snu-dhc 0.1.8

SNU_DHC Package

snu_dhc.tables

DemoTable generates baseline characteristic tables for medical studies and exports them to Excel. It supports continuous and categorical variables, grouping, and automatic p-value calculation.

Parameters:

• df: Input pandas DataFrame.
• table_name: Title for the Excel sheet.
• variables: List of dictionaries defining the variables to display.
• group_variable: Optional column to group comparisons.
• group_labels: Display labels for group values.
• show_total: Whether to show the total column.
• show_missing: Whether to show missing counts for categorical variables.
• percent_decimals: Decimal places for percentages.
• thousands_sep: Use comma separator for each thousand numbers.
• show_p_values: Whether to include a p-value column.
• p_value_decimals: Fixed or automatic formatting for p-values.

Test selection logic (for p-values):

• Continuous (mean):
  • 2 groups: Welch's t-test (with checks and warnings for normality)
  • 3+ groups: ANOVA (with checks and warnings for normality and variance)
• Continuous (median):
  • 2 groups: Mann-Whitney U test
  • 3+ groups: Kruskal-Wallis test
• Categorical:
  • Chi-square test by default
  • Fisher's exact test if 2x2 and <5 cell counts

Usage Example

from snu_dhc.tables import DemoTable

variables_config = [
    {"var": "age", "name": "Age", "type": "continuous", "stat": "median", "decimals": 0},
    {"var": "sex", "name": "Sex", "type": "categorical", "class_labels": {1: "Male", 0: "Female"}},
    {"var": "dm", "name": "Diabetes mellitus", "type": "categorical", "class_labels": {1: ""}},
    {"var": "init_rhythm", "name": "Initial rhythm", "type": "categorical", "class_labels": {1: "VF/VT", 2: "PEA", 3: "Asystole"}},
    {"var": "rti", "name": "RTI", "type": "continuous", "stat": "median", "decimals": 0},
]

table = DemoTable(
    df=ohca_group,
    table_name="Table 1. Baseline characteristics of study patients",
    variables=variables_config,
    group_variable="group",
    group_labels={"train": "Train", "val": "Validation", "test": "Test"}, 
    show_total=True,
    show_missing=True,
    percent_decimals=1,
    thousands_sep=True,
    show_p_values=True,
    p_value_decimals="auto"
)

table.save("table1.xlsx")

snu_dhc.statistics

This module includes tools to evaluate binary classifiers with confidence intervals and visualization utilities.

binary_outcome_analysis(true, **probs)

• true: Array-like of true binary labels.
• **probs: Dictionary of model names and their predicted probabilities. Methods:
  • .auc(): Print AUC and 95% CI for each model.
  • .roc_auc(**styles): Plot ROC curves and display AUC values. Customize plot style per model.
  • .classification(**cutoffs): Print classification metrics using thresholds like fixed float, 'youden', 'sen90', 'spe80', etc.

Example: binary_outcome_analysis

from snu_dhc.statistics import binary_outcome_analysis

probs = {
    'XGBoost': prob0,
    'Random Forest': prob1,
    'Logistic Regression': prob2
}
result = binary_outcome_analysis(true, **probs)

# Plot ROC curves
styles = {
    'XGBoost': {'color': 'C0', 'linestyle': '-'},
    'Random Forest': {'color': 'C1', 'linestyle': '-.'},
    'Logistic Regression': {'color': 'C2', 'linestyle': '--'}
}
result.roc_auc(**styles)

# Show classification metrics with thresholds
cutoffs = {
    'XGBoost': 0.5,
    'Random Forest': 'sen90',
    'Logistic Regression': 'spe80'
}
result.classification(**cutoffs)

auc_score(true, prob, method='delong', alpha=0.95, n_iterations=1000, show=False)

• Get AUC and 95% CI using either DeLong's method or bootstrap. method can be 'delong' or 'bootstrap'.

Returns: [auc, [ci_lower, ci_upper]]

classification_metrics(true, pred, method='wilson', ...)

• Returns sensitivity, specificity, PPV, NPV, and F1-score with confidence intervals. method can be 'wilson' or 'bootstrap'.

Returns: Dictionary like {'sen': [...], 'spe': [...], 'ppv': [...], 'npv': [...], 'f1': [...]}

youden_metrics(true, prob, ...)

• Computes metrics using Youden's index as the threshold.

Returns: (metrics_dict, youden_threshold)

bootstrap_ci(true, pred, fx, n_iterations=1000, alpha=0.95)

• Computes confidence interval for any function fx using bootstrap.

Returns: [estimate, [ci_lower, ci_upper]]

overview_df(df)

• Provides column-wise overview of a DataFrame.

Returns: DataFrame with columns for dtype, #unique, #missing, #zeros, etc.

multiclass_bootstrap_metrics(true, pred, n_iter=1000, alpha=0.95)

• Computes sensitivity, specificity, PPV, NPV, and F1-score with 95% confidence intervals for each class using bootstrapping.

• true: Array-like true labels.

• pred: Array-like predicted labels (e.g., argmax of softmax outputs).

• n_iter: Number of bootstrap iterations (default=1000).

• alpha: Confidence level (default=0.95).

Returns: Dictionary of the form:

{
  0: {'sen': [...], 'spe': [...], 'ppv': [...], 'npv': [...], 'f1': [...]},
  1: {...},
  ...
}