Evidence-based thresholds improve detection of medical students’ academic decline

Academic warning systems used by medical universities in Iran may miss most students whose grades improve or deteriorate sharply from one semester to the next. A retrospective study of 1,175 students at Mazandaran University of Medical Sciences found that the customary rule of a two-point grade-point-average change had exceptionally high specificity but identified only a small fraction of genuine academic progress and decline. Researchers used latent profile analysis to establish a model-based reference standard, then derived alternative thresholds with receiver operating characteristic curves and the Youden index. The results suggest that more sensitive, evidence-based cut-offs could make university monitoring more useful without turning routine grade fluctuations into false alarms.

The practice under examination compares grades from two consecutive semesters and classifies a change of at least two GPA points as academic progress or decline, with the direction of the change determining the label. It is widely used in Iranian academic education offices, partly because it is simple and administratively easy to apply. The study notes, however, that the rule had not previously been subjected to diagnostic validation using measures such as sensitivity, specificity and predictive values. As a result, its apparent precision did not necessarily mean that it reliably detected the students for whom intervention mattered most.

To evaluate the conventional rule, the team analyzed anonymized records from the fall 2022 and spring 2023 semesters at Mazandaran University of Medical Sciences in Sari. Students were selected by random sampling, and the principal measure was the difference between their semester GPAs. Because students may take different amounts of academic credit, the analysis was stratified by quartiles of change in academic credits. The study used latent profile analysis, or LPA, as a reference standard for assigning students to three groups: academic decline, no change and academic progress. LPA is a probabilistic classification method designed to identify groups that may not be apparent from a single fixed threshold.

According to the traditional plus-or-minus two-point rule, 14.7% of students were classified as making academic progress and 6.5% as declining, with 95% confidence intervals of 12.7% to 16.9% and 5.1% to 8.2%, respectively. The LPA reference standard produced a much smaller and different picture: only 29 students, or 2.47%, were assigned to the progress group, while 35 students, or 2.98%, were assigned to the decline group. This marked difference did not mean that only a few students experienced meaningful academic movement. Rather, it showed that a substantial number of records crossed the administrative threshold without belonging to the underlying trajectory groups identified by the model.

When the conventional threshold was tested against those LPA classifications, its positive predictive value and specificity were both 100% for progress and decline. In other words, students labeled by the two-point rule were highly likely to fall into the corresponding LPA group, and nearly all students the model classified as stable were not incorrectly labeled as changing. The weakness emerged in sensitivity. The traditional rule detected only 17.8% of students whom LPA classified as making progress, with a 95% confidence interval of 12.3% to 24.5%. Its sensitivity for decline was higher but still incomplete, reaching 53.0%, with a 95% confidence interval of 40.3% to 65.4%. Overall accuracy was 59% for progress and 77% for decline.

The authors then applied receiver operating characteristic analysis, which evaluates diagnostic thresholds across the full range of possible GPA differences, and used the Youden index to select points balancing sensitivity and specificity. The optimal overall cut-off for decline was a change of at least minus 2.14 GPA points. At that threshold, sensitivity reached 100% and specificity 88.21%. For progress, the corresponding cut-off was at least plus 2.41 points, also yielding 100% sensitivity, with 91.45% specificity. The thresholds are therefore not simply a looser version of the old rule. They are slightly more extreme, and their value lies in identifying boundary cases more accurately while retaining almost complete capture of the LPA-defined groups.

The analysis also produced cut-offs for quartiles defined by changes in academic credits, reflecting the possibility that the same GPA movement can carry different implications when students enroll in different academic workloads. The authors reported markedly improved specificity in several of these subgroups, reaching 100%. This stratification may help universities avoid applying one nominal GPA boundary to students whose credit loads and grading contexts differ. The study did not present the subgroup thresholds as a universal screening rule, however; rather, its results provide locally derived estimates that should be evaluated and recalibrated where course structures, curricula or grading distributions differ.

For medical education, the practical consequence is potentially substantial. A warning system that fails to detect most true cases of academic decline cannot reliably trigger counseling, mentoring, assessment of fitness to progress or other support. The same problem applies in the opposite direction: missing genuine academic progress may deny students useful recognition or obscure evidence that interventions or study strategies are working. The study therefore frames LPA-derived cut-offs as a potential basis for early-warning systems, but it does not establish that identifying a student as declining will itself improve outcomes. Prospective evaluation would still be needed to determine whether the thresholds lead to better academic decisions, timely support and sustained student progress.

The findings also illustrate why a simple administrative rule can remain popular while being diagnostically incomplete. A threshold of two points can look decisive because almost every person it flags belongs to the intended group in this dataset. Yet that apparent advantage conceals a major failure to detect change, especially for academic progress. LPA offered a different way of asking who truly followed distinct performance trajectories, and ROC analysis converted those trajectories into operating thresholds. The resulting combination of 100% sensitivity and specificity above 88% for the overall cut-offs was achieved with 1,175 students at one Iranian institution over two consecutive semesters, so broader validation is needed before national adoption.

For universities, the study’s clearest message is that grade-change screening should be tested rather than assumed to work. Administrative simplicity is valuable, but it cannot substitute for evidence about who is actually identified and who is overlooked. If replicated, the model-based approach could give educators a more defensible starting point for academic alerts, allowing scarce advising resources to reach students whose trajectories warrant attention. The authors expect the results to inform the design of early-warning systems in Iranian medical education. Whether they can do so in practice will depend on prospective implementation, local recalibration and assessment of the educational and clinical consequences of acting on the classifications.

Subject of Research: Validation of GPA-based thresholds for identifying academic decline and progress among medical sciences students.

Article Title: Optimal cut-off point for identifying academic decline and progress in medical sciences students: a latent profile analysis-based approach

Article References: Rezapour, M., Amuei, F., Mohammadi, H. R., Ghasempouri, S. K., & Moradi, S. (2026). Optimal cut-off point for identifying academic decline and progress in medical sciences students: a latent profile analysis-based approach. BMC Medical Education. https://doi.org/10.1186/s12909-026-10405-5

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10405-5

Keywords: Medical education, Academic decline, Academic progress, GPA, Latent profile analysis, Diagnostic accuracy, Early-warning systems, ROC analysis, Iranian universities, Student support, Optimal, cut-off