Application of Discriminant Analysis for Blood Pressure Classification Based on Vital Signs: Evidence from a Regional Hospital in Ghana

This retrospective study of 1,000 adult patients at a Ghanaian regional hospital demonstrates that linear discriminant analysis using routinely collected vital signs, particularly body weight, can accurately classify patients into hypotensive, normotensive, or hypertensive categories, supporting the feasibility of data-driven risk stratification for cardiovascular and metabolic conditions.

The Gist

Imagine you are a doctor in a busy hospital in Ghana. You have thousands of patients coming through the doors every day, but you don't have enough time or blood pressure cuffs to check everyone's blood pressure individually. You need a way to quickly spot who is at risk of having dangerously high blood pressure (hypertension) or dangerously low blood pressure (hypotension) just by looking at the basic information you already have.

That is exactly what this study did. The researchers acted like detectives trying to solve a mystery: Can we predict a patient's blood pressure category just by using a few simple clues they already have on file?

Here is the breakdown of their investigation in simple terms:

1. The Detective's Toolkit (The Data)

The researchers looked at the medical records of 1,000 adult patients. Instead of doing complex, expensive tests, they only used four pieces of "vital sign" data that are usually written down for every patient:

• Age: How old are they?
• Weight: How heavy are they?
• Heart Rate: How fast is their heart beating?
• Body Temperature: How hot or cold is their body?

Think of these four items as the ingredients in a recipe. The researchers wanted to see if mixing these specific ingredients together could perfectly predict the "flavor" of a patient's blood pressure (Low, Normal, or High).

2. The Magic Formula (Discriminant Analysis)

To solve the puzzle, they used a statistical tool called Linear Discriminant Analysis (LDA).

• The Analogy: Imagine you are sorting a mixed bag of red, blue, and green marbles. You don't want to look at every single marble one by one. Instead, you build a machine that looks at the marbles' size and weight. If a marble is heavy and large, the machine sorts it into the "Red" pile. If it's light and small, it goes into the "Blue" pile.
• In this study: The "machine" (the mathematical model) learned that Body Weight and Age were the most powerful sorting tools. It realized that heavier patients and older patients were almost always in the "High Blood Pressure" pile, while lighter, younger patients were in the "Normal" or "Low" piles.

3. The Results: A Shockingly Accurate Prediction

The results were surprisingly good. The model was able to sort the patients into the correct blood pressure categories with 99.1% accuracy.

• The "Star Player": Body Weight was the MVP (Most Valuable Player). It was the single strongest clue. If you knew a patient's weight, you could guess their blood pressure category almost perfectly.
• The "Supporting Actor": Age was also very helpful.
• The "Benchwarmers": Heart rate and body temperature didn't help much on their own. They were like teammates who didn't score many points, but they still fit into the team strategy.

4. Why This Matters (The "So What?")

This study is like discovering a shortcut for doctors and public health workers.

• The Problem: In many places, checking blood pressure requires special equipment, electricity, and trained staff. Sometimes, people skip these checks because they are too busy or the equipment is broken.
• The Solution: This study suggests that if a doctor knows a patient's weight and age, they can make a very educated guess about their blood pressure risk immediately.
• The Big Picture: It highlights that managing weight is a superpower for preventing high blood pressure. If you help people lose weight, you aren't just helping them look better; you are directly lowering their risk of heart attacks and strokes.

5. The Catch (Limitations)

The researchers were honest about the limits of their study:

• It's a Snapshot: They looked at past records (like looking at old photos), so they can't say for sure that weight caused the blood pressure change, only that they are strongly linked.
• Needs a Second Opinion: Before this "shortcut" is used in real hospitals everywhere, it needs to be tested on different groups of people to make sure it works for everyone, not just the 1,000 people in this specific Ghanaian hospital.

Summary

Think of this paper as a smart filter. It tells us that we don't always need a complex lab test to understand a patient's heart health. By simply weighing a patient and asking their age, we can build a highly accurate "risk radar" that helps doctors catch dangerous blood pressure issues early, especially in places where resources are tight. It's a reminder that sometimes, the simplest clues (like a scale and a calendar) hold the most powerful answers.

Technical Summary

1. Problem Statement

Hypertension and diabetes are leading causes of cardiovascular morbidity and mortality, particularly in low- and middle-income countries (LMICs) like Ghana. While direct blood pressure measurement is the clinical standard, there is a need for efficient, early screening tools that utilize routinely collected vital signs to stratify risk without requiring specialized equipment for every initial assessment.

Existing literature often focuses on isolated predictors or treats blood pressure as a continuous variable using regression. There is a gap in research regarding multivariate classification techniques that explicitly categorize patients into distinct clinical groups (hypotensive, normotensive, and hypertensive) using common physiological markers (age, weight, temperature, heart rate) to support decision-making in resource-constrained settings.

2. Methodology

• Study Design: A retrospective observational study utilizing secondary data from a regional hospital in Ghana.
• Dataset: Anonymized records of 1,000 adult patients (aged ≥18 years).
• Target Variable (Dependent): Systolic Blood Pressure (SBP) status, categorized into three groups:
  • Hypotensive: < 90 mmHg
  • Normotensive: 90–120 mmHg
  • Hypertensive: > 120 mmHg
• Predictor Variables (Independent):
  • Age (years)
  • Body Weight (kg)
  • Body Temperature (°C)
  • Heart Rate (beats per minute)
• Statistical Approach: Linear Discriminant Analysis (LDA) was employed to construct a model that maximizes the separation between the three blood pressure groups while minimizing within-group variance.
• Assumption Checking:
  • Multivariate Normality: Assessed via skewness and kurtosis.
  • Homogeneity of Covariance Matrices: Tested using Box's M test (noted as violated but deemed robust due to large sample size).
  • Multicollinearity: Verified using Variance Inflation Factor (VIF) and correlation matrices; no problematic collinearity was found.
• Validation: Model performance was assessed using classification matrices, cross-validation, and Receiver Operating Characteristic (ROC) analysis for individual predictors. Analyses were conducted using STATA and Python.

3. Key Contributions

• Development of a Classification Model: The study successfully developed a discriminant model capable of classifying patients into hypotensive, normotensive, and hypertensive categories using only four routine vital signs.
• Identification of Dominant Predictors: The research quantified the relative importance of specific vital signs, identifying body weight and age as the primary drivers of classification, while heart rate showed negligible influence.
• Feasibility in LMICs: It provides evidence that complex risk stratification can be achieved using simple, low-cost data points available in almost any clinical setting, reducing reliance on immediate, precise blood pressure cuff readings for initial triage.
• Methodological Application: Demonstrates the utility of LDA over standard regression for categorical clinical outcomes, offering a robust framework for multivariate physiological data.

4. Results

• Descriptive Statistics: Hypertension was the most prevalent condition (51%), followed by normotension (44%) and hypotension (5%).
• Discriminant Functions:
  • Function 1 accounted for >99% of the between-group variance.
  • Body Weight was the dominant positive contributor to Function 1.
  • Age contributed moderately.
  • Body Temperature had a negative loading.
  • Heart Rate had negligible influence on the discriminant functions.
• Classification Accuracy:
  • The model achieved an overall accuracy of 99.1%.
  • 100% of hypotensive cases were correctly classified.
  • Normotensive and hypertensive groups achieved >98% accuracy.
  • Cross-validation confirmed the stability and reliability of these results.
• ROC Analysis (Individual Predictors):
  • Body Weight: Near-perfect discrimination (AUC = 0.999).
  • Age: Excellent predictive ability (AUC = 0.978).
  • Temperature & Heart Rate: Showed limited standalone classification power (low AUC).

5. Significance and Implications

• Clinical Utility: The findings suggest that body weight and age are sufficient for highly accurate initial risk stratification. Clinicians can use these easily measurable indicators to prioritize patients for immediate blood pressure confirmation and intervention.
• Public Health Strategy: The strong correlation between weight and blood pressure status reinforces the necessity of weight management and obesity prevention as central strategies for controlling hypertension and related cardio-metabolic diseases in Ghana and similar contexts.
• Scalability: The model supports the integration of data-driven risk stratification into Electronic Health Records (EHR) and mobile health platforms, enabling early detection in community settings where specialized equipment may be scarce.
• Limitations & Future Work: As a retrospective study, it cannot establish causality and does not account for confounders like medication use or socioeconomic status. The authors recommend external validation in independent cohorts before widespread clinical implementation.

Conclusion: The study validates that a discriminant model based on routine vital signs (specifically weight and age) can classify blood pressure status with exceptional accuracy, offering a scalable, low-cost solution for hypertension screening in resource-limited environments.