RiskExtrapolation/31_ValidationModels.R at main · ehm-lab/RiskExtrapolation · GitHub

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
################################################################################
#
#  RiskExtrapolation
#
#  First-stage unobserved ERFs for comparison
#
################################################################################

#-------------------------
# Run first-stage
#-------------------------

#----- Prepare parallelisation
ncores <- detectCores()
cl <- makeCluster(max(1, ncores - 2))
registerDoParallel(cl)

#----- Perform first stage on each city

# Loop over cities, splitting the data table
validres <- foreach(dat = isplit(tspred, tspred$city_code), .combine = rbind,
  .packages = c("dlnm", "splines", "foreach", "mixmeta", "dplyr")) %dopar%
  {

  # Extract from iterator
  city <- dat$key[[1]]
  dat <- as.data.frame(dat$value)

  # Define crossbasis
  argvar <- list(fun = varfun, degree = vardegree,
    knots = quantile(dat$tmean, varper / 100, na.rm = T))
  cb <- crossbasis(dat$tmean, lag = maxlag, argvar = argvar,
    arglag = list(fun = lagfun, knots = lagknots))

  # Loop on deaths by age group
  agevars <- grep("deaths", names(dat), value = T)
  agegroups <- gsub("deaths_", "", agevars)
  out <- foreach(y = dat[, agevars], agegr = agegroups, .combine = rbind) %do% {

    # Run model
    res <- glm(y ~ cb + dow + ns(date, df = 7 * length(unique(year))),
      dat, family = quasipoisson)

    # Reduce coefficients to overall cumulative
    redall <- crossreduce(cb, res, cen = median(dat$tmean, na.rm = T))

    # Output
    coefs <- coef(redall); names(coefs) <- sprintf("coef%i", 1:nc)
    vcovs <- vechMat(vcov(redall))
    names(vcovs) <- sprintf("vcov%i%i",
      row(vcov(redall))[lower.tri(row(vcov(redall)), diag = T)],
      col(vcov(redall))[lower.tri(col(vcov(redall)), diag = T)])
    data.frame(t(coefs), t(vcovs),
      totdeath = sum(y, na.rm = T), conv = res$converged, agegroup = agegr)
  }

  # Add info about city and return
  mutate(out, city_code = city)
}

# Stop parallel
stopCluster(cl)

#----- Create new full dataset for risk prediction

# Merge full dataset
metafull <- select(metadf, city_code, agegroup, lon, lat, age, obs,
    all_of(unname(metaprednames))) |>
  merge(rbind(stage1res, validres))

#-------------------------
# Create predictive model including "unobserved" locations
#-------------------------

#----- New composite indices of vulnerability

# Compute PLSR (basic PLS computed as a linear model)
fullpls <- plsr(as.formula(plsform), data = metafull, scale = T)

# Extract scores for all locations and age
newcomp <- predict(fullpls, newdata = metafull, type = "scores")
colnames(newcomp) <- sprintf("Comp%i", 1:ncol(newcomp))
metafull <- cbind(metafull, newcomp)

#----- New predictive model

# Extract variance-covariance
sfull <- select(metafull, starts_with("vcov")) |> data.matrix()

# Fit model
validmeta <- mixmeta(as.formula(fixform), random = ranform,
  data = metafull, S = sfull, subset = conv)

# Extract BLUPs
validblups <- blup(validmeta, vcov = T)
blupcoefs <- sapply(validblups, "[[", "blup") |> t()
blupvcov <- sapply(validblups, function(x) vechMat(x$vcov)) |> t()
colnames(blupvcov) <- grep("vcov", names(stage1res), value = T)
validblups <- cbind(subset(metafull, conv, c(city_code, agegroup)), blupcoefs,
  blupvcov)


#-------------------------
# Alternative predictions
#-------------------------

#----- Fixed effect part only

# Rename and store into data.frame
names(fixcoef) <- coefvars
colnames(fixvcov) <- grep("vcov", names(stage1res), value = T)
fixedres <- cbind(metadf[, c("city_code", "agegroup")],
  fixcoef, fixvcov)

#----- Age-only model

# Refit model
ageform <- sprintf("cbind(%s) ~ ns(age, knots = 60)",
  paste(coefvars, collapse = ", "))
ageonlymeta <- mixmeta(as.formula(ageform), random = ranform,
  data = metadf, S = smat, subset = conv)

# Predict and store
ageonlypred <- predict(ageonlymeta, metadf, vcov = T)
ageonlycoef <- sapply(ageonlypred, "[[", "fit") |> t()
ageonlyvcov <- sapply(ageonlypred, function(x) vechMat(x$vcov)) |> t()
names(ageonlycoef) <- coefvars
colnames(ageonlyvcov) <- grep("vcov", names(stage1res), value = T)
ageonlyres <- cbind(metadf[, c("city_code", "agegroup")],
  ageonlycoef, ageonlyvcov)

#----- Components only model

# Refit model
compform <- sprintf("cbind(%s) ~ %s",
  paste(coefvars, collapse = ", "),
  paste(colnames(comps)[1:npc], collapse = " + "))
compmeta <- mixmeta(as.formula(compform), random = ranform,
  data = metadf, S = smat, subset = conv)

# Predict and store
comppred <- predict(compmeta, metadf, vcov = T)
compcoef <- sapply(comppred, "[[", "fit") |> t()
compvcov <- sapply(comppred, function(x) vechMat(x$vcov)) |> t()
names(compcoef) <- coefvars
colnames(compvcov) <- grep("vcov", names(stage1res), value = T)
compres <- cbind(metadf[, c("city_code", "agegroup")], compcoef, compvcov)

#----- NULL model

# Refit model
nullform <- sprintf("cbind(%s) ~ 1",
  paste(coefvars, collapse = ", "))
nullmeta <- mixmeta(as.formula(nullform), random = ranform,
  data = metadf, S = smat, subset = conv)

# Predict and store
nullpred <- predict(nullmeta, metadf, vcov = T)
nullcoef <- sapply(nullpred, "[[", "fit") |> t()
nullvcov <- sapply(nullpred, function(x) vechMat(x$vcov)) |> t()
names(nullcoef) <- coefvars
colnames(nullvcov) <- grep("vcov", names(stage1res), value = T)
nullres <- cbind(metadf[, c("city_code", "agegroup")], nullcoef, nullvcov)