clinipub 0.1.2

Transform clinical dataframes into publication-ready, beautifully styled tables for medical journals and manuscripts.

clinipub

A Python package to transform clinical dataframes into publication-ready, beautifully styled tables for medical journals and manuscripts

clinipub is a modern Python package designed for data scientists in the life sciences and clinical research fields. It bridges the gap between raw data analysis (pandas.DataFrame) and publication-ready medical writing reporting standards (matching NEJM, JAMA, The Lancet, and CDISC/STROBE guidelines).

Unlike existing packages, clinipub offers autonomous statistical decision-making, impeccable typography, and native exports to formats that medical writers can use out of the box without manual reformatting.

---

Key Advantages Over Other Packages

• Autonomous Statistics: Automatically detects variable types and evaluates normality (Shapiro-Wilk) to map the correct parametric or non-parametric test.
• Regulatory-Grade Precision: Automatically falls back to Fisher's Exact test if any contingency cross-tab count drops below 5.

---

Statistical Test Selection Matrix

The framework automatically maps and executes your bivariate comparative analytics using the following clinical logic:

Variable Type	Distribution	Group Count	Test Executed
Continuous	Normal (Parametric)	2 Groups	Independent Welch's $t$-test
Continuous	Normal (Parametric)	3+ Groups	One-way ANOVA
Continuous	Skewed (Non-Parametric)	2 Groups	Mann-Whitney U test
Continuous	Skewed (Non-Parametric)	3+ Groups	Kruskal-Wallis test
Categorical	Any	2x2 with Cell Count < 5	Fisher's Exact test
Categorical	Any	Standard Configurations	Chi-Square ($\chi^2$) Contingency

---

Developer Setup & Installation

This project is managed efficiently using the uv package manager.

Prerequisites

Ensure you have uv installed:

pip install uv

Installation & Environment Setup

Clone the repository and sync the isolated project development container:

git clone https://github.com/arsalananwar11/clinipub.git

cd clinipub
uv sync

Running Unit Tests

We maintain high test coverage for clinical accuracy using pytest:

uv run pytest

Pipeline Flow

1. MissingDataAuditor computes raw missingness and validates data completeness.
2. ClinicalDataAuditor classifies variables and checks continuous normality.
3. BivariateTestSelector chooses tests automatically based on group count and distribution.
4. TableOneAssembler builds a stratified Table 1 with descriptive statistics and p-values.

Quick Usage

import pandas as pd
from clinipub import (
    MissingDataAuditor,
    ClinicalDataAuditor,
    BivariateTestSelector,
    TableOneAssembler,
)

# Load your clinical dataset
df = pd.read_csv("baseline.csv")

# 1. Audit missing data and generate an HTML report
missing_auditor = MissingDataAuditor(df)
missing_df = missing_auditor.calculate_missingness()
html_report = missing_auditor.to_html_report(
    audit_df=missing_df,
    thresholds={"low": 1.0, "mid": 20.0},
)

# 2. Audit variable types and normality
auditor = ClinicalDataAuditor(df)
var_types = auditor.detect_variable_types()
normality = auditor.test_normality(var_types["continuous"])

# 3. Run autonomous bivariate tests
selector = BivariateTestSelector(df, stratify_by="treatment")
continuous_result = selector.test_continuous("age", is_normal=normality["age"])
cat_result = selector.test_categorical("smoker_status")

# 4. Build publication-ready Table 1
assembler = TableOneAssembler(df, stratify_by="treatment")
table1 = assembler.build()

Core API and Arguments

• MissingDataAuditor(data: pd.DataFrame)

  • calculate_missingness()
    • returns a raw DataFrame with missing_count and missing_percentage
  • to_html_report(audit_df: pd.DataFrame = None, thresholds: dict = {'low': 5.0, 'mid': 30.0})
    • returns styled HTML for publication-ready missingness reporting

• ClinicalDataAuditor(data: pd.DataFrame)

  • detect_variable_types(max_categorical_threshold: int = 10)
    • returns {'categorical': [...], 'continuous': [...]}
  • test_normality(continuous_cols: list, alpha: float = 0.05)
    • returns {column_name: bool}

• BivariateTestSelector(data: pd.DataFrame, stratify_by: str)

  • test_continuous(col: str, is_normal: bool)
    • returns {'p_value': float, 'test': str}
  • test_categorical(col: str)
    • returns {'p_value': float, 'test': str}

• TableOneAssembler(data: pd.DataFrame, stratify_by: str, columns: list = None)

  • build()
    • returns a styled pandas.DataFrame with stratified descriptive statistics and p-values

Contributing

Contributions are welcome. Please read CONTRIBUTING.md for guidelines on issues, pull requests, testing, and repository workflow.