Medical Stat Tool

1. Data

📁 Data Management

Import datasets (CSV, Excel), view summaries, and cleaning.

2. Table 1

📋 Table 1 & Matching

Generate baseline tables and perform propensity score matching.

3. General Stats

📊 General Statistics

Diagnostic tests, Correlation, Agreement (Kappa, Bland-Altman).

4. Modeling

🔬 Advanced Modeling

Regression (Linear, Logistic, Firth), Survival, Advanced Inference.

5. Clinical

🏥 Clinical Tools

Sample Size Calculator, Causal Inference methods.

6. Settings

⚙️ Settings

Configure application preferences and defaults.

🛠️ Data Management Tools

📝 Metadata & Type

Select Variable:

Categorical Mapping: Format as `0=Control`.

🔍 Missing Data

Missing Values:

🧩 Impute Missing Data

Method:

Columns:

📈 Outlier Handling

Columns:

Method:

Threshold:

Action:

🗺️ Missing Data Pattern

📊 Multiple Imputation with Rubin's Rules — Generate m imputed datasets and pool estimates

⚙️ MI Settings

Columns to Impute:

Imputations (m):

Iterations:

💡 Tip: Use m ≥ 5 imputations. More is better for high % missing data.

📈 MI Status

🔍 Imputation Diagnostics

Variable:

Transformation:

📊 Assumption Check

Transformation Preview

🛠️ Variable Config

Metadata: Define variable types (Categorical vs Continuous).
Missing Data: standardized coding (e.g., -99, NaN) ensures accurate analysis.

🧹 Cleaning & Imputation

Mean/Median: Simple, fast, but reduces variance. Use for low missingness (<5%).
KNN (K-Nearest Neighbors): Imputes based on similar rows. Preserves local structure better.
MICE (Multivariate Imputation): Models each variable using others. Best for complex datasets with random missingness (MAR).

📈 Outlier Handling

IQR (Interquartile Range): Robust method. Flags points < Q1-1.5IQR or > Q3+1.5IQR.
Z-Score: Parametric. Flags points > 3 SD from mean. Assumes normality.
Actions:
- Winsorize: Cap values at the thresholds (preserves sample size).
- Remove: Delete values (creates missingness).

⚡ Transformation

Log: Reduces right-skewness (e.g., income, CRP levels). Handles x > 0.
Sqrt: Moderate skew reduction. Handles x >= 0.
Z-Score: Standardizes to Mean=0, SD=1. Essential for algorithms sensitive to scale (e.g., KNN, Clustering).

📄 Data Preview

⚙️ Data Controls

📂 Upload CSV/Excel

Browse...

📊 Table 1 Options

Configuration

Group By (Column)

OR Style:

All Levels (Every Level vs Ref)

Simple (Single Line/Risk vs Ref)

Variables

Include Variables

📥 Download HTML

Table 1 Results

⚖️ PSM Configuration

💡 Need ATE directly? Use Clinical Tools → Causal Methods for Inverse Probability Weighting (keeps all data, no matching).

1. Select Variables

Treatment Variable (Binary)

Outcome Variable

Confounding Variables

2. Quick Presets

🔧 Custom (Manual)

👥 Demographics

🏥 Full Medical

⚙️ 3. Matching Settings

Caliper Width

Matching Results

✅ Matched Data Actions

Export Options:

📥 CSV Format
📥 Excel Format

Filter & Display:

Rows to display:

Compare Variable:

Reset:

📊 Summary Statistics

🔍 Data Preview

📈 Statistics by Group

📊 Descriptive Stats
📉 Visualization

🔢 Sample Size & Power Calculator

Means (T-test) Setup

Group 1

Mean Group 1:

SD Group 1:

Group 2

Mean Group 2:

SD Group 2:

⚙️ Statistical Parameters (Power, Alpha)

Power (1-β):

Alpha (Sig. Level):

Allocation Ratio (N2/N1):

Results

Proportions Setup

Expected Proportions

Proportion 1 (0-1):

Proportion 2 (0-1):

⚙️ Statistical Parameters

Power (1-β):

Alpha (Sig. Level):

Allocation Ratio (N2/N1):

Results

Survival Analysis Setup

Input Mode

Input Mode:

Hazard Ratio

Median Survival Time

Parameters

Hazard Ratio (HR):

Median Survival 1:

Median Survival 2:

Power (1-β):

Alpha (Sig. Level):

Ratio (N2/N1):

Results

Note: This calculates required number of EVENTS, not total subjects.

Correlation Setup

Parameters

Expected Correlation (r):

Power (1-β):

Alpha (Sig. Level):

Results

📚 Reference & Interpretation Guide

💡 Tip: This section provides detailed explanations and interpretation rules for Table 1 and Propensity Score Matching.

🚦 Quick Decision Guide

Question	Recommended Action	Goal
Do my groups differ at baseline?	Generate Table 1 (Subtab 1)	Check for significant p-values (< 0.05).
My groups are imbalanced. Can I fix?	Run PSM (Subtab 2)	Create a "synthetic" RCT where groups are balanced.
Did the matching work?	Check SMD (Subtab 2 - Results)	Look for SMD < 0.1 in the Love Plot.
What do I do with matched data?	Export / Use Matched Data	Go to Subtab 3 to export, or select "✅ Matched Data" in other analysis tabs.

📊 Baseline Characteristics (Table 1)

Concept: A standard table in medical research that compares the demographic and clinical characteristics of two or more groups (e.g., Treatment vs Placebo).

Interpretation:

P-value: Tests if there is a statistically significant difference between groups.
p < 0.05: Significant difference (Imbalance) ⚠️. This suggests confounding may be present.
p ≥ 0.05: No significant difference (Balanced) ✅.

Reporting Standards:

Numeric Data (Normal): Report Mean ± SD. (e.g., Age: 45.2 ± 10.1)
Numeric Data (Skewed): Report Median (IQR). (e.g., LOS: 5 (3-10))
Categorical Data: Report Count (%). (e.g., Male: 50 (45%))

⚖️ Propensity Score Matching (PSM)

Concept: A statistical technique used in observational studies to reduce selection bias. It pairs patients in the treated group with patients in the control group who have similar "propensity scores" (probability of receiving treatment).

Key Metric: Standardized Mean Difference (SMD):

The gold standard for checking balance after matching.
SMD < 0.1: Excellent Balance ✅ (Groups are comparable).
SMD 0.1 - 0.2: Acceptable.
SMD > 0.2: Imbalanced ❌.

Caliper (Tolerance):

Determines how "close" a match must be.
Stricter (0.1×SD): Better balance, but you might lose more patients (fewer matches).
Looser (0.5×SD): More matches, but balance might be worse.

📝 Common Workflow

Check Original Data: Run Table 1 on the "Original Data". Note any variables with p < 0.05.
Match: Go to Subtab 2, select Treatment, Outcome, and all confounding variables (especially those with p < 0.05).
Verify: After matching, check the Love Plot. Ensure all dots (Matched) are within the < 0.1 zone.
Re-check Table 1: Go back to Subtab 1, switch the dataset selector to "✅ Matched Data", and generate Table 1 again. P-values should now be non-significant (or SMDs low).

ROC Curve Analysis

Analysis Mode:

Single Test Analysis Compare Two Tests (Paired)

📥 Download Report

Chi-Square & Risk Analysis (2x2 Contingency Table)

📥 Download Report

Descriptive Statistics

📥 Download Report

Decision Curve Analysis

📥 Download Report

📚 Reference & Interpretation Guide

💡 Tip: This section provides detailed explanations and interpretation rules for all the diagnostic tests.

🚦 Quick Decision Guide

Question	Recommended Test	Example
My test is a score (e.g., 0-100) and I want to see how well it predicts a disease (Yes/No)?	ROC Curve & AUC	Risk Score vs Diabetes
I want to find the best cut-off value for my test score?	ROC Curve (Youden Index)	Finding optimal BP for Hypertension
Are these two groups (e.g., Treatment vs Control) different in outcome (Cured vs Not Cured)?	Chi-Square	Drug A vs Placebo on Recovery
Is my model clinically useful at a specific threshold?	Decision Curve (DCA)	Should we biopsy everyone with PSA > 4?
(For Agreement/Kappa, see "Agreement" tab)

⚖️ Interpretation Guidelines

ROC Curve & AUC

Single Test: Detailed analysis of one diagnostic test with threshold optimization.
Compare Tests: Uses Paired DeLong's Test to statistically compare two ROC curves properly.
AUC > 0.9: Excellent discrimination
AUC 0.8-0.9: Good discrimination
AUC 0.7-0.8: Fair discrimination
AUC 0.5-0.7: Poor discrimination
AUC = 0.5: No discrimination (random chance)
Youden J Index: Sensitivity + Specificity - 1 (higher is better, max = 1)

Comparison Interpretation (DeLong)

P-value < 0.05: Significant difference between the two ROC curves.
Z-score: Strength of the difference.

Chi-Square Test

P < 0.05: Statistically significant association
Odds Ratio (OR): If 95% CI doesn't include 1.0, it's significant
Risk Ratio (RR): Similar interpretation as OR
Use Fisher's Exact Test when expected counts < 5

Decision Curve Analysis (DCA)

Net Benefit: The benefit of treating true positives minus the harm of treating false positives.
Interpretation: The model is useful if the model's curve (Red) is higher than both:
- Treat All (Gray line): Treating everyone assuming they have the disease.
- Treat None (Horizontal line at 0): Treating no one.
Threshold Probability: The patient's/doctor's preference (e.g., how worried are they about missing a case vs unnecessary treatment?).

Descriptive Statistics

Mean: Average value (affected by outliers)
Median: Middle value (robust to outliers)
SD (Standard Deviation): Spread of data around mean
Q1/Q3: 25th and 75th percentiles

📈 Continuous Correlation Analysis

Correlation Coefficient:

Variable 1 (X-axis):

Variable 2 (Y-axis):

📥 Download Report

📊 Correlation Matrix & Heatmap

Select Variables (Multi-select):

Correlation Method:

📥 Download Report

📚 Reference & Interpretation Guide

📈 Correlation (Relationship)

Concept: Measures the strength and direction of the relationship between two continuous variables.

Warning: Correlation does not imply causation. A strong relationship between two variables does not mean one causes the other.

1. Pearson (r):

Best for: Linear relationships (straight line), normally distributed data.
Sensitive to: Outliers.
Returns: R-squared (R²) = proportion of variance explained

2. Spearman (rho) & Kendall (tau):

Best for: Monotonic relationships, non-normal data, or ranks.
Robust to: Outliers.
Kendall's Tau is often preferred for small datasets with many tied ranks.

Interpretation of Coefficient (r, rho, or tau):

+1.0: Perfect Positive (As X goes up, Y goes up).
-1.0: Perfect Negative (As X goes up, Y goes down).
0.0: No relationship.

Strength Guidelines:

0.9 - 1.0: Very Strong 🔥
0.7 - 0.9: Strong 📈
0.5 - 0.7: Moderate 📊
0.3 - 0.5: Weak 📉
< 0.3: Very Weak/Negligible

Confidence Intervals (95% CI):

Shows the range where the true correlation likely falls
Wider CI = less precise estimate (usually with small samples)

💡 Common Questions

Q: What is R-squared (R²)?

A: R² tells you the proportion of variance in Y that is explained by X. For example, R² = 0.64 means 64% of the variation in Y is explained by X.

Q: Why use ICC instead of Pearson for reliability?

A: Pearson only measures linearity. If Rater A always gives exactly 10 points higher than Rater B, Pearson = 1.0 but they don't agree! ICC accounts for this.

Q: What if p-value is significant but r is low (0.1)?

A: P-value means it's likely not zero. With large samples, tiny correlations can be "significant". Focus on r-value magnitude for clinical relevance.

Q: How to interpret confidence intervals?

A: If 95% CI includes 0, the correlation is not statistically significant. Narrow CI = more precise estimate, Wide CI = less precise (need more data).

Q: How many variables do I need for ICC?

A: At least 2 (to compare two raters/methods). More raters = more reliable ICC.

Categorical Agreement (Kappa)

Method:

Cohen's Kappa (2 Raters) Fleiss' Kappa (>2 Raters)

📥 Download Report

Bland-Altman Analysis (Continuous Data Comparison)

Show CI Bands

Confidence Level:

📥 Download Report

Intraclass Correlation Coefficient (ICC)

📥 Download Report

📚 Agreement & Reliability Reference Guide

🤝 Cohen's Kappa & Fleiss' Kappa

Cohen's Kappa: For agreement between two raters.
Fleiss' Kappa: For agreement between three or more raters.

Landis–Koch (1977) scale:

> 0.81: Almost perfect agreement ✅
0.61–0.80: Substantial agreement
0.41–0.60: Moderate agreement
0.21–0.40: Fair agreement ⚠️
< 0.20: Slight/Poor agreement ❌

📉 Bland-Altman Plot

Used for continuous data to compare two measurement methods.

Bias (Mean Difference): Systematic difference.
Limits of Agreement (LoA): Interval containing 95% of differences.
Confidence Intervals (Shaded): Shows the precision of the Bias and LoA estimates.

🔍 Intraclass Correlation (ICC)

Measures reliability/consistency.

ICC Forms (Shrout & Fleiss, 1979):

ICC1: One-way random effects (raters selected at random).
ICC2: Two-way random effects (raters and subjects random).
ICC3: Two-way mixed effects (fixed raters).

Interpretation (Cicchetti, 1994):

> 0.75: Excellent 🌟
0.60 – 0.75: Good
0.40 – 0.60: Fair ⚠️
< 0.40: Poor ❌

Kaplan-Meier & Nelson-Aalen Curves

Variable Selection

Time Variable

Event Variable (1=Event)

Compare Groups

Plot Settings

Select Plot Type:

Kaplan-Meier (Survival Function) Nelson-Aalen (Cumulative Hazard)

Survival Probability at (times)

📥 Download Report

Analysis Results

Landmark Analysis for Late Endpoints

ℹ️ Principle: Landmark analysis is useful when the treatment effect is delayed (e.g., immune-oncology) or violates proportional hazards initially.

How it works:

Select a "Landmark Time" (t).
Patients who died/censored before t are excluded.
Analysis is performed only on patients who survived to time t, resetting their "start" time to t.

Settings

Landmark Time (t)

Compare Group

📥 Download Report

Landmark Analysis Results

Cox Proportional Hazards Regression

Model Configuration

Fitting Method

Select Covariates (Predictors)

📥 Download Report

Cox Model Results

Cox Subgroup Analysis - Treatment Heterogeneity

Variables

Follow-up Time

Event Indicator

Treatment/Exposure

Stratification & Adjustment

Stratify By

Adjustment Variables

⚠️ Advanced Settings

⚙️ Minimum Counts

Min N per subgroup:

Min events per subgroup:

📥 Download Report

Subgroup Analysis Results

Restricted Cubic Spline (RCS) Analysis

ℹ️ Purpose: Visualize non-linear relationships between a continuous variable and the hazard ratio (HR). Useful when the risk does not increase linearly (e.g., U-shaped or J-shaped curves).

Variables

Continuous Variable

Time Variable:

Event Variable:

Settings

Number of Knots

Reference Value

Adjustment Covariates:

📥 Download Report

Select data format and variables to run the model.

📄 Data Format Selection

Select your data structure below. (See 'Reference' tab for detailed format specifications)

📊 Long Format (Multiple rows per patient - Recommended)

📈 Wide Format (One row per patient)

⚙️ Column Configuration

Columns for Time-Varying Cox analysis:

Identification & Time

🆔 Patient ID Column:

⏱️ Interval Start Time:

⏱️ Interval Stop Time:

🚫 Event Indicator (0=censored, 1=event):

Covariates

Time-Varying Covariates: Columns that can change over time within intervals (e.g., treatment status, lab values, symptoms)

Select Time-Varying Covariates:

Static Covariates: Columns constant within each patient (e.g., age at baseline, sex, initial diagnosis)

Select Static Covariates:

📥 Download Report

Advanced settings, risk intervals, and data inspection.

📊 Risk Intervals Definition (Wide Format Only)

Risk intervals divide the follow-up period into time windows. For example: [0, 1m, 3m, 6m, 12m] creates 4 intervals.

Last-observed values of time-varying covariates are "carried forward" within intervals.

How to define intervals:

🤖 Auto-detect (from TVC column names like 'tvc_3m', 'tvc_6m')

📈 Quantile-based (equal number of events in each interval)

✍️ Manual (specify time points below)

Preview: Intervals will appear here

🔧 Model Configuration

🤖 Fitting Method

Select algorithm:

Auto (CoxTimeVaryingFitter) CoxTimeVaryingFitter (recommended)

⚠️ Advanced Options

Ridge Penalizer (L2 regularization, 0=none):

0.0: Standard Cox (no regularization)
0.1-1.0: Small penalty (useful if unstable)
>1.0: Strong regularization (shrinks coefficients)

🛥️ Data Preview (First 50 Rows)

Data Structure Summary:

Note: Only first 50 rows displayed. Full dataset is used for analysis.

ℹ️ Information & FAQ

🆔 What are time-varying covariates?

Time-varying covariates are variables that can change their values during follow-up.

Examples:

Drug dosage (adjusted over time)
Lab values (measured periodically)
Treatment status (switched during study)

Unlike standard Cox regression (assumes constant covariates), TVC Cox allows modeling of dynamic effects.

👀 When to use Long vs Wide format?

Aspect	Long Format	Wide Format
Rows/Patient	Multiple	One
Structure	Interval-based	Measurement-based
Best for	Complex follow-up schedules	Regular measurements
Preparation	Ready to use	Requires transformation
Example	Medical visit notes	Quarterly labs

📊 How are intervals created?

Three methods to define intervals:

Auto-detect (Recommended):
- Extracts time points from column names
- E.g., 'tvc_3m' → extracts time=3
- Creates intervals: [0-3m], [3m-6m], [6m-12m], etc.
Quantile-based:
- Divides data into roughly equal event frequencies
- Ensures sufficient data per interval
Manual:
- User specifies exact time points
- E.g., "0, 1, 3, 6, 12, 24"
- Maximum control but requires planning

🛠️ Data transformation example

Input (Wide Format):

ID | time | event | tvc_0m | tvc_3m | tvc_6m
1  | 12   | 1     | 100    | 110    | 120
2  | 6    | 0     | 95     | 98     | NA

Output (Long Format):

ID | start | stop | event | tvc
1  | 0     | 3    | 0     | 100
1  | 3     | 6    | 0     | 110
1  | 6     | 12   | 1     | 120
2  | 0     | 3    | 0     | 95
2  | 3     | 6    | 0     | 98

Key Points:

Event=1 only in final interval
Covariate values carried forward (last observed)
Each patient split into multiple rows

⏱️ Time-Varying Cox Reference

1. What is Time-Varying Cox?

Standard Cox regression assumes predictors (like treatment) are constant over time. Time-Varying Cox allows values to change.

2. Data Formats

A. Long Format (Recommended)

Structure: Multiple rows per patient. Each row represents a specific time interval.
Columns: patient_id, start_time, stop_time, event, and [covariates].
Status: Ready for direct analysis.
Example: Patient A (0-6 months, Drug=0), Patient A (6-12 months, Drug=1).

B. Wide Format

Structure: One row per patient with multiple columns for covariate measurements at different times.
Columns: patient_id, followup_time, event, tvc_baseline, tvc_3m, tvc_6m, etc.
Status: Requires transformation (we convert it to Long Format using "Risk Intervals").

3. Interpretation

Hazard Ratio (HR): Represents the instantaneous risk.
HR = 1.5: At any given moment, having the condition (or 1 unit higher value) increases the risk of the event by 50% compared to not having it at that same moment.

4. Assumptions

Proportional Hazards: Still applies! The effect of the variable (HR) is assumed constant over time, even if the variable's value changes.

Analysis Results

📚 Quick Reference: Survival Analysis

🎲 When to Use What:

Method	Purpose	Output
KM Curves	Visualize time-to-event by group	Survival %, median, p-value
Nelson-Aalen	Cumulative hazard over time	H(t) curve, risk accumulation
Landmark	Late/surrogate endpoints	Filtered KM, immortal time removed
Cox	Multiple predictors of survival	HR, CI, p-value per variable + forest plot
Subgroup Analysis	Treatment effect heterogeneity	HR by subgroup, interaction test
Time-Varying Cox	Time-dependent covariates	Dynamic HR, interval-based risk

📝 Detailed Interpretation:

1. Hazard Ratios (HR)

The main measure of effect in survival analysis.

HR = 1: No difference in risk between groups.
HR > 1: Increased risk of event (e.g., HR 1.5 = 50% higher risk). Hazardous / Bad Outcome.
HR < 1: Reduced risk of event (e.g., HR 0.7 = 30% lower risk). Protective / Good Outcome.

2. Log-Rank Test

Used in Kaplan-Meier analysis to compare survival curves.

P < 0.05: There is a statistically significant difference between the survival curves of the groups.

3. Landmark Analysis

Addresses Immortal Time Bias or violation of Proportional Hazards.

By selecting a "Landmark Time" (e.g., 6 months), you exclude patients who died/censored before 6 months.
You then compare survival given that the patient has already survived to 6 months.
Note: This reduces your sample size but provides a fairer comparison for late-acting treatments.

🔍 Advanced Inference

Mediation Analysis

Variables

Outcome (Y):

Treatment (X):

Mediator (M):

Confounders (C):

Results

Collinearity Diagnostic

Variables

Predictors:

VIF Results

Model Diagnostics (OLS)

Model Specification

Outcome (Y):

Main Predictor (X):

Covariates:

Diagnostic Results

Meta-Analysis Data Entry

Enter comma-separated values for effect sizes and variances.

Data Inputs

Effect Sizes (e.g., 0.5, 0.4, 0.6):

Variances (e.g., 0.01, 0.02, 0.015):

Results

Advanced Inference Reference

Mediation Analysis:

ACME (Indirect Effect): The portion of the effect mediated by M. (Effect of X on Y via M).
ADE (Direct Effect): The effect of X on Y, keeping M constant.
Total Effect: ACME + ADE.

Collinearity Diagnostics:

VIF (Variance Inflation Factor):
- VIF > 5: Moderate multicollinearity (Caution).
- VIF > 10: Severe multicollinearity (Consider removing variable).
Tolerance: 1/VIF. Values < 0.1 indicate problems.

Model Diagnostics (OLS):

Residuals vs Fitted: Checks linearity. Ideally, points fluctuate randomly around 0 (horizontal line).
Q-Q Plot: Checks normality of residuals. Points should fall along the diagonal line.
Cook's Distance: Measures influence. Points with Cook's D > 4/n (or > 1) are highly influential and may skew results.

Heterogeneity (Meta-Analysis):

I-squared (I²):
- < 25%: Low heterogeneity.
- 25-75%: Moderate heterogeneity.
- > 75%: High heterogeneity.
Q-Statistic P-value:
- P < 0.05: Significant heterogeneity exists.

🔢 Sample Size & Power Calculator

Means (T-test) Setup

Group 1

Mean Group 1:

SD Group 1:

Group 2

Mean Group 2:

SD Group 2:

⚙️ Statistical Parameters (Power, Alpha)

Power (1-β):

Alpha (Sig. Level):

Allocation Ratio (N2/N1):

Results

Proportions Setup

Expected Proportions

Proportion 1 (0-1):

Proportion 2 (0-1):

⚙️ Statistical Parameters

Power (1-β):

Alpha (Sig. Level):

Allocation Ratio (N2/N1):

Results

Survival Analysis Setup

Input Mode

Input Mode:

Hazard Ratio

Median Survival Time

Parameters

Hazard Ratio (HR):

Median Survival 1:

Median Survival 2:

Power (1-β):

Alpha (Sig. Level):

Ratio (N2/N1):

Results

Note: This calculates required number of EVENTS, not total subjects.

Correlation Setup

Parameters

Expected Correlation (r):

Power (1-β):

Alpha (Sig. Level):

Results

🎯 Causal Inference

💡 Need matched dataset? Use Clinical → Table 1 & Matching to create balanced paired data first.

IPW & Balance Results

Model Specification

Treatment Variable (Binary)

Outcome Variable

Covariates for Matching

Truncate Extreme Weights (1%/99%)

Stratified Results

Variables

Treatment/Exposure (Binary)

Outcome (Binary)

Stratification Variable

E-Value Results

Love Plot (Covariate Balance)

Common Support (Overlap)

Causal Inference Methods

1. Propensity Score Methods (IPW/IPTW)

Goal: Estimate the Average Treatment Effect (ATE) by adjusting for confounding.
Mechanism: Observations are weighted by the Inverse Probability of Treatment Weighting (IPTW).
Diagnostics: Check Standardized Mean Differences (SMD). Ideally, all SMD < 0.1 after weighting.

2. Stratified Analysis (Mantel-Haenszel)

Goal: adjust for confounding by a categorical variable (stratum).
Mantel-Haenszel OR: A weighted average of stratum-specific odds ratios.
Homogeneity Test: Checks if the effect of treatment is consistent across all strata.

3. Sensitivity Analysis (E-Value)

Goal: Assess robustness to unmeasured confounding.
E-Value: The strength of association an unmeasured confounder must have to explain away the observed effect. Larger E-values imply more robust results.

📁 Data Management

📋 Table 1 & Matching

📊 General Statistics

🔬 Advanced Modeling

🏥 Clinical Tools

⚙️ Settings

🔍 Missing Data

📊 Assumption Check

Transformation Preview

🛠️ Variable Config

🧹 Cleaning & Imputation

📈 Outlier Handling

⚡ Transformation

Configuration

Variables

Table 1 Results

1. Select Variables

2. Quick Presets

Matching Results

🔢 Sample Size & Power Calculator

Group 1

Group 2

Results

Expected Proportions

Results

Input Mode

Parameters

Results

Parameters

Results

📚 Reference & Interpretation Guide

🚦 Quick Decision Guide

📝 Common Workflow

ROC Curve Analysis

Chi-Square & Risk Analysis (2x2 Contingency Table)

Descriptive Statistics

Decision Curve Analysis

📚 Reference & Interpretation Guide

🚦 Quick Decision Guide

⚖️ Interpretation Guidelines

ROC Curve & AUC

Comparison Interpretation (DeLong)

Chi-Square Test

Decision Curve Analysis (DCA)

Descriptive Statistics

Categorical Agreement (Kappa)

Bland-Altman Analysis (Continuous Data Comparison)

Intraclass Correlation Coefficient (ICC)

📚 Agreement & Reliability Reference Guide

🤝 Cohen's Kappa & Fleiss' Kappa

📉 Bland-Altman Plot

🔍 Intraclass Correlation (ICC)

Variable Selection

Method & Settings

Advanced Adjustments

🔗 Interaction Pairs:

Analysis Results

Variables

Stratification & Adjustment

Subgroup Analysis Results

Variable Selection

Model Refinement

🔗 Interaction Pairs:

Poisson Results

Variable Selection

Model Refinement

🔗 Interaction Pairs:

Negative Binomial Results

Variable Selection

Distribution & Link:

Predictors

GLM Results

Variable Selection

Method & Settings

Ad Hoc Exclusions

Variable Selection

Model Settings

📚 Core Regression Reference Guide

1. 📈 Binary Outcomes (Logistic Regression)

2. 📉 Continuous Outcomes (Linear Regression)