08 · Explorative Datenanalyse und Datenvisualisierung

import pandas as pd
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt

labs   = pd.read_csv("https://schradern.github.io/data-science-coach/data/labor.csv")

df = pd.read_csv("https://schradern.github.io/data-science-coach/data/kohorte.csv").merge(labs, on="patient_id", how="left")

library(tidyverse)

cohort <- read_csv("https://schradern.github.io/data-science-coach/data/kohorte.csv", show_col_types = FALSE)
labs   <- read_csv("https://schradern.github.io/data-science-coach/data/labor.csv", show_col_types = FALSE)

df <- left_join(cohort, labs, by = "patient_id")

numeric_cols = ["alter", "bmi", "sofa_score", "crp_mg_l",
                "verweildauer_tage", "laktat_mmol_l", "kreatinin_mg_dl"]
print(df[numeric_cols].describe().round(2))

# Missing values separately — describe() hides them
print(df[numeric_cols].isna().sum())

# Frequencies for categorical variables
print(df["aufnahmegrund"].value_counts())

numeric_cols <- c("alter", "bmi", "sofa_score", "crp_mg_l",
                  "verweildauer_tage", "laktat_mmol_l", "kreatinin_mg_dl")
summary(df[numeric_cols])
colSums(is.na(df))
df |> count(aufnahmegrund, sort = TRUE)

series = df["crp_mg_l"].dropna()
q1, q3 = series.quantile(0.25), series.quantile(0.75)
iqr = q3 - q1
lower, upper = q1 - 1.5 * iqr, q3 + 1.5 * iqr
n_out = ((series < lower) | (series > upper)).sum()
print(f"IQR={iqr:.1f}  Grenzen=[{lower:.1f}, {upper:.1f}]  Ausreißer: {n_out}")

vals <- df$crp_mg_l
q1 <- quantile(vals, 0.25, na.rm = TRUE)
q3 <- quantile(vals, 0.75, na.rm = TRUE)
iqr <- q3 - q1
sum(vals < q1 - 1.5*iqr | vals > q3 + 1.5*iqr, na.rm = TRUE)

relevant = ["alter", "sofa_score", "crp_mg_l", "laktat_mmol_l",
            "verweildauer_tage", "verstorben_30d"]
print(df[relevant].corr().round(2))

relevant <- c("alter", "sofa_score", "crp_mg_l", "laktat_mmol_l",
              "verweildauer_tage", "verstorben_30d")
round(cor(df[relevant], use = "pairwise.complete.obs"), 2)

fig, ax = plt.subplots(figsize=(7, 4))
for outcome, label, color in [
    (0, "Überlebt", "#2A5C8A"),
    (1, "Verstorben", "#B5482E"),
]:
    subset = df.loc[df["verstorben_30d"] == outcome, "alter"]
    ax.hist(subset, bins=20, alpha=0.65, label=label, color=color, edgecolor="white")
ax.set_xlabel("Alter (Jahre)")
ax.set_ylabel("Anzahl Patient:innen")
ax.set_title("Altersverteilung nach 30-Tage-Mortalität")
ax.legend()
fig.tight_layout()
fig.savefig("verteilung_alter.png", dpi=120)   # ohne savefig()/show() bleibt die Grafik unsichtbar!

library(ggplot2)
p_hist <- ggplot(df, aes(x = alter,
               fill = factor(verstorben_30d, labels = c("Überlebt", "Verstorben")))) +
  geom_histogram(bins = 20, alpha = 0.65, position = "identity", colour = "white") +
  scale_fill_manual(values = c("Überlebt" = "#2A5C8A", "Verstorben" = "#B5482E")) +
  labs(x = "Alter (Jahre)", y = "Anzahl Patient:innen",
       title = "Altersverteilung nach 30-Tage-Mortalität", fill = NULL) +
  theme_minimal()
ggsave("verteilung_alter_r.png", p_hist, width = 7, height = 4, dpi = 120)

import seaborn as sns

order = (
    df.groupby("aufnahmegrund")["laktat_mmol_l"]
      .median()
      .sort_values(ascending=False)
      .index.tolist()
)
fig, ax = plt.subplots(figsize=(8, 4))
sns.boxplot(data=df, x="aufnahmegrund", y="laktat_mmol_l",
            order=order, palette="Set2", ax=ax)
ax.set_xlabel("Aufnahmegrund")
ax.set_ylabel("Laktat (mmol/l)")
ax.set_title("Laktat nach Aufnahmegrund")
fig.tight_layout()
fig.savefig("verteilung_laktat_nach_grund.png", dpi=120)

order_lvl <- df |>
  group_by(aufnahmegrund) |>
  summarise(m = median(laktat_mmol_l, na.rm = TRUE)) |>
  arrange(desc(m)) |>
  pull(aufnahmegrund)

p_box <- ggplot(df |> filter(!is.na(laktat_mmol_l)),
       aes(x = factor(aufnahmegrund, levels = order_lvl),
           y = laktat_mmol_l, fill = aufnahmegrund)) +
  geom_boxplot(outlier.size = 1.5, outlier.alpha = 0.5, show.legend = FALSE) +
  scale_fill_brewer(palette = "Set2") +
  labs(x = "Aufnahmegrund", y = "Laktat (mmol/l)",
       title = "Laktat nach Aufnahmegrund") +
  theme_minimal()
ggsave("verteilung_laktat_nach_grund_r.png", p_box, width = 8, height = 4, dpi = 120)

import numpy as np

fig, ax = plt.subplots(figsize=(7, 5))
colors = {0: "#2A5C8A", 1: "#B5482E"}
for outcome, label in [(0, "Überlebt"), (1, "Verstorben")]:
    sub = df[df["verstorben_30d"] == outcome]
    ax.scatter(sub["crp_mg_l"], sub["verweildauer_tage"],
               c=colors[outcome], label=label,
               alpha=0.42, s=22, edgecolors="none")

# Gesamt-Trendgerade (lineare Anpassung über alle Punkte)
mask = df["crp_mg_l"].notna() & df["verweildauer_tage"].notna()
coef = np.polyfit(df.loc[mask, "crp_mg_l"], df.loc[mask, "verweildauer_tage"], 1)
x_range = np.linspace(df.loc[mask, "crp_mg_l"].min(), df.loc[mask, "crp_mg_l"].max(), 200)
ax.plot(x_range, np.poly1d(coef)(x_range), color="#555555",
        linestyle="--", linewidth=1.1, alpha=0.7, label="Trend (gesamt)")

ax.set_xlabel("CRP (mg/l)")
ax.set_ylabel("Verweildauer (Tage)")
ax.set_title("CRP vs. Verweildauer nach 30-Tage-Mortalität")
ax.legend()
fig.tight_layout()
fig.savefig("streu_crp_verweildauer.png", dpi=120)

p_scatter <- ggplot(df, aes(x = crp_mg_l, y = verweildauer_tage)) +
  geom_point(aes(colour = factor(verstorben_30d, labels = c("Überlebt", "Verstorben"))),
             alpha = 0.42, size = 1.5) +
  geom_smooth(method = "lm", formula = y ~ x, se = FALSE,
              colour = "#555555", linetype = "dashed", linewidth = 0.6) +
  scale_colour_manual(values = c("Überlebt" = "#2A5C8A", "Verstorben" = "#B5482E")) +
  labs(x = "CRP (mg/l)", y = "Verweildauer (Tage)",
       title = "CRP vs. Verweildauer nach 30-Tage-Mortalität",
       colour = NULL) +
  theme_minimal()
ggsave("streu_crp_verweildauer_r.png", p_scatter, width = 7, height = 5, dpi = 120)