Research Points to the Need for a More Nuanced View of Student Risk

For academic nursing leaders and faculty, the data and detail available for each student can produce information overload. Which data points really matter? What are they telling you? And how can your program use this data to support student and program outcomes?

Greater clarity is coming into view. Recent studies of students in nursing and health professions programs offer a better, albeit evolving, understanding. Researchers are using statistical modeling, predictive analytics and machine learning to identify patterns associated with attrition, future course performance, and program completion.

An overarching theme is that the most useful predictions may not come from a measure determined at a single point in the curriculum. Instead, risk tends to become clearer as evidence of learning and performance accumulates.

What Can Admissions Data Tell You About Student Risk?

Long before risk identification entered academic conversations, nursing programs sought to link student characteristics with academic success. This pursuit continues and has demonstrated the predictive value of some traditional measures.

In a retrospective 2022 study of 351 associate degree nursing students over seven cohorts, Eudy and Brooks examined whether 15 factors contributed to the successful completion of a nursing fundamentals course.¹ Their research focus was spurred by a previous finding showing that poor performance in fundamentals negatively affected program success.

The authors found that higher ACT composite scores and higher general psychology grades were significantly associated with success in the class. Neither SAT critical reading scores nor SAT math scores were statistically significant predictors of success, nor were grades in anatomy and physiology courses.

Among students who were not successful, the authors identified demographic, academic and nonacademic factors as contributors. They concluded that admissions criteria can be a useful data point, but that programs have a responsibility to identify and provide support that addresses more of the factors that influence success.

A 2023 retrospective review by Bruce, Mabizela and Tshabala examined admissions data from 317 bachelor of nursing students and found that scores on national benchmarking exams were significant predictors of first-year nursing success.² Math and academic literacy (English usage and reading skills) scores exhibited the highest correlation, followed by high school scores in life sciences.

In an unanticipated finding, the review found that admission test scores and high school grades together accounted for only 35% of the variation in first-year grades — leaving  65% largely unexplained. The authors concluded that while academic readiness measures can reveal important vulnerabilities, other academic and nonacademic factors warrant consideration.

The findings of these two studies provide a point of consideration for academic nursing programs: An early risk indicator can provide information, but it doesn’t determine a student’s trajectory.

Student Risk Becomes Clearer as Data Accumulates

In other recent research, authors have shifted the focus from what programs know about students when they enter to what they learn about them after classes begin.

A 2025 study by Crisp, Cook and Jones analyzed routinely collected administrative data from 1,840 nursing students and developed attrition models for different stages of their education.³

The researchers found that information available early in the program exhibited relatively weak predictive performance, while models incorporating students’ results as they progressed through coursework were substantially more predictive of later attrition and completion. After enrollment, student results and periods of interrupted study became substantially stronger predictors of outcomes.

This finding suggests that waiting for more data may produce a stronger prediction. But every program director and faculty member knows that waiting too long can reduce the amount of time available to respond. For nursing programs, this creates a practical tradeoff: Obtaining an accurate prediction is valuable, but equally important is getting a signal that is reliable enough to be useful to the faculty who can act on it.

Not All Student Risk Looks the Same

Another thread in recent research focuses on expanding the data used to understand student risk.

In a 2026 study of 527 undergraduate nursing students, Li and coauthors tested five machine-learning models to predict performance in the theoretical and practical components of a fundamentals of nursing course.⁴

The data encompassed GPA, periodic exams, assignments, practical assessments, comprehensive skills training, and practice frequency. The team found that different combinations of variables predicted differing outcomes.

The findings raise an important consideration: Being “at risk” isn't necessarily a single condition. For example, a nursing student may demonstrate the start of a theoretical knowledge problem and difficulty with performance of a clinical skill. Or they could exhibit inconsistent engagement and another pattern requiring a different response.

Research in a medical school setting reinforces this observation. Mastour and colleagues analyzed records from 1,005 medical students to predict performance on the Comprehensive Medical Basic Sciences Examination.⁵

Through application of nine machine learning algorithms, they determined that medical program GPA and performance in specific basic sciences subjects, including anatomy and biochemistry, demonstrated the strongest association with success on this standardized high-stakes exam for physicians.

The outcomes of these two studies don’t suggest that nursing programs need increasingly complex algorithms. Rather, they demonstrate that prediction can potentially become more useful to education when it helps faculty understand what the risk signal represents. Assigning a student to a risk category without other action isn't enough.

In their research, Li et al also noted the need for comprehensive consideration of student needs. They emphasized that student performance is shaped by multiple factors — some of which weren’t captured in the variables used to generate the study findings.

They recommend personalized teaching strategies that incorporate:

• differentiated instructional strategies that reflect individual learner characteristics
• integrated approaches that bridge theoretical instruction and practical application
• support that helps students balance academic demands with other commitments.

How Should Nursing Programs Think About Predictive Accuracy?

Another recent research focus stems from the very nature of prediction. Programs that use predictive data want earlier information, and that can introduce an element of uncertainty.

Nimy and coauthors studied 517 first-year university students using a probabilistic machine-learning model that incorporated data on students’ demographics, Moodle engagement and academic performance. ⁶ The team collected this data at five points during the semester. Rather than predicting a final grade, the model calculated the probability that a student would fail an upcoming assessment.

Early in the semester, engagement signals such as time spent in the course exhibited more predictive importance. But after students began taking assessments, those results became more influential. The strongest predictive accuracy occurred in the sixth week of the program. The researchers also determined that uncertainty in several model parameters decreased as performance data accumulated.

This finding suggests an approach to incorporating predictive systems. Accuracy matters, but so does understanding the conditions behind a prediction. For programs evaluating predictive data, this research suggests that several questions become important:

• When is the prediction being made?
• What information is being used to make the prediction?
• What is the degree of uncertainty in the prediction?

In this study, later predictions carried greater certainty, while earlier identification preserved more opportunity for intervention.

What Happens After a Student Is Identified as Being at Risk?

As the research on predicting nursing student success evolves, an important question continues to be a priority: What should happen after risk is identified?

A 2025 study of a learning analytics intervention provides a useful example outside of nursing education.⁷ Educators in a first-year computer course used continuous assessment data to predict student risk and provided personalized interventions during weeks 6 and 11. The historical control group received no intervention.

Among students identified as being at risk, 45.5% of the intervention cohort passed the class and only 10.9% of the historical control group passed. The failure rate in the intervention group was 28%, compared to the control group’s 72.7%.

Although the findings cannot be assumed to apply directly to nursing education, the study illustrates why identifying risk and responding to it should be viewed as two related but distinct parts of the process.

Identifying risk alone does not ensure improvements in student progression, retention or licensure outcomes. Those outcomes also depend on what programs do after educators receive the signal showing that a student may need attention.

What Can Academic Nursing Programs Do with Predictive Data?

Across the studies outlined here,^1-7 evidence suggests that risk information can change as students progress; actual performance can add information that isn’t available at admission; different data combinations can reveal various kinds of vulnerability; and later predictions may be more accurate than early ones.

For deans, directors and faculty, these observations expand the lens through which student retention strategies can be viewed. Rather than focusing solely on whether a predictive system can identify an at-risk student, programs can get a more comprehensive picture by asking:

• How early does a meaningful signal of risk emerge?
• Which student data points contribute to prediction of success?
• Does information become more precise as data accumulates?
• Can educators see where a potential weakness may be developing?
• What specific actions should the program take after an at-risk student is identified?

As nursing education moves from retrospective reporting toward more timely use of student performance information, academic programs are likely to explore these questions and more.

Evolving research is shaping new perspectives on student data, and the value of prediction is expanding beyond identifying who may struggle. Coming into view is the need to recognize developing risk early enough — and specifically enough — to create a meaningful opportunity for program leaders and educators to understand it and respond.

References

1. Eudy C, Brooks S. Factors impacting student success in a fundamentals course of an associate degree nursing program. Teaching and Learning in Nursing. 2022. doi:10.1016/j.teln.2021.05.004
2. Bruce J, Mabizela SE, Tshabalala AM. Selection tests and their predictive value in university nursing students' success in the first year of study. BMC Medical Education. 2023;23:176. doi:10.1186/s12909-023-04140-4
3. Crisp E, Cook R, Jones S. A model for predicting student nurse attrition during pre-registration training: A retrospective observational study using routinely collected administrative data. Nurse Education in Practice. 2025;85:104377. doi:10.1016/j.nepr.2025.104377
4. Li RL, Lan Y, Han YF, Lan HX, Chen DD, Jiang WT. Construction of a predictive model for theory and practical performance in Fundamentals of Nursing based on student behavior: A case study of nursing students. Iranian Journal of Nursing and Midwifery Research. 2026;31:502-512. doi:10.4103/ijnmr.ijnmr_38_25
5. Mastour H, Dehghani T, Moradi E, Eslami S. Early prediction of medical students' performance in high-stakes examinations using machine learning approaches. Heliyon. 2023;9:e18248. doi:10.1016/j.heliyon.2023.e18248
6. Nimy E, Mosia M, Chibaya C. Identifying At-Risk Students for Early Intervention: A Probabilistic Machine Learning Approach. Applied Sciences. 2023;13:3869. doi:10.3390/app13063869
7. Alalawi K, Athauda R, Chiong R, Renner I. Evaluating the student performance prediction and action framework through a learning analytics intervention study. Education and Information Technologies. 2025;30:2887-2916. doi:10.1007/s10639-024-12923-5