generated from opensafely/research-template
-
Notifications
You must be signed in to change notification settings - Fork 1
/
ITS_broad_model.R
250 lines (222 loc) · 11 KB
/
ITS_broad_model.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
#### this scirpt transfers the data to a prepared verson for ITS model
### load library ###
library("tidyverse")
library(foreign)
library(tsModel)
library(lmtest)
library(Epi)
library(multcomp)
library(splines)
library(vcd)
library(here)
library(lubridate)
library(stringr)
library(ggplot2)
library(patchwork)
library(dplyr)
library(tidyr)
### import data ###
all_files <- list.files(here::here("output"), pattern = "df.model_")
outcomes <- stringr::str_remove_all(all_files, c("df.model_|.csv"))
outcome_of_interest_namematch <- bind_cols("outcome" = outcomes,
"outcome_name" = (c("Asthma","Cold","COPD","Cough",
"LRTI","Otitis externa","Otitis media",
"Pneumonia","Renal","Sepsis","Sinusitis",
"Sore throat","URTI","UTI"))
)
bkg_colour <- "gray99"
# load data ---------------------------------------------------------------
for(ii in 1:length(outcomes)){
load_file <- read.csv(here::here("output", paste0("df.model_", outcomes[ii], ".csv")))
assign(outcomes[ii], load_file)
}
its_function <- function(outcomes_vec = outcomes,
display_from = as.Date("2019-01-01")){
plot_its <- function(outcome){
df_outcome <- get(outcome)
## model binomial
# Change in level + slope:
### include interaction with time (centred at end of Lockdown adjustment period)
ldn_centre <- df_outcome$time[min(which(df_outcome$covid == 1))]
## fit model, calculate lagged residuals to fit in final model
binom_model1 <- glm(as.matrix(cbind(numOutcome, numEligible)) ~ covid + I(time-ldn_centre) + I(time-ldn_centre):covid + as.factor(months) , family=binomial, data = filter(df_outcome, !is.na(covid)))
### confidence intervals for the coefficients
ci.exp(binom_model1)
binom_lagres <- lag(residuals(binom_model1)) %>% as.numeric()
res1 <- residuals(binom_model1,type="deviance")
## manipulate data so output looks cleaner
model_data <- df_outcome %>%
mutate(timeC = time - ldn_centre) %>%
mutate_at("months", ~as.factor(.))
## fit model with lagged residuals
binom_model2 <- glm(as.matrix(cbind(numOutcome, numEligible)) ~ covid + timeC + timeC:covid + as.factor(months) + binom_lagres, family=binomial, data = filter(model_data, !is.na(covid)))
ci.exp(binom_model2)
summary.glm(binom_model2)
## calculate dispersion adjustment parameter -- https://online.stat.psu.edu/stat504/node/162/
#Pearson Goodness-of-fit statistic
pearson_gof <- sum(residuals(binom_model2, type = "pearson")^2)
df <- binom_model2$df.residual
deviance_adjustment <- pearson_gof/df
## some manual manipulation to merge the lagged residuals varaible back with the original data
missing_data_start <- min(which(is.na(model_data$covid)))
missing_data_end <- max(which(is.na(model_data$covid)))
missing_data_restart <- max(which(is.na(model_data$covid)))
binom_lagres_timing <- bind_cols("time" = model_data$time[!is.na(model_data$covid)],
"binom_lagres" = binom_lagres)
## set up data frame to calculate linear predictions
outcome_pred <- model_data %>%
left_join(binom_lagres_timing, by = "time") %>%
mutate_at("binom_lagres", ~(. = 0))
## predict values adjusted for overdispersion
pred1 <- predict(binom_model2, newdata = outcome_pred, se.fit = TRUE, interval="confidence", dispersion = deviance_adjustment)
predicted_vals <- pred1$fit
stbp <- pred1$se.fit
## set up data frame to calculate linear predictions with no covid and predict values
outcome_pred_nointervention <- outcome_pred %>%
mutate_at("covid", ~(.=0))
pred_noCovid <- predict(binom_model2, newdata = outcome_pred_nointervention, se.fit = TRUE, interval="confidence", dispersion = deviance_adjustment)
pred_noCov <- pred_noCovid$fit
stbp_noCov <- pred_noCovid$se.fit
## combine all those predictions and convert from log odds to percentage reporting
df_se <- bind_cols(stbp = stbp, stbp_noCov = stbp_noCov,
pred = predicted_vals, pred_noCov = pred_noCov) %>%
mutate(
#CIs
upp = pred + (1.96*stbp),
low = pred - (1.96*stbp),
upp_noCov = pred_noCov + (1.96*stbp_noCov),
low0_noCov = pred_noCov - (1.96*stbp_noCov),
# probline
predicted_vals = exp(pred)/(1+exp(pred)),
probline_noCov = exp(pred_noCov)/(1+exp(pred_noCov)),
#
uci = exp(upp)/(1+exp(upp)),
lci = exp(low)/(1+exp(low)),
#
uci_noCov = exp(upp_noCov)/(1+exp(upp_noCov)),
lci_noCov = exp(low0_noCov)/(1+exp(low0_noCov))
)
## combine data set and predictions
outcome_plot <- bind_cols(outcome_pred, df_se) %>%
mutate(var = outcome)
## Get ORs for effect of covid
paramter_estimates <- as.data.frame(ci.exp(binom_model2))
vals_to_print <- paramter_estimates %>%
mutate(var = rownames(paramter_estimates)) %>%
filter(var == "covid") %>%
mutate(var = outcome)
## output
return(list(df_1 = outcome_plot, vals_to_print = vals_to_print))
}
# the plot ----------------------------------------------------------------
main_plot_data <- NULL
forest_plot_data <- NULL
for(ii in 1:length(outcomes_vec)){
main_plot_data <- main_plot_data %>%
bind_rows(
plot_its(outcomes_vec[ii])$df_1
)
forest_plot_data <- forest_plot_data %>%
bind_rows(
plot_its(outcomes_vec[ii])$vals_to_print
)
}
## convert proportions into percentage
main_plot_data <- main_plot_data %>%
mutate(pc_broad = (numOutcome/numEligible)*100) %>%
mutate_at(.vars = c("predicted_vals", "lci", "uci", "probline_noCov", "uci_noCov", "lci_noCov"),
~.*100) %>%
left_join(outcome_of_interest_namematch, by = c("var" = "outcome"))
## replace outcome name with the pretty name for printing on results
main_plot_data$outcome_name <- factor(main_plot_data$outcome_name, levels = outcome_of_interest_namematch$outcome_name)
abline_max <- main_plot_data$weekPlot[max(which(is.na(main_plot_data$covid)))+1]
abline_min <- main_plot_data$weekPlot[min(which(is.na(main_plot_data$covid)))-1]
if(is.na(abline_min) & is.na(abline_max)){
abline_min <- start_covid
abline_max <- start_covid
}
write_csv(main_plot_data, here::here("output", "its_main_plot_data.csv"))
main_plot_data$weekPlot <- as.Date(main_plot_data$weekPlot)
plot1 <- ggplot(main_plot_data, aes(x = weekPlot, y = pc_broad, group = outcome_name)) +
# the data
geom_line(col = "gray60") +
### the probability if therer was no Covid
geom_line(data = main_plot_data, aes(y = probline_noCov), col = 2, lty = 2) +
geom_ribbon(data = main_plot_data, aes(ymin = lci_noCov, ymax=uci_noCov), fill = alpha(2,0.4), lty = 0) +
### probability with model (inc. std. error)
geom_line(aes(y = predicted_vals), col = 4, lty = 2) +
geom_ribbon(aes(ymin = lci, ymax=uci), fill = alpha(4,0.4), lty = 0) +
### format the plot
facet_wrap(~outcome_name, scales = "free", ncol = 4) +
geom_vline(xintercept = c(as.Date(abline_min),
as.Date(abline_max)), col = 1, lwd = 1) + # 2020-04-05 is first week/data After lockdown gap
labs(y = "% of broad-spectrum prescription", title = "A") +
theme_classic() +
theme(axis.title = element_text(size =16),
axis.text.x = element_text(angle = 60, hjust = 1, size = 12),
legend.position = "top",
plot.background = element_rect(fill = bkg_colour, colour = NA),
panel.background = element_rect(fill = bkg_colour, colour = NA),
legend.background = element_rect(fill = bkg_colour, colour = NA),
legend.text = element_text(size = 12),
legend.title = element_text(size = 12),
strip.text = element_text(size = 12, hjust = 0),
strip.background = element_rect(fill = bkg_colour, colour = NA),
panel.grid.major = element_blank(),
panel.grid.minor.x = element_blank(),
panel.grid.minor.y = element_line(size=.2, color=rgb(0,0,0,0.2)) ,
panel.grid.major.y = element_line(size=.2, color=rgb(0,0,0,0.3)))
plot1
ggsave(
plot= plot1,
filename="predicted_plot.jpeg", path=here::here("output"),
)
# Forest plot of ORs ------------------------------------------------------
## clean up the names
forest_plot_df <- forest_plot_data %>%
rename("Est" = "exp(Est.)", "lci" = "2.5%", "uci" = "97.5%") %>%
left_join(outcome_of_interest_namematch, by = c("var" = "outcome"))
# changes the names of outcomes to full names
forest_plot_df$outcome_name <- factor(forest_plot_df$outcome_name, levels = outcome_of_interest_namematch$outcome_name)
# export table of results for the appendix
write_csv(forest_plot_df, here::here("output", "its_main_ORs.csv"))
forest_plot_df <- forest_plot_df %>% filter(outcome_name %in% c("Asthma","Cold","COPD","Cough",
"LRTI","Otitis externa","Otitis media",
"Renal","Sinusitis",
"Sore throat","URTI","UTI") )
forest_plot_df <- forest_plot_df %>%
mutate(dummy_facet = "A")
## Forest plot
fp <- ggplot(data=forest_plot_df, aes(x=dummy_facet, y=Est, ymin=lci, ymax=uci)) +
geom_point(size = 2.5, pch = 16, colour = "darkred") +
geom_linerange(lwd = 1.5, colour = "darkred") +
geom_hline(yintercept=1, lty=2) + # add a dotted line at x=1 after flip
coord_flip() + # flip coordinates (puts labels on y axis)
labs(x = "", y = "95% CI", title = "B: Reduction") +
facet_wrap(~outcome_name, ncol = 1, dir = "h", strip.position = "right") +
theme_classic() +
theme(axis.title = element_text(size = 16),
axis.text.y = element_blank(),
axis.line.y.left = element_blank(),
axis.line.y.right = element_line(),
axis.ticks.y = element_blank(),
axis.text.x = element_text(angle = 0),
legend.position = "top",
plot.background = element_rect(fill = bkg_colour, colour = NA),
panel.background = element_rect(fill = bkg_colour, colour = NA),
legend.background = element_rect(fill = bkg_colour, colour = NA),
strip.background = element_rect(fill = bkg_colour, colour = NA),
strip.text.y = element_text(hjust=0.5, vjust = 0, angle=0, size = 10),
panel.grid.major = element_blank(),
panel.grid.minor.x = element_blank(),
panel.grid.minor.y = element_line(size=.2, color=rgb(0,0,0,0.2)) ,
panel.grid.major.y = element_line(size=.2, color=rgb(0,0,0,0.3)))
fp
ggsave(
plot= fp,
filename="forest_plot_broad.jpeg", path=here::here("output"),
)
}
its_function(outcomes_vec = outcomes,
display_from <- as.Date("2019-01-01")
)