This multicenter, retrospective study analyzed post-marketing data from patients treated with ulinastatin (manufactured by Guangdong Techpool Bio-pharma Co., Ltd., China) between August 2014 and June 2017 (Li et al., 2022). The study cohort comprised patients admitted to general wards and ICUs across nine participating hospitals in China. The study protocol received approval from the Ethics Committee of Guangdong Provincial Hospital of Chinese Medicine (Approval No. B2014-056-01), which served as the lead center. All procedures were conducted in compliance with relevant guidelines and regulations. The Ethics Committee of Guangdong Provincial Hospital of Chinese Medicine granted a waiver for the requirement of informed consent.

Excluded patients with missing data, this study enrolled 11,218 patients and the following data were collected: ulinastatin dosage (in 1 × 10⁴ U units), history of cardiopulmonary diseases, ICU admission status, length of hospital stay, gender, concomitant medications, allergy history, and occurrence of cardiopulmonary adverse outcomes. Based on the occurrence of such outcomes, the patients were categorized into two groups: those with cardiopulmonary adverse outcomes and those without. The cardiopulmonary adverse outcomes were ascertained from the Phase IV clinical trial database, to ensure consistency and accuracy in assessment, the terminology for all adverse outcomes was described in accordance with the Common Terminology Criteria for Adverse Events version 4.0 (CTCAE v4.0). All outcomes were first evaluated by the attending physicians and subsequently reviewed by an independent expert panel to confirm their causality and relevance to ulinastatin. Subsequently, logistic regression analysis was employed to identify independent risk factors among the collected variables. Finally, these data were used to establish a risk scoring system (Figure 1).

Prior to regression analysis, all variables were assigned numerical values as follows: a history of cardiopulmonary disease was coded as 1, and its absence as 0; ICU admission as 1, and general ward admission as 0; allergy history as 1, and no allergy history as 0; male gender as 1, and female as 0; concomitant medication use as 1, and no concomitant medications as 0. Continuous variables were dichotomized based on clinical relevance: length of hospital stay greater than 7 days was assigned a value of 1 (≤7 days as 0), as most patients in the dataset had a hospital stay in 7 days. According to the prescribing information of ulinastatin (Lv et al., 2020), the recommended initial dose for adult patients with acute pancreatitis or chronic recurrent pancreatitis is 10 × 10⁴ units per administration, diluted in 500 mL of 5% glucose injection or 0.9% sodium chloride injection for intravenous drip infusion over 1–2 h, administered 1–3 times daily. The dose may be gradually reduced as symptoms subside. For acute circulatory failure, the same dosage and dilution are recommended, with the option of slow intravenous injection under certain circumstances. The prescribed daily dose ranges between 10 × 10⁴ and 30 × 10⁴ units (Lyu et al., 2022). Based on the collected data, the average daily dose administered was 30 × 10⁴ units. Therefore, a total daily dose of ≥30 × 10⁴ units was assigned a value of 1, and <30 × 10⁴ units as 0. Duration of medication >1 day was assigned a value of 1, and <1 day as 0. The variable ‘duration of medication >1 day’ was defined as the administration of ulinastatin on more than one calendar day, irrespective of the total infusion hours.

ROC curve analysis was constructed to evaluate the discriminatory performance of the predictive model by plotting the true positive rate (sensitivity, true positive rate, TPR) against the false positive rate (1 – specificity, false positive rate, FPR) across all classification thresholds (Nahm, 2022; Lai et al., 2025). The ROC curve is formed with the FPR as the abscissa (x-axis) and the TPR as the ordinate (y-axis), wherein the coordinates (FPR, TPR) traversed across every conceivable threshold are joined to constitute the complete curve (Zhou et al., 2024). The overall performance was quantified by calculating the AUC, with an AUC of 1.0 indicating perfect discrimination and lower than 0.5 representing no discriminative capacity. The optimal cutoff value was identified based on the point on the curve closest to the top-left corner, maximizing both sensitivity and specificity.

A risk scoring system was developed to transform the multivariable logistic regression model into a simple (Chen et al., 2023), clinically applicable tool for predicting cardiopulmonary adverse outcomes. Initially, independent risk factors were identified through multivariable logistic regression, retaining variables with a significance level of p < 0.05. Each factor’s regression coefficient (β) was used to determine its relative weight (Lee et al., 2022). To construct the score, the β coefficient of each significant predictor was divided by a reference value (often the smallest β in the model), and rounded to the nearest integer to assign point values. For instance, a factor with β = 0.70 assigned 1 point, while another with β = 1.40 received 2 points. The points for all predictors were summed to calculate a total risk score for each patient. This score was then categorized into risk strata based on the observed incidence of the outcome in each score range. The total risk scores were then used to generate the ROC curve using SPSS software (version 25.0) (Wu et al., 2021). The discriminatory power of the risk scoring system was evaluated by comparing the area under the curve (AUC) with that of individual risk factors. Finally, risk stratification into high- and low-risk groups was performed to support clinical decision-making and guide intervention strategies.

All statistical analyses were performed using IBM SPSS Statistics 25.0 and Microsoft Excel. Categorical variables were compared using the χ² test or Fisher’s exact test, as appropriate. Continuous variables were compared using the Mann-Whitney U test or t-test based on their distribution. Univariate and multivariate logistic regression analyses were employed to calculate ORs and 95% CIs for identifying independent risk factors. A ROC curve was generated to evaluate the predictive performance of both the final multivariate model and the derived risk scoring system. The AUC was compared between the model and individual predictors. A two-sided p-value <0.05 was considered statistically significant.

The analysis included a total of 11,218 patients treated with ulinastatin, after the exclusion of 34 individuals due to incomplete baseline information. Among this cohort, 152 patients (1.35%) experienced cardiopulmonary adverse outcomes (11,066 (98.65%) did not), establishing a clear patient safety endpoint for further investigation. The total dosage of ulinastatin administered ranged from 2 × 10⁴ to 1.30 × 10⁴ U, with a mean dose of 30.27 × 10⁴ U. Univariate analysis aimed to identify clinical practices and patient profiles associated with this risk. It revealed significant associations with several identifiable factors. Specifically, a higher incidence of adverse outcomes was observed in patients receiving a total ulinastatin dosage of ≤30 × 10⁴ U (93.42% vs. 69.00% in the general ward group; p < 0.001), those with concomitant medication use (61.84% vs. 40.59%; p < 0.001), ICU patients (77.63% vs. 62.10%; p < 0.001), and those with medication duration >1 day (69.10% vs. 40.20%; p < 0.001). In contrast, no statistically significant differences were found in sex (male: 65.8% vs. 65.10%; p = 0.850), allergy history (7.24% vs. 5.02%; p = 0.217), history of cardiopulmonary diseases (57.24% vs. 53.09%; p = 0.309), or length of hospital stay (>7 days: 71.71% vs. 67.24%; p = 0.244) between the two groups. The detailed distribution of all baseline characteristics is presented in Table 2.

To further delineate the clinical burden of these events, we categorized the specific types of cardiopulmonary adverse outcomes among the 152 affected patients. Among patients who experienced cardiopulmonary adverse outcomes, the top five conditions in the ICU setting were: severe pneumonia (39 cases, 33.05%), respiratory and circulatory failure (36 cases, 30.51%), respiratory and cardiac arrest (25 cases, 21.19%), cardiogenic shock and heart failure (each 10 cases, 8.47%), and myocardial infarction (9 cases, 7.63%). In the general ward group, the most frequent adverse outcomes were respiratory and circulatory failure (15 cases, 44.12%), severe pneumonia (10 cases, 29.41%), respiratory and cardiac arrest (9 cases, 26.47%), heart failure (3 cases, 8.82%), and myocardial infarction (2 cases, 5.88%). A complete breakdown of adverse outcome types is provided in Table 3, highlighting the severe nature of the events this study aims to help prevent.

To identify independent and potentially modifiable predictors for preventive health strategies, all variables significant in univariate analysis were entered into a multivariable logistic regression model. These variables assessed included ulinastatin dosage, duration of medication/day(s), combination medication, ICU patients, history of cardiopulmonary diseases, length of hospital stay/day(s), sex, allergy history. The analysis revealed several independent predictors of cardiopulmonary adverse outcomes in patients receiving ulinastatin. A total ulinastatin dosage (low-dose) of ≤30 × 10⁴ U was the strongest medication-related risk factor, associated with a more than five-fold increase in odds (OR = 5.570, 95% CI: 3.670–8.454, p < 0.001).

Admission to the ICU itself was the strongest overall predictor (OR = 3.737, 95% CI: 2.487–5.615, p < 0.001), highlighting a critical high-risk patient population. Prolonged medication duration (>1 day) was an independent risk factor (OR = 2.165, 95% CI: 1.480–3.166, p < 0.001). Concomitant medication use also significantly elevated the risk (OR = 2.088, 95% CI: 1.414–3.083, p < 0.001). In contrast, history of cardiopulmonary disease (OR = 0.962, 95% CI: 0.670–1.382, p = 0.846), length of hospital stay (OR = 0.780, 95% CI: 0.536–1.135, p = 0.194), sex (OR = 0.971, 95% CI: 0.689–1.367, p = 0.865), as well as allergy history (OR = 0.955, 95% CI: 0.505–1.805, p = 0.886) did not demonstrate statistically significant associations in the multivariable analysis (Detailed results are presented in Table 4). These results indicate that lower ulinastatin dosage (≤30 × 10⁴ U), longer medication duration, concomitant medication use, and ICU admission are significant independent risk factors for cardiopulmonary adverse outcomes in patients treated with ulinastatin.

To translate the identified risk factors into a practical tool for preventive health screening, we developed a simple risk scoring system based on the regression coefficients (β) from the multivariable model. The goal was to create an intuitive score that could be easily calculated at the point of care to guide clinical behavior. The smallest β-value (0.736, for concomitant medication use) was used as the reference to standardize point assignments. The relative weight for each variable was calculated by dividing its β-coefficient by this reference value, followed by rounding to the nearest integer to assign practical point scores. The resulting points-based tool integrates the four independent risk factors as follows: ulinastatin dosage <30 × 10⁴ U (β = 1.717, standardized value = 2.33, assigned 2 points) and ICU admission (β = 1.318, standardized value = 1.79, assigned 2 points) were the strongest predictors. Medication duration >1 day (β = 0.772, standardized value = 1.05, assigned 1 point) and concomitant medication use (β = 0.736, assigned 1 point) were also included in the scoring system. This yields a total risk score ranging from 0 to 6. The detailed derivation is shown in Table 5. This risk scoring system transforms complex statistical predictions into a readily applicable clinical tool. By allowing for the immediate stratification of patients into distinct risk categories based on routinely available data, it facilitates proactive, evidence-based decision-making. This directly supports the promotion of safer medication practices by enabling clinicians to identify high-risk patients prior to or during treatment, thereby targeting monitoring resources and guiding preventive strategies.

The predictive accuracy of the developed risk scoring system was evaluated using ROC curve analysis. Based on the scoring criteria established from multivariable regression, each of the 11,218 patients included in the study was assigned a total risk score ranging from 0 to 6. These scores were used as the test variable, with the actual occurrence of cardiopulmonary adverse outcomes serving as the state variable. The resulting ROC curve (Figure 2A) demonstrated favorable discriminatory performance, with an AUC of 0.779 (95% CI: 0.741–0.817). This indicates that the risk scoring system has good overall accuracy in distinguishing between patients who experienced cardiopulmonary adverse outcomes and those who did not. An AUC value of 0.779 suggests that the model possesses clinically useful predictive ability, significantly exceeding random chance (AUC = 0.5) (Thomas et al., 2015). The narrow confidence interval further supports the robustness of this estimate. These findings confirm that the risk assessment system, incorporating four readily available clinical variables, serves as a reliable tool for stratifying patients according to their risk of developing cardiopulmonary adverse outcomes following ulinastatin administration.

To assess its prognostic utility, the performance of the newly developed risk assessment system was quantitatively evaluated against individual clinical predictors using ROC analysis. The composite risk score demonstrated significantly superior predictive accuracy, yielding an AUC of 0.779 (95% CI: 0.741–0.817). In contrast, the predictive capacities of the individual risk factors were considerably lower: ulinastatin dosage <30 × 10⁴ U (AUC = 0.681, 95% CI: 0.640–0.721), combination medication use (AUC = 0.606, 95% CI: 0.561–0.651), ICU admission (AUC = 0.578, 95% CI: 0.535–0.620), and medication duration >1 day (AUC = 0.522, 95% CI: 0.477–0.568) (Table 6; Figure 2B). To transform the risk score into an actionable tool for clinical practice, an optimal cutoff value of ≥3.5 was established using Youden’s index, providing the optimal balance between sensitivity and specificity for identifying high-risk patients (Tang et al., 2024). As the total score is an integer, this cut-off is operationally applied such that a score of 0–3 is classified as low-risk, and a score of ≥4 is classified as high-risk. This threshold enables reliable stratification of patients into distinct risk categories, facilitating early identification of those who would benefit most from intensified monitoring and preventive care strategies. These findings affirm that integrating multiple predictors into a simple composite score provides a more robust foundation for clinical decision-making and resource allocation than relying on any single clinical indicator.