine (PDF name)
# baseline_cys_c: from cys_c_baseline (PDF name)
final_master <- final_master %>%
  rename(
    baseline_cre   = baseline_creatinine,
    baseline_cys_c = cys_c_baseline
  )

5.10 Electrolyte outcomes — removed

K, MG, NA not in current RPDR lab pull (only ALB, CRE, HGB, PLT, GHBA1C). Hypokalemia, Hypomagnesemia, Hyponatremia outcomes removed.

5.11 Thyroid disease (RPDR diagnosis + medication)

thyroid <- c("hypothyroidism", "hyperthyroidism", "thyroiditis")
thyroid_med_list <- c("Methimazole", "Tapazole", "Propylthiouracil", "Potassium Iodide",
                      "Radioactive Iodine ", "levothyroxine", "Euthyrox", "Synthroid",
                      "Eltroxin", "Desiccated Thyroid Extract", "Armour Thyroid")

thyroid_final <- dia_list_name(
  master_dia,
  listed_names = thyroid,
  med_data     = master_med,
  med_names    = thyroid_med_list,
  index_date   = baseline_sample_date
)

# thyroid: 1 if diagnosed or treated for thyroid disease before baseline
final_master <- final_master %>%
  mutate(thyroid = ifelse(EMPI %in% thyroid_final$EMPI, 1, 0))

5.12 Smoking status (RPDR diagnosis)

# smoking: 1 if tobacco/smoking-related diagnosis found before baseline
smoking_terms <- c(
  "tobacco use", "tobacco dependence", "tobacco abuse",
  "nicotine dependence", "nicotine use",
  "smoking", "smoker", "cigarette"
)

smoking_final <- dia_list_name(
  master_dia,
  listed_names = smoking_terms,
  index_date   = baseline_sample_date
)

final_master <- final_master %>%
  mutate(smoking = ifelse(EMPI %in% smoking_final$EMPI, 1, 0))

5.13 Corticosteroid use (RPDR medication)

# steroids: 1 if corticosteroid medication dispensed within 30 days before baseline.
# The exclusion list (no_steroid_names) follows the reference definition in
# "Reference code/prospective cys.Rmd": drop inhaled, nebulized, topical,
# ophthalmic, and Symbicort dispensings so the flag reflects systemic
# corticosteroid use, not puffers / creams / eye drops.
steroid_med_list <- c(
  "Prednisone", "Prednisolone", "Methylprednisolone", "Medrol",
  "Dexamethasone", "Decadron", "Hydrocortisone", "Cortisol",
  "Budesonide", "Triamcinolone", "Betamethasone", "Fludrocortisone",
  "Cortisone"
)
no_steroid_names <- c("Inhaler", "Nebulization", "Ointment", "Ophthalmic", "Symbicort")

steroid_final <- master_med %>%
  filter(!is.na(Medication_Date),
         Medication_Date <= baseline_sample_date,
         Medication_Date >= baseline_sample_date - 30) %>%
  filter(grepl(paste(steroid_med_list, collapse = "|"), Medication, ignore.case = TRUE),
         !grepl(paste(no_steroid_names, collapse = "|"), Medication, ignore.case = TRUE)) %>%
  distinct(EMPI)

final_master <- final_master %>%
  mutate(steroids = ifelse(EMPI %in% steroid_final$EMPI, 1, 0))

5.14 HIV from RPDR (hiv_rpdr)

# hiv_rpdr: HIV status from RPDR diagnoses + medications (Sherley's pipeline).
# Note: There is no hiv column in the REDCap CSV. The hiv variable in Copy 709 / cleaned_master
# was always derived from RPDR. Sherley's final code Rmd used the name hiv_new.
#
# Step 1: dia_list_name with HIV terms, excluding false-positive diagnosis names
# Step 2: Further exclude patients whose HIV diagnosis co-occurs with hepatitis B/C

no_hiv <- c(
  "1 - Hives-Oncall",
  "Contact with and (suspected) exposure to human immunodeficiency virus [hiv]",
  "Encounter for screening for human immunodeficiency virus [hiv]",
  "counseling",
  "Inconclusive laboratory evidence of human immunodeficiency virus [hiv]",
  "Asymptomatic human immunodeficiency virus [hiv] infection status",
  "Family history of human immunodeficiency virus [hiv] disease",
  "HIV exposure-Oncall"
)

hiv_terms <- c("hiv", "human immunodeficiency virus")

hiv_med_list <- c(
  "abacavir", "ziagen", "emtricitabine", "emtriva", "lamivudine", "tenofovir", "viread",
  "zidovudine", "retrovir", "doravirine", "pifeltro", "efavirenz", "sustiva", "etravirine",
  "intelence", "nevirapine", "viramune", "rilpivirine", "edurant", "atazanavir", "reyataz",
  "darunavir", "prezista", "fosamprenavir", "lexiva", "ritonavir", "norvir", "saquinavir",
  "invirase", "tipranavir", "aptivus", "enfuvirtide", "fuzeon", "maraviroc", "selzentry",
  "cabotegravir", "vocabria", "dolutegravir", "tivicay", "raltegravir", "isentress", "fostemsavir",
  "rukobia", "ibalizumab", "trogarzo", "lenacapavir", "sunlenca",
  "atripla", "biktarvy", "cabenuva", "cimduo", "combivir", "complera", "descovy", "dovato",
  "epzicom", "evotaz", "genvoya", "juluca", "kaletra", "odefsey", "prezcobix", "stribild",
  "symtuza", "triumeq", "truvada", "trizivir",
  "didanosine", "stavudine", "videx", "zerit", "delavirdine", "rescriptor", "indinavir",
  "crixivan", "nelfinavir", "viracept", "fortovase"
)

hiv_rpdr_initial <- dia_list_name(
  master_dia,
  listed_names  = hiv_terms,
  excluded_names = no_hiv,
  med_data       = master_med,
  med_names      = hiv_med_list,
  index_date     = baseline_sample_date
)

# Step 2: Further filter — exclude patients whose HIV diagnosis also mentions hepatitis B/C
hiv_rpdr_cleaned <- hiv_rpdr_initial %>%
  left_join(master_dia %>% select(EMPI, Diagnosis_Name), by = "EMPI") %>%
  filter(
    str_detect(Diagnosis_Name, regex("hiv|human immunodeficiency virus", ignore_case = TRUE)),
    !str_detect(Diagnosis_Name, regex("hepatitis B|hepatitis C", ignore_case = TRUE))
  ) %>%
  distinct(EMPI)

final_master <- final_master %>%
  mutate(hiv_rpdr = ifelse(EMPI %in% hiv_rpdr_cleaned$EMPI, 1, 0))

table(final_master$hiv_rpdr)

  0   1 
165   3 

5.14 Charlson Comorbidity Index

### Charlson comorbidity index
ici_diag <- dia %>% subset(select = c(EMPI, Diagnosis_Name, Date, Code, Code_Type))
ici_diag <- ici_diag %>% merge(final_master %>% select(EMPI, baseline_sample_date),
                                by = "EMPI", all.y = TRUE)
ici_diag$Date <- as.Date(ici_diag$Date, "%m/%d/%Y")
ici_diag_pre  <- subset(ici_diag, Date < baseline_sample_date)

# Leukemia and lymphoma must be extracted manually (ICD code map includes them under cancer)
leukemia <- ici_diag_pre[grep("leukemia", ici_diag_pre$Diagnosis_Name, ignore.case = TRUE), ]
leukemia <- leukemia[!grepl("lymphoma", leukemia$Diagnosis_Name, ignore.case = TRUE), ]
leukemia <- as.data.frame(table(leukemia$EMPI))
names(leukemia)[names(leukemia) == "Var1"] <- "EMPI"
names(leukemia)[names(leukemia) == "Freq"] <- "leukemia"
leukemia$leukemia[leukemia$leukemia > 0] <- 1

lymphoma <- ici_diag_pre[grep("lymphoma", ici_diag_pre$Diagnosis_Name, ignore.case = TRUE), ]
lymphoma <- lymphoma[!grepl("leukemia", lymphoma$Diagnosis_Name, ignore.case = TRUE), ]
lymphoma <- as.data.frame(table(lymphoma$EMPI))
names(lymphoma)[names(lymphoma) == "Var1"] <- "EMPI"
names(lymphoma)[names(lymphoma) == "Freq"] <- "lymphoma"
lymphoma$lymphoma[lymphoma$lymphoma > 0] <- 1

leuk_lymph <- merge(leukemia, lymphoma, by = "EMPI", all = TRUE)

# Compute comorbidity scores separately for ICD-9 and ICD-10
diag_icd9  <- subset(ici_diag_pre, Code_Type == "ICD9",  select = c("EMPI", "Code"))
diag_icd10 <- subset(ici_diag_pre, Code_Type == "ICD10", select = c("EMPI", "Code"))
diag_icd9  <- diag_icd9[!(diag_icd9$Code == ""), ]
diag_icd10 <- diag_icd10[!(diag_icd10$Code == ""), ]

charlson9  <- comorbidity(x = diag_icd9,  id = "EMPI", code = "Code",
                           map = "charlson_icd9_quan",  assign0 = FALSE)
charlson10 <- comorbidity(x = diag_icd10, id = "EMPI", code = "Code",
                           map = "charlson_icd10_quan", assign0 = FALSE)

charlson <- merge(charlson9, charlson10, by = "EMPI", all = TRUE)
charlson[is.na(charlson)] <- 0

# Take maximum across ICD-9 / ICD-10 for each domain
charlson$ami      <- pmax(charlson$mi.x,       charlson$mi.y)
charlson$chf      <- pmax(charlson$chf.x,      charlson$chf.y)
charlson$pvd      <- pmax(charlson$pvd.x,      charlson$pvd.y)
charlson$cevd     <- pmax(charlson$cevd.x,     charlson$cevd.y)
charlson$dementia <- pmax(charlson$dementia.x,  charlson$dementia.y)
charlson$copd     <- pmax(charlson$cpd.x,      charlson$cpd.y)
charlson$rheumd   <- pmax(charlson$rheumd.x,   charlson$rheumd.y)
charlson$pud      <- pmax(charlson$pud.x,      charlson$pud.y)
charlson$mld      <- pmax(charlson$mld.x,      charlson$mld.y)
charlson$diab     <- pmax(charlson$diab.x,     charlson$diab.y)
charlson$diabwc   <- pmax(charlson$diabwc.x,   charlson$diabwc.y)
charlson$hp       <- pmax(charlson$hp.x,       charlson$hp.y)
charlson$rend     <- pmax(charlson$rend.x,     charlson$rend.y)
charlson$canc     <- pmax(charlson$canc.x,     charlson$canc.y)
charlson$msld     <- pmax(charlson$msld.x,     charlson$msld.y)
charlson$metacanc <- pmax(charlson$metacanc.x, charlson$metacanc.y)
charlson$aids     <- pmax(charlson$aids.x,     charlson$aids.y)

charlson <- charlson %>%
  subset(select = c(EMPI, ami, chf, pvd, cevd, dementia, copd, rheumd, pud, mld,
                    diab, diabwc, hp, rend, canc, msld, metacanc, aids))
charlson$canc <- 1  # all patients have cancer by definition

charlson <- merge(charlson, leuk_lymph, by = "EMPI", all = TRUE)
charlson[is.na(charlson)] <- 0

ages <- final_master %>% subset(select = c(EMPI, age))
charlson <- merge(charlson, ages, by.x = 1, by.y = 1)

# Weighted Charlson score with age adjustment (Quan 2011)
charlson$score <-
  charlson$ami + charlson$chf + charlson$pvd + charlson$cevd +
  charlson$dementia + charlson$copd + charlson$rheumd + charlson$pud +
  ifelse(charlson$mld == 1 & charlson$msld == 0, 1, 0) +
  ifelse(charlson$msld == 1, 3, 0) +
  charlson$diab + charlson$diabwc +
  charlson$hp * 2 + charlson$rend * 2 + charlson$canc * 2 +
  charlson$metacanc * 4 + charlson$aids * 6 +
  charlson$leukemia * 2 + charlson$lymphoma * 2 +
  ifelse(charlson$age >= 50 & charlson$age < 60, 1, 0) +
  ifelse(charlson$age >= 60 & charlson$age < 70, 2, 0) +
  ifelse(charlson$age >= 70 & charlson$age < 80, 3, 0) +
  ifelse(charlson$age >= 80, 4, 0)

# cci: categorical severity
charlson_sc <- charlson %>% subset(select = c(EMPI, score))
charlson_sc$cci <- ifelse(charlson_sc$score < 3, "mild",
                   ifelse(charlson_sc$score >= 3 & charlson_sc$score < 5, "moderate", "severe"))

final_master <- final_master %>% merge(charlson_sc, by = "EMPI", all.x = TRUE)

5.15 Competing risk variables

# Apply cmp_variable() to build cmp_time_* and cmp_status_* for each outcome.
# AE outcomes (anemia, throm, aki, grade_2/3_aes) are now computed from RPDR pipeline
# (section 4.9) and their competing-risk variables are computed here.

# Adverse event competing risk
final_master <- cmp_variable(final_master, event_time = time_to_anemia_2, event_status = Anemia_g2_90days, name = "anemia_2")
final_master <- cmp_variable(final_master, event_time = time_to_anemia_3, event_status = Anemia_g3_90days, name = "anemia_3")
final_master <- cmp_variable(final_master, event_time = time_to_throm_2,  event_status = Throm_g2_90days,  name = "throm_2")
final_master <- cmp_variable(final_master, event_time = time_to_throm_3,  event_status = Throm_g3_90days,  name = "throm_3")
final_master <- cmp_variable(final_master, event_time = time_to_aki_2,    event_status = AKI_g2_90days,    name = "aki_2")
final_master <- cmp_variable(final_master, event_time = time_to_aki_3,    event_status = AKI_g3_90days,    name = "aki_3")
final_master <- cmp_variable(final_master, event_time = time_to_aki_kdigo, event_status = AKI_kdigo_90days, name = "aki_kdigo")
# Composite AE: use the earliest event time across anemia/throm/AKI
final_master <- final_master %>%
  mutate(
    time_to_grade_2_aes = pmin(time_to_anemia_2, time_to_throm_2, time_to_aki_2, na.rm = TRUE),
    time_to_grade_3_aes = pmin(time_to_anemia_3, time_to_throm_3, time_to_aki_3, na.rm = TRUE)
  )
final_master <- cmp_variable(final_master, event_time = time_to_grade_2_aes, event_status = grade_2_aes, name = "grade_2_aes")
final_master <- cmp_variable(final_master, event_time = time_to_grade_3_aes, event_status = grade_3_aes, name = "grade_3_aes")

# Platinum-related hospitalization (side effects: hosp reasons 1–7)
final_master <- cmp_variable(final_master,
  event_time = time_to_side_effect, event_status = side_effect, name = "side_effect")

# Hospitalization by reason 1–11
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_1,  event_status = event_hosp_1,  name = "hosp_1")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_2,  event_status = event_hosp_2,  name = "hosp_2")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_3,  event_status = event_hosp_3,  name = "hosp_3")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_4,  event_status = event_hosp_4,  name = "hosp_4")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_5,  event_status = event_hosp_5,  name = "hosp_5")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_6,  event_status = event_hosp_6,  name = "hosp_6")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_7,  event_status = event_hosp_7,  name = "hosp_7")
# hosp_8–11 use their own event_status
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_8,  event_status = event_hosp_8,  name = "hosp_8")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_9,  event_status = event_hosp_9,  name = "hosp_9")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_10, event_status = event_hosp_10, name = "hosp_10")
final_master <- cmp_variable(final_master, event_time = time_to_event_hosp_11, event_status = event_hosp_11, name = "hosp_11")

# Combined Immune-related AE (hosp_9) & Other (hosp_11): event = 1 if either
# occurred within 90 days, time = earliest of the two (90 if neither). All
# event/time variables already follow the convention "time = 90 if no event",
# so pmin() naturally yields the earliest event time when one or both occurred,
# and 90 when neither did.
final_master <- final_master %>%
  mutate(
    event_hosp_immune_other = pmax(event_hosp_9, event_hosp_11, na.rm = TRUE),
    time_to_event_hosp_immune_other = pmin(time_to_event_hosp_9,
                                           time_to_event_hosp_11, na.rm = TRUE)
  )
final_master <- cmp_variable(final_master,
  event_time = time_to_event_hosp_immune_other,
  event_status = event_hosp_immune_other,
  name = "hosp_immune_other")

# Overall hospitalization (any reason 1–11)
final_master <- cmp_variable(final_master,
  event_time = time_to_any_hosp, event_status = any_hosp, name = "any_hosp")

# Convert all cmp_status_* columns to factor (required by tidycmprsk)
final_master <- final_master %>%
  mutate(across(contains("cmp_status_"), as.factor))

# Drop record 911 — all RPDR pre-baseline labs are NA (see Section 6.1 note).
# This patient's nearest RPDR labs are ~189 days before baseline, well outside
# the 45-day pre_lab() window, so they are dropped from every Fine-Gray model
# via complete-case analysis anyway. Removing them up front keeps the cohort
# denominator consistent across tables and plots.
final_master <- final_master %>% filter(record_id != 911)

# Save final_master for use by analysis.qmd
saveRDS(final_master, "sherley_final_master.rds")

6 Missing Data — Table 1 Variables

Variables that are NA by design for a platinum subgroup (carbo_AUC/carbo_AUC_cat for cisplatin patients; cis_dose/cis_dose_cat for carboplatin patients) are excluded from this table.

# Variables shared by all patients (not platinum-type-specific)
shared_vars <- c(
  "age", "age_cat", "sex", "race", "ethnicity", "tumor_stage", "cancer_type_cat_7",
  "platin_group",
  "ici", "Bevacizumab", "Paclitaxel", "Pemetrexed", "Doxorubicin", "Etoposide",
  "Gemcitabine", "Pertuzumab", "Trastuzumab", "Docetaxel", "Administered_alone",
  "hiv_rpdr", "thyroid", "smoking", "steroids",
  "cci", "ecog_score", "bmi", "bmi_cat",
  "pre_HGB_45days", "pre_PLT_45days", "pre_ALB_45days", "pre_CRE_45days",
  "baseline_cys_c", "baseline_cre", "cr_cys",
  "baseline_cre_egfr", "baseline_cys_c_egfr",
  "ckd_epi_gfr_cre_cys", "ckd_epi_gfr_cre_cys_unindex",
  "cockcroft"
)

# Shared variables: missing across all patients
shared_missing <- final_master %>%
  dplyr::select(any_of(shared_vars)) %>%
  summarise(across(everything(), ~ sum(is.na(.)))) %>%
  tidyr::pivot_longer(everything(), names_to = "variable", values_to = "n_missing") %>%
  mutate(pct_missing = round(100 * n_missing / nrow(final_master), 1),
         note = "") %>%
  filter(n_missing > 0)

# Platinum-type-specific: check within their own subgroup
carbo_pts <- final_master %>% filter(platin_group == "Carboplatin")
cis_pts   <- final_master %>% filter(platin_group == "Cisplatin")

platin_missing <- tibble(
  variable   = c("carbo_AUC", "carbo_AUC_cat", "cis_dose", "cis_dose_cat"),
  n_missing  = c(sum(is.na(carbo_pts$carbo_AUC)),
                 sum(is.na(carbo_pts$carbo_AUC_cat)),
                 sum(is.na(cis_pts$cis_dose)),
                 sum(is.na(cis_pts$cis_dose_cat))),
  pct_missing = round(100 * n_missing /
                 c(nrow(carbo_pts), nrow(carbo_pts), nrow(cis_pts), nrow(cis_pts)), 1),
  note = c(paste0("of ", nrow(carbo_pts), " carbo pts"),
           paste0("of ", nrow(carbo_pts), " carbo pts"),
           paste0("of ", nrow(cis_pts), " cis pts"),
           paste0("of ", nrow(cis_pts), " cis pts"))
) %>% filter(n_missing > 0)

bind_rows(shared_missing, platin_missing) %>%
  arrange(desc(n_missing)) %>%
  knitr::kable(caption = "Missing data — Table 1 variables (non-zero only)")

Missing data — Table 1 variables (non-zero only)

variable	n_missing	pct_missing	note

6.1 Note: Record 911 — All RPDR Pre-Baseline Labs Missing

Record 911 (baseline_sample_date = 2025-08-19) has valid REDCap baseline labs (baseline_cre = 0.93, baseline_cys_c = 0.88, cockcroft = 83.4) but all RPDR pre-baseline labs are NA (pre_HGB_45days, pre_PLT_45days, pre_ALB_45days, pre_CRE_45days).

Root cause: This patient has 73 RPDR lab records, but the closest labs before baseline were drawn on 2025-02-11 — 189 days before baseline, well outside the 45-day pre_lab() window. No RPDR labs exist between 2025-07-05 and 2025-08-19.

Impact: This patient is dropped from Fine-Gray models via complete-case analysis whenever pre_HGB_45days, pre_PLT_45days, or pre_CRE_45days appears as a covariate (i.e., all models except base-only AKI models). Their cockcroft, ckd_epi_gfr_cre_cys_unindex, and egfr_ratio are valid.