Patient Characteristics

This notebook describes the 106,229 adult and pediatric hospitalized COVID-19 positive patients that were studied as part of the Consortium for Clinical Characterization of COVID-19 by EHR Neurology group. This analysis contains the results of patients from 21 different healthcare systems spanning 6 countries.

library(tidyverse)
library(gridGraphics)
library(gridExtra)
library(ggpubr)
library(DT)
library(kableExtra)
library(cowplot)
library(RColorBrewer)
library(glue)
library(egg)
source("R/plot_theme.R")
source("R/demo_tables.R")
source("R/utils.R")

Import Data from each Healthcare System

Each participating healthcare system ran the analysis locally on patient level data using our customized R package. The output of each local analysis consisted of aggregated counts and summary statistics (means, medians, hazard ratios, p-values, etc). Summary results from each healthcare system were then aggregated and are imported and then combined using the following code:

# read in files from results folder 
# this folder contains all of the local healthcare system level analyses
rdas <- list.files(
  path = "results",
  pattern = ".rda",
  full.names = TRUE
)


for (rda in rdas) {
  load(rda)
}

rm(rdas, rda)

# create a list of participating healthcare systems from our study tracking spreadsheet
site_google_url <- "https://docs.google.com/spreadsheets/d/1epcYNd_0jCUMktOHf8mz5v651zy1JALD6PgzobrGWDY/edit?usp=sharing"

# load site parameters
site_params <- googlesheets4::read_sheet(site_google_url, sheet = 1)
site_avails <- googlesheets4::read_sheet(site_google_url, sheet = 2)

# filter the list of sites who ran the analysis
sorted_sites <- site_avails %>%
  filter(!is.na(date_v4_received)) %>%
  pull(siteid) %>%
  paste("results", sep = "_")

# list sites without race
sites_wo_race <- site_params %>%
  filter(!include_race) %>%
  pull(siteid)

# combine all rda files with 'results' in name
results <- mget(ls(pattern = "results"))

## load in the pre-processed comorbidity and meta-data
load("processed/comorbidity_table_cnps.rda")
load("processed/neuro_pt_counts.rda") 

Pre-processing

The following code pre-processes the demographic and clinical course data in order to construct a Table 1.

1. Format primary demographic characteristics

# create list of hospitals for adult and pediatric analyses
adult_sites = sorted_sites[!sorted_sites %in% c("BCH_results", "GOSH_results")]

pediatric_sites = sorted_sites[!sorted_sites %in% c("VA1_results", "VA2_results", "VA3_results", "VA4_results", "VA5_results")]

demo_table_adult <- create_demo_tableone(sorted_sites = adult_sites, is_pediatric = FALSE)

demo_table_pediatric <- create_demo_tableone(sorted_sites = pediatric_sites, is_pediatric = TRUE)

demo_table_combine <- rbind(demo_table_adult, demo_table_pediatric)

2. Format clinical characteristics (i.e: continuous variables such as median time to discharge)

clinical_table_adult <- create_clinical_tableone(sorted_sites = adult_sites, is_pediatric = FALSE)

clinical_table_pediatric <- create_clinical_tableone(sorted_sites = pediatric_sites, is_pediatric = TRUE)

clinical_table_combine <- rbind(clinical_table_adult, clinical_table_pediatric)

3. Conduct additional pre-processing of clinical variables

clinical_table_adult_clean <- clean_clinical_tables(clinical_table = clinical_table_adult)

clinical_table_pediatric_clean <- clean_clinical_tables(clinical_table = clinical_table_pediatric)

clinical_table_clean_combine <- rbind(clinical_table_adult_clean, clinical_table_pediatric_clean)

4. Add Comorbidity data

Here we create separate lists of the top 4 comorbidities in the adult and pediatric populations

comorbidity_table_adults <- comorb_table_wide %>%
  filter(population == "Adult") %>% 
  arrange(desc(Comorb_Total)) %>% 
  slice(1:4) # return top comorbidities

comorbidity_table_pediatric <- comorb_table_wide %>%
  filter(population == "Pediatric") %>% 
  arrange(desc(Comorb_Total)) %>% 
  slice(1:4) # return top comorbidities

top_comorb_adults <- comorbidity_table_adults$Comorbidity
top_comorb_pediatrics <- comorbidity_table_pediatric$Comorbidity

Demographics

Table One - Combined

The below table contains the counts and summary stats of the entire cohort (adult & pediatric patients).

neuro_pt_counts_sum <- colSums(neuro_pt_counts[,-1]) %>% 
  data.frame() %>% 
  t() %>% 
  data.frame()

# specify the order of Table 1
row_order_adult <- c(
  "All Patients", "Female", "Male", "Unknown Sex",
  "0-2", "3-5", "6-11", "12-17", "18-25",
  "26-49", "50-69", "70-79", "80+", "Unknown Age",
  "American Indian", "Asian", "Black",
  "Hawaiian/Pacific Islander",
  "Hispanic/Latino", "White", "Other",
  top_comorb_adults, 
  "Median Elixhauser score  (sd) ",
  "Median pre admission cns  (sd) ",
  "Median pre admission pns  (sd) ",
  "Non-Severe", "Severe",
  "Median time to severe  (sd) ",
  "Alive", "Deceased",
  "Median time to death  (sd) ",
  "Discharged", "Not Discharged",
  "Median time to first discharge  (sd) ",
  "Not Readmitted", "Readmitted",
  "Median time to first readmission  (sd) ",
  "Median number of readmissions  (sd) "
  )


tableone_combine <- create_tableone(demo_table = demo_table_combine,
                                    clinical_table = clinical_table_clean_combine) %>%
  rbind(., comorb_table_wide %>%
          group_by(Comorbidity) %>%
          mutate(Comorb_Total = sum(Comorb_Total, na.rm = TRUE),
          None_sum = sum(None_Total, na.rm = TRUE),
          CNS_sum = sum(CNS_Total, na.rm = TRUE),
          PNS_sum = sum(PNS_Total, na.rm = TRUE),
          None_perc = round(None_sum/neuro_pt_counts_sum$None_n*100,1),
          Central_perc = round(CNS_sum/neuro_pt_counts_sum$CNS_n*100,1),
          Peripheral_perc = round(PNS_sum/neuro_pt_counts_sum$PNS_n*100,1),
          None = paste(None_sum, "(", None_perc, ")"),
          Central = paste(CNS_sum, "(", Central_perc, ")"),
          Peripheral = paste(PNS_sum, "(", Peripheral_perc, ")")) %>%
          distinct(Comorbidity, Comorb_Total, None, Central, Peripheral) %>%
          rename(`Table 1` = "Comorbidity",
          N = "Comorb_Total")) %>%
  slice(match(row_order_adult, `Table 1`))

kbl(tableone_combine %>%
      rename("No Neurological Condition (NNC)" = None)) %>%
      add_header_above(c(
      " ",
      "Total Patients" = 1,
      "Neurological Disease" = 3
      )) %>%
      kable_paper("striped", full_width = F) %>%
      pack_rows("Sex", 2, 4) %>%
      pack_rows("Age", 5, 13) %>%
      pack_rows("Race & Ethnicity", 14, 20) %>%
      pack_rows("Past Medical History", 21, 27) %>%
      pack_rows("Severity", 28, 30) %>%
      pack_rows("Survival", 31, 33) %>%
      pack_rows("Discharge", 34, 36) %>%
      pack_rows("Readmission", 37, 40)

	Total Patients	Neurological Disease
Table 1	N	No Neurological Condition (NNC)	Central	Peripheral
All Patients	106229	88340 (100%)	15101 (100%)	2788 (100%)
Sex
Female	37590	32399 (36.7%)	4247 (28.1%)	944 (33.9%)
Male	68638	55940 (63.3%)	10854 (71.9%)	1844 (66.1%)
Unknown Sex	1	1 (0%)	0 (0%)	0 (0%)
Age
0-2	769	747 (0.8%)	22 (0.1%)	0 (0%)
3-5	275	251 (0.3%)	24 (0.2%)	0 (0%)
6-11	403	368 (0.4%)	33 (0.2%)	2 (0.1%)
12-17	707	637 (0.7%)	60 (0.4%)	10 (0.4%)
18-25	2328	2138 (2.4%)	145 (1%)	45 (1.6%)
26-49	17799	16122 (18.3%)	1166 (7.7%)	511 (18.5%)
50-69	36885	31556 (35.7%)	4169 (27.7%)	1160 (41.9%)
70-79	24860	19672 (22.3%)	4508 (30%)	680 (24.6%)
80+	22094	16813 (19%)	4920 (32.7%)	361 (13%)
Race & Ethnicity
American Indian	350	295 (0.3%)	55 (0.4%)	0 (0%)
Asian	1694	1464 (1.7%)	205 (1.4%)	25 (0.9%)
Black	16815	13368 (15.1%)	2995 (19.9%)	452 (16.5%)
Hawaiian/Pacific Islander	297	255 (0.3%)	41 (0.3%)	1 (0%)
Hispanic/Latino	870	819 (0.9%)	29 (0.2%)	22 (0.8%)
White	41871	33360 (37.8%)	7401 (49.1%)	1110 (40.5%)
Other	44220	38750 (43.9%)	4336 (28.8%)	1134 (41.3%)
Past Medical History
Hypertension	35587	27323 ( 30.9 )	7277 ( 48.2 )	987 ( 35.4 )
Alcohol abuse	30130	23065 ( 26.1 )	6194 ( 41 )	871 ( 31.2 )
Drug abuse	26596	20287 ( 23 )	5528 ( 36.6 )	781 ( 28 )
Diabetes	21969	16905 ( 19.1 )	4429 ( 29.3 )	635 ( 22.8 )
Median Elixhauser score (sd)	0.3 (0.7)	0.2 (0.5)	1.4 (2)	0.6 (1.7)
Median pre admission cns (sd)	0 (0)	0 (0)	0 (0)	0 (0)
Median pre admission pns (sd)	0 (0)	0 (0)	0 (0)	0 (0.1)
Severity
Non-Severe	66252	57771 (65.4%)	7007 (46.4%)	1474 (52.7%)
Severe	39985	30576 (34.6%)	8084 (53.6%)	1325 (47.3%)
Median time to severe (sd)	0.6 (1)	0.8 (1.3)	0.4 (0.7)	0.8 (1.1)
Survival
Alive	87376	74389 (84.2%)	10392 (68.8%)	2595 (93.1%)
Deceased	18849	13953 (15.8%)	4705 (31.2%)	191 (6.9%)
Median time to death (sd)	15.3 (6.4)	15.9 (5.8)	22.4 (32.5)	36.9 (41.4)
Discharge
Discharged	103165	85639 (96.9%)	14801 (98%)	2725 (97.7%)
Not Discharged	3063	2701 (3.1%)	297 (2%)	65 (2.3%)
Median time to first discharge (sd)	8.4 (4.2)	6.5 (3.6)	12.8 (6.1)	14.5 (18.3)
Readmission
Not Readmitted	91646	76877 (87%)	12435 (82.4%)	2334 (83.4%)
Readmitted	14569	11456 (13%)	2650 (17.6%)	463 (16.6%)
Median time to first readmission (sd)	28.1 (20.3)	26.4 (17.6)	31.9 (33.7)	38.1 (28.4)
Median number of readmissions (sd)	0.1 (0.3)	0 (0)	0.1 (0.3)	0.1 (0.3)

write.csv(tableone_combine, 'tables/table1_combined.csv', row.names = FALSE)

Table One - Adults

tableone_adult <- create_tableone(demo_table = demo_table_adult,
                                  clinical_table = clinical_table_adult_clean) %>%
                                  rbind(., comorbidity_table_adults %>%
                                          select(Comorbidity, Comorb_Total, `NNC_N_%`, `CNS_N_%`, `PNS_N_%`) %>%
                                          rename(`Table 1` = "Comorbidity",
                                                 N = "Comorb_Total",
                                                 None = "NNC_N_%",
                                                 Central = "CNS_N_%",
                                                 Peripheral = "PNS_N_%")) %>% 
  slice(match(row_order_adult, `Table 1`))

kbl(tableone_adult %>%
    rename("No Neurological Condition (NNC)" = None)) %>%
  add_header_above(c(
  " ",
  "Total Patients" = 1,
  "Neurological Disease" = 3
  )) %>%
  kable_paper("striped", full_width = F) %>%
  pack_rows("Sex", 2, 4) %>%
  pack_rows("Age", 5, 9) %>%
  pack_rows("Race & Ethnicity", 10, 16) %>%
  pack_rows("Past Medical History", 17, 23) %>%
  pack_rows("Severity", 24, 26) %>%
  pack_rows("Survival", 27, 29) %>%
  pack_rows("Discharge", 30, 32) %>%
  pack_rows("Readmission", 33, 36)

	Total Patients	Neurological Disease
Table 1	N	No Neurological Condition (NNC)	Central	Peripheral
All Patients	104031	86321 (100%)	14938 (100%)	2772 (100%)
Sex
Female	36564	31445 (36.4%)	4185 (28%)	934 (33.7%)
Male	67466	54875 (63.6%)	10753 (72%)	1838 (66.3%)
Unknown Sex	1	1 (0%)	0 (0%)	0 (0%)
Age
18-25	2328	2138 (2.5%)	145 (1%)	45 (1.6%)
26-49	17799	16122 (18.7%)	1166 (7.8%)	511 (18.5%)
50-69	36885	31556 (36.6%)	4169 (28%)	1160 (42.1%)
70-79	24860	19672 (22.8%)	4508 (30.2%)	680 (24.7%)
80+	22094	16813 (19.5%)	4920 (33%)	361 (13.1%)
Race & Ethnicity
American Indian	349	294 (0.3%)	55 (0.4%)	0 (0%)
Asian	1633	1412 (1.6%)	196 (1.3%)	25 (0.9%)
Black	16631	13202 (15.3%)	2979 (20%)	450 (16.5%)
Hawaiian/Pacific Islander	296	254 (0.3%)	41 (0.3%)	1 (0%)
Hispanic/Latino	850	799 (0.9%)	29 (0.2%)	22 (0.8%)
White	41206	32752 (38%)	7354 (49.3%)	1100 (40.3%)
Other	42971	37587 (43.6%)	4254 (28.5%)	1130 (41.4%)
Past Medical History
Hypertension	35566	27302 ( 31.6 %)	7277 ( 48.7 %)	987 ( 35.6 %)
Alcohol abuse	29918	22879 ( 26.5 %)	6169 ( 41.3 %)	870 ( 31.4 %)
Drug abuse	26389	20097 ( 23.3 %)	5512 ( 36.9 %)	780 ( 28.1 %)
Diabetes	21952	16888 ( 19.6 %)	4429 ( 29.6 %)	635 ( 22.9 %)
Median Elixhauser score (sd)	0.3 (0.7)	0.2 (0.5)	1.4 (2)	0.6 (1.7)
Median pre admission cns (sd)	0 (0)	0 (0)	0 (0)	0 (0)
Median pre admission pns (sd)	0 (0)	0 (0)	0 (0)	0 (0.1)
Severity
Non-Severe	64578	56181 (65.1%)	6935 (46.4%)	1462 (52.6%)
Severe	39468	30140 (34.9%)	8008 (53.6%)	1320 (47.4%)
Median time to severe (sd)	0.6 (1)	0.8 (1.3)	0.4 (0.7)	0.8 (1.1)
Survival
Alive	85212	72388 (83.9%)	10247 (68.6%)	2577 (93.1%)
Deceased	18820	13933 (16.1%)	4696 (31.4%)	191 (6.9%)
Median time to death (sd)	15.3 (6.4)	15.9 (5.8)	22.4 (32.5)	36.9 (41.4)
Discharge
Discharged	101032	83675 (96.9%)	14650 (98%)	2707 (97.7%)
Not Discharged	3005	2645 (3.1%)	295 (2%)	65 (2.3%)
Median time to first discharge (sd)	8.4 (4.2)	6.5 (3.6)	12.8 (6.1)	14.5 (18.3)
Readmission
Not Readmitted	89802	75170 (87.1%)	12312 (82.4%)	2320 (83.5%)
Readmitted	14237	11151 (12.9%)	2626 (17.6%)	460 (16.5%)
Median time to first readmission (sd)	28.1 (20.3)	26.4 (17.6)	31.9 (33.7)	38.1 (28.4)
Median number of readmissions (sd)	0.1 (0.3)	0 (0)	0.1 (0.3)	0.1 (0.3)

write.csv(tableone_adult, 'tables/table1_adults.csv', row.names = FALSE)

Table One - Pediatrics

# specify the order of Table 1
row_order_pediatric <- c(
  "All Patients", "Female", "Male", "Unknown Sex",
  "0-2", "3-5", "6-11", "12-17", "18-25",
  "26-49", "50-69", "70-79", "80+", "Unknown Age",
  "American Indian", "Asian", "Black",
  "Hawaiian/Pacific Islander",
  "Hispanic/Latino", "White", "Other",
  top_comorb_pediatrics, 
  "Median Elixhauser score  (sd) ",
  "Median pre admission cns  (sd) ",
  "Median pre admission pns  (sd) ",
  "Non-Severe", "Severe",
  "Median time to severe  (sd) ",
  "Alive", "Deceased",
  "Median time to death  (sd) ",
  "Discharged", "Not Discharged",
  "Median time to first discharge  (sd) ",
  "Not Readmitted", "Readmitted",
  "Median time to first readmission  (sd) ",
  "Median number of readmissions  (sd) "
  )

tableone_pediatric <- create_tableone(demo_table = demo_table_pediatric,
                                      clinical_table = clinical_table_pediatric_clean)  %>%
  rbind(., comorbidity_table_pediatric %>%
          select(Comorbidity, Comorb_Total, `NNC_N_%`, `CNS_N_%`, `PNS_N_%`) %>% 
          rename(`Table 1` = "Comorbidity",
                  N = "Comorb_Total",
                  None = "NNC_N_%",
                  Central = "CNS_N_%",
                  Peripheral = "PNS_N_%")) %>% 
  slice(match(row_order_pediatric, `Table 1`))

kbl(tableone_pediatric %>%
      rename("No Neurological Condition (NNC)" = None)) %>%
  add_header_above(c(
  " ",
  "Total Patients" = 1,
  "Neurological Disease" = 3
  )) %>%
  kable_paper("striped", full_width = F) %>%
  pack_rows("Sex", 2, 4) %>%
  pack_rows("Age", 5, 8) %>%
  pack_rows("Race & Ethnicity", 9, 15) %>%
  pack_rows("Past Medical History", 16, 22) %>%
  pack_rows("Severity", 23, 25) %>%
  pack_rows("Survival", 26, 28) %>%
  pack_rows("Discharge", 29, 31) %>%
  pack_rows("Readmission", 32, 35)

	Total Patients	Neurological Disease
Table 1	N	No Neurological Condition (NNC)	Central	Peripheral
All Patients	2198	2019 (100%)	163 (100%)	16 (100%)
Sex
Female	1026	954 (47.3%)	62 (38%)	10 (62.5%)
Male	1172	1065 (52.7%)	101 (62%)	6 (37.5%)
Unknown Sex	0	0 (0%)	0 (0%)	0 (0%)
Age
0-2	769	747 (37.3%)	22 (15.8%)	0 (0%)
3-5	275	251 (12.5%)	24 (17.3%)	0 (0%)
6-11	403	368 (18.4%)	33 (23.7%)	2 (16.7%)
12-17	707	637 (31.8%)	60 (43.2%)	10 (83.3%)
Race & Ethnicity
American Indian	1	1 (0%)	0 (0%)	0 (0%)
Asian	61	52 (2.6%)	9 (5.8%)	0 (0%)
Black	184	166 (8.3%)	16 (10.4%)	2 (12.5%)
Hawaiian/Pacific Islander	1	1 (0%)	0 (0%)	0 (0%)
Hispanic/Latino	20	20 (1%)	0 (0%)	0 (0%)
White	665	608 (30.2%)	47 (30.5%)	10 (62.5%)
Other	1249	1163 (57.8%)	82 (53.2%)	4 (25%)
Past Medical History
Alcohol abuse	212	186 ( 9.2 %)	25 ( 15.3 %)	1 ( 6.2 %)
Drug abuse	207	190 ( 9.4 %)	16 ( 9.8 %)	1 ( 6.2 %)
Weight loss	123	109 ( 5.4 %)	13 ( 8 %)	1 ( 6.2 %)
Cardiac arrhythmias	109	94 ( 4.7 %)	12 ( 7.4 %)	3 ( 18.8 %)
Median Elixhauser score (sd)	2.2 (4.5)	0 (0)	4.4 (9.2)	2.5 (3.6)
Median pre admission cns (sd)	0.2 (0.7)	0.1 (0.2)	1.9 (3.9)	0 (0)
Median pre admission pns (sd)	0 (0)	0 (0)	0 (0)	0.2 (0.6)
Severity
Non-Severe	1674	1590 (78.5%)	72 (48.6%)	12 (70.6%)
Severe	517	436 (21.5%)	76 (51.4%)	5 (29.4%)
Median time to severe (sd)	2.4 (5.3)	3.8 (9.5)	10.8 (33.1)	0.9 (2.5)
Survival
Alive	2164	2001 (99%)	145 (94.2%)	18 (100%)
Deceased	29	20 (1%)	9 (5.8%)	0 (0%)
Median time to death (sd)	12.6 (36.3)	53.5 (76.3)	5.8 (10.2)	95 (268.7)
Discharge
Discharged	2133	1964 (97.2%)	151 (98.7%)	18 (100%)
Not Discharged	58	56 (2.8%)	2 (1.3%)	0 (0%)
Median time to first discharge (sd)	5.1 (4.2)	3.8 (4.2)	5.8 (2.7)	5.9 (4.5)
Readmission
Not Readmitted	1844	1707 (84.8%)	123 (83.7%)	14 (82.4%)
Readmitted	332	305 (15.2%)	24 (16.3%)	3 (17.6%)
Median time to first readmission (sd)	20 (20.2)	20.1 (19)	75.3 (86)	14.4 (22.8)
Median number of readmissions (sd)	0.1 (0.3)	0 (0)	0.2 (0.4)	0.4 (0.5)

write.csv(tableone_pediatric, 'tables/table1_pediatric.csv', row.names = FALSE)

Demographic & Clinical Course Analysis

Next we will evaluate characteristics among patients stratified by neurological status: 1) patients with no neurological condition (NNC), 2) patients with a central nervous system condition (CNS), and 3) patients with a peripheral nervous system condition (PNS).

Categorical variables will be evaluated with chi-square tests, while continuous variables will be evaluated with a non-parametric anova (Kruskal-Wallis) test.

1. Conduct chi-squared tests for categorical demographic variables

compute_chi_square <- function(tableone, demo_table) {
  
  N = tableone %>% select(N) %>% head(1) %>% as.numeric()

  # sum of all neuro condition groups
  cns_n = as.numeric(gsub( " .*$", "", tableone[1,"Central"]))
  pns_n = as.numeric(gsub( " .*$", "", tableone[1,"Peripheral"]))
  none_n = as.numeric(gsub( " .*$", "", tableone[1,"None"]))
  
  # sum up the total counts of each demographic variable across healthcare systems
  tableOne_sums <- demo_table %>%
    group_by(variable, Demo_var, Demo_var_i) %>%
    summarise(across(
      starts_with("n_var"),
      function(x) sum(x, na.rm = TRUE)
    ), .groups = "drop")
  
  # create list of demographic/clinical variables
  vars = tableOne_sums$variable
  
  # for binary variables, we should group by category 
  # Thus, only run the analysis for sex.male rather than both sex.female AND sex.male which would be redundant
  # we will also remove `age_group.unknown` since their is only 1 patient here
  exclude <- c("readmitted.false", "sex.female", "sex.other", "survival.deceased", "severity.non-severe",
               "covid_discharged.discharged", "age_group.Unknown")
  vars <- vars[!vars %in% exclude]
  
  # create empty list for chi.square test results
  chi_result_list = list()
  
  # for each variable, we will run the chi.square test
  for(i in vars) {
    test = calc_chisq(i, tableOne_sums, none_n, pns_n, cns_n)
    X2 <- round(test$statistic, 4)
    p_value <- test$p.value
    df <- test$parameter
    Variable = paste(i)
    
    chi_results <- cbind(Variable, X2, p_value, df) %>% 
      data.frame() %>% 
      mutate(p_value = as.numeric(p_value))
    
    rownames(chi_results) <- NULL
    
    chi_result_list[[i]] <- chi_results
  }
  
  # save list of chi.square test results
  chisq_results <- bind_rows(chi_result_list)
  
  return(chisq_results)
  
  
  }

Combined Adult & Pediatric Chi-Square Results

chisq_combine <- compute_chi_square(tableone = tableone_combine %>%
                                      filter(!`Table 1` == 'Discharged',
                                             !`Table 1` == 'Not Discharged'),
                                    demo_table = demo_table_combine %>%
                                      filter(!variable == 'covid_discharged.discharged',
                                             !variable == 'covid_discharged.not discharged'))

datatable(chisq_combine %>% arrange(p_value))

Adult Chi-Square Results

chisq_adult <- compute_chi_square(tableone = tableone_adult %>%
                                      filter(!`Table 1` == 'Discharged',
                                             !`Table 1` == 'Not Discharged'),
                                    demo_table = demo_table_adult %>%
                                      filter(!variable == 'covid_discharged.discharged',
                                             !variable == 'covid_discharged.not discharged'))

datatable(chisq_adult %>% arrange(p_value))

Pediatric Chi-Square Results

chisq_pediatric <- compute_chi_square(tableone = tableone_pediatric %>%
                                      filter(!`Table 1` == 'Discharged',
                                             !`Table 1` == 'Not Discharged'),
                                    demo_table = demo_table_pediatric %>%
                                      filter(!variable == 'covid_discharged.discharged',
                                             !variable == 'covid_discharged.not discharged'))

datatable(chisq_pediatric %>% arrange(p_value))

2. Conduct non-parametric anova (Kruskal-Wallis) tests for continuous clinical variables

compute_kruskal <- function(clinical_table) {

  # create empty list to save processed variables
  processed_list_figs <- list()
  
  # create a list of clinical variables to further pre-process
  vars_to_process <- unique(clinical_table$name)
  
  # for each variable, we will remove the [min, max] as before
  for (i in vars_to_process) {
    mod_table <- clinical_table %>%
      rename("var" = name) %>%
      filter(var == i) %>%
      mutate(
        site = toupper(site),
        None = sub("(\\(.*|\\[.*)", "", None),
        Peripheral = sub("(\\(.*|\\[.*)", "", Peripheral),
        Central = sub("(\\(.*|\\[.*)", "", Central)
      ) %>%
      mutate(across(None:Central, as.numeric))
  
    processed_list_figs[[i]] <- mod_table
  }
  
  clinical_table_anova <- bind_rows(processed_list_figs)
  
  # format the continuous variables
  cont_vars = clinical_table_anova %>%
    mutate(
      # create a variable without mean/median prefix
      grouped_var = gsub("Mean |Median | \\[Min, Max\\]| \\(SD\\)", "", var)
    ) %>%
    pivot_longer(
      cols = None:Central,
      names_to = "type"
    ) %>%
    mutate(type = factor(type, levels = c("None", "Central", "Peripheral"))) %>% 
    # select variables to analyze
    filter(grouped_var %in% c("Elixhauser score",
                              "pre admission cns", 
                              "pre admission pns", 
                              "time to death", 
                              "time to first discharge", 
                              "time to severe",
                              "number of readmissions",
                              "time to first readmission")) %>% 
    distinct()
  
  # create function to conduct anova
  cont_var_results <- function(variable) {
    
    df <- cont_vars %>% 
    filter(grouped_var == paste(variable))
    
    # kruskal.test calculates the kruskal-Wallis H-statistic 
    # does not assume normality between groups
    # also called the one-way ANOVA on ranks
    test = kruskal.test(df$value ~ df$type)
    
  }
  
  # create empty list for Kruskal-Wallis results
  cont_vars_list = list()
  
  # create list of unique variables
  outcome_cont_vars = unique(cont_vars$grouped_var)
  
  for(i in outcome_cont_vars) {
    
    test = cont_var_results(i)
    X2 <- round(test$statistic, 4)
    p_value <- test$p.value
    df <- test$parameter
    Variable = paste(i)
    
    kw <- cbind(Variable, X2, p_value, df) %>% 
      data.frame() 
    
    rownames(kw) <- NULL
    
    cont_vars_list[[i]] <- kw
  
    
  }
  
  anova_results <- bind_rows(cont_vars_list)
  
  }

Combined Adult & Pediatric Kruskal-Wallis Results

kw_combine <- compute_kruskal(clinical_table = clinical_table_combine)
datatable(kw_combine %>% arrange(p_value))

Adult Kruskal-Wallis Results

Note: median CNS is 0 for all sites (reason for the NA values)

kw_adult <- compute_kruskal(clinical_table = clinical_table_adult)
datatable(kw_adult %>% arrange(p_value))

Pediatric Kruskal-Wallis Results

kw_pediatric <- compute_kruskal(clinical_table = clinical_table_pediatric)
datatable(kw_pediatric %>% arrange(p_value))

3. Combine chi-square and Kruskal-Wallis results

We will evaluate significance when controlling for False Discovery Rate (FDR)

tableone_stats <- function(chisq_results, anova_results, population) {

  demo_stats <- rbind(chisq_results %>% select(Variable, p_value),
                      anova_results %>% select(Variable, p_value))
  
  # adjust p-values with FDR
  demo_stats$adj_p_value <- p.adjust(demo_stats$p_value, method = "fdr")
  
  # round & format p-value 
  demo_stats <- demo_stats %>% 
    mutate(p_value = as.numeric(p_value),
           pvalue = if_else(p_value < 0.001, "< 0.001", paste(round(p_value, 3))),
           adj.p_value = if_else(adj_p_value < 0.001, "< 0.001", paste(round(adj_p_value, 3))))
  
  # tidy up the demographics stat table
  demo_stats_tidy <- demo_stats %>% 
    rename(`Unadjusted P-value (raw)` = p_value,
           `FDR Adjusted P-value (raw)` = adj_p_value,
           `Unadjusted P-value (tidy)` = pvalue,
           `FDR Adjusted P-value (tidy)` = adj.p_value) %>% 
    select(Variable, `Unadjusted P-value (raw)`, `FDR Adjusted P-value (raw)`, `Unadjusted P-value (tidy)`, `FDR Adjusted P-value (tidy)`)

  write.csv(demo_stats_tidy, paste0("tables/Table1_pvals_", population, ".csv"), row.names = FALSE)
  
  return(demo_stats_tidy)
  
}

4. Incorporate Comorbidities

We will also add the top 4 comorbidities for each population to our table one. For the combined adult and pediatric table, we will use the top 4 comorbidities of the adult population.

# define matrix for each comorbidity
comorb_matrix <- function(comorbidity_table, comorbidity, population) {

    if(population=='Combined') {
      comorbidity_table = comorbidity_table
    } else if (population=="Adult") {
      comorbidity_table = comorbidity_table %>% 
        filter(population == "Adult")
    } else if (population=="Pediatric") {
      comorbidity_table = comorbidity_table %>% 
        filter(population == "Pediatric")
    } else {
      print('incorrect population specification')
    }
  
  mat_comorb <- comorbidity_table %>%
    filter(Comorbidity == paste(comorbidity)) %>%
    group_by(Comorbidity) %>% 
    mutate(None_comorb = sum(None_Total, na.rm = TRUE),
    CNS_comorb = sum(CNS_Total, na.rm = TRUE),
    PNS_comorb = sum(PNS_Total, na.rm = TRUE)) %>%
    ungroup() %>%
    distinct(None_comorb, CNS_comorb, PNS_comorb) %>%
    mutate(none_dif = neuro_pt_counts_sum$None_n - None_comorb,
    pns_dif = neuro_pt_counts_sum$PNS_n - PNS_comorb,
    cns_dif = neuro_pt_counts_sum$CNS_n - CNS_comorb) %>%
    data.frame() %>%
    #as.integer() %>%
    matrix(nrow = 2, ncol = 3, byrow = TRUE)

  comorb_results <- chisq.test(unlist(mat_comorb))
  
  results_df <- data.frame(Variable = paste(comorbidity),
                                X2 = round(comorb_results$statistic, 4),
                                p_value = comorb_results$p.value,
                                df = comorb_results$parameter)
  row.names(results_df) <- NULL
  
  return(results_df)
}

## combined comorbidities
comorbidity_combined_chi_list <- list()

for(i in top_comorb_adults) {
  
  comorb_chi_results <- comorb_matrix(comorbidity_table = comorb_table_wide, i, population = "Combined")
  comorbidity_combined_chi_list[[i]] <- comorb_chi_results
}

comorbidity_combined_chi <- comorbidity_combined_chi_list %>%
  bind_rows() 

chisq_combine_comorb <- chisq_combine %>%
  mutate(X2 = as.numeric(X2),
  df = as.numeric(df)) %>%
  bind_rows(comorbidity_combined_chi) 



## adult comorbidities
comorbidity_adult_chi_list <- list()

for(i in top_comorb_adults) {
  
  comorb_chi_results <- comorb_matrix(comorbidity_table = comorb_table_wide, i, population = "Adult")
  comorbidity_adult_chi_list[[i]] <- comorb_chi_results
}

comorbidity_adult_chi <- comorbidity_adult_chi_list %>%
  bind_rows() 

chisq_adult_comorb <- chisq_adult %>%
  mutate(X2 = as.numeric(X2),
  df = as.numeric(df)) %>%
  bind_rows(comorbidity_adult_chi) 

## pediatric comorbidities
comorbidity_pediatric_chi_list <- list()

for(i in top_comorb_pediatrics) {
  
  comorb_chi_results <- comorb_matrix(comorbidity_table = comorb_table_wide, i, population = "Pediatric")
  comorbidity_pediatric_chi_list[[i]] <- comorb_chi_results
}

comorbidity_pediatric_chi <- comorbidity_pediatric_chi_list %>%
  bind_rows() 

chisq_pediatric_comorb <- chisq_pediatric %>%
  mutate(X2 = as.numeric(X2),
  df = as.numeric(df)) %>%
  bind_rows(comorbidity_pediatric_chi) 

Table One - Combined Analysis

combine_tableone_stats <- tableone_stats(chisq_results = chisq_combine_comorb,
                                         anova_results = kw_combine,
                                         population = "combined") 
  

datatable(combine_tableone_stats)

Table One - Adult Analysis

adult_tableone_stats <- tableone_stats(chisq_results = chisq_adult_comorb,
                                         anova_results = kw_adult,
                                         population = "adult") 
  

datatable(adult_tableone_stats)

Table One - Pediatric Analysis

pediatric_tableone_stats <- tableone_stats(chisq_results = chisq_pediatric_comorb,
                                         anova_results = kw_pediatric,
                                         population = "pediatric") 
  

datatable(pediatric_tableone_stats)

Neurological Diagnosis Analysis

Next, we will evaluate the diagnoses captured with the CNS and PNS groups. Specifically, we will examine the frequency of CNS and PNS diagnoses by patient age group and by clinical outcomes (mortality and COVID-19 severity).

1. Format neurological icd code data

# import list of ICD-9/ICD-10 codes
neuro_icds_10 <-
  readxl::read_excel("public-data/2020-02-22_neuro-icd10_CNSvPNS.xlsx", sheet = 2) %>% 
  rename(
    "icd" = `ICD-10`,
    "pns_cns" = `Nervous system Involvement (1=central, 2=peripheral, 3=irrelevant)`,
    "type" = `ICD-10_type (1=first three alphanumeric code only, 2=digits after decimal point)`) %>%
  filter(type == 1) %>% 
  mutate(pns_cns = as.factor(pns_cns) %>% fct_recode(
    Central = "1",
    Peripheral = "2"
  )) %>%
  distinct(icd, `Neurological Disease Category`, pns_cns, `ICD-10 Description Revised`) %>%
  rename("description" = `ICD-10 Description Revised`) %>%
  mutate(concept_type = "icd-10")

neuro_icds_9 <- read.csv("public-data/icd9_tab_CNSvPNS.csv") %>%
  rename(
    "Neurological Disease Category" = "Neurological.Disease.Category",
    "pns_cns" = `Nervous.system.Involvement..1.central..2.peripheral.`,
    "icd_description" = `icd9_desc_revised`
  ) %>%
  mutate(
    pns_cns = as.factor(pns_cns) %>% fct_recode(
      Central = "1",
      Peripheral = "2"
    ),
    concept_type = "DIAG-ICD9"
  ) %>%
  select(icd, `Neurological Disease Category`, pns_cns, icd_description) %>%
  rename("description" = icd_description) %>%
  mutate(concept_type = "icd-9")

# bind list of ICD-9/ICD-10 codes
icds <- rbind(neuro_icds_10, neuro_icds_9)

2. Calculate total number of neurological diagnoses for each adults & pediatrics

compute_neuro_counts <- function(sorted_sites, sample_size, is_pediatric = FALSE) {
  
    # create empty list to save processed diagnostic data
    diag_table_list <- list()
  
    if(is_pediatric==FALSE) {
      population = "adults"
      } else {
      population = "pediatrics"
      }
  
  # for each healthcare system, we will calculate the total number of patients with each ICD code
  for (i in sorted_sites) {
    tmp <- get(i)
    tmp_diag <- tmp[[c(
      "first_hosp_results",
      "icd_tables",
      paste0("icd_tables_", population),
      "demo_table"
    )]]
    tmp_diag$site <- tmp[["site"]]
    diag_table_list[[i]] <- tmp_diag %>%
      filter(variable == "sex.female" | variable == "sex.male" | variable == "sex.other") %>%
      select(site, contains("n_var"))
  }
  
  diag_table_wide <- bind_rows(diag_table_list)
  
  # we will count the total number and percent across healthcare systems
  diag_counts <- colSums(Filter(is.numeric, diag_table_wide), na.rm = TRUE) %>% 
    data.frame() %>% 
    mutate(perc = round(./sample_size * 100, 1))
  
  # format diagnostic table
  diag_counts <- tibble::rownames_to_column(diag_counts, "icd") %>%
    mutate(icd = gsub("n_var_", "", icd)) %>%
    rename(`Number of Patients` = ".") %>%
    full_join(., icds, by = "icd") %>%
    dplyr::arrange(desc(`Number of Patients`)) %>% 
    mutate(`Number of Patients (% of Cohort)` = paste(`Number of Patients`, "(", perc, ")", sep = " ")) %>% 
    select(pns_cns, `Neurological Disease Category`, concept_type, 
           icd, description, `Number of Patients (% of Cohort)`) %>% 
      rename(Code = "icd",
           `Nervous System` = "pns_cns",
           Description = "description",
           `ICD Version` = "concept_type") %>% 
    mutate(`Number of Patients (% of Cohort)` = if_else(`Number of Patients (% of Cohort)` == "NA ( NA )", "0 ( 0 )", `Number of Patients (% of Cohort)`),
           Code = if_else(Code == "NN", "No Neurological Disease (NNC)", Code))
  
  return(diag_counts)
  
  }

# create vectors containing patient sample sizes
n_adult <- as.numeric(tableone_adult$N)[1]
n_pediatric <- as.numeric(tableone_pediatric$N)[1]

# calculate the total neurological patient counts 
neuro_counts_adult <- compute_neuro_counts(sorted_sites = adult_sites,
                                           sample_size = n_adult, 
                                           is_pediatric = FALSE)

neuro_counts_pediatric <- compute_neuro_counts(sorted_sites = pediatric_sites,
                                           sample_size = n_pediatric, 
                                           is_pediatric = TRUE)

neuro_counts_combined <- neuro_counts_adult %>% 
  rename(`Number of Adult Patients (% of Cohort)` = `Number of Patients (% of Cohort)`) %>% 
  left_join(., neuro_counts_pediatric %>% 
              select(`ICD Version`, Code, `Number of Patients (% of Cohort)`) %>% 
              rename("Number of Pediatric Patients (% of Cohort`)" = `Number of Patients (% of Cohort)`))

write.csv(neuro_counts_combined, paste0("tables/Supp.Table2_diagnoses", ".csv"), row.names = FALSE)

Number (%) of Adult Patients with each Neurological Diagnosis

datatable(neuro_counts_adult) 

Number (%) of Pediatric Patients with each Neurological Diagnosis

datatable(neuro_counts_pediatric) 

Evaluate Diagnoses by Age & Outcome

compute_stratified_neuro_counts <- function(tableone, sorted_sites, is_pediatric = FALSE) {
  
  # create empty list to save processed diagnostic data
    diag_table_list <- list()
  
    if(is_pediatric==FALSE) {
      population = "adults"
      } else {
      population = "pediatrics"
      }
  
  # for each healthcare system, we will calculate the total number of patients with each ICD code
  for (i in sorted_sites) {
    #print(i)
    tmp <- get(i)
    tmp_diag <- tmp[[c(
      "first_hosp_results",
      "icd_tables",
      paste0("icd_tables_", population),
      "demo_table"
    )]]
    tmp_diag$site <- tmp[["site"]]
    try(
    diag_table_list[[i]] <- tmp_diag %>%
      select(site, variable, starts_with("n_var")) %>% 
      pivot_longer(cols = starts_with("n_var"), names_to = "code", values_to = "freq") %>% 
      ungroup()
    )
  }

  diag_table_wide_all <- bind_rows(diag_table_list) %>% 
      group_by(variable, code) %>% 
    # total number of patients who had each code for each variable across sites
      mutate(total_n_var  = sum(freq, na.rm = TRUE)) %>% 
    distinct(variable, code, total_n_var) %>% 
    mutate(code = gsub("n_var_", "", code),
           variable = gsub("Severity.", "", variable),
           variable = gsub("Survival.", "", variable),
           variable = gsub("readmitted", "", variable),
           variable = gsub("age_group", "", variable),
           variable = gsub("covid_discharged", "", variable),
           variable = if_else(variable == ".TRUE", "Readmitted", variable)) %>% 
    left_join(., icds %>% rename(code = "icd"), by = "code") %>% 
    ungroup() %>% 
    filter(!concept_type == "icd-9")


  # refactor neuro status
  diag_table_wide_all$pns_cns <- factor(diag_table_wide_all$pns_cns,
                                        levels=c("Central", "Peripheral"),
                                        labels=c("CNS", "PNS"))
  
  
  return(diag_table_wide_all)
  
  }

stratified_neuro_counts_adult <- compute_stratified_neuro_counts(sorted_sites = adult_sites,
                                                                 is_pediatric = FALSE,
                                                                 tableone = tableone_adult)

stratified_neuro_counts_pediatric <- compute_stratified_neuro_counts(sorted_sites = pediatric_sites,
                                                                 is_pediatric = TRUE,
                                                                 tableone = tableone_pediatric)

compute_stratified_neuro_counts_age <- function(stratified_neuro_counts, tableone) {
  
  stratified_neuro_counts$variable <- factor(stratified_neuro_counts$variable, 
                                     levels=c(".00to02", ".03to05", ".06to11", ".12to17",
                                              ".18to25", ".26to49", ".50to69", ".70to79", ".80plus"), 
                                     labels=c("0-2 Years", "3-5 Years", "6-11 Years", "12-17 Years",
                                              "18-25 Years", "26-49 Years", "50-69 Years", "70-79 Years", "80+ Years"))
  
  # total counts
  n_0_2 <- get_table_n(tableone, var = "0-2")
  n_3_5 = get_table_n(tableone, var = "3-5")
  n_6_11 = get_table_n(tableone, var = "6-11")
  n_12_17 = get_table_n(tableone, var = "12-17")
  n_18_25 = get_table_n(tableone, var = "18-25")
  n_26_49 = get_table_n(tableone, var = "26-49")
  n_50_69 = get_table_n(tableone, var = "50-69")
  n_70_79 = get_table_n(tableone, var = "70-79")
  n_80 = get_table_n(tableone, var = "80+")
  
  age_df <- data.frame("variable" = c("0-2 Years",
                                    "3-5 Years", 
                                    "6-11 Years", 
                                    "12-17 Years",
                                    "18-25 Years", 
                                    "26-49 Years", 
                                    "50-69 Years", 
                                    "70-79 Years", 
                                    "80+ Years"),
                     "n_var" = c(n_0_2,
                               n_3_5,
                               n_6_11,
                               n_12_17,
                               n_18_25,
                               n_26_49,
                               n_50_69,
                               n_70_79,
                               n_80))

  
  stratified_neuro_counts <- stratified_neuro_counts %>% 
    filter(grepl("Years", variable),
                !code == "NN", 
                !total_n_var == 0) %>% 
         # total_n would be calculating total codes across each age group which is not true reflection of patient counts b/c patients can have more than one code
         #mutate(total_n = sum(total_n_var)) %>% 
         ungroup() %>% 
         left_join(., age_df, by = "variable") %>% 
         mutate(n_var = as.numeric(n_var),
                perc = as.character(round(total_n_var/n_var*100,1)),
                perc = if_else(perc < 0.1, '<0.1', perc),
                perc = paste0(perc, "%")) 
  
  return(stratified_neuro_counts)
  
  
}

stratified_neuro_counts_age_adult <- compute_stratified_neuro_counts_age(stratified_neuro_counts = stratified_neuro_counts_adult,
                                                                         tableone = tableone_adult)

stratified_neuro_counts_age_pediatric <- compute_stratified_neuro_counts_age(stratified_neuro_counts = stratified_neuro_counts_pediatric,
                                                                 tableone = tableone_pediatric)

Diagnoses by Adult and Pediatric Populations

n_0_2 <- get_table_n(tableone_combine, var = "0-2")
#n_3_5 <- get_table_n(tableone_combine, var = "3-5")
n_6_11 <- get_table_n(tableone_combine, var = "6-11")
n_12_17 <- get_table_n(tableone_combine, var = "12-17")

total_ped_age_count = as.numeric(n_0_2) +  as.numeric(n_6_11) + as.numeric(n_12_17)

stratified_neuro_counts_age_pediatric_reformat = stratified_neuro_counts_age_pediatric %>%
  select(variable, pns_cns, description, n_var, total_n_var) %>%
  mutate(variable = "0-17 Years") %>%
  group_by(pns_cns, description) %>%
  mutate(total_n_var = sum(total_n_var, na.rm = TRUE)) %>%
  ungroup() %>%
  # n_var is the total per `age_group` stratification
  select(-n_var) %>%
  distinct() %>%
  mutate(perc = round(total_n_var/total_ped_age_count*100,1),
         perc = if_else(perc < 0.1, '<0.1', as.character(perc)),
         perc = paste0(perc, "%"))

stratified_neuro_counts_age_combined <- stratified_neuro_counts_age_adult %>%
  select(variable, pns_cns, description, total_n_var, perc) %>%
  rbind(., stratified_neuro_counts_age_pediatric_reformat %>%
          select(variable, pns_cns, description, total_n_var, perc))

stratified_neuro_counts_age_combined$variable <- factor(stratified_neuro_counts_age_combined$variable,
                                                                      levels=c("0-17 Years", "18-25 Years", "26-49 Years", "50-69 Years", "70-79 Years", "80+ Years"))

group.colors <- c(NNC = "gray", PNS = "slateblue", CNS ="tomato")

adult_ped_combined_plot <- ggplot(stratified_neuro_counts_age_combined,
                                                aes(x = total_n_var, y = reorder(description,
                                                                                 total_n_var), 
                                                    fill = pns_cns)) +
  geom_bar(stat = "identity") +
  geom_text(
    aes(x = total_n_var, y = description, label = perc),
    hjust = -0.1,
    #hjust="inward",
    size = 6,
    inherit.aes = TRUE) +
  scale_x_continuous(expand = expansion(mult = c(0,0.2))) + # to ensure percent labels are within the figure
  facet_grid(pns_cns ~ variable, scales = "free", space = "free_y") + 
  scale_y_discrete("", position="left", labels = function(x) str_wrap(x, width = 100)) +   
  scale_fill_manual(values = group.colors) +
  xlab("Number of Patients") +
  theme(legend.position = "none",
        strip.text.y = element_text(size = 40),
        strip.text.x = element_text(size = 35),
        axis.text.y.left = element_text(size = 25, face = "bold"),
        axis.text.x = element_text(size = 25),
        axis.title.x = element_text(size = 35))

ggsave("figures/Fig3_diagnoses_by_age_adult_ped_18group.png", adult_ped_combined_plot, width = 40, height = 19, dpi = 300)

Evaluate Diagnoses by Outcome

stratified_neuro_counts_adult$population = "adult"
stratified_neuro_counts_pediatric$population = "pediatric"

stratified_neuro_counts_combine <- rbind(stratified_neuro_counts_adult, stratified_neuro_counts_pediatric)

## determine total number of patients who died or severe
n_death_adult = get_table_n(df = tableone_adult, var = "Deceased") %>% as.numeric()
n_severe_adult = get_table_n(df = tableone_adult, var = "Severe") %>% as.numeric()

n_death_pediatric = get_table_n(df = tableone_pediatric, var = "Deceased") %>% as.numeric()
n_severe_pediatric = get_table_n(df = tableone_pediatric, var = "Severe") %>% as.numeric()

stratified_neuro_counts_outcomes <- stratified_neuro_counts_combine %>% 
  filter(variable %in% c("Severe", "Deceased"),
         !code == "NN", 
         !total_n_var == 0) %>%          
         # total_n would be calculating total codes across each outcome group which is not true reflection of patient counts b/c patients can have more than one code
         #mutate(total_n = sum(total_n_var)) %>% 
         group_by(variable, description, population) %>% 
  mutate(total_n_var = sum(total_n_var)) %>% 
  ungroup() %>% 
  mutate(perc = case_when(variable == "Severe" & population == "adult" ~ round(total_n_var/n_severe_adult*100,1),
                                 variable == "Deceased" & population == "adult" ~ round(total_n_var/n_death_adult*100,1),
                                 variable == "Severe" & population == "pediatric" ~ round(total_n_var/n_severe_pediatric*100,1),
                                 variable == "Deceased" & population == "pediatric" ~ round(total_n_var/n_death_pediatric*100,1)),
         perc = if_else(perc < 0.1, '<0.1', as.character(perc)),
       perc = paste0(perc, "%")) %>% 
  select(-code, -concept_type) %>% 
  distinct()

adult_outcomes = ggplot(stratified_neuro_counts_outcomes %>% 
         filter(population == "adult") %>% 
           mutate(variable = as.factor(variable) %>% 
                                          fct_recode(
                                            Mortality = "Deceased")),
       aes(x = total_n_var, y = reorder(description, total_n_var), fill = pns_cns)) +
  geom_bar(stat = "identity") +
 geom_text(
    aes(x = total_n_var, y = description, label = perc),
    hjust = -0.1,
    #hjust="inward",
    size = 5,
    inherit.aes = TRUE) +
  scale_x_continuous(expand = expansion(mult = c(0,0.2))) + 
  facet_grid(pns_cns ~ variable, scales = "free", space = "free_y") + 
  scale_y_discrete("", position="left", labels = function(x) str_wrap(x, width = 100)) +   
  scale_fill_manual(values = group.colors) +
  xlab("Number of Patients") + 
  ggtitle("B. Adult") + 
  theme(legend.position = "none",
        plot.title = element_text(size = 44, face = "bold", hjust = -0.5), 
        strip.text.y = element_text(size = 35),
        strip.text.x = element_text(size = 30),
        axis.text.y.left = element_text(size = 20, face = "bold"),
        axis.text.x = element_text(size = 15),
        axis.title.x = element_text(size = 35))

ggsave("figures/SuppFig1_diagnoses_by_outcomes_adult.png", adult_outcomes, width = 27, height = 15, dpi = 300)

pediatric_outcomes = ggplot(stratified_neuro_counts_outcomes %>% 
         filter(population == "pediatric") %>% 
           mutate(variable = as.factor(variable) %>% 
                                          fct_recode(
                                            Mortality = "Deceased")),
       aes(x = total_n_var, y = reorder(description, total_n_var), fill = pns_cns)) +
  geom_bar(stat = "identity") +
 geom_text(
    aes(x = total_n_var, y = description, label = perc),
    hjust = -0.1,
    #hjust="inward",
    size = 5,
    inherit.aes = TRUE) +
  scale_x_continuous(expand = expansion(mult = c(0,0.2))) + 
  facet_grid(pns_cns ~ variable, scales = "free", space = "free_y") + 
  scale_y_discrete("", position="left", labels = function(x) str_wrap(x, width = 100)) +   
  scale_fill_manual(values = group.colors) +
  xlab("") + 
  ggtitle("A. Pediatric") + 
  theme(legend.position = "none",
        plot.title = element_text(size = 44, face = "bold", hjust = -0.5), 
        strip.text.y = element_text(size = 35),
        strip.text.x = element_text(size = 30),
        axis.text.y.left = element_text(size = 20, face = "bold"),
        axis.text.x = element_text(size = 15),
        axis.title.x = element_text(size = 35))

ggsave("figures/SuppFig1_diagnoses_by_outcomes_pediatric.png", pediatric_outcomes, width = 27, height = 9, dpi = 300)

Evaluate patients with “Both” CNS and PNS diseases

These patients were excluded from the primary analysis due to potential confounding.

# create empty list to store the counts of "both" patients
both_counts_list <- list()

# for each healthcare system, retrieve the number of excluded "both" patients recorded
for (i in sorted_sites) {
  n_both <- results[[paste(i)]][["first_hosp_results"]][["both_counts"]]
  
  both_counts_list[[i]] <- n_both
}

both_counts <- bind_rows(both_counts_list) %>% t() %>% data.frame() %>% sum()

print(paste(both_counts, "(", round(both_counts/(as.numeric(tableone_combine$N[1])+both_counts)*100,2), "%)")) 

## [1] "1990 ( 1.84 %)"

Comorbidity Analysis

Next, we will evaluate the frequency of comorbidities in our population and the associated risk that comorbidities may incur for acquiring a CNS or PNS diagnosis during COVID-19 hospitalization.

1. Create a table of counts for each comorbidity and stratified by neurological status.

comorbidity_table_adult <- comorb_table_wide %>%
  select(Comorbidity, population, Comorb_Total, `NNC_N_%`, `CNS_N_%`, `PNS_N_%`) %>%
  filter(population == "Adult") %>% 
  rename(N = "Comorb_Total",
         NNC = "NNC_N_%",
         CNS = "CNS_N_%",
         PNS = "PNS_N_%") %>% 
  mutate(N = paste0(N, " (", round(N/n_adult, 2)*100, "%)"))

comorbidity_table_pediatric <- comorb_table_wide %>%
  select(Comorbidity, population, Comorb_Total, `NNC_N_%`, `CNS_N_%`, `PNS_N_%`) %>%
  filter(population == "Pediatric") %>% 
  rename(N = "Comorb_Total",
         NNC = "NNC_N_%",
         CNS = "CNS_N_%",
         PNS = "PNS_N_%") %>% 
  mutate(N = paste0(N, " (", round(N/n_pediatric, 2)*100, "%)"))

supp_comorbidity_table <- rbind(comorbidity_table_adult, comorbidity_table_pediatric) %>% 
  arrange(Comorbidity)

write.csv(supp_comorbidity_table, "tables/SuppTable_Comorbidity_Counts.csv", row.names = FALSE)

2. Create vector counts of patients with comorbidity data

neuro_pt_counts_sum_adult <- neuro_pt_counts %>% 
  filter(population == "Adult") %>% 
  select(-population)

neuro_pt_counts_sum_pediatric <- neuro_pt_counts %>% 
  filter(population == "Pediatric") %>% 
  select(-population)

none_n_adult = neuro_pt_counts_sum_adult$None_n
cns_n_adult = neuro_pt_counts_sum_adult$CNS_n
pns_n_adult = neuro_pt_counts_sum_adult$PNS_n

none_n_pediatric = neuro_pt_counts_sum_pediatric$None_n
cns_n_pediatric = neuro_pt_counts_sum_pediatric$CNS_n
pns_n_pediatric = neuro_pt_counts_sum_pediatric$PNS_n

3. Calculate the relative risk (RR) of developing a CNS or PNS diagnosis for patients with each comorbidity

calculate_rr <- function(is_pediatric = FALSE) {
  
  if(is_pediatric == FALSE) {
    cns_n = cns_n_adult
    pns_n = pns_n_adult
    neuro_pt_counts_sum = neuro_pt_counts_sum_adult
    pop = "Adult"
    # create a list of comorbidities
    list_of_comorbs <- comorb_table_wide %>% 
      filter(population == paste(pop)) %>% 
      distinct(Comorbidity) 
    list_of_comorbs <- list_of_comorbs$Comorbidity
    } else  {
    cns_n = cns_n_pediatric
    pns_n = pns_n_pediatric
    neuro_pt_counts_sum = neuro_pt_counts_sum_pediatric
    pop = "Pediatric"
    # create a list of comorbidities
    list_of_comorbs <- comorb_table_wide %>% 
      filter(population == paste(pop)) %>% 
      distinct(Comorbidity) %>% 
      as.vector()
    list_of_comorbs <- list_of_comorbs$Comorbidity
  }
  
  
  # create empty list to save relative risk calculations
  rr_calcs <- list()
  
  # for each comorbidity, we will calculate the relative risk of having a CNS and PNS diagnosis
  for (i in list_of_comorbs) {
  
    # https://www.cdc.gov/csels/dsepd/ss1978/lesson3/section5.html
    # create matrix as:
    # exposed group - with outcome, without outcome
    # non exposed group - with outcome, without outcome
  
    # where variables ending with `_Total` are the total number of CNS, PNS, or NNC patients with the respective comorbidity.
    # variables ending with `_n` are the total numbers in the entire population
  
    # to calculate confidence intervals
  #https://sphweb.bumc.bu.edu/otlt/mph-modules/bs/bs704_confidence_intervals/bs704_confidence_intervals8.html
  
    rr_cns <- comorb_table_wide %>%
      filter(population == paste(pop),
             Comorbidity == paste(i)) %>%
      mutate(a = CNS_Total, # number CNS w/comorb
             b = Comorb_Total - CNS_Total, # number of comorb pts w/o CNS
             c = cns_n - CNS_Total, # number of CNS pts w/o comorb
             d = sum(neuro_pt_counts_sum) - Comorb_Total - c, # number of pts w/o comorb or CNS
             risk_cns = a/(a+b),
             risk_non_cns = c/(c+d),
             rr_CNS = risk_cns/risk_non_cns,
             # for confidence intervals
             ci_1 = (b/a)/(a+b), # (n1-x1)/n1 
             ci_2 = (d/c)/(c+d), # (n2-x2)/n2
             ci_root = sqrt(ci_1 + ci_2),
             ci_lower = log(rr_CNS) - (1.96*ci_root),
             ci_upper = log(rr_CNS) + (1.96*ci_root),
             l_ci_cns = exp(ci_lower),
             u_ci_cns = exp(ci_upper)) %>%
      select(Comorbidity, Comorb_Total, None_Total, CNS_Total, PNS_Total, rr_CNS, l_ci_cns, u_ci_cns)
  
      rr_pns <- comorb_table_wide %>%
      filter(population == paste(pop),
             Comorbidity == paste(i)) %>%
      mutate(a = PNS_Total, # number PNS w/comorb
             b = Comorb_Total - PNS_Total, # number of comorb pts w/o PNS
             c = pns_n - PNS_Total, # number of PNS pts w/o comorb
             d = sum(neuro_pt_counts_sum) - Comorb_Total - c, # number of pts w/o comorb or PNS
             risk_pns = a/(a+b),
             risk_non_pns = c/(c+d),
             rr_PNS = risk_pns/risk_non_pns,
             # for confidence intervals
             ci_1 = (b/a)/(a+b),
             ci_2 = (d/c)/(c+d),
             ci_root = sqrt(ci_1 + ci_2),
             ci_lower = log(rr_PNS) - (1.96*ci_root),
             ci_upper = log(rr_PNS) + (1.96*ci_root),
             l_ci_pns = exp(ci_lower),
             u_ci_pns = exp(ci_upper)) %>%
      select(Comorbidity, rr_PNS, l_ci_pns, u_ci_pns)
  
      rr <- rr_cns %>%
        left_join(., rr_pns, by = c("Comorbidity"))
  
    rr_calcs[[i]] <- rr
  
  }
  
  rr_results <- bind_rows(rr_calcs) %>%
      mutate(rr_CNS = round(rr_CNS, 2),
           l_ci_cns = round(l_ci_cns, 2),
           u_ci_cns = round(u_ci_cns, 2),
           rr_PNS = round(rr_PNS, 2),
           l_ci_pns = round(l_ci_pns, 2),
           u_ci_pns = round(u_ci_pns, 2),
           `RR CNS (95% CI)` = paste(rr_CNS, "(", l_ci_cns, ",", u_ci_cns, ")"),
           `RR PNS (95% CI)` = paste(rr_PNS, "(", l_ci_pns, ",", u_ci_pns, ")")) %>%
    select(Comorbidity, rr_CNS, l_ci_cns, u_ci_cns, `RR CNS (95% CI)`,
           rr_PNS, l_ci_pns, u_ci_pns, `RR PNS (95% CI)`)
  
  # tidy up data
  write.csv(rr_results %>%
              arrange(desc(rr_CNS)) %>%
              select(Comorbidity, `RR CNS (95% CI)`, `RR PNS (95% CI)`), 
            paste0("tables/Supp.Table4_RR_", pop, ".csv"), row.names = FALSE)
  
    # reformat data
  rr_results_tidy_rr <- rr_results %>%
    select(Comorbidity, rr_CNS, rr_PNS) %>%
    pivot_longer(cols = rr_CNS:rr_PNS, names_to = "Neuro Status", values_to = "RR") %>%
    mutate(`Neuro Status` = if_else(`Neuro Status` == "rr_CNS", "CNS", "PNS"))
  
  rr_results_tidy_l_ci <- rr_results %>%
    select(Comorbidity, l_ci_cns, l_ci_pns) %>%
    pivot_longer(cols = l_ci_cns:l_ci_pns, names_to = "Neuro Status", values_to = "l_CI") %>%
    mutate(`Neuro Status` = if_else(`Neuro Status` == "l_ci_cns", "CNS", "PNS"))
  
  rr_results_tidy_u_ci <- rr_results %>%
    select(Comorbidity, u_ci_cns, u_ci_pns) %>%
    pivot_longer(cols = u_ci_cns:u_ci_pns, names_to = "Neuro Status", values_to = "u_CI") %>%
    mutate(`Neuro Status` = if_else(`Neuro Status` == "u_ci_cns", "CNS", "PNS"))
  
  # combine all tidy results
  rr_results_tidy <- rr_results_tidy_rr %>%
    left_join(., rr_results_tidy_l_ci,
                               by = c("Comorbidity", "Neuro Status")) %>%
    left_join(., rr_results_tidy_u_ci,
                               by = c("Comorbidity", "Neuro Status"))
  
  # order by CNS diagnoses with higher RR
  cns_top_results <- rr_results_tidy %>%
    filter(`Neuro Status` == "CNS") %>%
    arrange(RR) %>%
    select(Comorbidity)
  
  rr_results_tidy$Comorbidity <-  ordered(rr_results_tidy$Comorbidity, levels = cns_top_results$Comorbidity)
  
  cns_rr_plot <- ggplot(rr_results_tidy %>%
           filter(`Neuro Status` == "CNS"),
         aes(x = RR, y = Comorbidity)) +
    geom_vline(aes(xintercept = 1), size = .25, linetype = "dashed") +
      geom_errorbarh(aes(xmax = u_CI, xmin = l_CI), size = 0.7, height = 0.2, color = "gray50") +
      geom_point(size = 1.5, color = "tomato") +
      scale_x_continuous(breaks = seq(0, 4, 1), labels = seq(0, 4, 1),
                         limits = c(0,4)) +
    ylab("") +
    xlab("Relative Risk") +
    ggtitle("CNS Patients") + 
    theme(axis.text.y = element_text(face = "bold"))
  
  # arrange by top PNS results
  pns_top_results <- rr_results_tidy %>%
    filter(`Neuro Status` == "PNS") %>%
    arrange(RR) %>%
    select(Comorbidity)
  
  rr_results_tidy$Comorbidity <-  ordered(rr_results_tidy$Comorbidity, levels = pns_top_results$Comorbidity)
  
  
  pns_rr_plot <- ggplot(rr_results_tidy %>%
           filter(`Neuro Status` == "PNS"),
         aes(x = RR, y = Comorbidity)) +
    geom_vline(aes(xintercept = 1), size = .25, linetype = "dashed") +
      geom_errorbarh(aes(xmax = u_CI, xmin = l_CI), size = 0.70, height = 0.2, color = "gray50") +
      geom_point(size = 1.5, color = "slateblue") +
      scale_x_continuous(breaks = seq(0, 1.5, 0.5), labels = seq(0, 1.5, 0.5),
                         limits = c(0,1.5)) +
    ylab("") +
    xlab("Relative Risk") +
    ggtitle("PNS Patients") + 
    theme(axis.text.y = element_text(face = "bold"))
  
  pns_rr_plot <- set_panel_size(pns_rr_plot,
                            width  = unit(7, "cm"),
                            height = unit(6.5, "in"))
  
  cns_rr_plot <- set_panel_size(cns_rr_plot,
                            width  = unit(7, "cm"),
                            height = unit(6.5, "in"))
  
  #grid.arrange(cns_rr_plot, pns_rr_plot, ncol=2)
  
  ggsave(paste0("figures/Figure4.relative_risk_", pop, ".png"), grid.arrange(cns_rr_plot, pns_rr_plot, ncol=2), width = 15, height = 8, dpi = 300)
  
  }

Relative Risk (RR)

Here we plot the relative risk of acquiring a CNS or PNS diagnosis in adult patients who have each comorbidity of the ECI compared to those who do not.

We are not able to compute this analysis in pediatric patients, due to the overall low incidence of comorbidities and neurological illness during COVID-19 hospitalization.

LPCA deviance explained

During the local analysis for each participating healthcare system, we computed the top 10 principal components using logistic principal component analysis (LPCA). LPCA was performed on the matrix of npatients x 29 ECI comorbidities. Across healthcare systems, 10 principal components explained at least 75% of the model deviance. 

This supplementary analysis was performed in order to determine the best way to adjust for each patient’s comorbidity burden in our downstream survival models.

compare_deviance <- function(sorted_sites, is_pediatric=FALSE) {
  
  pca_dev <- list()
  
  if(is_pediatric==FALSE) {
    population = "adults"
    sorted_sites = sorted_sites[!sorted_sites%in% c("GOSH_results", "BCH_results")]
    } else {
    population = "pediatrics"
    }

  # for each healthcare system, we will calculate the total number of patients with each ICD code
  for (i in sorted_sites) {
    print(i)
    tmp <- get(i)
    tmp_pca <- tmp[[c(
      "comorbidities",
      paste0("comorb_", population),
      "deviance_expl"
    )]] %>% 
      data.frame() %>% 
      rename(dev_expl = '.')
    tmp_pca$site <- tmp[["site"]]
    pca_dev[[i]] <- tmp_pca
    
  }
  
  pca_df <- bind_rows(pca_dev)
  
  return(pca_df)
  
}

pca_adult <- compare_deviance(sorted_sites = adult_sites, is_pediatric = FALSE)

## [1] "APHP_results"
## [1] "FRBDX_results"
## [1] "UKFR_results"
## [1] "ICSM_results"
## [1] "HPG23_results"
## [1] "NUH_results"
## [1] "MGB_results"
## [1] "UPENN_results"
## [1] "UMICH_results"
## [1] "NWU_results"
## [1] "UPITT_results"
## [1] "H12O_results"
## [1] "UKY_results"
## [1] "VA1_results"
## [1] "VA2_results"
## [1] "VA3_results"
## [1] "VA4_results"
## [1] "VA5_results"
## [1] "UCLA_results"

pca_data <- bind_rows(pca_adult)


theme_set(theme_classic())

pca_plot <- ggplot(pca_data, 
       aes(x = fct_reorder(site, -dev_expl), y = dev_expl, group = 1)) +
  geom_point(size = 2.5) +
  geom_line() +
  xlab("Healthcare System") + 
  ylab("Proportion of Deviance explained") +  
  ylim(0,1) + 
  theme(axis.text.x=element_text(angle=90, hjust=1, face = "bold", size = 12),
        axis.text.y=element_text(face = "bold"), 
        axis.title.y = element_text(face = "bold", size = 15)); pca_plot

ggsave("figures/SuppFig_pca_deviance.png", pca_plot, height = 6, width = 6)