The Journal of ExtraCorporeal Technology No 58-1 | Page 103

A. M. Palmer: J Extra Corpor Technol 2026, 58, 95--99 97

During model development for Research Question One, the Hosmer--Lemeshow test was used as a diagnostic tool to determine which academic variables-- specifically course grades-- contributed to a well-fitting model. Together, the binary logistic regression model and the Hosmer--Lemeshow test provided a comprehensive approach to modeling and evaluating the binary outcomes in statistical analysis. A high p-value(>. 05) suggested the model fits the data well, meaning there were no significant differences between the observed and expected outcomes among the variables. For this model, the p-value =. 164.

Because the Hosmer--Lemeshow test had to show the model was a good fit, additional undergraduate and graduate courses were not included in the test due to a low p-value(<. 05) indicating a poor fit of the model. This suggested that the predicted probabilities based on these variables did not accurately match the observed ABCP exam outcomes. Several attempts were made to use as many of the undergraduate and perfusion classes as possible. However, there was skewed data which impacted the overall fit of the regression models. The Hosmer--Lemeshow test thus helped identify which courses could be retained without compromising model fit. Based on these factors, all undergraduate courses were left out of the regression. These courses included anatomy and physiology, calculus, ecology, organic chemistry, medical physics, etc.

For research questions two and three, binary variables were created to convert categorical values such as“ Male” and“ Female” into numerical form, with“ male” coded as 1 and“ female” as 0. To avoid multicollinearity in the regression model, one fewer binary variable than the number of categories was used. The omitted category serves as the reference group for comparison. In cases with two categories, one category is designated as the reference code. For research question two, the demographic variable included“ male” and“ female.” The education variable categorized students as“ undergrad” if they attended Carlow University prior to entering the perfusion program, and“ transfer” if they did not. For the experience variable,“ yes” indicated students with prior medical experience, while“ no” indicated those without it. For the demographic variable,“ male” was used as the reference code and“ female” was the dummy variable. For the education variable,“ transfer” was used as the reference code and“ undergraduate” was used as the dummy variable. For the experiential variable,“ yes” was used as the reference code and“ no” was used as binary variable.

Results

To evaluate the first research question, what courses affect the ABCP certification exam pass / fail outcome during graduate( perfusion) school, the logistic analysis in Table 2 shows the quintiles created for each course included in the regression. To improve the model’ s performance and interpretability, course grades were transformed into quintiles, allowing for more balanced groupings and reducing skewed data. This transformation allowed for a better-fitting model( p >. 05) while preserving meaningful insights into the relationship between academic performance and exam outcomes. Each variable in

Table 1. General description of variables.

the equation is represented by the percentile group of each graduate school class. Because the model was not of sound fit( p <. 05), the undergraduate courses were left out of the regression. Students in the higher quintile were more likely to pass the exam on the first attempt( p <. 05). The unstandardized B( beta) = 1.002 indicated students in the higher percentile group of Intro to Cardiac Perfusion were more likely to pass the exam on the first attempt, p =. 008. The unstandardized B( beta) =. 636 indicated students in the higher percentile group of Hematology were more likely to pass the exam on the first attempt, p =. 028. The percentile group of Pulmonary Anatomy and Physiology was starting to reach statistical significance, with p =. 089. The remaining course variables such as Intro to Cardiovascular Surgery, Congenital Pathology, and Cardiovascular Pharmacology were not statistically significantly related to the dependent variable pass or fail the ABCP examination( p >. 05).

To answer the second research question, what applicant factors( demographic, educational, and experiential) affect pass / fail outcome on the ABCP certification exam, the logistic regression analysis in Table 3 shows the binary variables created for each of the applicant factors included in the regression. The regression identified an independent applicant variable, admission: undergraduate, which statistically significant( p =. 004) indicating students were more likely to pass the exam on the first attempt. The independent applicant variable, experience: no, was starting to reach statistical significance, because p =. 097.

The third research question sought to determine the impact of clinical experience, such as rotation site or cohort, on the ABCP certification exam pass / fail score; the logistic regression analysis in Table 4 shows the dummy variables created for each cohort from 2017 to 2021 for inclusion in the regression analysis, with the 2022 cohort used as the reference group. The regression showed that the independent variable, more clinical experience or cohort, was not statistically significantly related to the dependent variable, pass / fail the ABCP examination( p >. 05).