Scales balancing patient data and medical data, symbolizing propensity score analysis.

Decoding Propensity Scores: A Beginner's Guide to Reducing Bias in Health Studies

Theo Raines monogram
Theo Raines in Health & Wellness October 2026 • 4 min read.

"Learn how propensity score methods can help you make more accurate conclusions from observational clinical data."


In the world of health research, randomized controlled trials (RCTs) are often seen as the gold standard. In an RCT, participants are randomly assigned to different treatment groups, ensuring that any differences in outcomes are likely due to the treatment itself. However, RCTs aren't always feasible or ethical. Sometimes, researchers have to rely on observational studies, where they observe people in their natural settings without intervening.

Observational studies can be incredibly valuable, but they come with a challenge: bias. People in different treatment groups might also differ in other important ways (age, lifestyle, etc.). These differences, known as confounding variables, can cloud the true effect of a treatment.

That's where propensity scores come in. Propensity score methods are statistical techniques designed to reduce bias in observational studies. In essence, they help create a level playing field, allowing researchers to draw more reliable conclusions about treatment effects.

AI Search Multiple angles on this topic

Why Propensity Scores Matter

A propensity score is the probability of treatment assignment based on observed characteristics, and it can be used to reduce bias and improve causal inference datacamp.com. The JAMA guide describes propensity scores as a way to reduce bias in treatment-effect estimates when randomized trials are not feasible jamanetwork.com. A 2021 simulation study examined weighting and trimming strategies for reducing variance and bias in treatment-effect estimates .

Matching and Model Choices

Propensity score matching (PSM) is described as a way to control bias and achieve pseudo-randomization in retrospective observational studies sciencedirect.com. The propensity score is the conditional probability of treatment given an individual's covariates; balancing covariates between groups can reduce bias stat.ubc.ca. A review published in 2026 identifies a need to synthesize evidence on the performance of propensity score methods and offer practical guidance for estimating treatment effects in observational studies . A separate analysis compares eight propensity score models across simulated settings involving selection on observed versus unobserved variables, assessing bias reduction and sample-size retention .

What are Propensity Scores and How Do They Work?

Scales balancing patient data and medical data, symbolizing propensity score analysis.

At its core, a propensity score is a single number that represents an individual's likelihood of receiving a particular treatment, based on their observed characteristics. It's like a summary of all the factors that might influence treatment assignment. Think of it as a personalized probability score that answers this question: "Given everything we know about this person, what's the chance they ended up in the treatment group?"

Researchers use statistical models to estimate these scores. Logistic regression is a common choice. The model considers various covariates (age, sex, health history, etc.) to predict the probability of treatment assignment. Once propensity scores are calculated, researchers can use them in several ways to reduce bias:

  • Matching: Pair individuals in the treatment and control groups who have similar propensity scores. This creates balanced groups, making it easier to compare outcomes.
  • Stratification: Divide the study population into subgroups (strata) based on propensity score ranges. Analyze treatment effects within each stratum to minimize confounding.
  • Inverse Probability of Treatment Weighting (IPTW): Assign weights to individuals based on their propensity scores. This re-weights the sample to mimic a randomized trial.
  • Covariate Adjustment: Include the propensity score as a covariate in a regression model. This statistically adjusts for differences between treatment groups.
AI Search Multiple angles on this topic

Research on Model Specification

A chapter in “Aspects of causal inference” focuses on specifying propensity score models to achieve the best treatment-effect estimates doi.org. A 2026 review describes a need for a comprehensive synthesis of evidence on propensity score methods and practical guidance for estimating treatment effects in observational studies link.springer.com. The supplied Post-GWAS review snippet identifies propensity score approaches as addressing confounders such as disease and patient characteristics . The NEAT trial protocol describes a randomized study of carnosine versus placebo, with supplementation for 12 weeks and repeated assessments during peak particulate-matter exposure in Louisville, Kentucky .

Residual Confounding and Poor Overlap

Propensity score methods do not claim to remove confounding from unmeasured covariates, so observational findings require care because residual confounding may remain pmc.ncbi.nlm.nih.gov. Inadequate overlap in propensity score distributions is another warning: the cited review says that even a large dataset may not support a causal conclusion without relying on untrustworthy model assumptions when overlap is poor pmc.ncbi.nlm.nih.gov.

Balance and Matched-Data Reporting

A survey of cardiovascular research found that covariate balance was formally assessed using both standardized mean differences and hypothesis testing in 29% of articles; 16% did not report a measure of balance pmc.ncbi.nlm.nih.gov. In a separate appraisal of propensity-score matching articles published from 1996 to 2003, 13 articles, or 28%, explicitly used statistical methods appropriate for matched data when estimating treatment effects and their statistical significance pubmed.ncbi.nlm.nih.gov. These findings point to reporting and analysis practices as important considerations when evaluating propensity score studies.

Each of these methods has its strengths and weaknesses. The choice depends on the specific research question and the characteristics of the data. The goal, however, is the same: to create a fairer comparison between treatment groups by accounting for confounding variables.

The Power of Propensity Scores: Real-World Impact

Propensity score methods aren't just theoretical tools; they have real-world applications. Observational studies are essential for understanding treatment effectiveness, identifying risk factors, and informing public health policies. By using propensity scores, researchers can strengthen the validity of these studies, leading to more reliable evidence for decision-making in healthcare.

AI Search Multiple angles on this topic

Real-World Data and Practice

A 2020 paper introduces commonly used and emerging propensity score methods, discusses their pros and cons, and presents three case studies using real-world and historical data sciencedirect.com. A 2026 review says observational studies using real-world data have become increasingly popular and are susceptible to selection bias; it describes propensity score methods as balancing patient characteristics across treatment groups to mitigate bias journals.sagepub.com. An ISPOR presentation reports inadequate reporting and control of selection bias and confounding in real-world-data-based economic evaluations, and calls for improvement in propensity score implementation and reporting .

Where the HELP clinic analysis is most constrained

  • A variable associated with homelessness was left out of score estimation because many participants lacked its data.[1]
  • The score model was estimated after excluding participants with missing values on important covariates, so the analysis covered a reduced subset.[1]
  • In the real-data analysis, the treatment–outcome association was not statistically significant after propensity score adjustment, unlike the unadjusted association.[1]

The HELP clinic study in the paper's own terms

PopulationHealth Evaluation and Linkage to Primary care clinic clinical data[1]
Comparisontreated and untreated groups[1]
Times cited4[1]

What remains unanswered about the HELP estimates

  • How sensitive would the estimated association be to the excluded participants whose covariate information was missing?[1]
  • Would the treatment–outcome association remain non-significant under other score specifications or adjustment approaches?[1] [2]

Sources

  1. Review: propensity score methods with application to the HELP clinic clinical study (Informa UK Limited, 2018)
  2. Aspects of causal inference. (2020)

About this Article -

Written with AI assistance from published research, and reviewed by the Mystum team. See our About page for more information.

This article is based on research published under:

DOI-LINK: 10.2147/oams.s156704, Alternate LINK

Title: Review: Propensity Score Methods With Application To The Help Clinic Clinical Study

Subject: General Medicine

Journal: Open Access Medical Statistics

Publisher: Informa UK Limited

Authors: Shesh N Rai, Xiaoyong Wu, Deo K Srivastava, John A Craycroft, Jayesh P Rai, Sanjay Srivastava, Robert F James, Maxwell Boakye, Aruni Bhatnagar, Richard Baumgartner

Published: 2018-06-01

Everything You Need To Know

1

In health studies, what exactly are propensity scores and why are they useful?

Propensity scores are a statistical method used to reduce bias in observational health studies. A propensity score is a single number that represents an individual's likelihood of receiving a particular treatment based on their observed characteristics. By accounting for confounding variables, propensity scores help create a fairer comparison between treatment groups. This is particularly useful when randomized controlled trials (RCTs) are not feasible or ethical.

2

How are propensity scores calculated, and what data is needed for this process?

Propensity scores are calculated using statistical models, commonly logistic regression. These models consider various covariates, such as age, sex, and health history, to predict the probability of treatment assignment. The necessary data includes observed characteristics (covariates) for each individual in the study, along with information on whether they received the treatment.

3

What are the different ways researchers can use propensity scores to reduce bias in observational studies?

Researchers can use propensity scores in several ways: matching (pairing individuals with similar scores), stratification (dividing the population into subgroups based on score ranges), inverse probability of treatment weighting (IPTW, assigning weights based on scores), and covariate adjustment (including the score in a regression model). Each method has its strengths and weaknesses, and the choice depends on the research question and data characteristics. These methods allow researchers to create balanced groups, making it easier to compare outcomes.

4

How do propensity score methods improve the validity of observational studies and what impact does this have on healthcare decision-making?

Propensity score methods strengthen the validity of observational studies by reducing bias caused by confounding variables. By creating fairer comparisons between treatment groups, researchers can obtain more reliable evidence about treatment effectiveness, identify risk factors, and inform public health policies. This leads to more informed decision-making in healthcare, as the conclusions drawn from these studies are more likely to reflect the true effect of a treatment.

5

Considering the use of propensity scores in observational studies, what are some potential limitations or challenges that researchers should be aware of?

While propensity scores are powerful, researchers should be aware of potential limitations. The accuracy of propensity scores depends on the quality and completeness of the observed covariates; if important confounding variables are not included in the model, bias may persist. Additionally, propensity score methods assume that there are no unmeasured confounders that affect both treatment assignment and outcome. Furthermore, the choice of the specific propensity score method (matching, stratification, IPTW, covariate adjustment) can influence the results, so researchers should carefully consider the characteristics of their data and research question when selecting a method.

Newsletter Subscribe

Subscribe to get the latest articles and insights directly in your inbox.